reconsidering pay dispersion’s effect on the …leeds-faculty.colorado.edu/dahe7472/trevor...

RECONSIDERING PAY DISPERSION’S EFFECT ON THEPERFORMANCE OF INTERDEPENDENT WORK: RECONCILING

SORTING AND PAY INEQUALITY

CHARLIE O. TREVORUniversity of Wisconsin–Madison

GREG REILLYUniversity of Connecticut

BARRY GERHARTUniversity of Wisconsin–Madison

Pay dispersion in interdependent work settings is virtually universally argued to bedetrimental to performance. We contend, however, that these arguments often con-found inequality with inequity, thereby overestimating inequity concerns. Conse-quently, we adopt a sorting (attraction and retention) perspective to differentiatebetween pay dispersion that is used to secure valued employee inputs and pay disper-sion that is not so used. We find that the former is positively related to interdependentteam performance, the latter has no effect or is detrimental, and the approach itselfhelps to reconcile the pay dispersion literature’s disparate results. Curvilinearity testsreveal potential constraints on the sorting argument.

Because interdependent work requiring substan-tial employee interaction to complete core tasks iscommonplace in today’s organizations, it is criticalto understand how to most effectively manage em-ployees in this type of setting. Pay dispersion, atopic of great interest in the recent managementliterature, is virtually always presumed to be coun-terproductive when work is interdependent. Nu-merous authors from economics (e.g., Akerlof &Yellen, 1988; Hicks, 1963; Levine, 1991) and man-agement (e.g., Bloom, 1999; Ferraro, Pfeffer, & Sut-ton, 2005; Harrison & Klein, 2007; Shaw, Gupta, &Delery, 2002) have argued that pay dispersion in aninterdependent work context causes inequity per-ceptions corrosive to employee attitudes, commit-ment, and cooperation, thereby hampering collec-tive (i.e., team, unit, organizational) performance.In contrast, we argue here that the commonly heldconceptual arguments against pay dispersion in in-terdependent work settings conflict with both rel-evant theory and the extant empirical research. Itis, for example, difficult to reconcile the inequity-based denunciation of pay dispersion in interde-

pendent work settings with the substantial pay-for-performance literature that advocates payingunequal contributors unequally (as failure to do sois itself an inequitable practice [Brown, Sturman, &Simmering, 2003; Heneman & Werner, 2005]).

Such inconsistency, however, is critical to rede-veloping theory because it provides the opportu-nity to “challenge the value of a theory and toexplore its weaknesses and problems” (Alvesson &Karreman, 2007: 1265), which is our goal here. Webelieve the inconsistencies in this case to be con-sequences of conceptual and empirical approachesto employee inputs (e.g., equating pay inequalitywith pay inequity) that in fact are incompatiblewith key tenets of not only equity theory, but alsoof sorting principles that address the attraction andretention of quality employees. This conceptualand empirical confluence risks leading scholarsand professionals to adopt conceptual approachesand pay practices that are of questionable validity.Thus, we aim to challenge the conventional wis-dom with a theory-grounded framework that bothcontradicts the inequity-based critiques of pay dis-persion in interdependent work and largely re-solves the discrepancies in the relevant empiricalresearch.

Although authors studying the pay dispersion–performance relationship at all levels of work in-terdependence tend to pit the incentive potential ofpay dispersion against its potential inequity-driven

We thank Associate Editor Peter Bamberger for hisinsights during the review process. We also thank StefanKoehler for his assistance in data collection.

Editor’s note: The manuscript for this article was ac-cepted for publication during the term of AMJ’s formereditor-in-chief, R. Duane Ireland.

� Academy of Management Journal2012, Vol. 55, No. 3, 585–610.http://dx.doi.org/10.5465/amj.2006.0127

585

Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s expresswritten permission. Users may print, download, or email articles for individual use only.

disruptiveness (e.g., Bloom, 1999; Kepes, Delery, &Gupta., 2009; Pfeffer & Langton, 1993; Shaw et al.,2002), we move beyond these existing perspectivesby theorizing that sorting is a critical mechanismthrough which pay dispersion, even in highly in-terdependent work settings, can facilitate groupperformance. Additionally, we differentiate be-tween pay dispersion that is explained by produc-tivity-relevant employee inputs and pay dispersionthat is not to illustrate why both positive and neg-ative dispersion effects on performance haveemerged in previous research. We then test, in ahighly interdependent work context, whether paydispersion and team performance are positively re-lated when (and only when) pay dispersion is ex-plained by the sorting of employee inputs. Finally,given recent nonlinear approaches to the impacts ofpay dispersion (Brown et al., 2003) and “humancapital” (knowledge, skills, abilities, and other in-dividual characteristics) (Ployhart, Weekley, &Ramsey, 2009), we explore possible curvilinear dis-persion effects that would reveal constraints on thesorting rationale.

CONCEPTUAL DEVELOPMENT

Key Definitions and Model Components

We study pay dispersion in a team-based contextin which the tasks required for team success arehighly interdependent, defining interdependenceas the degree to which completion of their organi-zation’s tasks requires employees to interact (Cum-mings, 1978; Thompson, 1967). We investigate lat-eral pay dispersion within the dominant job in anindustry. Lateral (or horizontal) pay dispersion in-volves pay differences among employees withinthe same job or organizational level (Bloom, 1999;Pfeffer, 1994). In contrast to vertical pay dispersion(i.e., pay differences between employees in jobs atdifferent hierarchical levels), lateral pay dispersionis often the focus when many incumbents hold asingle core job, as is frequently the case for those inmore professionalized occupations (sports profes-sionals, physicians, nurses, lawyers, accountants,teachers, academicians, etc.).

Central to our thesis is an input-based differenti-ation of pay dispersion types. Dispersion in ex-plained pay (DEP) is the amount of variation in theemployee pay that is tied to productivity-relevantemployee inputs. Figure 1 depicts this and otherkey relationships. Specifically, because certain em-ployee inputs (typically, job performance) are asso-ciated with organizational productivity, they areproductivity-relevant (and thus strategically rele-vant) explanations for pay dispersion under both

independent and interdependent work. Pay disper-sion associated with securing these productivity-relevant employee inputs (via attraction and reten-tion) is, by definition, DEP, as illustrated in Figure1. In contrast, dispersion in unexplained pay (DUP)is the amount of variation in the employee pay thatis unexplained by these productivity-relevant in-puts. Thus, DUP is dispersion in the pay that isattributable to factors unrelated to employee pro-ductivity (e.g., politics, discrimination, favoritism,random decisions).

Researchers to date have typically studied over-all pay dispersion, with each study’s statisticalmodeling specification determining (at times inad-vertently, in our view) whether DEP or DUP wasprimarily at work. In Table 1 we summarize theempirical pay dispersion–performance researchand present our informal assessment of whether thepay dispersion modeled more closely representedDEP or DUP. Such classification should allow for abetter understanding of the disparate research find-ings (we return to Table 1 in more detail later).

Pay Dispersion and Sorting

The compensation literature (Gerhart & Rynes,2003; Lazear, 2000) classifies pay effects on em-ployee behaviors as associated with either incen-tives or the acquisition and retention of talent (i.e.,sorting). We believe that incentive pay can, undercertain circumstances, have positive implicationsfor team performance, but our emphasis here is onsorting, which, as an explanation for performancein interdependent work settings, has been virtuallyignored in the pay dispersion literature. The sortingpremise stipulates that pay linked to inputs canyield human capital advantages by attracting andretaining higher-ability, better-performing employ-ees (Gerhart & Rynes, 2003; Lazear, 2000). In partbecause pay dispersion, pay level, and pay-for-per-formance practices are jointly tied to exactly thesame aggregated individual pay level decisions,there is indirect evidence that greater dispersion inthe pay linked to productivity-relevant employeeinputs (i.e., DEP) should facilitate such sorting.Larger pay differentials (and thus more DEP) arefrequently characterized by high pay levels neededto secure the most talented employees, with re-search indicating that high pay level increases theability to attract and retain the most talented work-ers (Krueger, 1988; Shaw, Delery, Jenkins, & Gupta,1998). Similarly, pay-for-performance, as distinctfrom pay level, also appears to be associated withboth larger input-based pay differentials (and thusmore DEP) and sorting advantages, in that it posi-tively affects the attraction of high-ability, high-

586 JuneAcademy of Management Journal

TA

BL

E1

Key

Ele

men

tsof

Stu

die

son

Pay

Dis

per

sion

Eff

ects

onP

erfo

rman

cea

Stu

dy

Sam

ple

Per

form

ance

Mea

sure

Pay

Dis

per

sion

Typ

eL

evel

ofW

ork

Inte

rdep

end

ence

Stu

dy

Ap

pea

rsto

Par

tial

Ou

tD

isp

ersi

onin

Exp

lain

edP

ay(D

EP

)b

Est

imat

edP

ayD

isp

ersi

onM

odel

ed

Ove

rall

Eff

ect

ofP

ayD

isp

ersi

onon

Per

form

ance

Via

Mea

nP

ayL

evel

Cov

aria

te

Via

Pay

-for

-P

erfo

rman

ceC

ovar

iate

Via

Em

plo

yee

Inp

uts

Cov

aria

te

Bea

um

ont

&H

arri

s(2

003)

U.K

.fi

rms

infi

vein

du

stri

esP

lan

t-le

vel

labo

rp

rod

uct

ivit

yV

erti

cal

Un

know

nN

oN

oN

oD

EP

Pri

mar

ily

Pos

itiv

eB

ecke

r&

Hu

seli

d(1

992)

Pro

fess

ion

alau

tora

cin

gd

rive

rsR

ace

fin

ish

ing

pos

itio

nL

ater

alL

owN

oN

oN

oD

EP

Pos

itiv

e

Blo

om(1

999)

Pro

fess

ion

alba

seba

llp

laye

rs/t

eam

sV

ario

us

on-f

ield

and

off-

fiel

dm

easu

res

Lat

eral

Low

Yes

No

Yes

(tea

mta

len

t)D

UP

Neg

ativ

e

Bro

wn

,S

turm

an,

&S

imm

erin

g(2

003)

Acu

teca

rece

nte

rsin

Cal

ifor

nia

hos

pit

als

Ave

rage

len

gth

ofst

ay;

surv

ival

rate

;R

OA

Ver

tica

lU

nkn

own

Yes

No

No

DE

PM

ixed

Con

yon

,P

eck,

&S

adle

r(2

001)

U.K

.st

ock

mar

ket

firm

sR

OA

;to

tal

shar

ehol

der

retu

rnV

erti

cal

Un

know

nN

oN

oN

oD

EP

No

effe

ct

Cow

her

d&

Lev

ine

(199

2)B

usi

nes

su

nit

sw

ith

hea

dqu

arte

rsin

U.S

.an

dE

U

Pro

du

ctqu

alit

yV

erti

cal

Un

know

nN

oN

oY

es(b

usi

nes

su

nit

size

argu

edto

bea

pro

xyfo

rin

terc

lass

hu

man

cap

ital

dif

fere

nce

s;jo

bp

ayd

iffe

ren

ces

also

rem

oved

inon

eof

the

dis

per

sion

mea

sure

sby

usi

ng

rati

oof

two

pay

-rel

ativ

e-to

-mar

ket

ind

icat

ors

asd

isp

ersi

on)

DU

PN

egat

ive

Dep

ken

(200

0)P

rofe

ssio

nal

base

ball

team

sW

inn

ing

per

cen

tage

Lat

eral

Low

Yes

No

No

DU

PN

egat

ive

Din

g,A

khta

r,&

Ge

(200

9)C

hin

ese

man

ufa

ctu

rin

gan

dse

rvic

efi

rms

Su

bjec

tive

rep

orts

ofsa

les

grow

than

dp

rod

uct

orse

rvic

equ

alit

yL

ater

alan

dve

rtic

alU

nkn

own

No

No

No

DE

PM

ixed

Eri

ksso

n(1

999)

Dan

ish

firm

sP

rofi

ts/s

ales

Ver

tica

lU

nkn

own

cY

esN

oN

oD

EP

Pos

itiv

eF

ran

ck&

Nu

esch

(201

1)P

rofe

ssio

nal

socc

erte

ams

Win

nin

gp

erce

nta

ge;

leag

ue

stan

din

gL

ater

alH

igh

Yes

No

Yes

(tal

ent

het

erog

enei

ty)

DU

PN

egat

ive

Fri

ck,

Pri

nz,

&W

inke

lman

n(2

003)

Tea

ms

from

fou

rp

rofe

ssio

nal

spor

tsW

inn

ing

per

cen

tage

Lat

eral

Mix

edY

esN

oN

oD

UP

Mix

ed

Gru

nd

&W

este

rgaa

rd-

Nie

lsen

(200

8)D

anis

hfi

rms

Val

ue

add

edp

erem

plo

yee

Ver

tica

lU

nkn

own

Yes

No

Yes

(ed

uca

tion

and

edu

cati

ond

isp

ersi

on)

DU

PN

oef

fect

Hey

man

(200

5)S

wed

ish

firm

sP

rofi

tsp

erem

plo

yee

Ver

tica

lU

nkn

own

No

No

Yes

(hu

man

cap

ital

leve

ls)

DU

PP

osit

ive

Hib

bs&

Loc

kin

g(2

000)

Sw

edis

hfi

rms

Pro

du

ctiv

ity

Ver

tica

lU

nkn

own

Yes

No

No

DE

PP

osit

ive

Jew

ell

&M

olin

a(2

004)

Pro

fess

ion

alba

seba

llte

ams

Win

nin

gp

erce

nta

geL

ater

alL

owN

oN

oY

es(s

ever

alh

um

anca

pit

alan

don

-fi

eld

per

form

ance

mea

sure

s)D

UP

Neg

ativ

e

Kep

es,

Del

ery,

&G

up

ta(2

009)

Tru

cktr

ansp

orta

tion

firm

s(d

rive

rs)

Saf

ety;

RO

E;

pro

du

ctiv

ity;

oper

atin

gra

tio

Lat

eral

Low

Yes

No

Yes

(sen

iori

ty-b

ased

pay

)D

UP

Pos

itiv

e

Lee

,L

ev,

&Y

eo(2

008)

Top

man

agem

ent

grou

ps

inp

ubl

icfi

rms

Mar

ket

valu

atio

n;

RO

AL

ater

alan

dve

rtic

alU

nkn

own

No

No

No

DE

PP

osit

ive

Leo

nar

d(1

990)

U.S

.fi

rms

RO

E;

chan

gein

RO

EV

erti

cal

Un

know

nY

esN

oY

es(j

obp

ayp

arti

ally

cap

ture

dby

rati

oof

leve

l1

and

2p

ayto

leve

l5

and

6p

ay;

hie

rarc

hy

ind

exm

ayal

sop

arti

alou

tjo

bp

ay)

DU

PN

oef

fect

Mai

n,

O’R

eill

y,&

Wad

e(1

993)

U.S

.fi

rms

RO

AV

erti

cal

Un

know

nc

No

No

No

DE

PP

osit

ive

(Con

tin

ued

)

TA

BL

E1

(Con

tin

ued

)

Stu

dy

Sam

ple

Per

form

ance

Mea

sure

Pay

Dis

per

sion

Typ

eL

evel

ofW

ork

Inte

rdep

end

ence

Stu

dy

Ap

pea

rsto

Par

tial

Ou

tD

isp

ersi

onin

Exp

lain

edP

ay(D

EP

)b

Est

imat

edP

ayD

isp

ersi

onM

odel

ed

Ove

rall

Eff

ect

ofP

ayD

isp

ersi

onon

Per

form

ance

Via

Mea

nP

ayL

evel

Cov

aria

te

Via

Pay

-for

-P

erfo

rman

ceC

ovar

iate

Via

Em

plo

yee

Inp

uts

Cov

aria

te

Pfe

ffer

&L

angt

on(1

993)

Un

iver

sity

facu

lty

Pu

blic

atio

ns

Lat

eral

Low

No

Yes

(wit

hin

-d

epar

tmen

tco

rrel

atio

nbe

twee

np

ayan

dp

rod

uct

ivit

y)

No

DU

PN

egat

ive

Sh

aw,

Gu

pta

,&

Del

ery

(200

2)T

ruck

tran

spor

tati

onfi

rms

(dri

vers

);S

afet

y;p

rod

uct

ivit

y;p

erfo

rman

cep

erce

pti

ons

Lat

eral

Low

Yes

No

Yes

(sen

iori

ty-b

ased

pay

)D

UP

Pri

mar

ily

neg

ativ

e/n

oef

fect

Con

cret

ep

ipe

pla

nts

(pro

du

ctio

nw

orke

rs)

Lat

eral

Low

cY

esN

oY

es(s

enio

rity

-bas

edp

ay)

DU

PP

rim

aril

yn

egat

ive/

no

effe

ctS

iege

l&

Ham

bric

k(2

005)

Top

man

agem

ent

grou

ps

inp

ubl

icfi

rms

Mar

ket-

to-b

ook

valu

e;to

tal

shar

ehol

der

retu

rns

Lat

eral

and

vert

ical

Un

know

nN

oY

es(f

irm

size

and

nu

mbe

rof

leve

ltw

oex

ecu

tive

sar

gued

tobe

pro

xies

for

pay

-for

-p

erfo

rman

ced

iffe

ren

ces)

Yes

(top

man

agem

ent

grou

psi

zear

gued

tobe

ap

roxy

for

job

fun

ctio

nd

iffe

ren

ces

that

wou

ldle

giti

mat

ely

war

ran

tp

ayd

iffe

ren

ces)

DU

PN

oef

fect

Som

mer

s(1

998)

Pro

fess

ion

alh

ocke

yte

ams

Tea

mp

oin

ts(t

he

team

stan

din

gsd

eter

min

ant)

Lat

eral

Hig

hY

esN

oN

oD

UP

Neg

ativ

e

aW

eu

sed

two

dec

isio

nru

les

tom

ake

the

dis

per

sion

inex

pla

ined

pay

/dis

per

sion

inu

nex

pla

ined

pay

(DE

P/D

UP

)dis

tin

ctio

n:(

1)w

eju

dge

dla

tera

lpay

dis

per

sion

stu

die

sas

mod

elin

gD

UP

ifw

eco

ded

as“y

es”

atle

ast

one

ofth

eth

ree

colu

mn

sre

pre

sen

tin

gav

enu

esfo

rco

ntr

olli

ng

for

stra

tegi

cp

ayd

isp

ersi

onex

pla

nat

ion

s;(2

)gi

ven

that

vert

ical

pay

dis

per

sion

inco

rpor

ates

job

pay

dif

fere

nce

san

dth

atjo

bp

ayis

the

pri

mar

yd

eter

min

ant

ofac

tual

pay

(Ger

har

t&

Mil

kovi

ch,1

992)

,ver

tica

lp

ayd

isp

ersi

onst

ud

ies

that

did

not

par

tial

out

job

pay

orjo

ble

veli

nd

icat

ors

wer

eju

dge

dto

mod

elD

EP

.Stu

die

sin

clu

ded

nu

mer

ous

dep

end

entv

aria

bles

,qu

adra

tic

term

s,an

d/o

rin

tera

ctio

nte

rms;

tosi

mp

lify

our

sum

mar

y,th

efi

nal

colu

mn

refl

ects

our

imp

ress

ion

ofth

eov

eral

lli

nea

rp

ayd

isp

ersi

onef

fect

,u

sual

lyas

inte

rpre

ted

byth

est

ud

ies’

auth

ors.

bT

he

term

“par

tial

out”

mea

ns

tore

mov

eth

eef

fect

sof

;hen

cein

this

con

text

“par

tial

ing

out”

mea

ns

that

the

infl

uen

ceof

DE

Ph

asbe

enre

mov

edfr

omth

ep

ayd

isp

ersi

onm

easu

reby

stat

isti

call

yco

ntr

olli

ng

for

cert

ain

cova

riat

esin

the

anal

yses

.c

Alt

hou

ghth

ese

thre

est

ud

ies

incl

ud

ed,v

iain

tera

ctio

nte

rms,

atte

mp

tsto

exam

ine

dis

per

sion

effe

cts

atlo

wer

and

hig

her

inte

rdep

end

ence

,in

non

eof

the

thre

eis

itcl

ear

that

the

wor

kit

self

isev

ertr

uly

inte

rdep

end

ent

inn

atu

re(s

eeD

iscu

ssio

nsu

bsec

tion

“Pay

Dis

per

sion

and

Inte

rdep

end

ence

.”)

performance employees (Cadsby, Song, & Tapon,2007; Lazear, 2000) and the retention of “high per-formers” (e.g., Salamin & Hom, 2005; Trevor, Ger-hart, & Boudreau, 1997).

Although pay level and dispersion in explainedpay are strongly related and thus have some similarsorting effects, DEP also has unique sorting advan-tages. Individuals, including those with high per-formance and talent inputs, depend on social com-parisons to evaluate their pay (e.g., Adams, 1963;Barnard, 1938; Pfeffer & Langton, 1993) and react tothe degree of perceived advantage in these compar-isons. DEP results in those high in a pay distribu-tion (via performance or talent inputs) having highrelative pay, and evidence shows that pay disper-sion strengthens perceptions of favorable pay forthese employees (Trevor & Wazeter, 2006). As aresult, individuals located near the top of a highlydispersed pay distribution should feel relativelyadvantaged and, consequently, be less likely to quit(Pfeffer & Davis-Blake, 1992). Additionally, on theattraction side of sorting, because individuals enjoythe status associated with higher rank in a payhierarchy, they have an incentive to choose thehierarchy in which they will enjoy the most rela-tive pay and status advantages (Frank, 1985), whichwill be a distribution with high DEP. Finally, wenote that a low-DEP alternative, paying everyone atthe same high level, loses the social comparison–based sorting advantages, drives up labor costs, andimplies that all individuals and their roles areequally critical to success and thus require top-of-the-market pay.

Little empirical work, however, exists onwhether pay dispersion in general, or dispersion inexplained pay, in particular, has sorting conse-quences. Using a sample of high-level administra-tive personnel from multiple academic institutions,a context that appears to be very low in work inter-dependence, Pfeffer and Davis-Blake (1992) foundsupport for pay dispersion’s sorting potential; spe-cifically, for those high in the pay distributions,turnover diminished as pay dispersion (and pre-sumably DEP) increased. Similarly, Shaw andGupta (2007) found that, when performance-basedincreases were emphasized and the pay system waswell communicated (a likely context for DEP),greater pay dispersion predicted a reduction in thenumber of “high-performing” truck drivers whoquit. However, because this trucking sample (likethe Pfeffer and Davis-Blake sample) did not repre-sent an interdependent work setting (Shaw et al.,2002: 495), the relevance of pay dispersion to theretention of interdependent workers remains un-tested. Further, we found no research directly ad-dressing pay dispersion effects on employee attrac-

tion. To formally examine the sorting premise thatis the foundation for our unconventional team per-formance predictions, we propose that teams withhigh DEP (i.e., teams that more closely tie pay toproductivity-relevant inputs) will fare better in theattraction and retention of high-input employees(i.e., those who previously have performed at highlevels).

Hypothesis 1. Relative to teams in which dis-persion in explained pay is low, teams inwhich dispersion in explained pay is high at-tract more high-input players and retain morehigh-input players.

Dispersion in Explained Pay’s Sorting Effects(and Lack of Inequity Effects) on TeamPerformance

Sorting advantages that accrue to DEP should, allelse being equal, enhance team performance. Butdespite some recent progress in understanding thefavorable incentive implications of pay dispersionwhen organizations pay for inputs in a highly in-dependent work context (Kepes et al., 2009; Shawet al., 2002), scholars agree that, in interdependentsettings, pay dispersion is particularly detrimentalto aggregate performance (e.g., Akerlof & Yellen,1988; Bloom, 1999; Ferraro, Pfeffer, & Sutton, 2005;Franck & Nuesch, 2011; Harrison & Klein, 2007;Hicks, 1963; Levine, 1991; Kepes et al., 2009; Pfef-fer & Langton, 1993), even if the dispersion is tiedto inputs. Authors linking pay dispersion with un-fairness at any level of work interdependence typ-ically cite equity theory (Adams, 1963) principlesto argue that large pay differentials yield inequityperceptions, psychological distress, reduced coop-eration, disharmony, lower commitment, and in-creased turnover (e.g., Akerlof & Yellen, 1988;Bloom, 1999; Ferraro et al., 2005; Lazear, 1989;Levine, 1991; Pfeffer & Langton, 1993).

Pay dispersion, however, means only that payallocation is unequal, not that it is inequitable.Leventhal (1976) and Steers and Porter (1983) de-scribed pay equality as equal pay for all employees(i.e., low pay dispersion, regardless of pay equity)and pay equity as pay proportionate to employeeinputs (i.e., potentially any level of pay dispersion,as long as pay is tied to productivity-relevant in-puts). This equality versus equity difference,though often overlooked in pay dispersion re-search, is crucial: pay dispersion implies pay in-equality, DEP implies pay equity, and only DUPimplies pay inequity (see Figure 1). Scholars ad-dressing equity and fairness clearly stipulates thatit is pay inequity, rather than pay inequality, that

2012 589Trevor, Reilly, and Gerhart

prompts negative employee reactions (e.g., Am-brose & Kulik, 1999; Deutsch, 1985; Heneman &Judge, 2000; Leventhal, 1976). Therefore, pay dis-

persion that is explained by inputs regarded asproductivity-relevant should be perceived as equi-table and should not yield counterproductive re-

FIGURE 1Conceptual and Empirical Perspectives on Dispersion in Explained Pay (DEP)a

Conceptual Framework

Overall Pay Dispersion

Sorting and potential incentive benefits depend on relativeamounts of DEP and DUPPay inequality

Productivity-Relevant Employee Input

Job performance, abilityJob held

Dispersion in Explained Pay (DEP)

Pay disperson used to secureproductivity-relevant employee inputsSorting and potential incentives benefitPay equity

TeamPerformance

Dispersion in Unexplained Pay (DUP)

Pay dispersion independent of productivity-relevant inputsNo obvious sorting or incentive benefitsPay inequity

Empirical Approach 1

Empirical Approach 2

DEP(Directly measured or

through partialing)

DUP(Directly measured or

through partialing)

DEP Effect(positive, via sorting and,

potentialy, incentives)

DUP Effect(negative or zero)

DUP Effect(negative or zero,

direct effect)

Pay DispersionTeam

Performance

Productivity-Relevant Employee

InputsDEP effect

(positive, indirect [sorting] effectof pay dispersion through inputs)

a Both empirical approaches are consistent with the conceptual framework presented above (top).


sponses. Hence, the idea that DEP (or even overallpay dispersion if primarily DEP) should yield neg-ative reactions is not, contrary to what has oftenbeen contended, consistent with equity and fair-ness theories.

As long as individual contributions can be iden-tified, nothing in the above argument is particularto independent work. In fact, in interdependentcontexts where individual contributions are iden-tifiable, inequity perceptions are more likely with-out pay dispersion, as pay equality, assuming vari-ation in inputs, produces pay inequity. Indeed,researchers who focus on social loafing, which isthe undesired tendency for people to reduce effortand productivity when in groups, maintain thatmonetary rewards tied to individual inputs “canserve as powerful incentives for behavior, counter-ing the reduction in effort typically exhibited byparticipants who are combining their efforts”(Sheppard, 1993: 70). In this well-developed liter-ature, it is the task interdependence itself that leadsto motivation loss, while pay tied to individuals’productivity-relevant inputs, and the subsequentDEP, is a tactic to combat it. Thus, when work isinterdependent and pay is tied to individual in-puts, we find little conceptual support for the pre-dictions that pay dispersion will lead to perceivedinequity and its behavioral fallout.

In terms of empirical support, perhaps becausethe conventional wisdom has so often been pre-sumed to be true, only four Table 1 studies (Bloom,1999; Eriksson, 1999; Main, O’Reilly, & Wade,1993; Shaw et al., 2002) explicitly reported testingwhether work interdependence resulted in nega-tive pay dispersion effects on performance or mit-igated any otherwise favorable pay dispersion ef-fects. Of these, only Bloom (1999) reported clearsupport (later we argue that this study involvesneither interdependent work nor DEP). A numberof studies in which work interdependence is notaddressed have reported negative pay dispersioneffects, however, and it is likely that these havefueled the popular conception of pay dispersion ascorrosive to teamwork. As we focus on in the nextsection, though, most of the empirical literaturecited in support of detrimental pay dispersion ef-fects, regardless of work interdependence, appearsto be modeling dispersion in unexplained payrather than dispersion in explained pay. DUP, bydefinition, has none of the sorting advantages ofDEP, and it is more susceptible to the inequity-based problems so often cited (see Figure 1).

In sum, under our conceptual model, and as pre-dicted in Hypothesis 1, DEP should yield sortingadvantages when work is interdependent (or inde-pendent, for that matter), as larger pay differentials

based on productivity-relevant inputs facilitate toptalent attraction and retention. As is often stipu-lated (e.g., Barnard, 1938; Becker & Gerhart, 1996;Pfeffer, 1994), and as has been empirically demon-strated (e.g., Ployhart et al., 2009), enhanced hu-man capital results in greater aggregate-level pro-ductivity. Thus, to the degree that DEP sorts higheraggregate human capital to a team, the effect onperformance should be positive. Furthermore, DEP,by definition, entails pay inequality but not the payinequity that fairness theories identify as problem-atic. Finally, the social loafing literature (e.g., Shep-pard, 1993) stipulates that pay tied to inputs ininterdependent work has incentive value. Conse-quently, we do not see DEP in the presence of workinterdependence as likely to generate the consider-able inequity perceptions and effort reduction nec-essary to overwhelm DEP’s sorting benefits. To ad-dress this, we provide the first empirical test of theentire pay dispersion–inputs–group productivitycausal chain.

We first isolated DEP and DUP measures, thentested for a DEP effect and for differences betweenDEP and DUP effects. Second (see Figure 1), wetested whether the pay dispersion that is used tosecure (is mediated by) employee inputs (i.e., thesorting effect) has a positive effect on performance.This indirect effect, by definition, is a DEP effect.We then compared the indirect (DEP) effect of thepay dispersion used to secure inputs with the di-rect (DUP) effect of the pay dispersion that is inde-pendent of inputs; thus, we differentiated betweenpay dispersion that should facilitate team perfor-mance (DEP) and pay dispersion less apt to doso (DUP).

Hypothesis 2a. Pay dispersion explained byproductivity-relevant employee inputs (DEP) ispositively related to team performance; thispositive effect is more favorable than the effectof pay dispersion that is net of inputs(i.e., DUP).

Hypothesis 2b. Productivity-relevant employeeinputs mediate overall pay dispersion’s rela-tionship with team performance, resulting in apositive indirect sorting effect (i.e., a positiveDEP effect); this sorting (DEP) effect is morefavorably related to team performance than isthe direct (DUP) effect.

Modeling, Loss of Dispersion in Explained Pay,and Emergence of Dispersion in Unexplained Pay

Given our arguments that DEP should positivelyaffect organizational performance in interdepen-dent work settings, why do studies often report


negative overall pay dispersion effects? One reasonmay be that organizations actually do at times failto adequately tie pay to productivity-relevant in-puts such as performance (e.g., Kahn & Sherer,1990; Heneman & Werner, 2005; Schwab & Olson,1990); the resultant DUP negates sorting benefitsand makes more likely the often-cited inequity-driven problems. We argue that a second reason fornegative pay dispersion effects, at all levels of workinterdependence, is a methodological artifact, inthat DUP, which has none of DEP’s sorting benefits,is sometimes inadvertently modeled in this re-search. Our concern is that, unless explicitly study-ing DUP (e.g., as did Cowherd & Levine [1992]),researchers may mistakenly attribute DUP effects tototal pay dispersion or, worse yet, to DEP, therebyperpetuating the belief that pay dispersion is inher-ently detrimental.

Specifically, we contend that, via decisionsabout what covariates to include in regression mod-els, many authors have largely “partialed out” (i.e.,removed the influence of) the DEP from total paydispersion, thus primarily leaving DUP (i.e., Hy-pothesis 2b’s and Figure 1’s direct effect of disper-sion that is independent of inputs and not expectedto have positive effects). For example, pay-levelstrategy (mean pay) has appeared as a control vari-able in several analyses of pay dispersion effects(see Table 1). One argument for including it is toaccount for a high wage effect that would manifestin talent advantages to organizations that pay more(e.g., Bloom, 1999). This rationale, however, is pre-cisely why our emphasis on pay tied to productiv-ity-relevant inputs necessitates a different model-ing approach. Controlling for mean pay level“parcels out the positive effects of pay dispersion:attraction and retention of star players who are paida great deal, thus resulting in better team perfor-mance, higher team pay, and greater dispersion”(Gerhart & Rynes, 2003: 182). In a second argument,Harrison and Klein (2007) contended that research-ers should control for within-group mean becauseit may be confounded with a disparity measure. Ina pay-setting context, however, such confoundingis exactly why we examine pay dispersion effectsboth with and without the mean controlled. Be-cause paying for talent yields both high mean payand high pay dispersion (Gerhart & Rynes, 2003),we focus on distinguishing the dispersion that co-varies with high talent and mean pay (i.e., DEP)from the dispersion that does not (i.e., DUP). Iso-lating the degree of this covariation, or “confound,”is central to our contribution.

Similarly, partialing out a pay-for-performancestrategy measure risks parsing out DEP and increas-ingly leaving DUP as the dispersion modeled. De-

spite their insightful discussion regarding fair basesfor pay dispersion, Pfeffer and Langton (1993) par-tialed out the within-department correlation be-tween employee pay and productivity (which cor-related with pay dispersion at .31) when predictingfaculty productivity. Thus, they actually modeledonly the pay dispersion that was unrelated to thepay tied to job performance inputs, likely largelycapturing DUP effects. Moreover, the pay-produc-tivity correlation was positively related to produc-tivity in the Pfeffer and Langton data (Gerhart &Rynes, 2003), suggesting that, in contrast to thedetrimental pay dispersion effect reported by thestudy’s authors, conditions were present thatwould make a positive DEP effect likely.

Like partialing out pay strategies that yield pro-ductivity-relevant inputs, controlling for produc-tivity-relevant inputs themselves also can yield themodeling of DUP and pay dispersion effects that donot reflect dispersion’s sorting benefits. Bloom(1999), for example, in analyses leading to his re-porting that pay dispersion hindered performancein interdependent work, partialed out a measure ofbaseball team talent. Similar partialing of produc-tivity-relevant inputs occurs in several other stud-ies reporting negative pay dispersion effects (e.g.,Frank & Nuesch, 2011; Jewell & Molina, 2004;Leonard, 1990; Siegel & Hambrick, 2005). Such anapproach, however, serves to parcel out the posi-tive sorting effects of pay dispersion (Gerhart &Rynes, 2003), likely leaving DUP as the pay disper-sion modeled in the regression analysis.

As an informal exploration of our belief thatthese modeling issues have led to confusion in thefield, we applied our DEP/DUP logic to the extantwork in Table 1. Thus, the table includes our judg-ment of whether DEP or DUP was the primary typeof pay dispersion modeled in each study (see thetable note for our decision rules). According tothese judgments, although 3 studies yielded resultscounter to our framework and 3 were ambiguous,17 studies supported our positions: in general, DEPyields positive effects, DUP yields zero or negativeeffects, and partialing out pay strategies and theproductivity-relevant employee inputs they pro-duce leaves less DEP, more DUP, and a lower like-lihood of observing the sorting benefits that soundpay policy can provide. Thus, we contend thatresearch from both independent and interdepen-dent work settings is likely to reveal that pay dis-persion effects strongly depend on how pay disper-sion is modeled.

Hypothesis 3. In an interdependent work set-ting, controlling for pay level strategy, pay-for-performance strategy, and productivity-rele-


vant employee inputs partials out positiveeffects of pay dispersion on team performance,resulting in a negligible or negative pay disper-sion effect.

Pay Dispersion and Curvilinearity

Although we have thus far challenged the ineq-uity-based critique of pay dispersion and promoteda more favorable sorting-based characterization ofDEP, we also acknowledge that a more nuancedapproach to DEP may be necessary. Brown et al.(2003), for example, although recognizing the valueof linking pay to inputs, still theorized on inequitygrounds that pay that is too widely dispersed maybe detrimental. At some point, even pay differencesclearly explained by inputs may be seen as toolarge, and thus inequitable. Indeed, organizationsattempt to ensure that performance-based pay dif-ferences between individuals are not dispropor-tionate to actual performance differences (Milkov-ich & Newman, 2008). With such disproportion,pay dispersion may still be “explained,” but theexplanation may be deemed inadequate, and thepay inequitable, given the extreme differentials.Thus, individual incentive effects in groups, suchas those identified in the social loafing literature(e.g., Sheppard, 1993), could be tempered by ineq-uity concerns when the subsequent pay differen-tials (i.e., DEP) are very large. Moreover, any suchinequity perceptions driven by extreme DEP mayconstrain the sorting advantages DEP would nor-mally provide, as inequity perceptions based on thesense of having been underrewarded can lead em-ployees to quit (Gerhart & Rynes, 2003; Heneman &Judge, 2000).

An additional basis for expecting diminishingsorting returns to DEP exists. An organization’s de-velopment of certain capabilities often has inherentlimits, such as a limited capacity to utilize re-sources (Helfat & Peteraf, 2003). That is, as talentresources (inputs) grow beyond a certain point, theorganization may become less effective at managingthem, thus constraining the leveraging of DEP-driven talent into performance. For instance, workteams may perceive some level of a capability, suchas the ability to manage talent, as satisfactory andrefrain from developing the capability further(Winter, 2000). On the basis of this capability argu-ment, Ployhart et al. (2009) hypothesized andfound that the positive effect of sales force humancapital on store performance diminished at higherhuman capital levels. Similarly, Barry and Stewart(1997) found a curvilinear effect of the personalitydimension extraversion on group performance, per-haps indicating “too much of a good thing” (Ger-

hart & Rynes, 2003). Hence, despite DEP’s sortingadvantages, the enhanced inputs that greater DEPyields may not be accompanied by proportionateincreases in managing those inputs and, subse-quently, in team performance. This potential, cou-pled with the inequity-based argument for dimin-ishing returns to DEP, suggests the followingnonlinear relationship.

Hypothesis 4. The positive effect of dispersionin explained pay on team performance is at-tenuated at high levels of such dispersion.

METHODS

To test our hypotheses, we studied NationalHockey League (NHL) teams, a population in whichdata on pay and readily observable performance areavailable at both individual and organizational lev-els and in which player movement across teams isconsiderable. Also, team performance in hockeydepends on strong work interdependencies (Beau-champ & Bray, 2001; Foster & Washington, 2009;Frey, Kerr, & Lee, 1986; Gerhart & Rynes, 2003); asdiscussed earlier, work settings in which interde-pendence is high are regularly cited as the contextin which pay dispersion will be particularly dis-ruptive (e.g., Akerlof & Yellen, 1988; Bloom, 1999;Levine, 1991; Pfeffer, 1994; Pfeffer & Lang-ton, 1993).

Data

The individual player and team performancedata used for this study are from the official recordsof the NHL. We acquired these records from thewebsites hockeydb.com and hockeyzoneplus.comas well as from annual editions of the NHL’s Offi-cial Guide & Record Book (e.g., National HockeyLeague, 2002). The data set used for the team-levelanalyses consisted of pay and performance data foreach team in the NHL during each of the seasonsending in 1998 through 2004. Thus, the data setconsisted of 201 total team-years, as the leaguehoused 26 teams in 1998 and expanded to 27 teamsin 1999, 28 teams in 2000, and 30 teams in 2001–04. The usable sample for our team-level analysesdropped to 175, as we used year t – 1 measures ofproductivity-relevant employee inputs when pre-dicting year t team performance.

Our team-year measures were built from an ini-tial data set of 4,465 player-year observations,which included annual performance and salary sta-tistics for each individual nongoalkeeper (“non-goalie”) player who played in the league during ourstudy period. We focused on nongoalies because


performance criteria (and thus pay-for-performancestrategies) are distinctly different for goalies; more-over, the small number of goalies per team (two orthree) precludes reliable measures of within-teampay dispersion for that group of players. Also, toenhance reliability in our productivity-relevantmeasures, we limited inclusion in the database toindividual players who appeared in at least 20 oftheir team’s 82 games in any given season.

Dependent Variable

Team performance. Our first on-ice measure,points, is calculated by the NHL by summing twopoints for each regular season win, one point foreach tie, and one point for each overtime loss (theNHL added this last component two years into ourstudy window). Points determine position in thestandings during the regular season of play, whichteams make the play-offs, and how those teams areseeded in the play-offs. Because the ultimate on-icegoal of NHL teams is to win the Stanley Cup cham-pionship, however, our second on-ice measure,round, is each team’s final position in the play-offtournament bracket. Each year 16 teams make theplay-offs, with teams advancing to subsequentrounds only by winning a best-of-seven series ofgames. Round takes on the following values foreach year: 0 (all non-play-off teams); 1 (8 teams thatlost in the first round); 2 (4 teams that lost in thesecond round); 3 (2 teams that lost in the thirdround); 4 (1 team that lost in the fourth round); and5 (1 championship team). Because play-off injuries,opponent match-ups, and reduced variance in themeasure suggest that round will be less reliablypredicted than points, we expected that the supportfor our hypotheses might be weaker when predict-ing round. We restrict our analysis to the predictionof on-ice performance because the linkage betweenpay strategies and team off-ice performance (e.g.,profitability) is significantly more distal andtenuous.

Independent Variables

Because empirically isolating DEP and DUP ischallenging, we took three approaches—partialing,mediation, and predicted values/residuals—andassessed whether our findings are robust to tech-nique choice. We summarize our approach to thekey predictors in Table 2.

Inputs. Job performance, the classic employeeinput from a rewards perspective (Steers & Porter,1983), is the basis for our sorting arguments for paydispersion effects on team performance. Based on ameasure of individual player value officially ap-

proved by the National Hockey League Players’ As-sociation (NHLPA) for use in fantasy hockey leagueplay, our inputs measure represents the perfor-mance aspect of the productivity-relevant em-ployee inputs depicted in Figure 1. The sanctionedmeasure is the sum of seven on-ice performancecomponents: goals, assists, plus/minus (the differ-ential between goals scored and allowed when theindividual being rated is on the ice and the teamscoring is not on a power play), power play andshorthanded goals and assists (we use goals here, aspower play and shorthanded assist data were un-available for all years; our amended formula pro-duced scores that correlated with the official mea-sure’s scores at .994 in 2002–04), penalty minutes,shots on goal, and defensive goals and assists(ESPN Fantasy, 2007). Each component was stan-dardized (within-year) prior to the addition of thecomponents. To account for injury and othersources of unreliability among inputs, we used atwo-year average (from years t – 1 and t – 2) whenthe data were available. We used the within-team-year mean of individual inputs to create team-yearinputs (i.e., the team-year level of productivity-relevant employee inputs). Inputs correlated withraw salary at .69 at the individual level and .79 atthe team-year level. Year t – 1 inputs are used topredict year t performance.

Pay variance. We used pay variance in eachteam-year observation to operationalize pay disper-sion. Unlike some previously used dispersion mea-sures, variance does not explicitly factor out themean and is thus consistent with our emphasis onavoiding the partialing out of mean pay effects (asnoted in the Discussion section, alternative disper-sion measures yielded results highly similar tothose found with the variance operationalization).

Dispersion in explained pay (DEP). To estimateDEP, we used an individual-level, league-wide re-gression of logged (due to extreme positive skew)Consumer Price Index–adjusted pay on individual-level performance inputs. The regression equationused was:

Yit � Pit � 1A � XitB � eit, (1)

where Y is a vector of CPI-adjusted and loggedpay-level observations for player i in year t, A andB are regression coefficient vectors, X is a matrix ofdummy variables representing years, e is an errorterm reflecting the residual for player i in year t,and P is a matrix of values from the individual-level inputs measure and its square from year t – 1.

The value of R for this equation was .72, and theR2 was .52, indicating that just over half of thevariation in individual pay level in the population


was a function of the variation in the observableperformance data captured in the individual-levelinputs measure (and year effects). Each player’spredicted value of pay level (i.e., yit) from thisleaguewide regression represents his expected pay,given how the entire market rewards the observableinputs in the equation. Thus, within-team variationin these predicted pay values is variation in the paythat can be explained by observable productivity-relevant data. Consequently, our measure of pre-dicted DEP is simply the variance of the predictedvalues of individual pay level for all players on theteam-year observation (i.e., �2

yit).

Dispersion in unexplained pay (DUP). To mea-sure DUP, we again made use of the leaguewideindividual-level regression described in Equation1. Each residual from the analysis is the individualpay that is independent of the productivity-rele-vant player observables. Thus, the variance of theseresiduals (within team-year) represents pay disper-sion that is unexplained by these data. Because thisvariance of player residuals is dispersion in unex-plained pay, we refer to this measure as residualDUP.

DEP and DUP from model specification. Paydispersion used to secure productivity-relevantemployee inputs is, by definition, DEP. Conse-quently, in the mediation models in which wetested the indirect (sorting) effect of pay variancethrough employee inputs, this indirect effect is aDEP effect, but the remaining direct effect is a DUPeffect. In keeping with our critique of the modelingof pay dispersion in earlier research, we also as-sessed DUP via our independent variable combina-tions in our analyses. Controlling for pay strategiesand productivity-relevant employee inputs leavesDUP as the unpartialed component of pay variance(see Table 2 and Figure 1).

Pay-for-performance strategy. We also corre-lated, within each team-year, individual inputsfrom year t – 1 and logged individual pay from yeart to create the pay strategy variable, pay-for-perfor-mance (see Pfeffer and Langton [1993] for a similarmeasure of pay-for-performance). We used loggedsalary in these team-year correlations to account foran exponential increase in pay returns as individ-ual inputs increase.

TABLE 2Measurement and Modeling of Key Constructs

Variable Description of Team-Year Measurement/Modeling

Productivity-relevant reasons for paydispersion

Inputs Mean, within team-year, of individual performance inputs from the NHLPA measure.a

Overall pay dispersionPay variance Variance, within team-year, of player salary (raw CPI-adjusted dollars).

Dispersion in explained pay (DEP)Pay variance The pay variance term becomes dispersion in explained pay when it shares variation

with (is mediated by) productivity-relevant inputs; hence, the pay variance thataffects performance indirectly through inputs is actually dispersion in explained pay.

Predicted DEP Variance, within team-year, of predicted value from individual-player regression oflogged salary on productivity-relevant inputs.

Dispersion in unexplained pay (DUP)Pay variance The pay variance term becomes dispersion in unexplained pay when strategically

relevant reasons for pay dispersion are partialed out; hence, the pay variance term inregression models that include these reasons as covariates actually representsdispersion in unexplained pay. The pay variance term also becomes dispersion inunexplained pay when it shares no variation with (is not mediated by) productivity-relevant inputs; hence, the direct pay variance effect, when inputs are included inthe model, is a dispersion in unexplained pay (DUP) effect.

Residual DUP Variance, within team-year, of residuals from individual-player regression of loggedsalary on productivity-relevant inputs.

Pays strategies other than paydispersion

Pay level Mean, within team-year, of player salary.Pay-for-performance Within team-year correlation between player pay and individual inputs from the

NHLPA measure.

a Individual performance inputs are from a measure of player value that was officially approved by the National Hockey League Players’Association (NHLPA) for use in fantasy hockey league play. This sanctioned measure is the sum of seven standardized (within-year) on-icecomponents.


Pay level strategy. Mean pay level is the averagesalary for individual players within each team-yearand was used to measure pay level strategy. Beforeaveraging, we adjusted the salaries for the CPI, leav-ing all salaries in 1998 dollars.

Analysis

Dependent variable distributions. Our pointsdependent variable is relatively normally distrib-uted, but the rounds dependent variable is distrib-uted as event count data. The Cameron and Trivedi(1986) regression-based test for overdispersion in-dicated that the conservative negative binomial re-gression model, rather than Poisson regression, wasappropriate for predicting round.

Dependent observations. Because we needed toaccount for the fact that teams appeared in the dataset an average of 5.83 times, we used random- andfixed-effects models to reduce concern that anyunmeasured organization-level variable could bedriving both team pay and team performance. Ran-dom-effects models produce matrix-weighted aver-ages of the between-unit and within-unit effects,whereas fixed-effects models, by definition, partialout stable between-unit differences, leaving the re-gression coefficients as estimates of purely within-unit effects (e.g., Wooldridge, 2002). Thus, ourfixed-effects analyses yield within-team effects ofpay dispersion, with the coefficients indicating theexpected increase in a team’s performance whenthat team increases pay dispersion by one unit (i.e.,a fixed-effects model is essentially equivalent toestimating a separate regression of team perfor-mance on pay dispersion for each team, using themultiple years of data as observations, and thenaveraging these separate regressions’ dispersion ef-fects to obtain the fixed-effects estimates). We pres-ent both random- and fixed-effects estimation foreach of our models and, as will become evident, thepattern of results is quite robust to the tech-nique used.

Our fixed-effects (within-team) analyses evoke,but must be interpreted differently than, the with-in-individual analyses in Lazear’s (2000) influen-tial individual-level study of pay-for-performance’ssorting and incentive effects on productivity. Un-der the assumption that ability was stable overtime, Lazear added individual-specific dummyvariables (the equivalent of a fixed-effects analysis),thus controlling out stable between-individual abil-ity differences; this was interpreted as controllingout pay-for-performance’s sorting effects. The re-maining pay-for-performance effect (i.e., the pay-for-performance coefficient after controlling outability) was interpreted as the impact of a change in

an individual’s motivation brought on by a changeto his/her pay-for-performance status. In contrast,our analysis is at the team level, where rather thanremaining constant, each team’s “ability” (i.e.,player inputs) changes from year to year (46 per-cent of the year t variation in team inputs was leftunexplained by year t – 1 inputs). Thus, althoughstable individual differences were partialed out viafixed effects in Lazear’s study, stable team differ-ences, which do not include the changes in teaminputs, are partialed out here. That is, within-teamsorting effects are not partialed out from pay dis-persion coefficients via our use of fixed-effectsmodels.

Actually, it is our mediation analyses, in whichwe do partial out team inputs completely, that par-allel Lazear’s individual-level fixed-effects ap-proach. In both cases, the ability/inputs effect (sort-ing) is partialed, and the pay strategy coefficient isconsiderably reduced. In our study, the pay disper-sion effect essentially disappears when inputs arecontrolled, indicating sorting, rather than incen-tives, as responsible for positive pay dispersioneffects on team performance.

Mediation versus moderation. We used media-tion to isolate the effects of pay dispersion thatoperate either via inputs (i.e., DEP effects) or inde-pendent of inputs (i.e., DUP effects). A moderationapproach would have been less appropriate in thatit would tell us about the effects of overall disper-sion when a moderator is high or low, potentiallyindependent of dispersion’s actual explanations.For example, pay dispersion, even under high pay-for-performance, could still be largely attributableto other factors, as performance is often less impor-tant than factors such as seniority in the predictionof pay (Bishop, 1987; Medoff & Abraham, 1980).

Generated regressors. In some models, we usedresiduals and predicted values from an individual-level pay level regression (see Equation 1 above) inour attempts to isolate DEP and DUP effects insubsequent regressions. Because such generatedfirst-stage terms are subject to error when esti-mated, standard errors tend to be underestimatedin the second analysis (Pagan, 1984). Consequently,to account for this generated regressor bias, for allmodels in our primary analyses that included thegenerated predicted DEP and residual DUP terms,we used a bootstrapping technique to obtain appro-priate standard errors (e.g., Efron & Tibshirani,1993; Mooney, 1996; Stine, 1990). Specifically, weresampled the data 1,000 times, with replacement(by cross-sections to account for the panel struc-ture); in each of the 1,000 bootstrapped samples,we conducted both the stage 1 and stage 2 regres-sions. We then used standard deviations from the


stage 2 sampling distributions of regression coeffi-cients to provide appropriate stage 2 standard er-rors. This commonly used method is asymptoti-cally valid and has recently been shown to performacceptably when the number of clusters (teams inour study) approaches 30 (Cameron, Gelbach, &Miller, 2008), which is the number of teams inour data.

RESULTS

Our framework recognizes that organizationsmay implement decisions on pay from a strategicperspective (so as to better attract, retain, and mo-tivate talented employees). We can draw inferencesabout whether pay is in fact implemented strategi-cally from the stability of certain pay practices overtime. That is, relatively stable pay practices overtime would suggest that these practices reflect stra-tegic policy (Gerhart & Milkovich, 1990; Mintzberg,1978), rather than random fluctuation. Conse-quently, we computed the year-to-year correlationsin pay practice for several of our measures. Themean year-to-year correlations for pay variance(.86), pay level (.84), pay-for-performance (.51), andpredicted DEP (.69) support the idea that organiza-tions are consistently following particular policieswith regard to pay setting. Because productivity-relevant employee inputs should then reflect whatwe believe to be strategic approaches to pay setting,we should also see evidence of stability in thismeasure. Indeed, the mean year-to-year correlationfor employee inputs was .74.

Means, standard deviations and correlations arepresented in Table 3. Although we standardizedthe independent variables in all regression analysesto aid in interpretation, the Table 3 means andstandard deviations are in the original metrics.

There is a strong positive relationship between payvariance and pay-for-performance (r � .45), andbetween pay variance and mean pay level (r � .86).This is consistent with our contention that the twopay strategies and pay dispersion are jointly tied tothe exact same aggregated individual pay level de-cisions. Similarly, Table 3 reveals strong positiverelationships between pay variance and inputs (r �.65). This relationship is quite consistent with oursorting explanation of pay dispersion benefits. Incombination, these correlations suggest that oursample is characterized by a high degree of strate-gically allocated pay.

Pay Dispersion and Sorting (Hypothesis 1)

In Hypothesis 1, we predict that DEP will yieldadvantages in the attraction and retention of theproductivity-relevant inputs that subsequentlyfacilitate team performance. At the team level,three types of evidence support this hypothe-sized sorting advantage: the .68 correlation be-tween predicted DEP and inputs; the multivariateregression, which is one step in the Hypothesis2b mediation analysis and reveals a large posi-tive, statistically significant effect of overall payvariance on inputs; and the mediation resultsthemselves, in which overall pay dispersiondrives team performance through inputs (see theHypothesis 2b results). Additionally, becausesorting is an organization-level perspective thatis derived from aggregated individual-level em-ployee movement outcomes, particularly reveal-ing data are accessible at the individual playerlevel. To focus on the sorting of players of vary-ing value, we examined players in the upperquartile, the lower quartile, and the middle 50percent on our individual-level measure of player

TABLE 3Descriptive Statistics and Correlations for Team-Level Variablesa

Variableb Mean s.d. 1 2 3 4 5 6 7

1. Points 85.75 15.962. Round 1.06 1.31 .703. Pay variancec 1.90 1.83 .40 .264. Pay level strategy 1.35 0.49 .55 .32 .865. Pay-for-performance strategy 0.67 0.18 .31 .16 .45 .436. Inputs 0.57 1.37 .65 .43 .65 .79 .407. Predicted DEP 0.29 0.15 .44 .32 .72 .62 .55 .688. Residual DUP 0.28 0.13 .15 .12 .34 .38 -.33 .29 .18

a n � 175; correlations greater than .15 are significant at p � .05.b Although we standardized the independent variables in all regression analyses, the Table 3 means and standard deviations are in the

original metrics. Pay variance is in hundred billions; pay level strategy is in millions; predicted DEP and residual DUP are variances ofpredicted values and residuals from the prediction of logged pay.

c Because pay variance is in raw dollars, it does not equal the sum of the means of predicted DEP and residual DUP. The mean payvariance of logged pay, however, is .57, which is the sum of the means of predicted DEP and residual DUP.


inputs. All players continuing in the league fromone year to the next were characterized as eitherstayers (remained with the prior year’s team) ormovers (went to a new team via a trade or as a freeagent). Mover and stayer inputs were measuredas of year t – 1, and the DEP levels of the teams onwhich these movers and stayers flowed to weremeasured as of year t, thus allowing us to assesswhether low-, average-, and high-input playerstended to flow differentially to and from teamswith more DEP.

Table 4 reveals the input-specific movement pat-terns of movers, stayers, and total players. Each ofthe three breakdowns yields an overall chi-squarevalue that is statistically significant, meaning thatyear t player input level is related to whether play-

ers flowed to high-DEP or low-DEP teams in year t� 1. The nature of the relation is revealed throughthe percentage of players in each cell and througheach row’s contribution to the overall chi-square.These row contributions in each breakdown indi-cate that the high-input players (i.e., high perform-ers) are primarily responsible for the overall chi-square value and statistical significance; this high-input player sorting accounted for approximately71 percent (for stayers; i.e., 45.2/63.9), 79 percent(for movers), and 75 percent (for all players) of theoverall chi-square value that itself indicates a rela-tionship between DEP and player inputs. Focusing,then, on the high-input row, Table 4’s stayers anal-ysis (top) shows that 419 (65%) of the high-inputstayers spent the next year on teams above the DEP

TABLE 4Player Acquisition and Retention by Player Inputs (Performance) and Team Levels of Dispersion in Explained Paya, b

Where Year t – 1 Players That Stay Play Next YearYear t Teams Ratio of Stayers

at High DEP toStayers at Low

DEPRow Contribution toOverall Chi-Square

Low-DEPTeams

High-DEPTeams

High-input stayers 226 (35.0%) 419 (65.0%) 1.85 45.2***Moderate-input stayers 614 (54.3%) 517 (45.7%) 0.84 16.4***Low-input stayers 274 (51.5%) 258 (48.5%) 0.94 2.2Total 1,114 (48.3%) 1,194 (51.7%)

Overall �2 63.9***

Where Year t – 1 Players That Move Play Next YearYear t Teams Ratio of Movers

to High DEP toMovers to Low


Low-DEPTeams

High-DEPTeams

High-input movers 65 (38.5%) 104 (61.5%) 1.6 13.1***Moderate-input movers 270 (54.4%) 226 (45.6%) 0.84 0.8Low-input movers 161 (57.1%) 121 (42.9%) 0.75 2.5Total 496 (52.4%) 451 (47.6%)Overall �2 16.5***

Where All Year t – 1 Players Play Next YearYear t Teams Ratio of Players

at High DEP toPlayers at Low


Low-DEPTeams

High-DEPTeams

High-input players 291 (35.8%) 523 (64.3%) 1.8 61.2***Moderate-input players 884 (54.3%) 743 (45.7%) 0.84 15.4***Low-input players 435 (53.4%) 379 (46.6%) 0.87 5.2*Total 1,610 (49.5%) 1,645 (50.5%)Overall �2 81.8***

a Overall chi-square is a test of the null hypothesis that there is no relationship between player inputs and the DEP level of next year’steam. Hypothesis 1 is more specifically addressed by the row chi-square tests for high-input players, which indicate rejection of the nullhypothesis that high-DEP and low-DEP teams do not differ in their attraction and retention of high-input players.

b DEP is dispersion in explained pay. DUP is dispersion in unexplained pay. High-DEP (-DUP) and low-DEP (-DUP) teams are thoseabove and below the median, respectively.

High-input players are those in the top quartile in their league. Moderate input players are those in the 25th–75th percentile. Low-inputplayers are those in the bottom quartile in their league.


median, which is 25.6 percent greater than thecell’s expected value if DEP and stayer inputs wereunrelated (i.e., 51.7 percent of the 645 high-inputstayers, which is 333.5); as the 1.85 (i.e., 419/226)in the ratio column indicates, this means that 85percent more high-input stayers remained withhigh-DEP teams than remained with low-DEPteams (see the Discussion section for the nuances ofthe high-DEP retention advantage). Similarly, formovers (Table 4, middle), 104 (61.5%) of the high-input movers spent the next year on teams abovethe DEP median, which is 29.2 percent greater thanthe cell’s expected value if DEP and mover inputswere unrelated (i.e., 47.6 percent of the 169 high-input movers, which is 80.4); here, the 1.60 in theratio column reveals that 60 percent more high-input movers went to high-DEP teams than went tolow-DEP teams. Finally, Table 4’s total playermovement analysis (bottom) mirrors the mover andstayer sorting component analyses as, overall, 80percent more high-input players played the follow-ing year on high-DEP teams than on low-DEP teams(moderate- and low-input players were slightlymore likely to play on low-DEP teams the followingyear). Clearly, the statistically significant relation-ship between DEP and player inputs is driven byhigh-DEP teams faring better than low-DEP teamsin the attraction and retention of high-input players(additional regression analyses confirmed thispoint, as the DEP of the team that a player moved toor stayed with had a positive and statistically sig-nificant effect on the input level of the acquired/retained player). In short, as predicted in our frame-work and in Hypothesis 1, DEP yielded clearsorting advantages in our data. Not surprisinglygiven the high level of explained pay in our data,these results were replicated when we substitutedoverall pay variance for the DEP measure.

DEP and Team Performance (Hypotheses 2a and2b)

Tables 5 (points) and 6 (round) provide the re-sults of tests of our hypotheses that DEP should, viasorting effects, lead to more effective team perfor-mance and should yield more favorable effects thanDUP. Models 3 and 8 in each table support Hypoth-esis 2a. Specifically, under both random- and fixed-effects models, predicted DEP (i.e., the withinteam-year variance in predicted values from theindividual-level regression of pay on prior year per-formance inputs) had a statistically significant,positive association with points and round. Forexample, the predicted DEP coefficients frommodel 3 in Tables 5 and 6 reveal, respectively, thata one standard deviation increase in predicted DEP

predicts 5.04 additional points, which is .32 of astandard deviation, and a .34 increase in loggedrounds advanced in the play-offs (a 40 percent in-crease in actual rounds). Turning to the relativesizes of DEP and DUP effects, we tested for statis-tical differences between the two. As indicated inthe Wald tests summarized at the bottom of Table5’s models 3 and 8, the predicted DEP coefficientwas larger than the residual DUP coefficient underboth the random-effects (5.04 versus –0.12) andfixed-effects (3.64 versus –0.93) approaches. Simi-larly, in Table 6’s models 3 and 8, when predictinground, the predicted DEP coefficient was again sta-tistically greater than the residual DUP coefficientunder both random- (0.34 versus 0.04) and fixed-effects (0.28 versus –0.02) estimation. Thus, Hy-pothesis 2a was supported, as DEP effects on per-formance were positive and more favorable thanDUP effects.

Support for a pay dispersion effect through sort-ing also emerged in our mediation tests of Hypoth-esis 2b. Sobel (1982) tests consistently revealed astatistically significant, positive effect of pay dis-persion on team performance that operated throughthe securing of inputs. In three of the four media-tion scenarios, the nested models’ positive payvariance effects essentially completely disappearedonce inputs were incorporated as predictors. Forexample, in Table 5, the pay variance coefficientdeclined from 5.10 in model 1 to –0.05 when in-puts were added in model 2 (see also Table 5’smodels 6 and 7 and Table 6’s models 1 and 2). InTable 6’s models 6 and 7, the pay variance effectwent from near zero to negative when the inputsmediator variable was included, still revealing apositive indirect effect (see Mackinnon, Lockwood,Hoffman, West, & Sheets [2002] on how indirecteffects can be present in the absence of total ef-fects). The indirect effects are calculated by multi-plying the mediator (inputs) effect by the effect,from a separate (unreported) regression, of pay vari-ance on inputs. In terms of effect size, a one stan-dard deviation increase in pay variance leads toinput enhancements that ultimately translate, un-der random- and fixed-effects specifications, to5.39 points and 3.58 points, respectively, and to0.40 and 0.35 increases in logged rounds advancedin the play-offs, respectively (the latter two effectsrepresent 49 and 42 percent increases in actualrounds).

In terms of testing the relative DEP and DUPeffects through mediation, as described in Hypoth-esis 2b, we compared the indirect (DEP; sorting)effect of pay variance through inputs to the directpay variance (DUP) effect. Indirect effects appear tobe more favorable than direct effects under all con-


ditions, though available estimation techniqueslimited our testing of statistical differences to themore conservative fixed-effects models. Wald testsrevealed that pay variance produced indirect (DEP)effects through inputs that were statistically greaterthan direct (DUP) effects (3.58 versus –0.05 whenpredicting points and 0.35 versus –0.19 when pre-dicting round). Thus, we found support for Hy-pothesis 2b, as mediation analysis indicated thatoverall pay dispersion that secures productivity-relevant inputs produces more favorable effects onperformance than does DUP.

In sum, whether predicting points or round, andwhether using the mediated approach or the moredirect measure of predicted DEP, DEP was posi-tively related to team performance, and it was morefavorably so than DUP. Both approaches indicatethis is a manifestation of pay dispersion’s sortingadvantages when paying for productivity-relevantinputs.

Partialing Out Covariates and Dispersion inUnexplained Pay (Hypothesis 3)

In several tests of Hypothesis 3, we found thatcontrolling for pay level strategy, pay-for-perfor-mance strategies, and productivity-relevant in-puts removed the positive pay dispersion effectson team performance; presumably this occurredbecause what remained of pay variance was DUP,which resulted in a negative or negligible paydispersion effect. For example, whereas an in-crease of one standard deviation in pay variancein Table 5’s models 1 and 6 predicts statisticallysignificant increases of 5.10 and 3.76 points,models 5 and 10 reveal that the same pay vari-ance increase predicts statistically significant de-creases of 5.16 and 5.89 points when pay leveland pay-for-performance strategies were con-trolled. When inputs are added to the baselinemodels 1 and 6, the pay variance effects essen-

TABLE 5Regression Analyses of the Effects of Pay Dispersion Types and

Strategically Relevant Reasons for Pay Dispersion on Pointsa

Variable

Random Effects Fixed Effects

1 2 3 4 5 6 7 8 9 10

Constant 85.52*** 85.70*** 85.57*** 87.43*** 84.83*** 85.75*** 85.75*** 85.78*** 85.25*** 85.00***(1.94) (1.28) (1.95) (1.84) (1.74) (0.00) (0.00) (2.05) (1.94) (0.19)

Pay varianceb 5.10*** –0.05 –5.16* 3.76† 0.05 –5.89*(1.66) (1.71) (2.38) (1.85) (1.48) (2.76)

Predicted DEP 5.04** 6.38*** 3.68* 4.74**(1.71) (1.84) (1.67) (1.88)

Predicted DEP squared –1.80* –1.47*(0.80) (0.88)

Residual DUP –0.12 –0.31 –0.93 –1.03(1.36) (1.34) (1.36) (1.39)

Inputs 9.22*** 6.84***(1.24) (1.46)

Pay level 10.75*** 10.22***(2.20) (2.60)

Pay-for-performance 2.07† 2.03(1.22) (1.26)

R2 .16 .43 .19 .23 .33 .16 .43 .16 .21 .32�2 9.40** 69.17*** 8.84* 12.38** 32.91*** 4.92† 6.55†

F 4.14† 13.64*** 7.36***bDEP � bDUP ?c Yes Yes Yes Yes

a Robust standard errors are in parentheses for all but models 3, 4, 8, and 9, in which we used bootstrapped standard errors to accountfor generated regressor bias (bootstrapped and random-effects models yield model fit statistics in terms of �2 rather than R2); allindependent variables were standardized prior to the regressions. For all models, n � 175.

b As described in Table 2, pay variance becomes DEP when it affects points indirectly through inputs (see the Results section for theindirect effects calculated from models 1 and 2 and from models 6 and 7). Because of the covariates modeled, pay variance becomes DUPin models 2, 5, 7, and 10.

c “Yes” indicates a statistical difference (p � .05) between predicted DEP and residual DUP; tests were conducted in models 3, 4, 8,and 9.

† p � .10* p � .05

** p � .01*** p � .001


tially go to zero (see models 2 and 7). A similargeneral pattern emerges in the round models. Inall cases, the pay variance coefficient when paystrategies or inputs were present as covariateswas statistically different from the coefficientwhen pay strategies or inputs were absent.

These Hypothesis 3 findings illustrate the enor-mous impact of covariate decisions on whetherDEP or DUP is modeled, as well as the stark con-trast between DEP and DUP effects on performance.The results support not only Hypothesis 3, but alsoour interpretation of the Table 1 research. Control-ling for relevant inputs or the pay strategies de-signed to produce them partials out DEP from payvariance and largely leaves DUP to predict perfor-mance. Thus, negative pay dispersion effects insuch scenarios should not implicate overall paydispersion, or DEP in particular, as detrimental.

Curvilinearity (Hypothesis 4)

Table 5’s models 4 and 9 indicate that DEP has acurvilinear effect on team performance, as the

squared predicted DEP term, as predicted, is nega-tive and statistically significant when predictingpoints. Plotting the relationships indicates thatpositive predicted DEP effects not only diminish insize as predicted DEP increases, but also becomenegative as the curve turns slightly downward atthe higher predicted DEP levels in the data. Addi-tional analysis, however, qualifies this inference.Following the procedure in Aiken and West (1991)for analyzing curvilinearity, we computed severalsimple slopes (i.e., the predicted DEP linear effects,which are tangents to the curvilinear plot, at spec-ified values of predicted DEP) and their standarderrors. For Table 5’s model 4, this produced sta-tistically significant predicted DEP effects of13.08 when predicted DEP is at its minimum(–1.85 s.d.), 10.00 when at negative 1 s.d., 6.38when at the measure’s mean, and 2.76 at plus 1s.d.. But at values of predicted DEP that were atplus 1.5 s.d. and above, the predicted DEP effectswere not statistically significant, though they dobecome negative at 2 s.d. and above. Curvilinearity

TABLE 6Regression Analyses of the Effects of Pay Dispersion Types and Strategically Relevant

Reasons for Pay Dispersion on Rounda

Variable

Random Effects Fixed Effects

1 2 3 4 5 6 7 8 9 10

Constant 1.37* 1.16* 1.55 1.67 1.25* 1.64* 1.22* 1.55 1.69 1.56†

(0.65) (0.58) (4.37) (4.32) (0.63) (0.81) (0.62) (4.62) (4.86) (0.83)Pay varianceb 0.22* –0.12 –0.23 0.04 –0.19 –0.43†

(0.11) (0.12) (0.19) (0.15) (0.15) (0.24)Predicted DEP 0.34*** 0.52*** 0.28* 0.40*

(0.10) (0.14) (0.12) (0.17)Predicted DEP squared –0.16† –0.11

(0.12) (0.10)Residual DUP 0.04 0.01 –0.02 –0.04

(0.13) (0.12) (0.12) (0.12)Inputs 0.68*** 0.67***

(0.13) (0.16)Pay level 0.47** 0.49*

(0.17) (0.21)Pay-for-performance 0.15 0.15

(0.12) (0.14)�2 4.17* 30.92*** 12.21** 14.60** 13.59** 0.07 16.59*** 5.35† 5.64 6.59†

bDEP � bDUP ?c Yes Yes Yes Yes

a We used bootstrapped standard errors in models 3, 4, 8, and 9 to account for generated regressor bias; all independent variables werestandardized prior to the regressions; n � 175 for random-effects models, but n � 160 for fixed-effects models because we eliminated 15observations owing to teams’ all-zero values for round across years in fixed-effects models.

b As described in Table 2, pay variance becomes DEP when it affects points indirectly through inputs (see the Results section for theindirect effects calculated from models 1 and 2 and from models 6 and 7). Because of the covariates modeled, pay variance becomes DUPin models 2, 5, 7, and 10.

c “Yes” indicates a statistical difference (p � .05) between predicted DEP and residual DUP; tests were conducted in models 3, 4, 8,and 9.

† p � .10* p � .05

** p � .01*** p � .001


under fixed effects (model 9) followed a very sim-ilar pattern.

For the round regressions, the statistically signif-icant squared term in the random-effects modelalso indicated curvilinearity. Simple slope analysisof Table 6’s model 4 produces predicted DEP ef-fects of 1.10 when predicted DEP is at its minimum(–1.85 s.d.), 0.84 when at negative 1 s.d., 0.52 whenat its mean, and 0.37 at plus 0.5 s.d. At values ofpredicted DEP that were at plus 1 s.d. and above,the predicted DEP effects were not statistically sig-nificant, though, as in the points analysis, they dobecome negative at 2 s.d.’s and above . The squaredpredicted DEP term in the fixed-effects roundmodel was not statistically significant. We interpretthe round findings as supportive of curvilinearity,however, as the Hausman (1978) test indicated thatrandom-effects modeling, which is more efficient,was appropriate in this instance.

In sum, both the points and round analyses pro-duced an attenuated positive effect of predictedDEP on team performance. As predicted DEP in-creases, its positive effects diminish, eventually be-coming no different from zero (though negativein sign).

Supplemental Analyses with InstrumentalVariables

Although our primary analyses supported ourhypotheses, we also conducted supplementaryanalyses using instrumental variables to more con-servatively test causal direction. As instruments,we used metropolitan area population and metro-politan area income, which should (via incomestream, subsequent ability to pay, and the inputsultimately secured by pay) primarily affect teamperformance through pay and inputs. Becauselagged values of independent variables are fre-quently suitable as instruments (e.g., Blalock,1985), we also used one-year lags of the pay andemployee input variables. The two commonly citedrequirements for instrument suitability, associationwith the independent variable to be instrumented(once the other predictors have been partialed) andlack of correlation with the error term (e.g.,Wooldridge, 2002), generally appeared to be satis-fied here. These instrumental variable analyses(available upon request) generally supported ourpredictions of positive DEP effects on team perfor-mance, diminishing returns at high DEP, and lessfavorable DUP effects. Support was less consistentin round models, though round was expected to bemore difficult to predict because it is a less reliableindicator of team performance than is points.

DISCUSSION

In this study, we conceptually and empiricallydistinguished between dispersion in the pay ex-plained by productivity-relevant inputs and disper-sion in the pay left unexplained by such inputs. Wecouched this distinction in a sorting explanation inwhich, contrary to conventional wisdom, pay dis-persion when work is interdependent can facilitateteam performance. In support, and robust to a va-riety of empirical approaches (e.g., different vari-able operationalizations, estimation via fixed-ef-fects or random-effects techniques, addressing DEPeffects with mediation of pay dispersion effects viainputs or with direct measurement, the use of in-strumental variables), DEP (dispersion in explainedpay) in a highly interdependent work context pro-duced sorting advantages that resulted in team per-formance gains.

Pay Dispersion and Interdependence

The conceptual approach and empirical findingspresented here are not only new, but are in starkcontrast to the prevailing wisdom, as interdepen-dence is virtually always cited as the context inwhich pay dispersion will be particularly disrup-tive (e.g., Akerlof & Yellen, 1988; Bloom, 1999;Levine, 1991; Pfeffer & Langton, 1993; Pfeffer, 1994;Shaw et al., 2002). Indeed, we were unable to find asingle example of a hypothesized positive effect ofpay dispersion on performance in an interdepen-dent work setting. Empirical work on pay disper-sion, however, has focused primarily on indepen-dent settings, or has not addressed the issue. Forexample, of the 23 studies summarized in Table 1,we identified 2 as from high-work-interdependencesettings, 7 as from low-interdependence settings, 1that used samples from each type of work setting,and 13 in which interdependence was unknown.Some researchers have asserted that they tested paydispersion effects under interdependence but have,in fact, used only settings with unknown or lowlevels of interdependence. For instance, Bloom(1999) argued that his baseball performance findingunderscores the problem with pay dispersion ininterdependent work. However, we note that base-ball is a sport characterized by pooled interdepen-dence (Foster & Washington, 2009; Keidel, 1987),which is the least interdependent form of work asdescribed in Thompson’s (1967) hierarchical order-ing. Keidel further described baseball player inter-action as minimal and baseball itself as “a meta-phor for the autonomy of organizational parts”(1987: 592). In another attempt to (empirically) ex-amine the role of interdependence (Shaw et al.,


2002, study 2), the “high” interdependence condi-tion was actually not very high. Interdependencewas measured as “extent of use of self-managedteams.” The response options were 1 (“none”), 2(“almost none, 1–20%”), 3 (“some, 21–40%”), 4(“about half, 41–60%”), 5 (“most, 61–80%”), 6(“almost all, 81–99%”), and 7 (“all, 100%”). Theoverall mean score in the sample was 1.45. Thus,“high” interdependence (mean � 1 s.d.), 2.29, fellclosest to response option 2, “almost none.” Simi-larly, in two studies that failed to find evidence that“interdependence” moderated dispersion effects(Eriksson, 1999; Main et al., 1993), the work itselfwas never described. In contrast, one aspect of oursample that made it of considerable interest here isthat hockey is a highly interdependent work setting(Beauchamp & Bray, 2001), as hockey team perfor-mance is characterized by reciprocal interdepen-dencies (Foster & Washington, 2009; Frey et al.,1986; Gerhart & Rynes, 2003) in which the outputsof individual members become the inputs for othermembers and vice versa. Thompson (1967) charac-terizes such reciprocal interdependence as themost interdependent form of work.

Although the prevailing view of pay dispersionas detrimental in interdependent work contextsmay be a reasonable position when the disper-sion in question is in unexplained pay (DUP), forwhich equity theory predicts such problems, DEPappears unlikely to yield such disruption. In-deed, interdependent work may even allow em-ployee perceptions of coworker inputs to be moreaccurate (via greater observability), further re-ducing the likelihood of disruption due to DEP.Moreover, it may be that most team membersunderstand that success hinges on having tal-ented team members, which hinges on paying toattract and retain them. In sum, given our con-ceptual framework, the lack of prior work on paydispersion in truly interdependent settings, andour findings of DEP’s positive effects in an inter-dependent context, we believe the common char-acterization of interdependent work as incompat-ible with pay dispersion is unwarranted, unlessexplicitly addressing DUP.

Sorting Benefits of Pay Dispersion

The incorporation of sorting, a fundamental ave-nue for examining pay effects (Gerhart & Rynes,2003; Lazear, 2000), into the dispersion-perfor-mance debate informs the issue and maintains anemphasis on inputs. Our findings demonstrate apotent sorting role in explaining dispersion bene-fits to team performance, as well as in understand-ing what differentiates DEP from DUP. Hence, we

strongly encourage pay dispersion researchers toaddress sorting, which may well be the primarydeterminant of dispersion’s effect. Indeed, the ab-sence of a sorting role in prior pay dispersion worklikely contributed to the disparate results andoverly negative view of pay dispersion, even whenwork was independent in nature.

Our sorting findings indicate that, generally,high-DEP teams are better at retaining and attract-ing the talent that ultimately drives team perfor-mance. A closer look at the retention side, however,yields a more nuanced inference. High-DEP teamsretain many more high-input players than do low-DEP teams, but they actually retain only a slightlyhigher proportion of them (.80 for high-DEP teams;.78 for low-DEP teams). Thus, the retention advan-tage is that high- and low-DEP teams retain similarproportions of differently sized talent pools, ashigh-DEP teams tend to have more high-input play-ers to begin with.

In addition (or in contrast) to our position on DEPyielding sorting advantages, it might also be arguedthat simply paying well, and equally well (i.e., highpay, low DEP), for all talent (from a pay level per-spective) or for all performance (from a pay-for-performance perspective) would yield sorting ad-vantages as well. This approach, however, wouldcreate considerable labor cost problems, requireteams or organizations to be comprised of onlymembers with equally high talent or equally highperformance (otherwise, high pay for all necessi-tates DUP), and run contrary to the strong tendencyfor teams and organizations to be designed aroundroles that contribute unequally to success (e.g., inorganizational hierarchies, in surgical teams, andin law firms). Thus, we see little practical or con-ceptual reason to doubt the contention that highpay for talent and pay-for-performance suggest pos-itive DEP effects on sorting, rather than uniformlyhigh pay effects.

Sorting and Incentives

Although we have grounded this study in em-ployee sorting, with incentives conspicuouslydownplayed, we suggest that sorting is often con-sistent with the incentive approach. Expectancytheory (Vroom, 1964) stipulates that two majorinfluences on motivation to perform are valence(the attractiveness of a reward) and instrumental-ity (the perceived likelihood that performancewill be rewarded). DEP’s larger pay returns toproductivity-relevant employee inputs should re-sult in not only sorting, but also in enhancedmotivation via greater valence (assuming that alarger payoff is more attractive) and a clarified


line of sight (instrumentality) between such in-puts and pay. In our sample, however, control-ling for sorting (i.e., partialing the prior year’s[year t – 1] individual performance inputs) “con-trols out” (i.e., mediates) virtually the entire pos-itive pay dispersion effect (e.g., see models 1 and2, Table 5; and models 1 and 2, Table 6). Becausethe positive pay dispersion effect essentially dis-appears when players’ prior performance inputsare controlled, sorting, rather than incentives, isprimarily responsible for positive pay dispersioneffects on team performance in our study.

Another potential role of incentives in a sortingscenario emerges when considering more closelythe member characteristics of those that self-se-lect into teams. It could be that high-DEP teamsare more attractive to players who are more at-tuned to the pay-for-performance aspects of high-DEP environments. For example, as a result oftheir heightened attention to the proportionalityof rewards and inputs, players more sensitive topay equity may tend both to be more likely toself-select into high-DEP scenarios and to bemore motivated by the high instrumentality. Con-sequently, DEP could yield incentive benefits inaddition to sorting advantages (though our con-text yielded little evidence of such incentive ef-fects). To the degree, however, that those high inequity sensitivity also are high in raw ability, asorting explanation for DEP’s benefits becomesmore nuanced, as high performers are attracted tohigh-DEP teams, but we cannot necessarily attri-bute the presorting performance to ability ratherthan motivation. Thus, although it is clear thatDEP led to team performance through the sortingof prior high performers to current high-DEPteams (rather than through incentives per se, asexplained above), we cannot be certain of therelative contributions of ability and motivation tothe individuals’ prior performance.

The sorting and incentives synthesis suggestedhere is further complicated, and generalizability ispotentially constrained, under at least two condi-tions. First, recent research indicates that, for equi-ty-sensitive individuals, pay secrecy constrainsperceived instrumentality, thereby reducing taskperformance (Bamberger & Belogolovsky, 2010).Thus, for those individuals otherwise likely to self-select into high-DEP scenarios under highly visiblepay structures, a lack of transparency in payamounts may make high DEP less attractive (from asorting perspective) and less motivating. Second,the sorting and incentives synthesis is considerablymore viable when individual contributions areidentifiable. Absent this visibility, instrumentali-ties become less clear, and inequity perceptions

may increase. Organizations also then may find itmore difficult to align individual incentives withgroup objectives. Ultimately, we encourage a morebalanced view of pay dispersion that recognizesDEP’s sorting advantages, its potential incentivebenefits, and, when individual contributions can-not be identified or pay amounts are unknown, itspotential incentive liabilities.

An additional concern with a lack of some degreeof individual performance visibility is that it mayopen the door for the emergence of incentives forcounterproductive behaviors. The pay dispersionliterature we reviewed, as well as other work (Am-brose, Seabright, & Schminke, 2002; Latham & Pin-der, 2005), indicates that the main cause of coun-terproductive team behaviors, such as sabotage ofteam members or lack of cooperation, is perceivedinequity (which DEP, except perhaps when at ex-treme levels, should preclude). We acknowledge,however, that such behaviors can also result fromraw self-interest and/or incentives that strongly in-centivize only individual performance, rather thanteam performance. However, even as performancevisibility diminishes, it often may be unlikely thatthe conditions necessary for such problematic be-haviors to occur exist. These enabling conditionsinclude (Alchian & Demsetz, 1972; Prendergast,1999) a lack of penalties for and an inability toobserve such counterproductive behaviors (inter-dependent work will tend to allow team membersto be aware of each others’ actions, making behav-iors such as anonymous sabotage unlikely); littleimportance of reputation among peers; a significantlack of goal alignment between team members andthe organization; and a true zero-sum situation inwhich the performance and/or pay of one teammember comes at the expense of another (Kandel &Lazear, 1992). Most team settings, however, proba-bly do not satisfy these conditions very well. Inmost settings in which significant interdependenceexists, the success of one team member does notcome at the expense of another. The developmentand timely launch of a successful new product, forexample, will reflect well on all product teammembers, and success is most likely when every-one on the team performs at a high level; hence,even with some difficulty in identifying individualperformance contributions, DEP in such situationsshould not tend to result in counterproductive be-havior. Moreover, even if such behavioral incen-tives exist, teams have strong norms and expecta-tions of their members that can also act as powerfuldeterrents to behaviors that will harm the teamsand their members (Barker, 1993). Of course, care-ful consideration must always be given to how best


to balance individual-, team-, and organiza-tion-level incentives.

Context and Generalizability

Given our sample, considerable attention mustbe paid to various contextual and generalizabilityconcerns. Two key contextual elements whenconsidering dispersion’s effects are the degree ofpay-for-performance and, as discussed above, theidentifiability (i.e., measurability) of perfor-mance inputs, which have important implica-tions for sorting and pay system design, respec-tively. Professional hockey is very high on thepay-for-performance and individual contributionidentifiability dimensions. Hence, future re-search is needed on DEP and DUP effects whenthese two contextual dimensions are lower thanin our study.

Relatedly, although performance is the classicinput from an equity perspective (Steers & Porter,1983), other inputs that drive pay dispersion maywell be seen as productivity-relevant proxies andthus as acceptably strategic explanations for dis-persion (e.g., skills, tenure, formal certifications).It is likely, however, that the acceptability andvalue of dispersion lessen as the explanation forpay differentials increasingly deviates from ob-jectively assessed performance, as will occur asindividual performance contribution identifiabil-ity declines. And the greater this deviation, themore likely that the sorting, incentive, and equityarguments that we have attempted to synthesizewould diverge. Pay tied to seniority, for example,may be viewed as equitable if seniority is be-lieved to be a proxy for performance, but it mayneither attract and retain high performers norincentivize performance.

Additional generalizability questions arise be-cause professional athletes are extremely well paidpeople, have very short careers, and are among thevery best in the world at what they do. These threequalities in combination put them in a ratherunique position, relative to what we consider to bemainstream employees in commonly held jobs. Onthe other hand, in an economy in which individualtalent is increasingly at a premium (e.g., Frank &Cook, 1995; Michaels, Handfield-Jones, & Axelrod,2001) and a variety of occupations have star per-formers with star salaries (e.g., attorneys, consul-tants, executives, realtors, investment bankers, en-tertainers, salespeople), perhaps the situationoutside of sports is not always so different. In anycase, we welcome pay dispersion research that goesbeyond the sports realm to further our examinationof DEP and DUP. Moreover, although team perfor-

mance often defines organizational success in pro-fessional sports (Danielson, 2004), nonsports re-search can directly speak to DEP effects onfinancial outcomes by assessing whether DEP’sbenefits (i.e., sorting and incentive effects) out-weigh its costs.

Finally, we reiterate that our sample does havethe critical high work interdependence that muchof the prior pay dispersion research has eitherlacked or not addressed. Often cited as the contextin which pay dispersion will be particularly dis-ruptive, interdependence appears here to provideno constraint on pay dispersion’s potential to pro-duce, via sorting, positive effects on teamperformance.

The Pay Dispersion Construct and ModelingIssues

Our research suggests several meaningful con-ceptual and empirical considerations for pay dis-persion research that we have yet to fully address.An important concern here that has received littleattention elsewhere is the possibility of a curvilin-ear pay dispersion effect (see Brown et al. [2003] foran exception). Results suggest that, at least in oursample, at higher levels of dispersion in explainedpay (DEP), increases to this dispersion yield dimin-ishing returns and, ultimately, no additional per-formance advantages. Additional (unreported)analyses using mediation replicated this finding, aspay variance as a whole had a diminishing, thoughnever counterproductive, effect on points andround that operated through inputs. Future re-search into the generalizability of this curvilinearDEP effect is needed, as is exploration of the effect’sexplanation, which we speculated to be inequityperceptions or limitations in organizational capa-bilities to manage valued resources (see the deriva-tion of Hypothesis 4).

A second modeling issue is the operationaliza-tion of pay dispersion. We used pay variancebecause it does not factor out mean pay in anyway (and is thus consistent with our position onavoiding the partialing of mean pay effects indispersion modeling). We recognize, however,that most pay dispersion research has deployedalternative operationalizations that do, in theircomputation, account to some degree for meanpay. Hence, we reran our analyses after replacingpay variance with pay’s coefficient of variation(i.e., the standard deviation of pay divided bymean pay level) and the gini coefficient, twocommonly used pay dispersion measures thatcorrelate with pay variance in our data at .70 and


.74, respectively. These substitutions did notchange the overall pattern of results.

Third, productivity-relevant reasons for paydispersion, and thus the makeup of DEP andDUP, will rarely be known with complete cer-tainty. DUP measures probably will, to at leastsome degree, be associated with pay based onrational or logical pay elements that we do nothave data sophisticated enough to detect. Indeed,this may be one reason why DUP frequentlyyields zero, rather than negative, effects, both inthe extant research and in our study. Relatedly,even determining what exactly is “productivity-relevant” in principle and what is not is a sub-jective process likely to produce disagreementamong researchers. Despite this limitation, con-sistent results across the partialing, mediation,and predicted value/residual techniques lendsupport to our conceptual and empirical model-ing of DEP and DUP.

Finally, although our study involves lateral(within-job) dispersion, whether its logic general-izes to vertical (across-job) dispersion is also ofinterest. Certainly jobs vary tremendously in pro-ductivity implications, as evidenced by enormousdifferences in job pay. Because the job held is theprimary driver of pay (Gerhart & Milkovich, 1992),job differences may even surpass job performanceas the quintessential productivity-relevant expla-nation for pay dispersion, meaning that partialingout these differences could change overall pay dis-persion or DEP effects to DUP effects. Overall, theTable 1 studies appear to confirm this contention,as vertical pay dispersion and performance tendedto be positively related in studies in which aspectsof the job were not partialed out, but tended to beunrelated or negatively related in studies in whichthese job inputs were partialed. Hence, we encour-age researchers to consider the DEP/DUP implica-tions in vertical, as well as lateral, pay dispersionresearch.

Implications for Practice

One reaction to our results would be to pre-sume that we advocate pay dispersion. First, wereiterate that pay dispersion makes sense only tothe extent that it represents DEP; positive DEPeffects are what our conceptual framework andempirical analyses support. Second, it would beshortsighted to suggest that even DEP is the cor-rect strategic approach in all situations. Such asweeping recommendation conflicts with our be-lief that context is vital in terms of the efficacy ofhuman resource practices in general and pay

practices in particular. For instance, the SAS In-stitute is an intriguing, albeit rare, example of ahighly successful company that has been identi-fied as a proponent of a less-dispersed pay systemthat appears to effectively support its businessstrategy and company culture (Gerhart &Rynes, 2003).

That said, however, absent such a fit or contin-gency model that calls for the use of less paydispersion, we do believe that DEP, by virtue ofsorting (and incentive) benefits, is often a goodidea. Of course, DEP is not always viable. Thoughpay-for-performance systems often effectivelysort and motivate, they also can be fraught withproblems, particularly when individual perfor-mance differences are difficult to accurately inferand/or measure credibly in the eyes of employ-ees. Without such measurability, as well as theperception thereof, the rationales for the sortingand incentive benefits quickly disintegrate. Thus,possessing a high-quality performance assess-ment system and convincing employees of itsvalidity are primary concerns for organizationsfollowing a high-DEP approach. Additionally,when considering the investment into more DEP,management should be sure to consider wherethey already are with regard to the construct. Ourcurvilinear analyses indicate that for companiesalready high on the construct, more DEP maygarner few or no organization-level advantages.

Conclusions

In their 1993 study of pay dispersion effects,Pfeffer and Langton (1993: 382) quoted Barnard’s(1938: 145–146) statement that differentials inmoney “are a source of jealousy and disruption ifnot accompanied by other factors of distinction.”We believe that many scholars addressing paydispersion effects in interdependent settings mayhave focused on Barnard’s “source of jealousyand disruption” idea without adequate regard forhis “accompanying factors of distinction” contin-gency. Relative inattention to inputs (i.e., Bar-nard’s factors of distinction) in pay dispersionresearch is arguably at the heart of what we havecharacterized as the literature’s conceptual andempirical confounding of inequity and inequal-ity. In response, our framework’s input-baseddistinction between DEP and DUP provides note-worthy resolution to inconsistencies in prior em-pirical work and to ostensibly conflicting theo-retical perspectives. Specifically, incorporating asorting-based focus and isolating DEP, while lim-iting inequity concerns to DUP, ultimately leavesus with a considerably more favorable view of


pay dispersion in interdependent work settingsthan that found in previous research. Simply put,when work is interdependent, pay dispersion ex-plained by productivity-relevant employee in-puts provides sorting advantages that lead to apositive relationship with team performance, butpay dispersion net of these inputs does not.

REFERENCES

Adams, J. S. 1963. Toward an understanding of inequity.Journal of Abnormal and Social Psychology, 67:422–436.

Aiken, L. S., & West, S. G. 1991. Multiple regression:Testing and interpreting interactions. NewburyPark, CA: Sage.

Akerlof, G. A., & Yellen, J. L. 1988. Fairness and un-employment. American Economic Review, 78: 44–49.

Alchian, A. A., & Demsetz, H. 1972. Production, infor-mation costs, and economic organization. AmericanEconomic Review, 62: 777–795.

Alvesson, M., & Karreman, D. 2007. Constructing mystery:Empirical matters in theory development. Academyof Management Review, 32: 1265–1281.

Ambrose, M. L., & Kulik, C. T. 1999. Old friends, newfaces: Motivation research in the 1990s. Journal ofManagement, 25: 231–292.

Ambrose, M. L., Seabright, M. A., & Schminke, M. 2002.Sabotage in the workplace: The role of organiza-tional injustice. Organizational Behavior and Hu-man Decision Processes, 89: 947–965.

Bamberger, P., & Belogolovsky, E. 2010. The impact ofpay secrecy on individual task performance. Person-nel Psychology, 63: 965–996.

Barker, J. R. 1993. Tightening the iron cage: Concertivecontrol in self-managing teams. Administrative Sci-ence Quarterly, 38: 408–437.

Barnard, C. I. 1938. The functions of the executive.Cambridge, MA: Harvard University Press.

Barry, B., & Stewart, G. L. 1997. Composition, process,and performance in self-managed groups: The role ofpersonality. Journal of Applied Psychology, 82: 62–78.

Beauchamp, M. R., & Bray, S. R. 2001. Role ambiguityand role conflict within interdependent teams.Small Group Research, 32: 133–157.

Beaumont, P. B., & Harris, R. I. D. 2003. Internal wagestructures and organizational performance. BritishJournal of Industrial Relations, 41: 53–70.

Becker, B. E., & Huselid, M. A. 1992. The incentive ef-fects of tournament compensation systems. Admin-istrative Science Quarterly, 37: 336–350.

Becker, B., & Gerhart, B. 1996. The impact of humanresource management on organizational perfor-mance: Progress and prospects. Academy of Man-agement Journal, 39: 779–801.

Bishop, J. 1987. The recognition and reward of employeeperformance. Journal of Labor Economics, 5: S36–S56.

Blalock, H. M. 1985. Causal models in the social sci-ences. Chicago: Aldine.

Bloom, M. 1999. The performance effects of pay disper-sion on individuals and organizations. Academy ofManagement Journal, 42: 25–40.

Brown, M. P., Sturman, M. C., & Simmering, M. J. 2003.Compensation policy and organizational perfor-mance: The efficiency, operational, and financial im-plications of pay levels and pay structure. Academyof Management Journal, 46: 752–762.

Cadsby, C. B., Song, F., & Tapon, F. 2007. Sorting andincentive effects of pay for performance: An experi-mental investigation. Academy of ManagementJournal, 50: 387–405.

Cameron, A. C., Gelbach, J., & Miller, D. 2008. Boot-strapped-based improvements for inference withclustered errors. Review of Economics and Statis-tics, 90: 414–427.

Cameron, A. C., & Trivedi, P. K. 1986. Econometric mod-els based on count data: Comparisons and applica-tions of some estimators and tests. Journal of Ap-plied Econometrics, 1: 29–54.

Conyon, M. J., Peck, S. I., & Sadler, G. V. 2001. Corporatetournaments and executive compensation: Evidencefrom the UK. Strategic Management Journal, 22:805–815.

Cowherd, D. M., & Levine, D. I. 1992. Product quality andpay equity between lower-level employees and topmanagement: An investigation of distributive justicetheory. Administrative Science Quarterly, 37: 302–320.

Cummings, T. G. 1978. Self-regulating work groups: Asocio-techncial analysis. Academy of ManagementReview, 3: 625–634.

Danielson, M. 2004. Home team In S. Rosner & K. Shrop-shire (Eds.), The business of sports: 34–38. Sud-bury, MA: Jones and Bartlett.

Depken, C. A. 2000. Wage disparity and team productiv-ity: Evidence from major league baseball. EconomicsLetters, 67: 87–92.

Deutsch, M. 1985. Distributive justice: A social-psycho-logical perspective. New Haven, CT: Yale Univer-sity Press.

Ding, D., Akhtar, S., & Ge, G. 2009. Effects of inter andintra-hierarchy wage dispersions on firm perfor-mance in Chinese enterprises. International Jour-


nal of Human Resource Management, 20: 2370–2381.

Efron, B., & Tibshirani, R. 1993. An introduction to thebootstrap. New York: Chapman & Hall/CRC.

Eriksson, T. 1999. Executive compensation and tourna-ment theory: Empirical tests on Danish data. Journalof Labor Economics, 17: 262–280.

Fantasy, ESPN. 2007. Hockey rules scoring categories.http://sports.espn.go.com/fantasy/hockey/fhl/story?page�fhlrulesscoringcategories.

Ferraro, F., Pfeffer, J., & Sutton, R. I. 2005. Economicslanguage and assumptions: How theories can be-come self-fulfilling. Academy of Management Re-view, 30: 8–24.

Foster, W. M., & Washington, M. 2009. Organiza-tional structure and home team performance.Team Performance Management, 15(3/4): 158–171.

Frank, E., & Nuesch, S. 2011. The effect of wage disper-sion on team outcome and the way team outcome isproduced. Applied Economics, 43: 3037–3049.

Frank, R. H. 1985. Choosing the right pond: Humanbehavior and the quest for status. New York: Ox-ford University Press.

Frank, R. H., & Cook, P. J. 1995. The winner-take-all-society. New York: Free Press.

Frick, B., Prinz, J., & Winkelmann, K. 2003. Payinequalities and team performance: Empiricalevidence from the North American majorleagues. International Journal of Manpower, 24:472–491.

Fry, L. W., Kerr, S., & Lee, C. 1986. Effects of differentleader behaviors under different levels of task inter-dependence. Human Relations, 39: 1067–1081.

Gerhart, B., & Milkovich, G. T. 1990. Organizational dif-ferences in managerial compensation and financialperformance. Academy of Management Journal,33: 663–691.

Gerhart, B., & Milkovich, G. T. 1992. Employeecompensation: Research and theory. In M. D. Dun-nette & L. M. Hough (Eds.), Handbook of industrialand organizational psychology (2nd ed.), vol. 3:481–569. Palo Alto, CA: Consulting PsychologistsPress.

Gerhart, B., & Rynes, S. L. 2003. Compensation: Theory,evidence, and strategic implications. ThousandOaks, CA: Sage.

Gerhart, B., Rynes, S. L., & Fulmer, I. S. 2009. Pay andperformance: Individuals, groups, and executives. InJ. P. Walsh, & A. P. Brief (Eds.), Academy of Man-agement annals, vol. 3: 251–315. Essex, U.K.: Rout-ledge.

Grund, C., & Westergaard-Nielsen, N. 2008. The disper-sion of employees’ wage increases and firm perfor-mance. Industrial and Labor Relations Review, 61:485–501.

Harrison, D. A., & Klein, K. J. 2007. What’s the differ-ence? Diversity constructs as separation, variety ordisparity in organizations. Academy of Manage-ment Review, 32: 1199–1228.

Hausman, J. A. 1978. Specification tests in econometrics.Econometrica, 46: 1251–1271.

Helfat, C. E., & Peteraf, M. A. 2003. The dynamic re-source-based view: Capability lifecycles. StrategicManagement Journal, 24: 997–1010.

Heneman, H. G., & Judge, T. A. 2000. Compensationattitudes. In S. L. Rynes & B. Gerhart (Eds.), Com-pensation in organizations: Current research andpractice: 61–103. San Francisco: Jossey-Bass.

Heneman, R. L., & Werner, J. M. 2005. Merit pay: Link-ing pay to performance in a changing world.Greenwich, CT: Information Age.

Heyman, F. 2005. Pay inequality and firm performance:Evidence from matched employer-employee data.Applied Economics, 37: 1313–1327.

Hibbs, D. A., & Locking, H. 2000. Wage dispersion andproductive efficiency: Evidence for Sweden. Journalof Labor Economics, 18: 755–782.

Hicks, J. R. 1963. The theory of wages (2nd ed.). London:MacMillan.

Jewell, R. T., & Molina, D. J. 2004. Productive efficiencyand salary distribution: The case of US Major LeagueBaseball. Scottish Journal of Political Economy, 51:127–142.

Kahn, L., & Sherer, P. 1990. Contingent pay and mana-gerial performance. Industrial and Labor RelationsReview, 43: 107–121.

Kandel, E., & Lazear, E. P. 1992. Peer pressure and part-nerships. Journal of Political Economy, 100: 801–817.

Keidel, R. W. 1987. Team sports models as a genericorganizational framework. Human Relations, 40:591–612.

Kepes, S., Delery, J. E., & Gupta, N. 2009. Contingenciesin the effects of pay range on organizational effec-tiveness. Personnel Psychology, 62: 497–531.

Krueger, A. B. 1988. The determinants of queues forfederal jobs. Industrial and Labor Relations Re-view, 41: 567–581.

Latham, G. P., & Pinder, C. C. 2005. Work motivationtheory and research at the dawn of the twenty-firstcentury. In S. Fiske, D. L. Schacter, & C. Zahn-Wexler (Eds.), Annual review of psychology, vol.56: 485–516. Palo Alto, CA: Annual Reviews.


Lazear, E. P. 1989. Pay equality and industrial politics.Journal of Political Economy, 89: 841–864.

Lazear, E. P. 2000. Performance pay and productivity.American Economic Review, 90: 1346–1361.

Lee, K. W., Lev, B., & Yeo, G. H. H. 2008. Executive paydispersion, corporate governance, and firm perfor-mance. Review of Quantitative Finance and Ac-counting, 30: 315–338.

Leonard, J. S. 1990. Executive pay and firm perfor-mance. Industrial and Labor Relations Review, 43:S13–S29.

Leventhal, G. S. 1976. The distribution of rewards andresources in groups and organizations. In L.Berkowitz (Ed.), Advances in experimental socialpsychology, vol. 9: 92–131. New York: AcademicPress.

Levine, D. I. 1991. Cohesiveness, productivity, and wagedispersion. Journal of Economic Behavior and Or-ganization, 15: 237–255.

MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M.,West, S. G., & Sheets, V. 2002. A comparison ofmethods to test mediation and other interveningvariable effects. Psychological Methods, 7: 83–104.

Main, B. G. M., O’Reilly, C. A., & Wade, J. 1993. Topexecutive pay: Tournament or teamwork. Journal ofLabor Economics, 11: 606–628.

Medoff, J. L., & Abraham, K. G. 1980. Experience, perfor-mance, and earnings. Quarterly Journal of Econom-ics, 95: 703–736.

Michaels, E., Handfield-Jones, H., & Axelrod, B. 2001.The war for talent. Boston: Harvard BusinessSchool Press.

Milkovich, G. T., & Newman, J. M. 2008. Compensation(9th ed.). Boston: McGraw Hill.

Mintzberg, H. 1978. Patterns in strategy formation. Man-agement Science, 24: 934–948.

Mooney, C. Z. 1996. Bootstrap statistical inference: Ex-amples and evaluations for political science. Amer-ican Journal of Political Science, 40: 570–602.

National Hockey League. 2002. Official guide & recordbook. Kingston, NY: Total Sports Publishing.

Pagan, A. R. 1984. Econometric issues in the analysis ofregressions with generated regressors. InternationalEconomic Review, 25: 221–247.

Pfeffer, J. 1994. Competitive advantage through people:Unleashing the power of the work force. Boston:Harvard Business School Press.

Pfeffer, J., & Davis-Blake, A. 1992. Salary dispersion,location in the salary distribution, and turnoveramong college administrators. Industrial and LaborRelations Review, 45: 753–763.

Pfeffer, J., & Langton, N. 1993. The effect of wage disper-sion on satisfaction, productivity, and working col-laboratively: Evidence from college and universityfaculty. Administrative Science Quarterly, 38:382–407.

Ployhart, R. E., Weekley, J., & Ramsey, J. 2009. The con-sequences of human resource stocks and flows: Alongitudinal examination of unit service orientationand unit effectiveness. Academy of ManagementJournal, 52: 996–1015.

Prendergast, C. 1999. The provision of incentives infirms. Journal of Economic Literature, 37: 7–63.

Salamin, A., & Hom, P. W. 2005. In search of the elusiveU-shaped performance-turnover relationship: Arehigh performing Swiss bankers more liable to quit?Journal of Applied Psychology, 90: 1204–1216.

Schwab, D. P., & Olson, C. A. 1990. Merit pay practices: Im-plications for pay-performance relationships. Industrialand Labor Relations Review, 43: 237S–255S.

Shaw, J. D., Delery, J. E., Jenkins, G. D., & Gupta, N. 1998.An organization-level analysis of voluntary and in-voluntary turnover. Academy of Management Jour-nal, 41: 511–525.

Shaw, J. D., & Gupta, N. 2007. Pay system characteristicsand quit patterns of good, average, and poor perform-ers. Personnel Psychology, 60: 903–923.

Shaw, J. D., Gupta, N., & Delery, J. E. 2002. Pay dispersionand workforce performance: Moderating effects ofincentives and interdependence. Strategic Manage-ment Journal, 23: 491–512.

Sheppard, J. A. 1993. Productivity loss in performancegroups: A motivational analysis. Psychological Bul-letin, 113: 67–81.

Siegel, P. A., & Hambrick, D. C. 2005. Pay disparitieswithin top management groups: Evidence of harmfuleffects on performance of high-technology firms. Or-ganization Science, 16: 259–274.

Sobel, M. E. 1982. Asymptotic confidence intervals forindirect effects in structural equation models. Soci-ological Methodology, 13: 290–312.

Sommers, P. M. 1998. Work incentives and salary distri-butions in the National Hockey League. AtlanticEconomic Journal, 26: 119.

Steers, R. M., & Porter, L. 1983. Motivation and workbehavior. New York: McGraw Hill.

Stine, R. 1990. An introduction to bootstrap methods:Examples and ideas. Sociological Methods and Re-search, 18: 243–291.

Sturman, M. C., Trevor, C. O., Boudreau, J. W., & Gerhart,B. 2003. Is it worth it to win the talent war? Evalu-ating the utility of performance-based pay. Person-nel Psychology, 56: 997–1036.


Thompson, J. 1967. Organizations in action. New York:McGraw-Hill.

Trevor, C. O., Gerhart, B., & Boudreau, J. W. 1997. Vol-untary turnover and job performance: Curvilinearityand the moderating influences of salary growth andpromotions. Journal of Applied Psychology, 82: 44–61.

Trevor, C. O., & Wazeter, D. 2006. A contingent view ofreactions to objective pay conditions: Interdepen-dence among pay structure characteristics and pay

relative to internal and external referents. Journal ofApplied Psychology, 91: 1260–1275.

Vroom, V. 1964. Work and motivation. San Francisco:Jossey-Bass.

Winter, S. G. 2000. The satisficing principle in capabilitylearning. Strategic Management Journal, 21: 981–996.

Wooldridge, J. 2002. Econometric analysis of cross sec-tion and panel data. Cambridge, MA: MIT Press.

Charlie O. Trevor ([email protected]) is an associateprofessor in the Management and Human Resources De-partment at the School of Business, University of Wis-consin–Madison. He earned his Ph.D. in industrial andlabor relations from Cornell University. His current re-search interests include employee compensation, volun-tary turnover, and downsizing.

Greg Reilly ([email protected]) is an assistantprofessor in the Management Department at the Univer-sity of Connecticut School of Business. He received his

Ph.D. from the University of Wisconsin–Madison. Hiscurrent research interests include teams, goals, andcompensation.

Barry Gerhart ([email protected]) is the Ellig Dis-tinguished Chair at the School of Business, University ofWisconsin–Madison. He received his Ph.D. in industrialrelations from the University of Wisconsin–Madison. Hisresearch interests include compensation, human re-source strategy, and international human resources.


reconsidering pay dispersion’s effect on the …leeds-faculty.colorado.edu/dahe7472/trevor...

Documents