the irs research bulletin · 2017. 4. 8. · t his 1999 edition of the irs research bulletin...

The IRS Research BulletinPublication 1500 1999

Internal Revenue Service

Wayne ThomasAssistant Commissioner (Research and Statistics of Income)

Howard OglesbyDirector, Office of Research

Michelle KaplanDirector, Office of Planning and Finance

Russell GeimanChief, Projections and Compliance Studies Group

Javier FraminanTerry ManziMelissa KovalickEditors

Stuart SimpsonProduction Assistant

Product Available ElectronicallyThis publication is available on the IRS’s Digital Daily Web site. Non-IRS users can reach that site atwww.irs.gov. For IRS employees, that site can be accessed at www.irs.ustreas,gov. Select the “Tax Stats”option, and then the “IRS Research Bulletin” option (under the Statistical Publications heading).

Feedback WelcomedWe welcome user feedback on this publication. Please direct questions and comments toTerry Manzi at (202) 874-1083 or Javier Framinan at (202) 874-1104 of the Projections and ComplianceStudies Group, or by writing

Internal Revenue ServiceResearch and Statistics of Income OP:RSAttn.: Projections and Compliance Studies Pub. 1500 Staff1111 Constitution Ave., N.W.Washington, D.C. 20224

AcknowledgementsWe would like to thank the followingindividuals for their assistance in thereview of this publication.

Elinor ConveryDennis CoxCharles BennettKen HubenakJeff Butler

The IRS Research BulletinPublication 1500 1999

Suggested Citation:

United States Department of the TreasuryInternal Revenue ServiceThe IRS Research BulletinPublication 1500 (Rev. 11-99)

This 1999 edition of the IRS Research Bulletin (Publication 1500) updates a traditional publication of the IRSResearch organization. The Bulletin provides IRS executives, managers and staff, as well as interestedexternal stakeholders, insights into significant trends and major IRS research findings impacting Federal taxadministration. A lot has changed in the research area of IRS since our last 1995/1996 update of theBulletin. This includes the further maturation of our District Office Research and Analysis (DORA)capabilities, and the implementation of a more comprehensive and systematic approach to planning andallocating resources among the Service's operational functions (based on IRS research results). In addition,even greater changes are just around the corner with the phased-implementation of the IRS modernizationconcepts, including the movement to the four new operating divisions. Despite these changes, and indeedbecause of them, this release of the Bulletin is perhaps more timely than ever.

IRS’s vision for the future is supported by a new mission statement, a new balanced measures framework,and a renewed commitment to providing top quality service to taxpayers. This publication brings togetherextensive statistical information and other analytical materials which tie directly to this vision, including suchIRS strategic goals as making it easier to file; providing prompt, helpful treatment to taxpayers in balance duesituations; improving overall compliance; and increasing employee satisfaction. For example, the “trends”section contains a number of thought-provoking items relating to electronic commerce and e-file, customersatisfaction, and developments in the U.S. labor market, including some employee preferences.

In addition, among the research articles are two covering innovative taxpayer treatment programs designedto address noncompliance in a more “preemptive” and less intrusive fashion in the areas of self-employmenttaxes and duplicate claiming of dependents. A third article illustrates a new “risk-based” statistical approachto potentially screen accounts receivable cases at the point of initial assessment, thereby allowing for lessforceful IRS contacts for those likely to pay, and more accelerated processing for those not likely to pay. Afourth presents the first systematic estimate of the tax gap in the estate tax area. Other articles cover: IRSefforts to provide taxpayers with needed forms in a more effective manner; an improved methodology forselecting large corporations for examination; a statistical approach for deriving fewer, yet morecomprehensive, comparative measures of compliance; and a summary of the major IRS research findings inthe area of electronic filing.

This update of the Bulletin is timely from another perspective, as well. While the modernization efforts moveIRS toward new structures and new business practices, the first order of business quite often remainsmaking sure there is a firm understanding of the past. The Research Abstracts and the indices to priorresearch, also contained in this publication, provide a valuable reference guide to help answer the question“has any research ever been done on … ?”

The Bulletin is by no means the “last word” on IRS research. Knowledge acquisition is an ongoing,cumulative process, and some of the materials presented in this edition were prepared a year or so ago. Buthopefully you will find it to be the “first word,” i.e., the place you start when seeking out IRS research findings,and a vehicle to identify IRS staff and/or organizations that potentially could provide you with additionalinsights.

Wayne ThomasAssistant Commissioner(Research and Statistics of Income)

IRS Research Bulletin 1999 EditionTrends1

Articles

13 Alternative Treatment for Self – Employment Tax InventoryKay Anderson and Dan Beckerle

24 Duplicate Use of Dependent and Qualifying Children Social Security NumbersIvette Alamo-Tirado and Robert Holmes

31 High – Range Corporation Return Workload Selection System DevelopmentJames A. Wilhelm

39 Predicting Estate Tax Filings and Taxable GiftsJonathan Feinstein and Chih-Chin Ho

46 Payment Dynamics of Individual Accounts Receivable and a New Look at RiskJeff Butler

60 A Case Study of the Bank, Post Office and Library ProgramDenise York Young and Erika D. Alexander

65 Using Data Reduction Techniques to Analyze Baseline ProfilesLarry May and Anne Steuer

75 Review of the IRS’s Individual Return Electronic Filing and Related ResearchJavier Framinan

AbstractsReporting Compliance

89 Inadequate Compensation to 1120S Corporate Officers StudyBruce Korbesmeyer

90 Can Demographic Trends Predict Taxpayer Noncompliance?Kim M. Bloomquist

91 Pre-payment Position and Income Tax NoncompliancePeter D. Adelsheim, Ph.D.

92 Farm Labor Contractors Compliance with the Income Tax LawsJames L. Zanetti and Marlene Le

93 Erroneous PBA/PIA ProfileStan Griffin

94 Business Profitability and Income Tax ComplianceKim M. Bloomquist

95 Tax Table Clustering: Empirical Evidence of Noncompliance?Peter D. Adelsheim, Ph.D.

96 Empirical Goodness of Fit Scoring System (Applied to Tax Preparer Due Diligence)Curt Hopkins

97 Processing Year 1998 Criminal Investigation Questionable Refund Formula Development

James A. Wilhelm

98 Roadmap to the Future: Using Groupware as an Analytical Tool to Rank Emerging Compliance Issues

Deborah J. Meyers, Ph.D.

Payment Compliance

99 Closed Case Analysis to Support Collection of Agreed DeficienciesRonald L. Edgerton

100 Final Study Report: Self Employed Real Estate AgentsRick Denesha

101 Subchapter C Corporations with Low Retained EarningsMichael Hill and Richard Denesha

102 Estimated Collections on FY 1998 Accounts Receivable Dollar InventoryTodd Headrick

103 Decline in BMF TDA Inventory in Los Angeles DistrictScott Mendelson

104 Business Licensing as a Nonenforcement Approach to Increasing Tax Compliance in the Liquor Industry

Nancy Richman and Scott Mendelson

105 Analysis of Toll-Free Telephone DemandJeff Butler

106 Taxpayer Service Walk-in StudyCurtis R. Darling

Customer Service

107 Selected Estimates of Returns with Form 1099 InformationTerry Manzi

108 Alternative Signature Methods for Filing Individual Tax ReturnsT. Scott Shutt, Robert A. Kerr and Dennis L. Raup

109 Focus Group Report: On-Line Filing ProgramDru DeLong and David Browne

Statistical Tables

111 Table Notes

113 Table 1 Return and Economic/Demographic Data by

IRS Regions and States, 1991 – 2005

132 Table 2Return and Economic/Demographic Data by IRS Regions and Districts, 1991 – 2005

145 Table 3 Returns, FTDs, Withholding/Information Documents, and Economic/Demographic Data for the United States and IRS Centers, 1991 - 2005

Indexes

151 Index of Past and Current Articles

157 Index of Past and Current Abstracts

The views expressed in this publication represent the opinions and conclusions of the authors. They do not necessarilyrepresent the position of the Internal Revenue Service

By Melissa Kovalick and Russell Geiman

INTRODUCTION

Leaders in the Internal Revenue Service (IRS) long have been aware of the need topay attention to the larger economic and societal trends outside the organization.The Trends section of The IRS Research Bulletin is one tool designed to help IRSmanagement and staff with this important responsibility.

It is obvious that “external trends” impact federal tax administration and theachievement of our organizational goals. These include the types of new servicesthe IRS needs to provide customers, the recruitment and retention of IRS’s ownemployees, the emergence of new compliance issues, and the opportunities forimproved IRS products and services created by new technologies and otherdevelopments— to name just a few.

The Trends segment of the Bulletin is intended as a thought-provoking synopsis ofsome important trends impacting tax administration. On one level, the trendshopefully provide bits of useful information to IRS employees, such as by helpingexplain developments in the labor market and industry, or by highlighting somesimilarity between the IRS and the private sector. However, on another level, wealso hope these trends are a catalyst for further action by IRS readers. Forexample, a cited trend might spark an idea for a new IRS service to meet anemerging customer need. The reader then might research the matter morethoroughly and submit a formal employee suggestion or otherwise surface their ideato management. Similarly, a listed trend might suggest to the reader a possible rootcause for a particular compliance problem they have noticed recently. This reader,in turn, might seek the assistance of Research or other technical staff whosystematically could investigate their causal hypothesis.

To assist the reader in reviewing the following selected trends, we have groupedthem by six general areas. These are:

Internet – Electronic Commerce – Computer UseCredit Cards – Debt – BankruptciesLabor MarketWorkforce Characteristics and PreferencesEmployee SatisfactionCustomer Satisfaction

Within each general area, we have attempted to sequence the trend “bullets” suchthat related dimensions of specific developments are within proximity.

Internet - Electronic Commerce – Computer Use

• Although the number of electronic payment transactions is on the rise, 89 percent ofall payments are still cash. Of the remaining 11 percent, 90 percent are paperchecks or credit cards and the remaining ten percent are other noncashtransactions.(CNN interactive. October 1998)

• Forty-two percent of U.S. households pay 1.7 monthly bills electronically. Mostcommon methods of payment are direct debits from customers’ checking accountsmade to utilities, lenders, or insurers. Visa USA reports that the use of debit cardshas been growing by more than 50 percent a year.(American Demographics. March 1998)

• The total number of Internet users between the ages of 18 and 34 who went onlinein the past month is 17 million. 7.5 million were female, and the top reasons citedfor using the Internet were to seek information, e-mail, and use chat areas.(American Demographics. January 1999)

• America Online reports that the gender cybergap is narrowing, with female usersaccounting for 52 percent of total AOL users. Four years ago, only 16 percent ofAOL members were female. Since women purchase three times as much by mail orphone orders, this narrowing gender gap has significant ramifications for onlineretailers.(CNN interactive. August 1998)

• The majority of American students are using computers by the time they beginschool. Seven out of ten children aged six to 17 have used a computer in the past30 days; 85 percent used them in school and 50 percent used them at home.Fifteen percent look up information about possible items to purchase and another 15percent read periodicals online.(American Demographics. April 1998)

• Senior citizens are making huge leaps in the cyberworld. More than 20 percent owna computer, and an estimated nine million persons over the age of 50 go online.Seniors also comprise the group that most heavily uses financial service web sitesand online trading.(Business Week. March 1999)

• Households with annual incomes greater than $50,000 accounted for 74 percent ofonline sales. Although lower-income consumers are projected to be doing moreelectronic purchasing, upper-income households probably still will account for 66percent of sales by 2003.(Business Week. January 1999)

• An estimated 15 percent of all credit card enrollments will be initiated on the Internetby 2002, while less than one percent are initiated electronically today. Althoughmore than 15 million people conducted Internet searches for new credit cards, onlysix million completed online applications in the past two years.(American Demographics. January 1999)

Internet - Electronic Commerce – Computer Use (Continued)

• In a study of 700 Internet users, the average income of persons acquiring onlinecredit cards was $56,315, versus the $54,833 average income of active Internetusers. People who signed up for a card online were most likely white (81 percentversus 75 percent), and more likely to live on the West Coast. Of the $6.5 billion inonline purchases during the 1998 holiday season, more than 90 percent wascharged to credit cards. About half of all online shoppers say they use a single cardonline.(American Demographics. June 1999)

• Total electronic sales to consumers (“E-Commerce”) are expected to exceed $18billion for 1999, representing a $7.8 billion increase from 1998. By 2003, electronicsales may account for as much as six percent of total consumer retail spending.(Business Week. January 1999)

• Twenty percent of firms with fewer than 100 employees used the Internet in 1996;but by 1998, that number jumped to 41.2 percent. Thirty-six percent of smallbusinesses going online plan to use the Internet to sell their products. Revenues forsmall companies using the Internet average $3.79 million, while small companiesthat didn’t averaged $2.72 million.(CNN interactive. July 1998)

• A recent survey of 800 CEO’s by Price-Waterhouse Coopers revealed that half ofthe respondents believe online companies and other non-traditional rivals will pose athreat to their businesses in the future. However, the same group also foreseesmany corporate opportunities on the Internet; 40 percent think more than one-tenthof revenues will come from E-commerce in five years. Seventy-five percent of theparticipants reported that E-commerce currently accounts for five percent of theirtotal revenue.(Business Week. February 1999)

• Although the media has been focusing on online consumer sales, online commercebetween corporations also has been increasing significantly. Business-to-businesstrades over the Internet were expected to total over $15.6 billion in 1998 and shouldreach $175 billion by 2000.(Time Magazine. August 1998)

• Although E-commerce is on the rise, customer satisfaction for online purchases isgoing down. Cited problems include merchandise availability (15 percent), shippingand handling costs (14 percent), and slow site performance (13 percent). However,74 percent of the holiday season online shoppers were satisfied with their overallexperiences, and only five percent said they would spend less for online purchasesnext year.(CNN interactive. January 1999)

Internet - Electronic Commerce – Computer Use (Continued)

• Concern for privacy remains a key issue in making online purchases. Fifty-sixpercent of those polled by Business Week/Harris are “very concerned” their creditinformation will be misused by employees where they are making their purchases, ortheir credit information will be made available to others without their consent. Sixty-one percent said they would be more likely to use the Internet if their privacy couldbe protected. Concern for privacy extends to noncredit information as well. Fifty-nine percent of those surveyed claim they never register at sites which require thedisclosure of personal information, and 40 percent have given false informationwhen registering.(American Demographics. February 1999)

• According to a recent Business Week/Harris survey, 59 percent of those polledbelieve that the government should pass legislation regulating Internet privacy, andonly 19 percent felt that individual groups should be able to determine their ownstandards for privacy, without government intervention.(American Demographics. February 1999)

• In 1992, it was predicted that 100 million returns would be filed electronically by theyear 2000.(Fortune magazine. April 1998)

[Editor’s note: This 1992 prediction arose from a cross-functional IRS task forceoutside the IRS Research organization. Research forecasts of electronic filings(including those on magnetic tape) for 2000 prepared in fall 1999 project 33.6 millionindividual tax returns, 6.0 million business tax returns, and 0.1 million othermiscellaneous tax returns.]

• The IRS estimates that the error rate on paper returns is 20 percent, as opposed to0.5 percent for electronic returns.(Business Week. February 1999)

• One out of every five taxpayers waits until the last week to mail a return, and hasspent an average of nine hours and 54 minutes working on it.(CNN Interactive. April 1998)

Credit Cards – Debt – Bankruptcies

• Despite an average 2.5 percent fee for using credit cards to pay their taxes, moreand more wealthy filers are choosing to use plastic instead of checks. The reason?Credit card incentives, such as airline frequent flier miles and other rewards.(CNN Interactive. April 1999)

• After numerous bond defaults and bankruptcies in the 1980’s, corporations reducedtheir level of debt in the 1990’s. However, net new borrowing by nonfinancialcompanies hit a record $343 billion in 1998, rising by more than 10 percent annuallyfor the first time in a decade. Meanwhile, the difference between capitalexpenditures and cash flow grew from $5 billion in 1994 to $79 billion in 1998.(Business Week. April 1999)

Credit Cards – Debt – Bankruptcies (Continued)

• Seven out of ten Americans own at least one credit card. Thirty four percent ofthese cardholders do not know the interest rate of the card they use most often.People over the age of 55 are least likely to know their rate, at 44 percent. Mostspenders do not realize that there is a correlation between being unaware of thecost of charging purchases and the overall increase in consumer debt.(American Demographics. May 1997)

• From 1989 to 1995, the percentage of households with credit cards rose from 56 to67 percent. At the same time, the percentage of card-holders with incomes lessthan $25,000 increased from 22 to 28 percent. The average balance increased from$1,100 to $1,700 during this time, while the average card-holder’s liquid assetsdeclined by more than 25 percent.(Business Week. April 1999)

• The average American consumer carries more than $5,000 in credit card debt, whileover 1.35 million Americans filed for bankruptcy in 1998.(Money Magazine. April 1999)

• Fifty-five to sixty million American households have an average debt of more than$7,000 with over $1,000 in annual interest and fees. Those with the most debt arehouseholds with incomes at or slightly above the federal poverty income level of$16,036 per year for a family of four.(ABCNEWS.com. July 1998)

• Fourteen million students were projected to enroll in U.S. colleges and universities in1998. Two-thirds of college students own credit cards and carry phone cards. Theestimated spending power of all college students is more than $90 billion, with full-time, four-year enrollees spending over $30 billion a year.(American Demographics. March 1998)

• One in nine high school students has a credit card co-signed by a parent.(Business Week. February 1999)

• The average total debt of graduating college students in 1997 was $18,800,compared with $8,200, in 1991. Adding to the rising debt problem is the increasingnumber of students who are using credit cards to pay off tuition costs. It isestimated that $7.5 billion will be charged this year to pay for college bills. Sixtypercent of those who charge collegiate expenses pay their balances in full.(CNN interactive. October 1997)

• Approximately 1.35 million consumer bankruptcies were filed in 1997, whichrepresents a 49.9 percent increase over the number filed in 1992. As a share of allfilings, consumer bankruptcies are also higher than ever before. In 1987, consumerfilings represented 85.7 percent of total filings; by 1997, the share had grown to 96.1percent.(American Demographics. July 1998)

Credit Cards – Debt – Bankruptcies (Continued)

• Although every state in America saw an increase in consumer bankruptcy filingsfrom 1992 to 1997, Southern states have higher consumer bankruptcy rates overall.In 1997, Tennessee had the highest rate at 9.6 filings per thousand residents.Georgia ranked second, with 8.2 filings per thousand. Nevada, the third-rankedstate at 7.9 filings, was the only non-Southern state in the top five. Finishing fourthand fifth were Alabama and Mississippi, respectively.(American Demographics. July 1998)

• Bankruptcy rates are 18 percent higher in counties with one gambling facility, and 23percent higher in counties with five or more gambling facilities. Atlantic City has abankruptcy rate, which is 71 percent higher than any other county in New Jersey.Clark County, Nevada, where Las Vegas is located, has the highest bankruptcy ratein the state.(ABCNEWS.com. August 1997)

• With bankruptcy filings on the rise, banks must write off increasing amounts ofuncollectible debt from their credit card customers. In 1996, $17 billion was writtenoff as uncollectible. Regional areas with the highest numbers of filings share threebasic characteristics; a high divorce rate, lax rules on automobile insurance, and alarge population that lacks health insurance. (Fortune Magazine. March 1997)

• According to a study by Visa International and MasterCard International, nearly150,000 bankruptcy filers who were pardoned of their financial obligations last yearunder Chapter 7 plans could have repaid 64 percent of their unsecured debts. Thiswould have resulted in a potential recovery of more than $4 billion.(ABCNEWS.com. March 1998)

• Examining per-capita projections of spending can provide insight into the changingfortunes of consumer goods. In 1996, the average American spent $299 oncomputers (1992 dollars) and is expected to spend $2,953 by 2006. Car sales areexpected to suffer; average spending for people aged 16 and over is expected todrop from $360 to $302. In 1996, the average person spent $1,313 on clothing andshoes; by 2006, this figure will increase by 18 percent to $1,549.(American Demographics. August 1998)

• Between 1990 and 1998, the percent of owner-occupied housing units in UnitedStates counties increased from 63.9 percent to 66.3 percent. The number ofbusiness starts during this period decreased from 158.9 (thousands) to 155.1;however, the industrial production index (base year of 1987) increased from 98.9 to131.4.(Resident Population of the US, Bureau of the Census. 1998 Edition)

• In 1997, the three states with the highest homeownership rates were Minnesota,Kentucky, and Maine. States with the lowest rates of homeowners were California,New York, and Hawaii. Nationwide, 65.7 percent of the population are homeowners.

(1998 Statistical Abstract of the United States. Bureau of the Census.)

Labor Market

• In the 1970s and 1980s, the U.S. labor force grew at a pace of 2.3 percent. In thenext two decades, it is projected to grow less than one percent annually.(Business Week. March 1999)

• According to the Bureau of Labor Statistics, the U.S. labor force is expected to growmore slowly between 1996 and 2006 than it did in the previous decade. Between1986 and 1996, the growth rate was 14 percent. From 1996 to 2006, the labor isprojected to grow by 149 million participants, or 11 percent.(American Demographics. March 1999)

• Between 1996 and 2006, the labor force age 45 to 64 will grow faster than the laborforce of any other age group, while the labor force 25 to 34 years of age is projectedto decline by almost 3 million. The labor force participation rates of women in nearlyall age groups are projected to increase, while men’s labor force participation ratesare expected to continue to decline for all age groups under 45 years of age. TheAsian-and-other labor force and Hispanic labor force are projected to increase fasterthan other groups -- 41 percent and 36 percent, respectively.(Employment Projections, Bureau of Labor Statistics. December 1997)

• People residing in suburban areas are more likely to be in the labor force than thosewho live in central cities, regardless of age, sex, race, or ethnicity. Persons age 16and over living in the suburbs have a labor force participation rate of 69.9 percent,and an unemployment rate of 4 percent. However, central city dwellers have a 64.6percent participation rate, and a 7.3 percent unemployment rate.(“Issues in Labor Statistics,” Bureau of Labor Statistics. December 1998)

• According to Woods & Poole Economics, Inc., the number of U.S. jobs is projectedto increase 12.7 percent between 1998 and 2010, to almost 172 million. However,these jobs won’t be distributed evenly among the states; Summit County, Utah, isexpected to see a gain of 58 percent while Edgar County, Illinois may see a 10percent loss. Overall, southern and western counties will most likely see thegreatest increase in jobs, especially those in Florida, Georgia, Texas, Colorado,Alaska and New Mexico. The prospects for the East are not as great. No easterncounty appears in the top-100, and only four make the top-200: Burlington andSomerset, New Jersey; Washington, Rhode Island; and Saratoga, New York.(American Demographics. May 1998)

• As time goes by, some industries are shrinking rapidly while others are expanding.Since 1970, the number of general merchandise stores has declined from 25,032 to14,797 in 1996. There were 1,567 drive-in theaters in 1970, compared with 408 in1996. Some of the expanding industries include carpet/upholstery cleaning, whichgrew from 816 in 1970 to 8,879 in 1996. During this time frame, movie productionand services increased from 2,922 to 14,680, and the number of eating and drinkingestablishments went from 233,048 to 466,386.(Southwest Economy. January/February 1999)

• “New Economy” industries, such as software, communications, and consulting, areadding jobs to the labor market at a rate of 3.7 percent, twice as fast as the rest ofthe economy.(Business Week. February 1999)

Labor Market (Continued)

• Technical jobs comprise an average of 2.8 percent of total employment nationwide.Surprisingly, Massachusetts leads the states at 7.3 percent, beating out California at6.8 percent. However, San Jose leads the metro areas at 28 percent.(Business Week. April 1999)

• The number of jobs lost due to mergers almost doubled between 1997 and 1998,from 37,033 to 73,903. However, the number of mergers announced in 1998reflected only a 4 percent increase from 1997. Deregulated industries, such astelecommunications and financial services, received a disproportionate share oflayoffs. Throughout the 1990s, white-collar workers continued to share the badnews with blue-collar workers.(American Demographics. April 1999)

• The unemployment rate of 4.3 percent is at its lowest point in 28 years and theservice sector continues to show strong job growth. However, the manufacturingsector is continuing to cut jobs; steel makers have laid off 10,000 in the past year,while textile and apparel mills have cut 110,000 slots.(Business Week. February 1999)

• In March 1998, there were approximately 21,000 unemployed persons in the miningindustry. One year later, that number jumped to 32,000. Conversely,unemployment in the construction industry decreased from 593,000 in 1998 to490,000 in March 1999.(Employment Situation News Release, Bureau of Labor Statistics. April 1999)

• Between 1988 and 1996, the proportion of managers and professionals in theworking class increased four percent, to reach a total of 17 percent by 1996.Conversely, there has been a decline in the proportion of farm workers, serviceemployees, and craft/skilled workers.(American Demographics. March 1999)

• Faced with increasing competition as patented medications expire, drug companiesare adding to their marketing departments. The top 40 drug makers now deploynearly 59,000 representatives in the United States, up from 34,000 in 1994.(Business Week. May 1999)

• The number of automotive dealerships fell from 32,000 in 1972 to 26,000 in 1996.In contrast, employment in this area has grown from 800,000 to over 1 million duringthe same time. Since 1980, the occupational mix in the auto arena also has beenshifting; the demand for technicians has declined while the share of supervisors hasgrown. Salespersons and service/parts workers have shown very little change inemployment patterns.(“Issues in Labor Statistics,” Bureau of Labor Statistics. January 1999)

• About 12.6 million people, or one in ten workers, were classified into one of fouralternative employment arrangements in February 1997. Independent contractorswere the largest at 8.5 million, followed by on-call workers (2 million), temporary helpagency workers (1.3 million), and contract company employees (800,000).(Monthly Labor Review. November 1998)

Labor Market (Continued)

• After adjusting for inflation, doctors’ median net income has fallen 1.4 percent eachyear since 1993. Before the recent managed care reform that focused on how feeswere negotiated, doctors’ salaries sometimes climbed 10 percent annually.(CNN Interactive. May 1999)

• For the third straight year, Internet workers are making the largest gains in the payarena. Office managers and secretaries are expected to see a pay hike of 4.8percent, while information technology workers have the highest percentage increasein starting salaries among all industries, at 7.3 percent. The large increase insalaries represents the demand for IT workers, especially in the areas of finance,insurance, and real estate.(Business Week. February 1999)

Workforce Characteristics and Preferences

• The median job tenure of wage and salary workers with their current employeredged down to 3.6 years in 1998. In 1996, average tenure was 3.8 years. Formales, declines in tenure were apparent in almost all of the age groups. Amongfemales, there was very little change in overall tenure.(Monthly Labor Review. October 1998)

• In 1980, the average male college graduate earned about one-third more than theaverage male high school graduate. By 1993, the gap in earnings had increased tomore than 70 percent. However, this trend may be ending for three main reasons:continued low unemployment rates leave companies with no choice but to hire lessskilled workers; an increasing supply of skilled workers is holding down wage growthat the top; and information technology is becoming more user-friendly, thus enablingless-educated workers to access it more easily.(Business Week. March 1999)

• A recent Louis Harris poll discovered the workweek has increased by 15 percent inthe last 25 years, while leisure time has decreased 37 percent. Data from theFamilies and Work Institute indicates 13 percent of U.S. workers are holding downtwo jobs. This may explain why the National Institute of Occupational Safety andHealth reports that at least one quarter of today’s labor force feels stressed at work(CNN Interactive. April 1999)

• Skilled and highly educated workers are the most likely to work longer hours.Among those with managerial, professional, or technical jobs, more than 33 percentof men and 17 percent of women put in 50-hour plus weeks, compared with 20percent of men and 7 percent of women in other occupations. A study by Jacobsand Gerson finds almost half of the workforce would prefer to work fewer hours, andmore than a quarter said they would be willing to take a pay cut to make it happen.(Business Week. April 1999)

• The overall number of persons working at home between 1991 and 1997 did notgrow dramatically; but the number of wage and salary workers doing work at homedid. In 1991, 1.9 percent of at-home workers were wage and salary; by 1997 thatnumber had increased to 3.3 percent. Nearly nine out of ten people doing work athome were in white-collar occupations.(Population Survey, Bureau of Labor Statistics. March 1998)

Workforce Characteristics and Preferences (Continued)

• According to the White House Domestic Policy Advisor, Bruce Reed, 75 percent ofAmerican families are made up of two working parents. From 1985 to 1997, thenumber of women with children working at home increased 32 percent, from 18 to24 million.(Business Week. May 1999)

• Men are more likely (50 percent) to prefer full-time jobs outside the home thanwomen (19 percent). Of those who do desire to work at home, 34 percent are white-collar workers and 33 percent are dual-earner couples. Working at home is leastappealing to those age 60 and over (16 percent) and African-Americans (18percent).(American Demographics. May 1998)

• Women tend to be more conservative when it comes to selecting a workplace. Only49 percent of men insist on working for a profitable company, while 79 percent ofwomen do. Women also prefer to work in team-oriented environments and inmidsize companies with 100 to 500 employees.(Business Week. March 1999)

• Although the share of workers who were union members fell from 14.1 percent in1997 to 13.9 percent in 1998, the overall number of union members rose for the firsttime in five years. There was little change for private industry workers, but thenumber of union members among government employees rose for 6.7 to 6.9 million.Local governments were most likely to be unionized, followed by the federalgovernment.(Monthly Labor Review. January 1999)

• According to a recent AFL-CIO poll of the general public that was employed in non-supervisory jobs, 44 percent said that they would vote in favor of forming a union attheir workplace. An additional 20 percent were less certain, but still positive to theissue of a union, saying that it was better to join together at a work site to solveproblems.(American Demographics. March 1999)

• In 1996, six in ten job changers cashed out their retirement savings, instead ofrolling them over into IRA’s or employee-sponsored plans. Eighty-one percent ofworkers with savings less than $3,500 cashed out, while only 14 percent withsavings between $50,000 and $100,000 followed suit.(American Demographics. April 1999)

• In May 1991, 15.1 percent of full-time wage and salary workers were on a flexiblework schedule allowing them to vary their starting and ending times. In May 1997,27.6 percent (about 25 million workers) were on such a schedule. Executives,administrators, and managers are most likely to be on a flex schedule (42.4percent), while less than one-fourth of those employed in administrative support orservices have the option to do so.(Bureau of Labor Statistics News Release. March 1998)

Workforce Characteristics and Preferences (Continued)

• According to a 1990 to 1997 Federal Reserve Bank of St. Louis/U.S. Census studywhich focused on the migration of 59 metro areas with population over one million,the following are the top five most livable: Las Vegas, Nevada; Atlanta, GA; Phoenix,AZ; Austin, TX; and Raleigh-Durham, NC. The bottom five metro areas are OrangeCounty, CA; Miami, FL; San Jose, CA; New York City, NY; and Los Angeles, CA.(Business Week. May 1999)

Employee Satisfaction

• Starting in 2000, part of the compensation for top United Airlines Inc. executives willbe tied to worker satisfaction as measured by an outside survey firm. Their bonuspay also will be based on customer satisfaction and on-time performance. Together,the three new criteria will account for more than half of what the top 625 UALmanagers receive.(Business Week. March 1999)

• Chuck Knight, CEO of Emerson Electric, firmly believes sales growth leads to abetter bottom line. To help get his message across, more than half of executives’annual incentive pay is linked to revenue growth, as opposed to earnings growth,ROE, and other traditional measures.(Fortune Magazine. April 1999)

• FedEx’s “People-Service-Profit” philosophy is based on the idea that a motivatedand conscientious workforce will provide professional service to customers, which inturn will ensure profits and continued corporate growth.(FedEx Homepage – Work Culture and Diversity; September 1999)

• IBM offers different programs to compensate employees for their contributions.Base Pay reflects long-term responsibilities and skills, whereas the Variable PayProgram is designed to reward employees based on how they, their business units,and IBM perform against annual objectives in key areas.(IBM Homepage – Benefits Section; September 1999)

• Continental Airlines provides employees with the opportunity to share their successthrough the Stock Purchase Plan and the Profit Sharing Plan. In addition,Continental offers a cash bonus to employees on a monthly basis as a reward fortheir dedication and teamwork when their on-time goals are met.(Continental Airlines Homepage – Benefits and Incentives; September 1999)

• Companies with employee stock-ownership plans have average shareholder returnsof 26.1 percent; those without ESOPs have an average return of 19.2 percent.(Business Week. May 1999)

• Southwest Airlines allows eligible employees to participate in a Profit Sharing Plan,funded by company contributions to profit sharing accounts. Contributions are madeto the accounts when Southwest meets profitability goals set each year. Employeesalso can share in Southwest’s success by investing in their stock, which ispurchased through payroll deductions at a reduced rate.(Southwest Airlines HomePage - Employee Benefits Summary; September 1999)

Employee Satisfaction (Continued)

• The ownership stakes of directors and senior officers of publicly traded U.S.corporations rose from 12.9 to 21.1 percent of their company’s stock from 1935 to1995. Meanwhile, the average value of their combined holdings increased from $18million to $73 million (measured in 1995 dollars).(Business Week. April 1999)

• In 1997, seven percent of managers feared for their jobs due to poor corporateresults; by 1998, 24 percent expressed concern.(Business Week. March 1999)

• According to an audit by the General Accounting Office, approximately 15 percent ofIRS employees who were investigated in IRS thefts from 1995 to 1997 lackedadequate background investigation checks.(CNN Interactive. December 1998)

• Of the 87 claims of rights violations against IRS agents from 1996 to 1998, nonewere successful. Ninety-three percent of lawsuits filed against IRS enforcementagents during these years involved motor vehicle accidents, while only three casesinvolved the use of search warrants by agents.(CNN Interactive. April 1999)

Customer Satisfaction

• IRS is not the only public agency under watch; 36 states currently have annualreport cards on all of their schools, and 13 of these states require the evaluations tobe sent to all parents.(CNN Interactive. April 1999)

• According to a poll by Pew Research, 60 percent of people surveyed had anunfavorable opinion of the IRS. Forty-four percent gave poor ratings to Congress,and the Pentagon and Postal Service were viewed unfavorably by 19 and 11percent, respectively.(CNN Interactive. April 1998)

• According to a Harris Poll, over 75 percent of the population surveyed believe theyhave been treated fairly by IRS workers. Eighty-three percent used the word“courteous” to describe Service employees they had encounters with. The majorityof Americans also believe that evasion of taxes is more rampant than IRSharassment of taxpayers.(Money magazine. April 1998)

• Only 42 percent of Americans believe that they get their money’s worth from federalincome taxes, 48 percent say they do not, and eight percent do not know. Womenare less likely to think that their tax dollars are spent wisely, at 38 percent, comparedwith 46 percent of males. Forty-six percent of people age 65 and above feel positiveabout the way tax dollars are spent, compared with 39 percent of 18-to-24-year olds.Over half of the households with incomes under $15,000 believe that the benefitsare worth the expense, while only 40 percent of households with incomes over$65,000 agree.(American Demographics. April 1997)

Customer Satisfaction (Continued)

• According to a four-year assessment of 15 government agencies by SyracuseUniversity, the Internal Revenue Service ranked 12th on the basis of factors such asmanagement of finances, human resources, information technology, capitalinvestment, and managing for results. The IRS received a grade of “C” overall, withindividual factor grades of B/C/D/NA/B, respectively. Social Security Administrationranked first, and the Federal Aviation Administration received the lowest score.(Washington Post. February 1999)

• The U.S. tax code began in 1913 as a 14-page law with a one page form. In 1998,the simplest form, the 1040EZ, had a 28-page instruction book. It takes the averagetaxpayer ten hours to complete the regular 1040.(Fortune magazine. April 1998)

• In the past, the IRS often has been criticized for providing poor customer service;but things are improving. This year, taxpayers got 13 million fewer busy signals and91 percent of the toll-free calls are being answered. In 1997, callers got throughonly 66 percent of the time, and only 39 percent in 1995. Nationally, accuracyscores are up to 93 percent as opposed to 63 percent in 1989. IRS employees inBaltimore gave correct advice 100 percent of the time in a random test.(CNN Interactive. April 1998)

• In an effort to improve customer service, the IRS now is posting “special taxpayeralerts” on its Web site to describe errors and other problems, the number of peopleaffected, where they may live, and what they can do about it. In addition, the IRSunveiled a draft proposal that would provide $2 million in grants to organizationsproviding legal assistance for low-income taxpayers involved in tax disputes.(CNN Interactive. January 1999)

• A recent survey compiled by field Taxpayer Advocates listed the 20 most seriousproblems facing tax practitioners. Topping the list was complexity of the tax laws,including figuring out exemptions, filing status, and the EITC (Earned Income TaxCredit). In second place was customer service/telephone access, with mostcomplaints focusing on inconsistent answers and inconvenient times and locationsof help centers. In third place was the cost of electronic filing, followed by offer-in-compromise (OIC) program issues. And the fifth biggest problem area focused onpenalties, such as inconsistency in applying criteria and the use of penalties as anegotiation tool.(Daily Tax Report, Bureau of National Affairs. January 1999)

Alternative Treatment for Self - Employment Tax Inventory

By Kay Anderson and Dan Beckerle

The self-employment tax (SET) inventory consists of approximately 400,000 Individual Income Tax Forms1040 with reported income which appear to be subject to self-employment tax but do not have Schedules SEattached. Kansas City Service Center personnel and Kansas-Missouri District Office Research and Analysis(DORA) staff tested an educational letter on a sample of taxpayers from the SET inventory, trackingresponses and measuring the letter’s impact on compliance. Over 20 percent of the taxpayers that receivedthe educational letter voluntarily amended their returns and on average paid an additional $260 in tax andinterest due. The Kansas-Missouri DORA subsequently helped determine at which point in the processingstream the educational letter should be generated and mailed. The team also investigated ways to eliminatefrom the SET inventory those taxpayers not liable for self-employment tax. The success of this researchlead to its full implementation as one of IRS’s ten National Strategies to increase compliance using non-traditional enforcement methods.

Introduction

Individuals are required to pay self-employmenttax on their self-employment income.1 On anincome tax return, self-employment income isreported on either Schedule C, Schedule F,and/or line 21 of Form 10402. Self-employmentincome also should be reported on lines 1 and 2of Schedule SE. During filing year (i.e., returnprocessing year) 1996, approximately 19.7million individuals filed a Schedule C or F, orotherwise reported self-employment income ontheir return.3 In a typical year, however, morethan 400,000 individual tax returns are filed withwhat appears to be self-employment income butwithout a Schedule SE attached, as required.These returns comprise the self-employment tax(SET) inventory, also referred to as the V-codeinventory.4

1 United States Code, Title 26 Internal Revenue Code, Subtitle A,“Income Taxes,” Chapter 2, “Tax on Self-Employment Income.”2 “Other Income” is currently, and since tax year 1994 has been,reported on Line 21 of Form 1040. However, for the tax year onwhich the research study was conducted, 1993, and prior tax years,“Other Income” was reported on Line 22.3 Statistics of Income Division report “U.S. Total of All Returns:Selected Sources of Income, Exemptions, Deductions and Tax bySize of AGI,” report symbol NO-R:S-87.4 The SET inventory consists of all returns assigned the audit code“V” by the Martinsburg Computing Center. For tax year 1993,individual income tax returns were assigned the “V” audit code ifthere was net positive income in excess of $400 from eitherSchedule C, Schedule F, and/or Line 22 “Other Income,” wagessubject to FICA were less than $57,600, and there was no ScheduleSE with the return.

After the Martinsburg Computing Centeridentifies the SET inventory, it is made availableto the Service Centers as discretionary work fortheir Correspondence Examination (Corr Exam)programs. The SET inventory ordered by CorrExam at each Service Center for file years 1994,1995 and 1996 is shown in Table 1. MostService Centers audit some portion of theirordered SET inventory, depending on theiravailable staff and the returns’ potential dollaryield.

The Kansas-Missouri (KSMO) District OfficeResearch and Analysis (DORA) and the KansasCity Service Center (KCSC) tested the use of aneducational letter as a way to bring taxpayers inthe SET inventory into compliance without theuse of audits. If successful, this treatment wouldaddress the non-compliance of the unworkedSET inventory, as well as some of the inventorynormally ordered for examination (i.e., audit).

In this report, we (the research analysts from theKSMO DORA) first explain the method used totest the educational letter and the limitations ofthe test. Next, we report our findings and theestimated costs and benefits of fullimplementation of the tested treatment. Lastly,we discuss the conclusions we drew from thisresearch.

Research Articles

ALTERNATIVE TREATMENT FOR SELF-EMPLOYMENT TAX INVENTORY

The IRS Research Bulletin P u b l i c a t i o n 1 5 0 0 1999 Update21

Table 1. Self-Employment Tax (SET) Inventory by Service Center and Selected File Year.

Service File Year 1994 File Year 1995 File Year 1996

Center Ordered Total Ordered Total Ordered TotalAndover 8,036 38,764 6,000 34,168 21,681 30,459Atlanta 6,000 29,394 0 29,748 4,057 31,108Austin 8,531 76,210 0 77,610 0 68,698Brookhaven 0 19,433 0 21,266 3,235 22,492Cincinnati 10,000 74,312 10,000 76,718 0 83,228Fresno 19,998 36,364 0 32,044 7,721 33,181Kansas City 8,000 38,551 5,000 34,918 9,096 35,711Memphis 2,000 28,356 5,000 29,104 8,000 29,588Ogden 10,000 45,397 200 52,652 6,000 53,951Philadelphia 7,992 36,029 0 46,944 0 41,041

Total 80,557 422,810 26,000 435,172 59,790 429,457 Percent 19.0% 100% 6.0% 100% 13.9% 100%

Extraction Date 12-28-94 12-27-95 12-24-96

First Mailing – Pilot Test in the Kansas CityService Center

In 1993, the KCSC initiated a project that testedthe impact on compliance of sendingeducational letters to taxpayers with less than acertain self-employment income threshold. Thetest was limited to this group (herein referred toas the “lower strata”) because the KCSCgenerally orders SET inventory returns with self-employment income above the threshold (hereinreferred to as the “upper strata”) for possibleexamination. A group of taxpayers from thelower strata of the SET inventory for the 1991tax year was selected randomly to receive aneducational letter explaining self-employmenttax requirements. The letter encouragedtaxpayers to file amended returns if theydetermined they were liable for the tax. Themailing included the necessary forms andinstructions to make filing an amended returneasier. All responses to the letter werevoluntary. The educational letter proved to beeffective.5 Over 20 percent of the taxpayers

5 “Enhancing Voluntary Compliance with the Self EmploymentTax Provisions of the Internal Revenue Code.” Unpublished:Department of the Treasury, Internal Revenue Service, KansasCity Service Center, August, 1994. The differences between the1993 study by KCSC and the 1995 joint KCSC/DORA study arediscussed in the study plan for this project, “Study Plan: IssuesAssociated With Self-Employment Tax Compliance.”Unpublished: Department of the Treasury, Internal RevenueService, Kansas-Missouri District Office Research & Analysis,June 17, 1996.

receiving the educational letter took some formof corrective action.

Second Mailing – More Detailed Study

In 1996, KCSC personnel and the KSMO DORAstaff conducted a second test of an educationalletter mailing to the lower strata of the KCSCSET inventory. The primary objective of thisstudy was to determine more precisely theeffectiveness of an educational letter in bringingtaxpayers in the lower strata SET inventory intovoluntary compliance (i.e., “voluntarily” inresponse to an educational letter rather than aCorrespondence Examination). A secondobjective was to quantify the amount of self-employment tax not being collected fromtaxpayers in the lower strata of the SETinventory. Responses to the mailing alsoprovided information about why taxpayers didnot file the Schedule SE, what percent weremisclassified workers, what impact paid taxreturn preparers had on the SET inventory, andto what extent unclaimed but deductibleexpenses reduced the estimated employmenttaxes due.

For this project, we randomly selected a sampleof 2,189 taxpayers from the 25,469 returns inthe lower strata KCSC SET inventory for taxyear 1993,6and mailed them one of two letters

6 The unworked tax year 1993 returns were extracted on July 25, 1995,from returns processed during file year 1994.



on January 19, 1996 (see the Appendix forcopies of the letters).7 We included forms,publications, instructions, and a self-addressedreturn envelope in the correspondence to reduceburden for those taxpayers who needed to fileamended returns. All responses were voluntary.A control sample of 500 taxpayers not sent theletter was drawn from the lower strata forcomparison purposes.

We tracked both telephone and writtenresponses. The educational letters included atoll-free telephone number that routed calls to asite specifically designated for the test at theKSMO District’s Taxpayer Service location in theSt. Louis post of duty. The Taxpayer ServiceRepresentatives (TSRs) answering calls to thisphone were provided with a questionnaire toassist them in determining the caller’s self-employment tax liability. The TSRs recordedeach caller’s identification along with theirresponses to the questions. All writtenresponses received by the KCSC wereforwarded in their entirety (except for originalreturns and payments) to the KSMO DORA, aswell.

The KSMO DORA reviewed and analyzed thewritten and telephone responses. We thenconducted a study of those taxpayers that didnot respond to the letter. We assumed themajority of non-response to the mailing occurredfor two main reasons--either the letter’srecipients were not liable for the tax, or theywere liable but chose to ignore the letter. Tocorroborate these assumptions, we drew asubsample of 504 taxpayers from the originalsample of 2,189 that were sent the mailing, andordered their returns for delivery from theFederal Records Center. These 504 returnswere classified for audit. Taxpayers notselected for audit included those that (1)responded to the letter and were found notliable, (2) had already filed an amended return,and (3) had SET liability below audit productivitycriteria. The remaining were sent to KCSC’sCorr Exam unit for audit. Of the 504 returns,173 ultimately were audited. 7 If the taxpayer reported income on Line 22 of Form 1040(“Other Income”) and therefore possibly was liable for either Self-Employment Contributions Act (SECA) tax or their share ofFederal Insurance Contributions Act (FICA) tax, they were sent thefirst letter shown in the Appendix. If a taxpayer reported incomeonly on Schedules C and/or F, they were potentially liable for onlySECA tax and were sent the second letter shown in the Appendix .

Tracking Amended Returns. A number ofrespondents indicated they would file amendedtax returns; however, resources were notavailable to conduct Integrated Data RetrievalSystem (IDRS) research on the entire sample of2,189 taxpayers sent the letter to determine howmany amended returns were filed. Ultimately,we conducted IDRS research on arepresentative sample of 1,000 taxpayers sentthe letter (including the 504 considered foraudit), plus the 500 taxpayers in the controlgroup not sent the letter.8

The analysis showed some taxpayers filedamended returns for reasons other than receiptof the educational letter. For the purpose of thisstudy, only amended returns to tax year 1993with a code indicating a change to self-employment tax were classified as having beenaffected by our educational letter.

Also, we conducted IDRS research on thecontrol group of 500 not sent the letter toestimate the natural rate at which taxpayers fileamended returns without the influence of theeducational letter.

Finally, we obtained data for the KCSC SETinventory for years both preceding (tax year1992) and succeeding (tax year 1994) the taxyear 1993 under study. We matched these dataagainst the 1993 data to get a measurement ofthe amount of turnover that occurs in the SETinventory.

Taxpayers Liable for Self-Employment Tax

Of the 504 taxpayers drawn for the auditsubsample, 217 taxpayers (43.1 percent) werefound liable for self-employment tax, 200 (39.7percent) were found not liable, and nodetermination could be made on the remaining87 (17.3 percent).9

8 IDRS is the computer interface with Masterfile, IRS’ computerrecord of tax return and taxpayer information. IDRS is the systemused to retrieve tax return information from IRS’ computerrecords.9 We were unable to determine liability of 42 taxpayers becausetheir cases were unresolved as of February 1997. (They eitherwent from Examination Division to default status without makinga payment, were still open in Exam, or had statutory notices stillpending.) We did not make determinations on an additional 37taxpayers who did not meet audit productivity criteria and thuswere not examined. We were unable to make determinations oneight other taxpayers for various other reasons.



We categorized a taxpayer as liable in twocircumstances: either if they filed an amendedreturn, or as determined by an audit byExamination. We did not count as liable thosetaxpayers whose examinations closed in defaultstatus (i.e., the taxpayer did not respond to theRevenue Agent’s Report or the subsequentletter IRS sent assessing additional tax), unlessIDRS indicated a payment to the taxpayer’saccount.

Amended Returns. Of the 1,000 test subjecttaxpayers for which IDRS research wasconducted, 223 (22.3 percent) filed amendedreturns and paid an additional $58,094 (averageof $260 per amended return) in taxes andinterest. Projecting this proportion to allrecipients of the mailing, we estimate 488 (± 57at 95-percent confidence) taxpayers filedamended returns as a response to theeducational mailing and remitted an additional$127,000 (± $20,000 at 95-percent confidence)in taxes10 and interest.

Control Group. Of the 500 taxpayers in thecontrol group who did not receive theeducational letter, only six (1.2 percent)amended their 1993 returns. Three of thoseamended returns had adjustment reason code44, indicating a change to self-employment tax.Therefore, we conclude the educational letterhad a significant, measurable impact on the rateat which taxpayers filed amended returns.

Audited Returns. Of the 173 returns that wereselected for audit, Corr Exam assessed $33,214in taxes and interest – an average of $192 perreturn. Twenty of the audits resulted in nochange. As of the February 1997 conclusion ofthe study, several cases remained in openstatus and $29,438 of the assessments fromthese cases had been collected.

Estimated Revenue and Costs

Using our estimated percent of the lower strataliable for self-employment tax and the dollarscollected from taxpayers in our sample, weprojected the total amount of self-employmenttax not being collected from the lower stratanationwide. We used the file year 1995 SET 10 Taxpayers are allowed an adjustment to income in the amountof one-half of their self-employment tax. This reduction in taxableincome results in reduction in income tax liability. All paymentsof tax reported here are self-employment taxes net of any reductionin income taxes.

inventory levels (344,973 lower strata returns)and a two-sided, 95-percent confidence interval.For our subsample of 504 taxpayers consideredfor audit, the mean tax liability (not includinginterest) was $97.45, with a standard deviationof $151.74. Thus, we projected the uncollectedtax for the total population of lower strata returnsto be $33,600,000 ± $4,570,000.

We also projected the costs and benefits ofimplementing the tested treatment to the entireSET inventory nationwide. In making theseprojections, we estimated minimum revenueamounts by using the lower, one-sided, 95-percent confidence limits and maximum costs byusing the upper, one-sided, 95-percentconfidence limits. We also assumed theinventory could be reduced by filtering outtaxpayers who are under the age of 18 andtaxpayers who report non-passive losses onSchedule E which offset Schedule C and/orSchedule F income.

We projected the benefits of a nationwidemailing for the lower strata as described above.Although we did not test the treatment on theupper strata, we estimated the potential revenuethat the letter would generate from this strata byassuming that, except for the amounts of self-employment income, the upper strata wouldrespond in a manner similar to the lower strata.11

The results of the estimated revenues areshown in Table 2. These estimates do notinclude any audit revenues or costs.

11 The study team concluded that the most cost effective option fortesting (in effect) the educational letter on the upper strata was toadminister it to the entire upper strata inventory concurrently withtreatment of the lower strata. Such treatment at a nationwide levelwould add only $140,000 to the project cost while potentiallyadding significantly more revenue.



Table 2. Estimated Revenue, Cost, andReturn on Investment forNationwide Mailing.

Strata Tax &Interest

Collected

Costs toCollect

Return onInvestment

Projected(LowerStrata) $14,100,000 $536,000 26

Potential(UpperStrata)

$29,500,000 $140,000 211

Total $43,600,000 $676,000 64

Even if no revenue is generated by sending theletter to the upper strata, the return oninvestment is 21 ($14,100,000/$676,000 = 21).

Taxpayers Not Liable for Self-EmploymentTax

We found approximately 40 percent of the lowerstrata SET inventory not liable, based on the504 taxpayers considered for audit. Taxpayersprovided numerous reasons for not being liable.Two explanations accounted for 36 percent ofthe non-liable respondents. The first involvedtaxpayers claiming to have erroneously filed aSchedule F (reporting profit or loss from farmingon the Individual Income Tax Form 1040) ratherthan a Form 4835 (reporting farm rental incomeand expenses on the Form 1040). The secondwas for taxpayers under the age of 18 andemployed either by their parent(s) or as anewspaper carrier. Table 3 shows the variousexplanations (and associated frequencies)taxpayers provided. Further description of theseexplanations follow.

Table 3. Reasons for Not Being Liable for Self-Employment Tax.

Reason for not being liable Count PercentFiled Schedule F instead of Form 4835 41 20.5%Under Age 18 31 15.5%Included Both Spouses’ Incomes 16 8.0%Sched. E Loss Offset Sched. C or F Income 13 6.5%Administrator/Executor 7 3.5%Employee (Including Statutory) 7 3.5%Prizes/Awards 7 3.5%Minister/Religious Exemption 7 3.5%Sale of Asset 7 3.5%Schedule E Income 6 3.0%One-time Activities 5 2.5%Strike Benefits 3 1.5%Expense Reimbursement 3 1.5%Gambling Winnings 3 1.5%State Retirement System 3 1.5%Additional Expenses 3 1.5%All Others 38 19.0%Total 200 100%

Schedule F Returns. Of the 504 taxpayers inthe subsample, 75 filed Schedule F, Profit orLoss from Farming, with their Individual IncomeTax return (Form 1040). More than half of thesetaxpayers (41) were not liable for self-employment tax because they rent their farmsand rental income is not subject to self-

employment tax. However, these taxpayerserroneously reported their rental income onSchedule F; this income should be reported onForm 4835, Farm Rental Income and Expensesand Schedule E. In total, 76 percent (57 of theSchedule F filers in our subsample) were notliable for self-employment tax.



Taxpayers Under Age 18. IRS records indicatethat 28 of the 504 subsample taxpayers wereunder the age of 18 during the 1993 tax year.Of those, we determined 24 to be not liable,three to be liable, and one was indeterminable.Through direct communication with the taxpayer,we determined three others in the subsample,for whom there was no record of age, were notliable because they were under age 18 andemployed either by their parent(s) or as anewspaper carrier.

Spouses With Self-Employment Income LessThan $400. A tax return is included in the SETinventory when the combined total from allSchedules C and Schedules F included with thereturn exceed the self-employment incomethreshold of $434 per individual.12 Although thecombined self-employment income reported ontheir returns exceeded the threshold, 16taxpayers in the subsample (8.0 percent ofthose not liable) were not liable for self-employment tax because each spouse’sindividual self-employment income was lessthan $434.

Schedule E Losses. Thirteen respondents (6.5percent of those not liable) in the subsamplewere not liable for self-employment taxesbecause they had Schedule E losses whichoffset Schedule C and/or Schedule F income.Taxpayers should reduce their Schedule Cand/or Schedule F income subject to self-employment tax by non-passive partnershiplosses reported on Schedule E.

Administrator/Executor. Seven taxpayers in thesubsample indicated they had receivedcompensation as a result of serving as either anexecutor of a relative’s estate or as theadministrator of an incapacitated relative’saffairs. Since these taxpayers were not in thebusiness of serving as an executor oradministrator, they are not subject to self-employment tax.

Statutory Employees. Statutory employees13

accounted for 3.5 percent of the subsample that

12 Self-employment tax applies only to the amount of self-employment income in excess of $400. The amount of self-employment income subject to this threshold is equal to combinedSchedule C and Schedule F profit/loss reduced by 7.65 percent.Thus, the effective threshold is $400/(1-.0765) = $434.13.

was not liable for self-employment tax.Nationwide, approximately 65,900 workers areemployed as statutory employees. Althoughstatutory employees make up only 0.06 percentof all filers, they are disproportionatelyrepresented because they comprise 3.5 percent(± 1.6 percent at 95-percent confidence) of theSET inventory.

Minister/Religious Affiliation. Seven taxpayers inthe subsample were not liable for self-employment tax because they were eitherreligious clergy exempt from this tax or weremembers of a recognized religious group andhave personally waived their rights to SocialSecurity benefits. Taxpayers must obtainapproval from the Internal Revenue Service inorder to claim this exemption from self-employment tax. Although there is a ministerindicator field on the Individual Master File (IMF)to capture this exemption, it did not indicate thatany of these seven had a valid exemption. (Allseven provided proof of their valid exemption.)

Ministers are of interest in the SET inventorybecause they are routinely surveyed (i.e., thereturns are ordered from the records center butnot examined) by Corr Exam.14 If the ministerindicator on the IMF could be used as a filter toeliminate these returns from the SET inventory,the cost of ordering and then surveying thesereturns could be saved; however, the indicator isof questionable use. Over 62 percent of theministers in KCSC’s SET inventory are in theupper strata, but the IMF indicates only 32percent of them have an approved exemptionfrom self-employment tax.

13 Statutory employees include: (1) full-time traveling or citysalespeople who solicit orders from wholesalers, restaurants, orsimilar establishments, on behalf of a principal; (2) full-time lifeinsurance agents whose principal business activity is selling lifeinsurance and/or annuities for one life insurance company; (3)agent/drivers or commission-drivers engaged in distributing meat,vegetables, bakery goods, beverages (other than milk), or laundryor dry cleaning goods; and (4) home workers performing work onmaterial or goods furnished by the employer. Employers indicateon a Form W-2 whether a worker is classified as a statutoryemployee. Statutory employees report their wages, income, andallowable expenses on Schedule C, but they are not liable for self-employment tax because the employers are obligated to treatstatutory employees as employees for social security tax purposes.Source: 1993 U.S. Master Tax Guide, ¦ 941B.14 A taxpayer is considered a minister if he/she reports PrimaryBusiness Activity Code 8771 on his/her Schedule C.



Paid Preparers. Respondents whose returnswere completed by paid tax return preparerswere far less likely to be liable for self-employment tax than respondents who preparedtheir own returns. Only 25.5 percent oftaxpayers who used a paid preparer weredetermined to be liable whereas 51.0 percent oftaxpayers who prepared their own returns weredetermined to be liable. These results arepresented graphically in Figure 1.

Figure 1. Self-Prepared Returns versusPaid Preparer Returns.

0%

10%

20%

30%

40%

50%

60%

70%

No Preparer Paid Preparer

Not Liable Liable Liablity Undetermined

Inaccurately Filed Returns. Several reasons fornot being liable are grouped under the generalcategory of inaccurately filed returns. Theyrelate to issues associated with Schedule Eincome, sales of assets, state tax refund incomereported on Line 22, and scholarship income(the last two of which are included in the “AllOthers” category in Table 3). Taxpayers in thiscategory represent 67, or 33.5 percent, of thosein the subsample who are not liable.

One-time Activities. The “One-time Activities”category represents taxpayers who performedsome activity for a relative or friend for whichthey were paid; however, they were not in thebusiness of providing this particular service tothe general public. In many respects, some ofthe other categories listed in Table 3 alsorepresent one-time activities, includingAdministrator/Executor, Prizes/Awards, Sale ofAsset, and Strike Benefits. In total, 30 taxpayers(15 percent) in the subsample were not liablebecause of one-time activities.

Turnover in the SET Inventory

We evaluated the amount of turnover in theKCSC SET inventory. To do this, we matchedthe tax year 1993 KCSC SET inventory to the1992 and 1994 inventories. The followingshows the percentage of the 1993 SETinventory present in other years’ inventories:

SET Inventory (Year)

Percent of 1993SET Inventory Present

1992 25.3 %1994 22.3%

Both 1992 and 1994 10.8%Either 1992 or 1994 36.7%

We concluded there was a high degree ofturnover in the SET inventory with the possibilitythat a small core group remains in the inventoryyear to year.

Conclusion

The educational mailing proved to be aneffective alternative to audits as well as aneffective way to improve compliance amongtaxpayers in the SET inventory. Many of thetaxpayers liable for self-employment taxindicated they were unaware of their liability.Once made aware of their responsibility, manyfiled amended returns with full payments.

The amount of tax not being collected from thelower strata of the SET inventory is significant.We project it to be $33,500,000 (± $4,570,000 at95-percent confidence). Furthermore, much ofthe upper strata is not audited in any given year,thus likely adding millions of dollars to self-employment taxes not collected. However,limiting our revenue estimates to just the lowerstrata, we estimate the educational mailing willgenerate at least $14 million in additional taxreceipts.

In addition to responses from liable taxpayers,responses from those not liable providedvaluable information. This information indicatedways to reduce the SET inventory (i.e., byeliminating those not liable), as well as to handleit in a more cost-effective manner.



From Research to Operational Programs

Based on the effectiveness of the test, IRSadopted implementation of the educational lettertreatment to the entire SET inventory as aNational Strategy in Fiscal 1998. The NationalStrategies are a key component of IRS’sOperations Plan, the multi-year, multi-functionalplanning document for the Chief OperationsOfficer area. Fiscal 1998 served as a transitionyear where the letter was administered in amass mailing. For Fiscal 1999 and subsequentyears, mailing of the letter will be accomplishedthrough a computer generated notice, reducingthe time between filing of the original return andreceipt of the letter to less than four weeks.

Adoption of the strategy also includesidentification and removal from the inventorytaxpayers not liable for self-employment tax,review of the tax return processing proceduresconcerning self-employment tax, and possiblyconsideration of tax form changes to improvefiling accuracy. Preliminary indications suggestthe inventory can be reduced by as much as 25percent (62 percent of those not liable) by usinga more accurate selection process. Some taxreturn processing changes have been made,and tax form changes are under consideration.

About the Author(s):

Kay Anderson is an economist in the Kansas-Missouri District Office of Research andAnalysis. She received her M.A. degree ineconomics in 1988 from Northern IllinoisUniversity. She has been with the IRS since1995.

Dan Beckerle is an operations research analystTeam Leader in the Kansas-Missouri DistrictOffice of Research and Analysis. He receivedhis B.S. in Mining Engineering in 1979 from theUniversity of Missouri – Rolla and his M.B.A. in1990 from the University of Houston. He hasbeen with the IRS since 1982.


The IRS Research Bulletin P u b l i c a t i o n 1 5 0 0 1999 Update

28

Appendix A

Educational Letters



29

Kansas City, MO 64999 Mail Stop 4200

Person to Contact: CP:RAD:PIU

Telephone Number: District Office Toll Free Number: 1-800-736-5271Date:

Taxpayer Identification Number:

Dear Taxpayer:

We have analyzed the information you provided on your individual income tax return for 1993. You received andreported income which could be subject to self-employment tax or your share of FICA taxes not withheld fromyour wages. For example, Schedule C or F net profit or certain Form 1099 income is subject to either self-employment or FICA tax.

Enclosed is a copy of Schedule SE and instructions for self-employment tax, and a copy of EmployeeProcedures and Employee FICA Worksheet for your share of FICA taxes to assist you in determining if you areliable for either of these taxes.

If you determine that you owe self-employment or your share of FICA tax for 1993 or 1994 please complete aseparate Form 1040X, Amended US Individual Tax Return, for each year that you owe tax. The computationschedules (Schedule SE for Self-employed individuals or Employee FICA Worksheet for employees) should beattached to the amended return. If you owe self-employment or FICA tax for 1995 you should include acompleted Schedule SE or Employee FICA Worksheet with your original return. Please mail completed formsand any written correspondence to the IRS in the enclosed self-addressed envelope.

If you have any questions concerning this information or if you need help in any way please call the toll freenumber listed above. You may also contact your local Internal Revenue Service office in person.

Sincerely,

(signed) Nancy L. JonesNancy L. JonesChief, Examination BranchKansas City Service Center

Enclosures:Return envelopeCopy of this letterEmployee ProceduresEmployee FICA WorksheetForm SS-8Form 1040XSchedule SE and instructions

Kansas City, MO 64999 Mail Stop: 4200

Person to Contact: CP:RAD:PIU

Telephone Number: District Office Toll Free Number: 1-800-736-5271Date:

Taxpayer Identification Number:

Dear Taxpayer:

We have analyzed the information you provided on your individual income tax return for 1993. You received andreported income which could be subject to self-employment tax or your share of FICA taxes not withheld fromyour wages. For example, Schedule C or F net profit or certain Form 1099 income may be subject to self-employment tax if the total is $434.00 or greater.

Enclosed is an excerpt from Publication 533, “Self-Employment Tax”, to assist you in deciding whether you areliable for the self-employment tax.

If you determine that you owe self-employment tax for 1993 or 1994 please complete a separate Form 1040X,Amended US Individual Tax Return, for each year. Forms for 1993 are enclosed. If you owe self-employmenttax for 1995 you should include a completed Schedule SE with your original return. Please mail completedforms and any written correspondence to the IRS in the enclosed self-addressed envelope.

If you have any questions concerning this information or if you need help in any way please call the toll freenumber listed above. You may also contact your local Internal Revenue Service office in person.

Sincerely,

(signed) Nancy L. JonesNancy L. JonesChief, Examination BranchKansas City Service Center

Enclosures:Return envelopeCopy of this letterExcerpt from Publication 533Form 1040XSchedule SE and instructions

Duplicate Use of Dependent and Qualifying Children

Social Security Numbers

By Ivette Alamo-Tirado andRobert Holmes

Building upon prior research, including Internal Audit findings of an increasing trend in the use ofduplicate Social Security Numbers (SSNs), the North Florida District Office of Research and Analysis(NFL-DORA) profiled this population and tested alternative treatments to improve compliance. Ouranalysis of the tax year (TY) 1995 duplicate SSN population revealed it is composed largely of situationsinvolving two returns claiming one duplicated SSN. In particular, around 3.2 million individual returnscontained a duplicate SSN in TY 1995. This reflected around 1.8 million duplicated SSNs, of which 98.6percent were claimed on only two returns. The most frequent duplication (in 46.3 percent of the 3.2million returns) occurred where one return claimed a dependent and that dependent also filed a returnand claimed a self-exemption. To address the problem of duplicate SSN usage, we tested two majoralternative treatments designed to improve compliance, one involving a service center correspondenceexamination and the other involving a "soft" (i.e., educational) notice. Both treatments proved successfulin improving subsequent compliance, reducing instances of duplicate SSN usage in the subsequent filingyear by over 15 percentage points in the tests conducted. However, the finding of a relatively low year-to-year "repeater" rate among the population served to dampen some of the potential effectiveness of thetreatments. Still, the research revealed insights into new approaches to better target this noncompliantmarket segment, and has lead to several operational program initiatives. This project was adopted asone of the National Strategies within the multi-year Operations Plan for the Chief Operations Officer areaand is designed to improve compliance through innovative approaches.

Introduction

Only one return legally can claim an exemptionor the Earned Income Tax Credit (EITC) for aparticular person. Duplicate claims for the sameperson result in substantial revenue loss to thegovernment. By mid 1996, the IRS processed3.2 million tax year (TY) 1995 U.S. IndividualIncome Tax Returns that contained a duplicateclaim for a valid Social Security Number (SSN)as a dependent, qualifying child for EITC, and/ora dependent filing a return and claiminghim/herself.

The IRS checks the validity of an SSN duringreturn processing by confirming the SocialSecurity Administration issued one to thetaxpayer, dependent, or qualifying child.Presumably, half the population of identifiedduplicate SSN users is entitled to claim theexemption and/or qualifying child. With minorexception, however, data do not exist internallyto the IRS to determine who is entitled and whois not. Also, the IRS cannot afford to spendsignificant enforcement resources to bring thisentire population into compliance.

As a result, the North Florida District OfficeResearch and Analysis (NFL – DORA)undertook this research project to profile themarket segment of returns involving duplicateuse of SSNs. We (NFL – DORA staff) wanted tolearn more about this population, and measurethe effectiveness and efficiency of “soft notices”(i.e., educational-type letters) versuscorrespondence examinations (i.e., audits) inimproving future compliance and motivatingtaxpayers to file amended returns. This articlepresents the research results relating to the TY1995 valid duplicate SSN population of 3.2million returns.

Profile Highlights

Types of SSN Duplication

We identified the duplicate SSN populationbased on three duplicate SSN conditions:

1. duplicate EITC - two or more returnsclaiming the same qualifying child

2. duplicate dependent - two or more returnsclaiming the same dependent

DUPLICATE USE OF DEPENDENT AND QUALIFYING CHILDREN SOCIAL SECURITY NUMBERS

3. dependent/primary - a return claiming adependent and that dependent filing a returnand claiming a self-exemption.

Many returns in the population involvedcombinations of the three conditions, as thefollowing table illustrates.

Table 1. Number and Percentage of TY 1995 Returns by Type of Duplicate Condition

DUPLICATE TYPE NUMBER %

EITC Only 50,541 1.6%

EITC & Dependent 434,450 13.5%

Dependent Only 1,170,025 36.4%

Dependent &Dependent/Primary

55,878 1.7%

Dependent/PrimaryOnly

1,488,953 46.3%

Other 18,666 0.5%

Total 3,218,513 100%

Only 1.6 percent of the returns in the populationhad just a duplicate EITC condition; however,13.5 percent contained both a duplicate EITCand duplicate dependent condition. Thissituation could have involved the same SSNclaimed for both EITC and dependent, or couldhave involved two or more different SSNs.Duplicate dependents accounted for over 50percent of the population when consideringmultiple conditions (i.e., duplicate dependentand EITC, duplicate dependent and dependentfiling as primary). A dependent on one return,filing their own return and claiming the self-exemption, represents a significant portion (46.3percent) of the population.

Number of Returns Involved

There are 1,824,108 duplicated SSNs in the TY1995 population. Of these 98.6 percent wereclaimed on only two returns. Chart 1 illustratesthese data.

Chart 1. Percent of Returns by Number of Returns Per Duplicated SSN

0

10

20

30

40

50

60

70

80

90

100

2 3 >3Number of Returns Per Duplicated SSN

P e r c e n t

98.6

1.3% 0.1%

Of the 3,218,513 returns in the population, 87.8percent involves only one duplicated SSN and10.4 percent involves two. Only 1.8 percentinvolves more than two duplicated SSNs. Chart2 illustrates these findings.

Chart 2. Percentage of Returns by Number of Duplicated SSNs Per Return

1 Duplicated SSN 2 Duplicated SSNs

>2 Duplicated SSNs

10.4%

1.8%

87.8%


Considering 98.6 percent of the duplicatedSSNs was claimed on two returns and 87.8percent of the returns involves only oneduplicated SSN, clearly the population iscomposed largely of two returns claiming oneduplicated SSN.

Repeater Rate

The primary SSN on a tax return reflects theperson filing (or the person for whom the returnis being filed). We defined a repeater as aprimary SSN in the TY 1995 duplicate SSN filethat also was a primary SSN in the TY 1994duplicate SSN file. The following chart reflectsthe percentage of repeaters by group:

Chart 3. Percentage of Repeaters in Population by Group

0%

10%

20%

30%

40%

50%

60%

The repeater rate for all returns in the populationis only 34.1 percent. Conversely, 65.9 percentof the taxpayers in the TY 1995 file was not inthe TY 1994 population. Returns with aduplicate dependent condition had the highestrepeater rate at 44.5 percent. The repeater ratefor returns with a duplicate EITC condition had arepeater rate of 38.5 percent. The dependentfiling as primary condition had the lowestrepeater rate at only 25.1 percent.

These results indicate the duplicate SSNpopulation turns over rapidly. Hence, this limitsthe effectiveness of any treatment involving one-to-one contact. For all contacts, notice orexamination, only the portion made withtaxpayers that would repeat a duplicate SSNclaim can affect future compliance.

Treatment Tests

We conducted three distinct taxpayer treatmenttests. The first two involved two different “soft”(educational) notices sent to different samples ofthe population. Both notices informed thetaxpayer of the duplicate SSN condition, pointedthem to readily available information sourcesthat explained the proper circumstances forclaiming the personal exemption, dependents,etc., and advised them not to claim theduplicated SSN if not entitled. However, one ofthe notices also requested the filing of anamended return for TY 1995 if the taxpayer wasnot entitled to claim the duplicated SSN. Theother notice made no mention of filing anamended return.

The third test treatment involved sending astandard IRS Examination initial contact letter tosamples of the population. This initial contactletter advised taxpayers of the examination oftheir TY 1995 tax return and requestedtaxpayers verify with the IRS their dependencyand earned income credit claims.

We mailed the soft notices and Examinationcontact letters in late November and earlyDecember 1996, before taxpayers filed their TY1996 returns. The primary purpose of the testtreatments was to determine the relativeeffectiveness of the various contacts inimproving compliance on TY 1996 returns.Another objective was to measure theeffectiveness of the particular soft notice inmotivating taxpayers to file amended returns forTY 1995.

Methodology

Linking Related Returns

To enable selection of the test and controlgroups and evaluation of the test results, welinked taxpayers that claimed the sameduplicated SSN or SSNs to one another to forma “case.” We then assigned each case a controlnumber. Of the 3,218,513 returns in theduplicate SSN population, we successfullylinked 3,177,574, or 98.7 percent, to create1,530,230 cases.

Selection of Test and Control GroupsThe duplicate SSN population of returnssuccessfully linked then was segmented into sixgroups for sample selection. First, we split the


population into three groups based on theduplicate condition(s) – i.e., duplicate EITC,duplicate dependents, and dependent/primary.Since many cases involved multiple duplicateconditions, we used the following methodologyin assigning cases to a group:

• any case involving a duplicate EITCreturn was assigned to the EITCgroup,

• any remaining case with adependent/primary condition wasassigned to the dependent/primarygroup, and

• any case with only a duplicatedependent condition was assigned tothe duplicate dependent group.

We then segmented the three groups based onwhether the case was a repeater (i.e., alsopresent in the TY 1994 duplicate SSNpopulation) or non-repeater, resulting in sixmarket segments from which the test and controlgroup samples were selected. We selected arandom sample of approximately 550 cases foreach of the test and control groups. Casesselected for the control groups received no IRScontact (relative to this research).

Conducting the Tests and Measurement ofResults

As previously stated, in November/December1996 we mailed two different soft notices tosamples of the six segments of the population.We placed our return DORA address on theenvelope so we could measure and respond toall correspondence received. In addition, aunique toll-free telephone number wasestablished to route taxpayer inquiries to oneCustomer Service group for additional controland cost/benefit analysis purposes.

The third treatment test involved examinations ofreturns. However, this test included only thethree segments of the repeater population.Examinations were not initiated on non-repeaters. In November 1996, we furnishedidentifying information of the taxpayers in threerepeater segments to a Service CenterCorrespondence Examination Branch; and thenin early December 1996 mailed all contactletters.

Test Results

We measured the compliance improvementresulting from the test treatments by searchingthe TY 1996 valid duplicate SSN population forthe presence of the taxpayers in the 1995 testand control groups. Taxpayer presence in theTY 1996 valid duplicate SSN population requiredat least one other taxpayer to file a return andduplicate the SSN claim (and satisfy one of thethree duplicate conditions previously described).We measured the soft notice effect on amendedreturn filing by using IRS Master File data fromthe TY 1995 module of taxpayers in the test andcontrol groups.

Effectiveness of Notices and Examinations inImproving Compliance on TY 1996 Returns

Chart 4 shows results of the repeater controlgroup, soft notice, and Examination contactletter. Since the two soft notices were equallyeffective in improving compliance on TY 1996returns, only data for the notice that requestedthe filing of an amended return are presentedbelow. Again, the Examination initial contactletters were mailed to samples of the repeaterpopulation only.

Chart 4. Percentage of TY 1995 Repeaters Not Repeating a Duplicate SSN Claim in TY 1996 by Group

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

EITC Dependent Dep./Prim. All

Control Group Exam Group Notice Group


By definition, the control groups were notsubjected to the treatments. Therefore, thedifferences between the test and control groupfor each the soft notice and Examination contactletter represent the treatments’ effect inimproving compliance for TY 1996. Comparing the notice group with the controlgroup, it is clear the educational notice wassuccessful in reducing duplicate SSN claims inthe subsequent year for the repeater segment.In particular, the notice improved compliance27.9 percentage points (76.1 percent - 48.2percent) for the EITC segment. It increasedcompliance 36.9 percentage points (80.3percent - 43.4 percent) for the dependent-onlysegment, and 29.5 percentage points (88.3percent - 58.8 percent) for thedependent/primary segment. Complianceincreased 32.7 percentage points (82.3 percent -49.6 percent) for repeater taxpayers, overall. Considering the effects of the Examinationcontact letter, we again see a distinct reductionin the instances of duplicate SSNs, compared tothe control group. The Examination contactletter increased compliance 23.1 percentagepoints (71.3 percent - 48.2 percent) for the EITCgroup, 27.1 percentage points (70.5 percent -43.4 percent) for the dependent group, 27.9percentage points (86.7 percent - 58.8 percent)for the dependent/primary group, and 26.7percentage points (76.3 percent - 49.6 percent)for taxpayers overall. Interestingly, the percent of taxpayers notrepeating a duplicate SSN claim for the noticegroups was higher for all groups compared tothe Examination contact letter. However, thedifference was “statistically significant” only forthe dependent group. Among the non-repeater population, the noticealso proved effective in reducing duplicate SSNclaims compared to the control group. Asshown in Chart 5, the notice improved futurecompliance (i.e., increased the instances of norepeat duplicate SSN claims) by 16.7percentage points for the EITC group, 12.0percentage points for the dependent group, 5.2percentage points for the dependent/primarygroup, and 8.9 percentage points for non-repeattaxpayers overall.

Chart 5. Percentage of TY 1995 Non-Repeaters Not Repeating a Duplicate SSN Claim in TY 1996 by Group

0%10%20%30%40%50%60%70%80%90%

100%

EITC Dep. Dep./Prim. All

Control Group Notice Group

Chart 6 presents the consolidated results for allthe notice and control group taxpayers in thethree duplicate SSN categories. This chartweights and combines the results for repeatersand non-repeaters. Chart 6. Percentage of All TY 1995 Taxpayers Not Repeating a

Duplicate SSN Claim in TY 1996 by Group

0%10%20%30%40%50%60%70%80%90%

100%

EITC Dep. Dep./Prim. All


Considering the entire population of repeaters and non-repeaters, thenotice improved compliance by 19.7 percentagepoints for EITC taxpayers, 19.8 percentage pointsfor dependent taxpayers, 9.2 percentage pointsfor the dependent/primary taxpayers, and 15.7percentage points for taxpayers overall. For all notice groups, 13.1 percent repeated aduplicate SSN in TY 1996. This portion mayrequire enforcement action to treat. Effectiveness of Notices in Motivating Taxpayersto File Amended Returns for TY 1995

Since this project focuses on the duplicate use ofan SSN, there is an inherent presumption that oneof the taxpayers claiming the duplicated SSN isentitled to the claim, and one or more taxpayersare not. Since over 98 percent of the duplicatedSSNs involve two taxpayers, we assumeapproximately half of the population is entitled toclaim the duplicated SSN(s). Therefore, if all thetaxpayers not entitled to claim the duplicatedSSN(s) filed amended returns, the maximumamended return filings would be approximately 50percent.

Chart 7 presents the percentage of taxpayersfiling amended returns with a balance due orrefund claimed. Amended returns with no changein liability were excluded due to the likelihoodtaxpayers filed these to notify the IRS of an SSNerror. The notice did not discourage the filing ofan amended return to correct SSN errors, and wereceived several amended returns where thetaxpayer indicated they mistakenly used thewrong SSN (generally that of another child or ex-spouse).

The notice did precipitate filing of amendedreturns within the notice groups. Compared to thecontrol group, instances of filing an amendedreturn for the notice group was 0.9 percentagepoints higher for EITC, 3.0 percentage pointshigher for dependent, 3.7 percentage pointshigher for dependent/primary, and 3.0 percentagepoints higher for taxpayers overall. Still, theresults indicate the notice is more likelyassociated with future compliance than the filingof amended returns. This primarily is due to thetiming of the subsequent year extract. We knowfrom information copies of amended returnsmailed to NFL- DORA that not quite all amendedreturns had been processed as of the master fileextract date.

Chart 7. Percent of Amended Returns Filed by Group

0.0%

0.5%

1.0%

1.5%

2.0%

2.5%

3.0%

3.5%

4.0%

4.5%

EITC Dependent Dep./Primary All


Conclusions and Actions Taken

Conclusions

Composition of Population

The research results demonstrate the roughly3.22 million duplicate SSN population iscomposed largely of two groups. One group (1.17million) consists of two parents filing separately,each claiming one or more of their children as adependent and/or qualifying child. The other(1.49 million) consists of one parent (or parents inthe case of a joint return) claiming a dependentchild and that dependent filing a return that claimsa personal exemption for him-/herself. With minorexception, data do not exist in IRS databases todetermine who is entitled to the SSN claim andwho is not.

This population also turns over rapidly. Ourresearch indicates approximately two-thirds of theTY 1995 population was not in the TY 1994population. This limits the effectiveness of anytreatment involving one-to-one contact. Suchcontact with the duplicate SSN users wouldimpact only the remaining one-third of theidentified population that would repeat a duplicate


SSN claim. In addition, since so many taxpayersenter the population every year, absentdevelopment of a preventative treatment (such asa tax law, tax form, or schedule change), a yearlyapplication of a treatment would be necessary toreduce the size of the duplicate SSN populationfrom year to year.

Effectiveness of Notices and Examinations

Despite this inherent limitation (arising from thelow repeater rate), the soft notices werenevertheless effective at improving compliance.For example, the notices reduced the instances ofsubsequent duplicate SSN claims by 15.7percentage points for taxpayers overall, comparedto the control group. Also, there was an indicationthe notices were slightly more effective on thisdimension than correspondence examination. Inaddition, educational notices are less threateningand burdensome to taxpayers than examinationsand such examinations cost more than treatmentby notices.

Nevertheless, Examination contact letters alsoimproved compliance by reducing subsequentduplicate SSN claims among the repeatersub-population tested. In addition, not only do theexaminations result in improved future compliance(prevention), they also result in recovery of lostrevenue. The revenue potential likely is greater inexaminations because the number of taxpayersdisclosing deficiencies on previously filed returnsis greater than the apparent small number ofvoluntarily amended returns likely to arise from thenotice treatment.

Operational Actions Taken

The test results described above, as well asearlier research findings in the duplicate SSNarea, gave rise to a number of promising programadaptations (initiatives). These initiatives weretargeted to specific segments of the duplicateSSN population and incorporated into the NationalStrategies within the Operations Plan -- thecomprehensive, multi-year planning documentdesigned to improve compliance throughinnovative ideas for the IRS functions under theChief Operations Officer. Those initiatives

included the "Duplicate TIN Repeater Project,”which involved correspondence examinations forsome 145,000 TY 1995-1996 duplicate SSNrepeaters claiming the earned income tax credit,and a nationwide soft notice mailing to another 2.3million taxpayers claiming a duplicate SSN for TY1996. North Florida DORA staff currently aremeasuring the effectiveness of those operationalprogram initiatives.


Ivette Alamo-Tirado began her career as anoperations research analyst with the IRS in 1991,and is presently an ORA with the North FloridaDistrict Office of Research and Analysis. Sheholds a B.S. degree in industrial engineering fromthe University of Puerto Rico, an M.S. degree inengineering administration and systems analysisfrom George Washington University, and anM.B.A. degree from Turabo University, PuertoRico.

Robert Holmes is a program analyst TeamLeader in the North Florida District Office ofResearch and Analysis. He began his career withthe Internal Revenue Service in 1971 upongraduating from Jacksonville University with aB.S. degree in accounting. He is also a certifiedpublic accountant.

High-Range Corporation ReturnWorkload Selection System Development

By James A. Wilhelm

Prior to this research, IRS’s Examination function did not have a mathematical model to help it prioritize itshigh-range corporate workload. Examination defines high-range workload as audit work associated withcorporations and partnerships with over $10 million in total assets not designated as part of the CoordinatedExam Program (CEP) (i.e., its large case program). This article discusses the development of an operationalselection model for the high-range corporate income tax return (Form 1120) workload. When theDiscriminant Analysis System (DAS) scoring model becomes fully functional for these types of returns,Examination has the potential to assess almost $1.2 billion in additional tax revenue per fiscal year.Additionally, the DAS model will help Examination reduce taxpayer burden by reducing by half the number ofsuch corporations audited. The ability of this scoring system to identify more productive cases forexamination and reduce taxpayer burden led to its inclusion as one of IRS’s ten National Strategies designedto increase compliance through innovative approaches. With initial implementation in fiscal year 1998,Examination used the DAS scoring model to identify 25 percent of its high-range Form 1120 workload.

Background

After a 1995 General Accounting Office (GAO)audit, IRS Examination officials stated they“want better systems for selecting andclassifying returns.”15 They were referringspecifically to the high-range corporate andpartnership returns. These high-range returnsinclude those for corporations and partnershipswith $10 million and over total assets notdesignated as part of the large case program, orCEP. Approximately 53,000 such returns arefiled each year.

Currently, Examination does not have amathematical selection system for the high-range returns. As a result, selection of returnsfor examination in this category is not veryeffective. For fiscal years (FYs) 1993, 1994, and1995, 47.7 percent of all closed high-rangeexaminations resulted in no additionalassessments from the audits (i.e., no tax liabilityincrease over what the taxpayers reported ontheir returns). This suggests current auditselection is not much better than randomselection. Examination needs more effectiveselection systems for the high-range returns toensure it examines the most non-compliantreturns and improves yield in this area.

15 Tax Administration: Audit Trends and Taxes Assessed on LargeCorporations (GAO/GGD-96-6, October, 1995, p. 9)

Objective

The objective of this project was to develop anational scoring model for all non-CEP high-range corporate returns. This model wouldprioritize returns for the IRS Examinationfunction based on their probability of beingprofitable to audit (as defined later in this article).That is, the profitable to audit (PTA) cases willhave a high rank and the non-PTA cases willhave a low rank.

For this effort, we decided to look at variousalternative statistical modeling techniques,including discriminant function analysis andregression. With any of these techniques, ourgoal was to predict an outcome (i.e., PTA or notPTA) based on various predictor variables -- i.e.,tax return line items. Therefore, each of thesetechniques required a database of returns thatcontained both prior audit results and tax returndata. The model discussed in this report wasdeveloped to operate within the current IRSreturn submission, processing, and examinationsystems.

Data Used

The data we used for model development camefrom the following three sources.

1) Midwest Automated ComplianceSystem (MACS) MACScontains transcribed tax returndata for all high-range

HIGH-RANGE CORPORATION RETURN WORKLOAD SELECTION SYSTEM DEVELOPMENT


corporations. We usedprocessing year (PY) 1992,1993, and 1994 line item taxreturn data. (“Processing year”refers to the year in which thereturns were filed.)

2) Office of Foreign BusinessStudy (OFBS) This database isproduced by the Statistics ofIncome (SOI) Division at theIRS. It contains line iteminformation on corporationreturns with over $10 million intotal assets16. We used PY1989, 1990, 1991, 1992, and1993 data to create a long termportrait of the businesses understudy.

3) Examination Closed-CaseDatabase This databasecontains various data for allcompleted examinations for agiven fiscal year. A componentof the database is audit-relatedinformation for these returns.We used FY 1993, 1994, and1995 closed-case data.

We used these three databases to create ourdevelopment database. We first merged theMACS and SOI data with the Closed-Case datain order to match tax return line item data withtheir corresponding Examination results. Wethen eliminated all returns without correspondingExamination results. We also eliminated allreturns designated as CEP or duplicate returns(i.e., returns filed for the same tax period). Wehad 16,415 returns remaining in our databaseafter this merging and elimination.

The 16,415 returns represented 51.5 percent ofthe 31,855 high-range returns examined andcompleted by Examination in FYs 1993, 1994,and 1995. They were filed on or after PY 1989.The remaining 15,440 returns were filed beforePY 1989 (e.g., some returns were filed in 1968).We could not match any of these returns since

16 The SOI database contained information on 30 percent of AC219 returns and all AC 221 through AC 225 returns. ActivityCode (AC) is Examination’s way of categorizing its workload. Itis based on size of total assets. AC 219 includes returns withassets of $10,000,000 < $50,000,000. AC 221 includes returnswith assets of $50,000,000 < $100,000,000. AC 223 includesreturns with assets of $100,000,000 < $250,000,000. AC 225includes returns with assets of $250,000,000 or more.

we no longer have the tax return line item datafor these pre-PY 1989 returns. Since weneeded line item information for this project, wecould not use the information contained in thesepre-PY 1989 cases. The characteristics of thepre-PY 1989 returns may or may not be differentfrom those of the post-PY 1989 returns. Forpurposes of model development, we assumedthe differences were minimal.

Methodology

The methodology for developing workloadselection systems is straightforward. A keyassumption was our database consisted of arandom sample of high-range corporatereturns.17 We then developed a selection modelusing both a logistic regression approach and adiscriminant analysis approach. The followingdiscussion provides more details.

Determining “PTA” Criteria

Before we started model development, weneeded to determine the criteria for a “profitableto audit”18 return. A definition was essential foreach of the statistical techniques we used formodel development. In the high-range area,Examination has defined the PTA criteria as“$2,000 of audit results per hour” (i.e., a $2,000increase to a return’s tax liability resulting fromevery hour expended on an audit). Using thisdefinition, slightly over 10 percent of the returns(1,722 returns) in our project database werePTA returns.

Development Versus Test Files

Once we determined the PTA criteria, we splitour database into a formula development fileand a test file. This “splitting” was done so wecould evaluate, or measure, the predictedeffectiveness of the developed models on an 17 Ideally, the examination results we use to develop selectionsystems should be from a random sample of the population. Thisis so we can feel confident that the results are statisticallyrepresentative of the population allowing us to develop anunbiased selection system. In reality, we do not have results froma random sample. However, Examination believes that currentreturns selection in this area is not much better, if any, than randomselection. Therefore, we will make the assumption that examresults will reflect what we would see if Examination audited arandom sample of these cases.18 For our “profitable to audit” definition, we were concerned onlyabout the change in tax (audit results). We did not consider thechange to the Net Operating Loss Deduction (NOLD) as apotential PTA definition. However, other IRS research on“Revenue Protection” uses the NOLD as a potential PTAdefinition.



independent test file of cases not used in modeldevelopment. To further enhance thisevaluation, we required our two split files tosatisfy three conditions to ensure they were assimilar as possible. The conditions were:

1) the distribution of thedevelopment file returns besimilar to the distribution of thetest file returns (e.g., the percentof returns in the developmentfile with total assets between$50 million and $100 million hadto be approximately equal to thepercent in the evaluation file);

2) the average tax change of thereturns in the development filehad to be similar to the averagetax change of the returns in thetest file; and

3) the percent of PTA cases in thedevelopment file had to besimilar to the percent of PTAcases in the test file.

With these “splitting” conditions in mind, we letSPSS (our statistical analysis software package)randomly select 70 percent of the 16,415 returnsin our database for the development file. Wewanted the development file to contain a largeportion of the project database in order toincrease the chances of identifying datarelationships between examination results andtax return line items. The remaining 30 percent(4,899 examined returns) constituted the testfile. For all model development, we used datafrom the development file. We then used themodel to score and rank the returns on the testfile to predict how well the model would performif Examination used it to select returns.

Determining Sampling Weights

Our final data-preparation task was to determinethe sample weights. Since we assumed theavailable data was a random sample of theForm 1120 population, we needed a way for oursample data to represent the high-rangecorporate population for the most current year.We noted there is a multi-year lag betweenwhen a return is filed/processed and when it isselected and examined. Data from Examinationindicated the total population consisted of53,394 Form 1120 filers in PY 1994, the most

current processing year for which there arecompleted corresponding examinations. Wewanted our “samples” to represent thisprocessing year. Hence, the weight weassigned to each return in our overall file was3.25 (i.e., 53,394 divided by 16,415). Theweight we assigned to each return in our test filewas 10.899 (i.e., 53,394/4,899).

Model Development

After we completed this data preparation, wedeveloped the first set of models using only taxreturn line item data relative to dollars. For thefirst set of models, we found the discriminantfunction models performed better than thelogistic regression models. We found thelogistic models correctly classified the non-PTAreturns. However, the discriminant functionmodels did a better job of classifying both thenon-PTA returns and the PTA returns. Since itwas more important to classify correctly the PTAreturns (from an Examination standpoint), wecontinued our efforts using the discriminantanalysis technique.

We then considered “non-dollar” return lineitems as we continued our formula development.Two examples of non-dollar line item variablesare Accounting Code, which reflects the type ofaccounting method used by the taxpayer, andPrincipal Business Activity (PBA) Code, whichreflects the taxpayer’s type of business.

Next, we developed models for sub-segments ofour population. The reason for thissegmentation was to produce more refinedformulas. Using our same 70 percentdevelopment/30 percent test approach, wedeveloped segmented models for ExaminationActivity Codes (i.e., groups by size of totalassets), major PBA categories, andcombinations of the two. For each sub-segment, we developed both discriminantfunction models and logistic models. We againfound the discriminant-based modelsoutperformed the logistic models. We alsofound that grouping the returns into certaindistinct PBA Codes was the best groupingstrategy of those considered. That is, modelsthat distinguish specific groupings by PBACodes did the best job of correctly classifyingboth the non-PTA returns and the PTA returns.



Our final step in the development process wasthe inclusion of financial ratios and indicators inthe model development. These financialvariables were either basic economic financialratios or financial indicators used by the IRS.We developed discriminant-based models, forPBA Code groupings, using all transcribed lineitem data (both dollar and non-dollar) and thegenerated financial data. These modelsoutperformed all other models (e.g., thosewithout groupings by PBA Code) in classifyingthe PTA returns as PTA returns. They also did abetter job of classifying the non-PTA returns asnon-PTA returns.

Results of Model Development

Evaluation Approach

Our model development resulted in the creationof two alternatives for scoring and rankingreturns for examination. The first alternative(Georgia DORA Selection I) was a one-formulamodel. That is, one discriminant functionformula was applied to the entire high-rangecorporation return population. The secondalternative (Georgia DORA Selection II) was atwo-formula model. This model consisted of twodiscriminant function formulas, Formula A andFormula B, where Formula A applied to returnswith one particular set of major PBA Codes andFormula B applied to all other returns. ForFormulas A and B, we considered transcribeddata, financial ratios, and financial indicators inthe model development process.

For our evaluation, we replicated as closely aspossible a typical fiscal year of examinations.Our definition of a typical year of examinationswas the expenditure of 1,382,160 “Examinationhours.” We use 1,382,160 hours as arepresentation of a typical year for two reasons.First, it represents the average number of hoursExamination spent auditing (and ultimatelyclosing) high-range Forms 1120 tax returns inFY 1993, 1994 and 1995. Second, an hourrepresentation tempers “complexity of the audit”dilemmas, “case grade” dilemmas, and “indirectstaff time” dilemmas caused by a standard“case” fiscal year representation.

We used returns from the 30 percent test file forour evaluation19. These returns were not usedfor formula development. The sole purpose ofthese “test” returns was to determine how wellour formulas identify a “profitable to audit”return. This type of evaluation is commonplacefor statistical model development.

Our evaluation used four different approachesfor selecting returns to examine.

1) Current Examination SelectionFor this approach, we averagedthe results for three fiscal years(1993, 1994, and 1995) ofexaminations. These averageswill be the best estimate of howExamination currently conductsits high-range corporateprogram.

2) Georgia DORA Selection IIn this approach, we applied ourbest single-formula model to all4,899 returns on the test file.We sorted and ranked thereturns in descending scoreorder. We then selectedreturns from our ranked listinguntil we accumulated, at most,1,382,160 weighted hours.When we expended theseExamination hours, we recordedthe predicted audit results. Webelieve these results representwhat we could expect in atypical fiscal year inExamination, if Examinationused the one-formula model.

3) Georgia DORA Selection IIFor this approach, we appliedboth Formulas A and B to theirrespective “test file” returns. Wesorted and ranked both sets ofreturns in descending scoreorder. For this set of formulas,we needed a relative predictivemeasure for combining the twoscored groups into onesequenced listing. To helpdetermine this sequence, we

19 To be as conservative as possible, we eliminated one “outlier”case from the test file. This case took 2,560 hours and yieldedover $105 million in tax change.



decided to use a probabilitySPSS assigns to each scoredcase. This probabilityrepresents the likelihood of thereturn being profitable to audit.The higher the probability, themore likely the return is PTA.Thus, we created a singleranking of all the cases from thehighest to lowest under theSelection II approach using theSPSS probability assigned toeach return.

We then selected theappropriate returns from ourranked listing under Selection IIuntil the total hours weaccumulated equaled 1,382,160weighted hours. When weexpended these examinationhours, we recorded the results.These results are our bestrepresentation of a typical fiscalyear in Examination, ifExamination used the twoformulas to select returns.

4) Perfect SelectionThis approach represents the“ideal” scenario. To determineperfect selection, we sorted andranked the 4,899 returns in thetest file in descending auditresult order. That is, the largesttax change (per hour) return isfirst, the second largest taxchange return is second, etc.We then went down our rankedlisting until we accumulated1,382,160 weighted hours.When we expended these examhours, we recorded the results.

To better understand the ranking abilities of thevarious selection approaches, we evaluatedthem using five different measures.

1. Total Audit Results2. Dollars per Hour3. Profitable to Audit Rate4. No Tax Increase Rate5. Number of Cases Worked

Total Audit Assessments Can Be Increased byNearly $1.2 Billion

One measure of the success of the Examinationfunction is its ability to assess accuratelyadditional tax after audit. As seen from Figure 1,both the Georgia DORA Selection models canincrease total Examination assessments. Weestimate that Examination could generate anadditional $1.197 billion (i.e., $2,474,289,051 -$1,277,115,840) in assessments per year if theGeorgia DORA models were used to selectreturns for examination20. Based on ourmatched files, this is a 94 percent improvementover the historic Examination selectionprocedure. However, there is still room forimprovement. The Georgia DORA models onlyidentify half the total optimal tax change underperfect selection.

Figure 1. Comparison of the GeorgiaDORA Selection Models withCurrent ExaminationSelection and PerfectSelection for Total DollarsAssessed (Based onProjected/Actual Total AuditResults)

$1,277,115,840

$2,474,289,051

$2,417,453,622

$5,457,206,771

$0

$1,000,000,000

$2,000,000,000

$3,000,000,000

$4,000,000,000

$5,000,000,000

$6,000,000,000

CurrentExam

Selection

GeorgiaDORA

Selection I

GeorgiaDORA

Selection II

PerfectSelection

20 Total improvement will not be realized until the fiscal year inwhich all discretionary returns closed were selected forExamination using the models. Thus, improved results willaccumulate over time.



Dollars Per Hour Can Be Improved by 97Percent

Another measure of interest is the amount ofassessments an examiner can produce in anhour. This commonly is referred to as “dollarsper hour.” The average dollar per hour rate forhigh-range corporations based on our matchedfiles is $924. By using the Georgia DORASelection I model, Examination could realizeover $1,800 per hour (see Figure 2). This is a97 percent improvement in dollars per hour.However, as we saw in the total audit resultssection, there is room for even moreimprovement.

Figure 2. Comparison of the GeorgiaDORA Selection Models withCurrent ExaminationSelection and PerfectSelection for AssessedDollars Per Hour (Based onProjected/Actual Dollars PerHour)

$924

$1,827

$1,769

$4,031

$0

$500

$1,000

$1,500

$2,000

$2,500

$3,000

$3,500

$4,000

$4,500

Current ExamSelection

Georgia DORASelection I

Georgia DORASelection II

Perfect Selection

Profitable To Audit Rate Can Be Improved by100 Percent

Another goal for Examination is to work onlyprofitable to audit returns. From Figure 3 wesee the historic PTA rate for Examination is10.31 percent. That is, almost 11 returns out of100 audited have a tax change in excess of$2,000 per hour. We also note the PTA rate forboth of the Georgia DORA models is around 20percent. Thus, the use of a Georgia DORASelection model would improve the PTA rate byabout 100 percent.

Figure 3. Comparison of the GeorgiaDORA Selection Models withCurrent ExaminationSelection and PerfectSelection for PTA Percent(Based on Projected/ActualProfitable to Audit Rate (inpercent))

10.31%

21.98% 19.92%

64.92%

0%

10%

20%

30%

40%

50%

60%

70%




PerfectSelection



No-Tax-Increase Rate Can Be Decreased by 30Percent

Another measure to Examination is the no-tax-increase (NTI) rate. Examination does not wantto devote resources on audits that do notgenerate revenue. This wastes the taxpayer’stime and wastes Examination resources. Wewanted our models to select fewer NTI returnsfor examination. It appears the Georgia DORAmodels will reduce the number of theseerroneously selected returns in the auditinventory. Depending on which model is used,we see from Figure 4 the Georgia DORA modelscan reduce the NTI rate by anywhere from 26 to30 percent.

Figure 4. Comparison of the GeorgiaDORA Selection Models withCurrent ExaminationSelection and PerfectSelection for the No TaxIncrease Percent(Based on Projected/ActualNo Tax Increase Rate (inpercent))

43.34%

30.21% 31.94%

0.00%

0%

5%

10%

15%

20%

25%

30%

35%

40%

45%




Perfect Selection

Reduced Number of Cases Worked/TaxpayerBurden

Another measure important to Examination isthe allocation of cases to its staff and theassociated burden on taxpayers. From Figure 5we see Examination historically closes almost

12,000 high-range corporate audits each fiscalyear. By using a Georgia DORA model,Examination could expect to audit up to 52percent fewer returns, since better returns willbe selected for audit. This reduction wouldimpact work assignments both between andwithin IRS districts.

The reduced number of cases also reducestaxpayer burden. Using the Georgia DORAmodels, Examination should reduce overalltaxpayer burden in this Form 1120 group byone-half. Also, since the Georgia DORA modelswill identify fewer no-tax-increase returns,Examination further could reduce the number oftaxpayers experiencing an audit that results inno substantial change to their tax liability. Thisreduction will go from 5,148 audited no-tax-increase returns (i.e., 11,878 x .4334) to aprojected 1,722 (i.e., 5,700 x .3021). Therefore,by using the Georgia DORA models,Examination could reduce taxpayer burden onthis dimension by examining 70 percent fewerno-tax-increase returns.

Figure 5. Comparison of the Georgia DORASelection Models with CurrentExamination Selection and PerfectSelection for Number of CasesWorked (Based on Projected/Actual

Number of Cases Worked for a Fiscal Year)

11,878

5,700 5,798 5,253

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000




Perfect Selection

Num

ber o

f Cas

es W

orke

d

Conclusions and Operational Implementation

This research showed Examination could use theGeorgia DORA selection models to identify betterworkload in the high-range Form 1120 corporatearea. The new selection models promised notonly more productive use of Examination staff, interms of assessments, but also significantreductions in taxpayer burden. As a result of thispromising innovation, the high-range Form 1120scoring systems were adopted as one of theNational Strategies. The National Strategies are akey component of the Operations Plan, the coremulti-year planning document for the IRS’s ChiefOperations Officer.

In FY 1998, Examination began initialimplementation of the high-range Form 1120Discriminant Analysis System (DAS) model in alimited test environment. Examination examinedapproximately 1,200 high DAS scored cases.This represents 25 percent of Examination’s FY1998 high-range Form 1120 workload. Most ofthese large corporate examinations take severalyears to close. Therefore, the results of the testwill not be available for some time.


James A. Wilhelm is an operations researchanalyst in the Georgia District Office of Researchand Analysis. He received his M. S. in statisticsfrom West Virginia University. He has been withthe IRS since 1988.

Predicting Estate Tax Filings and Taxable Gifts

By Jonathan Feinstein andChih-Chin Ho

We develop a strategy for predicting filing for estate tax returns by integrating individual wealth and mortalitystatistics into a model framework, based on two non-IRS survey datasets: the Health and Retirement Study(HRS) and the Assets and Health Dynamics Among the Oldest Old (AHEAD). Both datasets provideextensive economic, financial, demographic, and health information about older Americans.

We use the HRS and AHEAD datasets for estimation of household assets subject to estate taxation anddevelop a prediction of estate tax filings among decedents who exceeded 50 years of age in 1992. Acomparison of our predictions with the IRS’s Statistics of Income (SOI) Division estate tax filing statisticsprovides an estimate of estate tax nonfilers. Our findings suggest very few nonfilings of estate tax returns fordecedents age 80 and above, for which the bulk of estate taxes are paid.

Since the lifetime giving of gifts is considered a preferred tool for estate planning among wealthy individuals,we also develop a strategy for estimating the incidence and magnitude of noncompliance with the gift tax.We tabulate responses to questions about gifts in HRS and AHEAD and construct preliminary estimates oftaxable gifts for 1992. A comparison of our results with IRS gift tax reporting statistics provides an estimate ofunreported gift giving above the threshold for paying tax. Our findings indicate more substantialunderreporting of gift tax liability. However, our estimation methodologies for both the estate and gift taxareas reflect new approaches with some known weaknesses. As a result, the tax gap estimate in this articleshould be viewed as tentative, with future research planned.

Data Sources

Both the Health and Retirement Study (HRS)and the Asset and Health Dynamics Among theOldest Old (AHEAD) are panel datasets fundedby the National Institute of Aging of theDepartment of Health and Human Services andadministered by the Institute of Social Researchof the University of Michigan. They provideextensive economic, financial, demographic, andhealth information about older-age population ofthe United States.

The HRS began in 1992. Beginning in that year,and every second year for 12 years thereafter,each household in the HRS has beeninterviewed and the household’s respondentshave been asked numerous questions abouthousehold income and assets, employment andretirement status of individual householdmembers. To be selected for participation, thehead of a household must have been between51 and 61 as of 1992.

AHEAD is a longitudinal study of U.S. populationcohorts born prior to January 1, 1924. FromOctober 1993 to April 1994, AHEAD Wave 1interviews were conducted with national

samples of these age 70-plus individuals andtheir spouses about major transitions in theirhealth and financial situations. The longitudinalstudy plan specifies a full-scale re-interview ofthe AHEAD panel every second year beginningin 1995.

We extracted information on age, race, sex,marital status, life insurance, and assets fromthe HRS and AHEAD datasets. We thenperformed numerous data manipulations inorder to impute missing data.

There is considerable overlap between theinformation recorded in HRS and thatin AHEAD. Both datasets contain informationabout household composition, householdincome and assets, including pension assets,employment and retirement status ofindividual household members, health care useand costs, health insurance and lifeinsurance of individual household members,housing, and the household's economicrelationship with other non-household familymembers.

However, despite the many similarities, thereare some differences between thedatasets in the information they gather. Oneimportant difference for work on estate

PREDICTING ESTATE TAX FILINGS AND TAXABLE GIFTS


48

taxation is that AHEAD asks about trust assets,while HRS does not. We discoveredthat trust assets are a significant proportion oftotal assets for many households in thetop 10% of the wealth distribution.

We computed several different measures of totalhousehold assets, differing in how weinterpreted the trust asset variables. We reportour results for the case that is biased in favor ofthe largest amount of trust assets, the case thatwe believe is associated with the most plausibleinterpretation of the respondents' answers to theseries of questions they were asked about trustassets.

Predicting Estate Tax Filings

An estate tax return, Form 706, must be filedwith the IRS for any decedent whose estatesatisfies certain conditions. For tax year 1992,the most important condition is that the grossvalue of the estate exceed $600,00; however,starting in 1998, recent tax legislation increasesthe threshold in steps to $1 million by 2006.

For all decedents meeting the filingrequirements, an estate tax return must be filedwithin 9 months of the date of death, althoughan extension of six months is allowed. Althougha return must be filed for every decedent whoseestate value exceeds this threshold, tax is notowed on all such estates. Certain deductionsmay be subtracted from the gross value, and taxis owed only on the net value of the estate.

Further, if a decedent is married an unlimitedmarital bequest may be made to his (her)spouse, in which case tax is owed only on thenet value of the estate not included in thebequest in excess of $600,000. If a decedent iswidowed and his or her spouse made a maritalbequest, tax is owed only on that portion of thenet value in excess of $600 thousand.

The HRS and AHEAD datasets can be used topredict estate filings among decedents whoexceeded 50 years of age in 1992. Our strategyis to divide the post-50 population into mutuallyexclusive cells, based on age, sex, and martialstatus groups.

The five age brackets include one group takenfrom HRS (50-61), three groups from AHEAD(70-74, 75-80, 80 and over), and the 62-69group interpolated between HRS and AHEAD.The two sex groups are male and female; and

the two race groups are white and non-white.The four marital status groups are married,widowed, divorced, and single and all others. Intotal, there are 80 separate cells.

We construct a measure of household assetsequal to the HRS/AHEAD household net worthvariable. Then we construct individual estateasset values. For non-married individuals, thisequals household net worth plus life insurance;for married heads of household and theirspouses, it equals one-half of household networth plus relevant life insurance.

We compute the predicted number of estate taxfilings in each cell in the following three steps:

1. We take the number of death in 1992directly from mortality statistics providedto us by staff at the National Center forHealth Statistics (NCHS).

2. We calculate the fraction of individualsfor which the estate value exceeds the1992 estate filing threshed of $600,000,using the estate asset value variablebased on information from HRS orAHEAD as described earlier. Thisfraction is weighted and therefore refersto the U.S. population in this cell.

3. We multiply the fraction from Step 2 bythe number of deaths determined inStep 1, thus providing an estimate of thenumber of deaths for which it ispredicted that an estate tax returnshould be filed.

Table 1 lists three estimates for each cell: (1) thenumber of deaths in 1992 reported by NCHS, (2)the proportion of individuals for whom estatevalue exceeds the filing threshold, and (3) thenumber of estate filings expected for 1992. Notethat we compute predicted estate tax filings, notfilings for which estate tax is due or for net taxliability. In particular, for decedents who weremarried, estate tax returns must be filed on theirbehalf although they need not pay any tax if theybequeath all or most of their estate to theirspouse.



49

Table 1 Deaths, Proportions of Individuals Exceeding the Estate Tax Filing Threshold,

Predicted Estate Tax FilingsFor Decedents Who Exceeded 50 Years of Age in 1992

MaleWhite

FemaleWhite

MaleNonwhite

FemaleNonwhite

Ages 51-61

Married 71,078 .077 5,473 39,159 .054 2,115 13,608 .025 340

7,582 .019 144

Widowed 4,146 .059 245 10,205 .043 439 1,662 .000 0 4,158 .000 0Divorced 20,390 .077 1,570 11,587 .026 301 4,826 .006 29 4,744 .000 0Never Married 11,358 .079 897 4,292 .037 159 4,744 .000 0 2,206 .000 0

Ages 62-69

Married 111,213 .068 7,596 53,000 .054 2,862 15,070

.012 188 7,361 .010 74

Widowed 13,504 .071 959 32,713 .047 1,538 3,477 .010 35 8,438 .006 51Divorced 18,975 .077 1,461 12,530 .046 576 3,753 .000 0 2,704 .006 16Never Married 12,480 .076 948 5,810 .046 267 3,473 .000 0 1,734 .000 0

Ages 70-74

Married 94,941 .059 5,602 40,288 .052 2,095 10,246

.000 0 4,150 .000 0

Widowed 18,745 .109 2,043 47,454 .059 2,800 3,620 .037 134

8,494 .014 119

Divorced 10,180 .100 1,018 8,507 .075 638 1,936 .000 0 1,380 .000 0Never Married 8,407 .051 429 5,595 .073 408 1,862 .000 0 1,040 .000 0

Ages 75-79

Married 97,514 .060 5,851 36,789 .055 2,023 8,836 .000 0 3,105 .000 0Widowed 28,521 .055 1,569 75,266 .042 3,161 4,355 .000 0 10,585 .011 116Divorced 7,869 .054 425 8,084 .058 469 1,342 .000 0 1,045 .000 0Never Married 8,518 .093 792 7,677 .038 292 1,302 .000 0 928 .000 0

Ages 80+

Married 148,141 .040 5,926 46,017 .032 1,473 12,137

.000 0 3,217 .000 0

Widowed 103,211 .076 7,844 356,589 .032 11,411

11,400

.000 0 33,125 .000 0

Divorced 9,033 .041 370 16,528 .018 298 1,522 .000 0 1,448 .000 0 Never Married 15,518 .095 1,474 33,454 .076 2,543 1,914 .000 0 2,308 .000 0Note:• In each cell, the first number is the number of deaths reported by NCHS, the second number is the

fraction of individuals in that cell category for whom estate value exceeds the filing threshold, basedon tabulations made using the HRS and AHEAD datasets, and the third number is the predictednumber of estate tax filings.

• For the interpolated cells (Ages 62-69), deaths from the NHCS are imputed as:0.65 X deaths among ages 60-64 + all deaths among ages 65-69;Wealth fractions are computed as: .50 X fraction of individuals’ estates valuedabove $600,000 among ages 51-61+0.25 X fraction of individuals above $600,000 among ages 70-74+0.25 X fraction of individuals above $600,000 among ages 75-79



50

Comparing Actual and Predicted Estate TaxFilings

We have combined the individual assetdistribution derived from HRS andAHEAD with the mortality data provided by theNCHS to develop preliminary estimatesof the number of estate tax filings expected foryear of death 1992. We also comparedthe expected filings with the actual filingstabulated by SOI for 1992. Such a comparisongenerates a prediction of the number of estatetax nonfilers in 1992.

This involves comparing our estimates ofpredicted filing for 1992 decedents with SOIstatistics about actual estate filings for 1992decedents. Since the SOI estate filing statisticsare available only for 3 age groups - 2 sexbrackets - 3 marital status categories (18 cells),we collapse over relevant cells to form an 18-cellcomparison.

Table 2 summarizes the analysis and providespreliminary estimates of the number of estatetax nonfilers, together with additional informationgleaned from Eller (1995)21. The table dividesthe data into three age groups (51-69, 70-79, 80and over), two sexes (male, female), and 3marital status groups (married, widowed,others).

Table 2 lists five statistics for each cell: (1) thepredicted number of estate filers, (2) the actualnumber of estate filers reported by SOI, (3) thepredicted number of estate nonfilers, (4) the totalgross estate value reported by the filers, and (5)the net estate tax reported by the filers.

The number of predicted estate tax filings issignificantly greater than the actual number fordecedents below the age of 80, but isapproximately equal to the actual number fordecedents 80 and above. If the finding is correctit suggests that filing noncompliance isconcentrated among families of youngerdecedents. However, since the AHEAD data isthin for individuals 80 and above, we may beunder-predicting estate values for this agegroup.

Approximately two-thirds of estate tax is paid onestates of decedents who were 80 or above.

21 Eller, Martha, “Federal Taxation of Wealth Transfers, 1992-1995,” Statistics of Income Bulletin (Winter 1997), pages 8-63.

Approximately one-third is paid for a singlecategory, widowed females 80 and above. It isclearly important to learn more about this group.

Overall, the preliminary estimates suggest thatthere may not be much of a nonfiler problem forestate taxes. The estimates do predict nearlytwice as many filings for decedents below theage of 80, but much of it may be due to thefailure to take into account the relationshipbetween wealth and mortality. The estimatessuggest that there are very few nonfilings fordecedents age 80 and above, for which the bulkof estate taxes are paid.

Estimating Taxable Gifts

A household must file a gift tax return, Form709, with the IRS and pay the gift tax if duringthe year it makes a gift or gifts to an individualwith total monetary value exceeding thehousehold’s gift tax threshold. In general, thedonor of the gifts must file Form 709 on or afterJanuary 1 but not later than April 15 of the yearfollowing the calendar year when the gifts weremade. Tax is due on the excess of the amountabove the threshold. A separate line item mustbe recorded for each individual donee for whomtotal monetary value of the gift or gifts exceedsthe threshold. The tax threshold is $20,000 for amarried couple and $10,000 for divorced,widowed, and never married individuals.

Both HRS and AHEAD ask about gifts inreasonable detail. HRS asks the householdrespondent about gifts given to parents andchildren during the past year, allowing forinformation about the value of gifts given to up tofour different children. For each child to whom agift was given, HRS asks whether the gifts werein part for education or housing. AHEAD asksabout gifts given to children and otherindividuals during the past year.

Table 3 presents statistics about gift giving asreported in HRS and AHEAD, both unweightedand weighted to reflect the U.S. population ofhouseholds. The percentage of householdsreporting making gifts to at least one individualfor which the total value of the gifts exceeds thehousehold's tax threshold is approximately onehalf percent in HRS and one and one-halfpercent in AHEAD.



51

Table 2Actual and Predicted Estate Tax Filings

Reported Gross Estate Values and Net Estate TaxesFor Decedents Who Exceeded 50 Years of Age in 1992

Age 69 - Age 70-79 Age 80 +

Male Female Male Female Male Female

Married

Predicted Number of FilersActual Number of FilersPredicted Number of NonfilersReported Gross Estate ValueReported Net Estate Tax

13,5978,2175,380

$15,486 M256 M

5,1952,4172,778

$3,866 M115 M

11,4537,0724,381

$14,333 M374 M

4,1182,1301,988

$3,198 M155 M

5,9267,046

(1,120)$13,827 M

768 M

1,4731,700(227)

$2,675 M198 M

Widowed


1,239513726

$760 M109 M

2,0281,155

873$1,749 M

314 M

3,7461,5382,208

$2,374 M443 M

6,1963,0843,112

$4,522 M805 M

7,8444,9772,867

$9,247 M1,727 M

11,41111,825

(414)$19,072 M

3,574 M

Single, Divorced, Other


4,9052,0782,827

$3,267 M393 M

1,359795564

$1,189 M185 M

2,6641,0211,643

$2,391 M323 M

1,807782

1,025$1,014 M

135 M

1,8441,504

340$2,455 M

333 M

2,8412,228

613$2,730 M

297 M

Total


19,74110,8088,933

$19,513 M760 M

8,5824,3674,215

$6,804 M615 M

17,8639,6318,232

$19,098 M1,140 M

12,1215,9966,126

$8,734 M1,095 M

15,61413,2572,357

$25,529 M2,828 M

15,72515,753

(27)$24,477 M

4,069 M

Male Female Age 69 - Age 70-79 Age 80 + AllTotal


53,21833,69619,522

$64,140 M4,728 M

36,42826,11610,312

$40,015 M5,779 M

28,32315,17513,148

$26,317 M1,375 M

29,98415,62714,356

$27,832 M2,235 M

31,33929,0102,330

$50,006 M6,897 M

89,64659,81229,834

$104,155 M10,507 M



52

We have used the HRS and AHEADstatistics to develop a very rough estimate ofaggregate gift reporting and tax liability forhouseholds in which the head is older than 50years of age, for tax year 1992. The combinedfigures from HRS and AHEAD suggest that atleast 329,000 households should have reportedgifts and paid the gift tax in 1993, with total valueof gifts on which tax is owed equal to $7.2billion. With an average marginal tax rate of31%, the tax liability on these gifts equals $2.23billion.

The Internal Revenue Service Data Book 1993-94 reports that 211,000 gift tax returns were filedin fiscal year 1993 with the aggregate tax paidequal to $1.46 billion. A comparison with ourresults suggests about one third of all large gifts(above the threshold for reporting and payingtax) are not reported, and the associated gift taxliability underreporting is likely to exceed $0.77billion.

Methodological Qualifiers and FutureRefinements

The most significant methodological weaknessof the estimation procedure used to generate thepreliminary results of estate tax nonfilings is thefailure to take into account the relationshipbetween wealth and mortality. A large body ofliterature suggests that individuals of highersocioeconomic status (SES) live longer thanindividuals of lower status. Most of these studiesuse current household income or educationalstatus as proxies for SES, variables that are notdirectly relevant for estate tax analysis.Recently Attanasio and Hoynes (1996)22 haveinvestigated the relationship between wealth andmortality using data from Survey of Income andProgram Participation (SIPP). Their findingsconfirm that individuals residing in wealthierhouseholds have a significantly lower mortalityrate than individuals residing in poorerhouseholds.

Another area of extension relates to the impactof marital status on mortality, since marriedpersons typically leave all or most of their assetsto their spouse, and therefore owe little if any

22 Attanasio, Orazio and Hilary Hoynes, “Differential Mortalityand Wealth Accumulation,” National Bureau of EconomicResearch Working Paper No. 5126 (May, 1995).

estate tax. Lillard and Waite (1995)23 indicatethat marital status exerts an important impact onmortality. From this perspective, incorporatingthe effect of marital status on mortality canimprove estimates of estate tax.

In future research efforts, we plan to use datafrom waves 2 of HRS and especially AHEAD toestimate models of relationship between wealth,marital status, and mortality. The estimates ofsuch models can be used to develop moreprecise predictions about the proportion ofdeaths among individuals in each age-sex-race-marital status cell for which the correspondingestate value exceeds the filing threshold. Inparticular, the model estimates can be used togenerate wealth dependent mortality curves,thereby allowing a correction to be made for thefact that wealthier individuals face a lowermortality rate.

Since the lifetime giving of gifts is considered apreferred tool for estate planning among wealthyindividuals, more careful analysis of responsesto gift giving related questions in HRS andAHEAD can shed insight on the relationshipamong household wealth, individual healthcondition, and gift giving.

As a result, we plan to explore gift statistics fromwaves 2 in HRS and AHEAD in a life cycleframework to analyze various tax planningmotives behind gift giving, such as estateplanning or income transfer, after controlling forhousehold wealth, family structure, andindividual health and demographic factors.

Conclusions

By combining data from HRS and AHEAD withmortality statistics provided by the NationalCenter for Health Statistics, we haveconstructed preliminary estimates of the numberof estate tax filings expected for 1992, by age,marital status, race, and sex. A comparison ofour estimates with SOI statistics on estate taxfilings for 1992 has enabled us to generatepreliminary estimates of the number of estatenonfilers. Our conclusion is that there are notmany estates for which an estate tax returnshould have been filed for 1992, but was not

23 Lillard, Lee and Linda Waite, “Till Death Do Us Part: MaritalDisruption and Mortality,” American Journal of Sociology(March, 1995), pages 1131-1156.



53

Table 3Related Gift Statistics Based on the HRS and AHEAD Datasets

For Heads of Household Who Exceeded 50 Years of Age in 1992Number of Households

Making at Least One Gift

Unweighted Weighted

TotalNumberof Gifts

Unweighted

TotalValue

of Gifts

Weighted

Total Valueof Gifts on

which Tax is Owed

WeightedHRS Dataset

Gifts Greater than Tax Threshold 29(.38%)

83,000(.47%)

34 $3.0 B $1.6 B

In Part for Education In Part for Housing

--

--

247

$2.3 B1.0 B

--

Gifts Equal to Tax Threshold 19(.25%)

58,000(.33%)

23 $1.0 B 0

In Part for Education In Part for Housing

--

--

114

$0.5 B0.2 B

--

All Gifts 1,364(18%)

3,300,000(19%)

2,050 $16.2 B $1.6 B

Ahead Dataset

Gifts Greater than Tax Threshold 90(1.4%)

246,000(1.4%)

99 $9.1 B $5.6 B

Gifts Equal to Tax Threshold 59(.91%)

169,000(.96%)

71 $2.4 B 0

All Gifts 1,549(24%)

4,400,000(25%)

1,965 $23.5 B $5.6 B

filed. As a result, the nonfiler tax gap is likely tobe small.

Based on the information from HRS and AHEADabout gifts made to family members and others,we have conducted a preliminary analysis onnoncompliance with the gift tax. Our findingssuggest that there may be a substantialunderreporting of taxable gifts. However, thesetax gap estimates for estate and gift taxes are justthe preliminary results from brand new estimationmethodologies. The methodologies have someidentified weaknesses, for which future researchwill be pursued in the hopes of improving thereliability of the estimates.


Jonathan Feinstein is a Professor of Economicsat Yale University. He received his doctorate ineconomics from MIT and taught

economics at Stanford before joining the Yalefaculty. He is a leading expert on public finance,economics of aging, and econometrics, and haspublished numerous articles in the leadingeconomics and econometrics journals. Hetestified before the U.S. Congress on behalf of theIRS on tax compliance measurement issues in1997.

Chih-Chin Ho is a Senior Economist in the IRS’sNational Office of Research. He received hisdoctorate in economics from the University ofMichigan at Ann Arbor. He worked as a SeniorResearch Associate for the U.S. SentencingCommission before joining the IRS. He haspublished actively in tax compliance, publicfinance, and econometric modeling. His mostrecent contributions include “An EmpiricalAnalysis of Gift Giving in Later Life,” “TowardBuilding a Profile of an Income Tax Nonfiler,” and“Predicting Individual Income Tax ReportingAccuracy: A Trichotomous Choice Model.”

Payment Dynamics of Individual Accounts Receivable

and a New Look at Risk

By Jeff ButlerThe decisions of when and how much to pay on balance-due assessments are studied retrospectively for apanel of individual taxpayers. Combining data from the individual’s current and prior accounts receivablestatus, observed tax return, and credit history, a heuristic assessment of risk is developed, and classificationmodels are proposed. The results of such models suggest that delinquent taxpayers should be scoredaccording to risk at the time of assessment, not after they have moved through the notice stream. In doingso, it may be possible to reduce intrusiveness on low-risk cases that would otherwise pay, and identify high-risk cases before they become a financial burden.

Section 1--Introduction

One of the most challenging problems facing theInternal Revenue Service is its AccountsReceivable Dollar Inventory (ARDI), the amountof outstanding taxes, penalties, and interestowed by taxpayers at a given time. At the endof 1997, gross total ARDI stood at over $228billion, with 52 percent of that owed byindividuals.

Individual ARDI continues to grow— from $93.6billion at the end of 1994 to $118.8 billion in1997. More serious, however, is that this three-year rate of growth (26.9%) significantlyexceeds the rate of growth of new individualfilers during that same period (5%). Equallyserious is that it also exceeds the growth innominal U.S. personal income during that time(18.7%), which is an important factor usedultimately to determine ability to pay.

Those familiar with these statistics continue toask the familiar questions: Why is individualARDI growing? Who are those most likely topay? What factors contribute to those with thegreatest risk of becoming a financial burden?What constitutes risk?

This paper addresses these questions forindividual accounts that file a return with abalance due. It does so by developing a newbut simple approach for assessing risk based onsuch factors as the timing of payments, returncharacteristics, current and prior accountsreceivable status, and credit reporting data. Theanalysis will guide the reader through thedynamics of payment decisions, identify factorsthat contribute to the likelihood of risk, anddescribe patterns that are ultimately of value forpredicting the likely disposition of cases.

Inferences from this analysis will also motivateother questions: How much of ARDI is asystemic problem related to outdated billingpractices or poor toll-free telephone accessrates? To what degree do repeaters and eventax law complexities contribute to the growth ofdelinquencies? And perhaps most importantly,what classification system is currently in place, ifat all, to accurately identify the relative risk ofaccounts at the time of assessment? Is itpossible that such a system could be used toreduce intrusiveness on low-risk cases thatwould otherwise pay? Identify high-risk casesbefore they become a financial burden (i.e., notfully paid within one year)? And allow for a moreflexible development of strategies and efficientprioritization of resources?

The remainder of this paper is organized asfollows. Section 2 introduces the data used forthis analysis. Section 3 examines selectedsummary statistics for the panel of individualsunder consideration. Sections 4 and 5 use thissummary information to identify and analyzesubgroups of relative risk. Section 6 isolatesand examines several issues related todeferrals; and Section 7 offers recommendationsbased on the results of this study.

Section 2--The Data

In order to comprehensively examine paymentdynamics and other characteristics of delinquentindividual accounts, a panel must be constructedand followed retrospectively through time. Forpurposes of this analysis, a panel comprising33,263 accounts from Maryland and the Districtof Columbia was developed. Each individualaccount in this panel has a tax delinquency dueexclusively to a balance-due return that has not

PAYMENT DYNAMICS OF INDIVIDUAL ACCOUNTS RECEIVABLE AND A NEW LOOK AT RISK


55

been fully satisfied. The date of assessment onthese accounts covers the last two weeks ofMay 1996.

Several points about the construction of thispanel are necessary. First, this analysisinvestigates the payment characteristics only ofbalance-due delinquencies, which are thedominant source of assessment for individualreceivables. Those that are the result of anaudit or other source-of-assessment are notincluded here, and a separate analysis isrecommended. Second, it would be ideal tocapture N cases at a specific point in time, sayMay 15, and track the payment dynamics ofthese cases over one or two years. However,the proportion of cases selected on a given dayrepresents, in effect, the probability of anassessment on that day, which is almost zero,thus resulting in very few cases to study.(Recall that the probability of a given point isundefined, but for a small neighborhood aroundthat point it is in fact well defined.) Hence, asmall neighborhood of two weeks was used.Given that these cases are being tracked for asmany as three years (156 weeks), it is unlikelythat the results of this analysis will be sensitiveto this two-week initial condition.

Finally, it should be noted that a representativenational sample could have been drawn for thisstudy. However, such a sample would have tobe designed to account not only for uniquedifferences across geographic locations but timeperiods as well; not knowing these uniquefeatures might lead to inefficient design and thuspoor sample representation. By analogy, anyother district office could have been chosen aswell, as there was no a priori information abouttaxpayers from Maryland and D.C. thatprejudiced their selection (except perhaps thatthe overall ARDI population was large relative toother districts, resulting in a larger panel). Theseissues, particularly the need for cross-validation,will be discussed later.

Section 2.1--Data Sources

This analysis combines data from the IndividualMaster File, Accounts Receivable Database,and Credit Reporting sources. It is admittedlylimited by the lack of additional data thought tobe relevant for certain subgroups, some of whichwill be discussed in Section 4. Notwithstandingthese limitations, however, it will be shown that

these data sources contain sufficient informationnecessary not only to identify and describeimportant characteristics of payment dynamicsand risk, but to predict outcomes of risk as well.

Because this is the first known effort to use third-party credit data for the purposes mentionedabove, a few comments will be made. First, nospecific taxpayer’s records are being analyzedhere; what follows is largely a descriptivestatistical analysis of averages, rates, andproportions. Second, while there is evidencefrom this analysis that credit data may in factimprove classification accuracy for certainsubgroups, its applicability is generally notwidespread; as a result, there is a cost/benefitquestion to be addressed concerning its futureuse. Finally, analysis of credit data during thecourse of this study has left the authorsuspicious about certain features that appearstatistically irregular, although no rigorousvalidation was performed to test thisassumption. As with the need for cross-validation mentioned above, future research inthis area using such data should proceed withcaution.

Section 3--Descriptive Statistics

The ultimate goal of this study is to contribute tothe ability of the IRS to determine accuratelywhich individuals are more likely to pay, whenand how much will be paid, and the likelydisposition of accounts that do not pay. Doingso means talking about the probability ofpayment, and hence risk groups that relatecertain individuals to certain outcomes based onobservable data. This section motivates theconstruction of such groups by exploring broadcharacteristics of the panel and identifyingsimple descriptive features that may be helpfulto classification.

For the panel of 33,263 new modules used inthis study, the total assessed amount was $52million, with a median module balance of $713.(The median— half of the sample above, halfbelow— is a more appropriate measure of‘average’ than the mean for income amounts,which tend to be highly skewed.) About 75percent of the cases had a beginning balance ofless than $1,500; a full 21 percent had balancesless than $250.



56

Just over 50 percent of all modules were fullypaid within one year, paying $21.4 million (41.2percent) of the total beginning module balance.Why is payment within one year important?Because this is the time between filing taxreturns; it will be demonstrated later thatindividuals who incur a new module balancebefore paying their old one are, all other thingsequal, at greater risk of not paying at all.Therefore, this study uses one year as therelevant time period for the analysis ofpayments.

Table 3.1 shows that, of those fully pay at oneyear, 78 percent had only one module; otherfeatures thought to be relevant for illuminatingdifferences among these two groups are alsopresented.

Table 3.1 Summary statistics for full payment within one year, by payment status

Feature Pay No Pay

Median Beginning Balance $531 $933Median Income 31,705 28,761Median Age 41 40% One Module Only 78 41.8% Filing Form 1040 59.5 48.7% Remittance with Return 32.6 15.2% Schedule C or F 21.6 18.6% Homeowners1 37.8 25.8% With Investment Income2 45.3 23.9

1 Presence of real estate taxes on Schedule A.2 Presence of interest, dividend, Schedule D, or other capital gains.

While this table does not include an exhaustivelist of factors useful for discriminating those whopay from those who do not, it appears somefactors may be more useful than others. Forexample, all other things equal, those who don’tpay will have almost twice the average modulebalance; those who pay are almost 2 times aslikely to have investment income or just onemodule, and are more than 2 times as likely toattach a remittance with their balance-duereturn.

While these results can be used to heuristicallymotivate the construction of classificationmodels, they also raise additional questions.For those who do not fully pay within one year,how much, if any, was paid during that period,and what is the disposition of the account at theend of that period? Table 3.2 reveals that nearlyhalf of all individuals not fully paid are in aninstallment agreement, paying on average about35 percent of their module balance over thecourse of the year. However, it appears thatcertain cases— in particular those in bankruptcy,currently not collectible (CNC), automatedcollection system (ACS), and collection fieldfunction (CFF)— are not paying down much, ifany, of their original balances. Are individuals inthese categories at a greater risk of not paying?If so, we might want to identify factors thatcontribute to predicting the likely disposition ofthese cases and make such informationavailable at the time of assessment. Doing somight permit a more flexible prioritization ofresources for those cases if in fact they dopresent the greatest risk of becoming a financialburden.

Table 3.2 Disposition at one year of cases not fully paid

% Of Median Median% Of Ending Starting Ending

Disposition1 Cases2 Balance3 Balance Balance

Bankruptcy 3 3.9 $1,315 $1,190CNC 4.2 8.2 1,579 1,808Deferral 29.6 7.6 407 321Installment 48.5 50.2 1,308 966ACS 9.8 13.8 1,336 1,340CFF 1.9 6.3 1,654 1,903

1 Based on TRCAT status codes.2 Percent of only those cases not fully paid at 1 year.3 Percent of total module balance remaining.

Finally, it may be instructive to explore the timingof payments, and ask whether certain individualshave a higher probability of paying earlier thanothers; it is possible that such information maybe useful in the classification sense. Figure 3.1shows the probability of making a payment attime T given that one has not already beenmade.



57

Figure 3.1 Probability of making a payment versus number of days after assessment1

1 The horizontal axis runs from 30 to 270 days.

This distribution reveals what might be intuitivelyexpected: the more time that elapses before anindividual makes at least one payment, the lesslikely it is that they will do so at any time afterthat. A crucial feature of Figure 3.1 is the rate ofchange of this function, which falls toapproximately zero after 180 days— the typicallength of the notice stream. After this time,virtually no new individuals will voluntarily beginmaking payments that have not already done so(unless contacted by ACS staff or a RevenueOfficer).

While it is evident that the probability of makingat least one payment during 180 days is adecreasing function of time from the date ofassessment, this does not imply that a responsehas not been made. Many individuals with anexisting delinquency, for example, respond andenter an installment agreement but make nopayments on their current module because theyare still paying off a previous balance. However,Table 3.3 shows that for those who do make atleast one payment during 180 days, all otherthings equal, the probability of fully paying theirbalance at one year is also a decreasingfunction of time. This too, seems obvious: thesooner I start paying, the sooner I finish. Butwho is making contact and beginning paymentssooner? Are there relevant data that can beused to distinguish this group from others?

As it turns out, 58.3 percent of individuals for thispanel make at least one payment within 180days after assessment. As expected, a majority(77 percent) have only one module. Also,excluding deferrals (for reasons to be discussed

below), all 21 percent of those with a balance of$250 or less belong to this group.

The broad characteristics described thus farseem to suggest several distinct patterns: 1)individuals with very small balances mayconstitute little or no risk if a significantpercentage fully pay within one year; 2)individuals with just one module are nearly twiceas likely to fully pay than not pay, and it seemsintuitively appealing to isolate such cases forfurther analysis; and 3) it follows that individualswith two or more modules should be separatelyanalyzed as well.

Table 3.3 Distribution of individuals making at least one payment within 180 days and the probability of full payment at one year

Number of % Making % Who MedianDays from at least 1 Full Pay StartingAssessment Payment1 at 1 Year Balance

30 Days 11.1 77.2 $62560 Days 19.7 75.5 84890 Days 10.2 64 785120 Days 5.9 64.5 699150 Days 5.7 60 593180 Days 2.7 49.5 694

1 Number of those making at least one payment who have not yet made a payment as a percent of panel.

Section 4--Low to Moderate Risk Groups

Of the original panel of 33,263 cases, 6,836 (21percent) had a beginning balance of $250 orless. Of these, 69 percent fully paid within oneyear. However, of those that did not fully pay atone year, 72 percent were deferrals and 20percent were in an installment agreement;adjusting for deferrals gives a payment rate of88 percent for this group. This can be seen inFigure 4.1, which shows the probability ofpayment within one year as a function ofbeginning module balance. It is evident theprobability of full payment is relatively constant(above 85 percent) for balances up to about$250, after which it falls off rapidly. As a result,it appears that module balance alone issufficient for describing the likelihood of paymentfor very small balances.



58

Figure 4.1 Probability of full payment at one year as a function of module balance, excluding deferrals1

1 Module balance is in logarithms to accentuate the shape of the distribution; the horizontal axis shows actual dollar amounts associated with its logarithm.

Virtually all individuals with balances under $250either fully pay or are deferred within one year.Therefore, excluding deferrals will not affect theresults. In fact, it actually helps classificationefforts: whoever is not deferred will fully pay (asmall percentage will be in an installmentagreement). However, two questions should beasked at this point: do those in deferral statuswith small balances truly represent low risk?After all, one reason they are in deferral status isbecause they failed to respond to one or morenotices, which might be considered a risk factorper se. And, will a significant percentage (say,85 percent or more) in fact fully pay when theIRS eventually contacts them or visa versa?The answer to the second question will beaddressed in Section 6, whereas the firstquestion is examined in Table 4.2 below, whichcompares those with balances under $250 whofully pay at one year with those who aredeferred.

Although dollar amounts of $250 or less mayseem insignificant, there are features in thistable that merit discussion. First, based onincome alone, these two groups appear to haveroughly the same ability to pay. If this is true,why are some responding to a notice and notothers? Second, a closer look at these featuredifferences may lead one to conclude that theIRS may be deferring, on average, cases thatare relatively more risky. For example, 19percent of those who fully pay have just one

module versus 42 percent for deferrals. Is theIRS allowing new deferrals on top of existingones, and if so, does this eventually increase therisk of non-payment? That is, while theseindividuals appear to have the ability to pay$250 today, will they have that same ability topay an accumulated deferral balance of say,$1,000 tomorrow?

Table 4.2 Selected characteristics of deferrals versus those who fully pay within one year for individuals with beginning module balances of under $250

Feature Deferral Full Pay

Median Starting Balance 136 124Median Income 19,181 22,592Median Credit Balance1 10,445 10,472Median Age 35 37% Only One Module 57.8 81% Filing Form 1040 26.9 44.5% Filing Status of Single 62.8 56.2% Remittance with Return 8.3 22.6% Schedule C or F 5.9 12.1% Homeowners 9 22.5% With Investment Income 12.7 35.9

1 Based on total outstanding credit as of 5/96.

Section 4.1--Cases with Only One Module

Of the 33,263 individuals in the panel with astarting balance exceeding $250, 11,976 (45.4percent) fully pay within one year. Of these, 77percent have just one module. On the surface, itwould seem that the number of modules is animportant indicator of who is likely to pay. As itturns out, sophisticated classification techniquesdiscussed later support this observation. Simplecounting methods bear this out as well:individuals with just one module are 1.6 timesmore likely to fully pay in one year than not pay;those with multiple modules are 3 times morelikely not to pay than pay. Intuitively, thesesimple odds would seem to suggest a separateinvestigation of cases with just one module.

To do so, several basic questions needattention: what characteristics, if any, distinguishthose who pay from those who do not; what



59

proportion of these cases are repeaters; and arethere other patterns not previously studied thatcan improve our understanding of this market?

Table 4.3 addresses the first issue by comparingthose who fully pay within one year from thosewho do not (for those with a single modulewhose balance exceeds $250). Patterns of riskare evident: those who fully pay are almost twiceas likely to submit a remittance with their return;will have a module balance relative to incomethat is, on average, 50 percent lower than thosewho do not fully pay; and will be 1.5 times morelikely to own a home or have investment income.Those who do not pay appear more likely to besingle and slightly younger as well. One featureof particular significance that will be discussed inmore detail below is the percentage ofindividuals who make at least one payment in180 days: those who do so are roughly 3 timesmore likely to pay.

Before turning to that issue, there is the questionof repeaters: of those cases with just onemodule, who had a delinquency on a balance-due return in the prior tax year? Of the 15,068cases with one module whose balance wasgreater than $250, 22.3 percent met this simplecriterion. What is somewhat more interesting isthat the repeater rate for those who fully paywithin one year is higher (25.6 percent) thanthose who do not (17.3 percent). Also ofinterest: of those who fully pay and have aremittance with their return, 61 percent use a taxpreparer. Are preparers correcting fornoncompliance, for example, obliging a taxpayerto file a balance due return rather thandeliberately overstate expenses or understateincome to avoid owing taxes? Or are thesesimply corrections to an unanticipated liabilitybased on a taxpayer’s misunderstanding of taxlaw or return instructions?

Finally, Section 3 asked whether the timing ofpayments is important: do certain individualshave a higher probability of paying earlier thanothers? What inferences, if any, can be drawnfrom such a market? Table 4.4 addresses thisquestion by looking at the distribution ofpayments within 180 days and the associatedprobability of full payment within 360 days.Selected financial and other characteristics areincluded in the hope of finding patterns thatmight also be useful for classification purposes.

Several insights can be drawn from Table 4.4.First, the obvious: those who delay payment (orpossibly even contact), all other things equal,will be less likely to pay within one year (seeTable 4.3). Second, there is evidence that thosewho delay payment are somewhat younger, filesimpler returns, have smaller incomes, and areless likely to own a home or have investmentincome. But what other factors determinewhether an individual delays their response to anotice? Could there be explanations that aresystemic and treatable, such as poor toll-freetelephone access or the ambiguity of a noticerelative to industry billing statements?

Finally, an examination of account dispositionsfor those not fully paid at one year is warranted.Table 4.5 shows that almost 95 percent of thesecases are in just three categories.

Table 4.3 Selected characteristics of individuals with one module, by

payment status1

Feature Pay No Pay

Median Beginning Balance $800 $1,015Median Income 36,612 29,656Median Balance Burden (%)2 2.2 3.4Median Credit Balance 10,114 10,023Median Credit Burden (%)3 2.8 3.4Median Age 41 38% Filing Form 1040 67.1 54% Remittance with Return 40.7 23.7% Who Use Preparer 51.7 43.6% With Home or Investments 64.5 43.2% Single Filers4 46.3 56.4% With Schedule C or F 25.8 21.1% Payment within 180 days 69.1 23.7

1 Module balance greater than $250.2 Starting module balance as a percent of income.3 Credit balance as a percent of income.4 Includes Head of Household and Widowers.



60

Table 4.4 Distribution of individuals making at least one payment within 180 days, theprobability of full payment at one year, and associated characteristics

Number of % Making % Who Median % With % WithDays from at least 1 Full Pay Starting Median Median %1040 Home or ReturnAssessment Payment1 at 1 Year Balance Income Age Filers Investments Remittance

30 Days 6.4 77 $967 $38,928 44 72.8 68.7 62.160 Days 12.9 74.2 1,034 37,904 42 69.1 66.3 45.490 Days 6.4 60.8 943 34,826 40 61.3 57.4 31.7120 Days 3.3 63.6 904 34,973 39 61.5 56.8 24.4150 Days 3.1 57.8 730 30,643 38 55.8 49.5 24.5180 Days 1.5 48.7 744 30,887 38 56.7 46.3 22.5

1 See Table 3.3 for definition.

Table 4.5 Selected characteristics of cases not fully paid one year after assessment

Feature Deferral Installment ACS

Percent of Cases1 38.9 47.5 7.1Median Balance 611 1,456 1,769Median Income 23,755 35,504 30,488Median Burden2 2.6 4.1 5.8Median Age 36 40 38% 1040 44.6 59.6 56.8% Single 49.2 35.1 38.1% Home/Invest 32.2 51.1 38.6% Schedule C or F 16.1 23.5 29.1

1 Percent of those not fully paid within one year.2 Median balance as a percent of median income.

There is evidence that cases in ACS present thegreatest risk: they can be readily distinguishedbased on their high Median Burden— thebalance owed as a percent of income. Whomakes up this group? Not surprisingly, asignificant percentage (54 percent) are thosesame individuals more likely to be young, single,and have no homeownership or investmentincome. As a result, a more detailed profile ofthis group is provided in Section 4.2.

It may also be of some interest that cases indeferral status have, on average, a modulebalance that is only slightly higher relative toincome (2.6 percent) than individuals who fullypay within one year (2.2 percent; see Table 4.3).Because of this, and since they have a balance

owed on just one module, they may representlittle or no risk. However, as discussed earlier,this may not be the case if the IRS is allowingconsecutive deferrals. This question will betaken up again in Section 6.

Section 4.2--Classification and Risk

In Section 3, it was demonstrated that thenumber of modules is an important factor, allother things equal, in describing the likelihood ofpayment. It was that result which in factmotivated the work in this section; that is, ofisolating and analyzing individuals with just onemodule.

Splitting the panel on this particular feature wasno accident or guess, but rather the outcome ofstatistical, machine learning, and tree-basedmethods designed to look for such effects.Although a full description of these techniques isbeyond the scope of this paper, several keyresults can be discussed.

First, individuals with just one module are fareasier to describe and classify relative to the riskof non-payment than those with more than onemodule, as will be seen in the next section. Infact, every modeling technique mentioned aboveconfirmed that only a few key features areneeded to develop accurate risk analysis modelsfor this group.

Second, and perhaps more importantly, suchmodels can be examined heuristically throughthe use of decision trees to describe the rulesinvolved, making them easy to interpret and



61

Remittance?

Yes No

Fully Pay Within360 Days: 76.4%540 Days: 85.5%

Homeownershipor Investments?

Yes No



Single Module

Balance Under $250?

Fully Pay Within360 Days: 79.2%

YesNo

understand. Finally, in light of this, such modelswould be very easy to implement in practice.

Combining the results of this section into such aheuristic decision tree gives Figure 4.1, whichreveals four distinct subgroups: those with smallbalances, and those who: 1) have a remittancewith their return; 2) have no return remittanceand no homeownership or investment income;and 3) have no remittance but do havehomeownership or investment income.

All other things equal, those who submit aremittance have a high probability of fullpayment within one year (76.4 percent), andmight be considered low risk; the same is truefor those who fail to remit but own a home orhave investment income (75 percent). Onegroup, however, has a distinctly different risk:those who have no return remittance and haveno homeownership or investment income (lessthan 60 percent fully pay within one year).

Table 4.5 examines these last three subgroupsin more detail, and it is not surprising that thesame group of younger taxpayers more likely todelay payment (from Table 4.4) are those leastlikely to fully pay in one year— and more likely tobe in ACS. Whether these results can begeneralized for assessments from districts otherthan Maryland and D.C. is a testable hypothesisthat warrants further analysis.

Several other hypotheses relating to thissubgroup (those with no remittance and nohomeownership or investment income) shouldbe tested as well. First, this group has asignificantly higher credit balance relative toincome (43.4 percent), and there is thepossibility that this burden, along with other debtnot used in this analysis (for example, studentloan payments), creates financial pressures toogreat for some. Second, is the repeater raterelatively higher for these individuals? (Thisquestion was not analyzed here.) Finally, giventhat they are younger and therefore more likelyto change jobs, is this simply a withholding (W-4) problem?

Figure 4.1 Example of a heuristic decision tree used to predict the likelihood of payment

Table 4.5 Selected characteristics of the subgroups represented in Figure 4.1

No Remit No Remit(No Home/ (Home/

Feature Remit Investment) Investment)

MedianBalance

1,027 673 1,056

Median Income 40,900 24,223 43,279Median Credit1 9,892 10,519 10,321Credit Burden2 24.2 43.4 23.8Median Age 42 35 44% 1040 74.3 30.9 81% Single 32.4 50 27.8% Head House 8.3 19.1 12.6

1 See Table 4.2 for definition.2 ‘Median Credit’ as a percent of median income.



62

Section 5--High Risk Groups

In the previous section, it was shown that theodds of not fully paying within one year weremuch lower for individuals with more than onemodule (3-to-1 versus those with just onemodule). Again, this makes sense if theindividual is still paying on a prior year balance.However, this raises several simple butimportant questions for those with more thanone module: 1) How much is being paid on priormodule balances; 2) What factors can be usedto predict the likely disposition of a new module;and 3) Is there a relationship between paymentactivity on prior modules and expected futurepayment on a new module?

Before exploring these questions, it is worthnoting that individuals with multiple openmodules can be studied in two ways: thecharacteristics of each module can be examinedseparately (module-level analysis), or incombination (entity-based analysis). Whilespecific needs often drive which approach totake, any attempt to construct reliable models forclassification purposes requires the use of both,and that is the direction taken here.

Of the 33,263 modules in this panel, 13,350(40.1 percent) had more than one module (withthe module balance on the new assessmentexceeding $250). Of these, 27 percent fully paythe most recent module within one year.Roughly 62 percent of those who fully pay theirnew module balance also fully pay their entitybalance. For those who do not fully pay theirnew module balance, Table 5.1 examines thequestion of how much is being paid on both themodule and entity balance over one year, bydisposition of the account at the end of that year.Aside from those who fully pay, only those in aninstallment agreement are, on average, payingdown their balances.

It is evident from the median balances reflectedin Table 5.1 that individuals with multiplemodules represent a significantly different riskthan those with just a single module. This is trueeven for individuals in installment agreements,which one may be inclined to consider relativelyless risky. This can be seen in Table 5.2, whichshows that, for those in an installmentagreement, the probability of making paymentsto any module is a decreasing function of thenumber of modules. That is, although the entity

balance is being paid down, the reduction isinversely related to the number of modules:those with two modules pay down, on average,almost 8 percent of their total entity balance overone year, while those with six pay down justover 2 percent.

Table 5.1 Payments made on current module and entity balances over one year for cases with more than one module, by disposition

Starting Ending Starting Ending

Module Module Entity Entity

Disposition Balance1 Balance Balance Balance

Bankruptcy $1,418 1,384 4,764 4,899

Deferral 532 487 1,108 1,115

CNC 1,657 1,971 9,886 11,167

Installment 1,342 1,236 3,599 3,332

ACS 1,192 1,292 3,531 4,053

CFF 1,655 1,994 10,105 11,635

1 All balance amounts are in medians.

Table 5.2 Installment agreement payments over one year, by number of modules

Number Starting Ending Starting EndingOf Module Module Entity EntityModules %1 Balance2 Balance Balance Balance

2 48.2 $ 1,427 1,119 2,513 2,3203 26.7 1,241 1,284 3,744 3,3424 13.1 1,208 1,331 5,677 5,2915 5.3 1,249 1,378 8,425 7,7336 2.7 1,338 1,542 14,872 14,536>6 3.6 1,682 1,984 34,969 35,286

1 Percentage of installment agreement cases.2 All balance amounts are in medians.

Another troublesome feature of Table 5.2 is thatfor those with just two modules, the entitybalance decreases by less than the currentmodule balance. It seems that not only is therisk of non-payment in installment agreements



63

higher for those with more modules, but even forthose with just two modules. Why is this?Accumulating interest alone? As it turns out, thereason can be attributed to new modulesopening at the end of 360 days— to repeaters.In fact, of the 4,721 individuals in an installmentagreement comprising Table 5.2, a full 2,232(48.1 percent) had a new balance-dueassessment the following year. The fact thatnearly one out of every two individuals withmultiple modules an installment agreement willbe a repeater raises serious questions about therole the IRS should play in preventionmanagement— especially for cases that mayoffer the greatest chance of recovery throughtreatment. (The question of whether a repeaterrate for this panel is representative of the nation,however, is beyond the scope of thisinvestigation.)

The next basic question that needs to beaddressed is what features, if any, can be usedto describe and predict the likely disposition ofaccounts. Table 5.1 clearly reveals thatindividuals in Bankruptcy, CNC, and CFF statushave significantly higher entity balances relativeto income. Is this feature alone sufficient forheuristically categorizing these individuals asinherently more risky? Probably not, althoughadditional characteristics of these cases shownin Table 5.3 seem to support this conjecture:those in Bankruptcy, CNC, or CFF also have ahigher number of modules, on average, thanindividuals in other dispositions; a higher relativeaccumulation of penalty and interest; and forbankruptcies in particular, relatively higher creditbalances. In fact, if further research cross-validated the association between high creditbalances and bankruptcies, the IRS might usesuch evidence to explore the impact of easycredit standards in the banking industry onpotential loss of revenue to the U.S. Treasury.One additional feature of interest from Table 5.3relates to individuals in CNC status: they tend tobe nearly twice as likely, on average, to have aSchedule C or F than other groups. However,their average income, homeownerhip, andinvestment rates seem to reflect an unstablesource of income. Clearly, this type ofinformation— were it available at the time ofassessment in the form of a risk analysis scoringsystem— might benefit collection activities in awide variety of ways.

The last question to be explored here is whethera relationship exists between prior paymentactivity— or the disposition of a prior module—and expected future payments. For individualswith just one module, it was seen that timingplayed a key role: the sooner contact is madeand payments started, the higher the probabilityof fully paying within a reasonable period oftime. (Ironically, it may be timing alone that isresponsible for a large percentage of multiple-module cases: if I begin payments late and don’thave last year’s balance paid off before beingassessed with a new delinquency this year, myoverall liability is compounded.) The questionhere is whether the disposition of the lastmodule will be a determinant, all other thingsequal, of the likely disposition of a new module.

Table 5.4 investigates this issue by constructingthe joint distribution of the probability that anindividual will have a particular disposition for anew module given the current disposition of theirmost recent module. Of value here is the maindiagonal of this table, which shows theprobability of a particular disposition for a newmodule given that the last module is currently inthe same disposition. For example, theprobability that an individual’s new module willbe in bankruptcy given that their most recentmodule is currently in bankruptcy is 42.4percent.

From this table, some useful results can becomputed: the odds of full payment are 2 timesgreater for those whose prior module in eitherdeferral or installment agreement than thosewith any other status; the odds of a new modulebeing in either bankruptcy, CNC, or CFF statusis over 11 times higher for those whose priormodule was in either of those three categories;and so on. In short, the joint probabilitydistribution in Table 5.4— in combination withadditional features from Table 5.3— appears tobe very useful for developing reliable modelsthat predict the likely disposition of multiple-module cases, to be discussed below.



64

Table 5.3 Selected characteristics of multiple-module account dispositions at one year, bydisposition category

Starting Starting Starting Median % With % With % With % With Module Entity Penalty/ Median Credit Schedule % 1040 Over 2 Home/ Return

Balance1 Balance Interest Income Balance2 C or F Filers Modules Invest2 Remit2

Disposition Full Pay $ 912 2,198 172 35,933 9,867 24.0 61.4 28.8 53.9 22.5Bankruptcy 1,418 4,764 639 38,091 12,127 28.6 72.0 65.6 60.8 11.1Deferral 532 1,108 98 25,955 10,462 15.5 40.2 30.2 26.7 13.4CNC 1,657 9,886 1,707 26,120 10,397 35.8 54.8 76.4 29.7 7.4Installment 1,342 3,599 342 32,877 10,303 17.4 50.5 51.8 39.6 11.4ACS 1,192 3,531 372 26,821 10,853 19.1 44.6 57.7 26.6 8.7CFF 1,655 10,105 835 33,341 10,562 22.0 57.4 69.3 41.7 9.4 1 All dollar amounts are in medians.2 See Table 4.3 for definition.

Table 5.4 Joint probability distribution of the disposition of a new module at one year giventhe disposition of most recent module

Disposition of Last Module1

Disposition of New Module Percentat One Year of Cases2 Bankruptcy Deferral CNC Installment Full Pay ACS CFF Bakruptcy 3.0 42.4 6.3 1.2 32.2 9.9 2.1Deferral 9.5 0.4 56.5 8.7 24.0 4.3 0.4CNC 5.1 1.1 5.0 28.0 20.9 24.3 4.1Installment 42.7 1.1 12.8 0.8 61.0 13.3 0.9Full Pay 24.9 1.9 22.2 1.7 53.6 9.9 1.2ACS 9.8 3.4 24.1 2.9 30.5 25.5 0.8CFF 2.3 1.7 10.9 0.4 13.4 19.7 26.4 1 At the time of assessment of the new module.2 Represents the percentage of cases for new modules at the end of one year.

Section 5.1--Classification and Risk

Section 4 outlined a simple, heuristic approachfor classifying individuals with one module basedon just a few features. For example, almost 80percent of those with one module who have aremittance with their return will fully pay in oneyear. Can a similar rule-based approach withsuch high accuracy be developed for individualswith more than one module— those representingsignificantly higher risk?

Unfortunately, very few heuristic rules could befound for this group: the information space isrelatively more complicated, and thus theclassification problem more difficult. However,

using statistical, machine learning, and tree-based techniques with many of the featurescovered in this section, risk assessment modelscan be developed with accuracy similar to thoseof the previous section (77 percent to 89percent). There are several models to consider:1) who is likely to fully pay their new modulewithin one year— a strong indicator of who willfully pay their entity balance; 2) of thoseremaining, who is likely to be disposed inBankruptcy, CNC, or CFF— cases with thehighest relative risk; and 3) of those likely toenter an installment agreement, whichindividuals are most likely to make payments.Of course, there are other viable classificationmodels as well, and those mentioned above are

by no means exhaustive. Those interested in a more detailed analysis of classification results from thisstudy should contact the author.

Section 6--Deferrals Revisited

Of the original 33,263 modules in thisretrospective panel, 16,570 (49.8 percent) did notfully pay their balance within one year; of these, afull 30 percent were deferred (see Table 3.2), withthe total amount deferred representing 7.7 percentof the total module balance remaining at one year.Of those in deferral at the end of one year, 35percent had more than one module.

Earlier sections presented a cursory examinationof deferrals and, based on those results, raisedseveral critical questions about their relative risk:What is the rate of payment on deferrals over say,two or three years? What are the risks, if any, ofdeferring balances for individuals with more thanone module? What is the probability that adeferral ends up in bankruptcy, ACS, or CFF? Dosuch probabilities depend on the number ofmodules?

To answer some of these questions, a separatepanel of individuals from Maryland and D.C. wascreated using the methodology outlined in Section2.1, with the only difference being the year ofassessment— 1995 instead of 1996. This earlierdate will allow for retrospective tracking over threeyears instead of two. The new panel contained27,305 cases, of which 14,508 (53.1 percent) didnot fully pay within one year; of these, 3,922 (27percent) were deferred. Of those in deferral atone year, 34.7 percent had more than onemodule.

At two years, only 1,491 (38 percent) of thoseoriginally deferred fully pay their balance; of these,70 percent have just one module. Of those whodo not pay, however, 61 percent have more thanone module at two years. Table 6.1 shows that ofthose who did not fully pay within two years, only79 percent remain in deferral status; 10 percentare in an installment agreement. However, almost9 percent are indispositions that might be considered categoriesof risk: bankruptcy, CNC, ACS, and CFF. Moreimportant is the percentage of originally deferredcases that have more than one module.

Table 6.1 Distribution of deferrals after two years, by disposition and number

of modules

Number of Modules

Disposition %1 1 2 3 4+

Bankruptcy 0.8 15 60 20 5CNC 0.7 19 50 25 6.2Deferral 79.3 46 31.5 16 6.5Installment 10.1 11 35.6 38.1 15.8ACS 6.4 12 34.6 32.7 21.2CFF 0.7 13 56.3 12.5 18.7

1 Percentage of all originally deferred cases that have not fully paid at two years.

At three years, 2,202 (56 percent) have fully paidtheir original balance, but as shown in Table 6.2,the proportion of cases in risk categoriesincreases by 84 percent, from 9 percent to 16.6percent. There is also a noticable shift in thepercentage of cases towards an even greaternumber of modules.

Table 6.2 Distribution of deferrals after three years, by disposition and number of modules

Number of Modules

Disposition %1 1 2 3 4+

Bankruptcy 2 17.6 20.6 29.4 32.4CNC 2.6 24.4 37.8 26.7 11.1Deferral 72.4 48 29.1 14.0 8.9Installment 9 10.6 19.4 35.6 34.4ACS 10.5 13.3 27.1 30.9 28.7CFF 1.5 4 20 44.0 32

1 See Table 5.5.

These results are troubling for a number ofreasons. First, Section 4 asked whether ahigh percentage of deferrals with balances of$250 or less were, on average, being paid in twoyears— a more than reasonable amount of time

66

for such an amount under most financialconditions. If so, then these cases might trulyrepresent zero risk. In fact, of the panel ofdeferrals above with beginning entity balances of$250 or less, only 43 percent fully paid in twoyears.

Second, as with the issue of timing discussed inSection 4, one might infer from the above datathat the probability of risk (i.e., of non-payment)for deferrals is an increasing function of time: thelonger the deferral is ignored, the greater thechances of it moving into a bankruptcy, ACS, orCFF status— all costly to the IRS.

Finally, there is strong evidence from the abovetables of consecutive deferrals, which by definitionincreases the number of modules and henceentity balance as well. Allowing a new deferral ontop of an old one may be compounding the risk ofnon-payment: as the number of modulesincreases, the accumulation in entity balance mayeventually become unmanageable relative to afixed income source.

Section 8--Discussion and Conclusion

One need only examine Table 7.1 to get arenewed sense of the challenges inherent indelinquent individual collections. For the panel of27,305 cases studied in Section 6 with newassessments in May of 1995, the total beginningentity balance was just over

Table 7.1 Payments on aggregate entity balances over three years

# of Percent Entity PercentDate Entities1 Change2 Balance3 Change

May-95 27,302 - $134,319 -May-96 19,061 30.2 125,902 6.3May-97 13,978 26.7 121,531 3.5May-98 10,918 21.9 120,507 1.0

1 Original number of entities is shown at 5/95.2 Percent reduction from previous year.3 In millions.

$134 million. After one year, 6.3 percent waspaid; after two years, an additional 3.5 percentwas paid; after 3 years, just 1 percent. If onewere to track this aggregate payment functionbeyond three years, it may show continued

payments, but at a decreasing rate. It would alsoshow dispositions for these cases in roughly thesame proportion as presented in Section 5.

Who are these individuals remaining after threeyears with large entity balances that are at risk ofnon-payment? What factors can be used todescribe such risk? Clearly, this researchprovides concrete answers to these basicquestions. Can we identify these cases earlyenough so that, where appropriate, differentstrategies or treatments can be pursued? Theanalysis presented in Sections 4 and 5 wouldindicate that the answer is yes.In short, the results of this study suggest that theability to identify and distinguish individuals whoare likely to pay from those that are not— in aframework of relative risk at the time ofassessment— may be an important ingredient forhelping reduce the inventory reflected in Table7.1. It also demonstrates that building accuratemodels for such purposes is not beyond the reachof the IRS. At a minimum, a system designed toscore delinquent accounts at the time ofassessment could offer:

§ Greater ability to identify and devoteresources to high-risk accounts before theybecome a financial burden.

§ Potential to reduce intrusiveness on taxpayerswho are otherwise likely to pay.

§ Ability to prioritize workload for more efficientutilization of IRS resources.

Before the merits of developing and implementingsuch a system are debated, the work of this papermust be cross validated against other districtoffice data and time periods. However, if thepatterns are favorably close, it would imply that anew risk analysis system such as that proposedshould be given serious consideration— perhapsfor businesses as well as individuals. It wouldalso present an opportunity for additional researchto study unique customer markets found inSections 4 and 5. For example, why is therepeater rate for individuals in an installmentagreement so high? And why are such a highpercentage of low-risk delinquencies (Section 4)associated with returns prepared by taxpractitioners?

Such new research might require a morequalitative framework— perhaps through the useof survey instruments or focus groups— toinvestigate such questions. Based on inferences

67

from this paper, it might also address, withmonetary quantification where possible, severalother topics:

§ Certain taxpayers demonstrate a higherpropensity for delaying payment on a notice.Would such taxpayers respond sooner to abilling statement such as those used byutilities, banks, and other lenders in industry?Such a bill might show a “Minimum AmountDue”, in effect permitting either full payment orautomatic installment.

§ Is the probability of delaying payment relatedto poor toll-free telephone access rates? If ataxpayer calls about a notice but is unable tospeak to a customer service representative,what percentage abandon their efforts as aresult?

§ The repeater rate for multiple-module cases inan installment agreement is nearly 50 percent.If states have the authority to mandate “DriverEducation” classes for those with poor drivingrecords, could the IRS secure the authority tomandate a similar program? Would such taxcounseling reduce repeater rates?

§ Individuals who tend to be younger and singlealso tend to have higher credit balancesrelative to their incomes. Would partnershipprograms with local financial firms helpcounsel these individuals through educationor credit consolidation?

§ If individuals are 30, 60, or 90 days late onmaking payments through a newer billingmechanism, could it be reported to a creditbureau? What benefits might accrue fromsuch action?

§ About two-thirds of individuals who will likelypay their delinquent balance also use a taxpreparer. What percentage of returns arebeing corrected for an anticipated tax liabilitybased on a misunderstanding of tax law orreturn instructions? That is, to what extentdoes tax law complexity contributesystemically to delinquencies?

§ Payment rates on deferrals are much lowerthan expected two or three years afterassessment. One reason for this may be thatthe IRS allows new deferrals on top of oldones, creating the potential for balances thatgrow too large relative to income. Could theIRS prevent this by simply not allowingconsecutive deferrals?

§ There is evidence that payment rates ondeferrals are inversely related to the time thecase remains in deferral. Could the IRSreduce its exposure to this risk by routingdeferrals to ACS after a specific period oftime, say one year?

It is hoped that this analysis, as well as resultsfrom any future research proposed above,provides more than just a new look at an oldproblem. Given the magnitude and direction ofaggregate delinquent balances, it would seem thatmuch more is needed. A modern system foraccurately identifying risk at the time ofassessment might just be a good place to start.

Acknowledgments

The author wishes to acknowledge theCompliance Data Warehouse, without which thisresearch would not be possible; the support ofMaury Harwood, Rudy Estrada, and Bruce Colton;and the helpful comments of Mike Graeber,George Coakley, Doug Beazley, and JoelFriedman.


Jeff Butler is an Operations Research Analyst inthe AC (Research and Statistics of Income) area.He received his M.A in Economics in 1988 fromPenn State University. He has been with the IRSsince 1988.

68

Application of Scientific and Statistical Methods to an Operational Program: A CaseStudy of the Bank, Post Office and Library Program

By Denise York Young andErika D. Alexander

Prior to 1991, the IRS recorded very limited data on individual tax products delivered through the Bank, PostOffice, and Library (BPOL) Program. Continual efforts since then to improve both the quantity and quality ofdata have enabled more systematic research on ways to improve the program. As described in more detailin this article, the use of quantitative methods has led to major modifications to the BPOL Program. Themodifications have served to reduce the number of forms remaining at BPOL outlets at the end of the filingseason, while increasing the number of taxpayers filing forms obtained from BPOL outlets. The use ofscientific and statistical methods in the BPOL Program demonstrates a model that can be applied to otheroperational programs. The primary components of this model are (1) initiation of data collection processes,(2) implementation of data-driven operational decisions, and (3) control of external factors.

Introduction

The Internal Revenue Service’s Bank, Post Office,and Library (BPOL) Program distributes Federaltax forms to the public through approximately45,000 outlets nationwide. A BPOL account is abank, post office, library, or other entity with aformal arrangement with the IRS to distribute taxforms through the BPOL Program. A BPOLaccount consists of one or more outlets. TheBPOL Program serves over 18 million taxpayers,and has a printing budget of $7 million. In 1991,in an effort to improve its efficiency, the IRS beganstudying this program using a variety of scientifictechniques and statistical methods. This work hasresulted in the following improvements from 1991to 1997:

• an increase in the number of taxpayers filingforms obtained at BPOL outlets (from 13million to 18 million),

• a decrease in the number of surplus formsremaining at BPOL outlets at the end of afiling season (from 77 million to 43 million),

• a decrease in the number of forms sent toBPOL outlets (from 394 million to 361 million),and

• a decrease in the number of BPOL locationsdistributing forms (from 100,000 to 45,000).

There are 10 standard items that all BPOLaccounts receive: Form 1040, Form 1040A, Form1040EZ, Schedule A&B, Schedule EIC, Schedule1, Schedule 2, Instructions 1040, Instructions1040A, and Instructions 1040EZ. Theimprovement efforts described in this paper were

aimed at all 10 products, but particularly the Form1040. Tables 1, 2, and 3 contain historical datarelative to Forms 1040, 1040A, and 1040EZ,respectively. They show the volumes distributedto BPOL outlets, the surplus amounts remainingat the end of each filing season, and the numbersof taxpayers filing the forms obtained from BPOLoutlets.

As a result of the research, the number of Forms1040 distributed to BPOL accounts decreasedabout 25 percent since the 1992 filing season,while the number of taxpayers filing a Form1040 obtained from a BPOL outlet increased.Moreover, the number of surplus Forms 1040remaining in BPOL outlets at the end of the filingseason was reduced by almost half. There wereno reductions in the number of Forms 1040Aand 1040EZ sent to BPOL outlets; however,their surpluses at the end of the filing seasonalso decreased.

Data Collection Processes

Prior to 1991, the IRS recorded only the mostbasic data needed for delivering forms to theoutlets. The program recorded the type ofaccount (i.e., bank, post office, library, other)and the number of locations a particular accountserviced. But this was often inaccurate andincomplete. Ordering information for specificaccounts was not retained from year to year,making analysis of historical trends at a microlevel impossible. Even though the Service knewthe quantity of forms sent to BPOL outlets, it did

A CASE STUDY OF THE BANK, POST OFFICE AND LIBRARY PROGRAM


69

not know the portion actually filed by taxpayers,or the unused portion at the end of the filingseason.

In order to increase efficiency, the BPOLProgram committed itself to increasing thequality and quantity of its operational data. Forinstance, to gain insight on taxpayer usage andend-of-season surpluses, the IRS began printingsource codes on the forms distributed by BPOLoutlets in 1991 (i.e., “B” for bank, “L” for library,and “P” for post office). From an operationalstandpoint, source codes added complexity to theBPOL Program. The source codes had to beadded manually to the proofs of each tax form,additional contracts had to be administered, andthe distribution center had to handle additionalproducts. However, the data collection aspects ofthis initiative were made easy by an agreementreached between the BPOL Program and theIRS’s Statistics of Income (SOI) Division. The SOIDivision conducts an Early Tax Estimates (ETE)Study that draws a random sample ofapproximately 20,000 returns each year. As aresult of the agreement, SOI modified its ETEStudy to capture source code information used toestimate the lower bound on the number of peopleserved by BPOL outlets. It should be noted,though, the ETE Study does not capture data onan undetermined (and probably large) number oftaxpayers that picks up forms and instructions atBPOL outlets but does not file them.

To supplement the information provided by theETE Study, the BPOL Program established anannual inventory report in April 1993 to determinethe end-of-filing-season surplus of the 10 standardproducts at each BPOL outlet. Because of theexpense of mailing the inventory report to allaccounts and the time burden placed onrespondents, a stratified sample was used toobtain information on accounts that received largequantities of forms. This sampling method alsorandomly selected accounts in other sizecategories, as well as estimated the total amountof surplus for each of the 10 products. Over time,the sampling process was refined to reduce thesampling error, by increasing both the sample sizeand the response rate. In 1993, 6,414 inventoryreports were mailed and 2,744 were returned withusable data. By 1997, the inventory report wasmailed to 10,088 accounts and usable data wereobtained from 7,292 of them.

Implementation of Data-Driven OperationalDecisions

The process of integrating data into BPOLoperational decisions has evolved since 1991. Atthe onset, much of the data was either unavailableor unreliable; however, the quantity and quality ofdata for making operational and managerialdecisions have increased greatly over time.

Development of Recommended Amounts

Statistical methods first were introduced to theBPOL Program to develop recommendeddelivery amounts for each account for the 1992filing season. Analysis of historical operationaldata suggested the number of forms distributedto BPOL accounts could be reduced. The ratioof number of Forms 1040 available in BPOLoutlets to total number of individual tax returnsfiled was developed as a measure of abundanceof forms in an area. This ratio was computed foreach county and metropolitan statistical area inthe country. Accounts in those counties andmetropolitan statistical areas whose ratioexceeded the median by a given amount had anadjustment factor applied to the total amount offorms received the previous year. For example,the adjustment factor initially was applied to allaccounts in counties and metropolitan statisticalareas whose ratio was more than 1.5 times themedian ratio. This methodology was appliedincrementally over time, reaching fullimplementation in the 1996 filing season whenthe adjustment factor was applied to all accountsin counties and metropolitan statistical areaswhose ratio was above the median. The use ofthis type of adjustment factor resulted inrecommended amounts that were less than orequal to the total amount of forms received byan account the previous year. Depending onthe year, the percentage of accounts whoserecommendation was reduced by this processranged from less than 1 percent to over 28percent.

The BPOL Program further modified the processof developing recommended amounts in the1994 filing season, by incorporating data fromthe inventory report. Prior to that, BPOLaccounts whose amounts were not adjusted bythe statistical ratio method (described in thepreceding paragraph) received the samerecommended amount as the previous year (i.e.,initial amounts plus any resupply amounts). Thiswas because no information existed to indicate



70

whether an account ran high or low on forms. Forthose accounts that returned an inventory report,the number of forms and instructions sent couldbe reduced or increased in response to reportedsurpluses and shortages. The use of this method,in conjunction with the Form 1040 ratio measures,generated annual savings of approximately$750,000 over the prior method of developingrecommended amounts.

Computation of Plan Ratios

The BPOL Program used data from theinventory report to compute plan ratios asanother way to improve efficiency. Plans aregroups of the 10 standard items packagedtogether in specific quantities. Prior to the 1994filing season, the number and mix of items in agiven plan were not based on quantitativeanalysis, but on past experience. Beginning inthe 1994 filing season, statistical models weredeveloped based on data from inventory reports.

Given the inventory report records the quantityof standard items that remain after the filingseason, product usage was obtained bysubtracting amounts remaining at the end of thefiling season from total amounts sent to eachaccount. Usage trends were developed andadjustments were made to each of the plansizes to better reflect the observed usagepatterns. By creating plans that more accuratelyreflected accounts’ needs, the BPOL Programreduced both surpluses and shortages. Thedata analysis provided a sound method todetermine product quantity in a plan.

Recommendation of Resupply Amounts

Once instituted, the dataset from the inventoryreport was used in a variety of ways -- some ofwhich had not been envisioned at the onset. Forexample, matching inventory report informationwith ordering data revealed BPOL accounts thatreordered forms throughout the filing seasonwere, in fact, more likely to have surplus formsleft at the end of the filing season. Thisdiscovery led to the development of suggestedresupply amounts, which decreased inaccordance with ordering trends as the filingseason neared end. Institution of these tables ofresupply amounts helped reduce the number offorms remaining at BPOL outlets at the end ofthe filing season.

Modification of Number of Outlets

The efforts described thus far concentrate on theuse of quantitative methods to modify thenumber of forms available at BPOL outlets.Statistical methods also were used to identifycounties either in need of additional BPOLoutlets or counties saturated with them. TheBPOL Program used a variety of measures anddata sources to identify such counties. Forexample, data from the three IRS DistributionCenters were analyzed to determine whichcounties had a high number of orders for taxforms on a per capita basis. In addition, ratios ofthe number of BPOL outlets to number of taxreturns and the number of forms available inBPOL outlets to the number of tax returns werecomputed for each county. The BPOL Programthen focused recruitment efforts for outlets incounties with values in the bottom 10 percent forsuch measures. All but 18 of the nation’s 3,140counties had at least one BPOL outlet for the1997 filing season.

Counties with an abundance of BPOL outletsexperienced limited additional growth. Forexample, in the 1997 filing season a county wasidentified as saturated if it had more than 25outlets and had more than 4.8 outlets per 100square miles. These values were derived fromstatistical analysis of the data, and representedvalues that were more than 1.5 interquartileranges above the 75th percentile (which is acommonly accepted statistical measure ofextreme values). Approximately 200 of thenation’s 3,140 counties fit this definition ofsaturation for the 1997 filing season.

Removal of Banks as Distribution Outlets

Removal of most banks as distribution outletswas perhaps the most visible result of the BPOLProgram analyses. Analysis of the source codedata from the ETE Study and the analysis ofBPOL data over several years led to theconclusion banks were the most ineffective ofthe three major outlet types. With the IRSCommissioner’s approval, in 1996 the BPOLProgram eliminated most banks as distributorsof tax forms and information. Statistical analysisdemonstrated no adverse effect on taxpayers orthe IRS as a result of this action. There werefewer BPOL outlets and fewer forms sent toBPOL outlets, yet more returns filed with BPOLsource codes (refer to Tables 1, 2, and 3).Furthermore, there was no appreciable increase



71

in the number of forms obtained at IRS officesnor in the number of orders placed into IRSdistribution centers. The BPOL Programsuccessfully moved taxpayers from banks topost offices and libraries with little negativeimpact.

Organizational Changes/External FactorsAffecting the BPOL Program

In spite of statisticians’ best efforts to controlexternal variables, an operational program can notexist in a completely controlled environment.Over the years there has been a variety ofuncontrolled factors, both external and internal tothe BPOL Program, that has affected the ability tomake and measure changes to the program.Budgetary restrictions limited the initial shipmentof forms to the BPOL accounts for the 1997 filingseason. Moreover, accounts were unable tochange the recommended amounts, and ordermore forms. When the budgetary restrictionswere lifted later in the 1997 filing season, reorderamounts were extremely high, partly as a result ofover-reaction to the initial restrictions.

At the same time the BPOL Program was strivingto increase efficiency, additional demand wasplaced on it by the decreased availability of formsin other places. The number of IRS officesdistributing tax forms decreased from 625 for the1991 filing season to 507 for the 1997 filingseason. As a result of the budgetary restrictionsfor the 1997 filing season, 46 million taxpayersthat filed a practitioner-prepared return theprevious year did not receive a postcard or taxpackage from the IRS. For that same filingseason, approximately 22 million taxpayersreceived 1040EZ Telefile packages that did notcontain the traditional paper Form 1040EZ andInstructions 1040EZ. (Since only 5 million filers ofForm 1040EZ used Telefile, we estimate 17million obtained paper forms elsewhere.) Theexclusion of these forms from tax packages addedover 3 million taxpayers to the BPOL Program.

Conclusions

IRS’s experiences with the BPOL Program since1991 demonstrate that the application ofscientific and statistical methods to an

operational program provides opportunities formore effective management, which can lead to amore efficient program. However, suchintegration requires commitment frommanagement and workers. Resources must bededicated to increasing both the quantity andquality of data available for statistically baseddecision-making. Conflicts sometimes occurbetween the allocation of resources for“operations” versus “research.” In reality thisdichotomy is artificial because the purpose of“research” is to improve the efficiency of theoperational program. Above all, it is important tomeasure the results of new methods andcompare them to previous ways of doingbusiness. In addition, efforts must be made tolimit the impact of external forces in order tomeasure results in a meaningful and objectivemanner.


Denise York Young worked as a MathematicalStatistician in the Tax Forms Marketing AnalysisProgram of the IRS's Multimedia ProductionDivision from 1990 to 1998. Currently, she isAssistant Director of Strategic Planning andAnalysis at the University of Texas at Dallas.She received her M.S. degree in AppliedStatistics and Animal Genetics from NorthCarolina State University and is enrolled in thePh.D. program in Higher Education andEducational Research at the University of NorthTexas.

Erika Alexander is a Mathematical Statisticianin the Tax Forms Marketing Analysis Program ofthe IRS's Multimedia Production Division. Shereceived her M.S. degree in Statistics in 1991from Texas A&M University and is currently adoctoral candidate at the University of NorthTexas in Educational Research and ComputerEducation/Cognitive Systems. She has beenwith the IRS since 1992.

Table 1 Historical Data for Form 1040 (in Millions)

Filing Amount Amount Number FiledSeason Sent Remaining with BPOL

Source Code

1991 96.5 n.a. 4.71992 101.1 n.a. 5.11993 91.8 14.8 5.01994 86.4 14.2 5.21995 81.0 12.0 5.31996 67.6 8.2 6.01997 73.3 7.6 6.4

Table 2 Historical Data for Form 1040A (in Millions)


Source Code1991 54.8 n.a. 3.01992 58.3 n.a. 4.01993 71.3 11.0 3.51994 81.2 16.8 3.61995 75.7 15.0 4.01996 62.5 10.0 3.81997 66.5 9.3 3.9

Table 3 Historical Data for Form 1040EZ (in Millions)


Source Code1991 53.5 n.a. 4.01992 57.1 n.a. 4.81993 70.4 11.5 4.51994 80.5 17.0 5.21995 74.7 15.4 5.01996 60.8 9.7 5.11997 65.6 6.9 7.8


73

Using Data Reduction Techniques to Analyze Baseline Profiles

By Larry May andAnne Steuer

The Baselining National Compliance Measures project generated individual tables consisting of elevencompliance measures across three years for each of 71 broad taxpayer market segments for 33 IRS districtoffices. Analyzing this copious output presented a considerable challenge. Data reduction techniques wereexplored as an alternative to commonly used ad hoc methods of interpretation. Factor analysis together withcluster analysis was used to identify associations between the eleven measures and among the broadmarket segments. Preliminary results demonstrate that these various measures can be reduced into thethree broad underlying compliance dimensions --market size, balance due, and accuracy. Further, marketsegments can be clustered into relatively homogenous groups with respect to these broad compliancedimensions and “compliance profiles” generated for each cluster. These results can be used to identify andtarget market segments with similar compliance characteristics.

Introduction

An expectation of the fiscal year (FY) 1997Research Plan was to identify major non-compliant market segments24 through baseliningaccuracy, timeliness, and payment of National,Regional, and District populations. The projectteam considered several sources in determiningwhich market segments and measures to use.These included, but were not limited to, thesegments and measures used in the FY 1996Research Plan baseline objective, feedbackfrom the members of the Market Segmentationand Profiling Cooperative Strategy WorkingGroup, District Office Research and Analysis(DORA) Chiefs and their staffs, the FY 1997Annual Compliance Plan and the FY 1997Research Plan. The source of data for thisproject was a subset of the full 1040 FilersModel of the Compliance Research InformationSystem (CRIS) known as CRIS-Lite. Twenty-sixmarket categories were identified in the originalplan.

These 26 categories further were subdividedinto 71 market segments for evaluation,including the total population for comparisonpurposes. District baseline tables were

generated, presenting eleven compliancemeasures for each of the 71 market segments.

24 Market segments are groups of taxpayers with some commoncharacteristic(s) such as all taxpayers claiming a refund, allbusinesses in the same industry, or all taxpayers with a ScheduleA, for example.

With 33 districts, there were a total of 2,343tables that needed to be analyzed for a nationalperspective. Analyzing this copious outputpresented a considerable challenge. Datareduction techniques were explored as analternative to commonly used ad hoc methods ofinterpretation. Factor analysis together withcluster analysis was used to identifyassociations between the eleven measures andamong the broad market segments.

Data Reduction Methodology

The eleven measures and their originalgroupings are listed below:

Market Size• Estimated Population

Timely Payment• Total Unpaid Tax at Time of Filing• Average Unpaid Tax at Time of

Filing• Percent Dollars Unpaid at Time of

Filing• Percent Returns Unpaid at Time of

Filing Timely Filing

• Percent Returns Late Tax Accuracy

• Total Taxes Reported• Total Predicted Tax Increase (PTI)• Average Predicted Tax Increase• Voluntary Compliance Level (VCL)• Estimated Percentage of Returns

Accurately Filed

USING DATA REDUCTION TECHNIQUES TO ANALYZE BASELINE PROFILES


74

Some of the measures assess the samecompliance aspect; for example, Percent ofDollars Unpaid at Time of Filing and Percent ofReturns Unpaid at Time of Filing. Othermeasures even have a direct mathematicalrelationship: Total Unpaid Tax at Time of Filingand Average Unpaid Tax at Time of Filing (theaverage is simply the total divided by thepopulation). All four of these measures areslightly different evaluations of the sameattribute: timely payment.

Relationships between some measures andattributes, or characteristics, are not as straightforward. For example, individuals who owetaxes may have a greater tendency to submit aless accurate return in an attempt to reduce theirbalance due. Accordingly, there could be arelationship between the four “payment”measures and the characteristic of “accuracy”.

In a traditional interpretation of these elevenmeasures two barriers are presented. First, howshould we compare two market segments wherethe various measures are presenting a mixedmessage — market segment A has a higherPercent of Returns Unpaid at Time of Filingwhile market segment B has a higher Percent ofDollars Unpaid at Time of Filing. Second, it isdifficult and subjective to infer the indirectrelationships. Exactly how much of an influencedoes Average Unpaid Tax at Time of Filing haveon accuracy? Factor analysis overcomes bothof these barriers. Factor analysis createscharacteristic scores that incorporate theinfluence of all the measures including the weakand indirect relationships.

Factor Analysis

Description and Methodology

Factor analysis explores the relationshipsbetween all the measures and attempts toidentify the underlying characteristics. Havingidentified them, it quantifies the associationbetween each measure and each characteristic.Measures with strong associations to acharacteristic contribute more to determining thefinal characteristic score; this is referred to asloading high. Other measures with weakassociations load low. Each measure caninfluence (positively or negatively) eachindividual characteristic. Usually a measure willload high on one particular characteristic and

load low to moderate on the othercharacteristics.

In our factor analysis each market segment wasregarded as an observation with elevenassociated measures. A principle componentsfactor analysis was run to identify underlyingcharacteristics that explain the correlationsamong the set of measures. Its purpose is tosummarize a large number of variables ormeasures with a smaller number ofcharacteristics while preserving as much of thetotal sampling variation, and thus originalinformation, as possible. For this analysis, eachindividual baseline table is regarded as amultivariate observation. There are 71 tablesper DORA; thus with 33 DORAs there are a totalof 2,343 tables nationally. Each table haseleven measures that jointly describecompliance behavior for that particular marketsegment.

Refer to Appendix A for a more completedescription of factor analysis, as it is used in thiswork.

Results

Initially the correlation matrix of the elevenmeasures was examined to assess thesuitability of the data for factor analysis. Of 55correlations, 23 are greater than 0.30 suggestingfactor analysis is appropriate for these data.25

Further, the Barlett Test of Sphericity issignificant at a level less than 0.000126. TheKaiser-Meyer-Olkin Overall Measure ofSampling Adequacy is 0.71034, which is withinthe acceptable range.27 The measures ofsampling adequacy for all variables except VCL

25 Hair, J.F., Anderson, R.E., Tatham, R.L. and Black, W.C. 1995.Multivariate Data Analysis, 4th edition. Prentice-Hall, Inc. Theauthors suggest the initial general rule of looking for a substantialnumber of correlations greater than 0.30 for applying factoranalysis.26 This test is used to access the overall significance of thecorrelation matrix -- that is the off diagonals are nonzero.Statistically significant results indicate that the null hypothesis ofzero correlations can be rejected. The results of this test indicatethat the non-zero correlations in the correlation matrix are mostlikely not due to random chance. This test, like most statisticaltests is sensitive to sample size. The larger the sample size, thegreater the ability of the test to detect smaller departures from zero.Thus it is also important to examine the magnitudes of thecorrelations.27 The Kaiser-Meyer-Olkin Measure of Sampling Adequacy is anindex ranging from zero to one. Values under 0.50 are consideredunacceptable. A value of one indicates that each variable can bepredicted perfectly using the other variables.



75

are greater than 0.50. Thus, only VCL falls inthe unacceptable range, suggesting thismeasure has little association or relationship tothe remaining ten measures. VCL was omittedfrom the data set and the correlation matricesand associated tests recomputed.

Under the revised data set, 18 of 45 correlationsare greater than 0.30 and the Kaiser-Meyer-Olkin overall measure of sampling adequacy isslightly larger. As before, the Barlett Test ofSphericity is significant at a level less than0.0001. All the variables in the reduced data setmeet the individual measure of samplingadequacy threshold. Based on these results,the reduced data set was used for the factoranalysis.28

Table 1 gives the results for the rotatedsolution.29 The columns listed -- Factor 1,Factor 2, and Factor 3 -- give the factor loadingsfor each variable and the communality columnprovides a measure of the degree to which eachmeasure is “explained” using the three factors.30

All the measures have a high communality withthe exception of Percent Returns Late. For thismeasure, the three factors account for onlyapproximately a third of its variability. The sumof squares and percentage of trace indicate therelative importance of each factor in accountingfor the variability of the eleven measures.

Four measures, Estimated Population, TotalPTI, Total Tax Due and Total Tax DollarsUnpaid at filing load highly for Factor 1 and allother measures have relatively low loads. Thissuggests Factor 1 characterizes marketsegment size. The four measures, Average TaxDollars Unpaid at filing, Percent Tax DollarsUnpaid at filing, Percent Returns Late, andPercent Returns Unpaid load highly for Factor 2.This suggests Factor 2 characterizes payment.The two measures, Average PTI and PercentReturns Accurately Filed load highly and inreverse directions for Factor 3. Average PTIloads positively and Percent Returns AccuratelyFiled loads negatively. This suggests Factor 3characterizes accuracy. Note, Average Tax

28 For computational details see Internal Revenue Service NorthCentral Office of Research and Analysis, 1997. Baselining CRIS-Lite RP97-1.02 Volume 2 Appendix B of the National Report.29 Ibid.30 The factor loadings give the correlation between the variablesand the factors. Communality ranges from 0 to 1 with 0 indicatingno variation is explained by the factors to 1 where all the variationis explained by the factors.

Dollars Unpaid at filing loads moderately on boththe payment and accuracy factors. Factor 1(market segment size) accounts for 34.4 percentof total variability, Factor 2 (payment) accountsfor 24.8 percent, and Factor 3 (accuracy)accounts for 21.2 percent.

Cluster Analysis

Description and Methodology

The other data reduction technique consideredwas cluster analysis. With limited resources, theIRS must attempt to address issues of non-compliance with the broadest possiblewholesale approach. If Retail Food andBeverage has the same compliance levels asHotels and Lodging it may be more appropriateto view these segments as sub-components of alarger market segment. Accordingly, the 71market segments were subjected to clusteranalysis in an attempt to identify marketsegments with common compliancecharacteristics.

Cluster analysis is a mathematical technique inwhich the difference between objects, relative tosome attribute or set of attributes, is quantified.Similar objects are grouped together to form acollective object, or cluster. A successful clusteranalysis will take a large number of observationsand classify them into meaningful groups withminimal loss of information.

As with factor analysis, one of the primaryobjectives of cluster analysis is the reduction ofdata to aid in the interpretation of results. Theterm cluster analysis actually refers to a numberof different techniques that all attempt to classifyobservations according to their commonrelationships. These techniques primarily differin the way they measure similarity or differenceand how they group the objects together.

After identifying the underlying characteristicsfrom the eleven measures, focus was directedtowards the 71 market segments that made upthe baseline study. The motivation forsegregating the population into these marketsegments was to answer specific questionsposed during the planning phase. The basis formany of these questions was exploratory innature. Also, the composition of the marketsegments was not mutually exclusive; ataxpayer may belong to two or more segments.



76

For example, a taxpayer could be a member ofthe Paper Filing market segment as well as amember of the Self Prepared market segment.With regard to compliance characteristics, were

these market segments truly different or werethey different “views” of substantially the samemarket?

Table 1. Rotated Component Analysis of Three Factors

Factor 1 Factor 2 Factor 3Variables Size Payment Accuracy Communality

X1 Estimated Population. 0.84 -0.19 -0.38 0.89X8 Total PTI 0.95 -0.16 -0.14 0.94X9 Total Tax Due 0.95 -0.18 -0.10 0.93X10 Total Tax $ Unpaid 0.86 0.31 -0.01 0.83

X2 Average Tax $ Unpaid 0.12 0.69 0.54 0.78X3 Percent Tax $ Unpaid 0.08 0.91 0.13 0.85X4 Percent Returns Late -0.27 0.51 -0.04 0.33X7 Percent Returns Unpaid -0.09 0.85 0.11 0.73

X5 Pct, Rtns. Accurately Filed 0.27 -0.12 -0.87 0.83X6 Average PTI -0.14 0.08 0.93 0.90

TotalSum of Squares(eigenvalues)

3.44 2.48 2.12 8.04

Percentage of Trace* 34.4 24.8 21.2 80.4Trace = 8.03747 (sum of the eignvalues)

Refer to Appendix B for further description ofcluster analysis.

Results

Cluster membership of each market segmentwas obtained for cluster solutions from 2clusters through 20 clusters. At each stage, anexisting cluster is segmented into two groups. 31

In some cases the new clusters have the samecompliance attributes as the “parent” cluster (acompliant cluster yielding two compliant clustersor a non-compliant cluster resulting in two non-compliant clusters). In these situations, thecluster was probably split due to the market sizecharacteristic. In other cases non-compliancewas further distinguished because the twoemerging groups are different: one compliantand one non-compliant. For selected clustersolutions, additional inquiry was conducted,

31 Since our objective was to identify the cluster solution with theminimum number of groups, our analysis evaluated the changes incluster groupings from the 2-cluster solution upwards to the 20-cluster solution — a divisive analysis. This is in contrast to theway the clusters were actually built; from 20 down to 2 — anagglomerative procedure.

including a cross-tabulation of clustermembership and market segment. The cross-tabulation identified which market segments areassociated with which clusters and permitted theassessment of optimization criteria two andthree.

The 17-cluster solution presented seven non-compliant groups: clusters 10, 11, 13, 14, 15, 16& 17. The population of the market segmentsincluded in cluster 14 was very small and thiscluster was ignored. This was deemed to bethe optimum cluster solution based on thecriteria previously defined.

Figure 1 is a condensed dendrogram of the 17-cluster solution. This shows the relativesimilarity of each of the seventeen clusters.Clusters 14 and 16 are the most similar, sincethey have the shortest horizontal lines prior tojoining together. These two clusters were theones separated when moving from the 16-cluster solution to the 17-cluster solution. Thisresulted in the small population marketsegments of cluster 14 being split from thebalance of cluster 16.



77

Figure 1: Condensed Dendrogram using Ward's Method –– 17 Clusters Shown

Rescaled Distance Cluster Combine Record 0 5 10 15 20 25Cluster Count --------------------------------------------------- 11 127 -----+ +---+ 12 260 -----+ ¦ +---+ 8 206 -----+ ¦ +---------+ +---+ ¦ ¦ 13 95 -----+ ¦ ¦ ¦ ¦ 15 34 ---+ ¦ ¦ +---------+ ¦ 10 119 ---+ ¦ ¦ 6 193 -----+ +-------------------------+ +-----+ ¦ ¦ 9 347 -----+ ¦ ¦ ¦ +-----------+ ¦ 7 256 -------+ ¦ ¦ +---+ ¦ 16 30 ---+ ¦ ¦ +---+ ¦ 14 33 ---+ ¦ ¦ 4 111 -------+ ¦ +-----------------------+ ¦ 5 137 ---+ ¦ ¦ ¦ +---+ +-----------------+ 1 158 ---+ ¦ ¦ 17 34 ---------------------+ ¦ ¦ ¦ 2 47 -----+ +---------+ +---------------+ 3 134 -----+

Table 2 provides summary information for thesix most non-compliant clusters of the 17-clustersolution. Using the types of market segmentsincluded in a cluster and the compliancecharacteristics summarized below, subjective

descriptive names can be assigned to each ofthe clusters. Although this typically is done aftercluster analysis, no attempt was made to labelthe clusters resulting from this analysis.

Table 2. Average standardized factor scores for the six most non-compliant clusters.

Standardized Factor ScoresNon-CompliantCluster Payment Accuracy Size17 6.27 -0.91 1.0816 2.64 -0.13 -0.5415 -0.73 2.31 1.2110 0.65 1.90 0.6311 -0.58 1.35 -0.3113 1.15 0.80 -0.25



78

Cluster 17 — Large markets with accuratereturns but totally unpaid balances.• Percent of returns unpaid at time of filing: 100%• Percent of dollars unpaid at time of filing: 32%• Average amount unpaid at time of filing: $1,987• Percent of returns accurately filed: 39%

Cluster 16 — Small markets with unpaidbalances.• Percent of returns unpaid at time of filing: 22%• Percent of dollars unpaid at time of filing: 38%

Cluster 15 — Large markets with inaccurate butfully paid returns.• Percent of returns unpaid at time of filing: 6%• Percent of dollars unpaid at time of filing: 1%• Average predicted tax change: $3,410• Percent of returns accurately filed: 18%

Cluster 10 — Medium to large markets withinaccurate and unpaid returns.• Percent of returns unpaid at time of filing: 18%• Average amount unpaid

at time of filing: $1,419• Average predicted tax change: $2,522• Percent of returns accurately filed: 17%

Cluster 11 — Small markets with inaccurate butpaid returns.• Percent of returns unpaid at time of filing: 9%• Percent of dollars unpaid at time of filing: 3%• Average amount unpaid at time of filing: $509• Average predicted tax change: $2,584• Percent of returns accurately filed: 12%

Cluster 13 — Small markets with inaccurateand unpaid returns.• Percent of returns unpaid at time of filing: 18%• Percent of dollars unpaid at time of filing: 16%• Average predicted tax change: $2,138• Percent of returns accurately filed: 11%

Discussion

The data reduction techniques presented in thispaper were successful at consolidating a lot ofdata into useful information. The original datawere represented by 11 measures for 71different market segments across 33 geographicareas. In total, these were over 25,000statistics. Factor and cluster analysis reducedthis to 3 characteristics across 17 groups — 51statistics. This becomes a much moreinterpretable set of data upon which to apply the

analysts’ wisdom and insight. The compliancecharacteristics of the six non-compliant groupseasily can be interpreted. With reference backto the original market segments, the analyst canformulate a comprehensive group comprised ofthe market segment intersections and overlaps.

Although the original categorization of thecompliance measures seems to make sense,the factor analysis found a better allocation.Originally the totals for each attribute werespread across the categories — total populationunder market size, total taxes unpaid undertimely payment, and total taxes reported alongwith total predicted tax increase under taxaccuracy. The factor analysis concluded thatthese measures of totals more accuratelyrepresented a single characteristic — marketsize.

The factor analysis also demonstrated thelimited usefulness of the VCL measure toidentify non-compliance in this setting. The VCLmeasure had relatively little variability and didnot associate or correlate with the other tenmeasures.

The factor analysis also demonstrated theindirect relationships of the measures. Forexample, from Table 1, it can be seen thataverage tax dollars unpaid has a moderateinfluence on the accuracy of a market segment:loading at 0.54. Conversely, the averagepredicted tax increase has relatively littleinfluence on the timely payment of taxes. Andlastly, while the percent of returns late has littlein common with the characteristics identified(communality of 0.33), it does have a moderateinfluence on the payment characteristic, loadingat 0.51.

The data reduction techniques used in thisanalysis can be applied to a wide variety ofsituations where the analyst is presented withlarge volumes of summary information thatneeds to be distilled into a smaller set of “keyinformation”.



79

Acknowledgments

The authors would like to thank Michael Gregoryof the North Central DORA and the entireBaselining National Compliance MeasuresProject Team. Additional thanks go to TeresaO’Hearn, our summer intern, for all herenthusiastic and helpful research assistance. Aspecial thanks to Jeff Butler for his review andinput.


Larry R. May is an operations research analystTeam Leader in the Pennsylvania District Officeof Research and Analysis. He received his B.S.degree in Business Administration (Accounting)in 1984 from Glassboro State College. He hasbeen with the IRS since 1983.

Anne Steuer is a statistician in the NorthCentral District Office of Research and Analysis.She received her Ph.D. degree in NaturalResource Economics and Policy in 1986 and herM.S. degree in Statistics in 1997 from theUniversity of Minnesota. She has been with theIRS since 1995.

Appendix A -- Description of Factor Analysis

Let the random variables X X X1 2 11, , ,Krepresent the eleven compliance measures andlet the random vector [ ]Χ T X X X= 1 2 11, , ,Khave covariance matrix ∑ with eigenvaluesλ λ λ1 2 11 0≥ ≥ ≥ ≥L . (Eigenvalues are

scalars λ λ λ1 2, , ,K k which are solutions to the

polynomial equation Α Ι− =λ 0 where Α is a k

x k matrix and Ιis the k x k identity matrix.Details are available in any linear algebra textsuch as: Graybill, F.A. 1969. Introduction toMatrices with Applications in Statistics, Belmont,California: Wadsworth.) The ith principlecomponent, or factor, is Υ Χi i

Te= where ei is theith eigenvector. (An eigenvector is a nonzerovector x such that for a k x k matrix Α andeigenvalue λ, Αx x= λ . Since eigenvectorsare, by convention, set to length one,

e xx xT

= and is the eigenvector associated

with the eigenvalue λ.)

Further the ( )Var e ei iT

i iΥ = ∑ = λ and since

( )Cov e ei j iT

jΥ Υ, = ∑ = 0 , the principle

components, or factors, are uncorrelated.

The eleven compliance measures are varied fromtotals, averages, and percentages of both countsand dollars. Because of the wide range ofmeasurement scales, the eleven measures werestandardized prior to the factor analysis. Therandom vector [ ]Χ T X X X= 1 2 11, , ,K was

transformed to ( ) ( )Ζ Χ= −−

V1

21

µ where V1

2 is an

11 x 11 matrix with standard deviations on thediagonal and zeros on the off-diagonals and µ isa vector of means. This transformation is the

familiar standardization formula zx= − µσ

written in matrix notation. Here σ is the standarddeviation and µ is the mean of the randomvariable x.

The random vector [ ]Ζ Ζ Ζ ΖT = 1 2 11, , ,K has

mean ( )E Ζ = 0 and ( ) ( ) ( )Cov V VΖ = ∑ =− −1

21

21 1

ρwhere ρ is the correlation matrix for the original

random vector [ ]Χ T X X X= 1 2 11, , ,K . Thus theprinciple components or factors for the elevenstandardized measures are derived from thecorrelation matrix for the unstandardizedmeasures. These factors are generally not thesame as the factors obtained using theuntransformed variables. Thus standardizationwill yield different results. See Johnson, R.A. andWichern, D.W. 1992. Applied MultivariateStatistical Analysis 3rd edition. Prentice-Hall, Inc.


81

Appendix B -- Description of Cluster Analysis

Without a predetermined number of groups, ahierarchical approach was used to classify thecases into groups.32 Specifically, Ward’s Methodclustering algorithm was used with the factorscores from the preceding factor analysisdescribing each market segment and Euclideandistance as the measure of similarity. Ward’smethod was selected since the distance measureis suitable for use with factor scores. Ward’smethod is biased towards clusters of similarsizes.33

The first step is to calculate the dissimilaritybetween two observations. This is calculated asthe Euclidean distance between the two objects inP dimensional space where P is the number ofcharacteristics being considered.

Although Euclidean distance is a commonmeasure of similarity in cluster analysis, it has twoweaknesses: the calculation is sensitive tomagnitudes of scale and correlation between thevariables34. In this baseline study, both of theseconcerns are alleviated as a result of the factoranalysis. This is because, as previouslydiscussed, factor scores are standardized; henceno scale problems. Furthermore, factor scores

32 Cluster analysis techniques can be broken down into two generalcategories; hierarchical procedures and nonhierarchical procedures.In hierarchical procedures the technique moves up through thegrouping of individual cases into clusters until all cases are in onegroup (agglomerative methods) or down through all cases in onegroup to all cases in separate groups (divisive methods). Thedecision at each stage of which clusters to combine or split is basedon the similarity measure. In nonhierarchical procedures a specifiednumber of clusters or maximum distance is established.Observations that meet the algorithm’s criteria are assigned to theclosest cluster. A critical step in the nonhierarchical procedures isthe specification of cluster centers. See Joseph F. Hair, Jr. et al.,Multivariate Data Analysis: with Readings -- 4th ed. (Upper SaddleRiver, New Jersey: Prentice-Hall) for a complete discussion of thecommon clustering techniques.33 Joseph F. Hair, Jr. et al., Multivariate Data Analysis: withReadings, 4th ed., (Upper Saddle River, New Jersey: Prentice-Hall),p. 440.34 Brian S. Everitt, Cluster Analysis , 3rd ed., (New York: HalstedPress), pp. 46-47.

are jointly uncorrelated; hence no collinearityproblem.

After the dissimilarity is quantified, the data areanalyzed to group similar cases. In Ward’sMethod this is done by evaluating the within-cluster variation versus the between-clustervariation. Figure B1 demonstrates this concept35

Figure B1: Within vs. Between Cluster Variation

In hierarchical methods of cluster analysis thegrouping of data is exhaustive. Clusteringcontinues until all the data is grouped into onecluster. For example the above diagram showsthree groups. At the next stage of the analysistwo of these groups would be joined leaving twoclusters, and in the final stage those two clusterswould be joined into one collective group. Whileindividual cases may be too much data tointerpret, a single cluster provides little informationas well. The optimum number of clusters issomewhere in between. Determining the optimumnumber of clusters is a matter of considerabledebate36. This remains a subject of deliberationbecause the context of the clustering applicationand the structure of the data have at least as

35 Adopted from Joseph F. Hair, Jr. et al., Multivariate Data Analysis:with Readings, 4th ed., (Upper Saddle River, New Jersey: Prentice-Hall), p. 437.36 See Milligan, G. & Cooper, M. 1985, "An Examination ofProcedures for Determining the Number of Clusters in a Data Set,"Psychometrika, Vol. 50, No. 2, pp. 159-179. andDubes, R. & Jain, A. (1979), "Validity Studies in ClusteringMethodologies," Pattern Recognition, Vol. 11, pp 235-254.

( )d X Xij jk ikk

p= −

=

∑ 2

1

Within-Cluster Variation

Between-Cluster Variation


82

much influence on the decision as mathematicaltheory and the technique applied.

In this analysis, traditional cluster output ofendrograms and agglomeration schedules werereviewed along with summary statistics for eachgroup at each stage of the clustering process.Since the objective of this analysis was to identifygroups of non-compliant market segments at theNational level, three subjective operational criteriaalso were considered in determining the finalnumber of clusters:

Optimization Criteria

1. A non-compliant cluster was definedas having a median factor score foreither the Payment or Accuracycharacteristic in excess of 0.8.

2. The collective number of marketsegments contained in the non-compliant clusters was approximately10.

3. A market segment was consideredpart of a cluster if at least half of thedistricts appeared in that cluster.

The cluster solution that minimized the overallnumber of clusters while meeting the three criteriaabove was deemed the optimum solution.

Review of the IRS’s Individual Return Electronic Filingand Related Research

By Javier Framinan

The Internal Revenue Service has profiled and studied individual income tax return filers in an effort to learn how toimprove and market its electronic filing products. This article reviews and summarizes that research, examining whofiles electronically, why they do, why they do not, the costs of electronic filing, and the results of various electronicfiling marketing initiatives. Overall, the various research efforts show electronic filers are young, have lower income,and have simple tax situations compared to the general individual filer population. In 1998, 20 percent of individualincome tax returns were filed electronically. The biggest motivation for filing electronically is a fast refund. Notsurprisingly, over 30 percent of taxpayers with refunds in the $1,000 to $3,000 range elect to file electronically. Over40 percent of taxpayers receiving the Earned Income Tax Credit file electronically. The IRS wants to maintain a currentknowledge base of the individual taxpayer market segment in order to hone its electronic filing marketing strategy.Also, as the IRS shifts some of its attention to the electronic filing of business returns, it will need to conduct similarresearch in the profiling and study of business entities for strategic planning purposes.

Introduction

The Internal Revenue Service recognizes thatmanual processing of paper-based interactionwith the public is a resource-intensive activity,prone to errors. (The “public” includes individualand business taxpayers, taxpayerrepresentatives, tax practitioners, and othergovernment entities.) Data entry error rates forpaper individual income tax returns areapproximately 20 percent, compared to about 2percent for electronically filed returns.37 The IRSalso recognizes taxpayer information (stored inIRS databases) will improve with electroniccommerce and communications. The IRScaptures 100 percent of the informationrecorded on electronically filed returns,compared with approximately 40 percent frompaper returns.38 More information and easieraccess to it, in turn, improves the IRS’s ability toserve the public. In addition, from the public’sperspective, electronic exchange offers easierways to file returns and pay tax liabilities,confirms the IRS’s receipt of returns, speeds uprefunds, and enables electronic retrieval of

37 Internal Revenue Service, Electronic Tax Administration.(1997). Critical Issues for Development of an Electronic TaxAdministration Strategy: A Plan for Moving Federal TaxAdministration into the Information Age. Washington, DC. &General Accounting Office. (1993). Opportunities to Increase theUse of Electronic Filing. Washington, DC.38 General Accounting Office. (1995). Electronic Filing FallingShort of Expectations. Washington, DC.

forms, publications and other information fromthe IRS (e.g., from the IRS Internet homepage).

This paper gives a brief historical review of theIRS’s electronic filing activities with individualincome taxpayers, but primarily serves to reviewthe research conducted to date. In general, thegoal of the research related to electronic filing,or e-file, has been to understand its marketsegments and increase participation. Thediscussion is limited to the electronic filing ofindividual income tax returns, as this is wherethe IRS has concentrated its efforts. In fact, thisreview of the literature confirms the limitedresearch done to date on the electronic filing ofbusiness returns, and points to a clear need formore IRS study in that area.

Background

In 1998, 20 percent of individual income taxreturns were filed electronically. However, inthat same year the Congress stated its intentthat by 2007 the IRS should conduct 80 percentof its interactions with the public electronically.39

This will require aggressive development of theIRS’s electronic commerce. The Office of theAssistant Commissioner (Electronic TaxAdministration), or ETA, is the lead IRSorganization

39 The IRS Restructuring and Reform Act of 1998.

REVIEW OF THE IRS’S INDIVIDUAL RETURN ELECTRONIC FILING AND RELATED RESEARCH


84

in attempting to meet this challenging 80 percentgoal. In December 1998, it submitted to theCongress its first installment on its strategic planto revolutionize how taxpayers transact andcommunicate with the IRS.40

The ETA’s strategic planning depends on anunderstanding of its customers -- who uses thecurrently available electronic commerceproducts and why, and who does not and why.Data-driven research helps the ETA understandthe technological, legal, and financial barriersfacing tax-related electronic commerce. It usesresearch results to identify programs that giveIRS customers incentive to shift from traditional(i.e., paper-based) interaction to electronic.

Individual Return Electronic Filing

The IRS started electronic filing of individualincome tax returns as a test in 1986, andexpanded it nationwide in 1990. Individualelectronic filing is one of the IRS’s mostdeveloped forms of electronic commerce.Consequently, most of the IRS’s electronic filingresearch relates to the individual income taxreturns. Individual e-file can be divided into twodistinct categories: “standard e-file” andTeleFile. Until 1994, standard e-file includedonly electronic returns prepared and filed for thetaxpayer by an authorized professional taxpractitioner, or returns prepared by taxpayersthemselves and taken to an approvedtransmitter who transformed the returninformation into the necessary electronic formatand transmitted the return electronically. (This“standard e-file” also was known as “ELF,” orELectronically Filed.) Presently, it also includesOn-Line filing, comprised of returns prepared bytaxpayers that use tax preparation software andtransmit on-line through an authorized electronicreturn filer. TeleFile returns include all thosetransmitted over the telephone using touch-tonetechnology. Table 1 presents historical annualvolumes of individual e-file, as well as its overallmarket penetration rates. Also, more detaileddescriptions of these categories are provided inthe gray box.

40 Internal Revenue Service, Electronic Tax Administration.(1998). A Strategy for Growth. Washington, DC.

Individual Return e-file Categories and Sub-categories

Electronic Return Originator/Practitioner e-filing

In 1986, before the proliferation of personal computers and modems,individual taxpayers could file electronically only if they went to aprofessional tax practitioner (also referred to as a preparer). The IRScoordinated with electronic return originators (EROs) and thepreparer community, and set technical and procedural standards forelectronic information exchange. Although taxpayers incurredadditional fees filing electronically, faster refunds provided incentivefor many to e-file. By 1992, after only two years of nationwideimplementation, almost 11 million taxpayers were filing electronicallythrough a preparer. The attraction of faster refunds gave rise to therefund anticipation loan (RAL) market. For an extra fee (in addition tothe return preparation and transmission charges), electronic filerscould secure a RAL, where in coordination with the preparer, a bankadvances the anticipated tax refund amount. In essence, this givesthe taxpayer an instant refund upon tax return transmittal.

On-Line Filing

On-Line filing has grown with the popularity of tax preparationsoftware. In 1996, in its second year of existence, 158,000 taxpayersparticipated in the On-Line filing program. By 1998, 942,000 werefiling using this method. The volume of On-Line filers is projected togrow to over 2 million in 1999.41

To file on-line, the taxpayer must have a computer, modem, and taxpreparation software from a certified private vendor. On-Line filingalso requires use of an IRS-accepted on-line service company ortransmitter. After completing a tax form electronically, the on-line filerpays a fee directly to a return transmitter company (or indirectly to thetransmitter, through the purchase of the tax preparation software) thattranslates the return information into an IRS readable format. On-Line filing basically has the same incentives and restrictions aspractitioner electronic filing – i.e., faster refunds, higher accuracy, IRSconfirmation of receipt, but extra cost.

Electronically Filed, but not Prepared

A small number of taxpayers take a hard copy of their returns to apractitioner or other transmitter to submit it electronically to the IRS.During filing year 1997, 2.1 percent of all individual tax returns werefiled this way.42 Again, the taxpayer motivation is an acceleratedrefund.

TeleFile

In 1996, the IRS offered TeleFile nationwide. TeleFile employstouch-tone telephone technology to transmit returns, using IRS-issued customer service numbers for authentication. The IRS haslimited TeleFile’s availability to filers of simple returns (i.e., Form1040EZ), as research shows taxpayers’ unwillingness to key enterinformation for longer returns on a touch-tone pad. This filing optioncaters to taxpayers unwilling to use a preparer and/or unwilling to paytransmission fees. Every year since 1996, the IRS has mailedTeleFile tax packages to approximately 25 million taxpayers identifiedas eligible. In 1998, almost 6.0 million taxpayers filed this way,making up 24 percent of the individual return e-file market.

41 Internal Revenue Service, Research and Statistics of Income.(1998). Calendar Year Projections of Individual Returns by MajorProcessing Categories Selected Years and Areas Fall 1998Update. Washington, DC.42 Internal Revenue Service, Southwest District Office Researchand Analysis. (1997). A National Profile of IRS e-file Users in1997. Phoenix, AZ.



85

Table 1. Electronically Filed Individual Returns: 1986 - 1998

* On-Line Filing volumes are a subset of “Standard e-file.”† Indicates first year of nationwide implementation.‡ The 1995 drop in standard e-file volume was due to the IRS Revenue Protection Strategy, instituted to combat refund fraud associated with electronic filing.

Who Files Individual Returns Electronically?

Initial Research

The IRS understood relatively little about itsindividual electronic filing market when it startedthe program in 1986. The Service dependedprimarily on the tax practitioner community topromote its growth. However, by the 1990s, theIRS recognized the need for research, not onlyfor marketing purposes, but also for resourceallocation planning. (With taxpayers switchingfrom paper to electronic filing, resourcesformerly committed to the manual processing ofreturns need to be reallocated.)

In 1991, the IRS Research Bulletin featured anarticle by Bryan Musselman that profiledindividual return electronic filers. The articledescribed electronic filers from the 1990 and1991 filing seasons in terms of age, education,income, geographic location, and form type.Since the IRS lacked internal data related toelectronic filers, the article relied heavily onsurvey data gathered by the RoperOrganization. Despite the reliance on externaldata, much of the article’s general profile and

conclusions remain valid today. The researchfound the typical individual electronic filer files arelatively simple return, is young, has a lowerincome, most likely lives in the southeast, and ismotivated by a quick refund. Barriers toelectronic filing are awareness and cost.

A Lack of Information Leads to e-file ResearchDatabase

The absence of an IRS database for profilingtaxpayers eligible to file electronically hinderedefforts to market e-filing during the early to mid1990s. IRS data to corroborate the 1991 Ropersurvey results were scant, and any annualchanges in the e-filer profile were difficult todetermine. These deficiencies did not gounnoticed. The General Accounting Office’s TaxAdministration, Electronic Filing Falling Short ofExpectations (1995) concluded the lack ofadequate data and inability to performcost/benefit analyses hampers IRS decision-makers, and contributes to their lack of strategicfocus. A year later, the ELF Profiling ProjectTeam’s Profile Report: Current and PotentialMarket Segments for Electronic Filing (1996)warned “further research and analysis of theelectronic filing program will be hindered by thelack of a timely database.”43 The reportrecommended development of a database toenhance the profiling of electronically filedreturns. By 1997, the IRS Southwest DistrictOffice Research and Analysis (DORA) wastasked with coordinating research in the“alternative ways of filing” area (i.e., electronicfiling). However, in its National Profile of IRS e-file Users in 1997 (1997), the DORA admitted toan as yet inadequate understanding of theindividual e-file market segments. Theresearchers cited the need to build an individuale-file database to enable the study of the marketsegment.

In 1997, the Southwest DORA took a steptoward filling the data void, by constructing the“national e-file research database,” whichcontained tax year (TY) 1996 individual incometax return data on characteristics such astaxpayer location, age, income, taxes,deductions and exemptions. The database was“intended as a prototype for an annual

43 Internal Revenue Service, ELF Profiling Project Team. (1996).Profile Report: Current and Potential Markets for ElectronicFiling. Washington, DC.

e-file as a

Percent of

Filing Standard On-Line Total All Individual

Year e-file TeleFile Filing* e-file Returns

1986 24,814 24,814 0.02%

1987 77,612 77,612 0.08%

1988 583,462 583,462 0.55%

1989 1,160,516 1,160,516 1.06%

1990 4,204,188† 4,204,188 3.74%

1991 7,567,116 Did Not Exist 7,567,116 6.65%

1992 10,919,281 125,981 11,045,262 9.63%

1993 12,333,750 148,585 Did Not Exist 12,482,335 10.97%

1994 13,502,055 518,693 285 14,020,748 12.23%

1995 11,126,885‡ 680,010 1,372† 11,806,895 10.17%

1996 12,128,969 2,839,437† 158,284 14,968,406 12.65%

1997 14,449,712 4,685,959 366,727 19,135,671 15.90%

1998 18,625,689 5,954,564 942,176 24,580,253 20.06%



86

construction of similar databases.”44 The reportdescribes the database development andsources. It also considers research questionsposed by the ETA, and profiles the practitioner,On-Line, and TeleFile market segments, as wellas the paper return filers for comparisonpurposes. More recently, the Southwest DORAupdated the national e-file research database,and renamed it the “ETA Market ResearchDatabase” (see On-Line Filers: ETA MarketResearch for 1998 (1999)).

Profiles and Other Facts Regarding IndividualTaxpayers

The IRS has conducted several studies, andcontracted private vendors to conduct severalmore, to understand its e-file markets. Much ofthis research has been of a profiling nature –i.e., it describes the demographic characteristicsof taxpayers who e-file, as well as eligibletaxpayers who do not. The findings vary little,even across time. Following are various e-filerdemographics as reported in two major profilingreports: Profile Report: Current and PotentialMarket Segments for Electronic Filing (1996)and A National Profile of IRS e-file Users in1997 (1997).

Profile Report: Current and Potential MarketSegments for Electronic Filing

In 1995, prior to the establishment of the ETA,the IRS’s electronic filing executive requestedthe IRS Director, Compliance Research profileelectronic filing market segments. Despite theabsence of a database specifically designed tostudy the e-file market, the National OfficeResearch and Analysis (NORA) and the ELFProfiling Project Team (consisting of IRSNational Office and District Officerepresentatives) under Compliance Researchconducted a national profile of taxpayers filing in1994 and 1995. Meanwhile, the IRS’s DistrictOffices Research and Analysis (DORAs)produced similar local profiles. The ProfileReport: Current and Potential Market Segmentsfor Electronic Filing (1996) reported findingsfrom both efforts, predominantly on returns filedin 1994. This research profiled all taxpayerseligible to file electronically, including those thatdid and those that did not, using TY 1993

44 Internal Revenue Service, Southwest District Office Researchand Analysis. (1997). A National Profile of IRS e-file Users in1997. Phoenix, AZ.

sample data from the interim ComplianceResearch Information System (CRIS) file anddata from the Automated Wage Information File(Autowif). The analysis excluded TeleFile, sinceit still was not available nationwide at the time.

The resulting report essentially confirmedMusselman’s 1991 research in profiling theelectronic taxpayer. Electronic filers of individualincome tax returns are predominantly younger,lower income, simpler return, and motivated by afaster refund. The report cited 98.4 percent ofTY 1993 individual return filers were eligible tofile electronically; however, only 12.2 percentactually did.

Tables 2 through 5 present selected statisticsfrom the Profile Report: Current and PotentialMarket Segments for Electronic Filing (1996).They reflect returns filed in 1994 (for TY 1993)and show participation among eligible taxpayersby category. In Table 2, for instance, of all taxreturns eligible for e-file with adjusted grossincome (AGI) less than $13,000, 14.3 percentactually were e-filed.

Table 2. e-file Penetration by Adjusted Gross Income

Adjusted GrossIncome

Participation Rateof Eligibles

< $13,000 14.3%$13,001 - $26,000 16.3%$26,001 - $39,000 10.8%$39,001 - $52,000 9.2%> $52,001 5.1%

Table 3. e-file Penetration by Age

Age Participation Rateof Eligibles

< 16 0.6%16 – 20 8.1%21 - 24 18.8%25 – 44 17.2%45 – 64 7.9%> 64 2.2%



87

As Tables 2 and 3 show, e-file participationdrops dramatically among taxpayers with AGIabove $26,000 and older than 44 years of age.

Table 4 presents information contained in thereport regarding return complexity and itsrelationship to e-file participation. In general, e-filers are characterized by simpler returns.Stated differently, e-filing does not attracttaxpayers with more complex income taxsituations.

Table 4. e-file Penetration by Return Complexity

Return Schedules Participation Rateof Eligibles

No Schedules 15.3%Schedule A 7.2%Schedule C 5.7%Schedule F 3.9%More than OneSchedule

4.5%

The report also highlights various othercharacteristics regarding e-file, some of whichare contained in Table 5. Not surprisingly, only0.7 percent of eligible TY 1993 balance duereturns were filed electronically. Also notsurprisingly, of all taxpayers receiving a refundof less than $300, only 2.8 percent filedelectronically. The relatively high costs ofelectronic filing dissuaded these taxpayers, asthe net refund amounts after payment of e-filingfees would be small. Conversely, of thosetaxpayers receiving a $1,001 to $3,000 refund,31.4 percent apparently desired a fast refundand could justify the fees. The profile of theearned income tax credit (EITC) recipients tellsthe same story. (The earned income tax creditis a Federal government credit provided to lowerincome individuals. It is not related to incometax, but is deducted from income tax liabilitiesand disbursed through income tax refunds.)Over 42 percent of the taxpayers receiving theEITC e-filed to get their money faster. Thelarger check from the government apparentlyoffsets the costs associated with e-filing.

Table 5. Miscellaneous e-file Penetration Characteristics

Characteristic Participation Rateof Eligibles

Refund Return 17.2%Balance Due Return 0.7%

Paid Prepared Return 18.0%Non-Paid Prepared Return 6.6%

Refund < $300 2.8%Refund $300 - $1,000 16.2%Refund $1,001 - $3,000 31.4%Refund > $3,000 16.6%

Form 1040 – Type Return 12.3%Form 1040A – TypeReturn

19.0%

Form 1040 EZ – TypeReturn

6.6%

Single Filing Status Return 7.0%Joint Filing Status Return 9.4%Head of Household FilingStatus Return

38.3%

Return with EarnedIncome Tax Credit

42.1%

Return without EarnedIncome Tax Credit

7.6%

A National Profile of IRS e-file Users in 1997

The IRS’s Southwest District Office Researchand Analysis conducted further individual e-filemarket profiling on TY 1995 and 1996 returnsand produced A National Profile of IRS e-fileUsers in 1997 (1997). The report confirmedmost of the e-filer profile information presentedin earlier research, but contains more in terms ofdemographic detail and distinction between“standard e-file,” On-Line filing, and TeleFile.

According to the 1997 report, TeleFilers areyoung, single, have low income, and have lowrefund amounts (compared to the average for allindividual returns filed). Non On-Line standarde-filers (i.e., originating from EROs/practitioners)are a little older, have higher incomes andrefunds, and consist of more EITC recipients.On-Line filers have even higher income, higherrefunds, and more complex returns.



88

Following are some of the 1997 report’s generalfindings, by demographic characteristic.

Age: E-filers are predominantly younger. Thelargest age group of TeleFilers wasconcentrated in the 18 to 29 year old range,while the majority of standard e-filers was in the30 to 44 year old range. Conversely, paperreturn filers are more evenly distributed by age.

Income: E-filers’ average income is lower thanthat of the entire taxpayer population’s. MoreTeleFilers and non On-Line filers fell in the lessthan $15,000 adjusted gross income categorythan in any other AGI group defined in the study.However, On-Line filers have higher AGIs.Forty-five percent of On-Line filers had an AGI of$50,000 or more.

Refunds: Standard e-file returns (not includingOn-Line) had an average refund of over $1,827in 1997, while TeleFile returns averaged $400.The average refund for all individual returns filedthat year was $627. The researchers attributethe high refunds for non On-Line standard e-filereturns, at least in part, to the earned income taxcredit from the government. The e-file returnsoriginating from tax preparers werecharacterized by lower income, yet higherrefunds due to the EITC payments. Almost 49percent of the non On-Line standard e-filereturns claimed EITC. The average EITCpayment for all non On-Line standard e-filereturns (including returns not claiming EITC)was $880, compared to On-Line filers’ (the nextclosest group) $214. The combination of lowincome and high refund amount apparentlycreates the demand for quick refunds,regardless of the extra preparation and filingexpense. This phenomenon holds true for thehigher income market segment characterizingthe On-Line filers, as well. The On-Line FilingProgram: Focus Group Report (1997) by the IRSOffice of Opinion Research concluded sometaxpayers more readily pay the extra cost of e-filing if they expect a large refund check.

Professionally Prepared: Almost 49 percent ofall individual income tax returns filed in 1997were prepared professionally. Professionalsprepared 81 percent of all electronically filedreturns, compared to 42 percent of all paperreturns.

Return Complexity: On-Line filed returns were

the most complex of all returns (paper andelectronic combined). The researchers used thenumber of schedules attached as an indicator ofcomplexity. They found 80.2 percent of On-Linereturns had at least one schedule attached.Further IRS success with On-Line filing wouldaddress concerns noted by the U.S. GeneralAccounting Office (GAO). In 1995 the GAOreported only about 20 percent of the e-filereturns in 1994 were Form 1040-type -- i.e.,more complex -- even though 59 percent of allindividual returns filed in 1994 were Forms 1040.The GAO report suggests IRS had focused onan e-file return volume goal, rather than anoperating cost reduction one. It recommendsthe marketing strategy consider targetingtaxpayers with more complex returns that aremore expensive for the IRS to process manually.

Repeat Rates: In general, taxpayers remainloyal to their filing method from year to year.Paper filers had the highest repeat rate, at 90percent, while On-Line filers showed the lowest,with 55 percent. (On-Line Filers: ETA MarketResearch for 1998 (1999) subsequently profiledOn-Line filers for TY 1996 and 1997. It foundoverall the repeater rate rose to 59 percent forthe period from TY 1996 to 1997 filings.) Fornon-On-Line standard e-file returns the repeaterrate is 74 percent. For TeleFile, the overallrepeater rate is 56 percent. Many TeleFileusers, however, can not repeat the followingyear due to TeleFile restrictions on income level,complexity of returns, change of address, age,and dependents.

Regional Differences: There are some regionaldifferences with respect to electronic filing. Theprofile confirmed earlier studies that thesoutheast experiences a higher participation rate(24 percent), while the west has a lower rate (13percent). Returns e-filed from the IRS’sWestern Region were more complex, and hadlower refunds. E-file returns from the SoutheastRegion had the highest refunds, on average,and had a higher rate claiming the EITC. Thestudy defined the TeleFile participation rate asTeleFile returns filed divided by TeleFilepackages mailed. The Northeast Regionexperienced the highest TeleFile participationrate at almost 20 percent, compared to WesternRegion’s 15 percent. (However, anexceptionally aggressive state-level telefileprogram in Massachusetts has a unique effecton the Federal TeleFile participation rate in thatstate, boosting it to almost twice the national



89

average. The author suggests this situationlikely contributes to the study’s findings relativeto the Northeast Region.)

Other Profiling Research

The Pacific Consulting Group profile of TY 1995On-Line filers, Minimizing Taxpayer Burden andReducing Costs at the IRS: Analysis ofCustomer Experience with On-Line Filing(1996), identified On-Line filers as a veryspecialized group. The survey research foundthe On-Line filer to be computer literate, highlyeducated, and with a high income. Unlike themarket of lower income e-filers that uses apractitioner, the On-Line filer market is moresavvy and will require more incentive (in theform of lower perceived costs) to e-file.

A Bonney & Co. survey of TY 1995 TeleFilers(1996) found these filers to be young andsomewhat educated, and identified a substantialuntapped market of less educated eligibletaxpayers that lack confidence using acomputerized data entry system. The reportsuggests the IRS direct TeleFile marketingefforts at these non-participants to overcometheir reluctance.

Cost/Benefit of e-file

All of the IRS’s efforts in the electroniccommerce arena are predicated on theassumption such activity will save resources.However, little (and incomplete) cost-benefitanalyses have been conducted to showdefinitively electronic filing, and electroniccommerce in general, saves IRS resources inthe short term. One of the reasons for the lackof research in this area is the complexity of thetask. Estimation of upstream and downstreamcosts and benefits (e.g., those related tofacilities, equipment, storage, archiving,subsequent adjustment activity, training, auditsand other compliance-related activities) requiresdata not readily available to the IRS, ortechniques as yet undeveloped. Nevertheless,the GAO (1995) reported the following per returnprocessing costs for the 1993 filing season.

1993 Individual return processing costs (perGAO):Form 1040 - $4.53Form 1040A - $3.95Form 1040EZ - $3.36e-file (excluding TeleFile) - $3.08

These cost figures do not consider the up- anddownstream cost, but just those associated withreturn processing – i.e., the opening and sortingof mail, coding, editing, data entry, validitychecks, and error correction at the IRS ServiceCenters and Processing Centers. Examples ofupstream costs include design, printing, anddistribution of tax forms and instructions;examples of downstream costs include archivingand retrieval of tax returns, and complianceactivities.

Steuer and Benson (1996) attempted to rank e-file market segments based on cost savings.They used the GAO’s per return cost data (alongwith other processing year 1994 data onindividual taxpayers in the state of Minnesota) ina “tree-structured analysis” that used SPSSCHAID, or Chi-squared Automatic InteractionDetector. They defined e-filer market segmentsalong the following taxpayer/returncharacteristics: adjusted gross income, age,refund/balance due, tax credits (including EITC),paid preparer/self prepared, and geographiclocation. The analysis identified someinefficiencies in the IRS’s e-file marketing. Theresearch suggests the ETA use rankingtechniques such as those based on CHAID toprioritize market segments and direct its e-filemarketing strategies with a cost-benefitorientation.

More recent work by the Office of Cost Analysisunder the IRS’s Chief Financial Officerestimated the following per return processingcosts for fiscal year (FY) 1996.45

FY 1996 Individual return processing costs (perCFO):Form 1040 - $4.44Form 1040A - $3.58Form 1040EZ - $3.54Form 1040PC - $3.44e-file (excluding TeleFile) - $4.73TeleFile - $3.88

These figures also exclude up- and downstreamcosts; but they show e-filing is relatively moreexpensive than paper filing alternatives. Furtherpreliminary analysis by the CFO suggestsinclusion of the up- and downstream costs would

45 Internal Revenue Service, Office of Cost Analaysis. (1998). FY1996 Cost of Submission Processing in the Service Centers.Washington, DC.



90

not change this conclusion. However, the IRS,and the ETA in particular, argue e-file’s costeffectiveness will be realized with economies ofscale, when e-filing becomes the predominantform of filing and the large sunken costsassociated with the IRS’s paper-based systemare eliminated.

Also, it is important to note the processing costsfor paper returns in the studies described aboveare based on a system that captures only about40 percent of the information on returns. Incontrast, the e-file cost figures reflect 100percent data collection. A more balancedcomparison requiring 100 percent data collectionfor paper processing would increase significantlythe corresponding cost figures for paper returnscited above.

Still, more complete cost/benefit analysis isnecessary to enable IRS management, as wellas the Congress, to make informed businessdecisions regarding e-file expansion.

After the IRS Knows Who e-files, It MustStudy Why

While profiling efforts enable IRS researchers todescribe taxpayers that electronically file andthose that do not, they can not fully explaintaxpayers’ motivations. Why does a taxpayerchoose to file electronically? More importantly,why does a taxpayer not file electronically? Thefollowing discussion addresses each e-filecategory (i.e., ERO/practitioner e-file, On-Line,and TeleFile) separately.

Why Do Taxpayers e-file?

ERO/Practitioner Standard e-filing

After describing electronic filers as young, lowerincome, simpler return taxpayers, Musselman(1991) speculated various reasons for theirparticipation. “First, most electronic transmittersalso offer a ‘refund anticipation loan’ . . .Although this option generally costsapproximately $60 . . ., it requires no upfrontcash, and thus may be the only way manypeople can afford tax preparation services.Second, the temptation to have one’s money in

a matter of days instead of weeks may be apowerful incentive for many . . .”46

In its report to the U.S. Senate FinanceCommittee, the GAO (1993) cited the samelatter motivation behind e-file participation:quicker refunds. This study involved GAO visitsand interviews at four regional IRS offices, eightdistrict offices, and three service centers, as wellas a survey of over 1,000 preparers andtransmitters that participated in electronic filingand 1,000 that did not.

On-Line Filing

The IRS’s Strategic Planning Division’s FocusGroup Report: On-Line Filing Program (1997)cited quicker refunds and less paper to keep asOn-Line filing selling points. The focus groupparticipants (made up of taxpayers who hadaccess to a computer and modem but did not e-file on-line) also perceived IRS acknowledgmentthat a return is received as an advantage of On-

Line filing. An earlier study by the KlemmAnalysis Group (1996) and Pacific ConsultingGroup (1996) had similar findings. Their surveyfound accuracy of the filed return mostinfluenced the decision to use the On-Lineprogram. Also, they found convenience andspeed of refund as the top “filing image items.”On-Line Filers: ETA Market Research for 1998(1999) also found repeat On-Line filers areattracted by conveniences such as the ability todirect deposit their refunds.

TeleFile

The Bonney & Co. focus group (1996) andsurvey (1996) research cite ease of use, fasterrefunds, and no need for a paid preparer astaxpayers’ incentives for using TeleFile. Earlierprofiling work from the Southwest DORAsupports this conclusion.

46 Internal Revenue Service, Planning and Research. (1991).Electronic Filing – Who’s Participating and Who Isn’t.Washington, DC.



91

Why Taxpayers Do Not e-file -- Barriers toElectronic Filing


Even before the IRS had a complete picture ofwho files electronically (i.e., before it had acomprehensive market profile), it began to studywhy taxpayer behave the way they do, andspecifically the barriers to participation growth.In 1991, Musselman cited Roper survey datasuggesting unawareness was electronic filing’sbiggest barrier. This seems logical in the earlyyears of electronic filing. The Roper surveyfound 34 percent of individual taxpayers werenot aware of the possibility to file electronically.The next largest barrier to the return filer wascost, cited by 14 percent of taxpayers.

As electronic filing became better known, costreplaced unawareness as the major impedimentto participation. The GAO (1993) cited cost ase-file’s main deterrent. Almost three years later,in its report to the U.S. Senate GovernmentAffairs Committee, the GAO (1995) concludedthe same. Taxpayers had to pay $15 - $40 tofile electronically through a preparer orelectronic filing transmitter. The GAO concludede-filing appeals primarily to taxpayers most inneed of a quick refund – those that disregardcost considerations.

Tax practitioners have had their own barriers tothe electronic filing business. Nelco Inc.’s WhyTax Preparers Do Not Offer Electronic Filing(1994) explored the reasons for preparers’reluctance to e-file. Its survey researchconcluded mainly two interrelated factors affectparticipation: client demand and cost. Theresearch recognizes the many dimensions ofpractitioners’ cost – costs related to thepurchase of software and hardware,transmission, training, etc. The preparerseventually must pass these costs onto the client.Higher fees, in turn, put the preparer at acompetitive disadvantage if there is little clientdemand for e-file. Unwillingness to learn thenew system and lack of confidence in it weretwo other reasons cited, though less frequently,by the survey respondents.

In September 1994, the Chairman of theAmerican Institute of Certified PublicAccountants (AICPA) provided GAO with the

following corroborating information onpractitioners’ barriers to e-filing:47

• e-filing does not fit into their “office routine,”• clients do not perceive any additional benefit

to offset the additional cost,• e-filing requires additional input,

transmitting, and monitoring time,• e-file is not yet truly paperless – preparers

particularly have a problem with thesignature requirement. (Until filing year1999, all such electronic filing still requiredthe practitioner to prepare, sign and mail apaper signature document to authenticateeach return, as well as attach Form W-2earning statements and other documentsrequiring signatures.)

On-Line Filing

The proliferation of personal computers in the1990s has created a huge potential for the On-Line electronic filing market. However, when theIRS made On-Line e-filing available nationwidein 1995, there was little promotion aimed at On-Line filers either from the software vendors orthe IRS. Some tax preparation softwarevendors offered electronic filing at no extra cost(e.g., “first one free” offers), but this feature wasnot advertised by the vendors as a major sellingpoint.

The 1996 On-Line filer customer satisfactionsurvey research by the Pacific Consulting Groupand the Klemm Analysis Group foundsatisfaction among current users generally high,but also found high retransmission rates and lowcustomer support ratings discourage repeat useor initial entrance into the program.

To devise an effective marketing plan to attractOn-Line filers, the IRS conducted focus groupswith eligible taxpayers that did not file On-Line todetermine barriers. The Strategic PlanningDivision’s Focus Group Report: On-Line FilingProgram (1997) reported taxpayer awareness asthe biggest issue. Only half the focus groupparticipants knew the existence of On-Line filing,and then not much beyond that. General lack ofknowledge – how the program works, itsrequirements, and where to get moreinformation – manifested itself into fear andanxiety. Participants expressed concern

47 General Accounting Office. (1995). Electronic Filing FallingShort of Expectations. Washington, DC.



92

regarding data security, technical problems withtransmission, lack of paper documentation, aswell as software and transmission costs. Somecited their techno-phobia associated with theuse of computers on so complicated a task astax return preparation. Also, some thought sinceelectronically transmitted information is easier toaccess, it would be easier for the IRS to conductrandom audits on such returns, and thusconsidered electronic filing disadvantageous.

TeleFile

The Bonney & Co. focus group (1996) andsurvey (1996) research discovered similar fearsamong potential TeleFilers. Anxiety resultedfrom the lack of paper documentation, difficultyusing the alpha code on the return envelope,and general “techno-phobia.” In addition, theSouthwest DORA’s 1997 profile determinedsome of the barriers to TeleFile participation arecreated inadvertently by the IRS. For the 1997filing season, the IRS mailed 26.6 millionTeleFile packages to taxpayers. However, 8.2million of those taxpayers were ineligible. Theresearchers also found 5.0 million qualified touse TeleFile were not identified or mailed aTeleFile package by the IRS.

These reports and findings provide the IRS withmany insights to the barriers to electronic filinggrowth in the 1990s. They also point to theneed for independent marketing strategies forthe distinct e-file markets – non On-Line (i.e.,preparer-originated), On-Line, and TeleFile.

Breaking the Barriers – Marketing e-file


The North Florida and Kentucky-TennesseeDORAs’ Alternative Ways of Filing ResearchReport (1997) presented research that testedthe effectiveness of different treatments directedat electronic return originator participation in theelectronic filing program. These treatmentsincluded various forms of direct contact with theEROs designed to encourage greaterparticipation. The treatments failed to motivatethe EROs, who cited costs and lack of clientdemand as impediments. The researcherstherefore recommended marketing e-file directlyto taxpayers, who in turn would create demandfor e-file services from their preparers.

Not coincidentally, in 1997 the ETA begandeveloping marketing plans directed at both thetax practitioner community and taxpayers thatused practitioners. First, the ETA re-emphasized its marketing to practitioners,highlighting strides toward complete officeautomation and reduction of paper, confirmationof receipt, and reduced errors. The practitionershad the lucrative RAL (and RAL-like) markets asadded incentive. Second, the IRS marketeddirectly to taxpayers through print advertising, aswell as selected radio and television spots,promoting faster refunds and suggesting theyask their preparers about e-filing options.

On-Line Filing

The North Florida and Kentucky-TennesseeDORAs’ recommendations for theERO/practitioner e-file marketing apply to theOn-Line market, as well. A large portion of theIRS’s marketing promotes e-filing in generalterms, whether it is through a practitioner, on-line, or by telephone. However, the IRS needsto develop a more directed effort towards theOn-Line market segment. Among suggestionsmade by the participants of the 1997 On-Linefiling focus groups conducted by the StrategicPlanning Division: the IRS should run amarketing campaign simply to inform the publicabout On-Line filing. Given the participants’misperceptions and anxiety, this seems a goodidea. During the 1998 and 1999 filing seasons,the IRS partnered with major tax preparationsoftware companies to advertise On-Line filingcapabilities as a selling point for the software.

TeleFile

In 1996, the IRS contracted Price Waterhouse todevelop a marketing and communicationsstrategy to promote e-file use. In its report,Electronic Filing Marketing and CommunicationPlan (1997), the consultants concluded TeleFilewas the only IRS electronic filing product readyfor full marketing. It found 99 percent of allTeleFile users in 1996 planned to use TeleFileagain the following year. However, citing the1996 Automated Survey of TeleFile Users, italso found 85 percent of those eligible to useTeleFile did not do so. These two facts pointedto a marketing deficiency.

The DORAs explored this deficiency. Citing thatonly 11 percent of all taxpayers receiving



93

TeleFile packages used TeleFile in 1996, theConnecticut-Rhode Island DORA conductedlocal research aimed at increasing this rate. ItsResearch Report: The Effect of a Mailing onTeleFile Usage (1998) describes the test itconducted to measure the effect of flyerdistribution on TeleFile usage in 1997. Thoughlimited in scope, the researchers concludedmailing reminder flyers in addition to TeleFilepackages has no effect on participation. In1997, the North Florida and Kentucky-Tennessee DORAs conducted a similar test inthe IRS Southeast Region, using postcardsrather than flyers. Their Alternative Ways ofFiling Research Report (1997) arrived at thesame conclusion: a separate reminder mail outhas no effect on TeleFile participation. Thesetests confirmed the 1994 TeleFile CustomerSatisfaction Survey and Bonney & Co. survey(1996) findings that taxpayer awareness ofTeleFile results from the receipt of the TeleFilepackage itself.

In its TeleFile Marketing Strategies for theGeorgia District (1998), the Georgia DORAtested three low- or no-cost TeleFile promotionstrategies: 1) community and student cabletelevision messages, 2) an advertising campaignimplemented one week before the TeleFilepackage mail out, and 3) stuffers in on-campusstudent mailboxes. The research found thecombination of the first and second strategiesincreases TeleFile participation the most.

However, the researchers acknowledge lack ofcontrol on various aspects of the test. They alsoconclude there may be a natural 25 percentparticipation rate ceiling, given the programscurrent restrictions.

The Georgia DORA issued another report,Expanding TeleFile Eligibility (1998), thatexamined the impact of removing variousTeleFile eligibility conditions. The researchfound under 21 percent of Georgia districttaxpayers were eligible to use TeleFile in 1997.By removing the taxable income, interestincome, and filing status limitations, as well asthe restrictions based on age and blindness,eligibility rose to over 32 percent. If in addition,the Service were to allow TeleFilers to claim twodependents and file a Schedule A for itemizeddeductions, eligibility would increase to almost48 percent. The report acknowledges there are

costs associated with lifting the currentrestrictions, and recommends further study.

Conclusion

The IRS works with a variety of stakeholders inan effort to expand electronic informationexchange. A vital step is the identification andremoval of barriers. Barriers exist relative to theIRS’s tax administration duties, and to thepublic’s acceptance and use of newtechnologies. From the public’s perspective, theIRS will have to attract electronic exchange byreducing the public’s burden and costs indealing with the IRS. The IRS also mustaddress such issues (real or as perceived by thepublic) as fraud prevention, electronicauthentication, taxpayer privacy and informationsecurity, rules for certifying electronic taxadministration participants, and how to regulatethird parties that wish to exchange information.

This article provides a review of the researchdone in the individual e-file area. Further workshould include annual maintenance of thedatabases developed to study the individual e-file market, as recommended by the SouthwestDORA and others. Also, more research-oriented work is needed in the testing ofmarketing treatments (e.g., testing differentadvertising approaches to determine which workbest, through the use of test and control groups)and product development. To meet the 80percent participation goal set by the Congress,the IRS is preparing to implement many newinitiatives intended to expand e-file. The IRSshould precede these initiatives with small-scale(i.e., low cost) tests, such as those conducted bythe Georgia DORA for TeleFile, that quantifytheir impact. After such tests, the IRS canchoose and implement the most effectiveinitiatives.

Of equal importance is the need to start basicresearch in the business return e-file area. TheIRS currently accepts Form 1041, U.S. FiduciaryIncome Tax Return (for estates and trusts),Form 1065. U.S. Partnership of Income,Schedules K-1 and series 5500 returns, AnnualReturn Report of Employee Benefit Plan, oneither magnetic tape or via electronic filing. Itaccepts Form 941, Employers Quarterly FederalTax Return via magnetic tape, electronic filing,and TeleFile, as well as Form 940 via magnetictape. These electronic commerce programs



94

(excluding the magnetic media tape) are new,and still in the developmental stages in terms ofthe ETA’s marketing initiatives. Marketinginitiatives and further product developmentrequire the support provided by the profiling,study, and treatment testing that havecharacterized the individual e-file research.

Finally, the Price Waterhouse report, ElectronicFiling Marketing and Communication Plan(1997), recommends the IRS fully integratemarket research into e-file product developmentfrom the start to insure products and servicesare developed with the customer in mind. Itgoes on to recommend the IRS institutionalizecustomer feedback into all product developmentand refinement activities. Such information isvital to planning effective marketing strategies.

Research Currently in Progress

The IRS’s Office of Research and DistrictOffices Research and Analysis each yearpresent IRS management a Research Plan thatlists significant research projects planned andunderway.48 The ETA Research StrategyProjects listed on the fiscal year 1999 ResearchPlan reflect a marked increase in e-file researchactivity. Notably, in the business return area,Project 1.02 supports the development of an e-file database that will have the same purpose asthe individual return e-file database created andused by the Southwest DORA. The ResearchPlan also contains projects that support analysisof recent ETA pilots and initiatives designed tofurther expand individual e-file participation.Projects 1.11 and 1.12 entail survey research todetermine the impact and success of thealternative signature pilots conducted during the1999 filing season for the On-Line, as well aspreparer/ERO, filed returns. Project 1.05 willcontinue the exploration of CHAID to project thecost savings of individual e-file returns. Thereare a total of ten projects in the FY 1999 Plandesigned to better understand and promote e-filing.

48 For IRS employees, the FY 1999 Research Plan is available athttp://www.research.irs.gov/PLANS/plans.htm.


Javier Framinan is an economist in theProjections and Compliance Studies Group,National Office of Research. He received anM.A. degree in economics in 1987 from theUniversity of Virginia. He has been with the IRSsince 1987

The IRS Research Bulletin 95

1999 Update

References

Angell & Co. (1998). IRS PSA Tracking Study ‘Benchmark’Wave Table Report. Washington, DC (?)

Bonney & Co. (1996). Findings of Focus Groups Regardingthe TeleFile Program. Virginia Beach, VA.

Bonney & Co. (1996). Results of Surveys Among PersonsWho Filed their 1995 Personal Federal Income Taxes Usingthe TeleFile Program. Virginia Beach, VA.

General Accounting Office. (1993). Opportunities to Increasethe Use of Electronic Filing. (GAO/GGD-93-40). Washington,DC

General Accounting Office. (1995). Electronic Filing FallingShort of Expectations. (GAO/GGD-96-12). Washington, DC.

Internal Revenue Service. (1991). Musselman, Bryan.Electronic Filing – Who’s Participating and Who Isn’t.Publication 1500, The IRS Research Bulletin. Planning andResearch. Washington, DC.

Internal Revenue Service. (1994). 1994 TeleFile CustomerSatisfaction Survey. Statistics of Income. Washington, DC.

Internal Revenue Service. (1996). Incentives andDisincentives for Electronic Filing: A Survey of Studies andReports to Date. Compliance Research. Washington, DC.

Internal Revenue Service. (1996). Profile Report: Current andPotential Markets for Electronic Filing. ELF Profiling ProjectTeam. Washington, DC.

Internal Revenue Service. (1996). Steuer, Anne and Benson,Michelle. Exploring Current and Potential Markets forElectronic Filing (ELF) Using Tree-Structured Analysis. NorthCentral District. St. Paul, MN.

Internal Revenue Service. (1997). A National Profile of IRS e-file Users in 1997. District Office Research and Analysis,Southwest District. Phoenix, AZ.

Internal Revenue Service. (1997). Critical Issues forDevelopment of an Electronic Tax Administration Strategy: APlan for Moving Federal Tax Administration into theInformation Age. Electronic Tax Administration. Washington,DC.

Internal Revenue Service. (1997). Identification of Barriers toOn-Line Filing Study. District Office Research and Analysis,Pennsylvania District. Philadelphia, PA.

Internal Revenue Service. (1997). On-Line Filing Program:Focus Group Report. Opinion Research. Washington, DC.

Internal Revenue Service. (1997). Alternative Ways of FilingResearch Report. District Offices Research and Analysis,North Florida and Kentucky-Tennessee Districts. Jacksonville,FL; Nashville, TN.

Internal Revenue Service. (1998). FY 1996 Cost ofSubmission Processing in the Service Centers. Document10806 (03-98). Office of Cost Analysis. Washington, DC.

Internal Revenue Service. (1998). A Strategy for Growth.Publication 3187. Electronic Tax Administration. Washington,DC.

Internal Revenue Service. (1998). The Effect of a Mailing onTeleFile Usage. District Office Research and Analysis,Connecticut-Rhode Island District. Hartford, CT.

Internal Revenue Service. (1998). TeleFile MarketingStrategies for the Georgia District. District Office Researchand Analysis, Georgia District. Atlanta, GA.

Internal Revenue Service. (1998). Expanding TeleFileEligibility. District Office Research and Analysis, GeorgiaDistrict. Atlanta, GA.

Internal Revenue Service. (1999). Fiscal Year 1999 ResearchPlan. Office of Research. Washington, DC.

Internal Revenue Service. (1999). On-Line Filers: ETA MarketResearch for 1998. District Office Research and Analysis,Southwest District. Phoenix, AZ.

Klemm analysis Group, Inc. (1996). ‘On-Line’ Filing ProgramCustomer Satisfaction Survey. Tir-95-0038. Washington, DC.

Nelco, Inc. (1994). Why Tax Preparers Do Not OfferElectronic Filing. Green Bay, WI.

Pacific Consulting Group. (1996). Minimizing TaxpayerBurden and Reducing Costs at the IRS: Analysis of CustomerExperience with On-Line Filing. Palo Alto, CA.

Price Waterhouse LLP. (1997). Electronic Filing Marketingand Communication Plan. Tir-95-0037. Bethesda, MD.

Tax Systems Modernization Institute, IIT Research Institute.(1997). Tax Administration Electronic Commerce StrategicImplementation Plan. Tir-93-C-00026. Lanham, MD.


96

Inadequate Compensation to 1120S Corporate Officers Study Report

By Bruce KorbesmeyerKansas-Missouri District Office

Research and AnalysisJune 1998

Research indicated that nationwide almost $284 million of additional employment tax was due becauseofficers of small corporations underreported their compensation.

The Kansas-Missouri and Ohio District OfficesResearch and Analysis (DORAs) conducted a studyof Form 1120S Officer Compensation. Theobjectives were: (1) to quantify the amount ofunderreported officer compensation and theassociated employment tax gap, and (2) to identifythe underlying causes of underreporting and the taxgap associated with each cause. Using the tax year(TY) 1994 Business Returns Transaction File (BRTF)to frame the population, a random sample of 528 TY1995 returns was chosen for classification. Weselected 141 returns for audit.

The study results indicated that nationwide almost$284 million of additional employment tax was duebecause officers underreported their compensation.Major causes of underreporting and the associatedtax gap for each cause were:

§ IntentionalUnderreporting $106,800,000

§ Misinterpretation of Reporting Requirements $32,500,000

§ Taxpayer’s Lack ofTax Law Knowledge $ 76,300,000

(All estimates use the lower confidence limit. Thetotal does not add up to $284,000,000 dueto additional sampling error that occurs whenestimating the projected employment tax gap bycause.)

The study proposed: (1) the DORAs partner withstakeholders of the potential treatments in the nextphase of the research process, and (2) potentialtreatments be refined through a detailed cost/benefitanalysis.

Research Abstracts

RESEARCH ABSTRACTS


97

Can Demographic Trends Predict Taxpayer Noncompliance?

By Kim M. BloomquistIllinois District Office

Research and AnalysisJune 1998

The Bureau of Economic Analysis’ wage and salary adjusted gross income (AGI) “gap” has exhibited astatistically significant relationship with the percent of US population between 35 and 54 years of age. Thisrelationship is used to predict the direction of taxpayer noncompliance to the year 2020.

This analysis examined how would the aging of the“baby boom” generation influence individuals’compliance with tax laws. The paper introduced a life-cycle view of taxpayer noncompliance that associatednoncompliance behavior with age-related factors suchas an individual’s knowledge of tax issues, motivationto comply and opportunity to underreport earnings.This study also compared IRS and Bureau of EconomicAnalysis (BEA) estimates of misreported wage andsalary income and recommended use of the BEA’sAdjusted Gross Income (AGI) Gap for wages andsalaries as an indicator of the direction, but not themagnitude, of noncompliance. Regression analysisfound that the percent of US population between theages of 35 and 54 accounted for 84 percent of thevariance in the relative wage and salary AGI gap from1947 to 1995.

Based on these empirical findings, individualnoncompliance was projected to increase until the turnof the century as the last of the baby boom generationentered middle age. This trend is expected to reverseearly in the next century and aggregate compliancegradually improve as babyboomers leave their peak earning years and begin toretire.

The paper noted several important policy implicationsfor compliance research. First, a prior IRS goal of athree percent reduction in the “tax gap” by the year2001 is made more difficult because of demographictrends that augur more reporting noncompliance, notless, at least untilthe turn of the century. Second, IRS researchers

will need to net out the influence of an agingpopulation on aggregate compliance behavior tocorrectly evaluate the effectiveness of proposedtreatment programs. Third, IRS should explore waysto motivate younger taxpayers to comply. Wholesale

education and treatment efforts focusing on youngertaxpayers may potentially pay for themselves manytimes over for a relatively modest cost. Finally,methodology used by the BEA to estimate the AGIGap points to the need for an industry by industryreview of reporting compliance behavior byemployers. Major structural changes in the USeconomy over the last twenty years could mean that asignificant portion of employee income is not beingreported to the IRS by employers.

RESEARCH ABSTRACTS


98

Pre-Payment Position and Income Tax Non-Compliance

By Peter D. Adelsheim, Ph. D.,Pacific Northwest District Office

Research and AnalysisJanuary 1997

This research examined the relationship between income tax withholding and compliance.

Although only 26 percent of all individual returns areunder-withheld, they account for over 69 percent ofthe understated tax. Furthermore, the under-withheldreturns have an average tax change of $820 comparedto only $128 for the over-withheld. This study soughtto answer two questions. First, can under-withholdingbe regarded as a cause of noncompliance? That is, ifwithholding procedures were more accurate (so as toreduce the number of under-withheld returns), wouldwe expect an improvement in compliance? Second,what is the compliance impact of under-withholding?What incremental revenue might be expected frommaking particular changes to the system?

Our conclusions were as follows. First,the under-withholding phenomenon wasreal: the under-withheld were lesscompliant. Second, this relation was notmerely statistical. While the currentanalysis provided ample evidence thatwithholding is an

important causal agent in the complianceprocess. Third, changes to thewithholding system could be devised thatwould have significant impact onrevenues at little real cost to thegovernment.

Further study should focus on the followingquestions. First, are there important, identifiablesub-segments among the 27 million under-withheldreturns that are relevant to tax administration? Canresearch identify groups of taxpayers that arehomogeneous with respect to the causes of theirunder-withholding? Second, although this reportfocused almost entirely on reporting accuracy, it isimportant to know whether under-withholding isrelated to other forms of non-compliance (i.e., filing

and payment). Third, can the alternative withholdingsystems suggested here and other places be morecompletely defined? Can estimates of the costs andbenefits of these systems be developed?

RESEARCH ABSTRACTS


99

Farm Labor Contractors Compliance with the Income Tax Laws

By James L. Zanetti andMarlene Le

Central California District OfficeResearch and Analysis

January 1999

Farm labor contractors in California have a generally low level of compliance with the income tax laws whencompared with other segments the IRS has studied.

The United States is the largest agricultureproducer in the world, and the State of Californiais the largest agriculture producer in the UnitedStates. According to the 1992 Census ofAgriculture, California had over 77,000 farmerson 28.9 million acres growing products with amarket value of $17.0 billion. Wage and salaryemployees in agriculture accounted for 361,000(or 2.3%) of California’s civilian labor force of15.3 million. The wages paid to grow and harvestthe State’s agriculture products were $2.9 billion.

Much of the labor in the agricultural segment isprovided by farm labor contractors (FLCs). Theagricultural laborers are employed directly by theFLCs, and subcontracted to farm enterprises. Inorder to measure FLCs’ compliance with theincome tax laws, we obtained a database of all of

them registered in Arizona, California, Nevada,and New Mexico from the U.S. Department ofLabor (DOL). IRS data for all registered farmlabor contractors came from the IndividualReturns Transaction File (IRTF) for the calendaryears ending December 31, 1992 and December31, 1993. Secondary data sources included the1992 Census of Agriculture, the Bureau of LaborStandards Consumer Expenditure Survey, andthe California Statistical Abstract - 1994.

The overall U.S. population’s level of compliancewith the income tax laws is in the mid-eightypercentile. The following table showscomparatively lower compliance levels forCalifornia farm labor contractors.These results were used to support localcompliance projects in this market segment.

IncomeTaxYear

TotalNumberof FLCs

Total IncomeTax Reported

EstimatedUnreportedIncome Tax

AverageUnreported

Tax

VoluntaryCompliance

Level1994 3,588 $8,670,268 $3,808,317 $1,061 69%1993 3,519 $7,258,998 $3,460,331 $983 68%

RESEARCH ABSTRACTS


100

Erroneous PBA/PIA Profile

By Stan GriffinCentral California District Office


Analysis suggests invalid PBA/PIA codes are not an important factor in taxpayer compliance.

Prior IRS studies indicated a significant portion (up to30 percent) of Principle Business Activity/PrincipleIndustry Activity (PBA/PIA) codes on filed returns areinaccurate. This research focused on the relationshipbetween PBA/PIA code accuracy and taxpayercompliance. Were taxpayers that used invalidPBA/PIA codes less compliant than those that usedvalid ones?

We profiled Form 1040 (with Schedule C), Form1120, Form 1120-S and Form 1065 filers using fiscalyear 1994 Interim Compliance Research Informationsystem (ICRIS) sample data files from five states(Massachusetts, Maine, New Hampshire, Vermont,and Pennsylvania) plus Central California.Categorizing results by either valid or invalidPBA/PIA codes, we measured noncompliance severalways.

Analysis of all the measures suggested invalidPBA/PIA codes were not an important factor in taxnoncompliance in the Form 1040 Schedule C, Form1120, Form 1120-S, or Form 1065 populations.However, returns with invalid PBA/PIA codes tendedto be filed a few days later than valid PBA/PIA codedreturns. This was especially true for business returns(i.e., Forms 1120, 1120-S, and 1065), which hadapproximately double the filing delinquency rate ofbusiness returns with valid PBA/PIA codes.Practically speaking, the difference in timeliness forvalid versus invalid PBA/PIA coded returns in thisinstance was approximately one week or less -- not asignificant factor in terms of compliance.

All the districts studied reported on average a 5percent higher rate of invalid codes on Form 1040Schedule C returns prepared by a paid tax preparer.However, we could not demonstrate a causalrelationship between paid tax preparer returns, invalidPBA/PIA codes, and noncompliance.

RESEARCH ABSTRACTS


101

Business Profitability and Income Tax Compliance

By Kim M. BloomquistIllinois District Office

Research and Analysis January 1999

From 1957 to 1995, sole proprietor net profit margins and reporting non-compliance exhibited a strong negativecorrelation. This relationship suggested that as long as competitive market forces act to squeeze small business ratesof return, sole proprietor tax compliance will continue to erode.

In 1992, the IRS estimated the “tax gap”, the amountof tax legally owed but not voluntarily paid, at $95.3billion. Of this amount, $39.9 billion, or 42 percent,was due to unreported small business income. Thehigh rate of noncompliance by small business is oftenattributed to the lack of third party reporting ofbusiness income. In the absence of enforcedwithholding, the IRS continues to rely on voluntaryreporting by business taxpayers backstopped bytraditional enforcement techniques (e.g., audits). Amore recent emphasis on “wholesale” compliancemeasures, coupled with a renewed emphasis oncustomer service, was also hoped to reducenoncompliance.

However, the rate of small business noncompliancehas not been static over time. From 1957 to 1995, theBureau of Economic Analysis (BEA) estimates theshare of sole proprietor income not reported on federaltax returns more than doubled -- from 26% to 53% fornon-farm proprietors and from 60% to 125% for farmproprietors (if income not reported exceeds reportedincome, the ratio can exceed 100%). During the sametime, net profit margins shrank an average of 30percent for a core group of non-farm soleproprietorships and 50 percent for farm proprietors.

The high statistical correlation between net profitmargin and percent underreported income (-0.90 fornon-farm proprietors and -0.77 for farm proprietors)may indicate some businesses initiate or increase theirlevel of tax evasion activity in response to leaneconomic times. This finding was similar to thatreported by Rice (1992) for small corporations.Assuming a continuation of the historical trend ofnarrowing profit margins, it was projected thatreporting noncompliance of non-farm sole proprietorswould rise 36 percent from 1995 levels by the year2020. Similarly, underreporting among farmproprietors in 2020 was expected to rise 27 percentfrom 1995 levels. This conclusion stood in contrast toa recent analysis of reporting noncompliance of wageand salary income which foresees a gradualimprovement in compliance by wage earners over thenext 25 years (Bloomquist, 1998).

The paper recommended that the IRS track industryprofit trends to provide possible early warning ofindustries with emerging compliance issues.Alternative tax systems appropriate for small business,such as the Value Added Tax (VAT), also weresuggested for further study as a complianceimprovement strategy.

RESEARCH ABSTRACTS


102

Tax Table Clustering: Empirical Evidence of Non-compliance?

By Peter D. Adelsheim, Ph. D.,Pacific Northwest District Office

Research and Analysis May, 1996

This research examined the question of whether taxpayer distribution within tax table brackets wasevidence of non-compliance and whether this information could be a benefit to the IRS.

In a 1985 article, Joel Slemrod(University of Michigan) developed theidea of secondary evasion in order tostudy the compliance impact of severalpolicy variables: most notably, marginaltax rates. Secondary evasion committedby taxpayers finding themselves in thebottom of one tax bracket under-reporttaxable income just enough to slip intothe top of the next lower bracket.Empirically, Slemrod does find a largerthan expected proportion of taxpayers inthe upper parts of the tax table brackets.Theoretically, he argues secondaryevasion is a sure sign of primary evasionand uses the proportion of taxpayers inthe upper quintile of each bracket as anempirical measure of noncompliance (inreporting accuracy).

The Pacific Northwest District OfficeResearch and Analysis (DORA) studiedSlemrod’s idea of secondary evasion todetermine its validity and applicability inthe IRS. In particular, we sought toanswer the following four questions.First, does clustering exist? Using the1988 Taxpayer ComplianceMeasurement Program (TCMP)database, we too found adisproportionately large number oftaxpayers in the upper quintiles of the taxbrackets in virtually every marketsegment examined. Second, isclustering evidence of secondaryevasion? Although at the outset it wasimportant to answer this question, weconcluded it can not be answeredempirically. It is difficult to imagine theresearch design that would separateprimary evasion from secondary. Third,is clustering evidence of primaryevasion? Using the TCMP database, we

could not find any association betweenclustering and under-reporting of taxes.Fourth, can clustering be used to refineIRS’s Discriminate Function (DIF) systemof case selection? Unfortunately,qualifying DIF selection by reportingincome in the upper quintile of a tax tablebracket lowered the expected tax changeand increased the estimated level ofcompliance.

Finally, Charles Christian (Arizona StateUniversity) has provided an alternativeexplanation for clustering. He hassuggested taxpayers finding themselvesnear the bottom of a tax table bracketexpend the extra energy to find legitimatedeductions to move them into the lowerbracket. In summary, we found little toencourage further research into theconcept of secondary evasion.

RESEARCH ABSTRACTS


103

Empirical Goodness of Fit Scoring System(Applied to Tax Preparer Due Diligence)

By Curt HopkinsCentral California District Office


The results of an Empirical Goodness of Fit (EGoF) scoring system were compared with the opinions ofExamination Division Return Preparer Coordinators. The purpose of this effort was to either support orrefute the use of EGoF as a method to classify tax preparers as either ‘questionable’ or ‘not questionable’for research purposes.

The Internal Revenue Service spends a greatdeal of resources identifying, penalizing, andprosecuting professional tax return preparers whofail to use due diligence. The ExaminationDivision of each District Office maintains a staff ofone or more Return Preparer Coordinators(RPCs) to monitor questionable preparers. Thisproject, which was limited to individual incometaxes related to Form 1040, compared the resultsof an Empirical Goodness of Fit (EGoF) scoringsystem with the relatively resource-intensive workby Examination Division RPCs.

The premise behind EGoF was that tax preparerscannot make up numbers at random on a client’sreturns, especially if their goal is to minimize thetaxes paid by their clients. To this end, theleftmost (first significant) digit of a line item isobserved and compared to all leftmost digits ofthe same line item on returns filed through all taxpreparers in a district. The technique used inscoring a specific preparer’s change from theexpected proportions of ones, twos, threes, etc. isPearson’s Goodness of Fit. Thus the empiricaldistribution of digits scored by this goodness of fittechnique gave each preparer a single score ofrelative ‘fit’ within the preparer community as awhole

The purpose of this initial effort was to eithersupport or refute the use of EGoF as a method toclassify tax preparers as either ‘questionable’ or‘not questionable’ for research purposes. Thisdetermination was made without reference to themagnitude of tax changes or penalties assessed.It is not intended for use in selecting workload forenforcement purposes.

For this project we gave similar packets of taxpreparer client data to recognized experts, the

RPCs, in the examination divisions of four testdistricts. The 366 responses of the experts werecompared to the predictions of the mathematicalscoring system, and the comparison analyzedwith nonparametric statistical techniques. Wefound the EGoF scoring technique performed aswell as the experts. We also found that atsuccessively higher EGoF scores (indicating agreater likelihood of a questionable preparer),fewer mismatches occurred between Goodnessof Fit and experts’ conclusions. False positivesdrop below 3 percent at higher scores. TheEmpirical Goodness of Fit system passed all ofthe statistical tests for use in grouping preparersfor research use.

RESEARCH ABSTRACTS


104

Processing Year 1998 Criminal Investigation Questionable Refund FormulaDevelopment

By James A. WilhelmGeorgia District OfficeResearch and Analysis

April, 1998

The development of updated Questionable Refund Formulas helped Criminal Investigation identify and stop moreerroneously filed refund returns.

Using a mixture of Criminal Investigation (CI)questionable refund data with Processing Year1997 (PY1997) individual master file data, theGeorgia District Office Research and Analysis(DORA) developed the PY1998 QuestionableRefund (QR) formulas. These formulas weredesigned to help CI detect and “freeze” therefund of filers who have filed erroneous refundreturns. The PY1998 QR formulas were animprovement over the operational PY1997 QRformulas. In PY1997, CI reviewed more than 3.2

million returns to find 16,532 erroneously filedreturns. By using the developed PY1998 QRformulas, CI was expected to be able to reviewless than 1.5 million returns to identify about18,000 erroneously filed returns. Thus, thedeveloped PY1998 QR formulas indicated that CIcould detect and “freeze” more erroneously filedreturns while reviewing over 50 percent fewerreturns.

RESEARCH ABSTRACTS


105

Roadmap to the Future: Using Groupware as an Analytical Tool to RankEmerging Compliance Issues

By Deborah J. Myers, Ph.D.North Texas District Office


Collaborative software aids in prioritizing emerging compliance issues.In 1997, the North Texas District culminated a two-year effort to study the District’s demographic makeupand economic trends by recommending a series ofimplementation steps the District could take to addressemerging compliance issues. A critical step wasidentifying and prioritizing the top issues for theDistrict. A non-subjective and quantifiable approachwas sought to accomplish this task. Once issues werefully brainstormed and fleshed out, a way to rank thevarious issues was employed. Decision-makingsoftware -- known as groupware -- was used to aidgroup collaboration and facilitate the prioritizationprocess.

A survey of both implementation team members andupper-level District management helpedidentify the criteria that would be used to evaluate 29emerging issues. The criteria were: risk to the IRSorganization; impact on the IRS; durability of the issueover time; financial cost of implementation; impact onthe taxpayer; and, resource availability. In order torank the issues based on these weighted criteria, theteam used groupware as a computer aid. Theadvantages of using the groupware tool werethreefold: 1) team members’ votes on the importanceof issues remained anonymous throughout the rankingprocess; 2) the tool allowed a qualitative, potentiallycontentious ranking session to be quantified in anobjective way; and 3) it saved many hours of meetingtime. Using groupware proved to be a valuable part ofour effort.

RESEARCH ABSTRACTS


106

Closed Case Analysis to Support Collection of Agreed Deficiencies

By Ronald L. EdgertonOhio District Office

Research and AnalysisMay 1997

A closed case database analysis revealed opportunities to improve Examination Division’s collection ofagreed assessments.

The Ohio district Office Research and Analysis(DORA) conducted a study for the Ohio DistrictExamination Division of “Trends in the Collectionof Agreed Examination Deficiencies” tosupplement Examination’s efforts to enhancethe percentage of ‘agreed’ dollars it collects.Utilizing variables in Examination Division’sfiscal year (FY) 1996 Closed Case Database(CCDB), the Ohio DORA explored a wide arrayof characteristics of assessments such as thesize of assessment, organizational unit makingthe assessment, business activity of thetaxpayer, case cycle time, and the CCDB“collectibility indicator.”

Using a variety of illustrative charts and graphs,the Ohio DORA provided Examination Divisionwith a report that highlighted those areas thatprovide the best opportunities for improving thecollection of agreed deficiencies.

Combining this information with input frommanagers and employees, Examination Divisioncrafted an action plan involving all Examinationfunctions, to take advantage of the low and nocost actions the study brought to light.

RESEARCH ABSTRACTS


107

Final Study Report: Self-Employed Real Estate Agents

By Rick DeneshaUpstate New York District Office


Reporting noncompliance and payment noncompliance by self-employed real estate agents appeared to becaused by poor internal accounting controls, a lack of knowledge of the rules for substantiating deductions,and the lack of withholding at source. Traditional enforcement programs did not appear to be the mostefficient way to manage noncompliance within this type of small business market segment.

A profile report issued in fiscal 1996 indicatedself-employed real estate agents were at risk forfiling, payment and reporting accuracy problemsrelative to their income tax. However, TaxpayerCompliance Measurement Program (TCMP) andAutomated Issue Identification System (AIIS)data strongly indicated the major issuesassociated with reporting noncompliance (i.e.,accuracy problems) were not unique to realestate agents, but appeared characteristic of allsole proprietorships. The TCMP data furtherindicated the causes of reporting noncompliancemay be the result of taxpayers’ poor internalaccounting controls and lack of knowledge ofsubstantiation rules for business deductions.

Analysis of Audit Information ManagementSystem closed case data in conjunction with AIISdata indicated traditional audit programs were notthe most suitable mechanism for managing andimproving reporting compliance within this marketsegment. Regarding payment, the researchsuggested noncompliance is associated with alack of withholding on commission payments.Regarding filing, professional license data maybe used to identify potential nonfilers. Theresearch concluded compliance problems forthese types of small business market segmentwere best addressed by a proactive, wholesaleapproach rather than one-on-one enforcement.

RESEARCH ABSTRACTS


108

Subchapter C Corporations with Low Retained Earnings

By Michael HillCentral California District Office

Research and AnalysisRichard Denesha

Upstate New York District OfficeResearch and Analysis

June 1997

The Central California and Upstate New York DORAs found corporations with a history of losses were atrisk for filing and payment noncompliance, and understatement of tax liability was a function of returncomplexity.

Baselining of corporate tax reporting revealedcorporations that reported little retained earningson Schedule L of their Form 1120 had aVoluntary Reporting Percentage (VRP) materiallylower than the VRP for other corporations. TheCentral California and Upstate New York DistrictOffices Research and Analysis (DORAs)conducted a national profile to find whethercorporations in this market segment were at riskfor compliance problems relative to filing,payment, or accuracy. We also wanted todetermine whether these corporations constitutedan identifiable and potentially reachable marketsegment that merited future research.

We used the Interim Compliance ResearchInformation System (ICRIS) databases to analyzethe Central California, Upstate New York andNational market segments. These data filesconsisted of a random sample of corporate taxreturns processed during 1994. We found thefollowing.• Over sixty percent of all Internal Revenue

Code subchapter C corporations have lowretained earnings.

• The majority of these corporations are closelyheld.

• Over 80 percent report $500,000 or less intotal income.

• Approximately 90 percent report the bookvalue of their assets to be $250,000 or less.

• Depending on their geographic location, 40 to50 percent classify themselves as service orretail establishments.

• The effective tax rate for over 95 percent ofthese corporations was 5 percent or less.

• More than half of these corporations pays noFederal Income tax.

From our profile we concluded (a) corporationswith a history of losses, as evidenced by negativeretained earnings, are at risk for filing andpayment problems; and (b) accuracy issues andunderstatement of corporate income tax liabilitywere a function of the complexity of the return.However, one can not completely understand thecompliance risks associated with a corporationwith low retained earnings without consideringthe other tax returns associated with thatcorporation. These would include other returnsfiled by the corporation, as well as the returns ofrelated persons. Finally, (C) corporations withlow retained earnings do not constitute a unique,reachable market segment. The complianceproblems we identified were common to all smallbusinesses. These include complying withcomplex tax laws and the difficulties associatedwith properly accounting for transactions amongrelated parties and entities.

RESEARCH ABSTRACTS


109

Estimated Collections on FY 1998 Accounts Receivable DollarInventory

By Todd HeadrickResearch and Statistics of Income

Office of Planning and FinanceProjections and Compliance Studies Group

March 1999

Estimates indicate that of the $246.1 billion in gross FY 1998 ARDI, the IRS can expect to collect $41.4billion over the next ten years, or just under 17 percent. This implies an ADA for gross FY 1998 ARDI ofabout 83 percent.

At the request of the Office of AccountsReceivable Analysis, Projections and ComplianceStudies Group staff developed estimates ofcollections on fiscal year (FY) 1998 accountsreceivable dollar inventory (ARDI).

These estimates of future collections werederived from historical experience. They can beused to compute an associated allowance fordoubtful accounts (ADA) for current and futureARDI levels. The data and extrapolations weregrouped by source of assessment and by age ofassessment. The estimates indicated that of the$246.1 billion in gross FY 1998 ARDI, the IRScould expect to collect $41.4 billion over the nextten years, or just under 17 percent. This impliedan ADA for gross FY 1998 ARDI of about 83percent.

RESEARCH ABSTRACTS


110

Decline in BMF TDA Inventory in Los Angeles District

By Scott MendelsonLos Angeles District Office

Research and AnalysisMarch 1997

BMF TDA inventory, primarily comprised of employment tax accounts, declined unexpectedly compared toIMF TDA inventory in Los Angeles District during fiscal years 1996 and 1997.

This study investigated causes of the substantialchange in BMF (Business Master File)/IMF(Individual Master File) Taxpayer DelinquencyAccount (TDA) inventory mix during fiscal years(FYs) 1996 and 1997. BMF balance due accountscomprised 54 percent of the total TDA inventoryduring the first quarter of FY 1996. However, IMFTDAs increased by more than 900 percent inJanuary 1996 and by another 600 percent inSeptember 1996. Following these changes, only 39percent of the total balance due inventory consistedof BMF TDAs.

The research uncovered four reasons for thischange. The first relates to the Resource WorkloadManagement System (RWMS) “queue” criteria.Analysis showed the three Western Region districtswith the lowest RWMS criteria (one being LosAngeles) had the highest percentages of IMF TDAinventory.

Second, more than 50 percent of Form 941(employment tax) TDAs in the Automated CollectionSite (ACS) for Los Angeles District had been therefor at least 6 months, and 36 percent had been inACS for at least 1 year. Many of these were overduefor transfer to TDA status, thus lowering BMF TDAinventory. Also, a repeat delinquency strategy forcertain Form 941 taxpayers had been in place sinceJune 1994, which further impacted the BMF TDAinventory in Los Angeles.

Third, the match of BMF Tax DelinquentInvestigations (TDIs) with a database provided by theState of California Employment DevelopmentDepartment (EDD) allowed for a more expedientclosure of ACS, queue, and Collection Field function(CFf) TDIs relating to taxpayers that have gone out of

business. This process prevented quite a few BMFTDIs from erroneously evolving into BMF TDAs. Atotal of 11,385 BMF TDIs were closed during FY1996 as “final” based on this match, with anadditional 7,654 BMF TDIs closed during the firstquarter FY 1997.

Finally, the economic downturn of the early 1990s,particularly severe in Southern California, contributedto the reduction in BMF balance due accounts. Thiswas indicated by a downtown office vacancy rate inLos Angeles County that was well above the nationalaverage, and a record number of bankruptcy filings,including a Chapter 7 (No Asset) increase of 475percent in just one year. High city taxes and feesleading to an exodus of businesses from this area,high numbers of business closures, and a smallernumber of newly formed businesses also contributedto the reduction in BMF TDAs.

RESEARCH ABSTRACTS


111

Business Licensing as Nonenforcement Approach to Increasing TaxCompliance in the Liquor Industry

By Nancy RichmanNorth-South Carolina District Office

Research and AnalysisScott Mendelson

Los Angeles District OfficeResearch and Analysis

November 1998

The South Carolina legislature enacted a bill that requires full compliance with federal and state income taxes priorto granting a new liquor license. Results of this legislation have encouraged efforts by the IRS and the State ofCalifornia Alcoholic Beverage Control (ABC) to enact a licensing agreement.

The North-South Carolina IRS District OfficeResearch and Analysis (DORA) researched theeffect of licensing legislation on taxpayer compliance.In 1993 South Carolina implemented legislation thatrequires local and federal tax compliance forbusiness licensing. Since only individuals may applyfor liquor licenses in South Carolina, the legislationaffects only the compliance of the individualapplicant, not the business entity using the license.As an apparent result, from 1992 to 1994, thenumber of applicants not related to the businessusing the license increased from 15 percent to 21percent.

Using a random sample of liquor license applicants in1992 (pre-legislation) and 1994 (post-legislation),researchers found the percentage of licenseapplicants who failed to file Individual Income Taxreturns by late 1996 declined from 12.6 percent forthe 1992 applicants to 8.8 percent for the 1994applicants. Late, or overdue, filing declined from14.4 percent for 1992 to 12.1 percent for 1994applicants. The percentage of license applicantswho did not pay their income tax by the due datedeclined from 10 percent for 1992 to 5.5 percent for1994 applicants.

None of these compliance measurementsshowed a statistically significant improvement at the95 percent confidence level between the pre- andpost-legislation years. However, liquor licensees inSouth Carolina were relatively compliant, comparedto other market segments, even before enactment ofthe

legislation. Lack of significant improvementin compliance also may be attributable to the factsome license applicants may not have been aware ofthe fairly new law, and perhaps applicants had moretime to file their 1992 returns before the filingmeasurement was taken for the purposes of thestudy.

In related research, the Los Angeles and SouthernCalifornia DORAs profiled the liquor industry in a sixcounty area of Southern California. The primarypurpose was to identify the types and levels ofcompliance within the population of liquor licensees,which in turn could be used to help IRS and the Stateof California Alcoholic Beverage Control (ABC)consider a licensing agreement or other strategy.Findings indicated there is room for improvement inthe area of compliance. The percentage of liquorlicensees who have filed all required Tax Year 1994income, partnership and employment tax returns is83.7%, while only 80.0% have done so timely. Thepercentage of licensees who have filed timely andpaid timely (all required returns) and have madetimely Federal Tax Deposits is only 58.4%.

One of the categories for which the State ofCalifornia ABC suspends or revokes liquor licensesis “moral turpitude.” This currently does not include arequirement for a licensee to remain compliant withtax laws and regulations. IRS has begun using theresults of this research to suggest states makeFederal tax compliance a condition for securing orrenewing a liquor license.

RESEARCH ABSTRACTS


112

Analysis of Toll-Free Telephone Demand

By Jeff ButlerResearch and Statistics of Income

Office of Planning and FinanceProjections and Forecasting Group

November 1997

A comprehensive analysis of toll-free telephone demand led to recommended changes thatincreased national access rates from 62 percent in 1997 to 91 percent in 1999.

At the request of the Executive Officer forCustomer Service, a comprehensive analysis oftoll-free telephone Level Of Access (LOA) wasdeveloped for the 1040, 8815, and 4262 programareas— tax law, notices, and refunds,respectively. Using detailed call volume datafrom 1996 and 1997, this study examined: 1) theimpact of using new LOA criteria in 1998 on toll-free performance and tracking measures; 2) thedynamics of access rates within a peak/off-peakframework; and 3) the interrelationships betweenaccess rate factors such as overflows, abandons,and calls answered. Combining the results fromthese areas ultimately led to a quantitativeanalysis of how to raise the Level of Access.

For example, an aggregate model combiningdata for all three program areas showed that forevery 1 percent increase in the percentage ofcalls answered, the abandon rate drops 0.2percent, the ratio of overflows to abandons drops0.15 percent, and the LOA increases by 1.1percent. For every 1 percent decrease in theabandon rate, the LOA increases 6.8 percent;and for each one unit decrease in the ratio ofoverflows to abandons, the LOA increases byalmost 10 percent. A more detailed analysisalong these lines was conducted for eachprogram area separately in a peak/off-peakframework. Isolating factors in each time periodeither increased or decreased the LOA.

Using this approach, an average annual LOAcould be computed for a given percentage ofcalls answered. In the 8815 program area, for

example, the models showed that 65 percent ofall calls needed to be answered for an average

annual LOA of 80 percent. Since only 34.4percent of calls were answered in 1997, theimplication was a needed increase in staffingproportional to the increase in calls answered toreach a 71 percent LOA. (Adjustments weremade for the fact that this relationship wasnonlinear.) Scenarios were developed for targetLOAs of 75, 80, and 85 percent, andrecommendations made for increases orreallocations in staffing by time period. Theserecommended changes were put into place in FY1998, during which each program saw a dramaticimprovement. For the three program areascombined, the IRS increased its LOA from 62percent in 1997 to over 90 percent the followingyear.

RESEARCH ABSTRACTS


113

Taxpayer Service Walk-in Study

By Curtis R. DarlingOhio District Office

Research and AnalysisOctober 1996

An Ohio DORA and Taxpayer Service study of walk-in customers revealed taxpayers continue to look towalk-in sites as a source of tax forms, and promotion of VITA and TCE services could reduce substantially thedemand for return preparation services at walk-in sit

The Ohio District Office Research and Analysis(DORA) and Taxpayer Service Divisionconducted a study to secure informationregarding walk-in customers. Taxpayer assistorsrecorded the type of assistance provided and theZip Code of the taxpayer on log sheets for nineweeks – six weeks during the 1996 filing periodand three following the filing period. The studyincluded nearly 50,000 filing period contacts (30percent of all 1996 filing period contacts) and7,500 post-filing period contacts (12 percent of all1996 post-filing period contacts). In addition,information from a survey initiated by InternalAudit at walk-in sites in the Northeast Region,including six sites in Ohio, was included in thestudy.

The study found 51 percent of all contacts duringthe filing period were for tax forms and 30 percentwere for return preparation services. The percentof contacts for these services after the filingperiod dropped to 23 percent and 10 percent,respectively. Profiles for each walk-in siteprovided increased insight into why taxpayersvisit particular sites. Contacts regarding formsdistribution, return preparation, notices, lienclearances, individual income tax and business-related tax questions, filing issues, highway usetax and total visits were displayed in a collectionof tables, colored graphs, and maps that allowedIRS management to identify what taxpayerpopulation were served at each site.

Based on this information, locations wereidentified where an increase in Volunteer Income

Tax Assistance (VITA) and Tax Counseling forthe Elderly (TCE) services could reduce thedemand for return preparation services. Sitesalso were identified which would be ideallocations to develop or improve alternativemethods for tax form distribution, therebyreducing the demand for walk-in services relatedto tax forms. A self-help kiosk to distribute taxforms was one of the possible applications atthose sites that have a high demand for taxforms.

RESEARCH ABSTRACTS


114

Selected Estimates of Returns with Form 1099 Information

By Terry ManziResearch and Statistics of Income

Office of Planning and FinanceProjections and Compliance Studies Group

June, 1999

This analysis estimated which types of information documents were prevalent among individual returnsfiled on or before February 15.

The IRS Restructuring and Reform Act of 1998required that a study be done of the impact ofextending the deadline for providing taxpayerswith copies of information returns (other thanForms W-2) from January 31 to February 15.Staff from the Office of the AssistantCommissioner (Electronic Tax Administration)were responsible for preparing that report. They,in turn, requested that the Projections andCompliance Studies Group develop a profile ofthe early return filers in terms of their informationreturn characteristics.

The Projections and Compliance Studies Groupanalysis provided the following broad profile, interms of approximate return filing volumes incalendar year 1999:

§ number of returns filed by February 15reflecting strictly wages and salaries (i.e.,Form W-2 information, only) -- 16,649,000;

§ number of returns filed by February 15 with atleast some "Form 1099-type" information(i.e., other than Form W-2) -- 14,807,000

§ number of "Form 1099-type" returns filed byFebruary 15 with interest income(Form 1099-INT) -- 10,168,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with unemploymentcompensation or taxable state refund (Form1099-G) -- 5,006,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with home mortgage interestdeduction (Form 1098) -- 4,104,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with pension/IRA distributions(Form 1099-R) -- 2,937,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with taxable social securityincome (Form 1099-SSA/RRB) -- 1,144,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with dividend income(Form 1099-DIV) -- 1,868,000;

§ number of "Form 1099-type" returns filed byFebruary 15 with capital gains information(Form 1099-B) -- 1,492,000

RESEARCH ABSTRACTS


115

Alternative Signature Methods for Filing Individual Tax Returns

By T. Scott Shutt,Robert A. Kerr and

Dennis L. RaupResearch and Statistics of Income

Artificial Intelligence Lab 2November 1998

Both Voice and Digitized Signature methodologies are viable as alternative signature methods to create atotally paperless filing experience.

Beginning in 1995 two alternative signaturemethods were tested to provide information as tousing voice or a dynamic digitized signature tomake electronically filing individual income taxreturns totally paperless. The impetus behindthese efforts was to (1) broaden electronic filingopportunities, (2) eliminate the required papersignature document which is costly to processand unwieldy to manage, and (3) provide aforensically sound electronic signature as legallybinding as a pen to paper signature.

The voice signature test required taxpayers, oncetheir returns were filed electronically, to initiate atelephone call, state their name and identifyinginformation, and read the affirmation text from thereturn. The affirmation text and the nameconstituted the signature. All signatures weretested for cooperation, identification and potentialfraud (as an impostor or a multiple filer) usingspeaker verification and recognition techniques.Test forgeries were introduced into the testpopulation for fraud tests.

Focus groups with the test participants indicatedlittle aversion to using voice as a signaturemethod. Results of operations-oriented researchindicated the use of voice was technically feasibleas an alternative signature method; however thelarge size of the stored voice signature wouldhave a significant impact on storagerequirements if implemented. Speakerverification easily detected cooperativeness.Speaker recognition easily determined impostors

during the fraud tests but had more difficulty withthe multiple filers. Also, the test indicatedmultiple samples of the same person’s voice overseveral years could become a fraud detectiontool that could provide indications of suspiciousreturns quickly. The use of voice as analternative signature method is viable but needsmore study. Voice is a no cost solution to thetaxpayer and practitioner in creating a totallyelectronic return. Voice could provide additionalfraud detection and prevention tools.

The Dynamic ELF Signatures Test (DigEST)used routines embedded in tax preparationsoftware to capture the taxpayers’ signature froma digitizing pad at the time of tax returnpreparation.

The major conclusions from the DigEST were (1)the digitized signature technology works andprovides more security than a paper signature,(2) the digitized signatures are not as strongforensically as paper signatures but are as strongas any electronic method in a criminal trial, (3)the public accepts digitized signatures whenpresented positively, (4) IRS can processdigitized signatures successfully, at much lowercosts than paper signature documents, and (5)practitioners must perceive a business reason forusing any electronic signature or they will not useit.

The test demonstrated DigEST is a workablesignature methodology for totally paperless taxfiling.

RESEARCH ABSTRACTS


116

Focus Group Report: On-Line Filing Program

By Dru DeLong andDavid Browne

Management and AdministrationStrategic Planning Division

July 1997

Focus group research revealed taxpayers’ perceptions of benefits and barriers to individual tax return On-Line Filing.

Trained moderators from the Strategic PlanningDivision conducted six focus group interviews withtaxpayers eligible to participate in the On-LineFiling Program but did not. These focus groupswere conducted at the request of thePennsylvania Taxpayer Education/ElectronicFiling Staff.

Roughly half the participants knew filing incometax returns using home computers was possible,but knew no detailed information beyond that.Most expressed a willingness to use thisalternative method of filing. They believedtransmitting to IRS directly rather than through athird party would remove most of the barriers toparticipating in this program. Researchers alsofound the following.

• Two barriers were common to most of theparticipants: mistrust of data security andthe yearly cost of tax preparationsoftware.

• Publicizing information about the serviceprovider, explaining security measures,and establishing a procedure ofaccountability could alleviate the datasecurity barrier.

• A clear benefit over filing a paper return isreceiving notification that IRS received thereturn. Other benefits included recordstorage on disk rather than paper, andreceiving refunds faster.

• To be most effective, a marketingcampaign should provide informationabout the on-line Filing Program, how itworks, its requirements, and where to getmore information.

RESEARCH ABSTRACTS


117

1. Detail may not add due to rounding.

2. Table 1 shows tax return and economic/demographic data for IRS regions and states.

3. Table 2 shows tax return and economic/demographic data for IRS regions and the new consolidated IRSdistricts.

4. The years displayed represent the calendar year in which the tax return was filed, except (table 3)Federal tax deposits, which are presented on a fiscal year basis.

5. Economic and demographic data projections were made in July 1999 by Data Resources, Inc.,(DRI)/McGraw Hill. This information is not available for the Assistant Commissioner (International). Formost recent economic and demographic projections, please direct questions to the contacts listed on theinside cover of this publication.

6. The selected employment by industry variables (e.g., construction employment, mining employment,etc.) do not sum to “Civilian Employment” since they are only a subset of all industries.

7. Federal tax deposit projections were made in March 1999. These figures are organized by the ServiceCenter Recognition Imaging Processing System (SCRIPS) alignments and include projections ofelectronic fund transfers as specified in the original North American Free Trade Agreement legislation.Withholding and information document projections were made in May 1999 and are also sorted by aSCRIPS grouping of service centers. The tax return projections reflect the fall 1998 editions of IRSDocument 6149 (districts and regions) and Document 6186 (service centers). For more recentprojections of tax returns, please direct questions to the contacts listed on the inside cover of thispublication.

8. Total returns consist of the following tax forms:

Individual Paper and electronic Forms 1040, 1040A, 1040EZ, 1040PC, 1040NR,1040PR, and 1040SS

Estimated Tax Form 1040ES

Fiduciary Form 1041 and Form 1041ES

Partnership Form 1065

Corporation Forms 1120, 1120A, 1120F, 1120FSC, 1120H, 1120POL, 1120REIT,1120RIC, 1120S, 1120PC, 1120L, and 1120SF

Estate Forms 706, 706NA, 706GS(D), and 706GS(T)

Gift Form 709

Employment Forms 940, 940EZ, 940PR, 941, 941E, 941PR, 941SS, 942, 942PR, 943,943PR, 945, and CT-1

Form 1042

Table Notes

RESEARCH ABSTRACTS


118

Form 8752

Exempt Organization Forms 990, 990C, 990EZ, 990PF, 990T, 4720 and 5227

Employee Plans Forms 5500, 5500C, 5500EZ, and 5500R

Excise Forms 11C, 720, 730, and 2290

Form 8752

Supplemental Documents Forms 1040X, 1120X, 2688, 4868, 7004, and 1041(A) prior to 1993

9. Withholding Documents consist of the following: Forms W2, W2P (prior to 1992) and W2G.

10. Information documents consist of the following: Forms 1098, 1099A, 1099B, 1099DIV, 1099G, 1099INT,1099MISC, 1099C, 1099OID, 1099PATR, 1099R, 1099S, 1099SSA/RRB, 5498, 1096, Schedules K-1and foreign information.