The supply & demand of motivated labor:When should we expect to see nonprofit wage gaps?

Daniel JonesUniversity of Pittsburgh

October 23, 2012IUPUI - WIMPS

Nonprofit wages

Nonprofit workers on average earn less than for-profit workers –Why?

1 Composition of nonprofit sector

2 Labor donation hypothesis: Nonprofit firms typicallythought of as providing some service, so (some) workersmay be willing to accept lower wages.

Discussed by multiple authors: e.g., Hansmann (1980),Preston (1989), Rose-Ackerman (1996)

Empirical literature: clearly wage gaps in some industries, butnot others (Leete, 2001)

In this talk...

Compare nonprofit and for-profit workers to determine:

To what extent are workers willing to sacrifice earnings towork in a job they “care about”?

When should we expect such wage differences to emerge asa result?

How does sorting into jobs on the basis of motivationimpact firm productivity?

Motivated labor

Recent theoretical work examines implications of motivation forlabor market outcomes(Besley & Ghatak, 2005; Delfgaauw & Dur, 2007; Ghatak & Mueller, 2011;

Heyes, 2005)

“Mission-oriented” (MO) firms and “profit-oriented” (PO)firms

“Motivated” workers enjoy supplying effort to the MO firmand are willing to work for less

Result: MO firm can offer lower wages, minimize costs, andattract only motivated applicants

Nonprofit wages

When should we expect motivation to generate observable wagegaps?

Argument: Building on a point made by Preston (1989),nonprofit (or “mission-oriented” firms more generally) canonly maintain lower wages when there are enoughintrinsically motivated workers to meet labor demand

Hypothesis: Nonprofit wage gaps should arise whennonprofit labor demand is low relative to for-profit labordemand

Preview of approach and results

Compare nonprofit wage gaps across localities but withinindustries as a function of nonprofit share

1 Economy-wide (indvidual-level Census data)2 Specific industry (Nursing homes)

Results:

Wage gaps exist when nonprofit share is lowWages roughly equal when nonprofit share is highIn nursing homes, quality of work is higher when nonprofitshare is low

Remainder of talk

1 Existing literature

2 Theoretical framework

3 Wage gaps in economy-wide data

4 Wage gaps & quality differences in nursing home data

5 Additional results: Differences in worker characteristics?

6 Conclude

Existing literature: Motivation in the nonprofit sector

What would motivate nonprofit workers?

“Warm glow” or moral satisfaction from contributing toproduction of public good / social service (Preston (1989),Frank (1996))

Elsewhere, lack of profit motive tends to lead to(inefficiently) high quality output (Hansmann, 1980)

Existing literature: Motivation in the nonprofit sector

Is there direct evidence of motivation?

Nonprofit workers more willing to engage in unpaidovertime (Gregg et al., 2011)

Report a higher number of “ideal number of hours worked”(Lanfranchi et al., 2011)

Existing literature: Nonprofit wages

Testing labor donation hypothesis within particular industries:

Significantly negative wage differential amongst publicinterest lawyers (Weisbrod (1983), Frank (1996))

Little/no wage differential amongst nonprofit nursing homeworkers (Holtmann & Idson (1993)) and amongst day careworkers in competitive markets (Preston (1993))

Existing literature: Nonprofit wages

Testing labor donation hypothesis at economy-wide level:

Preston (1989): some evidence of a negative wage gap amongstmanagers and professionals

Ruhm and Borkoski (2002): no “general” wage gap

Suggest that “compensation is primarily determined incompetitive markets without ‘labor donations’ to nonprofitemployers”

Leete (2001): no general wage gap, but significant gaps inparticular industries

Particularly prominent gaps amongst highly educatedworkers

Theoretical framework

When will presence of motivation generate observable wagegaps?

Use a simple model to add precision to the argument

Use Delfgaauw & Dur’s (2007) model of signaling andscreening worker motivation as a starting point

Theoretical framework: Job types

Two jobs available: for-profit (FP) and non-profit (NP)

Required level of effort identical across the two jobs

Wages (may) vary across the two jobs

Wage only received if “required level of effort” is provided

Theoretical framework: Worker types

Workers care about two things: how much they get paid andeffort.

Two types of workers: “standard” and “motivated”

Standard: Effort is costly; benefit of a job is (Wage - Costof effort) Figure

Motivated: To some extent, enjoy supplying effort to NPFigure

Theoretical framework: Worker types

Workers care about two things: how much they get paid andeffort.

Two types of workers: “standard” and “motivated”

Standard: Effort is costly; benefit of a job is (Wage - Costof effort) Figure

Motivated: To some extent, enjoy supplying effort to NPFigure

Workers’ decisions: With wage gap

Workers’ decisions: With large gap

Workers’ decisions: With no wage gap

Nonprofit firm behavior

Nonprofit firm’s task: Hire enough workers to produce q̄with goal of minimizing costs

Ideally: Keep wage in range that attracts only “motivated”workers:

Nonprofit firm behavior

Problem: Only some fraction of workers are motivatedtypes

If fraction of available workers demanded by NP < fractionof motivated workers, lower wage is possible

Nonprofit wage < for-profit wage:Mo'vated#

Workers:#

NP#Jobs:#FP#

Nonprofit firm behavior

Problem: Only some fraction of workers are motivatedtypes

If fraction of available workers demanded by NP > fractionof motivated workers, nonprofit must raise wages to attractstandard workers and meet production quota

Nonprofit wage < for-profit wage:Mo'vated#

Workers:#

NP#Jobs:#FP#

Nonprofit firm behavior

Problem: Only some fraction of workers are motivatedtypes

If fraction of available workers demanded by NP > fractionof motivated workers, nonprofit must raise wages to attractstandard workers and meet production quota

Nonprofit wage = for-profit wage:

Workers:#

NP#Jobs:#

Mo'vated#

FP#

Resulting predictions

NP can minimize costs by offering lower wage only whenNonprofit Share of jobs is low; attracts only “motivated”workers

Predictions:

NP wage gap when nonprofit share is low → wages equalizeas nonprofit share increasesHigher average effort in NP when nonprofit share is low

Empirical strategy and hypotheses

Expect industries with higher nonprofit shares to have smallerwage differentials

Data from Leete (2001) roughly consistent with the hypothesis

-.6-.4

-.20

.2wa

ge d

iffer

entia

l

0 .2 .4 .6 .8nonprofit industry share

But comparing across industries is not ideal

Empirical strategy and hypotheses

Instead, compare nonprofit wage gaps within industries butacross localities

Construct industry-specific nonprofit labor shares bylocality

Within an industry –

Nonprofit wage differentials should be largest in localitieswith low (industry-specific) nonprofit labor shareAs nonprofit share increases, wages should be roughly equalacross nonprofit and for-profit

Outline of empirical approach

1 Using microdata from 5% sample of 2000 U.S. Census –examine relationship between “nonprofit share of labor”and the existence of a nonprofit wage differential withinindustries

2 Using much richer data on a particular industry (nursinghomes) – further explore relationship and attempt to ruleout alternative explanations for economy-wide findings

Data: Economy-wide analysis

2000 Census 5%-sample public use microdata

Construct nonprofit shares by industry at theSuper-PUMA level. Sum the total number of nonprofit andfor-profit workers for a particular industry i andSuper-PUMA s and construct:

NPSis =(total NP emp.)is

(total NP emp.)is + (total FP emp.)is

Main empirical specification

Restrict attention to full time workers in private or non-profitsectors

Include Super-PUMA, industry, occupation fixed effects andcontrols for education, gender, (potential) experience, and race

ln(earnings) = α+ β1nonprofit+ β2nonprofit X NPS

+β3NPS + [controls/FEs]

Nonprofit wage differential is measured by β1 + β2 ×NPS

Argument suggests that β1 < 0 and β2 > 0

Baseline specification

(1) (2) (3) VARIABLES Log(Wage) Log(Wage) Log(Wage) Nonprofit -0.0273*** -9.20e-05 -0.0830*** (0.00408) (0.00449) (0.00639) Nonprofit X NP share 0.126*** 0.0764*** 0.192*** (0.00955) (0.0115) (0.0136) NP share -0.0668*** -0.0395** -0.116*** (0.0197) (0.0172) (0.0316) Constant 2.768*** 2.694*** 3.081*** (0.0164) (0.0235) (0.0317) Educ. group All Low High Observations 899,124 564,020 335,104 R-squared 0.450 0.363 0.335

*** p<0.01, ** p<0.05, * p<0.1“High educ.” = Bachelor’s degree or higher.

Specifications include industry, occupation, & super-PUMA FEs and worker-level controls.

High educ.: Wage gap of -8% when NPS is low. Wage gapof +0.015% when NPS is 0.50.

Specific industries

(1) (2) (3) Industry: Clinics Hospitals Research/Dev. Nonprofit -0.0535* 0.0272 -0.234*** (0.0319) (0.0262) (0.0486) Nonprofit X NPS 0.236 -0.00275 0.413*** (0.207) (0.0567) (0.145) Observations 13,200 51,024 5,286 R-squared 0.454 0.286 0.367 (4) (5) (6) Industry: Nursing homes Media Legal services Nonprofit -0.0700* -0.227*** -0.485*** (0.0400) (0.0330) (0.151) Nonprofit X NPS 0.123 0.584* 0.457 (0.171) (0.328) (2.373) Observations 6,562 13,141 1,454 R-squared 0.426 0.330 0.322 *** p<0.01, ** p<0.05, * p<0.1

!

Sample split by “collectiveness”

Industries in the nonprofit sector vary in the degree to whichthey produce public good or provide social services. Are resultsdriven by one sort of nonprofit?

Weisbrod (1988) proposes index of “collectiveness” = fraction ofindustry’s revenue from public sources

I construct publicness index using NCCS data on nonprofit taxreturns from 2000

Sample split by “collectiveness”

Collectiveness: Low Collectiveness: High VARIABLES ln(Wage) ln(Wage) Nonprofit -0.0569*** -0.138*** (0.00930) (0.0106) Nonprofit X NP Share 0.174*** 0.245*** (0.0248) (0.0192) NP Share -0.146*** -0.0931** (0.0422) (0.0363) Constant 3.386*** 3.201*** (0.0325) (0.0373) Observations 151,369 143,295 R-squared 0.304 0.350

*** p<0.01, ** p<0.05, * p<0.1

Nonlinear impact of nonprofit share

Model predicts:

1 Large wage differential with low nonprofit share2 Wage equalization as nonprofit share increases

(2) ⇒ positive relationship between wage and “Nonprofit XNP Share” when NP share is low, little relationship whenNP share is high

Two approaches:

Cubic specification in nonprofit shareSpline regression: Let the impact of “Nonprofit X NPshare” vary within each quartile of nonprofit share

Nonlinear impact of nonprofit share

! 18

Finally, while the preceding results document the predicted directional impact of nonprofit share

on wages, we should expect to see the wage differential decreasing with nonprofit share within

some range but ultimately flattening out as wages equalize. To assess this, I modify the baseline

specification to allow for a cubic relationship between nonprofit share and wages:

!" !"#$%$&'= ! + !!nonprofit + !! nonprofit!×!NPS + !!NPS + !! nonprofit!×!NPS! + !!NPS!

+ !! nonprofit!×!NPS! + !!NPS! + [controls/FEs]

The resulting wage differential (as a function of nonprofit share) is plotted in Figure 2, with

nonprofit share along the x-axis and wage differential on the y-axis. With low nonprofit share,

highly educated nonprofit workers earn roughly 12% less than their for-profit counterparts.

However, this differential shrinks as nonprofit share increases until nonprofit share reaches a

certain point, after which there is essentially no difference in wages across the two sectors.

Figure 2: Wage differential as a function of nonprofit share

(Shaded area represents 95% confidence interval)

-.15

-.1-.0

50

.05

Non

profi

t wag

e di

ffere

ntia

l (%

)

0 .1 .2 .3 .4 .5 .6 .7 .8Nonprofit share

Shaded area is 95% confidence interval

Spline regression

Let the impact of “Nonprofit X NP share” vary within eachquartile of nonprofit share

NPS quartile range ln(Wage) Nonprofit (0) -0.103** (0.0471) Nonprofit X NP Share (1st quartile) 0-0.02 -0.702 (2.746) Nonprofit X NP Share (2nd quartile) 0.02-0.07 1.244*** (0.378) Nonprofit X NP Share (3rd quartile) 0.07-0.43 0.199*** (0.0322) Nonprofit X NP Share (4th quartile) 0.43-1 -0.0211 (0.0395)

*** p<0.01, ** p<0.05, * p<0.1

Initial results

Economy-wide results consistent with hypothesis:

Negative wage differential when nonprofit share is low

Wage differential minimized as nonprofit share increases

Particularly true for highly educated workers (consistentwith previous literature)

Initial results

Concerns remain:

Despite theoretical prediction, are results driven bylower-quality nonprofit workers in low NPS areas?

Different competitive environment with low NPS?

High NPS driven by nonprofit-friendly politicalenvironment, in turn making it easier for firms to payhigher wages?

Nursing home analysis

Why nursing homes?

Detailed dataset from Department of Health & HumanServices that covers 95% of nursing homes in the US

Rich existing literature (e.g., Borjas et al. (1983),Holtmann & Idson (1993)) – some disagreement, but:

Researchers typically find very weak positive wagedifferential or no difference at all

Nursing home data

Firm-level data includes:

Number of residents and bedsNumber of labor hours utilizedNon-profit/for-profit statusQuality measures

Construct precise nonprofit labor share assessed atCommuting-zone (CZ) level

NPSCZ =(sum of staff labor hours across NP firms in CZ)

(sum of staff labor hours across all firms in CZ)

Nursing home data does not include any information aboutwages

I pair this data with data from the 2010 AmericanCommunity Survey (1%-sample of US population)

Market characteristics

CZ fixed-effects eliminate concerns about “general” labormarket characteristics

Still concern about differential impact of marketcharacteristics on nonprofit & for-profit wages

With this in mind, I include interactions of nonprofit andvarious market characteristics

number of NH residents, NP firms, and FP firms,(evaluated at the mean) ormeasure of market concentration (HHI)

Market characteristics

(1) (2) (3)

Nonprofit -0.173*** -0.186*** -0.175***(0.0533) (0.0635) (0.0543)

Nonprofit X NPS 0.386*** 0.418** 0.389***(0.141) (0.180) (0.144)

Nonprofit X NH residents 2.77e-07(2.58e-06)

Nonprofit X NP firms -0.000194(0.00128)

Nonprofit X FP firms 0.000139(0.000296)

Nonprofit X HHI -0.135(1.131)

Observations 2,082 2,082 2,082R-squared 0.506 0.506 0.506

Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Endogeneity of nonprofit share

Two instrumental variables for NPS:

1 Growth in elderly population in first half of the decade (as in Sloan etal. (2001), Grabowski & Hirth (2003))

2 10-year lag in nonprofit share of non-medical industries

(1) (2) (3) (4) IV: Growth in elderly population IV: Lagged unrelated-industry NPS VARIABLES First stage Second stage First stage Second stage Nonprofit -0.352** -0.298*** (0.163) (0.101) Nonprofit X NPS 0.996* 0.812** (0.558) (0.339) Instrument -0.708*** 3.081*** (0.247) (0.687) Observations 2,084 2,084 2,084 2,084 R-squared 0.890 0.503 0.903 0.505

*** p<0.01, ** p<0.05, * p<0.1

Endogeneity of nonprofit share

Two instrumental variables for NPS:

1 Growth in elderly population in first half of the decade (as in Sloan etal. (2001), Grabowski & Hirth (2003))

2 10-year lag in nonprofit share of non-medical industries

(1) (2) (3) (4) IV: Growth in elderly population IV: Lagged unrelated-industry NPS VARIABLES First stage Second stage First stage Second stage Nonprofit -0.352** -0.298*** (0.163) (0.101) Nonprofit X NPS 0.996* 0.812** (0.558) (0.339) Instrument -0.708*** 3.081*** (0.247) (0.687) Observations 2,084 2,084 2,084 2,084 R-squared 0.890 0.503 0.903 0.505

*** p<0.01, ** p<0.05, * p<0.1

Quality of output

Theoretical predictions: “average effort” should be highestwhen nonprofit share is low

Compare the quality of output at NP & FP firms usingnursing home “quality ratings”

Quality ratings are derived from safety inspections and acombination of health measures of residents (e.g., “Percentof residents with pressure sores”)

Quality of output

(1) (2)

Nonprofit 0.712*** 0.386***(0.0820) (0.0819)

Nonprofit X NP Share -0.797*** -0.398*(0.236) (0.221)

Firm-level controls X

Observations 13,393 13,089R-squared 0.117 0.182

Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1

Both specifications include CZ fixed effects and are clustered atCZ level

Firm-level controls include firm’s share of their market, RNhours, LPN hours, other staff hours, total residents, and ratio ofresidents to beds

Summarizing results

!!!!!

!

-0.2!

-0.16!

-0.12!

-0.08!

-0.04!

0!

0.04!

1st quartile! 2nd quartile! 3rd quartile! 4th quartile!Nonprofit share!

Wage gap by quartile!

0!

0.1!

0.2!

0.3!

0.4!

0.5!

0.6!

1st quartile! 2nd quartile! 3rd quartile! 4th quartile!Nonprofit share!

Quality rating gap by quartile!

Are “motivated types” really driving results?

Results: Negative wage gap and positive quality gap inlow-NPS areas; both go away in high-NPS areas

Claim: Driven by different “types” of workers choosing NPjobs in low- versus high-NPS areas

Any observable differences in workers?

Are “motivated types” really driving results?

Use National Longitudinal Survey of Youth (1997) toprovide suggestive evidence of differences in workers

Fewer observations, but much richer information aboutworkers available

1 How does job satisfaction vary with NPS?(NLSY: Reported job satisfaction)

2 How do prosocial motivations vary with NPS?(NLSY: Any volunteering or charitable activities)

Results

(1) (2) (3) VARIABLES ln(Wage) Satisfied? Prosocial? Nonprofit -0.354*** 0.168** 0.679*** (0.125) (0.0788) (0.258) Nonprofit X NPS 0.629** -0.234 -1.142** (0.280) (0.181) (0.536) Observations 1147 1046 401 R-squared 0.385 0.272 0.574

“Satisfied?” = 1 if the worker likes their job “very much” or “fairly well”; 0 otherwise. “Prosocial?” =1 if the worker reports any volunteering or donating; 0 otherwise.

Main findings

Wage gaps are largest when nonprofits’ share of labor is low

In economy-wide data: 8-12% wage penalty when nonprofitshare is very low(Roughly speaking –) Wages equalized beyond nonprofit share of0.3

In nursing homes: Quality of output in nonprofits is highest whenwage gap is largest

NP workers more satisfied with jobs and more prosocial in low-NPS

Consistent with the argument:

Wage gaps emerge when there enough “intrinsically motivated”workers to meet labor demandIn turn, wage gap leads to screening out workers not motivatedby the “mission” (increasing average effort)

Implications: Nonprofit wages literature

Exisiting literature leads to mixed results: wage gaps insome industries but not others – but unclear why

Accounting for nonprofit share may reconcile these results

Illustrative example:

No wage gap in hospitals ↔ Nonprofit share > 0.4Wage gap in legal services ↔ Nonprofit share < 0.05

Thus, absence of wage gap is not evidence against the“labor donation hypothesis”

Implications: Motivated labor literature

Provides empirical support for recent theoretical models(e.g., Besley & Ghatak, 2005; Delfgaauw & Dur, 2007;Ghatak & Mueller, 2011; Heyes, 2005)

Motivation-based labor market sorting has importantimplications for wages and the quality of workers firmsattract (and speaks to when we should expect such sortingto be important)

When motivation is important, higher pay may not be theclear solution to improving worker quality

Parallels recent result: Suggestive evidence that higher payfor politicians reduces politician quality (Fisman et al.,2012)

Standard workers

Effort

Bene

fit

FP wage

NP wage

Benefit from NP

Benefit from FP

Required effort

Back

“Motivated” workers

Effort

Bene

fit

FP wage

NP wage

Benefit from NPBenefit from FP

Required effort

Back