175

A P P L Y I N G

T H E N E W

A D A M M E T H O D

IIIIIIPART

* Phyllis J. Newton, M.A., of Abt Associates Inc., was the contracting Project Manager of the ADAM program; Margaret E. Townsend, M.A., is FieldOperations Manager with Abt.

A D A M 2 0 0 0 A N N U A L R E P O R T

177

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

VII. Implementing the New ADAMStudy Design at the Local Level

by Phyllis J. Newton and Margaret E. Townsend*

When the National Institute ofJustice (NIJ) decided to strength-en the Drug Use Forecasting

(DUF) program, it envisioned a way tomeasure drug use and drug-related behav-ior among arrestees that could withstandmethodological scrutiny and would be aneven better tool for local policymakers thanin the past. That meant developing a statis-tically sound method of data collection,improving the way the local sites collecteddata, enhancing the survey instrument(questionnaire), and increasing the numberof sites. The premise underlying the changewas straightforward: to build an infrastruc-ture that ensured standard data collectionprotocols; an unbiased, probability-basedsample of arrestees; and a data manage-ment system that generated standardizeddata for use by the sites.

The theoretical ideals underpinning thenew program, ADAM, have now beenapplied in the practical world of the jailenvironment in 35 sites nationwide, andthe program has had one full year of expe-rience administering a new collectioninstrument and probability-based samplingin all the sites. That application of researchin a real-world setting raised several ques-tions, which are explored here. Among thequestions are whether it is possible todevelop data collection protocols applica-ble in all jurisdictions, whether the ADAMprogram can ensure adherence to thesestandardized protocols, whether method-ologically sound sampling strategies can beimported into local jails and still retaintheir scientific rigor, what adaptations themethodology can tolerate before it no

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

longer meets ADAM’s standards, andwhether probability-based sampling canwork in a jail environment.

Explaining the full the transition from DUFto ADAM requires discussing why standard-ization is important; the reasons for proba-bility-based sampling; how the new, county-level and facility (jail)-level samplingdesigns were implemented; and what chal-lenges are posed by the jail environment.Once the transition is complete and all sitesoperate with probability-based sampling forfemale arrestees as well as male arrestees,ADAM should have even greater potentialfor generating information that will assistlocal officials at the sites in making policydecisions affecting these at-risk populations.

From DUF to ADAMUntil ADAM was established, researchershad never before attempted to use the settingof the jail as the focus of ongoing, standard-ized data collection and application of rigor-ous sampling procedures. NIJ did so, creat-ing a program at multiple sites nationwidethat included the following components:

■ Data collection procedures common to allsites

■ Probability-based sampling that wouldallow the sites to place confidence inter-vals around their findings on drug abuseand related behavior

■ Enhanced data collection capabilities,with questions about drug treatment anddrug markets

■ The ability to compare ADAM data tothose from national surveys of drug use.1

178

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

The importance of standardization, orcommon data collection procedures, can-not be overstated. Regardless of the sophis-tication of the sampling process, the find-ings are not reliable unless procedures arethe same and carried out the same way inall the sites. All interviewers must admin-ister the survey questionnaire (instrument)in the same way in all the sites. No step inthe data collection process can be omitted.The population from which the data aredrawn must be identifiable and the sametype in all sites. Sampling must always beconducted the same way.

The data collection instrument, or ques-tionnaire, was expanded from its initialfocus on drug use, and now covers treat-ment and the dynamics of drug markets.And with the greater methodologicalsophistication of probability-based sam-pling, the interview also needed to includeinformation required for weighting casesso that they are generalizable to a largerpopulation—the county. Thus, ADAMneeded a mechanism for estimating suchinformation as the number of arrestees inthe county who used drugs and the num-ber who needed treatment. And becauseADAM data are a measure of drug useamong a limited spectrum of Americans,NIJ decided to build into the new collec-tion instrument the ability to comparefindings to those of other, national surveysof drug use.

Achieving standardizationCreating standardized data collection pro-cedures involved first making sure the“catchment areas,” or regions from whicharrestees are drawn at the sites, are definedthe same way in all the sites. Because arepresentative mix of the types of offensescommitted is needed for the sample, stan-dardization also meant resolving varyingways in which the sites define and dealwith crime.2

Redefining the catchment areasArrestees who participate in ADAM areselected from people brought to bookingfacilities—generally a jail. In DUF, the sitescollected data from arrestees booked at one

or two jails, but they did not necessarilyreflect all arrestees in the community. Forexample, in Philadelphia data were col-lected in one facility, but because there areseven, the data could not represent druguse among all people arrested citywide. InLos Angeles, data were collected at twobooking facilities out of nearly 100 in thecounty. At the other extreme, in NewOrleans Parish (equivalent to a county), thesheriff’s department operates the sole hold-ing facility, so it is possible to make asser-tions about the generalizability of the datato the county’s arrestee population.

In fast-growing cities in the West, the num-ber of facilities grew with the population,but data continued to be collected in onlyone. DUF began operations in 1988.Phoenix, whose population increased 40percent in the past 10 years,3 continued tocollect data at one facility. In sprawlingcommunities in Texas, California, and else-where in Arizona, as metropolitan areasgrew and encompassed more and morelocalities, it became less clear which local-ity or localities the arrestees at the centraljail represented.

ADAM defined the common catchmentarea as the county. County lines generallyserved as a reasonably common demarca-tion, though there were exceptions.Atlanta, for example, extends across twocounties, but because the site felt the cityshould be covered as a single entity,ADAM included both Fulton and DeKalbcounties in the definition of this catchmentarea. Because the city and county ofPhiladelphia are coterminous, the city lim-its define this catchment area. New YorkCity consists of five boroughs, making itnecessary to attempt data collection in allof them.4

Not all counties are the same. Sites in theWest manage catchment areas that are con-siderably larger, geographically, than in theEast, with local law enforcement practices,such as deployment of officers, sometimescontingent on the number of miles to coveror amount of time it takes to bringarrestees to holding facilities. The catch-ment areas also vary in the number oflocal, county, and State law enforcement

officials having arrest authority, with theresult that procedures might differ by facil-ity. There were also variations in the num-ber of booking facilities capable of holdingarrestees. The ADAM program aimed toensure that each booked arrestee in a catch-ment area had some probability of beingselected for participation, and these twovariables affected that probability. In allsites, the first step for ADAM was to findout how many booking facilities there werein the defined area and where arrestingofficials took arrestees to be booked.

Variations in “arrest” and otherterms Every State and local jurisdiction in thecountry has its own laws and system of jus-tice, and while there are commonalities,the system of legal requirements in eachreflects local conditions. For example, whatone jurisdiction calls “breaking and enter-ing” another may call “burglary.”

What one community may call an arrestanother may call a citation; for some com-munities booking and arrest are synony-mous. Cite and release in one communitymay mean a street release/field release andin another may mean coming to the sta-tion/facility for booking and release. Inorder to understand which arrestees con-stituted the sample in each site, ADAM

needed common definitions so that seg-ments of the arrestee population were nei-ther over nor underrepresented.

The goals of ADAM dictated that allbooked arrestees—from low-level misde-meanants to serious felons—have someprobability of being selected for inclusionin the sample interviewed. If arrestingauthorities in one community bring allarrestees to booking facilities, the result forADAM’s purposes would be a reasonablemix of types of offenses among thearrestees. However, if another communitycites and then releases most arrestees whocommit only misdemeanor and city ordi-nance offenses, the arrestee population thatremained to be interviewed would overrep-resent those booked for more seriousoffenses. Even “misdemeanor” and“felony” are not defined the same by alljurisdictions. In some, such a label referssolely to the potential length of a jail sen-tence, while in others it refers more gener-ally to the seriousness of the offense.

Adopting probability-based samplingThe ADAM procedure was also redesignedto account for the variations in the structureand size of local criminal justice systemsand processes. That involved designing

A D A M 2 0 0 0 A N N U A L R E P O R T

179

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

Setting and Ensuring Performance StandardsTo promote standardization of ADAM data collection at the sites, NIJ established basic criteria for data col-lection procedures. At each site, the agency also implemented training in the procedures.

The contractor that administered the ADAM program (Abt Associates) developed the training materials thatwere used at all sites. It ensured the procedures for data collection were followed and observed data collec-tion. Editing instructions for the questionnaires were provided to all sites as a tool for ensuring quality data.

Common performance standards were established for all sites. They included requirements for meetingdata collection targets (number of arrestees), for training, for minimum error rates in conducting inter-views, and for adherence to a specified fiscal standard. A sampling plan to be used by all sites was devel-oped and implemented.

When data collected from the interviews at the sites were received at the ADAM Data Center, they were allsubjected to the same data management procedures by means of automated editing and entry programs.

By establishing these quality controls, NIJ anticipated that comparison of findings from site to site wouldbe more defensible than in the past and that ADAM data would be that much more useful in informing localand national discussions of substance abuse.

180

The four sampling modelsVariations among the catchment areasrequired developing a county-level sam-pling model that was flexible enough to beapplied to the specific counties/sites. Thisin turn required that no matter the numberof booking facilities in a county, each coun-ty would have a sampling plan that gener-ated estimates that could be extrapolated tothe entire arrestee population of the coun-ty. For counties with only a single bookingfacility, this was easily accomplished, butfor those with multiple facilities, ADAMdeveloped a procedure that would not biasthe sample against certain facilities or cer-tain types of offenses.

To accommodate the variations, four sam-pling models were developed. The simplestplan was the “single jail” design, in whichcollection took place at one jail only. Onlyslightly more complex was the “stratified”design, used for counties/sites with six orfewer facilities. This design called for datacollection at all jails, with target numbersof arrestees selected proportionate to thenumber of arrestees processed at each jail.

For counties with more than six jails, a“stratified cluster” sample design wasdeveloped. In this type of design, every jailwas assigned to one of a small number ofstrata, with one or two jails sampled fromeach stratum. This design generated esti-mates for all jails, even though only somejails were included in the sample. In a fewsites, this model needed to be furtherrefined into a “feeder” design. It wasapplied in counties where a large numberof jails quickly transported arrestees to acentral holding facility, which for ADAMreduced the probability of interviewingarrestees in the outlying jails to virtuallyzero. In counties where the feeder designwas used, interviews took place at the cen-tral holding facility (which represented alljails in the county) as well as at the “feed-er” jails, so that arrestees who were nottransported to the central facility could besampled.

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

county-level sampling models and strate-gies for sampling arrestees in each selectedfacility.

The goal of sampling in ADAM is to beable to estimate with a known probability,the likelihood that an arrestee will beselected for the ADAM interview and to beable to use that information to weight thesample data.5 To avoid biasing the samplein favor of the types of arrest charges andtypes of arrestees represented at the facilityduring the times when interviewers are notcollecting data, the sampling plan mustreflect all days of the week and all times ofthe day. Further, the plans must representall of the facilities in the county, whetherlarge or small, or urban, suburban, or rural,in order to represent all types of offendersand offenses.

The “stock” and the “flow”For each adult male arrestee booked in a24-hour period,6 ADAM had to determinethe probability of his being selected for thesample. The approach had to be one thatcould be readily implemented in 35 sites inwhich there were in all more than 100booking facilities. The easiest way—sam-pling and interviewing around the clock—would not be cost-effective. Another optionwould be to choose blocks of time random-ly during the day and sample arresteesbooked during those time periods. This wasalso too costly.

The sampling plan adopted covered thefull 24-hour period by splitting the day intotwo parts: a “flow” and a “stock” period. Inthe flow period were the arrestees bookedduring each daily ADAM data collectionshift—the 8-hour period when the flow ofarrestees was highest. In the stock periodwere arrestees booked during the 16 hourswhen data collection did not occur. Usingsystematic selection procedures, interview-ers sampled from both groups, with theresult an arrestee sample that representedeach 24-hour period of the one- to two-week ADAM data collection period.

To determine which model was appropriatewhere, the ADAM sites were asked to iden-tify every booking facility in the countyand furnish information about the numberof adult males booked at each in the pastyear. These two pieces of information wereused to select facilities and facility sampletarget numbers, proportionate to size.

In hindsight, it would have been helpful tohave worked with each site to documentthe movement of arrestees through the vari-ous booking facilities in the county beforedeveloping county-level sampling plans.Had that been done, it would have beenpossible to adjust sampling plans to accom-modate the county-specific variations, andthus implementing the plan would havebeen easier. However, that would havedelayed implementation and would nothave guaranteed immunity from other idio-syncracies.

Weighting the data to ensure representativenessIn order to demonstrate the extent to whichnumbers in a sample represent a largerpopulation, they are weighted. In ADAM,this requires identifying the probability ofeach adult male arrestee’s being includedin the sample, based on information aboutthat larger population. The ADAM datarequired a nontraditional approach toweighting. The main reason is that there isno way to know, before sampling takesplace, who will be arrested and thus noway to assign a probability of any givenarrestee’s being included in the samplingframe (the list of cases of interest). Theissue was resolved by using post-samplingstratification to identify the probability ofinclusion in the sample of like groups ofarrestees.

This required obtaining information aboutthe total population arrested during eachADAM data collection period in each site(the “census”). In ADAM, all sites provid-ed, as the census data, information aboutall adult males booked into the facility

during the one- to two-week ADAM datacollection period. These data include eacharrestee’s date of birth, race/ethnicity, dateand time of booking, and description of thearrest charge and its severity. They are usedto match the sample of arrestees inter-viewed with the larger population and toseparate the population into strata based onthese characteristics. The number ofarrestees in each stratum was then counted,and the probability of their being includedin the sample was determined in compari-son to that of everyone else in that stratum.Essential to this approach was ensuringthat bias could be minimized; that is, thatfor everyone in each stratum the probabili-ty of selection was the same.

Using “census” dataSeveral assumptions were used in weight-ing the data:

■ Arrestees charged with more seriouscrimes spend more time in the jail facili-ty than those arrested on less seriouscharges.

■ Arrestees booked at the same time of dayare processed similarly; that is, they allspend approximately the same amount oftime in the jail before arraignment and/ortransfer to another holding facility.

■ The stock and flow model (describedabove) may mean more serious offenderswill be over-represented in the stock pop-ulation, while the flow sample shouldrepresent the full range of charges. Thus,an assumption was made that for all peo-ple arrested on like charges and bookedat about the same time (stock or flow),the probability of selection for the sampleis similar.

■ Because the number of interviewers isstatic, the day of the week affects theprobability of an arrestee’s being selected.For arrestees booked on days when manypeople enter the system, the probabilityof their being selected is lower than forthose booked on slow days.

A D A M 2 0 0 0 A N N U A L R E P O R T

181

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

182

In stratifying the population, time of book-ing is used to separate stock and flow, andcharge severity and day of the week areused to further separate the population intostrata having approximately equal probabil-ities of selection. For example, all felony“stock” booked on Friday and Saturdaymay be in one stratum, while misde-meanants booked Monday throughWednesday are in another.

The importance of census data to theweighting process makes accuracy andcomprehensiveness of those data essential.Some sites find it more difficult than othersto meet these requirements. Some facilitieseither do not store booking informationelectronically, or do not allow the jail data-base to be queried at the user end. Forthese sites, census data must be collectedmanually. Thus, on each day of ADAMdata collection, in addition to orchestratingthe sample selection and the interviewing,the ADAM site staff keeps a running censusof all adult male arrestees booked into thejail. They then submit these data to theADAM Data Center at the end of the one-or two-week collection period. Althoughthis is typically the procedure in small or“low-flow” facilities, some jails where caseflow is very high are not automated andrequire this manual approach.

For sites with only one facility or for thosethat use a stratified sampling design(because they have six or fewer facilities),all facilities are sampled, because there areso few. However, sites that use stratifiedcluster design (because they have morethan six facilities) and feeder designs mustsubmit census data for all booking facilitiesin the county. For a few sites, where thecounty maintains a countywide criminaldatabase, this is not an arduous task. Forother sites, collecting census data from allfacilities in the county is difficult and thecost prohibitive. When it is not possible forthem to collect census data, annual book-ing statistics for each facility or county-wide booking data are used to developannualized countywide estimates.

Resolving census problemsThree problems that can adversely affectthe weighting process often arise after thecensus data are transmitted to the DataCenter. The data may include ineligibles,duplicates, and inconsistently recordedbooking times.

Ineligibles and duplicates. Not everyonebooked into jail is eligible to participate inADAM. (Those ineligible include Federalholds, extradition holds, and court holds,all of which involve people detained in alocal facility before trial). ADAM has toensure that ineligibles are identified andremoved from the sample so that they donot inflate the number of arrestees in thecounty. For example, an offender releasedon bond who arrives at court for arraign-ment may be remanded into custody andbooked into the jail for holding pendingtransfer to another facility. This personwould not be eligible for ADAM, but inmost facilities would appear in the bookingsystem. Some facilities track the type ofbooking and can include it as a variable insubmitting their census data. In othercases, eligibility for the ADAM sample canbe deciphered only by reviewing the chargeand the information from the arrestingagency and making educated guesses. It isthe sites that confirm that ineligible popu-lations are excluded from the census sub-mission (and that there are no missingcases).

Duplicate cases also inflate the arresteepopulation. Cases are duplicated whenmore charges are added to a previous book-ing of an arrestee or an arrestee’s use of analias is discovered. The sites will need tolearn to recognize the potential for dupli-cates and work with the ADAM DataCenter to merge or delete duplicate records.

Inconsistent booking times. There can bevariation in and confusion about the defini-tion of “booking time” at the local level.This is a variable essential to the weightingprocess because in order to assign cases tothe correct stratum, the time of the day

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

when arrestees are selected for data collec-tion must match the time the census dataindicate as booking time. However, in somecases, data submitted to the ADAM DataCenter indicate booking times differentfrom the times when the ADAM sample isselected. For example, many sites use thejail’s intake log to identify stock and flowand select their sample by using intaketime as a proxy for booking time. In thecensus data, the booking time indicatedoften reflects the time at which bookingdata were entered into the facility’s com-puter system, rather than the intake time.This poses a problem for weighting becausein some cases data are not entered into thebooking system for several hours afterintake, which means some stock might beweighted as flow and some flow as stock.In such situations, the sites will work withthe Data Center to find a solution to thediscrepancy.

The county-level sampling plandesignIn adopting probability-based sampling forADAM, there are two stages in planningthe sampling at the sites: the broad, county-level stage that determines a site’s generalsampling design, and a more specific, facil-ity-level stage that specifies the actualmechanics of drawing the sample. To deter-mine which of the four sampling strate-gies/models is to be used in a site, it wasfirst necessary to identify the number ofbooking facilities in each site’s catchmentarea. Once this number was known, each sitewas “labeled” with the type of county-levelsampling plan to be used in that catchmentarea (single-jail design, stratified design,stratified cluster design, or feeder design).

Implementing the county-level samplingdesign involved setting target numbers ofarrestees to be included in the sample;understanding who, on the basis of variouscharacteristics, the population of arresteesin county facilities represent; and ensuringfiscal accountability. Each ADAM siteneeded its own unique sampling plan,

A D A M 2 0 0 0 A N N U A L R E P O R T

183

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

based on the number of people arrestedand booked in the county and on the pro-cedure each arresting agency follows afterciting or arresting suspected offenders. Andalthough the plans had to be unique, adher-ence to standardization required that varia-tions among the sites did not compromisethe overall sampling principles.

Setting target numbers for thesampleThe basis of the sample size was the totalnumber of bookings in the county and thenumber of facilities in the county, becausethis permits calculating a sample equiva-lent to the variance that results from sam-pling proportionate to size. In order to per-form this calculation, the sites needed toprovide information about the total numberof arrestees booked in each facility in thecatchment areas, or at least the total num-ber of people booked in the county.However, this information was not alwaysavailable to the site staff and, in siteswhere booking data were not available, theFBI’s Uniform Crime Reports (UCR) wereused to document the number of arrests ineach county.

UCR data theoretically include all arrestsnationwide, but they have well-known lim-itations that derive primarily from differ-ences in reporting. Other limitations in theUCR data meant they could be used toidentify arrest numbers in only someADAM counties. First, the UCR includesarrests that do not result in bookings. Thisis principally the case for minor crimes inwhich the individual receives only a cita-tion. Second, the UCR excludes somearrests that do result in bookings. Warrantsand revocations are examples. Third, somearrests are double-counted in UCR data.

Despite these limitations, UCR data wereused in some sites to develop county-levelsampling plans. At the same time, ADAMworked with the sites to identify alternatesources of data to validate the preliminarytarget numbers. On the basis of these num-bers, ADAM set initial targets for each site

184

in number of adult male arrestees whoshould be interviewed, although this wasdone with the understanding that moreaccurate data, provided later, might lead tochanges in the sample sizes.

Defining the population to besampledSuccess in implementing the ADAM sam-pling designs requires understanding themovement of arrestees from one facility toanother and the length of time they spendat each facility. Arresting authorities identi-fied which facilities were those wherearrestees were booked. They also providedin-depth information about booking andarrest procedures, including potentialpoints of release in the field (that is, releasewhere the arrest took place) and releasefrom local booking facilities. ADAM need-ed to know about the extent of law enforce-ment’s discretion in arrest and release deci-sions in each jurisdiction and have someunderstanding of the sites’ transfer andhold procedures. It is important to know,for example, whether local booking facili-ties have holding capabilities and, if not,how soon and to what facility arrestees aretransferred.

In general, law enforcement agencies in allcounties can exercise some discretion inwhether to release on citation people whoviolate city ordinances. They can do soeither in the field or from local bookingfacilities. Whether an arrestee is processedin the field or is booked and released hasimplications for ADAM sampling, becauseit affects the size of the arrestee populationavailable to participate in the program.Arrestees released on a field citation arenot available to be interviewed. This makesit necessary to understand the categories ofarrestees who have no probability of beingselected and what proportion of thearrestee population is processed this way.

Just as important as understanding releaseon field citations is understanding the pro-cedures used to book and release arresteesfrom the local booking/holding facilities orstations. This includes knowing how much

time the procedures take and where bookand release occurs. Because the populationof arrestees who are booked and releasedgenerally consists largely of misde-meanants, it must be included in theADAM sample. Without it the representa-tiveness of the sample would be called intoquestion. Both options—field citations orbook and release—are typically available forprocessing the misdemeanor population butnot the felony population. Systematicallyexcluding arrestees who are released wouldheavily bias the sample against peoplearrested for minor offenses and would thuspotentially inflate the extent of drug useamong ADAM respondents.

Ensuring fiscal accountabilityCost was a consideration in redesigningADAM. At some ADAM sites, especiallythose where there are several facilities,expanding the catchment area to the entirecounty significantly increased the cost ofdata collection. Cost considerations necessi-tated a series of trade-offs between maintain-ing expenditures at a reasonable level andretaining the overall goals of the program.

The first trade-off involved setting the tar-get numbers of interviewees. All DUF sitestried to obtain 250 interviews per quarter, atarget goal that often took several weeks toachieve in some sites. Any increases ordecreases in this target number underADAM would have cost implications,favorable or unfavorable. The best justifica-tion of the cost of probability-based sam-pling was that it produced, at each site, asample size large enough to ensure a rea-sonable level of confidence in using thedata (in other words, a sample large enoughthat the level of variance would be accept-able). Logic dictated that the Los AngelesCounty sample would be larger than that ofWebb County (the primary city is Laredo,Texas), for example, but the question was:how much larger? The decision to basesample size on total number of bookings inthe county and the number of facilities inthe county (based on a sample equivalentto the variance resulting from samplingproportionate to size) met two needs. It

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

generated a sufficient number of cases fromwhich to estimate the number of people ina county who have certain characteristicsand to distribute the cases equitably amongthe sites.

Some smaller sites, understanding thattheir target numbers needed to be smallerthan those of larger sites, were at the sametime concerned that it might be difficult tomake credible statements about drug useand related behavior on the basis of thenumbers. A target of 75 cases, for example,might be large enough for sampling purpos-es, but local officials might be reluctant touse it to shape policy. Here the trade-offwas meeting local needs while developingstandard errors (such as those for confi-dence intervals) that were reasonable inrelation to those of other sites.

Cost considerations also affected the num-ber of interviewers working in each jail. Inorder to contain costs, it became necessaryfor NIJ to set the number of interviewersper shift per facility. In doing so, NIJ alsocreated a built-in mechanism for ensuringthat sites adhered to their own specificsampling plans. The plans emphasizedselecting the sample of arrestees to be inter-viewed rather than the total number ofinterviews completed. However, staff at thesites continued to focus on completing theprescribed number of interviews. The con-ceptual shift from quantity to quality wasdifficult for many sites.

Practical issues in implementingthe county-level designConceptually, the new approach to creatingcounty-level sampling designs was relative-ly straightforward: identify the number ofbooking facilities, establish the generaldesign, find out how many arrestees movethrough each facility each day, and set sam-ple targets for the county and each facilitywithin the county. In practical terms, thenew approach was not so straightforward.Except in a few sites, the UCR data werenot useful in identifying the number ofarrestees booked at each facility or thenumber of booking facilities in a county.

The sites found it difficult to obtain infor-mation about the flow of cases througheach facility and to gain access to the facili-ties selected.

Identifying facilitiesAlthough seemingly a straightforward task,finding out how many facilities there werein a catchment area was not easy. TheUniform Crime Reports do not contain thisinformation. Counts by county law enforce-ment authorities varied with the definitionof a booking facility. Further complicatingthe task of counting facilities was thatarrestees are often booked many times, firstin the local jail and again in the countyfacility. Thus, although county facility staffmight be correct in their assessment that allarrestees in the county are booked in thecounty facility, arrestees may be booked inlocal facilities as well.

These approaches failing, the one that suc-ceeded was to contact the source; that is,the arresting agency authorities in eachcounty. Each ADAM site thus systematical-ly contacted all arresting agencies in itscatchment area to determine wherearrestees are booked.

Gaining accessIn the past, jurisdictions wishing to partici-pate in the DUF or ADAM program submit-ted to NIJ letters of agreement from localjail facilities that ensured access to thosefacilities to conduct ADAM interviews.These letters generally applied only to theprimary county facility but did not guaran-tee the site would be permitted to collectdata in all facilities selected in the sam-pling plan.

To gain access to all facilities selected,local ADAM staff contacted facility admin-istrators, explaining the program and thereasons for including their facility. In gen-eral, jail administrators’ initial reaction wasto question the program or deny accessbecause it would delay the bookingprocess, interfere with operations, and raisesecurity concerns. Often, NIJ or NIJ’sADAM contractor intervened. When jail

A D A M 2 0 0 0 A N N U A L R E P O R T

185

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

186

administrators understood that the programhad been in existence for more than 10years and that numerous facilities nation-wide permitted data collection withoutadverse effects, they were more amenableto working out an arrangement. In a veryfew cases, access was still denied, andhigher law enforcement officials or othercity/county officials were contacted. In rarecircumstances, despite all efforts, therequest was denied.

When access is deniedWhen access is denied, replacing the site ora specific facility is easily accomplishedwithin the parameters of the samplingdesign. Increasingly, however, replace-ments are required repeatedly where thereare several small facilities in the stratum.One cause may be lack of security at thefacility. ADAM requires that to ensure safe-ty and security, law enforcement officersobserve the interviewers and arrestees dur-ing the interview process. Often, smallfacilities could not participate because of ashortage of officers. When a department isshort-staffed and the jail or booking area isnot sufficiently staffed to provide adequatesecurity for civilians (ADAM interviewers),the program cannot continue at that facili-ty. Even if an overtime incentive is offered,there may not be enough officers to workthe additional hours.

The other reason repeated replacementsmay be necessary is low case flow, whichcan make the cost of data collection prohibi-tive. Interviewers may be at the facility forseveral hours or even an entire shift withoutconducting a single interview, because noarrestees have been booked. In Bexar County(San Antonio), Texas, for example, there areapproximately 25 booking facilities, butmost book fewer than one arrestee per day.Again, the issue was resolved by trade-offs.The sampling plan was adjusted to elimi-nate facilities that produced fewer thanthree cases a day. In making all such trade-offs, the balance is between cost and risk tothe integrity of the overall sampling plan.

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

“Specialty” facilities Some jurisdictions have facilities wherepeople arrested for only certain types ofcrimes are booked. For example, there maybe a facility dedicated to booking peoplearrested on domestic violence charges. IfADAM interviews do not take place atthese facilities, people charged with thesetypes of offenses will not be included inthe arrestee populations for sampling pur-poses. On the other hand, if ADAM uses itsresources to interview at these facilities,only those specific types of arrestees willbe interviewed. The challenge is to reachthese arrestees before they are brought tothe specialty facility, but this is not feasiblein all jurisdictions. For now, the sites hav-ing specialty facilities must adapt by limit-ing the representative nature of their sam-ple through adding written caveats to theirfindings; for example, by making it clear intheir reports that their sample does notinclude domestic violence cases.

Defining the arrestee populationCounts of arrestees, essential to sampling,were difficult to ascertain. Arresting agen-cies maintain numbers for operational rea-sons, not for research purposes. ADAM hadto use the operational numbers as the basisfor constructing a mechanism to create arepresentative sample.

One reason it is difficult to count arrestees isthat their movement throughout a county isconsiderable. They are often transferredthroughout the criminal justice system,sometimes very quickly. Many factors con-tribute to this movement, including geo-graphic imperatives, municipal require-ments, and overlapping jurisdictions.Additionally, the criminal justice process hasseveral stages, with each one at a differentlocation. The possibility of double-countingwhen arrestees move is inevitable because agiven facility may not be aware that thearrestee has already been counted forADAM’s purposes. These factors affect allcounties, but become more problematic asthe number of law enforcement agencies andbooking facilities in the county increases.

Movement also means an arrestee whoshould be counted may not be. In mostfacilities, an arrestee’s booking sheet fol-lows him or her to each succeeding stage ofthe criminal justice process. Therefore, ifan arrestee has left the intake area or beentransferred to another facility, no record ofdemographic data and offense characteris-tics may be available. The booking log mayindicate that the arrestee was booked intothe facility, but if that arrestee (with hisrecords) has gone to court or another facili-ty, it would be difficult to include him inthe sampling plan.

In some ADAM sites, the individual idio-syncracies either in the facility popula-tions, the movement of arrestees withinand among facilities, and the bookingprocesses can only be documented. Thesesites must satisfy themselves that theirsample has limitations and must make itclear they exist. In some sites, such docu-mentation will lead to adaptations thatenable their samples to be consistent withthe overall sampling plans.

The facility-level sampling plandesign Before data collection began at the facili-ties, a facility-level sampling plan wasestablished for each one. The plans hadseveral steps: setting the targeted numberof interviews, determining what time ofday the interviews take place, and identify-ing the number of interviewers needed oneach shift.

Setting a targeted number ofinterviews The number of interviews to be conductedquarterly was identified for each site. Thiswas the site’s target number. Then, to iden-tify the number of interviews to be con-ducted at each facility in the site, the targetwas simply divided by the number of sam-pled facilities, proportionate to size. Forsites having stratified and stratified clusterdesigns (those with, respectively, 6 or fewerand more than 6 facilities), the county

sample target was divided among the strataand/or facilities, proportionate to the num-ber of bookings each contributed to thewhole. A site with a city jail and two sub-urban jails, for example, might have a sitesample target of 168 completed interviews.Annual booking statistics there might indi-cate that 50 percent of the county’s arresteepopulation is booked at the city jail and theother 50 percent at the two suburban facili-ties (25 percent at each). For the city jailthe sample target would be 84 completedinterviews and for each suburban jail thetarget would be 42.

Once the total sample per facility wasdetermined, the number of days requiredfor data collection was set. Collection musttake place every day of the week in order toaccount for variations, by day of the week,in the type of crimes for which arresteesare charged. The length of time data arecollected is based on the average number ofbookings per day. Using the example citedabove, and assuming the daily flow in thecity jail is significant, a 7-day collectionperiod, with a daily sample target of 12completed interviews, will meet the target-ed 84 interviews. (ADAM assumes eachweek in a given calendar a quarter is gener-ally like any other, so the sites do not col-lect data on holidays or during days whenspecial local events take place, such asIndependence Day, and Mardi Gras in NewOrleans.)

What time of day should data becollected? When the optimal time of the day for inter-viewing is determined, that becomes thedata collection shift. In the probability-based design, an 8-hour shift represents a24-hour period, and all arrestees bookedduring that period have a known probabili-ty of being selected for the sample.

The “stock and flow” design of the sam-pling plan (described above) increases thelikelihood that the sites will sample andinterview arrestees charged with lesseroffenses, whose numbers are typically larger

A D A M 2 0 0 0 A N N U A L R E P O R T

187

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

188

interviewer is assigned to stock and anoth-er to flow. The difference is that the flowinterviewer works the entire 8-hour shift,regardless of whether the target is met orsurpassed, and the stock interviewer worksto meet the daily stock quota and then endsthe shift.

This means the number of resources orinterviewers in a given facility will be con-stant, regardless of whether, on a given day,the flow of arrestees is high or low.Additionally, because an interviewer isalways needed throughout the flow shift,the probability is high that arresteesbooked in low-flow facilities will be inter-viewed. The interviewer can usually inter-view all of the remaining stock and eachflow as they are booked. The requirementthat the number of interviewers be keptconstant made implementation easier.Predicting daily flow activity in a jail andmodifying the number of interviewersaccordingly would have been difficult.Having the number of interviewers remainconstant also is important in weighting thedata because otherwise it would affectassumptions about the probability of selec-tion in a way that could not be predicted.

Practical issues in implementingthe facility-level design In implementing the facility-level design,there are a number of factors essential tothe success of the sampling. These includegaining access to booking data, identifyingarrestees ineligible for the sample, deter-mining when and where arrestees are to beinterviewed, tracking arrestees’ where-abouts as they are processed, ensuring theinterview space is secure and, in general,adapting to the jail environment.

Access to the booking data A site’s ability to implement its samplingplan is directly related to the facilities’ability to provide the necessary materialswith which to select stock and to establishan efficient process for identifying andinterviewing flow. Access to booking andcensus information is essential if the facili-ty is to participate in ADAM.

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

in a county arrestee population. Because ofthe less serious nature of their charges,these arrestees will be released more quick-ly than the more serious offenders andtherefore the window of availability forinterviewing them is smaller. The data col-lected from the “flow” cases are weightedto represent lower-level offenders bookedduring the “stock” period who werereleased before collection began. The suc-cess of this process relies on a site’s abilityto maximize data collection from arresteescharged with all types of offenses by col-lecting during the busiest 8-hour period ofthe day.

The “flow” period, which begins themoment the data collection team enters thefacility, represents the period of the daywhen the number of bookings is highest.After an interview is completed, the inter-viewer then selects the arrestee whosebooking time was closest to the time of thatinterview. This procedure ensures theinterviewer works throughout the shift,regardless of the number of interviewscompleted. For “stock” (which comprisesarrestees who were booked and whosenumbers accumulated during the timewhen data were not collected at the site),interviewers work with the facility todevelop a list of all arrestees booking dur-ing the stock period, organizing it chrono-logically by the time each arrestee wasbrought to the facility. Arrestees to be inter-viewed are selected at intervals determinedby the stock sample target.

Number of interviewersOnce the interview time (shift) is identi-fied, the sample targets for stock and floware calculated. The basis is the number ofdaily bookings estimated during each ofthese periods. For example, if 50 percent ofthe daily bookings occur during the flowtimeframe, a site with a daily sample targetof 12 completed interviews would haveflow and stock sample targets of 6. Thesetargets are the basis for determining thenumber of interviewers. In calculating thisnumber, the assumption is that one inter-viewer can complete approximately oneinterview per hour. In most cases, one

Stock samples are drawn from a list of allarrestees booked during the stock period(the period after the data collection shiftended the previous evening). Obtainingthis list can be particularly difficult forfacilities that do not have an automatedbooking system. It often requires consultingone or more handwritten logs. For exam-ple, the interviewer would use the intakelog, which identifies everyone booked dur-ing the stock period, and the inmate log,which identifies stock arrestees who arestill in the jail. ADAM site staff mergethese two lists to select the stock sampleand the replacements for released arrestees.

It is typically easier to obtain stock samplesin jails that have automated booking sys-tems. In the many facilities where bookingstaff do not have the authority to query thesystem, ADAM staff rely on command-levelstaff or department programmers to gener-ate a report that can be used to create astock list. However, many of these depart-mental reports either exclude arresteesreleased from the facility or do not coverthe full stock period, which often begins inthe middle of the night.

In many sites, the stock selection processdoes not end at this stage. This is becausethe information used to identify and selectstock often does not include such items asthe arrest charge, the specific location ofthe arrestee in the jail, and other basic vari-ables. Without this information, it is notpossible to know whether an arrestee isavailable or eligible to participate in theADAM program. Often, negotiations withfacility staff are necessary to develop a pro-cedure for obtaining information in a time-ly manner.

Because many facilities purge the bookinginformation after an arrestee has beenreleased, site staff often work with facilitystaff to manually look up charge and loca-tion information for each person beforebeginning the selection process. In othercases, site staff return later in their shiftand refer to the booking slips and/or thejail management system to fill in the infor-mation needed for the facesheet of thequestionnaire for selected and replacement

stock. This process is particularly time-con-suming in jails where the booking informa-tion physically accompanies the arrestee ashe is transferred to various locations in thejail for various purposes (for example,intake, fingerprinting, booking, classifica-tion, housing).

Flow selection is often easier because itrequires only access to arrestees as they arebooked into the facility. In low-flow facili-ties, where arrestees are booked sporadical-ly, this is often done by observation, but inmost facilities it must be more system-atized. In some cases, sites use an intakelist or medical screening list to identifyarrestees as they are booked or they use thebooking slips, as they are generated, toselect flow arrestees for interviewing. Themajor concern in flow selection is to ensurethat certain types of cases that might beoverlooked (for example, arrestees whosebooking sheets/cards take a relatively longtime to generate) are not.

The method used to develop the stock listshould also drive the flow selectionprocess, because the time of day whenstock and flow are selected must be thesame for both groups. In most ADAM sites,this is the time when the arrestee comesinto the facility (the intake/booking time)or the time when arrest and booking infor-mation is entered into the computer.Systematically recording time for allarrestees, no matter what charge or disposi-tion, is essential. Using different times canaffect the sample. Thus, for example, insome sites arrestees charged with minoroffenses might not be screened for medicalproblems. If a site uses medical screeningtime as the time for selecting stock andflow, lower-level arrestees may be systemat-ically excluded from the sample.

Identifying eligibles and ineligibles When drawing the sample, the ineligiblesmust be excluded. In some facilities, it isparticularly difficult to identify them.ADAM focuses on people arrested forcrimes committed in the local jurisdic-tion—the county. Thus, arrestees held forcrimes committed in other jurisdictions,

A D A M 2 0 0 0 A N N U A L R E P O R T

189

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

190

including other counties, as well as forFederal crimes (for example, arresteestaken into custody by the INS, DEA, FBI, orthe U.S. Marshal’s Service), or arresteesremanded into custody by the court, arenot of interest to ADAM. In general, theseineligibles are not represented in localarrest statistics or do not reflect an arreston a new charge.

Gaining access to the arrestees To implement the facility-level plan suc-cessfully, sites must understand the book-ing process and the restrictions it imposeson ADAM’s ability to access and interviewarrestees. In some facilities, booking maytake only 20 minutes, and in others it maytake as long as 8 or 10 hours—decreasingor increasing the possibility of the person’sbeing interviewed. The amount of time canbe determined by the number of arresteesbrought in at once: more arrestees meansmore time is needed. In some facilities,local ADAM staff are permitted to inter-view arrestees before the booking process iscompleted. When this happens, the ADAMarrestees are more likely to represent abroader range of offenses because thosewho may be released on bond or on theirown recognizance (that is, those chargedwith minor offenses) can be interviewed. Insome facilities, the intake areas are chaotic,making it unreasonable to expect ADAMstaff to conduct interviews in them.However, to the extent that ADAM inter-views are conducted after the originalintake, the likelihood increases thatarrestees charged with minor offenses havebeen released and thus not represented inthe sample.

For stock and flow collection, ADAM inter-viewers need access to arrestees who are atvarious stages in the booking process, inorder to sample a broad range of arrestees.In the past, many sites conducted DUFinterviews at times and in locations con-venient to facility staff, typically in thehousing areas. Frequently, the result was asample consisting only of arrestees in thegeneral population of the jail or housingunit of the jail. Although acceptable for

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

stock interviewing, this approach does notwork for flow collection because arresteesmay be released before they reach the hous-ing cells.

Other sites experienced the opposite prob-lem: obtaining interviews from arresteesnot in the intake area. DUF interviews hadbeen conducted in intake, because facilitystaff were particularly concerned for thesecurity of civilians in the housing area, orthey were reluctant to move housedarrestees back to the intake area for inter-viewing. In almost all cases, site staff havenegotiated a reasonable solution, usuallyinvolving selecting an interview locationfor stock and another for flow, or bringingone population into another area to accom-modate the interviewers.

In addition to negotiating with ADAM staffto determine when and where interviewstake place, facility officials also determinethe broad categories of arrestees who maybe interviewed. In almost all cases,arrestees in medical units and psychologi-cal units do not participate in ADAM.Some arrestees are deemed too inebriatedor are verbally abusive or violent and can-not be interviewed. Local site staff workwith facility staff to expand the interviewpopulation as much as possible and toensure that certain populations are notunintentionally eliminated. To eliminateunnecessary bias, sites complete “face-sheets” (forms containing information) onunavailable selected arrestees whosebehavior initially prevents their beinginterviewed, and attempt to interview themlater in the shift.

Adjusting to arrestees’ movementsin the system Arrestees may be released or transferredbefore they can be interviewed. Becausethe stock is selected as many as 16 hoursafter arrestees arrive in the facility, somemay have been released before data collec-tion begins. As a result, sites may find itdifficult to meet their quota of stock inter-views. When this happens, the sites mustfind out whether the arrestees are being

SecuritySecurity plays a major role in the successof sampling at the site level. Interviewersneed a safe and private environment inwhich to conduct the interviews, but theyalso must meet any conditions the facilitysets. In some sites, this may mean there is aglass partition between interviewer andarrestee; in others it may mean there arecell bars. Security may also affect the site’sability to implement the sampling plan ifthere are limitations on the number ofhours permitted for stock interviewing or ifinterview times must accommodate mealschedules or lights out hours. These condi-tions can usually be met, but they requireplanning on the part of site staff.

Security may play a major role in selectingarrestees to be interviewed because manyjails do not allow civilians access to theinformation management systems neededto select stock or flow interviews or tocomplete facesheets (information forms). Inthese situations, it is essential that thefacility officer understands what informa-tion is needed and does not unwittinglybias the sample in an attempt to assist inthe research. This might happen, for exam-ple, if the officer excludes certain arresteeshe or she deems unruly.

Adapting to the jail environmentEven if all processes specific to the facilityare carried out in the sampling plan, thedynamic nature of the jail environmentmay adversely affect a site’s ability to meetits daily target of interviews. For example,delays occasioned by lockdowns, meals,counts of arrestees, and fights amongarrestees may in turn delay interviewing. Inthe case of regularly scheduled events, thedata collection shift can be changed tolimit the effects of the delays. But even reg-ularly scheduled events are not always pre-dictable. For example, the amount of timeneeded to count arrestees may vary signifi-cantly from one night to the next, lasting aslong as two hours in some cases.

A D A M 2 0 0 0 A N N U A L R E P O R T

191

Imp

lem

en

tin

g t

he

Ne

w A

DA

M S

tu

dy

De

sig

n a

t t

he

Lo

ca

l Le

ve

l

released into the community or transferredto another jail; whether certain groups ofarrestees are missed when those releasedare not interviewed; whether arrestees arereleased, either after arraignment orthrough some other process; and whetherthere is a better time or location for con-ducting stock interviews.

Depending on the answers, collection pro-cedures may change so that stock are inter-viewed before a possible transfer or select-ed stock are interviewed at the new loca-tion. If large numbers of arrestees arereleased after arraignment, procedures maybe changed to interview before arraign-ment. One change would be to split thecollection shift into two periods of the day.It is more difficult to address the situationif arrestees are released on a preset bondand therefore not arraigned.

Information about arrestees’ movements isalso important to ensure comparabilityfrom site to site. In one county, for exam-ple, close to half the arrestee population,regardless of arresting agency, receives cita-tions in the field. The UCR data includethese cases, which means they were alsoincluded in the statistics used to developthe ADAM sample targets. Because thismade the site’s sample targets too high,they had to be adjusted to account for theunique nature of the arrest process there,which generated a large proportion ofarrestees who were not available for ADAMinterviewing.

It may or may not be possible to solve theproblem of arrestees released from stock.But if the release problem arises during theflow period, it must be resolved, because itmeans the flow selection process is notbeing implemented as designed. Ultimately,it may not be feasible to interview all thesecases, because the facility strives to processthem as quickly as possible and will notallow interviewing to interfere. In mostinstances, the problem can be solved bynegotiating access to different intake lists orholding areas in the facility.

192

Conducting research in a jail environmentis different from and often more challeng-ing than conducting the same research inprisons. Jail administrators act as the gate-keeper of the local criminal justice system.They are responsible for intake, medicalscreening, classification, release, feeding,and moving to and from court and other-wise transporting people who have onlyrecently been arrested and are often unruly,intoxicated, or violent. To protect the bal-ance of power and order in the jail, localADAM staff will invest whatever effort isnecessary to build relationships with facili-ty staff, understand the rules and restric-tions, and adopt procedures that adhere tothose rules without transgressing theADAM protocols.

The result: more reliable and useful data Standardized data collection protocols andprobability-based sampling will increasethe reliability of ADAM’s data and subse-quent findings. With the adoption of proba-bility-based sampling and the expansion ofthe questionnaire, ADAM will be more use-ful than in the past as a platform on whichresearchers can conduct studies of variousaspects of drug use and related behavior.The expanded questionnaire enables localcommunities to validly estimate the preva-lence of a variety of measures, including theproportions of arrestees who test positivefor drug use by urinalysis and the propor-tions who need treatment. And for the firsttime, the questionnaire enables the ADAMsites to develop prevalence estimates ofdrug use in the nonarrestee population.

A D A M 2 0 0 0 A N N U A L R E P O R TIm

ple

me

nt

ing

th

e N

ew

AD

AM

St

ud

y D

es

ign

at

th

e L

oc

al

Le

ve

l

NOTES1. These national collection programs are the National Household Survey on Drug Abuse (NHSDA), the DEA’s System to Retrieve

Information from Drug Evidence (STRIDE), and the Treatment Episode Data Set (TEDS). Other surveys used by ADAM are those con-ducted by the Center for Substance Abuse Treatment and the U.S. Census Bureau.

2. Variations in the booking process are discussed in the section on implementing facility-level sampling plans.

3. U.S. Census Bureau, 2000 data.

4. Data were collected during the first quarter of 2000 from all five boroughs. Because of unresolved sampling issues and cost constraints,data were collected in the second, third, and fourth quarters only in Manhattan.

5. A more in-depth discussion of the sampling and weighting method is in Methodology Guide for ADAM, by Dana Hunt and WilliamRhodes, Washington, D.C.: U.S. Department of Justice, National Institute of Justice, May 2001. The Guide can be downloaded from theADAM Web page (http://www.adam-nij.net) on the NIJ Web site (http://www.ojp.usdoj.gov/nij).

6. Cost constraints and other practical limitations led ADAM to adopt probability-based sampling for adult male arrestees only.

A D A M 2 0 0 0 A N N U A L R E P O R T

193

VIII. “Calendaring” in ADAM:Examining Annual Patterns ofDrug Use and Related Behavior

“C

al

en

da

ri

ng

”

in

A

DA

M

Alternatively, research designs may focuson drug use over a longer period, presumedto be more typical of use patterns. Theadded advantage is that people more readi-ly admit to drug use in the more distantpast than to use in the past few days orweeks.1 In this approach, arrestees might beasked to describe their annual consumptionof drugs. But when frequency of drug use isvery high or if it changes, recall can becompromised. It would not be difficult fora teetotaler to remember drinking no alco-hol, but someone who has several drinkson certain days and fewer on others wouldfind it difficult to come up with an accu-rate count of overall consumption. Askingpeople about an even longer period of timeintroduces significant chances of error andpromotes guessing or mental averaging.

The calendar approach to collecting datawas developed to deal with recall andrelated issues. It was used in early researchon drug careers2 as well as in other fieldsin which researchers are interested inrecall of events over an extended period.3

Calendaring has been shown to increasethe accuracy of recall over even longerperiods than 12 months (the period used inADAM).4 It promotes recall by dividing theyear into units conceptually manageable bythe arrestee and then “anchoring” his mem-ory around interconnected, real-life eventsthat occur in these units of time.

ADAM’s calendar information consists ofdata covering events as they occur in one-month units of time in a 12-month period.For each unit of time, data on drug use,

by Dana E. Hunt, Sarah Kuck, and Patrick Johnston*

Among the unique features of theADAM survey instrument (question-naire) fielded in 2000 is a technique

called “calendaring.” It is designed to exam-ine drug use and related behavior over theperiod of an entire year and month to monthwithin a year. To measure drug use only inthe recent past (30 days) does not conveythe complexity of the behavior. To measuredrug use over a year addresses the complex-ity issue, but events that took place as longago as a year may be difficult to remember.Calendaring helps solve the problem ofrecall by prompting the arrestee, during theinterview, with questions about events thattook place at about the same time duringeach period of time. The prompts includequestions that can be used to “crosswalk”ADAM data with data in other surveys.

Calendaring promotes recall Research designs often relinquish anyattempt at long-term recall, instead askingpeople about events that took place in asingle, relatively brief period of time (thepast month, for example). Drug use, thougha chronic behavior, changes over time.Thus, the recent past is not likely to be rep-resentative or typical, either of use or treat-ment history. In ADAM, it may be that atthe time the arrestee was interviewed, he orshe was in a period of escalated drug use(one that resulted in the arrest). Moreover,because drug use is a socially stigmatizedbehavior, people may be reluctant torespond to questions about recent use.

* Dana E. Hunt, Ph.D., of Abt Associates Inc., is Principal Scientist with Abt and was the contracting Project Director of the ADAM program; Sarah Kuck is aResearch Assistant with Abt; and Patrick Johnston, M.S., is an Associate with Abt.

194

A D A M 2 0 0 0 A N N U A L R E P O R T“

Ca

le

nd

ar

in

g”

i

n

AD

AM

amount of cocaine consumed or purchasedby arrestees in an area in a year, ADAMdata on total annual consumption ofcocaine among users involved in the crimi-nal justice system could be examined. Ifdata from a single point in time were usedto extrapolate to the entire year, that esti-mate would likely be biased. An arresteemight say he used cocaine 15 days in thepast 30. In the previous months, however,he might have used it only one or two daysin some months and not at all in others.Using past 30-day patterns as typical of ayear would seriously inflate calculations ofannual consumption of cocaine.

Measuring “typical” drug use That the commonly used 30-day recallperiod may not substitute for typical ormodal behavior is evident from self-report-ed drug use by adult male arrestees in theADAM sample. Data from 2000 for select-ed sites—New York, Phoenix, and LasVegas5—show that past-30-day crack usepatterns were not in all cases the same aslonger-term patterns. Crack users amongadult male arrestees in New York were farmore consistent in their long-term use thanwere their Phoenix or Las Vegas counter-parts. Ninety-five percent of crack users inNew York said they used the drug in the30 days before their arrest and a similarlyhigh proportion—82 percent—reportedusing it at least once in each of the previ-ous 12 months (only 9 percent reportedusing crack in fewer than six months of theprevious year). (See Table 8–1.)

In Phoenix and Las Vegas, by contrast, theproportions who used crack throughout theyear were far different from the proportions

SELF-REPORTED CRACK USE, SELECTED SITES, BYCALENDAR PERIOD—ADULT MALE ARRESTEES, 2000*Table 8-1

Primary City

* Questions were asked of adult male arrestees who said they used crack cocaine at any time in the 12 months before their arrest.

New York, NY 94.9% 81.8% 8.9% 1.6%

Phoenix, AZ 80.7 41.0 40.5 16.4

Las Vegas, NV 77.9 43.9 33.0 9.6

Percent Who UsedCrack in 30 DaysBefore Arrest

Percent Who UsedCrack in All 12Months Before Arrest

Percent Who UsedCrack in Fewer than 6Months Before Arrest

Percent Who Used Crackin Only One Month of Year Before Arrest

drug treatment, mental health treatment,number of arrests, and residency status ofthe arrestees are obtained during the inter-view. Each one-month “box” in effect con-tains sets of interrelated life events andbehaviors. These events vary in duration(number of months) and intensity (level ofdrug use, for example) and may be corre-lated with or conditioned on each other.Thus, for example, the likelihood of usingdrugs at Time 1 is related to the likelihoodof using drugs at Time 2. The relationshipof one event to another can also be con-temporaneous (occurs in the same month)or delayed (occurs in subsequent months).

The ADAM interviewer records the behav-ior or events throughout the time period ofinterest (one year, month by month). Theconnections among the behaviors andevents are presented visually (in a calen-dar) and mentally through a series of inter-related questions about life events. The lifeevents themselves serve as cues to recall ofother events. For example, the arrestee isasked where he was living, whether hewas in treatment, whether he was arrested,and the approximate extent of his druguse. These questions are asked for each ofthe 12 months of the past year. As eachtype of behavior is recalled, it becomes afurther anchor for recalling the next set ofevents. The result is a grid ofevents/behaviors, related to each andoccurring over a period of time.

Calendaring increases accuracy With calendaring, information becomesmore accurate because it reflects aidedrecall of patterns unfolding over relativelylong periods. For example, to calculate the

A D A M 2 0 0 0 A N N U A L R E P O R T

195

“C

al

en

da

ri

ng

”

in

A

DA

M

New York

San Diego

Oklahoma City

New Orleans

Sacramento

Atlanta

0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00

■ Heroin ■ Cocaine ■ Marijuana ■ Methamphetamine

Note: Use of drugs was detected by urinalysis. Numbers are mean number of arrests.

who used it in the past month. The 81 per-cent of crack-using arrestees in Phoenix whosaid they used the drug in the past 30 daysdropped to 41 percent for use in all 12 previ-ous months (41 percent had used crack half theyear or less and 16 percent in only one monthof the year). In Las Vegas, where the proportionwho used crack in the recent past 30 days wasalso relatively high (78 percent), the figure foruse in all 12 months of the past year droppedto 44 percent, and to one-third for use in fewerthan six months of the past year.

New York’s 30-day figures could reasonablybe used as the basis of an assessment ofcrack use at that site, because the level ofuse was consistent over time. In the othertwo sites, the picture was different. Whatdifference do these differences make? Ifpast-30-days data (when the proportionswho used crack were high) were used toassess the amount of crack consumed inthese sites, the result might be serious over-estimation. While it might be safe toassume that the most recent month’s use ofcrack in New York was like any othermonth’s use, it would not be so for Phoenixand Las Vegas. At those two sites, crack useappeared more variable or episodic. A usermight, for example, use crack one month,skip a month, return to use, and so forth.

Calculating repeated arrestsCalendaring also increases accuracy in calcu-lating the rate at which people in the ADAMsample are arrested.6 DUF and ADAM havealways reported number of arrests ratherthan number of arrestees.7 When number ofarrestees rather than number of arrests arereported, this conceals repeat offending. Infact, law enforcement has long noted thatrepeat offenders are overrepresented inarrestee statistics. With calendaring, the newADAM survey instrument makes it possibleto examine the number of repeat arrests of asingle arrestee in the sample.

Many variables affect rates of arrest. ADAMmeasures some of them, including age andprevious arrest history. Data from selectedADAM sites also show that the number oftimes someone in the ADAM sample hasbeen arrested varies with type of drugused.8 (See Exhibit 8–1.) In the cities where

Exhibit 8-1: Annual rates of arrest, by selected drugs, byselected sites–adult male arrestees, 20002000

methamphetamine use was detected amongadult male arrestees—San Diego, OklahomaCity, and Sacramento—the rate at whichusers were arrested was higher than therate for users of any other drug. OklahomaCity’s methamphetamine users were arrest-ed almost three times more often than hero-in users; in Sacramento, methamphetamineusers were arrested more than twice asoften as heroin users. In sites where nomethamphetamine use was found amongarrestees (New York, New Orleans,Atlanta), cocaine users were arrested mostoften, though the difference between usersis not as dramatic as for methamphetamine.

ADAM data support the law enforcementobservation that many arrestees come backto the criminal justice system again andagain. Nevertheless, about half the adultmales in the sample in these selected sitesindicated that their current arrest was theironly arrest in the past 12 months. (See Table8–2.) In New York, for example, almost halfthe adult male arrestees said the currentarrest was their only arrest in the past year.By contrast, slightly more than half thearrestees in New York were arrestedbetween one and five times in the past yearand 1 percent said they were arrested 10 ormore times in the past 12 months.

196

A D A M 2 0 0 0 A N N U A L R E P O R T“

Ca

le

nd

ar

in

g”

i

n

AD

AM

at least some part of that time.10 (See Table8–3.) The range among these sites was 4 per-cent (Birmingham) to 32 percent (Honolulu).While the NHSDA is the country’s premiersurvey of drug use in the general popula-tion, it may miss some people who are theheaviest consumers of illegal drugs.

Crosswalking ADAM with TEDS reveals asimilar pattern. The Treatment Episode DataSet contains information about extent of drugand alcohol treatment. ADAM mirrors theTEDS definition of an “episode” of treatment,with questions about treatment received bydrug users month by month in the past year.11

In the sites selected for examination, the pro-portions of adult male arrestees who useddrugs in the year before their arrest and didnot receive any inpatient or outpatient treat-ment were high.12 In New York, 81 percent ofthem did not participate in treatment; in NewOrleans and Birmingham, the figures were 90percent or higher. Thus, the TEDS data donot reflect large proportions of drug users inthe arrestee population.

The State Treatment Needs AssessmentProgram (STNAP) is a CSAT program thatcollects information about drug abuse anduses it to estimate the need for substanceabuse services. ADAM crosswalkedSTNAP data to identify the proportions ofdrug-using arrestees who might be exclud-ed from STNAP. Because the STNAP sur-vey is conducted by phone, anyone whodoes not have a phone cannot be contact-ed. Having a phone is a proxy for inclu-sion in STNAP, so the redesigned ADAMsurvey instrument included a questionabout the number of noncommercial phone

ARREST RATES IN PAST 12 MONTHS, SELECTED SITES—ADULT MALE ARRESTEES, 2000Table 8-2

Primary City

Percent WhoReported CurrentArrest as Sole Arrest

Percent WhoReported 1–5Previous Arrests

Percent WhoReported 6–9Previous Arrests

Percent WhoReported 10 or MorePrevious Arrests

Atlanta, GA 48% 48% 2% 3%

New Orleans, LA 41 55 3 1

New York, NY 47 51 1 1

Oklahoma City, OK 48 51 1 0

Sacramento, CA 44 55 1 0

San Diego, CA 56 42 2 0

How ADAM data can be used withother measures of drug useCalendaring also offers the opportunity touse ADAM data in conjunction with othermeasures of drug use and related behaviorto find out whether arrestees are covered inthese counts. Among these are the NationalHousehold Survey on Drug Abuse(NHSDA), the Treatment Episode Data Set(TEDS), and the State Treatment NeedsAssessment Program (STNAP).9 In develop-ing the new survey instrument, the ADAMprogram made certain that the variableswere defined in a way that would enablethe resultant data to be “crosswalked” with(directly compared to) data in these andother relevant surveys and data sets.

Crosswalking makes it possible, for exam-ple, to identify people not counted byNHSDA but included in ADAM. TheNHSDA examines drug use by all peoplein the general population age 12 andolder who are members of a household. Itexcludes people who are homeless, livingin a temporary shelter, confined in jail,or in like circumstances. In ADAM, thequestions about residence, whoseanswers are recorded on the month-by-month “calendar,” were designed tomatch those in NHSDA.

Comparing NHSDA and ADAM data revealsthat in most of the selected sites examined,large percentages of arrestees who useddrugs in the year before their arrest wouldnot have been included in the HouseholdSurvey. They were excluded because theywere transient or lived in unstable housing

A D A M 2 0 0 0 A N N U A L R E P O R T

197

“C

al

en

da

ri

ng

”

in

A

DA

M

the dynamics of substance abuse. (Forother ways in which calendaring is beingexplored by ADAM, see “Further Potentialof ADAM Calendaring.”)

Patterns of heroin and cocaine use Data on self-reported past-year use ofheroin and cocaine by adult male arresteesin New York reveal different levels orintensities of use. The percentages ofarrestees who used these drugs most heav-ily were higher than the percentages whoused them less frequently. Less than 20percent of both heroin and cocaine userswere involved with these drugs 1 to 7days in each month of the past year, andsimilarly small proportions were involved8 to 14 days per month. By contrast, fully55 percent of heroin users and more than40 percent of cocaine users were involved15 to 30 days in each month of the year.(See Exhibit 8–2.)

Information from three heroin users amongthe adult male arrestees in New Yorkreveals one of many patterns, in this caseescalating use in the months before arrest.(See Exhibit 8–3.) “User A” began the yearusing heroin fairly heavily, became absti-nent, and escalated to a higher level of usein the months before his arrest. “User B”also escalated to a higher level before he

lines where the arrestee was living. By thismeasure, the proportions of arrestees13 whowere not counted in the assessment oftreatment needs were high in the selectedsites examined.

The way in which these other national sur-veys of drug use are designed tends toexclude certain subpopulations. Analysisof the 2000 ADAM data revealed the extentto which this was so. ADAM provides the“missing” data on some of these drug users.Because DUF and ADAM have over theyears consistently shown drug use amongthis at-risk population to be high, theADAM findings are an important basis forState and Federal assessment of need andresource allocation.

Using calendaring to examine patterns of drug useMuch of the research literature on sub-stance abuse describes it as a chronic andrelapsing condition that involves cycles ofmoderate use, abuse, and abstinence. Bypermitting analysis of drug use and relatedbehavior over time, calendaring affordsinsights into patterns of use, whether con-sistent or episodic. Setting these patternsagainst the backdrop of concurrent eventsin the lives of the arrestees can illuminate

ADULT MALE ARRESTEES NOT INCLUDED IN OTHER MEASURESOF DRUG USE AND RELATED BEHAVIOR, SELECTED SITES, 2000Table 8-3

Primary City

Percent of Drug Users inUnstable Residence inPast Year—Not Reflectedin NHSDAa

Percent of Drug UsersNot in Treatment inPast Year—NotReflected in TEDSb

Percent of ArresteesHaving No Phone in PastMonth—Not Reflected in STNAPc

Albuquerque, NM 10.1% 82.3% 25.0%

Birmingham, AL 4.3 90.0 13.2

Honolulu, HI 32.3 80.3 34.3

New Orleans, LA 5.7 94.2 18.3

New York, NY 11.7 80.5 33.5

Phoenix, AZ 14.5 85.6 27.1

Portland, OR 23.5 77.2 26.1

San Antonio, TX 10.1 87.8 20.4

San Diego, CA 21.6 82.5 24.2

a. NHSDA is the National Household Survey on Drug Abuse.

b. TEDS is the Treatment Episode Data Set.

c. STNAP is the State Treatment Needs Assessment Program.

Note all adult male arrestees, not just those who used drugs, were included in this analysis.

198

A D A M 2 0 0 0 A N N U A L R E P O R T“

Ca

le

nd

ar

in

g”

i

n

AD

AM

1-7 Days Per Month

8-14 Days Per Month

15-30 Days Per Month

0 10 20 30 40 50 60%

■ Heroin ■ Cocaine

Note: Questions were asked of adult male arrestees who said they usedheroin or cocaine in the year before they were arrested. Because recallis particularly difficult when an event occurs frequently, they wereasked to state the number of days of involvement with the drug (thatis, any amount of use on a given day) rather than the number of timesthe drug was used.

Percent of ArresteesUsing Drug

Exhibit 8-2: Percentages of adult male arrestees who usedheroin or cocaine in past year, by level of use, New York–adult male arrestees, 2000

Further Potential of ADAM Calendaring The ADAM program is exploring the potential for using calendaring to build predictive models. Howwould calendaring be different from simple regression analysis in studying the effect of one variable—such as treatment—on another? The difference lies in the nature of the data available to use in develop-ing the model. A simple regression model might look at the impact of drug treatment on subsequentdrug abuse, examining data on level of use before and after treatment and amount of treatment received.In “real life,” however, the effects of treatment are conditioned on many variables, including the patternsof use the person brings to treatment, not simply the level of use at entry. Calendaring in ADAM makesavailable more dynamic information about these patterns.

Calendaring in ADAM does have limitations that need to be taken into account in developing a predictivemodel. First, the ADAM data cover only a 12-month period, which is a relatively small window into thenumber of fluctuations or events in the career of the drug user. Second, the 12-month period is not likelyto be “typical” in that in ADAM it always terminates with an arrest. In addition, the data are “left-cen-sored”; that is, projections or predictions would be made about behavior with no information about thembefore the time frame studied.

Clinicians often say that users have to “hit bottom” in drug use and other life crises before their treat-ment experience will be successful. One way to use a predictive model based on calendaring would be totry to measure “hit bottom.” It may mean a series of experiences before treatment—escalating arrests,increasing drug use, increasing transience. These variables (number of arrests, level of drug use, resi-dence status) are all measured in the ADAM calendar in each month before the event of interest (in thisexample, entry into treatment).

The more traditional approach is to summarize these “events” without regard to either when they hap-pened or to their interrelationship. By using the calendar data, it is possible to build models that canaccount for previous experiences and concurrent events and activities. For a given arrestee, the level ofheroin use in June might be directly related to whether he was in treatment or in jail in March, April, orMay, as well as conditioned on the amount of heroin he was consuming in those months. Level of usemight also be correlated with arrests in those months. The effects of some interrelated events (beingjailed, for example) might be immediate in reducing the amount of heroin consumed, while the effects ofothers (treatment, for example) might lag or be delayed by a month or two.

was arrested, although his use over the 12-month period was more consistent—there wasno period of abstinence. “User C” was a heavyuser who nonetheless managed to abstain forthree months but returned to heavy use in thetwo months before being arrested.

Calendaring can document events in thelives of the arrestees that might affect theirdrug use. These include other arrests, jailconfinements, and treatment experience.For “User B” (from Exhibit 8–3), time injail affected the level of heroin use. Theperiod in which this user reported reduceddrug use corresponds to his arrest and jailtime in month 7 of the 12-month period.(See Exhibit 8–4.) It is worth noting thatthis user said he was being treated for druguse (on an outpatient basis) during theentire 12-month period.

Level of Heroin Use ArrestOutpatient Treatment Jail Stay

“C

al

en

da

ri

ng

”

in

A

DA

MA D A M 2 0 0 0 A N N U A L R E P O R T

199

“C

al

en

da

ri

ng

”

in

A

DA

M

2.5

2.0

3.0

3.5

1.5

1.0

0.5

0Month 1 Month 2 Month 3 Month 4 Month 5 Month 6 Month 7 Month 8 Month 9 Month 10 Month 11 Month 12

Exhibit 8-4: Heroin use by “User B” in context of treatment and involvement in criminal justice system,New York–ADAM data, 2000

Note: On the y axis, “0” = no drug use; “1” = 1 day/week or 1–7 days/month; “2” = 2–3days week or 8–12 days/month; “3” = more than 3 days/week or 13–30 days/month.

LEVE

L OF

DRU

G US

E

Exhibit 8-3: Patterns of heroin use for 3 adult male arrestees in year before arrest, New York–2000

Note: On the y axis, “0” = no drug use; “1” = 1 day/week or 1–7 days/month; “2” = 2–3days week or 8–12 days/month; “3” = more than 3 days/week or 13–30 days/month. User A User B User C

2.5

2.0

3.0

3.5

1.5

1.0

0.5

0Month 1 Month 2 Month 3 Month 4 Month 5 Month 6 Month 7 Month 8 Month 9 Month 10 Month 11 Month 12

LEVE

L OF

DRU

G US

E

200

A D A M 2 0 0 0 A N N U A L R E P O R T“

Ca

le

nd

ar

in

g”

i

n

AD

AM

NOTES1. Anglin M., Y. Hser, and C. Chou, “Reliability and Validity of Retrospective Behavioral Self-Report by Narcotics Addicts,” Evaluation

Review 17 (1993): 91–108.

2. McGlothlin, W., M. Anglin, and B. Wilson, “Narcotic Addiction and Crime,” Criminology 16 (1978): 292–315; and Nurco, D., I. Cisin,and M. Balter, “Addict Careers: A New Typology,” The International Journal of the Addictions 16 (1981):1305–1325.

3. Freedman, D.A., et al., “The Life History Calendar: A Technique for Collecting Retrospective Data,” Sociological Methodology 18 (1988):37–68; and Furstenberg, F., et al., Adolescent Mothers in Later Life, New York: Cambridge University Press, 1987.

4. Anglin et al., “Reliability and Validity of Retrospective Behavioral Self-Report”; Freedman et al., “Life History Calendar”; and Collins, L.,et al., “Agreement Between Retrospective Accounts of Substance Abuse and Earlier Reported Substance Abuse,” Applied PsychologicalMeasurement 9 (1985): 301–309.

5. A subset of ADAM sites was selected for ease and simplicity of presentation. No other selection criteria were used.

6. See Chapter 9 for an example of how essential rates of arrest are in estimating hardcore drug use.

7. In DUF and subsequently ADAM, number of arrestees is used as a proxy for number of arrests.

8. In the ADAM program, a user of one drug may also be using another; that is, these categories are not mutually exclusive. Research hasshown that drug users may use a particular substance in preference to others, but they may also use other drugs.

9. The National Household Survey on Drug Abuse is conducted by the Substance Abuse and Mental Health Services Administration(SAMHSA), U.S. Department of Health and Human Services; the State Treatment Needs Assessment Program is administered by theCenter for Substance Abuse Treatment (CSAT) of SAMHSA; SAMHSA also maintains the Treatment Episode Data Set (TEDS).

10. For NHSDA purposes, to be considered a member of a household a person need not own or rent the residence in which he or she is liv-ing, and the residence can be one of a variety of types (for example, a trailer, apartment, or house). However, to be included in NHSDA, aperson cannot be a transient member of a household. If, for example, the survey respondent is living briefly (less than three months)with a girlfriend or a relative, he or she would not be considered a member of that household.

11. TEDS data come from reports of drug and alcohol treatment as measured by intake in treatment programs. An “episode” of treatment ismeasured as entry in an outpatient program and/or an overnight stay in an inpatient program. In the ADAM questionnaire, arrestees whosay they used drugs are asked, for example, how many nights they stayed in inpatient treatment and how many times they entered out-patient treatment.

12. These drug users include people who may not need treatment.

13. The question was asked of all arrestees, not just those who said they used drugs in the year before their arrest.

A D A M 2 0 0 0 A N N U A L R E P O R T

201

IX. Estimating Hardcore DrugUse in the Community

* William Rhodes, Ph.D., is a Principal Scientist and Fellow, and Ryan Kling is a Senior Analyst, at Abt Associates Inc.

by William Rhodes and Ryan Kling*

Scientists and nonscientists alike seekto estimate prevalence—from thenumber of stars in the sky to the

number of angels on the head of a pin.When policy analysts estimate prevalence,they focus on more prosaic topics. Theyseek to find out how many times a condi-tion is occurring or an event is takingplace because their clients—policymak-ers—need the information as the basis ofreasoned decisions.

Researchers have used data from ADAMand DUF (the Drug Use Forecasting pro-gram, ADAM’s predecessor) to understandthe prevalence of drug use and relatedbehaviors among arrestees. The redesignedADAM program now makes it possible toprovide additional prevalence estimates,including the number of hardcore drug usersin a county that has an ADAM program.

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

How prevalence is estimated1

“Hardcore” can be defined in any appropri-ate way. For example, a hardcore usermight be someone who uses illicit drugsmore often than some threshold number;alternatively, a hardcore user might besomeone who is seen as needing treatment.To explain the estimation technique usedhere, a large rectangle represents the num-ber of hardcore users in any ADAM coun-ty—the object of the estimation exercise.(See Exhibit 9–1.) Household surveys offerone way to estimate this number. However,these surveys would exclude a large num-ber of hardcore drug users, either becausethey do not live in a household (as definedby the survey), because they are typicallynot at home when interviewers call, orbecause they lie about their drug use.

Instead, inferences must be drawn about thelarge rectangle from information providedby the small square, which represents

It is possible to count hardcore drug userswho are arrested. Is it possible to estimatethe prevalence of those in the population

based on those who are arrested?

Exhibit 9-1: Hardcore drug users are assumed to all have the same arrest rate

Countypopulation

of hardcore

drug users

# ofHardcore

users in jail

202

A D A M 2 0 0 0 A N N U A L R E P O R TE

st

im

at

in

g

Ha

rd

co

re

D

ru

g

Us

e

Creating a model of the arrestprocessFor the sake of simplicity, the explanationdisregards the two complications of sam-pling and denial of drug use, and assumesthe availability of information about allhardcore drug users arrested and bookedin a given year (the square).2 In practice, ofcourse, neither sampling nor denial can beignored during estimation.

Estimating the composition of the rectan-gle from the composition of the squarerequires a mathematical model of the arrestprocess through which hardcore drug usersmove from the rectangle to the square. (SeeExhibit 9–2.)

Conceptually, modeling treats the police asif they were samplers, conducting a survey.The goal of the modeling is to determinehow hardcore users in the booking popula-tion (the square) should be weighted torepresent hardcore users in the countypopulation (the rectangle). Data about pre-vious arrests of hardcore drug users (in thesquare) provide a basis for estimating therate at which hardcore drug users getarrested and booked. The inverse of thatestimated arrest rate provides the means toweight the square to estimate the rectangle.Exhibit 9-3 is a simple illustration.

Exhibit 9-2: The estimate requires creating a model of the arrest process

Arrest Process

Countypopulation

of hardcore

drug users

THE BASIC APPROACH:Model the rate at which hardcore users arearrested. Use that model to infer the size of

the population.

# ofHardcore

users in jail

hardcore users who are arrested andbooked. An initial assumption is thathardcore drug users all have the sameprobability of being arrested and booked.(They are said to be “homogenous.”) Thesquare is smaller than the rectangle toillustrate the fact that not all hardcoreusers are arrested and booked during aspecific time period (one year, for exam-ple). The composition of the rectangle isinferred from information about the com-position of the square.

The ADAM data do not enumerate hardcoreusers booked into jail, because they are asample. The sample is depicted as a triangle.Because the sample is probability based, thetriangle can be weighted to estimate thesquare. An additional problem is that somehardcore users will deny their drug use orthe level of their drug use, so admitted hard-core users in the sample underrepresent theactual number in the sample. The latter isrepresented with a cross.

The problem stemming from underreportingcould be overcome by estimating how fre-quently hardcore drug users are truthfulabout using drugs at the hardcore level. If the“truthfulness rate” could be calculated, thetriangle could then be estimated on the basisof the cross and, if the sampling weightswere known, it would then be possible toestimate the square from the triangle.

A D A M 2 0 0 0 A N N U A L R E P O R T

203

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

The illustration assumes that the rectanglecomprises 1,000 hardcore drug users(called “H”). By assumption they are iden-tical with respect to the probability andrate of arrest; that is, each is arrested 0.3times per year on average (the arrest rate is“R”). Thus, the 1,000 hardcore drug usersgenerate about HxR=300 arrests per year(the number of arrests is “A”).

Continuing this illustration, the ADAM datawould indicate that the square comprises300 arrests of people who self-report hard-core drug use, and for them the interviewswould reveal that the average arrest rate is0.3 per year. Because H x R = A, then A/R =H. Thus, 300/0.3 = 1,000—the number ofhardcore drug users in the county.

This algebraic calculation illustrates thefundamentals of the estimation. Identifyingand counting the 300 hardcore drug usersrequires a tabulation of the ADAM data(though keeping in mind the two complica-tions, noted above, introduced by samplingand underreporting). The estimate of 0.3arrests per year is the result of analysis ofdata about arrest histories obtained duringthe ADAM interviews.

Assumptions about hardcore drug users inthe community■ They are homogenous with respect to

probability of and rate of arrest.■ Their average annual arrest rate = 0.3 (R).■ They generate about 300 arrests per year (A).

H x R = A.

Data from the ADAM calendar*■ ADAM counts 300 arrests per year.■ On the basis of ADAM interviews, the

number of arrests per year per arrestee isestimated at 0.3.

The estimate of the number of hardcore drug users in the community■ H = A/R = 300/0.3 = 1,000.

* See Chapter 8 for an explanation of “calendaring” in ADAM.

Exhibit 9-3: The basic logic of the estimationmodel illustrated

For the calculation to be correct, not allhardcore drug users need to be arrested. Infact, a hardcore drug user could eludearrest through the entire length of his orher drug use career and still be representedby the “sample” of people booked into jail.The estimation procedure makes noassumption that the booking populationenumerates all hardcore drug users.

Introducing measured heterogeneity.Abandoning the unrealistic assumptionthat hardcore drug users are homogenousin rate of arrest, the estimation methodolo-gy remains conceptually the same. Thistype of heterogeneity is called “measuredheterogeneity” because a number of vari-ables could explain the differences in arrestrates of hardcore users. Operationally themethodology is more complex, however.

To illustrate the consequences of measuredheterogeneity, the rectangle is divided intotwo parts (Exhibit 9-4), with the top halfrepresenting hardcore users of cocaine andthe bottom half representing hardcore usersof heroin. (There could, of course, be manytypes of hardcore drug users, but for illus-tration purposes, only two types areassumed.) The assumption is that each sub-group is homogenous; that is, all cocaineusers are alike and all heroin users arealike. But hardcore heroin users are differ-ent from hardcore cocaine users; the formerare arrested on average 0.4 times per year,while the latter are arrested on average 0.2times per year. In this respect, arrest ratesare heterogeneous but that heterogeneitycan be explained by an observed factor—type of drug use.

Given these assumptions, the booking pop-ulation (the square) is not representative ofthe county population (the rectangle). Inthe booking population, there are two hard-core cocaine users for every hardcore hero-in user. By contrast, in the county, there isone hardcore cocaine user for every hard-core heroin user. Thus, the equation usedpreviously, H = A/R, will not produce anaccurate estimate because A is the totalnumber of hardcore drug users (A = 300in this illustration), while R is the averagearrest rate in the booking population (R = (2/3)0.4 + (1/3)0.2 = 0.333 in this

204

A D A M 2 0 0 0 A N N U A L R E P O R TE

st

im

at

in

g

Ha

rd

co

re

D

ru

g

Us

e

are employed). When the number ofcovariates expands beyond a single vari-able, however, dividing the booking popu-lation into mutually exclusive homogenousgroups becomes impractical or even infea-sible, and estimation then requires the useof regression models. Nevertheless,increased complexity of the model doesnot alter the basic logic of the approach.

Introducing unmeasured heterogeneity.Not all factors that explain the arrestprocess are measurable, or at least not allfactors are identified in ADAM. These toomust be taken into account in calculatingthe number of hardcore drug users. Toillustrate unmeasured heterogeneity, therectangle is again redesigned. The upperpart still represents hardcore cocaine usersand the lower part hardcore heroin users(as in Exhibit 9–4). These two groups arefurther divided, with the divisions depict-ed as triangles. Hardcore cocaine users inthe upper triangle represent those whohave a serious mental illness. This duallydiagnosed group has an average annualarrest rate of 0.5. Hardcore cocaine users inthe lower triangle represent hardcorecocaine users who do not have a seriousmental illness. Their arrest rate is 0.3.Hardcore heroin users are similarly classi-fied. (See Exhibit 9–6.)

illustration). The previous calculation wouldproduce an estimate of H = 300/0.333 = 900hardcore drug users in the county popula-tion. Obviously, this is not correct.

Accommodating measured heterogeneity.The correct approach accommodates meas-ured heterogeneity in the entire populationof hardcore drug users. (See Exhibit 9–5.)The type of drug (cocaine or heroin) is thevariable differentiating the two groups ofhardcore users. ADAM identifies hardcorecocaine and heroin use; hence, heterogene-ity is explained by measured variables.

In this example, the booking populationconsists of 200 hardcore cocaine users,whose average arrest rate is 0.4 per year.This means there are about 200/0.4 = 500in the county. The booking populationalso includes 100 hardcore heroin users,whose arrest rate is 0.2. There are 100/0.2= 500 of them in the county. Thus, thereare 500 + 500 = 1,000 hardcore drug usersin the county.

Other than type of drug (in this example,heroin and cocaine), there are many meas-urable variables that should be taken intoaccount when modeling the arrest process.The ADAM interview instrument wasdesigned to obtain information about othermajor covariates (such as whether arrestees

THE PRINCIPAL DIFFICULTY:

A heterogeneous hardcoreuser population generates

different arrest rates.

Exhibit 9-4: Introducing measured heterogeneity into the estimation

Hardcore users of heroin

Arrest Rate:0.2/yr

Arrest Rate:0.4/yr

Arrest Rate:0.2/yr

Hardcore users of cocaine

Arrest Rate:0.4/yr

A D A M 2 0 0 0 A N N U A L R E P O R T

205

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

Exhibit 9-6: Introducing unmeasured heterogeneity into the estimation

Assumptions■ 300 arrests ■ 200 hardcore cocaine users—arrest

rate 0.4 per year■ 100 hardcore heroin users—arrest

rate 0.2 per year

The new calculation■ 200/0.4 = 500 hardcore cocaine users■ 100/0.2 = 500 hardcore heroin users■ Total = 500 + 500 = 1,000 hardcore

drug users in county

Exhibit 9-5: Changing the calculation to accommodate measured heterogeneity

Hardcore usersof cocaine—mentally ill

Arrest Rate:0.5/yr

Hardcore users of cocaine—not

mentally ill

Arrest Rate:0.3/yr

Hardcore usersof heroin–mentally ill

Arrest Rate:0.3/yr

Hardcore users of heroin—not

mentally ill

Arrest Rate:0.1/yr

The booking population (the square) stilloverrepresents hardcore cocaine users, butnow it also overrepresents hardcore userswho are mentally ill. ADAM does not includedata about mental illness, so the modelingcannot account for this condition in the sameway that it accounts for the cocaine-heroindistinction. Although the approach is concep-tually the same, taking unmeasured hetero-geneity into account requires mathematicalmodeling that defies a simple illustration. Atechnical report provides details.3

Preliminary estimates of numberof hardcore drug users Because the mathematical and statisticallogic of the estimation technique is com-plex, the estimates of number of hardcoreusers are best presented visually, in theform of graphics (Exhibits 9–7 and 9–8).(The algebraic terms used are presented in“Key to Algebraic Terms.”)

Arrest rates and user ratesAs noted, “hardcore drug user” could bedefined differently depending on policyobjectives. The methodology is invariantwith respect to definition. For illustration,hardcore use is defined here as the use ofcocaine, heroin, or methamphetamine onten or more days during the month beforethe ADAM interview.

Exhibit 9-7 shows estimates of the arrestrate for the average hardcore drug user ineach ADAM site. The estimate varies fromsite to site. Confidence intervals are appre-ciable, but most of them overlap a value of0.75 arrests per year. On the basis of thesedata, it might be said that, in most sites,there are about 750 arrests and bookings peryear for every 1,000 hardcore drug users.

For some sites, the confidence intervals areextremely wide. This is largely because sam-ple sizes are small, since several sites did nothave data for all four calendar quarters. Thepresumption is that the confidence intervalswill narrow as the ADAM samples grow.

AN ADDITIONAL DIFFICULTY:Heterogeneity in the

hardcore user populationgenerates different arrestrates. Some heterogeneityis measured by ADAM and

the rest is not.

0.3/yr0.5/yr

0.1/yr0.3/yr

Arrest Rate:

Arrest Rate:

206

A D A M 2 0 0 0 A N N U A L R E P O R TE

st

im

at

in

g

Ha

rd

co

re

D

ru

g

Us

e

hardcore drug users in the county population. Inother words, letting H be the represent the esti-mate of the number of hardcore drug users in thecounty, then: H =∑ Wi/[ÂiP]

The sum above is over the hardcore drug usersin the sample.

The average arrest rate for hardcore drug usersin the community is depicted as  without a sub-script:  = ∑ [Wi/P]/ H

The above equation is the observed number ofarrests divided by the estimated number of hard-core drug users in the community. (The arrestrates of hardcore drug users in all ADAM sitesare presented in Exhibit 9–7.)

Finally, let G represent the average number ofhardcore drug users in the county per hardcoredrug users in the arrestee population (see Exhibit9–8): G = ∑ Wi(1/Âi)/Wi.

2.50

2

1.50

1

.50

0

.25

1 4 7 10 13 16 19 22 25 28

Exhibit 9-7: Average annual arrest rate of hardcore drug users in a county

Note: The vertical axis shows a point estimate (a single number) and 95 percent confidence interval for the number of arrests andbookings per year. The confidence interval is truncated at 2.5 to preserve scale in this chart.

HARD

CORE

DRU

G US

ERS–

ANNU

AL A

RRES

T RA

TE

ADAM Sites1 = Detroit2 = Seattle3 = Spokane4 = Charlotte5 = Minneapolis6 = Denver7 = Des Moines8 = New York

9 = Portland10 = Birmingham11 = Phoenix12 = Salt Lake City13 = San Antonio14 = Omaha15 = Las Vegas16 = Indianapolis17 = Cleveland

18 = Miami19 = Honolulu20 = Albuquerque21 = New Orleans22 = Sacramento23 = Oklahoma City24 = San Jose25 = Houston26 = Dallas

27 = Albany28 = Tucson29 = Philadelphia30 = San Diego31 = Anchorage32 = Chicago33 = Atlanta34 = Fort Lauderdale35 = Laredo

31

Key to Algebraic TermsUsed in the EstimationLet Âi represent the estimated rate at which hard-core users in the county, who are represented bythe ith hardcore user in the sample of bookedindividuals, are arrested. (See model discussed inthe text.)

Let Wi represent the ADAM sampling weightassociated with the ith hardcore drug user inthe sample. This is derived from the ADAMsampling design.

Let P represent the proportion of hardcore drugusers who answer truthfully about whether theyare hardcore drug users. The assumption is theyeither tell the truth or deny drug use altogether.

Then, the ith hardcore drug user in the ADAMsample for a specified site represents Wi/[Â P]

34

.75

1.75

2.25

1.25

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

A D A M 2 0 0 0 A N N U A L R E P O R T

207

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

Note: The vertical axis shows a point estimate (a single number) and confidence interval for the number of hardcore drug users in thecounty per hardcore drug user in the booking population.4

Exhibit 9-8 shows estimates of the numberof hardcore drug users in the county perhardcore drug user in the booking popula-tion. The average of the 35 ADAM sitesappears to be about 1,600 hardcore drugusers in the county for every 1,000 hardcoredrug users who appear among arrestees.

Number of hardcore users in thecommunity Table 9-1 shows what might be consideredthe ultimate estimate—the number of hard-core drug users in the county. Caution isadvised in interpreting these estimates.One reason is the use of “stratified clustersampling” in some sites. That is, whencounties have many jails, ADAM inter-viewers cannot go to each of them, so jailsthemselves must be sampled. (These sitesare marked on the table as “S/C,” for strati-fied cluster sampling.) At the time of thisstudy, ADAM had not yet made provisionsfor adjusting its estimates to account for

nonsampled jails. Thus, in sites that usestratified cluster sampling, the estimatesare too low. In one site, New York, thenumber of hardcore drug users is under-represented because only in Manhattanwere data collected in all four calendarquarters; in only one quarter were all fiveboroughs in the study.

To adjust for the limitation, in the sites thatuse stratified cluster sampling (the S/Csites) the estimates should be doubled. Thatis only a crude adjustment. In Cleveland,Des Moines, and Minneapolis, doublingmight produce too high a number. By con-trast, the Detroit figures probably should bemultiplied by a factor of five. For New York,the estimate might be roughly two and one-half times as great as indicated here.

Another reason for caution in interpretingthe estimates is that they are of adult malesonly. Because women constitute about 20percent of the arrestee population,5 the

Exhibit 9-8: Hardcore users in the community per hardcore user in the booking population

ADAM Sites1 = Charlotte2 = Seattle3 = Spokane4 = Des Moines5 = Minneapolis6 = Denver7 = Portland8 = New York

9 = Phoenix10 = Birmingham11 = Salt Lake City12 = San Antonio13 = Miami14 = Indianapolis15 = Omaha16 = Cleveland17 = Detroit

18 = Las Vegas19 = Honolulu20 = New Orleans21 = Albuquerque22 = Houston23 = Oklahoma City24 = San Jose25 = Sacramento26 = Dallas

27 = Tucson28 = Albany29 = San Diego30 = Anchorage31 = Laredo32 = Atlanta33 = Chicago34 = Philadelphia35 = Fort Lauderdale

7

6

2

0

1

1 4 7 10 13 16 19 22 25 28 31 34

3

5

4

# OF

HAR

DCOR

E DR

UG U

SERS

IN C

OMM

UNIT

Y PE

R HA

RDCO

REUS

ERS

IN T

HE B

OOKI

NG P

OPUL

ATIO

N (M

ULTI

PLIE

R)

208

A D A M 2 0 0 0 A N N U A L R E P O R TE

st

im

at

in

g

Ha

rd

co

re

D

ru

g

Us

e

own calculations. Researchers may want access tothe code to improve it. Most sites are likely not tochoose to use the code to replicate or extend theanalysis of hardcore drug use. Moreover, it is unnec-essary because ADAM will do it for them.

As ADAM data accumulate over the years, the modelshould be reassessed and improvements made fol-lowing the most recent assessment. Once theimprovements reach a certain point, the only remain-ing challenge would be to develop weights suitable forestimating the number of hardcore users. After theyare developed, the ADAM sites could estimate thenumber of hardcore drug users, along with confidenceintervals, using software designed for statistical calcu-lations. The degree of difficulty would be no greaterthan what is currently required for the sites to producethe more routine weighted estimates of drug use.

Primary City

Table 9-1

Estimate Lower 95% Upper 95%

* A hardcore drug user is someone who used cocaine, heroin, or methamphetamine in at least 11 of the past 30 days before being interviewed by ADAM.

Note: S/C indicates counties where stratified cluster sampling is used. Numbers are underestimations of level and standard error for prevalence estimates.INC indicates incomplete data. The numbers underrepresent hardcore drug use because the New York sample is limited to Manhattan.

What Do the EstimationTechniques Mean for theADAM Sites?Most nonstatisticians, and perhaps even many stat-isticians, would consider the model/technique forestimating hardcore drug use sophisticated. Moststatisticians would likely agree that applying it isbeyond the ability of anyone not trained in statistics.That raises the question of how the approach can behelpful to the ADAM sites.

Although the estimation procedure requires somefamiliarity with statistics, computing code containinga program to calculate these estimates will be avail-able from NIJ for anyone who requests it. SomeADAM sites may want to use the code to make their

ESTIMATED NUMBER OF HARDCORE DRUG USERS* IN THE ADAMSITES—ADULT MALE ARRESTEES, 2000

Albany/Capital Area, NY 6,879 2,077 11,680Albuquerque, NM 8,605 6,825 10,386Anchorage, AK 1,629 824 2,434Atlanta, GA 34,836 16,346 53,326Birmingham, AL 5,129 4,142 6,117Charlotte-Metro, NC 4,422 -1,353 10,198Chicago, IL (S/C) 27,469 5,779 49,160Cleveland, OH (S/C) 11,561 8,082 15,041Dallas, TX (S/C) 31,662 26,364 36,959Denver, CO 2,122 1,446 2,798Des Moines, IA (S/C) 2,013 1,518 2,508Detroit, MI (S/C) 6,048 4,599 7,496Fort Lauderdale, FL 17,394 2,272 32,515Honolulu, HI 5,145 3,193 7,096Houston, TX 12,402 7,693 17,110Indianapolis, IN 8,001 4,915 11,087Laredo, TX 7,226 4,647 9,806Las Vegas, NV 17,223 13,714 20,732Miami, FL 13,441 8,510 18,373Minneapolis, MN (S/C) 1,538 752 2,324New Orleans, LA 12,674 7,792 17,556New York, NY (INC) 125,844 117,465 134,222Oklahoma City, OK 3,656 2,444 4,868Omaha, NE 6,436 5,146 7,727Philadelphia, PA (S/C) 27,847 24,478 31,216Phoenix, AZ (S/C) 30,200 24,181 36,219Portland, OR 4,842 3,450 6,234Sacramento, CA 24,991 20,540 29,443Salt Lake City, UT 2,668 1,880 3,456San Antonio, TX 12,098 8,776 15,421San Diego, CA 42,140 33,948 50,332San Jose, CA 13,693 7,272 20,114Seattle, WA 6,934 5,849 8,018Spokane, WA 2,147 1,484 2,811Tucson, AZ 4,961 3,268 6,653

A D A M 2 0 0 0 A N N U A L R E P O R T

209

Es

ti

ma

ti

ng

H

ar

dc

or

e

Dr

ug

U

se

estimates should be inflated by about 1.25.However, precise adjustments wouldrequire taking into account the number ofbookings of hardcore female drug usersand the rate at which they are arrested andbooked. When the ADAM redesign extendsto women arrestees, the adjustment can bemore informed.

The estimates also exclude juveniledetainees, but this is a minor problembecause there are good estimates of juvenilehardcore drug users in the annualMonitoring the Future (MTF) survey.Estimates based on the highly selectivesample of hardcore drug users among juve-nile detainees would probably not be muchimprovement over estimates based on MTF.

One other limitation is that the methodolo-gy used to adjust for underreporting hard-core drug use is provisional. (Recall thestep of drawing an inference about the tri-angle using information provided by thecross.) The ADAM project has not yetdeveloped adjustments for underreportingdrug use extending beyond the two orthree days before the interview. Suchadjustments would be welcome adjuncts tothe hardcore user estimation methodology.

Overcoming limitations in the estimation methodEven beyond the limitations, the estimatesshould be considered provisional. ADAMis in its infancy. Improvements will lead toadvances in estimation based on theADAM data. There is no reason to assumethat the method is immutable. Researcherswill undoubtedly find ways to improve it.But even with the current limitations, themethod of estimating the number of hard-core drug users at a given ADAM site cangenerate credible figures. And there is noreason that better estimation methodscould not be applied retrospectively.

What about people whose risk ofarrest is low?The hardcore user estimation methodolo-gy does not require that all hardcore drugusers be arrested. In fact, as noted above,

many hardcore users may elude arrestthroughout their entire drug-use careersand yet still be represented by the esti-mates. If the police are viewed as sam-plers, they no more need to arrest every-body than a sampler conducting a conven-tional survey needs to interview every-body for the resulting sample to representthe population.

There may be subsets of the population,however, whose risk of arrest is so smallthat they would not, practically speaking,be represented in the police sample.Celebrities might be an example of onesuch group. They would either avoid arrestentirely or else the probability of theirbeing arrested would be so small that theresulting prevalence estimate would be tooimprecise to be useful. How is this poten-tial “residual” to be handled?

Arguably, a subset so immune to arrest issmall or otherwise of marginal interest topolicymakers. It is not a group that wouldmake heavy demands on the criminal jus-tice system, the publicly financed treat-ment system, or the public health system.Perhaps from the standpoint of public poli-cy, it is sufficient to estimate the preva-lence of hardcore drug users who run anappreciable risk of arrest.

Avoiding undercountingIf the above argument is not convincing,then undercounting could be corrected byextending the hardcore user methodology.The extension is best explained by way ofexample. In the example, the current cal-culation method produces an estimate of80,000 hardcore drug users in a county ina given year. ADAM data, in this example,indicate that those 80,000 hardcore drugusers generate about 20,000 drug treatmentadmissions per year. A final assumption inthe example is that data from local treat-ment programs show that hardcore usersactually account for a higher number oftreatment admissions—25,000 per year.This means the estimates of hardcore druguse based on ADAM data understate hard-core drug use by 25,000/20,000 = 1.25.The ADAM-based estimates would be

210

A D A M 2 0 0 0 A N N U A L R E P O R TE

st

im

at

in

g

Ha

rd

co

re

D

ru

g

Us

e

recalculated, with the result 1.25 x 80,000= 100,000 hardcore drug users in a county.The “missing” hardcore drug users are“found” by looking at a second set of data.This would seem to be a way to avoid agross undercount of hardcore drug use.

Other applicationsThe estimation method has other applica-tions. Variations could be used, for exam-ple, to calculate the proportion of the gen-eral population with specific infectiousdiseases who come into contact with thecriminal justice system, and to estimatethe price of illicit drugs and the amount ofmoney users spend purchasing drugs. Theapproach would seem to be useful for ana-

lyzing illicit drug markets and for estimat-ing the number of career criminals in apopulation, among other applications.

There are applications for which theapproach is not suited. It would not beuseful for estimating the prevalence of gen-eral drug use (that is, hardcore and occa-sional use combined) in a county. The rea-son is that for many occasional users ofdrugs the risk of arrest is negligible, so thenumbers would be very imprecise. But forany population that runs an appreciablerisk of arrest, this approach to estimationwould seem to provide tolerably good esti-mates of counts of what are otherwisehard-to-study populations.

NOTES1. Developing prevalence estimates in the ADAM population requires both mathematical modeling and the application of statistical sam-

pling theory. To make this chapter accessible to nonstatisticians, a heuristic explanation rather than a mathematically rigorous justifica-tion is presented. The latter is available in a separate publication, Rhodes, W., and R. Kling, “Estimating the Prevalence of Hardcore DrugUse Using ADAM Data,” final report submitted to the National Institute of Justice by Abt Associates Inc., Winter 2002 (NIJ grant88–IJ–CX–C001).

2. These last two complications are discussed in the section of this chapter on preliminary estimates.

3. Rhodes, W., and R. Kling, “Estimating the Prevalence of Hardcore Drug Use Using ADAM Data,” final report submitted to the NationalInstitute of Justice by Abt Associates Inc., Winter 2002 (NIJ grant 88–IJ–CX–C001).

4. Hardcore drug users in the community are unique individuals. Hardcore users in the booking population are not necessarily unique; thatis, for example, 300 arrests could represent 200 arrestees. If 100 hardcore users are arrested once, 100 are arrested twice, and 100 arearrested three times, then the estimation is 300 hardcore users in the booking population but 100 + 200 + 300 = 600 arrests.

5. Crime in the United States, 2000: Uniform Crime Reports, Washington, DC: U.S. Department of Justice, Federal Bureau of Investigation,2001: 233.