insights from the delaware study and the education policy ... · comm/media studies english ++...

Insights from the Delaware Study and the Education Policy Initiative

Instructional Cost Drivers 1992-2017

Tom Eleuterio

Higher Education ConsortiaUniversity of Delaware

Ti Yan

Higher Education ConsortiaUniversity of Delaware

Two decade study, focused at the academic discipline level, of faculty instructional workload and costs, sponsored research and public service from over 700 four-year, public and private non-profit higher education institutions

“The National Study of Instructional Cost & Productivity”, conducted annually by the Higher Education Consortia at University of Delaware. Results reported here from a grant for public policy research by the Smith Richardson Foundation

Research collaboration with the Education Policy InitiativeDr. Kevin Stange, University of Michigan and Dr. Steven Hemelt University of North Carolina- Chapel Hill

Presenter

Presentation Notes

Who in the audience has been part of the study? The NSICP is conducted in the spirit of education as a public good. The academic research by the EPI is also contributing to our shared knowledge. Please support our work by being a participant.

All Institutional Costs

Costs Allocated to Departments

Direct Instructional

Costs

Personnel: Faculty and

support staff

Salary

Benefits

Non-Personnel

Direct Public Service Costs

(sep. budgeted)

Direct Research Costs (sep. budgeted)

Indirect CostsExternally Funded Research & Dept.

Match Academic

Administration

Central Costs

Presenter

Presentation Notes

What disciplines have the lowest cost per SCH?

College prices ↑ 36% between 2008 and 2018 (College Board, 2018)

− Student and parents now pay for >50% of costs (Desrochers & Hurlburt, 2016)

− Concerns about access, persistence, and indebtedness

Improved understanding of cost differences by field, trends, and cost drivers is key tool for tempering future cost growth

Better understating of costs fuller picture of effects of policies such as financial aid, free college, and incentives to major in specific fields

− Example: High cost of engineering majors makes increasing STEM much less welfare-enhancing than earnings differences would imply (Altonji & Zimmerman, 2017)

Presenter

Presentation Notes

The AZ study combines 2-digit CIP cost analysis and labor outcomes for FL.

In other words, by coverage we answer questions like:− What do DCS participants look like?− How representative are they of the 4-yr sector as a whole?− How can we account for discrepancies between DCS sample and 4-yr sector?

“Because the population for this study is self-selected, it is, by definition, not a random sample. Descriptive statistics are applied to data from responding institutions to describe instructional expenditures for those institutions, but the findings cannot be inferentially generalized to the larger population of all Title IV-eligible 4-year colleges and universities in the United States” (Middaugh et al, 2003, p. viii).

What extent is voluntary participation a significant concern ?

Presenter

Presentation Notes

How much of IR survey data is based on random samples? Do non-random samples bias conclusions? Original DCS report published with NCES reflects this concern. We want to know, at least for measurable characteristics of participants, whether DCS is all that far off, how it matters, and what we can do about it.

Delaware Cost Study Participation 1998 - 2015

Presenter

Presentation Notes

What is the average percent of IPEDS participants covered in the DCS? Additional details in Appendix slides #

Who’s most likely to participate?− Public institutions− Research/master’s universities− Institutions with larger enrollments− Institutions with higher tuition than average− Expenditures per student in line w/ IPEDS data

Who’s less likely to participate?− Selective private institutions across all degree levels

Presenter

Presentation Notes

Please raise your hand if you are from an institution in each of these subgroups

How do we account for the pattern of participants?

Goal of weighting is to make sample look “more” like nation as a wholeGive greater weight to underrepresented schools, less weight to

overrepresented schools

Based on model of participation, predict Probability(DCS)

Calculate inverse of probability of participation 1/(Prob.(DCS))Can think of this is as, how many colleges does one respondent

stand in for?

Presenter

Presentation Notes

What is the inverse probability of getting a heads when you flip a fair coin?

Presenter

Presentation Notes

Weighted sample covers 90% of degrees, 82% of all CIP2s,

Does the weighting matter?

What type of weighting in the sample normalizes costs the most?

1. By the type of student credit hour (SCH)2. By the type of institution under consideration (IPW)3. By the interaction between student credit hour type and

institution type (IPW*SCH)

Process of constructing weights honed our sample for future analyses

− Identified outliers, missing data patterns

We focus a subset of 20 disciplines in the study 2000-2015− Cross-section (2013 – 2105): 7,245 obs → 3,307 programs → 256

institutions − Panel (2000-2015): 32,496 obs → 6,443 program, → 486 institutions

Supplemented by IPEDS, OK Salary Survey, other sources Focus on 20 programs (CIP4 codes) using English (CIP 23.0101) as

a reference for the other 19.

• Selected largely based on size, coverage, policy relevance

1) What are the cost differences, direct instructional expenses (DIE) by field of study?At point in time as well as trends over time

2) What drivers account for cost differences by field? Decompose level and trend differences into 4 candidate

drivers:(A) Personnel costs per FTE instructor(B) Non-personnel costs(C) Faculty workload(D) Class size

3) Lessons and next steps

0 50 100 150$1000s

AccountingMechanical Engineering

Electrical EngineeringBiz Admin/Mgmt/Operations

EconomicsComputer/Info Sciences

PhysicsNursing

ChemistryPoli Sci/Government

BiologyPsychology

SociologyMathematics

HistoryPhilosophy

Fine/Studio ArtsEducation

Comm/Media StudiesEnglish

++ denotes 95% CIX-section sample. Weight=SCH*IPW.

Average Non-personnel Share of DIE Expenses

Average Personnel Expenses

0 5 10% of DIE

ChemistryEducation

BiologyNursing

Mechanical EngineeringFine/Studio Arts

Electrical EngineeringComputer/Info Sciences

PhysicsComm/Media Studies

Biz Admin/Mgmt/OperationsPsychology

Poli Sci/GovernmentMathematics

EconomicsEnglish

AccountingSociology

HistoryPhilosophy

++ denotes 95% CIX-section sample. Weight=SCH*IPW.

Decomposition Decompose within-institution cross-field cost differences using accounting identity (which we log):

Simulate what DIE/SCH would be if actual average class size for field c at institution i were replaced with average for English at that same institution (but other factors held constant)

Cost for field c at institution i:

Non-personnel expenses

Average salary + benefits

1/workload 1/class size

Presenter

Presentation Notes

A slight mathematical digression; the formula allows us to control all variables but one in a simulation

0.74

0.630.60

0.310.26

0.23 0.21 0.200.16

0.110.07

0.00

-0.02-0.05

-0.09 -0.11-0.18 -0.18

-0.22 -0.25

-.4-.2

0.2

.4.6

.8%

diff f

rom E

nglis

h

Electrical E

ngineering

Nursing

Mechanical E

ngineering

Education

Fine/Studio Arts

Biz Admin/M

gmt/Operations

Computer/Info Scie

nces

Accounting

Physics

Chemistry

BiologyEnglis

h

Poli Sci/G

overnment

EconomicsHisto

ry

Comm/Media Studies

Psychology

Sociology

Philosophy

Mathematics

-0.051 -0.050

-0.431

-0.119

0.447

-.6-.4

-.20

.2.4

.6%

diff

from

Eng

lish

1. Bas

e

2. Othe

r exp

ense

s

3. Sala

ries

4. Fac

work

load

5. Clas

s size

Economics cheaper on average than English

Faculty get paid a lot more

But teach bigger classes that slightly more than offset salary difference

Presenter

Presentation Notes

First consider economics, which is approximately 5% less expensive than English in the raw data (bar 1). This difference remains at about 5% even when non-personnel expenses are equated between the two fields (bar 2), because they do not differ much on this dimension. However, economics faculty are much more highly paid, so if salarieswere equated in English and economics, the cost gap would increase from 5% to 43% (bar 3). On the other hand, economics classes tend to be much larger than English classes, so equating class size (but holding other attributes unchanged) would make economics 45% more expensive than English (bar 5). Faculty workload is a little lighter in economics than English, so equating the two fields on this dimension would make economics 12% more expensive than English. Thus,economics departments are able to increase class size to more than offset the higher salary and (slightly) lower workload, resulting in slightly lower average costs than English.

0.6010.557

0.051

0.510

0.684

-.6-.4

-.20

.2.4

.6.8

% d

iff fr

om E

nglis

h

1. Bas

e

2. Othe

r exp

ense

s

3. Sala

ries

4. Fac

work

load

5. Clas

s size

M.E. more expensive than English

Faculty get paid a lot more

With slightly lower teaching loads

But class size similar to English

Presenter

Presentation Notes

Mechanical engineering, which is 60% more expensive than English, provides a counter example. Like economics, mechanical engineering professors are also better paid and have lower teaching loads than English faculty. As a result, equalizing these two factors, especially salaries (bar 3), would greatly reduce the cost difference with English. Unlike economics, however, the classes are only modestly larger in mechanical engineering than in English. Class size differences are not large enough to offset the higher salary and lower teaching load, thus mechanical engineering remains much more expensive than English.

-.50

.51

-.50

.51

-.50

.51

-.50

.51

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5

Accounting Biology Biz Admin/Mgmt/Operations Chemistry Comm/Media Studies

Computer/Info Sciences Economics Education Electrical Engineering Fine/Studio Arts

History Mathematics Mechanical Engineering Nursing Philosophy

Physics Poli Sci/Government Psychology Sociology

% d

iff re

lativ

e to

Eng

lish

1: Actual difference; 2: Equate other exp; 3: Equate salary; 4: Equate workload; 5: Equate class size

Decomposition of cost differences

Potential policy implications

Longitudinal panel 2000through 2015

050

100

150

200

250

300

DIE/

SCH

2000 2005 2010 2015Year

Notes: Expressed in constant 2015 dollars; weighted by SCH*IPW.

Modest increase in average cost…

200

400

600

200

400

600

200

400

600

200

400

600

2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015


Computer/Info Sciences Economics Education Electrical Engineering English

Fine/Studio Arts History Mathematics Mechanical Engineering Nursing

Philosophy Physics Poli Sci/Government Psychology Sociology

DIE/

SCH

Year

0.5

11.

5An

nual

% c

hang

e

Accou

nting

Econo

mics

Biz Adm

in/Mgm

t/Ope

ration

s

Electric

al Eng

ineeri

ng

Comm/M

edia

Studies

Fine/Stud

io Arts

Mathem

atics

Poli Sci/

Govern

ment

Psych

ology

Educa

tion

Sociol

ogy

History

Chemist

ry

Philos

ophy

Compu

ter/In

fo Scie

nces

English

Biolog

y

Physic

s

Mecha

nical

Engine

ering

Nursing

Nursing

0.2

.4.6

.80

.2.4

.6.8

0.2

.4.6

.80

.2.4

.6.8

2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015 2000 2005 2010 2015


Computer/Info Sciences Economics Education Electrical Engineering English

Fine/Studio Arts History Mathematics Mechanical Engineering Nursing

Philosophy Physics Poli Sci/Government Psychology Sociology

Tenure/TT Other Supplemental

Teaching assistants

% F

acul

ty s

hare

Year

Weighted by SCH*IPW

Presenter

Presentation Notes

Average faculty that are Tenured or tenure track in nursing has declined from about 65% to about 40% while non-tenured , continuing track faculty has increased from 20% to 40%. Supplemental and TA faculty has remained flat for the remaining proportions.

0 20 40 60 80 100percent

SociologyPsychology

Poli Sci/GovernmentPhysics

PhilosophyNursing

Mechanical EngineeringMathematics

HistoryFine/Studio Arts

EnglishElectrical Engineering

EducationEconomics

Computer/Info SciencesComm/Media Studies

ChemistryBiz Admin/Mgmt/Operations

BiologyAccounting

Lower division Upper divisionGraduate

How do institutions respond to large increases in total and discipline-specific enrollment demand?

Do institutions Expand number of sections?Hire adjuncts?Increase class size?Increase faculty workload?Other strategies?

How do these patterns differ across discipline?

Have you experienced any large increases in student enrollment?Institution-wide?For specific majors/programs?

What caused the increase? Was it expected?

How did you respond? What levers did you adjust? Were you able to maintain quality?

Thank youPlease remember to submit your

evaluation for this session.

Vast majority of studies use institution-level data from IPEDS/Delta Cost

A few existing case studies focus on a subset of elite institutions− Clotfelter (1997): costs rise due to broad investments in quality, financial aid− Ehrenberg (2000): Lack of central control fails to restrain spending by units

Most similar previous work focused on department level analysis− Middaugh et al (2003): Cost differences across fields are sizable and explain

much of the variation in costs across institutions− Johnson and Turner (2009): Large differences in students per faculty across

departments− Courant and Turner (2017): Departments with expensive faculty economize

by having larger classes− Altonji and Zimmerman (2017): Cost of producing a major has welfare

implications for college major returns

Sample weighting and coverage

In IPEDS In DCS % covered % covered w/ weight

Public 582 333 57.2% 73.3%Private 1,204 278 23.1% 36.5%ALL 1,786 611 34.2% 60.1%

-.4-.2

0.2

.4.6

.81

% d

iff fr

om E

nglis

h

Electric

al Eng

ineeri

ng

Nursing

Mecha

nical

Engine

ering

Educa

tion

Fine/Stud

io Arts

Accou

nting

Chemist

ry

Biz Adm

in/Mgm

t/Ope

ration

s

Physic

s

Biolog

y

Compu

ter/In

fo Scie

nces

Poli Sci/

Govern

ment

Econo

mics

Sociol

ogy

Comm/M

edia

Studies

English

History

Mathem

atics

Psych

ology

Philos

ophy

Pri research Pri comprehensive Pri bacc

Pub research Pub comprehensive Pub bacc

-1-.5

0.5

1An

nual

% ch

ange

Chemistry

Nursing

Accounting

Mathematics

Economics

Computer/Info Scie

nces

Biz Admin/Mgmt/O

perationsBiology

Physics

HistoryEnglish

Fine/Studio Arts

Mechanical E

ngineering

Philosophy

Sociology

Psychology

Poli Sci/G

overnment

Education

Electrical E

ngineering

Comm/Media Studies

-2-1

01

2An

nual

% c

hang

e

Fine/Stud

io Arts

Educa

tion

Poli Sci/

Govern

ment

History

Biz Adm

in/Mgm

t/Ope

ration

s

Comm/M

edia

Studies

English

Econo

mics

Accou

nting

Philos

ophy

Sociol

ogy

Electric

al Eng

ineeri

ng

Psych

ology

Compu

ter/In

fo Scie

nces

Mathem

atics

Biolog

y

Physic

s

Chemist

ry

Nursing

Mecha

nical

Engine

ering

Mechanical Engineering

0.5

11.5

Annu

al %

chan

ge

Accounting

Economics

Biz Admin/M

gmt/Operations

Electrica

l Engineerin

g

Comm/Media Studies

Fine/Studio Arts

Mathematics

Poli Sci/G

overnment

Psychology

Education

Sociology

History

Chemistry

Philosophy

Computer/Info Scie

nces

English

Biology

Physics

Mechanical E

ngineering

Nursing

-10

12

3An

nual

% ch

ange

Mechanical E

ngineering

Nursing

Electrical E

ngineering

Physics

Biology

Chemistry

Psychology

Accounting

Mathematics

Economics

Comm/Media Studies

Biz Admin/M

gmt/Operations

Sociology

Philosophy

Poli Sci/G

overnment

Computer/Info Scie

ncesHisto

ry

Fine/Studio Arts

English

Education

-2-1

01

2An

nual

% c

hang

e

Fine/Studio

Arts

Educa

tion

Poli Sci/G

overn

ment

History

Biz Adm

in/Mgm

t/Ope

rations

Comm/M

edia

Studies

English

Econo

mics

Accounti

ng

Philoso

phy

Sociolo

gy

Electrica

l Eng

ineeri

ng

Psycholo

gy

Compu

ter/Info Scie

nces

Mathemati

cs

Biology

Physic

s

Chemistr

y

Nursing

Mechan

ical E

ngine

ering

English

0.5

11.

5An

nual

% c

hang

e

Accou

nting

Econo

mics

Biz Adm

in/Mgm

t/Ope

ration

s

Electric

al Eng

ineeri

ng

Comm/M

edia

Studies

Fine/Stud

io Arts

Mathem

atics

Poli Sci/

Govern

ment

Psych

ology

Educa

tion

Sociol

ogy

History

Chemist

ry

Philos

ophy

Compu

ter/In

fo Scie

nces

English

Biolog

y

Physic

s

Mecha

nical

Engine

ering

Nursing

-1-.5

0.5

1An

nual

% c

hang

e

Chemist

ry

Nursing

Accou

nting

Mathem

atics

Econo

mics

Compu

ter/In

fo Scie

nces

Biz Adm

in/Mgm

t/Ope

ration

s

Biolog

y

Physic

s

History

English

Fine/Stud

io Arts

Mecha

nical

Engine

ering

Philos

ophy

Sociol

ogy

Psych

ology

Poli Sci/

Govern

ment

Educa

tion

Electric

al Eng

ineeri

ng

Comm/M

edia

Studies

-10

12

3An

nual

% c

hang

e

Mecha

nical

Engine

ering

Nursing

Electric

al Eng

ineeri

ng

Physic

s

Biolog

y

Chemist

ry

Psych

ology

Accou

nting

Mathem

atics

Econo

mics

Comm/M

edia

Studies

Biz Adm

in/Mgm

t/Ope

ration

s

Sociol

ogy

Philos

ophy

Poli Sci/

Govern

ment

Compu

ter/In

fo Scie

nces

History

Fine/Stud

io Arts

English

Educa

tion

Class sizes, 2000-2015

Faculty salaries, 2000-2015 Faculty workload, 2000-2015

Presenter

Presentation Notes

Three cost components for English; small increase in salaries, small decreases in workload and class size

0.5

11.

5An

nual

% c

hang

e

Accou

nting

Econo

mics

Biz Adm

in/Mgm

t/Ope

ration

s

Electric

al Eng

ineeri

ng

Comm/M

edia

Studies

Fine/Stud

io Arts

Mathem

atics

Poli Sci/

Govern

ment

Psych

ology

Educa

tion

Sociol

ogy

History

Chemist

ry

Philos

ophy

Compu

ter/In

fo Scie

nces

English

Biolog

y

Physic

s

Mecha

nical

Engine

ering

Nursing

-1-.5

0.5

1An

nual

% c

hang

e

Chemist

ry

Nursing

Accou

nting

Mathem

atics

Econo

mics

Compu

ter/In

fo Scie

nces

Biz Adm

in/Mgm

t/Ope

ration

s

Biolog

y

Physic

s

History

English

Fine/Stud

io Arts

Mecha

nical

Engine

ering

Philos

ophy

Sociol

ogy

Psych

ology

Poli Sci/

Govern

ment

Educa

tion

Electric

al Eng

ineeri

ng

Comm/M

edia

Studies

-10

12

3An

nual

% c

hang

e

Mecha

nical

Engine

ering

Nursing

Electric

al Eng

ineeri

ng

Physic

s

Biolog

y

Chemist

ry

Psych

ology

Accou

nting

Mathem

atics

Econo

mics

Comm/M

edia

Studies

Biz Adm

in/Mgm

t/Ope

ration

s

Sociol

ogy

Philos

ophy

Poli Sci/

Govern

ment

Compu

ter/In

fo Scie

nces

History

Fine/Stud

io Arts

English

Educa

tion

Class sizes, 2000-2015

Faculty salaries, 2000-2015 Faculty workload, 2000-2015

Student Credit Hours (thousands)

Unemployment Rate

5 ppt increase in unemployment rate

20% increase student credit

hours

Note: Sample restricted to balanced panel of programs participating in Delaware Cost Study at least 13 times

insights from the delaware study and the education policy ... · comm/media studies english ++...

Documents