Performance pay and employeeturnover

Patrick L. OHalloranDepartment of Economics, Finance and Real Estate, Leon Hess Business School,

Monmouth University, West Long Branch, New Jersey, USA

Abstract

Purpose The purpose of this paper is to explore how various performance related pay (PRP)schemes influence employee turnover. It also tests whether profit sharing has a differential impact onturnover in comparison to other forms of PRP.

Design/methodology/approach Utilizing a nationally representative longitudinal dataset ofindividuals, analysis begins with a parsimonious specification of the determinants of turnover and thenprogressively adds various sets of controls known to influence turnover decisions to observe how theirinclusion influences PRP coefficients. Estimations employ both standard probits and panel data models.

Findings Empirical evidence reveals a negative relationship between an aggregate measure of PRPand turnover. Disaggregating performance pay measures by type reveals a robust negativerelationship between profit sharing and turnover. Although one would expect the influence of otherPRP schemes to mimic that of profit sharing, evidence suggests otherwise.

Research limitations/implications Data lack information on how much earnings are based onPRP. Consequently, estimates may be biased when combining those who receive little earnings fromPRP with those who receive substantial amounts of PRP into a single PRP measure.

Practical implications Although PRP schemes are often introduced to improve incentives andproductivity, profit sharing based on firm profitability may allow labor costs to vary with firm profitshence enhancing retention and reducing the incidence of unemployment during recession.

Originality/value This paper adds to the literature and fulfils an identified need to study howother types of PRP besides profit sharing influence turnover.

Keywords Performance pay, Turnover, Job satisfaction, Profit sharing, Performance management,Company profit sharing schemes

Paper type Research paper

I. IntroductionDespite vast literatures on both employee turnover and the many impacts ofperformance related pay (PRP), very few studies explicitly investigate how employeeturnover is affected by other PRP schemes besides profit sharing. Prior researchsuggests that workers receiving individual PRP and profit sharing are more satisfiedwith their jobs (Heywood and Wei, 2006; Green and Heywood, 2008; Kruse et al., 2010)than those receiving pay based on traditional time-rates, even after accounting for thehigher pay levels associated with PRP. Moreover, it seems clear that there existspositive selection on PRP whereby workers who are relatively less risk averse, more

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0144-3585.htm

The author would like to thank Dr John S. Heywood for his continued guidance and supportas well as an anonymous referee for constructive comments. The author would also like to thankhis beloved wife Susanne for her continual encouragement. All remaining errors are the authorsown.

Performance payand employee

turnover

653

Received 19 March 2011Accepted 24 October 2011

Journal of Economic StudiesVol. 39 No. 6, 2012

pp. 653-674q Emerald Group Publishing Limited

0144-3585DOI 10.1108/01443581211274601

confident and more able receive a return or rent on these characteristics in PRP jobs thatthey cannot obtain in a standard time-rate job (Curme and Stefanec, 2007).Consequently, if PRP workers are more satisfied with their jobs, they mayexperience lower turnover than traditional time-rate workers (Clark et al., 1998)making a result thought limited to workers on profit sharing apply more broadly to allworkers receiving PRP. Furthermore, prior studies find mixed support for Weitzmans(1985) hypothesis that stock options and profit sharing enhance job security byallowing marginal labor costs to vary with profitability (Kruse, 1993; Kruse et al., 2010).Also, comparing turnover propensities between PRP and time-rate workers providesfurther evidence as to which of the two broad models of PRP may be most applicable;the classic agency theory or the sorting theory of Lazear (1981, 1983). According toclassic agency theory, PRP workers are fully compensated for the inherent higher risksassociated with PRP, leaving them indifferent between PRP jobs and standard time-ratejobs, implying no difference in turnover. Alternatively, according to the sorting theory,PRP tends to attract the less risk averse and more able, who receive a rent they wouldnot obtain in a time-rate position, implying lower turnover among PRP workers.

Research presented herein estimates the impact various forms of PRP has on USemployee turnover including both quits and layoffs (those who report job terminationdue to firing, layoff, or plant closing), by analyzing data from six waves of the NationalLongitudinal Survey of Youth 1979 (NLSY79). Although there is an abundance ofresearch on how profit sharing affects job turnover, very little research focuses on howother PRP schemes such as piece rates or bonuses influence job turnover. Specifically,the empirical analysis focuses on whether workers receiving PRP, both aggregated anddisaggregated by type, experience differing rates of turnover than those paidtime-rates. Critically, few studies explicitly explore how other types of PRP besidesprofit sharing influence job turnover.

There are several different PRP schemes. PRP can be based on individualperformance as is typical with piece rates and commissions or based on collectiveperformance as is typical with profit sharing and stock options. Furthermore, PRP can bebased on subjective measures of effort such as how hard one appears to be working, orbased on objective measures of effort such as the number and quality of units produced.Given the varied incentives provided from different PRP schemes, one would expecteach to affect employee turnover differently.

Empirical analysis of the nationally representative sample reveals that those whoreceive any form of PRP experience lower rates of job turnover than those who do notreceive any PRP, even after controlling for many commonly known determinants ofturnover. Most importantly, when PRP is disaggregated by type, significant differencesemerge. Workers who receive pay in the form of stock options and profit sharingexperience significantly lower rates of turnover than those who do not, supporting theWeitzman hypothesis. Conversely, there are no significant differences in turnover betweennon-PRP workers and those receiving piece rates, commissions, and tips. Also, there isweak evidence that those who receive bonuses experience fewer layoffs. Consequently, thelower turnover rates observed using the aggregated PRP measure are mainly driven bystock options and profit sharing, suggesting that those forms of PRP may be unique.Splitting turnover into quits and layoffs reveals that workers participating inprofit-sharing plans are significantly less likely to experience a layoff or quit whileworkers receiving stock options are significantly less likely to quit. Findings also

JES39,6

654

corroborate those of Azfar and Danninger (2001) and Green and Heywood (2010), whotheorize that profit sharing allows for increased expected returns on firm-specific humancapital investments such as on-the-job training due to higher expected tenure. AlthoughPRP schemes are typically introduced to increase worker productivity, employers utilizingprofit sharing or stock option schemes may experience lower labor force turnover.

The paper proceeds as follows: Section II will provide background on the relationshipbetween PRP, job satisfaction, and turnover; Section III will review the data and providerationales for covariate selection; Section IV will summarize the empirical results;and Section V offers conclusions.

II. Past researchA. The relationship between PRP, job satisfaction and turnoverThere are two broad models of PRP. The classic agency model of PRP presumes thatfirms pay workers just enough to compensate them for the greater variability in pay thatis associated with PRP jobs, as well as rewarding greater effort typically put forth in PRPjobs. This model implies that, since workers are fully compensated for the greater riskand effort PRP entails, workers retain the same level of utility that they would have hadin a standard time-rate position. Consequently, according to agency theory, jobsatisfaction would be no different in a PRP job as opposed to a time-rate job. Hence, onewould expect similar rates of job turnover between PRP workers and time-rate workers.

Alternatively, Lazear (1981, 1983) argues that firms face a zero profit constraint andworkers sort to capture rents they would not be able to obtain in a standard time-rateposition. These rents arise because PRP is thought to attract those workers who areless risk averse and more able, and who will be more satisfied in jobs where theiradditional effort is commensurately rewarded. If workers cannot obtain similar rents ina time-rate position, one would expect to observe lower rates of turnover in jobsassociated with PRP. Recent research by Cornelissen et al. (2011) make it clear thatwhile the Lazear theory of PRP implies rents, other models such as the classic agencymodel do not. Accordingly, this paper provides further evidence as to which theory ismost appropriate in modeling workers reactions to PRP. If PRP is associated withlower turnover, the Lazear sorting model would better depict how PRP influencesworkers. If PRP is not associated with lower turnover, the classic agency theory ofworker response to PRP would appear a better depiction of reality.

A growing body of literature finds that PRP tends to increase overall jobsatisfaction. Heywood and Wei (2006) find that both individual performance pay andprofit sharing are positively related to job satisfaction while Green and Heywood (2008)find that PRP workers are more satisfied with their jobs, pay level, job security andhours than non-PRP workers. These findings support the theory that PRP allowsworkers to more freely optimize on various dimensions without crowding out intrinsicmotivation, thereby generating positive rents for workers unavailable in non-PRP jobs.Furthermore, Brown and Sessions (2006) argue that workers prefer work environmentswhere their earnings equal their marginal revenue product, which they find improvesoptimism concerning future employment relationships. Additionally, both Goddard(2001) and Bauer (2004) find that PRP schemes are often part of a larger package ofhuman resource management practices that are known to improve job satisfaction.Critically, prior evidence demonstrates that less satisfied workers typically experiencemore job turnover than more satisfied workers (Freeman, 1978; Akerlof et al., 1988;

Performance payand employee

turnover

655

Clark et al., 1998; Garboua et al., 2007). Therefore, presuming more satisfied workersare less likely to experience turnover, one would expect to observe a negativerelationship between PRP and job turnover.

It is also well known that introducing PRP schemes causes large amounts ofself-selection to occur, whereby workers preferring PRP schemes self-select into jobs thatoffer PRP schemes. Sliwka and Grund (2006) find the higher inherent risk associated withPRP schemes attract a disproportionate number of less risk-averse workers who aremore satisfied accepting greater pay variability in return for higher pay levels. Manystudies identify a positive correlation between PRP and earnings (Pencavel, 1972; Seiler,1984; Brown, 1990; Ewing, 1996; Booth and Frank, 1999; Parent, 1999). Indeed, Weitzman(1985) and Kruse (1993) argue that PRP schemes such as profit sharing and stock optionsmay result in a decreased probability of layoffs, since firms are able to decrease paymentswhen profitability is lower. Consequently, the increased job security associated with PRPbased on firm performance may result in greater job satisfaction, especially satisfactionwith job security previously observed by Green and Heywood (2008).

However, much research focuses on how PRP decreases job satisfaction. In particular,Baker (1992) finds that PRP fails to increase job satisfaction when performancemeasures are overly subjective and when evaluations are poorly tied to actualperformance. Also, PRP decreases job satisfaction if used solely to intensify low skilledor menial workers effort. This is especially true when increased efforts are not rewardedwith higher earnings due to ratcheting or when the PRP scheme does not permit greaterflexibility to optimize (McCausland et al., 2005). In these situations, workers will be lesssatisfied under the PRP scheme, since they exert more effort without commensuraterewards. Furthermore, when PRP is sensitive to external phenomenon beyond theworkers control, job satisfaction declines because of uncontrollable pay variability(Milgrom and Roberts, 1992). Additionally, Kennedy (1995) finds PRP typicallyincreases pay dispersion, creating a disincentive to workers receiving less than theirpeers, reducing satisfaction with pay (Brown, 2001). Psychological studies also revealthat PRP is a form of extrinsic motivation which may crowd out intrinsic motivation toperform well, resulting in lower satisfaction among PRP workers (Frey, 1997; Kreps,1997; Benabou and Tirole, 2003). Therefore, presuming less satisfied workers are morelikely to experience job turnover, one would expect to observe a positive relationshipbetween PRP and turnover.

Consequently, whether PRP increases or diminishes satisfaction is theoreticallyambiguous. Therefore, assuming that job satisfaction influences turnover, PRP could beassociated with either higher or lower turnover. Observing lower turnover among thoseworking in PRP jobs would tend to support the sorting hypothesis, while observing nodifference in turnover or higher turnover would tend to support classic agency theory.

B. The effect of specific PRP plans on turnoverDifferent PRP schemes likely impart contrasting incentives and hence may result indissimilar impacts on satisfaction and turnover. Unsurprisingly, prior findings reveal amixed picture concerning the relationship between various PRP schemes and turnover.Prior evidence reveals that piece rates and commissions are associated with higherturnover (Lazear, 1986; Golden, 1986; Geddes and Heywood, 2003), while bonuses aretypically associated with lower turnover (Hashimoto, 1979; Blakemore et al., 1987;Guthrie, 2000). In particular, Hashimoto (1979) provides evidence that flexible bonus

JES39,6

656

payments enhance investments in on-the-job training, which is known to reduceturnover. Blakemore et al. (1987) observe lower turnover among those paid a two-partcompensation scheme composed of fixed pay plus flexible bonuses. Additionally,Guthrie (2000), drawing on organizational economics literature, shows that firms usingskill-based pay systems such as bonuses improve employee retention.

Prior evidence reveals that PRP based on firm profitability is negatively associatedwith turnover. Klein and Hall (1988) and Saneyoshi (2001) observe that participation inEmployee Stock Ownership Plans (ESOP) reduce turnover. Saneyoshi observes thatESOP workers are 71 percent more likely to stay with their current employer than thosewithout any ESOP. Azfar and Danninger (2001) find that profit-sharing plans arenegatively associated with quits and layoffs, using a subset of the data analyzed below.Furthermore, they observe increased skill accumulation among profit sharing workerssince they receive more intensive on-the-job training, leading to higher future wagegrowth. Analyzing UK data, Green and Heywood (2008) find that profit sharing increasesthe incidence of on-the-job training which is known to reduce turnover (Royalty, 1996).Others also identify a negative relation between various PRP schemes and turnover.Chelius and Smith (1990) provide suggestive evidence that profit sharing reduces theincidence of layoffs in the face of decreased product demand. Furthermore, Scoppa (2003)theorizes that firms employ PRP when the costs associated with turnover are high andfixed pay when the costs associated with turnover are low. Dale-Olsen (2004) findsevidence in Norway of a negative association between turnover costs and the use ofalternative forms of remuneration such as fringe benefits. He finds that fringe benefitshave a stronger negative impact on turnover than what would be indicated by the fringebenefits monetary value, supporting the idea that fringe benefits have an independentinfluence on turnover. In sum, literature reveals a strong negative relationship betweenaggregate measures of PRP and turnover, lower turnover among those receiving stockoptions, profit sharing and bonuses, and higher turnover among those receivingpiece rates and commissions. The analysis below will attempt to add to this literature byfurther exploring which PRP schemes most significantly influence turnover.

III. Data and the determinants of labor market turnoverThe NLSY79 contains information on both worker turnover and PRP as well otherinformation on workers job market experiences. Analysis will focus on a nationallyrepresentative sample of 6,111 men and women who were between the ages of14-22 years old as of their initial interview in 1979. These individuals were interviewedannually up to 1994 and biennially thereafter. The NLSY79 includes information onrespondents labor force experiences and attachment, demographics, earnings andinvestments in both education and training, as well as information on job satisfaction.Only six waves contain PRP data: 1988, 1989, and 1990, as well as 1996, 1998, and 2000.Although data provides information on up to five jobs per respondent, focus will be oneach respondents main CPS job. Weights are not employed when reporting means orperforming estimations since elimination of the supplemental over-samples of poor,Hispanic, black and military personnel result in a nationally representativecross-section. However, sample size decreases due to attrition, missing data and theelimination of those who report being self-employed or in the active military. Those inthe active military are eliminated since turnover decisions are likely much differentthan in the civilian sector, while the self-employed are eliminated since job turnover

Performance payand employee

turnover

657

likely has a much different meaning. Lastly, dropping observations with missing dataleaves an unbalanced panel of 17,114 observations across 4,925 individuals.

The NLSY asks respondents the following question: The earnings on some jobs arebased all or in part on how a person performs the job. On this card are some examples ofearnings that are based on job performance. Please tell me if any of the earnings on yourjob are/were based on any of these types of compensation. for the years 1988-1990 and,On this card are some examples of earnings that are based on job performance.Please tell me if any of the earnings on your job are/were based on any of these typesof compensation. for the years 1996-2000. Possible responses were piece rates,commissions, bonuses, stock options, tips and other. Furthermore, the survey collectedinformation on participation in profit sharing and pension plans. PRP workers are thosewho receive any earnings in the form of piece rates, commissions, bonuses, stockoptions, tips, profit sharing and other forms of PRP.

Although respondents were asked about participation in various payment schemes,no information was collected about how much of their earnings were based on theseschemes. Earnings could be entirely based on PRP or only marginally affected by PRP.Another problematic issue is what constitutes other forms of PRP. Fully 2.4 percentof the sample employed in analysis report receiving other forms of PRP. Although itis impossible to interpret what these other types of PRP entail, they will be includedwhen constructing the aggregated PRP variable and also included as a separate PRPscheme in the disaggregated estimations. Given these restrictions, it is impossibledetermine to what extent earnings are PRP based. The lack of more detailed PRPinformation limits the ability to effectively assess the relation between job turnoverand PRP. This becomes even more problematic if some jobs entail combinations of PRPschemes. Restrictions aside, one can arguably use the incidence of PRP to examine therelationship between PRP and other labor market outcomes.

The survey also asks respondents why their job terminated[1]. The survey asksrespondents which of the reasons on this card best describes why you happened toleave this job? Listed reasons include layoffs, plant closing, end of temporary/seasonaljob, firing, program ending, quitting for family reasons, quitting to find or take anotherjob, and quitting for other reasons. From these questions, the aggregate indicator ofturnover equals one if the respondent reported a quit or layoff for any reason, zerootherwise. Analogously, indicator variables for quits and layoffs are created todistinguish between reasons for turnover. Layoffs are equal to one if the respondentreports a layoffs, plant closing, end of temporary/seasonal job, firing, or program ending.

Many other individual and firm level characteristics are available in the NLSY79and are summarized by payment scheme in Table I. Since PRP is known to attractmore educated and able workers, measures of human capital and ability are includedas controls. Education is measured as the highest grade completed as of May 1 of eachsurvey year. Ability is proxied by each individuals percentile score on the ArmedForces Qualification Test. As Table I displays, educational levels and ability areslightly higher among PRP workers, except among those receiving piece rates and tips.

Given the close relationship between earnings, satisfaction and turnover, analysiswill include an earnings measure created by taking the logarithm of the individuals realwage. The real wage is obtained by dividing the individuals nominal hourly rate of payby the respondents regions Consumer Price Index All Urban Consumers using1982-1984 as the base period. Furthermore, because on-the-job training and turnover are

JES39,6

658

Var

iab

leP

RP

Non

-PR

PF

-rat

ioP

iece

rate

sC

omm

issi

ons

Tip

sB

onu

ses

Sto

ckop

tsP

rofi

tsh

arin

gM

ult

iple

PR

Psc

hem

es

Tu

rnov

er6.

59%

10.5

1%81

.27

11.0

3%7.

82%

14.7

1%5.

28%

1.91

%5.

13%

4.63

%Q

uit

4.46

%6.

69%

38.9

05.

13%

5.06

%10

.87%

3.90

%1.

67%

3.38

%3.

37%

Lay

off

2.12

%3.

82%

40.7

15.

90%

2.76

%3.

84%

1.38

%0.

24%

1.74

%1.

26%

Lik

ejo

bv

ery

mu

ch45

.78%

44.8

1%1.

5834

.87%

52.5

3%39

.23%

50.4

5%55

.26%

45.5

4%50

.91%

Lik

ejo

bfa

irly

wel

l46

.31%

45.2

1%2.

0552

.56%

40.4

6%50

.32%

43.9

9%40

.19%

46.4

3%43

.01%

Dis

lik

ejo

bso

mew

hat

6.10

%7.

11%

6.92

9.49

%5.

86%

8.74

%4.

22%

4.07

%6.

02%

4.68

%D

isli

ke

job

ver

ym

uch

1.82

%2.

87%

19.8

13.

08%

1.15

%1.

71%

1.34

%0.

48%

2.01

%1.

41%

Ten

ure

(in

wee

ks)

255.

3123

6.79

24.3

623

3.50

199.

9013

7.70

262.

0031

1.64

274.

4626

4.64

AF

QT

50.4

246

.53

79.4

336

.44

55.2

240

.36

56.1

863

.01

50.2

954

.90

Ed

uca

tion

13.4

413

.28

20.0

212

.01

13.7

912

.43

13.9

714

.69

13.4

613

.88

Rea

lW

age

$9.6

4$8

.00

294.

71$8

.31

$10.

73$5

.88

$11.

12$1

5.51

$9.7

5$1

1.48

Log

Rea

lW

age

$2.0

9$1

.94

350.

97$1

.90

$2.1

6$1

.58

$2.2

3$2

.56

$2.1

3$2

.25

On

-th

e-Jo

bT

rain

ing

22.5

8%18

.38%

46.0

314

.36%

26.2

1%12

.37%

27.2

5%27

.99%

22.9

1%27

.92%

Fem

ale

48.5

4%51

.52%

15.0

344

.62%

37.2

4%65

.88%

41.4

3%33

.49%

50.1

2%40

.74%

Non

wh

ite

18.2

3%19

.78%

6.55

21.2

8%11

.15%

20.2

6%16

.94%

12.2

0%18

.68%

16.1

5%B

lack

11.6

1%12

.87%

6.19

13.0

8%5.

86%

10.4

5%10

.15%

6.94

%12

.34%

9.46

%H

isp

anic

6.63

%6.

92%

0.56

8.21

%5.

29%

9.81

%6.

78%

5.26

%6.

34%

6.69

%A

ge

32.2

932

.66

22.2

532

.06

31.9

931

.42

32.4

212

34.6

032

.17

32.4

4U

nm

arri

edw

/oC

hil

dre

n28

.76%

28.9

1%0.

0427

.69%

32.3

0%42

.22%

30.0

6%28

.95%

26.5

2%28

.87%

Un

mar

ried

w/C

hil

dre

n,

6y

ears

4.20

%5.

12%

8.06

5.64

%3.

56%

6.61

%3.

05%

2.63

%4.

40%

3.67

%U

nm

arri

edw

/Ch

ild

ren.

6y

ears

6.93

%7.

74%

4.04

9.74

%4.

02%

9.59

%5.

08%

4.55

%7.

01%

4.88

%M

arri

edw

/oC

hil

dre

n15

.19%

13.4

8%10

.13

10.0

0%17

.82%

10.6

6%16

.37%

11.2

4%15

.68%

15.5

9%M

arri

edw

/Ch

ild

ren,

6y

ears

27.2

4%25

.67%

5.36

23.5

9%27

.59%

15.1

4%29

.65%

33.2

5%28

.48%

30.3

8%M

arri

edw

/Ch

ild

ren.

6y

ears

17.6

8%19

.07%

5.47

23.3

3%14

.71%

15.7

8%15

.80%

19.3

8%17

.91%

16.6

0%

(continued

)

Table I.Descriptive statistics

Performance payand employee

turnover

659

Var

iab

leP

RP

Non

-PR

PF

-rat

ioP

iece

rate

sC

omm

issi

ons

Tip

sB

onu

ses

Sto

ckop

tsP

rofi

tsh

arin

gM

ult

iple

PR

Psc

hem

es

Occupation

Pro

fess

ion

al16

.87%

23.8

0%12

4.19

3.59

%6.

67%

1.71

%19

.90%

22.0

1%17

.79%

15.6

4%M

anag

eria

l20

.33%

11.2

0%22

7.24

3.85

%24

.14%

8.32

%30

.50%

40.4

3%20

.38%

28.7

7%S

ales

7.44

%2.

14%

281.

711.

79%

35.0

6%0.

64%

8.16

%6.

22%

6.58

%12

.17%

Cle

rica

l17

.34%

18.9

0%6.

829.

49%

7.24

%2.

99%

13.2

4%9.

09%

21.1

5%11

.87%

Cra

ft10

.20%

12.6

9%25

.52

21.5

4%8.

97%

1.49

%8.

37%

6.70

%10

.19%

8.10

%O

per

ativ

e10

.07%

8.11

%19

.84

40.5

1%2.

18%

0.85

%6.

54%

7.42

%10

.86%

8.45

%T

ran

spor

tO

per

ativ

e3.

52%

3.77

%0.

735.

90%

5.29

%4.

90%

3.13

%2.

39%

3.24

%3.

67%

Lab

orer

3.57

%5.

47%

34.3

48.

97%

1.15

%2.

99%

2.92

%1.

20%

3.46

%2.

46%

Far

m0.

19%

0.09

%2.

660.

26%

0.00

%0.

00%

0.28

%0.

48%

0.18

%0.

25%

Far

mL

abor

er0.

39%

0.72

%8.

311.

03%

0.00

%0.

21%

0.77

%0.

00%

0.18

%0.

30%

Ser

vic

e9.

93%

12.7

5%32

.19

3.08

%8.

85%

75.9

1%5.

93%

3.83

%5.

79%

8.00

%P

riv

ate

Hou

seh

old

0.01

%0.

31%

20.9

80.

00%

0.00

%0.

00%

0.04

%0.

00%

0.00

%0.

00%

Industry

Ag

ricu

ltu

re1.

57%

2.08

%6.

022.

82%

0.69

%0.

64%

2.27

%0.

96%

1.34

%1.

56%

Min

ing

0.58

%0.

89%

5.16

0.00

%0.

00%

0.21

%0.

65%

0.24

%0.

69%

0.40

%C

onst

ruct

ion

4.24

%8.

54%

126.

5510

.77%

3.56

%0.

64%

5.24

%1.

67%

3.46

%3.

62%

Man

ufa

ctu

rin

g25

.48%

15.8

7%24

6.18

45.1

3%12

.07%

0.85

%21

.28%

30.3

8%29

.78%

23.4

9%T

ran

spor

tati

on7.

98%

6.41

%15

.87

6.92

%7.

59%

2.77

%8.

08%

12.6

8%8.

71%

9.26

%W

hol

esal

e23

.01%

13.8

0%24

7.37

14.6

2%32

.53%

63.9

7%21

.36%

21.5

3%21

.37%

25.9

1%F

inan

ce10

.29%

4.38

%22

9.78

2.56

%16

.90%

1.28

%11

.37%

12.4

4%10

.78%

12.0

7%B

usi

nes

sS

erv

ices

7.76

%6.

38%

12.3

88.

72%

12.3

0%2.

77%

9.99

%11

.48%

6.87

%9.

41%

Per

son

alS

erv

ices

3.20

%2.

53%

7.02

1.54

%9.

08%

22.3

9%2.

36%

1.44

%1.

56%

4.02

%E

nte

rtai

nm

ent

1.09

%1.

36%

2.47

1.28

%1.

72%

4.05

%1.

14%

0.48

%0.

75%

1.11

%P

rofe

ssio

nal

Ser

vic

es12

.52%

29.3

5%72

6.59

3.08

%3.

10%

0.43

%13

.48%

5.98

%13

.11%

8.20

%P

ub

lic

Ad

min

istr

atio

n1.

74%

8.17

%35

1.94

2.56

%0.

11%

0.00

%2.

11%

0.00

%1.

05%

0.45

%H

ours

per

Wee

k42

.69

40.7

617

8.05

42.7

844

.19

36.8

844

.76

46.5

342

.71

44.6

4U

nio

n9.

12%

18.0

5%28

1.63

19.4

9%1.

95%

3.62

%6.

17%

5.50

%9.

68%

5.68

%S

mal

lE

mp

loy

er(l

ess

than

ten

emp

loy

ees)

19.2

7%26

.38%

120.

2122

.56%

39.8

9%31

.56%

22.7

5%12

.20%

14.2

0%22

.43%

(continued

)

Table I.

JES39,6

660

Var

iab

leP

RP

Non

-PR

PF

-rat

ioP

iece

rate

sC

omm

issi

ons

Tip

sB

onu

ses

Sto

ckop

tsP

rofi

tsh

arin

gM

ult

iple

PR

Psc

hem

es

Med

ium

Em

plo

yer

(bet

wee

nte

nan

d49

emp

loy

ees)

23.4

8%25

.07%

5.81

17.6

9%35

.06%

37.3

1%25

.18%

17.4

6%20

.68%

23.4

4%L

arg

eE

mp

loy

er(b

etw

een

50an

d20

0em

plo

yee

s)25

.94%

24.7

3%3.

2621

.53%

17.5

9%21

.96%

25.0

6%25

.84%

28.4

0%24

.65%

Ver

yL

arg

eE

mp

loy

er(a

bov

e20

0em

plo

yee

s)31

.31%

23.8

2%12

0.65

38.2

0%7.

47%

9.17

%27

.01%

44.4

9%36

.71%

29.4

8%P

rom

otio

n25

.78%

17.0

5%19

6.58

13.5

9%20

.80%

17.7

0%32

.29%

34.6

9%26

.90%

30.5

3%F

rin

ge

Ben

efits

96.8

7%88

.48%

420.

8889

.23%

94.4

8%84

.65%

97.8

5%10

0.00

%99

.70%

98.5

9%E

xp

erie

nce

(in

wee

ks)

636.

8461

6.27

26.7

760

1.11

628.

0953

0.78

655.

5178

0.63

643.

0166

4.98

No.

ofJo

bs

5.58

5.79

21.9

45.

586.

216.

745.

625.

915.

365.

71U

nem

plo

ym

ent

Rat

e(c

ateg

oric

al)

2.29

2.37

45.8

22.

362.

292.

412.

262.

112.

262.

24U

rban

75.5

2%71

.67%

32.0

960

.26%

80.8

0%80

.38%

78.1

1%79

.19%

75.5

9%77

.52%

Yea

r

1988

14.0

1%12

.44%

9.15

16.1

5%13

.79%

13.0

1%13

.69%

7.42

%14

.79%

13.3

8%Y

ear

1989

17.6

1%17

.14%

0.64

19.2

3%19

.89%

23.2

4%16

.37%

9.57

%17

.75%

16.6

0%Y

ear

1990

18.4

1%17

.25%

3.88

20.0

0%21

.84%

22.8

1%18

.12%

8.85

%18

.48%

17.6

6%Y

ear

1996

18.3

1%19

.26%

2.48

15.9

0%17

.47%

16.6

3%17

.55%

18.6

6%17

.87%

18.7

6%Y

ear

1998

16.7

1%17

.51%

1.91

16.6

7%14

.83%

13.0

1%17

.51%

28.7

1%16

.23%

17.4

0%Y

ear

2000

14.9

5%16

.40%

6.67

12.0

5%12

.18%

11.3

0%16

.77%

26.7

9%14

.89%

16.2

0%N

o.of

obs.

7,53

09,

584

390

870

469

2,46

241

84,

936

1,98

8

Table I.

Performance payand employee

turnover

661

strongly negatively correlated (Royalty, 1996), an indicator in the estimations denotingwhether the respondent received any form of on-the-job training since their lastinterview is included.

Female workers often experience higher turnover than males for reasons such aschildbirth and rearing (Golden 1986, 1990). Within the sample, 16 percent of those whoreport leaving an employer quit for pregnancy or family reasons. Hence it seemsessential to control for the influence gender and family status have on turnover.Besides an indicator for gender, six other indicator variables are created to represent anindividuals current family situation. These six indicators include:

(1) unmarried without children;

(2) unmarried with children under six years of age;

(3) unmarried with children six years of age or older;

(4) married without children;

(5) married with children under six years of age; and

(6) married with children six years of age or over.

Furthermore, since larger establishments are more likely to make use of PRP schemes(Brown, 1990) and have lower turnover (Evan and MacPhearson, 1996), four firm sizeindicators are included. A firm is categorized as small if it employs less thanten workers, medium if it employs between ten and 49 employees, large if it employsbetween 50 and 200 employees, and very large if it employs more than 200 workers[2].

Respondents are also asked:

How (do/did) you feel about (the job you have now/your most recent job)? (Do/Did) you like itvery much, like it fairly well, dislike it somewhat, or dislike it very much?

Since job satisfaction is known to influence turnover decisions, four different indicatorsto control for global job satisfaction are created from the four possible responses above.Within the sample, 7 percent of those who report liking their job very much experienceturnover while those who report disliking their job very much have a turnover rate of28 percent. Also, quits are more numerous than layoffs among workers who reportdisliking their job very much.

Firms with internal labor markets typically experience lower turnover. This isbecause an alternative external wage offer is typically lower than the internal wage offersince the worker has accumulated firm specific knowledge that would not affectpay levels elsewhere. Also, fringe benefits such as insurance or pensions typically lowerturnover. Consequently, to proxy for the presence of an internal labor market, I create anindicator that equals one if a position change within an employer was a promotion, zerootherwise. The fringe benefit indicator takes a value equal to one if the individualreceives either employer provided insurance (medical, dental, and life), maternity leave,firm sponsored training and education, or employer-provided child care, and zero ifthey receive none of these benefits.

Also, the nature of production critically determines whether various PRP schemes arepracticable. Much of the prior literature points to the fact that PRP jobs are not availablein every occupation or industry. Elliott and Murphy (1986) observed that workers may nothave the option to choose whether to obtain a PRP job. For example, they find that manualworkers are much more likely to be paid based on some performance measure in

JES39,6

662

comparison to non-manual workers. Similarly, Parent (1999) observes that PRP in theform of piece rates are concentrated in particular occupations such as sales and operativeoccupations, while bonuses tend to be more evenly spread across occupations. This is dueto the nature of the job and the difficulty involved in structuring the PRP scheme insituations where effort or output are extremely difficult or impossible to measure.

Consequently, to proxy for the nature of production, I include occupational/industrialindicators at the one-digit level. Estimations also include an indicator for unionaffiliation, as well as a count variable for the unemployment rate for the workers currentresidence[3]. Additionally, since it is well known that older, more experienced workersare much less likely to experience job turnover, we include a measure of the respondentsage. Lastly, estimations also include the respondents tenure in weeks with a particularemployer, average hours per week worked, urban status, number of jobs held, andindicators for each year of the panel.

Table I displays summary statistics combining all six years of data in which PRPquestions were administered. Absent controls, PRP workers experience statisticallysignificant lower mean levels of job turnover in the form of quits and layoffs than workerswho do not receive any PRP. Among PRP workers, an average of 6.6 percent report jobturnover over the period considered, while the average turnover rate of those who do notreceive any PRP is 10.5 percent. Furthermore, F-tests for the difference in the means arestatistically significant. PRP workers experience on average 1.7 percent fewer layoffs and2.2 percent fewer quits than non-PRP workers. Differentiating between PRP schemesreveals that layoffs and quits are much lower for workers receiving stock options, profitsharing, bonuses, commissions, and those covered by multiple PRP schemes. Comparingmeans, piece rate workers are no different in terms of turnover rates than workers notreceiving PRP, while tip workers have higher rates of turnover than non-PRP workers.

The statistics in Table I are roughly consistent with those reported by Bonars andMoores (1995) analyze of PRP within the 1988-1990 waves of the NLSY79. Confirmingtheir observations, PRP workers receive higher wages and have higher levels ofeducation in comparison to non-PRP workers. Additionally, piece rate, commission andtip workers have shorter average tenures, while all other PRP workers have longeraverage tenures. However, contrary to their findings, Table I reveals that PRP workershave longer tenures than non-PRP workers. This is likely due to their exclusion ofgroup-based PRP plans from their measure of performance pay.

Furthermore, summary statistics reveal that PRP workers are more satisfied,receive more promotions, are more likely to be employed in larger establishments, andare more likely to receive fringe benefits than non-PRP workers. Turnover rates arehighest among piece rate workers and lowest among workers receiving stock options,followed by multiple schemes, profit sharing and bonuses. Among those who receiveany PRP, 26.4 percent report receiving more than one type of PRP. Among those whoreceive multiple types of PRP, the most common forms are profit sharing and bonuseswith many fewer workers receiving tips piece rates or multiple forms of PRP. Withinthe sample 80 percent receive profit sharing and some other form of PRP while72 percent receive bonuses and some other form of PRP. Among the various forms ofPRP, stock options and bonuses have the highest correlation coefficient at 0.25,followed by stock options and bonuses at 0.14, and profit sharing and bonuses at 0.12.

As one would expect, the vast majority of piece rate workers are found in craft andoperative occupations. Commissions are most prevalent in sales occupations. Tips are

Performance payand employee

turnover

663

concentrated in service occupations while bonuses and stock options are prevalentamong managerial occupations. Profit sharing appears to be spread evenly overprofessional, managerial, and clerical workers. Workers receiving multiple forms of PRPappear to be most prevalent among professional, managerial, sales and clericaloccupations. Furthermore, summary statistics reveal women are less likely to be paidPRP and are less likely to receive bonuses, commissions and stock options as wellreceiving multiple forms of PRP in comparison to men. With regard to firm size, piecerates are most prevalent among the largest employers, while tips and commissions aremore common among smaller firms. Also, bonuses are more predominant among thelargest employers. Additionally, summary statistics show that unionized employees aremost likely to be paid either piece rates or profit sharing. Over time, the incidence of PRPamong this cohort increased from 14 percent in 1988 to a high of 18 percent in 1990, thenfell back down to 15 percent in 2000. Within the different PRP schemes, the percentreceiving stock options increased from 7 percent in 1988 to 29 percent in 1996 beforetapering off slightly in 2000.

IV. Empirical resultsAnalysis begins with a parsimonious specification of employee turnover andprogressively adds various sets of controls known to influence turnover to observe howtheir inclusion influences the PRP coefficients. The base specification includes controls forthe presence of PRP, plus basic demographics such as race, gender, human capital andability, as well as industry and occupation. Specification 2 adds non-compensationvariables known to influence turnover including employer size, usual hours of work perweek, tenure and its square, union affiliation (whether member or covered), age, urbanstatus, number of jobs ever held, the local unemployment rate and family status indicators.Specification 3 adds compensation variables such as the log hourly real wage, as well asindicators for fringe benefits and promotions. Lastly, measures of global job satisfactionare included in Specification 4. Adding additional controls reduces size and significance ofthe PRP coefficient but increases the goodness of fit. Results of probit estimations onturnover, quits and layoffs are presented in Tables II-IV[4].

Table II presents probit results where the dependent variable is the turnoverindicator and the independent variable of interest is the aggregate measure of PRP. Theparsimonious specification presented in Table II, Specification 1, reveals that PRPworkers are 3.9 percent less likely to experience turnover than non-PRP workers. Most ofthe covariates included in this specification appear as expected. Females and nonwhiteshave greater turnover, while those who accumulate more human capital in the form ofeducation and on-the-job training experience less turnover. Workers in professional,managerial, clerical and craft occupations experience lower turnover, as do thoseworking in manufacturing, transportation, finance, professional services and publicadministration industries.

Table II, Specification 2 adds non-compensation related variables known to influenceturnover. Adding these non-compensation variables reduces the PRP coefficient from3.9 to 2.7 percent. Once non-compensation variables are controlled for, the size andsignificance of several variables in Specification 1 are reduced. Inclusion of firm sizereveals that workers at smaller employers have higher turnover rates, while beingcovered by a collective bargaining contract reduces turnover. As expected, higherunemployment rates are associated with higher turnover. Unmarried and married

JES39,6

664

(1) (2) (3) (4)

PRPa 2 0.039 (9.20) * * 20.027 (6.81) * * 20.015 (3.75) * * 20.013 (3.31) * *

Femalea 0.051 (10.80) * * 0.050 (10.76) * * 0.042 (9.22) * * 0.042 (9.30) * *

Nonwhitea 0.012 (2.14) * 0.011 (2.14) * 0.008 (1.57) 0.009 (1.75)OJTa 20.028 (5.39) * * 20.027 (5.63) * * 20.021 (4.43) * * 20.020 (4.29) * *

AFQTa 20.000 (4.51) * * 20.000 (2.63) * * 20.000 (0.60) 20.000 (0.69)Education 20.003 (2.51) * 20.003 (3.16) * * 20.002 (1.88) 20.002 (2.25) *

Professionala 20.029 (4.02) * * 20.019 (2.80) * * 20.007 (0.95) 20.004 (0.55)Managera 20.036 (5.21) * * 20.024 (3.58) * * 20.006 (0.79) 20.005 (0.70)Salesa 20.015 (1.39) 20.013 (1.32) 20.002 (0.20) 20.002 (0.17)Clerica 20.023 (3.52) * * 20.016 (2.61) * * 20.009 (1.39) 20.007 (1.20)Crafta 20.022 (2.70) * * 20.011 (1.30) 20.001 (0.14) 0.002 (0.21)Operativea 0.007 (0.76) 0.017 (1.86) 0.017 (1.98) * 0.016 (1.90)Laborera 0.021 (1.88) 0.024 (2.28) * 0.020 (2.01) * 0.020 (1.99) *

Transport operativea 20.006 (0.52) 20.007 (0.70) 20.007 (0.65) 20.005 (0.52)Farm laborera 0.019 (0.65) 0.018 (0.64) 0.001 (0.05) 0.006 (0.23)Private householda 20.012 (0.35) 20.015 (0.51) 20.027 (1.05) 20.023 (0.92)Aga 20.022 (1.45) 20.021 (1.58) 20.020 (1.51) 20.019 (1.50)Mininga 20.021 (1.04) 0.002 (0.11) 0.017 (0.78) 0.017 (0.78)Constructiona 20.007 (0.83) 20.013 (1.58) 20.003 (0.30) 20.003 (0.31)Mfga 20.023 (3.55) * * 20.006 (0.97) 0.001 (0.22) 0.001 (0.23)Transportationa 20.032 (4.00) * * 20.018 (2.28) * 20.010 (1.21) 20.010 (1.20)Financea 20.018 (2.19) * 20.012 (1.49) 20.002 (0.21) 20.002 (0.23)Businessa 0.009 (1.05) 0.001 (0.08) 0.005 (0.62) 0.002 (0.25)Personal serv.a 0.012 (1.02) 0.007 (0.68) 0.005 (0.54) 0.005 (0.55)Entertainmenta 20.004 (0.27) 20.000 (0.01) 20.003 (0.18) 0.002 (0.12)Prof. servicea 20.039 (6.27) * * 20.025 (4.28) * * 20.018 (3.06) * * 20.015 (2.57) *

Pub admn.a 20.059 (7.26) * * 20.039 (4.36) * * 20.032 (3.48) * * 20.029 (3.27) * *

1989a 0.039 (4.69) * * 0.030 (3.94) * * 0.017 (2.40) * 0.020 (2.80) * *

1990a 0.013 (1.60) 0.009 (1.19) 20.000 (0.06) 0.002 (0.30)1996a 0.023 (2.92) * * 0.019 (1.81) 0.010 (1.05) 0.015 (1.52)1998a 0.026 (3.12) * * 0.031 (2.44) * 0.023 (1.96) 0.029 (2.39) *

2000a 0.020 (2.39) * 0.032 (2.18) * 0.025 (1.81) 0.034 (2.39) *

Small empa 0.021 (3.57) * * 0.005 (0.84) 0.009 (1.69)Medium empa 0.016 (2.83) * * 0.009 (1.67) 0.011 (2.09) *

Large empa 0.014 (2.38) * 0.011 (2.08) * 0.013 (2.41) *

Hrs work per week 0.000 (0.10) 0.000 (0.75) 0.000 (0.67)Tenure 20.000 (13.26) * * 20.000 (10.35) * * 20.000 (10.51) * *

Tenure squared 0.000 (8.36) * * 0.000 (6.29) * * 0.000 (6.18) * *

Uniona 20.011 (1.97) * 20.001 (0.20) 20.001 (0.11)Age 0.001 (1.23) 0.001 (0.67) 0.000 (0.56)Urbana 0.009 (2.04) * 0.011 (2.72) * * 0.009 (2.38) *

Num. of jobs everheld 0.002 (2.26) * 0.002 (2.53) * 0.002 (2.11) *

Unemployment rate 0.008 (3.59) * * 0.007 (3.39) * * 0.008 (3.49) * *

Unmarried w/chd.,6 yrsa 0.020 (2.36) * 0.018 (2.12) * 0.018 (2.18) *

Unmarried w/chd..6 yrsa 20.017 (2.56) * 20.016 (2.44) * 20.016 (2.59) * *

Married w/o chd. a 20.021 (3.60) * * 20.017 (3.04) * * 20.017 (2.97) * *

(continued )

Table II.Determinants of

turnover probitanalysis marginal

effects shown

Performance payand employee

turnover

665

workers with young children are more likely to experience turnover, while parents witholder children and those without children are less likely to experience turnover.

Adding the log hourly real wage, fringe benefits, and promotions further reduces thePRP coefficient reported in Table II, Specification 3, from 2.7 to 1.5 percent. The largereduction in the PRP coefficient between Specifications 2 and 3 is likely due to the largeimpact compensation has on job satisfaction and turnover. Adding the compensationvariables turns many of the previously significant covariates insignificant. For example,coefficients for professional and managerial occupation fall to insignificance between2 and 3. A likely explanation is that professional and managerial professions payrelatively more than other occupations. Once controls for compensation are added,the coefficients for managerial and professional occupations become insignificant.

(1) (2) (3) (4)

Married w/chd., 6 yrsa 0.013 (2.60) * * 0.016 (3.20) * * 0.018 (3.67) * *

Married w/chd.. 6 yrsa 20.022 (3.89) * * 20.018 (3.34) * * 20.017 (3.20) * *

Log real wage 20.034 (8.12) * * 20.031 (7.44) * *

Fringe benefitsa 20.076 (10.35) * * 20.067 (9.45) * *

Promotiona 20.036 (7.88) * * 20.032 (7.06) * *

Like job very mucha 20.007 (1.83)Dislike jobsomewhata 0.062 (7.88) * *

Dislike job verymucha 0.125 (9.25) * *

Log likelihood 24,722.817 24,463.95 24,333.17 24,255.37Pseudo R 2 0.072 0.123 0.149 0.164Observations 17,114 17,114 17,114 17,114

Notes: Significant at: *5 and * *1 percent; adF/dx is for discrete change of dummy variable from 0 to 1;z and P . jzj correspond to the test of the underlying coefficient being 0; absolute value of z-statisticsin parentheses

Table II.

(1) (2) (3) (4)

QuitsPRPa 20.021 (6.38) * * 20.014 (4.53) * * 20.007 (2.12) * 20.005 (1.69)Log likelihood 23,498.14 23,342.45 23,265.38 23,194.54Pseudo R 2 0.066 0.108 0.128 0.147Observations 17,114 17,114 17,114 17,114LayoffsPRPa 20.015 (6.32) * * 20.010 (4.82) * * 20.007 (3.18) * * 20.006 (3.05) * *

Log likelihood 22,192.83 22,080.94 22,038.99 22,031.31Pseudo R 2 0.067 0.114 0.132 0.135Observations 17,114 17,114 17,114 17,114

Notes: Significant at: *5 and * *1 percent; adF/dx is for discrete change of dummy variable from 0 to 1;z and P . jzj correspond to the test of the underlying coefficient being 0; absolute value of z-statisticsin parentheses; regressions (1)-(4) include the same set of controls displayed in Table II (1)-(4)

Table III.Determinants of quitsand layoffs probitanalysis marginaleffects shown

JES39,6

666

The same types of interactions also cause other coefficients to become insignificant dueto the high correlation between pay and other aspects of job quality. However, the PRPcoefficient remains statistically significant. As shown in Table II Specification 3, a higherlog real wage is associated with lower turnover, while the presence of other fringebenefits or a recent promotion reduces the probability of turnover.

Lastly, adding job satisfaction controls causes the PRP coefficient to fall from1.5 to 1.3 percent, as displayed in Table II, Specification 4. Those that state that they eitherdislike their job somewhat or very much have higher rates of turnover. Consequently,initial estimations reveal that PRP generally reduces turnover, despite the inclusion ofmany known major determinants of turnover. Moreover, female workers remain morelikely to experience turnover as do those at larger employers and those caring for youngchildren. Workers who receive on-the-job training, work in public administration, or donot have young children are less likely to experience turnover. Also, the compensatoryvariables remain negatively significant, but at slightly reduced magnitudes.

Does the aggregated PRP measure have a different impact on quits than layoffs?Table III separates turnover into quits and layoffs while including identical controls asthose presented in Table II[5]. As previously observed, adding additional sets ofvariables known to influence turnover reduces the PRP coefficient on both quitsand layoffs. Adding non-compensation and compensation-related variables in Table IIISpecifications 1-3 reduce the PRP coefficient on quits from 2.1 to 0.7 percent.

(1) (2) (3) (4)

TurnoverBonusesa 20.017 (2.76) * * 20.013 (2.32) * 20.005 (0.80) 20.003 (0.47)Stock optionsa 20.041 (2.57) * 20.037 (2.54) * 20.032 (2.22) * 20.032 (2.19) *

Other PRPa 20.009 (0.68) 20.009 (0.75) 20.004 (0.38) 20.002 (0.19)Profit sharinga 20.043 (9.67) * * 20.029 (6.70) * * 20.018 (4.06) * * 20.017 (3.92) * *

Log likelihood 24,700.63 24,451.31 24,326.30 24,248.40Pseudo R 2 0.077 0.126 0.150 0.166QuitsBonusesa 20.004 (0.89) 20.002 (0.54) 0.003 (0.73) 0.005 (1.07)Stock optionsa 20.022 (1.80) 20.020 (1.81) 20.017 (1.49) 20.016 (1.43)Other PRPa 0.008 (0.72) 0.007 (0.67) 0.009 (0.95) 0.011 (1.18)Profit sharinga 20.027 (7.53) * * 20.018 (5.30) * * 20.011 (3.29) * * 20.010 (3.16) * *

Log likelihood 23,482.35 23,333.23 23,259.11 23,187.83Pseudo R 2 0.070 0.110 0.129 0.149LayoffsBonusesa 20.011 (3.31) * * 20.009 (3.15) * * 20.007 (2.42) * 20.007 (2.34) *

Stock optionsa 20.019 (1.87) 20.015 (1.78) 20.014 (1.70) 20.014 (1.70)Other PRPa 20.014 (2.07) * 20.011 (2.02) * 20.010 (1.85) 20.010 (1.81)Profit sharinga 20.014 (5.50) * * 20.008 (3.59) * * 20.005 (2.09) * 20.005 (2.03) *

Log likelihood 22,180.52 22,071.27 22,031.73 22,024.26Pseudo R 2 0.072 0.118 0.135 0.138Observations 17,114 17,114 17,114 17,114

Notes: Significant at: *5 and * *1 percent; adF/dx is for discrete change of dummy variable from 0 to1; z and P . jzj correspond to the test of the underlying coefficient being 0; absolute value of zstatistics in parentheses; regressions (1)-(4) include the same set of controls displayed in Table II,(1)-(4); also, coefficients for piece rates, commissions and tips were omitted from the above table forbrevity since they were not significantly different from zero in any of the above estimations

Table IV.Turnover by PRP

type probitanalysis marginal

effects shown

Performance payand employee

turnover

667

However, adding job satisfaction variables reduces the PRP coefficient toinsignificance as shown in Specification 4 for quits. With regards to layoffs, thePRP coefficient falls from 1.5 percent in Specification 1 to 0.6 percent in Specification 4.Nonetheless, the PRP coefficient remains negatively significant even after inclusion ofmeasures of job satisfaction, as well as non-pecuniary and pecuniary aspects of the job.Consequently, estimations support the hypothesis that PRP participation results inreduced turnover, especially layoffs. However, the observed negative correlationbetween PRP and turnover does not provide any evidence concerning which specificPRP schemes matter most in influencing turnover decisions.

Including indicators for the various PRP schemes as well as an indicator forparticipation in multiple PRP schemes reveal that profit sharing tends to reduceturnover, especially quits. Table IV displays the effect of the different PRP schemes onturnover when all PRP schemes are added into the specification together[6]. Profitsharing and stock options significantly reduce turnover, even after inclusion of allavailable controls in Specification 4. Bonuses initially reduce turnover, but this fallsto insignificance with inclusion of the compensatory covariates. Moreover, piece rates,commissions, tips, and other forms of PRP have no significant impact on turnover.Also, the coefficient for multiple PRP plans is insignificant. Thus, except for profitsharing and stock options, there may be correlated omitted variables that, when included,cause the negative correlation between the other PRP schemes and turnover to vanish.

Separating turnover into quits and layoffs reveal a slightly different pattern wherestock options no longer appear to significantly influence quits. However, profit sharingremains negatively significant, suggesting it has an influence distinct from the otherPRP schemes. Additionally, bonuses have a slight negative impact on layoffs whichfails to vanish in the full specification. Moreover, including profit sharing alone inSpecification 4 without the other PRP variables reveals that profit sharing significantlyreduces the probability of both quits and layoffs, while stock options have a weaknegative impact on layoffs. All other PRP coefficients remain insignificantly differentfrom zero when added separately into the fullest specification. Consequently, the profitsharing result is unlikely to be due to multi-collinearity; however, the other resultscould be insignificant due to multi-collinearity. If one expected the profit sharing resultto arise out of multi-collinearity, one would expect the profit sharing result to besignificant when included with the other PRP schemes, but insignificant when includedindividually. Whether included individually or together with the other PRP schemes,the level of significance and size of profit sharing coefficient do not change.Consequently, the aggregated PRP measure appears to generally reduce turnover. Yet,breaking down the PRP measure by type, I find that PRP schemes dependent on afirms profitability such as profit sharing have a consistently significant negativeimpact on turnover and layoffs, and to some extent, quits.

Given the large amount of self-selection known to be associated with PRP, it is essentialto control for individual effects. This is done by estimating fixed and random-effectsprobit models including all available determinants of turnover. The fixed-effect probitmodels are estimated using LIMDEPs fixed-effects probit procedure[7].When estimating fixed-effects models, sample sizes shrink because workers with novariation in status are dropped from the sample. Additionally, coefficients for timeinvariant determinants such as gender, race, and AFQT score cannot be estimated.Furthermore, fixed-effects estimates increase measurement error which often makes it

JES39,6

668

difficult to obtain precise and statistically significant results. Initially, to check therobustness of earlier results, all prior estimations were repeated using a fixed-effectsprocedure. These estimations reveal a similar pattern as shown in Tables II-IV, except fora few minor deviations in magnitude and significance[8]. In the fixed-effects estimations,the size of the fixed-effects is not of great consequence since they are individualcomponents of turnover propensity not associated with the independent variables andhence will not be discussed.

Another difficulty inherent in the fixed-effects estimations is that these estimationsmay be picking up movement between the differing forms of PRP as opposed tocapturing the intended influence of movement from non-PRP to PRP jobs on turnoverproclivities. If moving from a non-PRP to a PRP job is associated with positive rents,one would expect to see lower turnover among PRP workers. However, if one is movingfrom one PRP scheme to another, there may be no rents and hence no difference inturnover associated with the PRP scheme. Consequently, the fixed-effects estimationsshould ideally only identify movements from a non-PRP job to a PRP job. Since thefixed-effects estimations involve a small and select sample, I do not expect there to bemuch movement across PRP schemes. Nevertheless, to further limit the possibility ofcapturing movement across PRP schemes, I created three categories: profit sharing,all other PRP besides profit sharing, and those who have neither, and then includedthem in the fixed-effects estimations. This should minimize the possibility of capturingmovement from one PRP scheme to another since there are now only two schemes,profit sharing and all other PRP schemes besides profit sharing.

As shown in Specifications 1-3 in Table V, controlling for individual effects revealsthat PRP workers remain less likely to experience job turnover and layoffs, but notsignificantly less likely to quit. Disaggregating PRP by scheme as described abovereveals that those receiving profit sharing are less likely to experience turnover in theform of both quits and layoffs (Table V, Specifications 4-6)[9]. The magnitude ofthe marginal effect in each case actually increases after controlling for fixed-effects inthe estimations. Although not shown in Table V, the marginal effect for the PRPcoefficient in Specification 3 is now larger in magnitude (20.006 vs20.046) than in the

(1) (4) (5) (6)Job

turnover(2)

Quits(3)

LayoffsJob turnover by

PRP schemeQuitsbyPRP

schemeLayoffs by

PRP scheme

PRP 20.240(3.44) * *

20.139(1.63)

20.357(3.16) * *

Profit sharing

20.376(4.31) * *

20.294(2.74) * *

20.432(3.09) * *

All PRP schemesbesides profitsharinga 20.092 (1.03) 0.017 (0.16)

20.268(21.80)

Number ofindividuals 1,484 1,080 574 1,484 1,080 574Log likelihood 21,928.71 21,341.59 2790.07 21,925.26 21,338.71 2789.65

Notes: Significant at: *5 and * *1 percent; absolute value of z-statistics in parentheses; all estimationincludes the full set of controls as indicated in Table II, column 4; athis category excludes those whosimultaneously receive profit sharing as well as alternate forms of PRP

Table V.Fixed-effects probit

analysis of job turnover,quits, and layoffs

Performance payand employee

turnover

669

estimations without the fixed-effects, as are the marginal effects on the profit sharingcoefficients for turnover (20.017 vs 20.027), quits (20.010 vs 20.014), and layoffs(20.005 vs 20.055) in Specifications 4-6, respectively. As shown in Specifications 4-6,all other forms of PRP besides profit sharing have no significant impact on turnover orquits. However, the coefficient for other PRP besides profit sharing does come close tohaving a significant impact on layoffs.

In addition to the fixed-effects estimations, random-effects estimations were alsoperformed to check the robustness of the findings. These estimations reveals that PRPworkers are less likely to experience turnover and layoffs, but not significantly lesslikely to quit[10]. Breaking down PRP into different schemes, then adding themtogether with all other determinants of turnover reveals that those receiving profitsharing as a form of compensation experience lower rates of turnover as well as layoffsand quits. In addition, those receiving stock options are significantly less likely toexperience turnover and layoffs, but not quits.

After including a multitude of factors known to influence turnover and attemptingto control for individual heterogeneity, the negative correlation between turnover andprofit based PRP remains. Therefore, profit sharing, and to some extent stock options,appears to have an independent influence on job turnover that other forms of PRP donot. However, other forms of PRP do not have a significant impact on turnover afteraccounting for measures intended to capture the pecuniary and non-pecuniary aspectsof the job while controlling for job satisfaction. Thus, for these other PRP types,evidence suggests that there are correlated omitted variables that, once included, causethe negative association between PRP and turnover to vanish.

V. ConclusionAlthough there is a substantial literature on the relationship between profit sharing andturnover, there is very little research on the influence of other types of PRP on job turnover.This paper explores how various PRP schemes influence employee turnover. It also testswhether profit sharing has a differential impact on turnover in comparison to other formsof PRP. Analysis of NLSY79 data suggests that PRPs impact on labor market turnovercritically depends on which type of PRP the worker receives. I find that workers receivingsome form of PRP, regardless of type, have lower rates of employee turnover than workerswho do not, within the cohorts analyzed. This finding provides more empirical support forthe sorting model of PRP than the classic agency model. Furthermore, workers receivingsome form of PRP are less likely to quit and are much less likely to be fired than those notreceiving any form of PRP. Confirming Azfar and Danninger (2001), Green and Heywood(2008) and Kruse et al. (2010), these results appear to be mainly driven by profit based payschemes, especially profit sharing[11]. Although one would expect the influence of otherPRP schemes to mimic that of profit sharing, evidence suggests otherwise. When oneisolates the impact of the separate PRP schemes on turnover, only those receiving profitsharing are significantly less likely to experience turnover.

Empirical evidence also supports the Weitzman theory (1985) that profitsharing enables marginal labor costs to vary with profitability, hence reducing theincentive to shed workers during downturns. Accordingly, further reliance on sharedcapitalism in the form of profit sharing based on firm profitability may help mitigatethe problem of unemployment as well as enhancing productivity and incentivizingemployees.

JES39,6

670

Notes

1. Light and McGarry (1998) note the difficulty in distinguishing voluntary and involuntary jobseparations within the NLSY. Specifically, they observe that a majority of workersexperiencing turnover report voluntary separations and a significant proportion (25 percent)report separations for other reasons. However, they do note that replacing their measuresof total separations with voluntary separations other than discharges or layoffs does notqualitatively change their inferences concerning the influence job mobility has on wagepaths. Although potentially arbitrary, I code those who report separations for reasons oflayoffs, plant closings, end of temp or seasonal jobs, firings, or program endings asinvoluntary separations (layoffs) while those who report separations as quits, regardless ofreasoning, as voluntary separations (quits).

2. Although the common practice is to take the log of employer size, results are not sensitive towhether estimations include the log of establishment size or four separate size indicatorsemployed here.

3. The NLS constructs the unemployment rate by using state and area labor force data fromthe May publication of Employment and Earnings for the month of March of each surveyyear. If a respondent resides in a metropolitan area, then the unemployment rate is theunemployment rate for that metropolitan area. Otherwise, the unemployment rate is thecomputed balance of state unemployment rates for the state in which the respondent resides.

4. Logit estimations with and without fixed effects reveal a similar pattern as that shown usingprobit analysis and are available from the author upon request.

5. The full estimation results, including all covariates, are available from the author uponrequest.

6. Estimations including each PRP scheme separately in the four specifications reveal a similarpattern as when they are added jointly. These estimations are available from the authorupon request.

7. See LIMDEP Manual Version 9 (2007) Panel Data Models for Binary Choice, p. E20-1-E20-17.

8. The full estimation results are available from the author upon request.

9. Fixed-effects estimations including all the PRP measures together in a single model were alsorun for turnover, quits and layoffs. These estimations reveal that profit sharing reduces jobturnover and quits, but not layoffs. However, the profit sharing coefficient would be statisticallysignificant at the 10 percent level, suggesting a weak negative correlation between profit sharingand layoffs. Accounting for individual heterogeneity causes the stock option coefficient tobecome statistically insignificant, although it retains sign and size. This result suggests that itmay be imprecisely estimated due to the small number of changes occurring within individuals.In a similar vein, fixed-effects estimations reveal a weak positive correlation between piece rateand layoffs and a weak negative correlation between commissions and quits.

10. These estimations are available from the author upon request.

11. For theoretical explanations and further background concerning the influence of profitsharing on productivity and employment stability, readers can reference Kruse (1993) andKruse et al. (2010).

References

Akerlof, G.A., Rose, A.K. and Yellen, J.L. (1988), Job switching and job satisfaction in the USlabor market, Brookings Papers on Economic Activity, Vol. 2, pp. 495-594.

Azfar, O. and Danninger, S. (2001), Profit sharing, employment stability, and wage growth,Industrial and Labor Relations Review, Vol. 54 No. 3, pp. 619-30.

Performance payand employee

turnover

671

Baker, G.P. (1992), Incentive contracts and performance measurement, Journal of PoliticalEconomy, Vol. 100 No. 3, pp. 598-614.

Bauer, T.K. (2004), High performance workplaces and job satisfaction: evidence from Europe,IZA Discussion Paper No. 1265.

Benabou, R. and Tirole, J. (2003), Intrinsic and extrinsic motivation, Review of EconomicStudies, Vol. 70 No. 3, pp. 489-520.

Blakemore, A.E., Low, S.A. and Ormiston, M.B. (1987), Employment bonuses and laborturnover, Journal of Labor Economics, Vol. 5 Nos 4, Part 2, pp. S124-35.

Booth, A.L. and Frank, J. (1999), Earnings, productivity, and performance-related pay, Journalof Labor Economics, Vol. 17 No. 3, pp. 447-63.

Bronars, S.G. and Moore, C. (1995), Incentive pay, information, and earnings: evidence formthe NLSY, NLS Discussion Paper 95-23, US Department of Labor, Bureau of LaborStatistics.

Brown, C. (1990), Firms choice of method of pay, Industrial and Labor Relations Review, Vol. 43No. 3, pp. 165-82.

Brown, M. (2001), Unequal pay, unequal response? Pay referents and their implications for paylevel satisfaction, Journal of Management Studies, Vol. 38 No. 6, pp. 879-96.

Brown, S. and Sessions, J.G. (2006), Some evidence on the relationship between performancerelated pay and the shape of the experience-earnings profile, Southern Economic Journal,Vol. 72 No. 3, pp. 660-76.

Chelius, J. and Smith, R.S. (1990), Profit sharing and employment stability, Industrial and LaborRelations Review, Vol. 43 No. 3, pp. 256S-73S.

Clark, A., Georgellis, Y. and Sanfey, P. (1998), Job satisfaction, wage changes, and quits:evidence from Germany, Research in Labor Economics, Vol. 17, pp. 95-121.

Cornelissen, T., Heywood, J.S. and Jirjahn, U. (2011), Performance pay, risk attitudes, and jobsatisfaction, Labour Economics, Vol. 18 No. 2, pp. 229-39.

Curme, M. and Stefanec, N. (2007), Worker quality and labor market sorting, Economic Letters,Vol. 96 No. 2, pp. 202-8.

Dale-Olsen, H. (2004), Wages, fringe benefits and worker turnover, Labour Economics,Vol. 13 No. 1, pp. 87-105.

Elliott, R.F. and Murphy, P.D. (1986), The determinants of the coverage of PBR systems inBritain, Journal of Economic Studies, Vol. 13 No. 3, pp. 38-50.

Evan, W.E. and MacPhearson, D.A. (1996), Firm size and employer turnover, Industrial andLabor Relations Review, Vol. 49 No. 4, pp. 707-28.

Ewing, B.T. (1996), Wages and performance-based pay: evidence from the NLSY, EconomicLetters, Vol. 51 No. 2, pp. 241-6.

Freeman, R.B. (1978), Job satisfaction as an economic variable, American Economic Review,Vol. 68 No. 2, pp. 135-41.

Frey, B.S. (1997), On the relationship between intrinsic and extrinsic work motivation,International Journal of Industrial Organization, Vol. 15 No. 4, pp. 427-39.

Garboua, L.L., Montmarquette, C. and Simonnet, V. (2007), Job satisfaction and quits, LabourEconomics, Vol. 14 No. 2, pp. 251-68.

Geddes, L.A. and Heywood, J.S. (2003), Gender and piece rates, commissions, and bonuses,Industrial Relations, Vol. 42 No. 3, pp. 419-44.

JES39,6

672

Goddard, J. (2001), High performance and the transformation of work the implications ofalternative work practices for the experience and outcomes of work, Industrial and LaborRelations Review, Vol. 54 No. 4, pp. 776-805.

Golden, C. (1986), Monitoring costs and occupational segregation by sex: a historical analysis,Journal of Labor Economics, Vol. 4 No. 1, pp. 1-27.

Golden, C. (1990), Understanding the Gender Gap: An Economic History of American Women,Oxford University Press, New York, NY.

Green, C. and Heywood, J.S. (2008), Does performance pay increase job satisfaction?,Economica, Vol. 75 No. 300, pp. 710-28.

Green, C. and Heywood, J.S. (2010), Profit sharing, separation, and training, British Journal ofIndustrial Relations, July.

Guthrie, J.P. (2000), Alternative pay practices and employee turnover: an organizationaleconomics perspective, Group & Organization Management, Vol. 25 No. 4, pp. 419-39.

Hashimoto, M. (1979), Bonus payments, on-the-job training, and lifetime employment in Japan,Journal of Political Economy, Vol. 87 No. 5, pp. 1086-104.

Heywood, J.S. and Wei, X. (2006), Performance pay and job satisfaction, Journal of IndustrialRelations, Vol. 48 No. 4, pp. 523-40.

Kennedy, P.W. (1995), Performance pay, productivity and morale, Economic Record, Vol. 71No. 3, pp. 240-7.

Klein, K.J. and Hall, R.J. (1988), Correlates of employee satisfaction with stock ownership: wholikes an ESOP most?, Journal of Applied Psychology, Vol. 73 No. 4, pp. 630-8.

Kreps, D.M. (1997), Intrinsic motivation and extrinsic incentives, American Economic Review,Vol. 87 No. 2, pp. 359-64.

Kruse, D. (1993), Profit Sharing: Does It Make a Difference?: The Productivity and Stability Effectsof Employee Profit-sharing Plans, W.E. Upjohn Institute for Employment Research,Kalamazoo, MI.

Kruse, D.L., Freeman, R.B. and Blasi, J.R. (2010), Do workers gain by sharing?, in Kruse, D.L.,Freeman, R.B. and Blasi, J.R. (Eds), Shared Capitalism at Work: Employee Ownership,Profit and Gain Sharing, and Broad-based Stock Options, NBER, Chicago, IL, pp. 257-89.

Lazear, E.P. (1981), Agency, earnings profiles, productivity, and hours restrictions, AmericanEconomic Review, Vol. 71 No. 4, pp. 606-20.

Lazear, E.P. (1986), Salaries and piece rates, Journal of Business, Vol. 59 No. 3, pp. 405-31.

Light, A. and McGarry, K. (1998), Job change patterns and the wages of young men, Review ofEconomics & Statistics, Vol. 80 No. 2, pp. 276-86.

McCausland, W., Pouliakas, K. and Theodossiou, I. (2005), Some are punished and some arerewarded: a study of the impact of performance pay on job satisfaction, InternationalJournal of Manpower, Vol. 26 Nos 7/8, pp. 636-59.

Milgrom, P. and Roberts, J. (1992), Economics, Organization and Management, Prentice-Hall,New York, NY.

Parent, D. (1999), Methods of pay and earnings: a longitudinal analysis, Industrial and LaborRelations Review, Vol. 53 No. 1, pp. 71-85.

Pencavel, J.H. (1972), Wages, specific training, and labor turnover in US manufacturingindustries, International Economic Review, Vol. 13 No. 1, pp. 53-64.

Royalty, A.B. (1996), The effects of job turnover on the training of men and women, Industrialand Labor Relations Review, Vol. 49 No. 3, pp. 506-21.

Performance payand employee

turnover

673

Saneyoshi, Y. (2001), Employee behavior and company stock ownership, doctoral dissertation,Harvard University, Cambridge, MA.

Scoppa, V. (2003), The role of turnover costs in the enforcement of performance-related paycontracts, Metroeconomica, Vol. 51 No. 1, pp. 60-78.

Seiler, E. (1984), Piece rates vs. time rates: the effect of incentives on earnings, The Review ofEconomics and Statistics, Vol. 66 No. 3, pp. 363-76.

Sliwka, D. and Grund, C. (2006), Performance pay and risk aversion, IZA Discussion PaperNo. 2012, March.

Weitzman, M.L. (1985), The simple macroeconomics of profit-sharing, American EconomicReview, Vol. 75 No. 5, pp. 937-53.

Further reading

Blinder, A.S. (1990), Paying for Productivity: A Look at the Evidence, The Brookings InstitutionPress, Washington, DC.

Green, W.H. (2007), Econometric Modeling Guide: Volume 1, LIMDEP Version 9, EconometricSoftware, Inc., Plainview, NY.

Corresponding authorPatrick L. OHalloran can be contacted at: pohallor@monmouth.edu

JES39,6

674

To purchase reprints of this article please e-mail: reprints@emeraldinsight.comOr visit our web site for further details: www.emeraldinsight.com/reprints