3.5%....
3%”...
vim.

. 2‘. ’21:.

ﬁﬁééam
: v. . . .
32%.. p.
"3.433..
2.. 3. $5..

a. .21.")...
.2. .. .
. ﬁummmuéaanz
«my; .24

3:;
3.1.

w...” a”
“s“...mzwzhumuﬁmww
. . v3 . 15.3: as:
1.. : .x... a...
3.2.2 :3. . iii)...» “.2:
.rtlu 931...: .mumm..urxmﬂﬂ
a. . is
. P‘mln !
z . .
:21$.51.:jm.eﬂ H 92L“...

....i..iii. . .115»

I! L
. u
an.
.. ... 5.. E2131:
5 2.1.3.1....
.. «.3332
e

. .tvsz..nmm.g. .. .3: V
1).}. 13-5"
..h.............:..£xyt,a
. .r. 5.6

. .. (1...... i .

2}. .: ..:..:..v.i..r.:

3... . ,9...
.... . ﬂuépmariu

3.115,. 1
34.. dtﬁmu
at... £3.93. .2”

r1
hm...“
..rnlio-...Ha
Januﬁ
u}.u\\

 

:- 3.
’31:... IP.
:hh”.d.v:-I\2“.1t
.lIr. 1.2: 1‘!
(«3‘1

. .1. 7.... (Axaaﬁﬁﬂhui.
.. :1." .... .1 ...... . .......:_ ,. . .
f... “g.“ .i....m.§..w.... , . ..

 

iCHiG m LIBRARIES

 

Iii i lililllﬂllllllli Hill

ﬂ “1293 048 5920

 

 

 

 

 

 

 

 

LIBRARY

Michigan State
University

 

 

This is to certify that the

dissertation entitled

CEO AND AVERAGE EMPLOYEE PAY DURING THE 19808:

FIRM—LEVEL DETERMINANTS, DEVELOPMENTS
AND EFFECTS

presented by

JESS REASER

has been accepted towards fulﬁllment
of the requirements for

.D.

degree in

 

 

ECONOMICS

o prof or

Date izlmlﬂ?

MS U is an Afﬁrmative Action/Equal Opportunity Institution O-12771

PLACE lN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

11100 W.“

 

CEO AND AVERAGE EMPLOYEE PAY DURING THE 19808:
FIRM-LEVEL DETERMINANTS, DEVELOPMENTS, AND EFFECTS

By

Jess Reaser

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

1999

ABSTRACT

CEO AND AVERAGE EMPLOYEE PAY DURING THE 19808:
FIRM-LEVEL DETERMINANTS, DEVELOPMENTS, AND EFFECTS

By

Jess Reaser

This dissertation presents two separate analyses of chief executive and average
employee compensation. For both studies, data on executive compensation collected by
the author from archived proxy statements are matched with ﬁrm-level compensation,
balance sheet, and market data provided by Standard and Poor’s Compustat. The ﬁrst
study looks at the determinants of executive compensation, focusing primarily on how
ﬁrm-level unionization impacts executive pay. The second study looks at relative pay
developments over the 19803 and how those developments affected employee output.

The implicit regulation hypothesis proposed by Jensen and Murphy (1990)
suggests that CEO pay may be lower and less sensitive to ﬁrm performance than is
optimal due to political constraints posed by negative public reactions to high executive
pay. Chapter 1 uses firm-level measures of unionization in a sample of manufacturing
firms to test the predictions of the implicit regulation hypothesis. Results suggest that
executives in densely unionized ﬁrms have signiﬁcantly lower pay than executives in
nonunion and less unionized ﬁrms. However, the performance sensitivity of CEO pay in
unionized ﬁrms is higher than that in nonunion firms. Additionally, it is shown that at
high levels of unionization, unions successfully decrease the difference between

compensation expenditures made to chief executives and to average employees.

It is also shown that the pay of chief executive officers in the United States has
grown at a much faster rate than the per-employee compensation expenditures in the
same ﬁrms. Chapter 2 examines whether the divergence in pay levels has had an impact
on the productivity of ﬁrm employees. According to equity theory in industrial
psychology, workers will withdraw productive work effort as the wages for workers in
comparison groups grow at a relatively faster rate. According to tournament theory in
economics, however, relatively high wages for chief executives are consistent with
efﬁcient incentive mechanisms. The impact of pay inequality, measured as the difference
between executive and average employee pay, is gauged by including measures of
relative pay as explanatory variables in a production function framework. Efforts are
made to address potential sources of bias in the estimates, including bias associated with
the simultaneous determination of pay and productivity and bias associated with
unmeasured worker quality. Empirical tests do not indicate that a high ratio of executive
to average employee pay is associated unequivocally with either lower or higher per

worker output.

For Lian, remember that there are many ways to think about the world...

RETIRED MINERS

In Dr. Capailetti’s office,
crippled and wheezing:

“if any guy tells you
he got rich through hard work
ask him whose?”

From Miracle Mile by Ed Ochester, published by Carnegie-Mellon University
Press, Pittsburgh, 1984. Used with the author’s permission.

iv

ACKNOWLEDGEMENTS

An accounting of my indebtedness to the people who have helped me get to this
page would be an exhausting exercise. These acknowledgements are therefore, perhaps
necessarily, incomplete. For those of you who are included here with insufﬁcient
recognition, and for those of you who have been left out altogether, you have my deepest
gratitude for your help.

I could not have had access to a better group of scholars than those I have had the
privilege of working with while here at Michigan State. I especially would like to thank
Harry Holzer, my advisor, for his help over the last several years. His guidance in
bringing together these papers and getting me through the program is greatly appreciated.
Harry’s assistance in interpreting results, giving direction, and providing feedback has
been instrumental in developing the papers here. I also would like to thank him for
employing me as a research assistant for the past three years. Working for Harry has been
a boon to my development as an economist, and the fact that I beneﬁted while working
on topics I find interesting and important has made the experience extraordinary. Thank
you Harry for your help, your generosity, your patience, and for all the opportunities you
have provided me.

I would also like to thank David Neumark for his help. David’s contributions to
these papers have been fundamental. I have beneﬁted in innumerable ways from the
breadth and depth of his knowledge and interests and from his willingness to explain
concepts and give direction. Wally Mullin, as the ﬁnal member of my committee, has

provided much needed insight and context for the papers here. I appreciate his attention

to detail and his willingness to sit and help me work through ideas. Perhaps more
beneﬁcial than the comments each of my committee members has offered on the various
drafts of these papers has been their demonstrated work ethic, commitment to
scholarship, and passion for what they do.

This dissertation has also benefited from the comments of John Strauss and Jeff
Wooldridge among the faculty here at MSU. Hailong Qian and Sang Hyop Lee made
comments that materially improved the analysis presented in these papers, and Kathleen
Beegle and Heather Bednarek spent much time helping me edit. Professor Barry Hirsch at
Florida State generously provided me with the unionization data used in Chapter 1. The
usual disclaimer applies, however: the aforementioned are in no way responsible for the
errors that remain.

It would be preposterous not to acknowledge the friendships I have beneﬁted
from during graduate school. Kathleen, Heather, Sang Hyop, Qian, Dan, Paul, and 800
have made this endeavor sufﬁciently fun that I continued long after it otherwise ceased
being optimal. For now you get my gratitude, though I reckon the future holds your
rightful share of the blame. I would also like to thank Shwu Ying Chen for her help and
patience while I made my way through this process.

My deepest thanks go to my family. I do not possess the words that express the
admiration and love I feel for them. I would like to thank my parents and my sister Abby
and brother in law John for providing free childcare, making things much easier than they
otherwise would have been. And I would like to thank my mom and dad, brothers and
sisters, and nieces and nephews for providing me a comfortable place to belong.

Finally, I would like to thank Lian. Thanks boy, you make it all worthwhile.

vi

TABLE OF CONTENTS

LIST OF TABLES ix

CHAPTER 1
UNIONS AND CEO COMPENSATION: ARE UNIONS INIPLICIT REGULATORS?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Introduction 1
The compensation of chief executive ofﬁcers and the regulation of pay 6
Executive productivity 7
Executive incentives 8
Regulation of executive pay and welfare analysis 9
Unions and compensation 12
Unions, the median voter, and relative wages 13
Unions and firm performance 16
Countervailing factors ' 17
Data 19
Econometric Model and Predictions 21
Econometric problems 26
Effect of Unions on Chief Executive Pay 28
Unionization and pay equality 37
Fixed effects estimates 40
Conclusions 47
Bibliography 50
CHAPTER 2
WIDENING PAY GAPS AND FIRM PRODUCTIVITY: DOES FAIRNESS MATTER?
Introduction 63
Pay, relative pay, and productivity: theory and evidence 66
Compensation systems and equity theory 66
Pay variation in economics: the role of incentives 69
Related Evidence 72
Data and estimating framework 76
Econometric problems 81
Predictions 87
Empirical results 87
Summary statistics 87
Regression results 90
Industry pay variables 92
Instrumental variables estimation 93
Firm size 97
Ability to pay 98
Employee quality 101
Productivity levels as a ﬁrnction of changes in pay 104

Conclusions 105

vii

Bibliography 108

APPENDIX A 120

viii

LIST OF TABLES

CHAPTER 1
Table 1.1: Means (Standard Deviations) of Compensation Levels and Compensation
Growth for CEOs and Average Employees in Manufacturing Firms from 1980 to

 

 

 

 

 

 

 

 

 

1987 53
Table 1.2: Summary Statistics, Firm and CEO Characteristics 62
Table 1.3: Determinants of Annual CEO Salary + Bonus, OLS Estimates 55
Table 1.4: The Effect of Unionization on Per Employee Pay and the CEO-Average
Employee Pay Differential 57
Table 1.5: Determinants of Annual CEO Salary + Bonus, Fixed Effects Estimates 58
Table 1.Al: Changes in Unionization Between 1977 and 1987 60
Table 1.A2: Changes in Other Categorical Variables Between 1980 and 1987 __ 62
CHAPTER 2
Table 2.1: Literature Review 111
Table 2.2: Pay Levels and Pay Growth for CEO’s and Average Employees, _ 114
Table 2.3: Summary Statistics, Means (Standard Deviations) 115
Table 2.4: The Effect of Pay Differences on Output 116
Table 2.5: The Effect of Pay Differences on Output, Alternative Lag Structures _ 117
Table 2.6: The Effect of Pay Differences on Output, Instrumental Variables _ 118
Table 2.7: The Effects of Pay Differences on Output, Difference Estimates 119
APPENDIX A

Table A.1: Production Function Estimates, Full Sample 126
Table A2: Production Function Estimates, Text Sample without Relative
Compensation Variables 127

 

Table A3: Production Function Estimates, Text Sample with Relative Compensation
Variables 128

ix

CHAPTER 1

UNIONS AND CEO COMPENSATION: ARE UNIONS IMPLICIT REGULATORS?

Introduction

Recent economic analysis of the compensation packages received by CEOs in
large US corporations has explored the possibility that CEO pay is affected by its public
disclosure. Jensen and Murphy (1990) have proposed that political forces both inside and
outside the ﬁrm engage in “implicit regulation” of CEO pay at least in part due to SEC
requirements that the pay of executives be recorded in annual proxy statements, which
are sent to shareholders. This paper explores the relationship between ﬁrm-level
measures of unionization and executive compensation to deterrrrine if the effects of
unionization on executive pay are consistent with the predictions generated by the
implicit regulation hypothesis.

According to the implicit regulation hypothesis, average CEO pay is both lower
and less sensitive to ﬁrm performance than it would be in the absence of public
disclosure due to the negative reactions that greet announcements of high executive pay
levels. Such reactions, from shareholders and “employees, labor unions, consumer
groups, congress, and the media (Jensen and Murphy, p.254, 1990),” are believed to
insinuate themselves into the pay determination process for CEO’s, making it difﬁcult for
boards of directors to devise mechanisms which reward executives with extraordinary
pay levels for extraordinary market performance. The truncation at the high end of the

executive pay distribution, in turn, mitigates the possible “punishments” boards can mete

out for poor executive performance, resulting in a downward limit to executive pay and
less risk of executives losing their jobs.

Implicit regulation, according to Jensen and Murphy, may explain their ﬁnding
that executive pay is relatively insensitive to changes in the market valuation of ﬁrms.
This insensitivity raises questions concerning the importance of executives to ﬁrm
economic performance and, consequently, the optimality of executive pay. Since
economic theory suggests that, especially for chief executives, optimal labor contracts
will tie employee compensation to ﬁrm performance, failure to ﬁnd economically
substantial links between them may imply that markets disregard the inputs of chief
executives and that boards of directors therefore have little reason to tie executive pay to
firm performance'. Jensen and Murphy, among others, ﬁnd this interpretation
unsatisfying based in part on empirical evidence showing that the market valuation of
firms changes significantly with unexpected changes in ﬁrm leadership.

Since markets appear to pay attention to ﬁrm leadership, implicit regulation is
arguably a better explanation for the observed relationship between performance and pay.
Jensen and Murphy ﬁnd some support for this explanation by comparing executive pay
levels and sensitivity during the 19703 and 19803 to the higher levels and greater
sensitivity of executive pay during the 19303, a period when firms were not required to
publicize the pay of their top executives.

Testing a similar hypothesis using executive pay over the 19703 and 803, Joskow,
Rose, and Shepard (1993) show that the extent of government economic regulation

¥

1 Other economists question Jensen and Murphy’s interpretation of their ﬁndings either
on the basis of econometric misspeciﬁcation (Rosen, 1992, and Holmstrom, 1992) or on
theoretical grounds (Haubrich , 1994).

impacts the levels of executive pay and its sensitivity to ﬁrm market performance.
Executives of ﬁrms subject to ﬁrm-level price regulation, such as electric utilities, are
compensated dramatically less than executives of nonregulated firms. Furthermore, these
same executives have pay that is much less sensitive to ﬁrm performance than executives
of ﬁrms in nonregulated industries. J oskow, Rose, and Shepard propose that this may be
due to the political-economic nature of regulatory agencies, with regulatory ofﬁcials
being especially sensitive to public disapproval of high executive pay and especially
opposed to pay schemes which focus executives’ attention on increasing shareholder
value at the expense of consumer surplus.

This paper proceeds down a sirrrilar avenue of analysis by using firm measures of
union density to determine if unions affect the levels and manner in which executives get
paid. Arguably, if there is a political effect on executive pay independent of regulatory
authorities, it will be stronger in unionized ﬁrms. The motivation for this assertion is
provided by a combination of analyses presented in the theoretical and empirical analyses
of unions. Specifically, median voter models of unionization suggest that, provided union
members have a general preference for equality between their pay and that of executives,
the compensation of executives in unionized ﬁrms should be lower due to the enhanced
voice mechanism of unions allowing members to assert their preferences (Freeman and
Medoff, 1984).

Related economic reasons stem from the fact that management in unionized ﬁrms
must repeatedly interact with unions in contract negotiations. High executive pay may
serve as a signal to unions of the ﬁnancial health of the ﬁrm and as a symbol of the

allocation of proﬁts between capital and labor, making directors of unionized ﬁrms wary

of pay schemes which result in huge payoffs for chief executives. High executive pay,
Jensen and Murphy (1990) argue, may “provide emotional justification for increased
labor demands in labor negotiations.” Finally, unions play an additional political-
econorrric role in society through lobbying campaigns and policy advocacy (Addison and
Hirsch, 1986). If executive pay in unionized ﬁrms becomes too far out of line with union
members’ preferences, then unions may be more inclined to make efforts to curtail it
through political channelsz.

Results presented in this paper do not support the notion that unions, per se, play
a regulatory role in the pay of chief executives. In fact, executives of ﬁrms with moderate
levels of unionization seem to make more than the executives of nonunion ﬁrms.
However, executives in firrrrs with very high levels of unionization are paid signiﬁcantly
less than executives in nonunion and less unionized ﬁrms. Like the results in Joskow,
Rose, and Shepard, it is difficult to attribute this exclusively to a “regulatory” effect,
since unions may also decrease the potential impact that executives can have on ﬁrm
performance, though the bias associated with unobserved executive productivity is likely
to be a less important factor for these nonregulated ﬁrms. Fixed effects procedures
suggest that unobserved ﬁrm heterogeneity explains the negative impact of unions on
executive pay, though a negative impact on pay relative to less unionized ﬁrms remains.
Fixed effect results are statistically inconclusive, however, and their usefulness may be

limited due to data limitations and measurement error. There is also evidence that unions

 

2 See, for example, the International Union of Operating Engineers’ World Wide Web
site, entitled “America Needs a Raise,” at www.ioue.org/html/oe5.htm; and the AFL-
CIO’s “Executive Paywatch,” at www.aﬂci0.paywatch.org/ceopay/front.html, which
greets users with the statement, “Welcome to Executive Paywatch -- a working family’s
guide to monitoring and curtailing the excessive salaries, bonuses and perks in CEO
compensation packages.”

are able to narrow the pay difference between the average worker and the chief executive
when they have organized a sufﬁcient proportion of workers.

While unions are shown to impact the pay levels of chief executive officers, there
is not evidence that they impact the pay-performance relationship in the way predicted by
the implicit regulation hypothesis. The estimated semi-elasticity associated with market
returns is larger in unionized ﬁrms than in nonunion firms, and among unionized ﬁrms
there is little indication that this relationship declines as firms become increasingly
unionized. This result suggests that the formulation of the implicit regulation hypothesis
suggested by Jensen and Murphy may not fully capture the preferences that unions have
regarding how executives are paid.

While the results presented in this paper do not provide unequivocal support for
the implicit regulation hypothesis, neither do they lend themselves to direct conclusions
regarding the welfare effects of chief executive pay packages. Previous analysis has cited
concern in the popular media that high pay for executives in large US. corporations
reﬂects executive capture and control of boards of directors, not relative productivity
(J oskow, Rose, and Shepard (1993) and Joskow, Rose, and Wolfram (1996)). It cannot be
concluded from this analysis, for example, that the high executive pay in some unionized
ﬁrms is inefficient or unmerited on the basis that executives are participants in rent
sharing arrangements, since the executives may be responsible for identifying,
maintaining, or capturing those rents. Nevertheless, the results do show that executives in
unionized ﬁrms are paid differently, and in some cases more, than the executives in

nonunion ﬁrms.

The following analysis proceeds by reviewing the economic literature on
executive pay. In addition, relevant research on unionization is reviewed, with special
attention paid to their effects on wages and relative wages. The data used in this analysis
are then introduced, followed by the econometric model used and predictions for the
estimates in the model. Statistical results are then presented, followed by concluding

COITIITICIIIS.

The compensation of chief executive ofﬁcers and the regulation of pay

Gomez-Meija and Balkin (1992) would impose a moratorium on the analysis of
executive pay pending the emergence of data superior to those currently available from
traditional sources. Other authors, however, have noted that the compensation of
executives in large US corporations probably merits the academic attention it has
received given the enormous levels of resources at the command of these executives3.
This argument points out that there are substantial welfare implications from devising
efﬁcient incentive mechanisms for these employees. Rosen classiﬁes the economic
analysis of executive pay in terms of three fundamental problems which an efficient
executive labor market must solve: “the distribution and control among executives;
providing performance incentives; and identifying talent and reassigning control in the

course of career development (1992, p. 182).” These “problems” have been provided a

 

3 Rosen (1992), Joskow, Rose, and Shepard (1993).

theoretical framework through models examining marginal productivity in hierarchical

organizations, principal-agent theories, and tournament theories, respectively“.

Executive productivity

Competitive labor markets will result in workers being paid their marginal
product. The high pay of chief executive ofﬁcers reflects the higher productivity of these
individuals as well as the importance of their positions. Generally, this is illustrated in
two ways. First, models incorporating the hierarchical nature of production show that
individuals sorted into the top positions of firms have higher pay because their actions are
magniﬁed through all subordinate levels of the hierarchy. Because of this magnification
of productivity, decisions made at the top are especially important and there is strong
incentive for the hierarchy of the ﬁrm to sort the most talented people into the highest
positions, where “a little extra talent. . .can have enormous effects on total output (Rosen,
1992, p.185).” Thus, when there are complementarities between individual talent and
control of resources, the people in charge of those resources will receive substantially
more pay than others in the same organization, the most talented people will be allocated
into the top positions of the largest ﬁrms, and the returns to devising efﬁcient incentives
will be highest in large ﬁrms.

Additionally, the high concentration of pay in the top positions of ﬁrms reﬂects
the need to structure pay mechanisms which induce greater effort from younger

executives as they move through their career paths. Rosen (1986) presents a model where

 

4 See Gomez-Meija and Balkin (1992) for a multidisciplinary discussion of executive
compensation.

players vie for promotion through several stages of a tournament. Promotion into the next
stage increases the remuneration of the winner and allows her to continue to compete for
further promotions. As individuals proceed toward the top of the hierarchy, the number of
possible promotions decreases which decreases the incentive effect of participating in
further contests. Thus, ﬁrms must increase the payoff to promotions in the ﬁnal stages of
play to offset the decline in the future income stream associated with fewer potential
promotions. That is, the payoff must increase so that incentives are maintained for
individuals near the top of the promotion ladder, who may otherwise "rest on their
laurels" and reduce effort. This framework explains the positive skew in the distribution
of pay among top executives, where CEO’s frequently make substantially more than even
the second highest paid executive.

Empirical tests of these theories have taken several forms. In his review of the
literature, Rosen (1992) ﬁnds that estimates of the size elasticity for chief executive pay
is consistently estimated at around .25, implying that the pay of executives increases 2.5
percent on average for every 10 percent increase in ﬁrm size. Tests of tournament models
are less numerous, though papers exploring the structure of the pay hierarchy among

executives and its relationship to ﬁrm performance have found mixed resultss.

Executive incentives

Recently, econorrrists have expended much energy exploring the mechanisms
which can be used to tie managerial behavior to ﬁrm performance. In principal-agent
theory, the objectives of managers can be brought into alignment with those of owners

using the appropriate incentives. The primary result of principal-agent models is that non-

owning managers whose compensation is based on organizational performance will be
more diligent in maximizing shareholder value. Typically in these models, informational
asymmetries between the principals and agents regarding the agents’ effort levels (or
circumstances leading to some observed action) necessitate pay structures which result in
risk-averse agents exerting the effort which will maximize the expected value of the
principal’s assets. In absence of these pay structures, agents will be free to shirk or
engage in other utility maximizing behavior at the expense of firm performance.

With risk neutral agents, the efficient behavior is ensured by the transfer of
ownership from the principal to the agent. With risk averse agents, principals must insure
agents against risk by paying some base salary but will otherwise make pay contingent on
performance. Empirical analysis has measured the pay performance relationship for chief
executives by including accounting or market rates of return on the ri ght hand side of
earnings regressions. Positive and significant results are interpreted as indications that

managers’ pay is aligned with the interests of the ﬁrm.

Regulation of executive pay and welfare analysis

In their analysis of the pay-performance relationship, Jensen and Murphy argue
that the evidence suggests that executives have a only a small portion of their pay at risk
for poor performance. While principal agent theory provides no deﬁnitive rule regarding
the optimal portion of pay which should be tied to the performance of the ﬁrm, estimates
provided by Jensen and Murphy suggest that executives receive (or forego) only $.30 in

lifetime wealth for every $1000 generated (or lost) in shareholder value. Adding in

 

5 See Main, O’Reilly, and Wade (1993) and Leonard (1990).

another $.45 for the estimated increase in executive turnover after bad performance
brings the total to 75 cents for every $1000 change in shareholder value. Their high end
estimates, including the effect of poor performance on the value of personal holdings in
ﬁrm shares, brings the total to $3.25 per $1000 change.

Jensen and Murphy argue that, while the pay-performance relationship is positive
and signiﬁcant, these adjustments to pay (and employment) for good or bad performance
constitute very little incentive for executives relative to their base pay. Their empirical
evidence indicates that on average the pay-performance relationship for CEOs in large
US ﬁrms has little economic signiﬁcance. This is not due to the importance of the
executives to firm performance, since there remains an active takeover market which
disciplines against managerial performance becoming too inefﬁcient. Rather, they argue
that implicit regulation may be decreasing pay levels and muting its performance
sensitivity.

Joskow, Rose, and Shepard (1993) use interindustry variation in regulatory
behavior to identify the impact of economic regulation on executive pay. For example,
executive pay in the electric utilities industry is subject to ﬁrm-level regulation, and pay
levels are therefore subject to the scrutiny of state regulatory agencies. Owners of
regulated ﬁrms in the absence of this scrutiny presumably would engage the same
incentive mechanisms used in nonregulated industries in order to prompt efﬁcient
executive performance. Social welfare maximizing agencies, on the other hand, may not
approve of high payoffs, or at least of rewards based on stock performance, because those
agencies must act in the interest of consumers. Recognizing this conﬂict, regulatory

agencies may use their authority over rate setting or cost allowance decisions to prohibit

10

firms from establishing compensation packages that result in high payoffs. Testing these
hypotheses, they find substantial cuts in the pay levels for executives in regulated
industries, as well as signiﬁcantly lower pay—performance sensitivity for these ﬁrms.

The ﬁnding that regulatory agencies strongly impact the levels and performance
sensitivity of CEO pay relative to nonregulated ﬁrms does not necessarily imply that the
effect comes fully from political constraints on executive pay. Such an interpretation
depends on whether the pay of executives in nonregulated markets is efﬁcient or merely
indicative of how adept CEOs are at controlling the board of directors. Joskow, Rose, and
Wolfram observe, “to the extent that the political constraints become binding, they may
either limit ‘excessive’ CEO compensation (i.e. reduce any shareholder rents
appropriated by the CEO) or distort CEO performance incentives and the allocation of
managerial talent by reducing compensation from efﬁcient levels (p 166, 1996).” If the
latter is the case, then lower executive pay in regulated industries may be attributable to
productivity differences between executives in regulated versus nonregulated industries.
This will be true if one of the effects of regulation is to decrease the impact that
executives can have on the performance of the firm, making it efﬁcient for boards of
directors “to pay CEO’s less, to tie pay less closely to ﬁrm performance, and to hire less
able CEO’s (Joskow, Rose, and Wolfram, p.166, 1996).”

The effect of economic regulation on CEO pay is consistent with the predictions
of implicit regulation. However, it is difficult to discern if these effects occur because of
political restraints operating through the formal regulatory process or because regulation
decreases the relative productivity of CEOs in regulated ﬁrms. This paper uses a measure

of political variation across ﬁrms which presents a potentially cleaner test of the implicit

11

regulation hypothesis in that most ﬁrms are not subjected to external authorities which
have control over price setting and cost recovery. Jensen and Murphy (1991) propose
several sources of political regulators, including employees, unions, the media, and the
threat of legislative responses. The latter of these presumes that nonregulated firms
independently regard such a threat as real, though the magnitude of the benefits to pay
incentives and the costs of plausible government responses to high pay are not well
understood6. It is likely that in a Nash equilibrium sense, it would be suboptimal for any
ﬁrm to respond to the threat of a government response to a potentially lucrative pay
contract if all other ﬁrms are responding to the threat, especially if beneﬁts to such a
contract are substantial. There is also little indication that the public responds negatively
to high executive pay, at least in terms of purchasing decisions. The responses of
employees and unions are likely to be more important considerations in the determination
of executive pay, especially in the context of deteriorating or stagnating standards of

living.

Unions and compensation

Compensation specialists frequently discuss the various needs for ﬁrms to
develop compensation strategies which incorporate and balance the often conﬂicting
elements of equity, stability, and responsiveness to market forces], while noting that
unions will frequently impact compensation systems, work design, and management
strategies. In this context, the direct impact of unionization on CEO pay is of interest for

several reasons. First, unions are by their nature political organizations which seek to

 

6 For analysis of the beneﬁts of pay that is highly sensitive to performance, see Abowd
(1990).

12

increase the welfare of their members. Second, union members seem to prefer pay
equalitys, at least among members in particular bargaining units. Evidence of desire for
greater pay equality between executives and their employees is more anecdotal, but is
nonetheless a frequently voiced concern which is linked to both considerations of fairness
and productivity. According to Milkovich and Newman, “others, including unions, have
long held the belief that more egalitarian pay structures support team workers, high
commitment to the organization, and improved performance (p.52, 1992).” Finally,
managers of unionized ﬁrms must accommodate work rule constraints imposed by
collective bargaining agreements, as well as respond to the added variables of union rent

seeking and the strike threat.

Unions, the median voter, and relative wages

According to the exit-voice theory of unionization (Freeman and Medoff, 1984),
unions impact a ﬁrm’s industrial relations policies by providing a “voice” to members
which allows them to communicate collectively with management. This contrasts with
nonunion employees, who may only be inclined or able to communicate with
management by leaving, or exiting, the employment relationship9. In the absence of
unions, ﬁrms will construct pay packages, work rules, and other employment policies in

order to attract the marginal worker, though these policies may be undesirable to existing

 

7 See Levine (1993) for a discussion and empirical analysis of these.

8 See Parsons (1991) for indirect evidence of this, and see discussion below.

9 The exit-voice distinction is useful to the extent that the preferences of the average
worker in a potential collective bargaining unit are different from the preferences of the
marginal worker, which is likely to be the case if mobility differs by worker
characteristics like tenure and skill speciﬁcity.

13

employees. In such cases, workers have the option of leaving the employment
relationship or establishing a collective bargaining unit.

In cases where the beneﬁts of unionization to the median worker are greater than
the costs, unions may be established, becoming the democratic representatives of covered
workers. As such, unions are thought to assert the preferences of average, or
inframarginal, workers in negotiations with management over pay, employment, and
work rules. The democratic nature of unions reasonably assures that the preferences of
average workers are pursued subsequent to union certification. Hirsch and Addison
(1986), argue that the empirical evidence is generally supportive of the median voter
model of unions, suggesting that they do indeed pursue the interests of their average
workers.

In a test of the median voter model which pertains to the wages of workers, Farber
and Saks (1980) show that an individual’s position in a potential bargaining unit’s
earnings distribution is negatively related to her likelihood of voting for a union in
certiﬁcation elections. Unions, through various functions"), tend to pursue “egalitarian”
compensation policies through increasing the mean wage and narrowing the distribution
of earnings. They accomplish this by shifting the left tail of the earnings distribution
toward the center. Thus employees in the lower end of the wage distribution are likely to
gain substantially more in income due to unionization than those in the upper end. In
context of median voter models, this interpretation implies that ﬁrms whose production

units include low—skill majorities, that is ﬁrms with many workers in the left tail of the

 

10 For example, through rate standardization across and within establishments. See the
discussion in Freeman (1980).

14

earnings distribution, will be more likely to vote for unions due to their self interested
pursuit of union wage gains, which is what Farber and Saks ﬁnd.

However, Parsons (1991) has shown that the relatively egalitarian wage policies
of unions are not merely a function of the skill distribution of union members. Parsons
demonstrates that, while ﬁrms with a high concentration of highly skilled workers are
less likely to vote for unions, they are not likely to change the tendency of unions to
redistribute rents disproportionately to low wage workers once unions are established.
One explanation of Parson’s results is that union employees are frequently concerned
with matters of fairness in the distribution of wages, perhaps to preserve solidarity with
low-wage workers in order to maintain a tenable strike threat.

Unions also tend to close the pay gap between union and nonunion workers
within ﬁrms (Freeman, 1980). In Freeman’s estimates, unionized workplaces in the
manufacturing sector have a white collar-blue collar pay differential that is 30 percent
lower on average than that in nonunion workplaces. It is not clear from his analysis if the
reduction in the occupational wage differential comes fully from increases in union
wages, or if it comes partly through decreases in white collar wages. That is, unions could
have a negative or positive spillover effect on the wages of white collar workers within
ﬁrms if wages among white collar workers in those ﬁrms are not determined purely by
the market”.

Unions also decrease the use of performance incentives among members. This

may be due to perceptions by union members that management is incapable of

 

H Hirsch and Addison (1986) note that union wage gains must come from reductions in
returns to capital, product price increases, or lower pay to nonunion labor. The high
magnitude of the Freeman estimate, however, suggests that unions on average have a
negative spillover effect.

15

impartially evaluating individual performance, perhaps because of favoritism, team
oriented production technology, or because such mechanisms may weaken the strike
threat by introducing competition among workers and negatively impacting solidarity.
Farber and Saks (1980) present evidence that workers who believe that unions will
improve fairness, vis a vis their treatment by supervisors, are more likely to vote for
unions. Again, to the extent that such adjustments in the compensation of union members
are generalized to white collar workers in the same firm, executives in unionized ﬁrms

will likely have a weaker pay-performance relationship”.

Unions and ﬁrm performance

If political factors impact the pay of executives, then the union voice function will
likely affect the pay of executives in the manner predicted by the implicit regulation
hypothesis. However, the voice mechanism is not the only factor at work which will
lower the pay of executives. As in the case of economic regulation, unionization may
impact the decisions that executives make, thereby decreasing their relative productivity.
Hirsch (1991) presents a model of union rent seeking behavior where unions negatively
impact the investment and research and development decisions that are made in
unionized ﬁrms. Unions, through the exercise of the strike threat and myopic behavior
due to the nontransferability of union membership rights, will try to capture the rents
generated through such expenditures and investments. Thus, unions essentially impose a

tax on investments, resulting in lower equilibrium levels of investment in union ﬁrms. In

 

‘2 Indirect evidence of this comes from the “Multi-City Study of Urban Inequality,”
where recent hires hired into ﬁrms with collective bargaining units are substantially less
likely to receive pay increases based on performance. This is true, as well, for individuals
hired into managerial and professional jobs whether the job requires a college degree.

16

empirical tests, Hirsch shows that the level of unionization negatively affects the capital
investments a ﬁrm makes (though unions will still increase the capital labor ratio), its
research and development expenditures, and a host of other related “behavioral”
outcomes dependent on executive decisions”.

As in the problem outlined in other papers (J oskow, Rose, and Shepard, 1992, and
Joskow, Rose, and, Wolfram, 1996), identiﬁcation of a negative union impact on
executive pay may imply a regulatory effect or a productivity effect. To the extent that
boards of directors recognize the impact on executive behavior imposed by union rent
seeking, they will adjust compensation packages accordingly and hire lower quality
executives. Again, this distinction is important to the extent that one would like to use
variation in political institutions to identify rent appropriation on the part of chief
executive officers. That is, if executive talent is roughly equal across ﬁrms that differ
only in their levels of unionization, higher executive pay in unionized ﬁrms may indicate
that those executives are successfully appropriating shareholder rents. Practically,

measuring all the variables necessary to make such a ceteris paribus comparison is likely

to be prohibitively expensive and is not undertaken in this analysis.

Countervailing factors

The effect of union voice on executive pay will also be difﬁcult to identify since
there are likely to be numerous other unobserved factors that are correlated with
unionization and which may increase the pay of the executives in union ﬁrms. For

example, executives with the skills needed to interact effectively with a unionized work

 

'3 For example, investment intensity, patents per dollar spent on research and
development, advertising intensity, and debt-equity ratios.

17

force may command a premium in the market for executives. Or rather, the added
responsibilities of negotiating with unions may necessitate a compensating differential be
paid to executives in unionized ﬁrms. Additionally, ﬁrms that are targets of union
organization may also be those with more able managers. Managers that are especially
adept at generating extraordinary economic proﬁts through acquisition of market power,
or maintenance of other quasi-rents, may be paid more and be the leaders of ﬁrms that
are likely to be targets of certiﬁcation drives. If such is the case, higher executive pay can
be expected in unionized firms. Hirsch and Addison (1986) also note that management
can become more productive after employees have established collective bargaining
units, citing evidence that managers respond to decreased proﬁt expectations after the
establishment of unions by cutting costs, improving monitoring, and making other
productivity enhancing adjustments. Thus, one might expect higher executive pay in
unionized ﬁrms, which may be merited based on the higher productivity of those
managers.

Alternatively, high executive pay in unionized firms may reﬂect rent sharing with
union members due to product market power or the existence of quasi-rents that are not
necessarily related to the quality of the current executive. Firms in less competitive
product markets, with high ﬁxed-cost production technology, or with lower cost
structures may be more likely to be targets of successful union membership drives.
Executives in such firms may be paid more than their counterparts in more competitive
product markets due to rent sharing arrangements among the various stakeholders in the

ﬁrm.

18

These various possibilities make it difficult to predict the effect of unionization on
CEO pay. CEO pay should be negatively affected by the presence of unions through the
inﬂuence of social pressures or through their impact on executive investment behaviors
and the ﬁnancial performance of the firm. On the other hand, differences in unobserved
ability and rent sharing may result in higher executive pay in unionized ﬁrms. In the
absence of good measures of executive ability and ﬁrm-specific rents, it may be difﬁcult
to discern the possible political impact that unions have on executive pay. With these
limitations in mind, the goal of this paper is to assess the direct effect of unionization on
CEO pay and to explore the possibly different avenues through which the CEOs of more

unionized ﬁrms are paid relative to those of less unionized ﬁrms.

Data

Data for this analysis are drawn from several sources. Executive compensation
data were collected from ﬁrm proxy statements ﬁled with the SEC and archived by the
Q-sup data company. When proxy statements were unavailable, form lO-k were checked
for executive compensation information. Annual data on CEO compensation were
collected for 12 years (1980-91), with a mean of 9.8 years per ﬁrm available. Additional
information on the percentage (if greater than 1 percent) of the firm’s common shares
owned by both the CEO and the board of directors was collected, as well as information
on the CEO’s tenure as CEO, his age, and the number of years as a board member.
Missing observations in these data are the result of missing microﬁche, proxy statements

not archived by Q-sup data company, ﬁrm attrition from the sample due to mergers,

19

bankruptcies and closings, or some other event which removed the firm from SEC
reporting requirements.

Measures of unionization come from a 1987 mail-phone survey of publicly traded
manufacturing ﬁrms conducted by Barry Hirsch. Hirsch asked ﬁrms the following
question: "To the best of your knowledge, approximately what percentage of your
corporation’s total North American work force is covered by collective bargaining
agreements (p.26, 1991) ?” Respondent were also asked to recall as best as they could the
proportion of covered employees in their firm in 1977 (as it existed then). The measures
of unionization and sample methods are explained in greater detail in his book, Labor
Unions and the Economic Performance of Firms (1991). In total, executive compensation
information and unionization measures were matched for 249 ﬁrms. Of these, roughly
150 ﬁrms had observations for executive compensation in both 1980 and 1987. For the
bulk of analysis below, the sample is constituted of these 150 firms, with the 1977
measure of unionization assigned to 1980 measures of executive compensation and the
1987 measure matched to 1987 measures of executive compensation.

The third source of data is Standard and Poor’s Compustat. Compustat provides
ﬁnancial information from 10K forms, quarterly reports, and annual reports submitted to
the SEC by publicly traded firms. A wide variety of information is available from this
source. The following variables are used for the present analysis: annual sales, annual
market return, number of employees, labor related expenditures, state where corporate
headquarters are located, and SIC. All dollar measures are adjusted to 1991 values using

the consumer price index for urban areas.

20

Econometric Model and Predictions

The model used here to analyze the compensation of chief executive officers
follows that presented in Joskow, Rose, and Shepard (1993). The following equation will
be used to estimate the determinants of chief executive pay,

$8,], = a0 + (1, SIZE,, + a2 PERFﬁ + a3 HC,-f, + a4 OWNm +

a5 OUT”, + or, UNIONif, + a7 Xif, + em, (1)

where i denotes chief executive, f denotes ﬁrm, and t denotes year. SB represents the
natural log of chief executive salary plus bonus. The total of salary plus bonus must be
used, because salaries and bonuses were tabulated separately in only 18 percent of proxy
statements. While this is a standard problem when using information from proxy
statements (or other sources which use proxy statement as their source), it is unfortunate
since the use of base salary versus bonuses might be substantially different between
union and nonunion ﬁrms.

Also, this measure excludes noncash compensatory rewards associated with long
term incentive programs, such as the cash value of stock appreciation rights, stock
options, and other related stock grants. Reporting requirements during the sample period
make it difﬁcult to identify annual measures for this variable. The SEC required that
companies report the number and value of options granted as well as the realized value
from previously awarded options separately from other forms of compensation, though
companies frequently failed to tabulate the awards and were allowed to report cumulative
awards over several years. This particular shortcoming is especially important given
recent ﬁndings by Hall and Liebman (1997) that failure to incorporate the valuation of

longterrn awards results in a serious understatement of the sensitivity of executive wealth

21

to ﬁrm performance. The literature on executive compensation has long recognized the
inherent difﬁculty in combining long term awards with annual compensation, though
Jensen and Murphy (1990) and Hall and Liebman (1997), among others, have
successfully merged them. '4

SIZE denotes scale variables, which will be represented by the natural log of
annual sales in the following analysis. In the empirical literature, scale is controlled for in
several ways. Joskow, Shepard, and Rose (1993) use the natural log of employees and the
natural log of total assets together to control for scale. They argue that these measures
better control for the heterogeneity in the capital-labor ratio among the different
industries in their sample. Since the sample used in this analysis is constituted only of
manufacturing ﬁrms and is therefore more homogeneous in terms of production
technology, the sales measure will be used to control for scale effects. With these data,
speciﬁcations that include assets and employees instead of sales generally have a worse
fit in terms of R2 than the specifications that include the sales variable. Annual sales is
predicted to have a positive effect on the pay of executives, while there is no a priori
reason to believe that the scale elasticity will be affected by unionization.

PERF represents performance measures and is measured by annual market return.
Annual market return is provided by Compustat and is deﬁned as the percentage change
in closing share price from the previous year, accounting for per share dividends paid.

Results incorporating accounting rates of return are not presented at this time, though

 

'4 The conversion of long term awards to current cash value is not possible with these
data, however, due to missing information on the number of outstanding awards. For the
present, it is assumed that the error in measurement for the dependent variable is
uncorrelated with the measure of unionization, so that estimates presented below are
unbiased. Whether in fact this is the case is an interesting empirical matter.

22

such measures are likely to be relevant sources of information regarding the performance
of chief executive ofﬁcers (Rosen, 1992, and Holmstrom, 1992). Performance measures
are hypothesized to have a positive effect on executive salary. However, the implicit
regulation hypothesis suggests that unionized ﬁrms should have a weaker pay-
performance relationship since performance based pay should result in larger payoffs for
executives and higher average pay.

Human capital variables are represented by HC. The linear and quadratic
measures of tenure in the position of CEO and potential experience (defined as CEO age
minus 22) are used in this analysis. Both tenure and experience are expected to have a
positive effect on compensation and are interpreted in the usual Mincer (1974)
framework, with the coefﬁcient on experience representing returns to general human
capital, and the coefﬁcient on tenure representing the returns to job-speciﬁc human
capital. Joskow, Rose, and Shepard (1993) ﬁnd that the return to tenure in regulated ﬁrms
is twice that in nonregulated ﬁrms. They argue that this may represent a shift to more
bureaucratic forms of pay, for which the public has less objection. An alternative, though
not contradictory, explanation is that increased returns to tenure with a simultaneous
decline in the strength of the pay-performance relationship represents a shift from the use
of performance based incentive mechanisms to the use of bonding mechanisms.

Such a model has been presented by Lazear (1979), who argues that ﬁrms may
induce effort from employees by paying them less than their productivity early in their
careers and more than their productivity later in their careers. This will result in the slope
of the pay-tenure relationship being steeper than the slope of the productivity-tenure

relationship. In such a scenario, an employee essentially posts a bond early in her career

23

by working for less than she is worth. If she is caught shirking, she will be fired and
consequently have to forego the higher pay late in her career. Thus, if unions perform an
implicit regulatory role as described by Jensen and Murphy (1990) and have less
objection to predeﬁned, or bureaucratic, increases in pay, the tenure-pay relationship
should vary with unionization”.
OWN represents two sets of dummy variables for various levels of chief
executive common stock ownership and the common stock ownership of the board of
directors. The sign of the CEO variables is difficult to determine, a priori. Since high
levels of CEO ownership represent a substantial vested interest in the success of the ﬁrm,
executives who own more of the ﬁrm’s stock should have greater incentive to perform
well and therefore should have higher pay. However, failure to ﬁnd a positive
association between these variables and pay may not accurately reﬂect the relative
productivity of high ownership executives since they may have greater lifetime wealth
which annual pay measures are not accurately reﬂecting. It may be that for high
ownership variables, pay in ﬁrm shares substitute for annual cash compensation. An
alternative interpretation of these variables is that very high levels of ownership represent
executive entrenchment. Presumably, executives can enter their positions in the ﬁrm
through various avenues. Theoretically, the market for executives sorts the most
productive people into their most productive positions. Executives can enter into jobs less

efﬁciently, however, through mechanisms such as old boy networks or nepotism. Also,

executives who are ﬁrm founders may stay in their positions long after their value in the

 

'5 This prediction follows from J oskow, Rose and Shepard’s (1993) ﬁndings.
Theoretically, it is not clear to what extent one might expect one form of incentive
mechanism to substitute for others. In fact, the use of one mechanism may be
complementary to the use of others, or they may be independent.

24

market for executives has declined by virtue of their control over large portions of firm
stock”. The market for corporate control should discipline such inefficiencies through
takeover bids, though takeovers are likely to have substantial transactions costs and will
only be taken if the beneﬁt of acquiring the ﬁrm’s assets (the increase in the ﬁrm’s
market valuation resulting from the takeover) exceeds the associated transactions costs.

Ownership variables for the directors (net of CEO ownership) is expected to
increase pay. This is due to the likelihood that directors with a large portion of their
personal assets vested in the performance of the ﬁrm will be more diligent in searching
for skillful executives. Thus, the OWN variables should proxy for executive effort and
ability.

OUT is a dummy variable indicating whether the executive officer was a member
of the board of directors for three or less years before acquiring his position as CEO.
Since the average board tenure before becoming CEO is roughly six years (though the
median is slightly more than 4 years), short board tenure may be an indication of
executive quality. This variable measures whether the executive is on a “fast track” and
should indicate some extraordinary ability. It is predicted to have a positive effect on
executive compensation, and may mitigate any omitted variable bias from executive
quality which the measures for unionization pick up”.

UNION represents the measure of unionization previously described. The

variable, following Hirsch, is included in the following analysis in its continuous form

 

16 The current data set does not have information either on whether the CEO is the
founder of the company or on the familial relations among the members of the board of
directors.

'7 This will be the case if the variable is also positively correlated with unionization, that
is if unionized ﬁrms are more likely than nonunion firms to select executives from
sources outside the boardroom.

25

and in several categorical breakdowns. The implicit regulation hypothesis suggests that
the pay of chief executives in unionized ﬁrms should be lower due to union social
inﬂuence and union effects on ﬁrm performance. As outlined above, however, the sign of
this measure is difficult to determine, a priori, since factors such as unobserved executive
ability and union selectivity suggest a positive relationship between unionization and
executive pay.

The remaining control variables, represented by X, include year controls, industry
controls (9 categories), and a dummy variable for the ﬁrst, incomplete year of tenure for

the chief executive.

Econometric problems

The most important issues in the following analysis are related to measurement
error in the independent variable and omitted variable bias, most notably pertaining to
unmeasured executive ability and unmeasured ﬁrm-speciﬁc rents. Measurement error in
explanatory variables causes bias in OLS estimates. In the present analysis, this may be
especially important given there are three possible sources of measurement error in the
measure of unionization. First, the measure of unionization for 1977 is used with other
variables which are measured in 1980. Thus, there will be error in the measure of
unionization for those firms whose unionization rate changed between 1977 and 1980.
Second, both of the measures of unionization were acquired in a 1987 mail and phone
survey. Since the variable was directed to individuals in surveyed ﬁrms, as opposed to
objective measures of unionization, the measure is subject to respondent error (since any

two respondents in the same ﬁrm are likely to give different estimates). The third source

26

of measurement error is due to recall error, since the 1977 measure of unionization was
collected in a 1987 survey. Measurement error in OLS regressions is generally thought to
bias the estimated coefﬁcient toward zero and may cause bias in the estimated effects of
variables not measured with bias. The first of these problems will not be addressed in this
paper and the second does not seem to be a problem.

Estimating differences in compensation levels among different groups based on
some observable characteristic which varies across those groups is problematic whenever
unobserved characteristics exist that are correlated with both the dependent and
independent variables. As outlined above, potential sources of omitted variable bias may
come from factors at both the executive and ﬁrm level. At the firm level, the existence of
ﬁrm-speciﬁc rents will be correlated with unionization and with the compensation of
executives. Since the regressions reported here do not control for such rents”, the
estimates on unionization will be biased upward. At the level of the executive, if there are
unobserved executive characteristics positively correlated with firm performance and
unionization, then the coefficient on unionization will be biased upward if that ability is
not adequately measured by observable variables. Omitted variable bias can be addressed
by several methods. To the extent that those variables are ﬁxed over time, they can be
controlled for using firm-level or executive-level fixed effects.

The existence of union density measures for two years affords the panel nature of
the data to be exploited for this purpose. In such a case, the error term in Equation 1 is

assumed to take the structure 8,], = ¢f+ Um, where ¢f represents the unobserved, firm-

 

‘8 Rents are inherently difﬁcult to identify, ex ante, and are generally inferred from
performance or wage changes following changes the economic environment. For
example, see Rose (1983).

27

speciﬁc component. Fixed effects estimates are eliminated by differencing the time
varying variables or including ﬁrm-specific dummy variables in regressions. In order to

control for ﬁrm-level omitted variable bias, firm ﬁxed effects estimates are presented

below.

Effect of Unions on Chief Executive Pay

Table 1.1 presents sample means and standard deviations for compensation
variables. Results are presented for all firms and broken down by unionization categories,
where union categories are deﬁned as of 1977. The sample consists of averages on both
executive and average employee compensation for the years between and including 1980
and 1987. The dollar values are adjusted to 1991 dollars using the consumer price index
for urban areas. The sample for chief executive ofﬁcers includes observations for only
those years where the CEO worked the entire year, and the growth estimates for CEO’s
are CEO speciﬁc. Sample sizes differ between levels and changes because of missing
values, and they differ from regression samples in subsequent tables because the years
between 1980 and 1987 are included in Table 1.1.

On average, executives in this sample earned a little over $700 thousand per year,
while average compensation expenditures for other employees averaged $41 thousand per
annum, a ratio of 17 to 1. The ratio of CEO to average pay increased over the years,
going from 15 to l in 1980 to 19 to l in 1987, which is reﬂected in the percentage change
measures.

Levels of CEO pay differ across unionization category, though much of this

difference can be accounted for by ﬁrm size. Still, among unionized ﬁrms, executive pay

28

does decline as unionization increases, though the differences are not especially large. Per
employee expenditures also vary by union category, with the highest level occurring in
the most densely unionized firms. Interestingly, compensation expenditures per employee
do not rise monotonically with union density, since ﬁrms with 1 - 40 percent of their
work force unionized have lower average expenditures than nonunion ﬁrms.

Results in Table 1.1 also show that CEO compensation grew over the sample
period at an annual rate of 7 percent on average. This contrasts with the .5 percent growth
in average employee compensation expenditures during the same period. This sharp
contrast in the growth of pay understates the true difference in compensation
expenditures since the measures for executive compensation do not include the realized
or awarded value of stock grants and options for the chief executives. By unionization,
growth in pay differs for both executives and employee expenditures. Executives in ﬁrms
with medium levels of unionization had the highest rate of growth, as did the pay of their
employees. Employees in high union ﬁrms had negative compensation growth, though
they had the highest average pay levels. These results suggest that unions decrease the
executive employee pay ratio, but that the ratio is increasing faster than ﬁrms with less
unionization.

Table 1.2 presents summary statistics for firm and executive characteristics. In
this table, it is interesting to note the nonunion ﬁrms seemed to enjoy the greatest success
in terms of sales, employment, and asset growth during the sample period. These firms
also had higher average market returns than the unionized ﬁrms. Firms with medium
unionization, on the other hand, had the most dramatic drop in sales and employment,

suggesting negative demand shocks, though this is in contrast to the positive wage growth

29

reported in Table 1.1. Densely unionized firms also had negative average sales and
employment growth. Nonunion firms are smallest on average. Means on densely
unionized ﬁrms indicate they are the largest, though standard deviations suggest that the
means are heavily affected by outliers.

Consistent with the results presented in Becker (1987) and Hirsch (1991),
unionization is negatively associated with market returns. Average market return
decreases monotonically with the level of unionization. Executives of nonunion ﬁrms
take their positions at an earlier age and have longer tenure, though this result is likely to
be a function of ﬁrm size. Likewise, executives and board members of nonunion ﬁrms
hold a large portion of the ﬁrm stock than executives and board members in unionized
ﬁrms. Among unionized ﬁrms, there is no clear pattern of share ownership which
emerges. Since differences in the summary statistics are only suggestive (and frequently
not statistically signiﬁcant), the differences evident in tables 1 and 2 must be examined
further using regression analysis.

Table 1.3 presents results from OLS regressions where the dependent variable is
the log of annual salary plus bonus for chief executive officers. The sample size is
smaller than in Tables 1 and 2 because it includes data from 1980 and 1987 only. This
change is due to the fact that there are only two years of observations for the union
variable. Four speciﬁcations are presented, with the ﬁrst excluding unionization; the
second including the linear measure of unionization; the third including the linear and the
quadratic measures of unionization; and the fourth including categorical variables for

unionization.

30

Estimates of the size elasticity are .26 and strongly signiﬁcant. This estimate
matches the .25 - .30 values presented in other analyses (Rosen, 1992), and says that the
pay of chief executives increases roughly 2.6% with every 10% increase in ﬁrm size.
This elasticity changes little across the OLS speciﬁcations presented. The sensitivity of
executive pay to changes in the market value of the ﬁrm is estimated at .06, implying a .6
percent increase in executive pay for every 10 percentage point increase in market
returns. This point estimate is smaller than expected, being about half that estimated in
other papers that use roughly contemporaneous data. J oskow, Shepard, and Rose (1993),
for example, report pay sensitivity estimates for market performance of around .10 from
their OLS estimates. Rosen reports estimates from the literature generally range from .10
- .15 in both OLS and ﬁxed effects estimates. The low estimated market return increases
to .12 when pooling across additional years of data, suggesting that the selection of years
may have some impact on this estimate. Estimates on human capital variables are the
expected sign and a reasonable magnitude. Estimates suggest that returns to tenure peaks
at 28 years for executives, while returns to experience peak at 36 years.

The relationship between unionization and executive pay in column two suggests
that unionization is negatively associated with executive pay. The OLS estimate suggests
that going from zero to a work force that is fully unionized results in a decline in
executive pay of roughly 10 percent, though the estimate is only marginally signiﬁcant.
The negative coefficient is consistent with the implicit regulation hypothesis, though it is
also consistent with executives in unionized ﬁrms being less productive, perhaps because

union rent-seeking reduces the impact executives can have on ﬁrm success.

31

Results in Column 2 mask a nonlinear relationship between unionization and
executive pay that is apparent in closer inspection of the data. Column 3 presents results
from regressions including the continuous measure of unionization and its squared value.
The inclusion of the squared measure of unionization has little impact on the estimates of
other variables, though it does suggest a rather strong nonlinear relationship between the
level of unionization and executive pay. The estimate on the linear term of unionization is
positive and signiﬁcant at the 10 percent level (the linear and quadratic terms are jointly
signiﬁcant at the 5 percent level of significance). Estimates suggest that executive pay
rises sharply with unionization of the work force, until it is 28 percent unionized, at
which point further unionization starts to decrease executive pay. The relationship
between unionization and executive pay falls quickly, though it remains positive until
roughly 60 percent of the work force is unionized”. This nonlinear relationship suggests
that unions, per se, do not have a negative impact on executive pay until they reach fairly
high levels of union density. This has also been found by DiNardo, Hallock, and Pischke
(1997) in an analysis similar to this one.

The fourth column of Table 1.3 inspects the effect of unionization on CEO
compensation using categorical variables for unionization. The four categories of
unionization are nonunion ﬁrms, ﬁrms with 1 to 40 percent of their work force unionized,
ﬁrms with 41 to 70 percent of their work force unionized, and ﬁrms that are more than 70

percent unionized. Results reﬂect those presented in Column 3, since there appears to be

 

'9 Plotting the residuals of a regression excluding any of the union variables against
unionization reveals that the negative relationship between unionization and executive
pay is not a function of outliers. Eliminating one observation that has an especially
negative error term (a ﬁrm with a high level of unionization) changes the point estimates
somewhat, but the results are qualitatively the same.

32

a strong nonlinear relationship between union density and the pay of executives. OLS
results show that ﬁrms with l - 40 percent of their work force unionized tend to pay their
executives seven percent more on average (with marginal signiﬁcance) than nonunion
ﬁrms. Firms with 41 to 70 percent unionization pay their executives roughly the same as
nonunion ﬁrms, and ﬁrms with the highest levels of unionization pay their executives
signiﬁcantly less than the executives of nonunion ﬁrms. Executives in highly unionized
ﬁrms are being paid nearly 17 percent less than executives in nonunion ﬁrms (the entire
set of dummy variables are jointly different from zero at 1 percent conﬁdence level).

The initial positive association between unionization and executive compensation
is not likely to reﬂect a causal relationship. Such an interpretation would require that
either unions are increasing the productivity of chief executive ofﬁcers, or, at least, that
boards of directors must compensate executives for the disutility of interacting with
unions. Neither of these explanations would explain why the initial positive effect
decreases with further increases in unionization. More plausible explanations of high
executive compensation with low levels of unionization are that unions target ﬁrms with
more capable managers or that the presence of unionization indicates that there are
unobserved rents that are also increasing executive pay. The two sets of ownership
variables and the dummy variable for short board tenure are intended to proxy for
executive ability. Since they are not likely to be correlated with unionization, they will
not control for those characteristics which unions presumably target. In fact, their
exclusion has little impact on the presented estimated effects for unions.

The initial positive association between unionization and executives most

probably reﬂects the existence of unobserved rents, while the decline in executive pay

33

with further unionization indicates unions capturing those rents. Whether this eventual
negative impact of unions is also due to unions exerting a regulatory effect on the pay of
chief executives or a productivity effect is not immediately apparent. Arguably, the role
of productivity differences between executives within the manufacturing industry is
likely to be less serious than that difference between executives in regulated and
nonregulated industries (although the estimated difference is smaller, as well). In
regulated industries, the cost allowance and price setting functions of regulatory agencies
may allow those monopoly ﬁrms to overcome the impact that relatively poor executive
performance has on the ﬁrms’ ﬁnancial market performance. Manufacturing ﬁrms do not
have such latitude, since poor performance which results in price increases or the decline
in returns to assets will quickly drive ﬁrms from the market. The fact that chief
executives in unionized ﬁrms seem to behave differently in terms of investment decisions
and other matters does not necessarily imply that they are less productive in maximizing
shareholder value, since maximizing shareholder value implies that executives will
engage in only those activities where the marginal value exceeds the marginal cost.
Knowing which projects not to undertake is as important an knowing which projects to
undertake. Thus the negative impact of unions on executive pay is likely to be the result
of a regulatory effect.

To discern whether the second prediction of the implicit regulation is at work in
unionized ﬁrms, separate regressions were run by unionization category on a model
similar to that presented in Table 1.320. The implicit regulation hypothesis predicts that

the pay of executives will be less tied to the performance of the ﬁrm as political forces

 

2° The model exclude ownership variables and the dummy variable for less than three
years board tenure. The union categories are those presented in Table 3, Column 4.

34

exert pressure on ﬁrms to temper executive pay. The sample for these regressions
includes all observations from 1980 through 1987 for those firms where reported levels
of didn’t change categories unionization (with the levels defined as in Column 4 of Table
1.3). The results of both OLS and ﬁxed effects regressions suggest that pay-performance
sensitivity is highest for executives with the highest levels of unionization and lowest for
nonunion ﬁrms, though the difference between them is not statistically significantZI. Low
and medium unionized ﬁrms do have pay that is more sensitive to performance than is
the pay of executives in nonunion firms, at marginal levels of signiﬁcance (the point
estimates lie below the levels of the most unionized ﬁrms, however).

These results suggest that relative to nonunion ﬁrms, executives in unionized
ﬁrms have stronger pay performance relationships in the measure of salary plus bonus, in
direct contradiction to the implicit regulation hypothesis. Among unionized ﬁrms the
pay-performance relationship doesn’t seem to fall substantially with unionization, though
estimates initially fall and then become statistically weaker as ﬁrms become more
unionized. On the other hand, the coefﬁcient on the tenure variable does increase
monotonically with the level of unionization in both OLS and ﬁxed effects results.

The combination of results from these regressions suggests that boards of highly
unionized ﬁrms may use Lazear-type incentive mechanism as pay for performance

relationships become less reliable. In this sense, the results are weakly consistent with the

 

2‘ The estimates are (standard errors in parentheses):

Nonunion Low Medium High

OLS: Tenure .003 (.009) .017 (.012) .034 (.015) .043 (.027)
Return .057 (.043) .205 (.037) .153 (.026) .248 (.171)
FE: Tenure .016 (.006) .026 (.006) .052 (.014) .068 (.037)
Return .067 (.025) .147 (.027) .126 (.039) .269 (.114)

35

steeper tenure proﬁle for executives in regulated firms found by J oskow, Rose, and
Shepard (1993). The results may also indicate that the complexity of the job for
executives increases with unionization, and that executives in these ﬁrms experience
much on the job training.22

The ﬁnding that the pay of executives in unionized firms is more sensitive to
performance also suggests that the implicit regulation hypothesis needs to be further
formalized. It is not necessarily the case that union members will not want the pay of
executives to be insensitive to performance, especially if making pay sensitive to
performance is an effective method by which ﬁrms generate extraordinary proﬁts. If this
is the case, then unions can capture some of these rents through exercise of the strike
threat. Of course, reconciling higher pay variability with lower average pay is
problematic, since on average greater income risk should be offset by higher average
executive pay.

The preceding analysis must be qualiﬁed with respect to the failure to incorporate
long term incentive pay into the dependent variable. As Hall and Liebman point out, “the
large literature that measures pay-to-performance sensitivity with salary and bonus
elasticities should be interpreted with the important caveat that salary and bonus

sensitivity is only the tip of the iceberg (p.30 1997).”

 

22 This result implies that, ceteris paribus, the union pay differential closes over time,
which is exactly the case. Executives in the ﬁrms with the highest level of unionization
and with greater than the median tenure have a smaller wage differential than executives
in high union ﬁrms with low levels of tenure. The reason for this is not immediately clear
and merits further consideration, though the use of Lazear-type bonding mechanisms
seems a compelling explanation.

36

Unionization and pay equality

Table 1.4 presents results for the effect of unionization on the relative pay of ﬁrm
employees. Freeman has shown that unions tend to decrease the pay differential between
unionized workers and white collar workers by 30 percent. Joskow, Rose, and Shepard
(1993) note that the pay of blue collar workers in regulated industries is generally higher
than the pay in nonregulated industries, and that regulatory agencies tend to be more
tolerant of relatively high compensation for average employees (presumably for
efﬁciency wage types of reasons, because of regulator capture, or because of
redistributive preferences of regulators) than for executives. Thus, part of the regulatory
effect may be in the emphasis on decreasing the relative pay of executives within ﬁrms.
High pay for chief executive officers may be palatable to union members if their pay is
also relatively high. In order to examine the effect that unions have on pay inequality
within the ﬁrm, the following equation will be used to estimate the ﬁrm level-
deterrninants of compensation expenditures per employee,

PAY}; = 80 + [31 SIZE,«, + [3; UNION]; + B3 X]; + pm (2)

where f represents ﬁrms and t represents years; PAY is per employee labor related
expenditures; SIZE is the log of annual sales, and is used to control for firm size, UNION
represents the three measures of unionization which were used in the analysis of
executive compensation; and X represents dummy variables for year, industry, and region
of ﬁrm headquarters. These estimates are then subtracted from the corresponding
estimates presented in Table 1.3, which is comparable to regressing the log of the ratio of

executive pay to average employee compensation on the control variables”.

 

23 In fact, regressing the log of the ratio on the variables in (2) changes the point estimates

37

Results in Column 1 show that the pay of employees is positively and
signiﬁcantly related to ﬁrm size, with the point estimate roughly consistent to that found
in other studies (for example, Hansen, 1996). This effect is attributable to a number of
explanations, such as the use of efficiency wages to solve monitoring problems in large
organizations, rent sharing arrangements, or union threat effects.

Including the linear measure of unionization decreases the size elasticity by .008.
The coefﬁcient on unionization is nearly positive 20 percent, reﬂecting the differences in
employee pay levels reported in Table 1.1. However, this speciﬁcation also masks an
apparent nonlinearity in the effect of unionization on average employee pay. Linear and
quadratic estimates suggest that the effect of increasing union density from zero is to
initially decrease the average compensation expenditures, suggesting that there is not
necessarily a strong union threat effect within ﬁrms. Alternativey, this may indicate that
at low levels of unionization, management is able to adjust its behavior to eliminate
inefﬁciencies elsewhere in the ﬁrm in order to maintain ﬁrm proﬁtability (Addison and
Hirsch, 1986). However, as union density increases (estimates suggest that the average
pay is at its lowest when 30 percent of the work force is unionized), further unionization
increases average compensation expenditures. Estimates in Column 4 underscore this
pattern, with low levels of unionization decreasing average compensation expenditures
(though insigniﬁcantly) and high levels increasing the average (with marginal

significance).

 

little, though sample size falls and standard errors are larger. Also, regressing CEO
compensation on the variables in (2) and subtracting the estimates for average employees
changes the results little.

38

It is immediately apparent that the effects of unions on the pay differential
between the executives and average employees is going to reﬂect the results in Table 1.3.
Subtracting the coefﬁcients from those results gives the effect of the variable on
increasing or decreasing relative pay. These results, and their standard errors (calculated
as (varcgo+varEMp)'5), are given in columns 1’ - 4’. Estimates suggest that ﬁrm size
increases executive pay faster than it does the pay of average employees. For
unionization, the estimate in Column 2’ suggests that unionization dramatically decreases
the inequality between average and executive pay. This occurs both through decreasing
executive pay and increasing the pay of average employees. The estimate presented in
Column 2’ of -.31 is close to the effect reported in Freeman (1980) for the decrease in the
union employee-white collar pay differential. Column 3’ estimates include both the linear
and squared measure of unionization. These results indicate that relative pay in ﬁrms with
low unionization is higher than in nonunion ﬁrms. It is also substantially higher than in
highly unionized ﬁrms. The categorical variables also reﬂect this relationship, with the
highest union category closing the gap between executives and their employees by 28%,
which is nearly the same as the Column 2 estimate. These results suggest that unions may
act as implicit regulators by decreasing the pay gap between union members and
executive ofﬁcers. To the extent that the pay of unionized employees in low-union ﬁrms
is increased to the levels reﬂected in densely unionized ﬁrms, there may be little union

pressure to restrain executive pay in those ﬁrms.

39

Fixed effects estimates

The results presented above indicate that at high levels of unionization, unions
have moderately strong negative effects on the pay levels of chief executive ofﬁcers,
while at low levels there is a positive association between unionization and chief
executive pay. The positive results in these lower categories may be biased due to
unmeasured variables correlated positively with both unionization and executive pay. To
the extent that such variables are time invariant, they can be controlled for using ﬁrm
ﬁxed effects techniques.

Before turning to the results from fixed effects regressions, however, it is useful
to inspect the degree of variation in the unionization variable over time. Appendix Table
1.1 provides summary statistics on the changes in unionization that were reported by the
ﬁrms in the sample. For the sample in Table 1.3, changes in unionization are given in
Panel A. Of the 147 ﬁrms with observations on all variables in both 1977 and 1987, the
mean change in unionization was a decline of nearly 5% (given in Panel A.l). Forty eight
percent of ﬁrms reported a decline in unionization, while only 12% reported an increase.
In total, 60 percent of ﬁrms reported a change in unionization over the sample period.

Changes in categories are given by the cross tabulation in Panel A.2. Of the 147
ﬁrms in both years, 30 of them reported a change which put them into a different
category. Of these, 8 ﬁrms increased in category, while 22 ﬁrms reported a decrease in
category. It is worth noting that changes between categories can be smaller than changes
within categories. For example, ﬁrms that report a decline in unionization from 41% to
39% will be coded as changers, while firms that report a decline from 39% to 1% will

not. On average, however, the changers between categories experienced much larger

40

changes than those that remained in their categories. For example, the average change
for ﬁrms that declined categories was a 23 percentage point drop, while the mean for
those ﬁrms which did not change categories, but which reported a decline in
unionization, was -10 percentage points.

In ﬁxed effects estimates, the effect of an explanatory variable on the dependent
variable is identified only by those firms that report a change. Thus, much rides on the
way the regression is speciﬁed. In this case, using the continuous variable of unionization
will provide relatively more changers (88 ﬁrms report a change) than the speciﬁcation
which uses the categorical variables (30 changers). However, intuitively it seems as
though large changes are going to have more impact on the “political” inﬂuence unions
wield, or cease to wield, in determining chief executive pay than small changes in
unionization. In the regressions in Table 1.5, the results are presented for the
corresponding speciﬁcations in Table 1.3 and encompass both types of changers.

An additional concern in ﬁxed effects estimates is the assumption that the
structural relationship between unionization and executive compensation is stable over
time. The fact that the measures of unionization are ten years apart may be some cause
for concern, especially given the dramatic changes which occurred in collective
bargaining outcomes during the 1980s“. Since the estimates presented in this paper
presume to measure a “political” effect on the compensation of executives, it is
questionable whether the political impact was the same in 1987 as it was in 1980 given
the important developments that took place during that time in the outcomes of union

contract agreements. OLS regression results by year suggest, however, that the estimates

 

24 See Bell (1989) for an analysis of collective bargaining outcomes during the 19803.

41

are relatively stable both in terms of sign patterns and magnitudes. F tests on the entire
model and for the unionization variables do not reject pooling over the two years”.

Column 2 presents results from including the linear measure of unionization in
fixed effects estimates. Relative to ordinary least squares, the point estimate changes sign
and the standard error increases dramatically. The result indicates that the negative effect
in OLS estimates is biased downward and that small changes in unionization have little
impact on the pay of chief executive ofﬁcers. The standard error increases in magnitude
dramatically, indicating that the measure of unionization is very noisy, has too little
variance, or that there are two few changers to get reliable estimates. Given the survey
methods used to gather the measure on unionization and the fact that the initial year of
compensation does not correspond to the 1977 measure of unionization in these
regressions, it is unsurprising to ﬁnd such a noisy estimate. Furthermore, ﬁxed effects
estimates may be less reliable than OLS estimates in the presence of measurement error
(Hsiao, 1986). That is, the bias associated with measurement error in OLS is exacerbated
in ﬁxed effects estimates. Unfortunately, the data set used here affords only two years of
data on unionization and therefore provides little opportunity to address this issue.

Including the quadratic term allows the ﬁrms who experience big changes in
unionization to have more impact on the estimated effect. Consequently, the results seem
to reﬂect those in Table 1.3, Column 3 in that there remains a positive coefﬁcient on the
linear term and a negative term on the quadratic, though the standard errors have

increased dramatically in size. The relationship between unionization and executive

 

25 The only variable for which pooling is rejected over 1980 and 1987 is the measure of
market performance. In 1980 the effect is positive and signiﬁcant, while for 1987 it is
negative and indistinguishable from zero.

42

compensation is positive initially, but it peaks and eventually declines. Notably, these
results suggest that the effects of unionization on executive pay are positive over the
entire range of unionization relative to nonunion ﬁrms. Thus, the ﬁxed effects estimates
retain the upside-down U shaped relationship exhibited in Table 1.3, but it is much ﬂatter
than OLS estimates indicate with the ﬂatness resulting from a moderate decline in the
positive effect and a relatively large increase in the previously negative effect. Fixed
effects estimates indicate that the relatively low pay in highly unionized ﬁrms is less a
function of unions than a function of ﬁrm heterogeneity, though such an interpretation of
this bias is tempered given that the estimates are likely to be seriously biased toward zero
due to measurement error. In addition, given that the standard errors in the ﬁxed effects
estimates are rather large, too much should not be read into them.

This relationship is also reﬂected in Column 4 estimates. Although the point
estimates are much higher than the estimates presented in Table 1.3, they do increase
initially and then slowly decline. The standard errors remain very large in this
speciﬁcation. Together, fixed effects estimates suggest that OLS estimates are biased.
The bias story does not ﬁt well with that hypothesized: that ﬁrm level rents associated
positively with both unionization and executive pay were biasing the estimates of the
effect of low levels of unionization on executive pay upward. In fact, while the estimated
impact of unions does seem to shift down somewhat at lower levels of unionization, the
effect at higher levels of unionization become more positive, and at the highest levels of
unionization the effect becomes positive. For categorical measures, the effect become
larger for all levels of unionization, and becomes positive for ﬁrms with high levels of

unionization (though only one firm enters that level of unionization over the time period,

43

while seven leave). To the extent that rents are time invariant, then the results from ﬁxed
effects regressions suggest that OLS estimates are not biased upward due to unobserved
rents. However, if rents vary over time, then estimates from both ﬁxed effects and OLS
regressions are biased.

Fixed effects estimates control for unobserved ﬁrm heterogeneity. Also of interest
is how changes in unionization effect the pay of executives. To explore the effect that
increases or decreases in unionization, or large versus small changes, have on the pay of
executives, the following model was estimated:

2153);: [30 + ,6, AX,-f+ [32 AUNIONf+ vif, (3)
where .458 represents the difference in the log of executive pay between 1987 and 1980,
AX represents the difference in other time varying variables (here X represents log of
sales, market return, tenure, and experience and their squared terms), and AUNION
represents various speciﬁcations of the changes in unionization. For example, AUNION
may represent the simple difference in the measures of unionization between time
periods. More importantly, AUNION is specified as dummy variables indicating whether
the change in unionization was positive or negative, or whether the change was sufﬁcient
to change the category of unionization (for the present analysis, categories are deﬁned
differently than above, with the categories here being nonunion, 1 - 50 percent union, and
greater than 50 percent).Using this speciﬁcation allows one to isolate the effects that
changes of different direction and magnitudes had on the pay of chief executives.

Results from this exercise suggest that there is a difference between ﬁrms that
change categories and ﬁrms that report changes which don’t result in categorical changes,

though t-statistics rarely reach traditional magnitudes of signiﬁcance. Speciﬁcally,

executives of ﬁrms that report that the level of unionization declined experienced a 3
percent (s.e.=.06) increase in pay, while those in ﬁrms experiencing an increase in
unionization reported a decrease in pay of 3 percent (s.e.=10), both relative to ﬁrms that
report no change in unionization. Executives of ﬁrms that declined a category, however,
experienced a decrease in pay of 10 percent (s.e.=.09), while increasing a category
resulted in a 7 percent decrease in pay”. In specifications distinguishing between
categorical changers and ﬁrms reporting simple changes27 the effects of reporting a
change large enough to change categories is not so negative, while within category
changes resulted in a slightly higher increase in pay. In all, these results suggest that
moderate decreases in reported levels of unionization result in higher executive pay. This
result is consistent with the implicit regulation hypothesis, though such an interpretation
is unsatisfying to the extent that the political effect associated with small changes in
unionization is likely to be negligible. The negative effect of categorical declines in
unionization is inconsistent with implicit regulation, though the variable may be
indicating ﬁrms that are experiencing negative economic shocks which cause both the
level of unionization and the pay of executives to fall.

A ﬁnal specification uses 4 dummy variables indicating whether the ﬁrm reported
a large change (positive and negative) or a small change (positive and negative),

irrespective of the categories for unionization. Results from these regressions indicate

 

26 Several other speciﬁcations have been analyzed, including an exhaustive representation
of the cross tabulation matrix of possible changes (9 categories) and their interactions
with continuous changes. Due to small sample sizes, these speciﬁcations frequently
provide very noisy estimates.
7 That is, a speciﬁcation with four dummy variables (with no overlap between them)

taking the form (standard errors in parentheses):

-.O76 * Dec. Cat + .051 * Dec - .029 * Inc - .047 * Inc. Cat.

(.099) (.067) (.119) (.140)

45

that executives of firms reporting the largest declines in unionization experienced a 1
percent increase in pay (relative to executives in ﬁrms reporting no changes in
unionization), while those experiencing the largest increases in unionization experienced
a decrease in pay of four percent. Both results are not statistically different from zero.
Point estimates suggests that the magnitude of the change is less important than whether
the change put the firm in a different category for unionization. Since the categorical
break points were not formally determined in this analysis and there are not statistically
important differences between speciﬁcations, it is difﬁcult to. reach strong conclusions
from these results.

Running ﬁxed effects on per employee compensation results in similar patterns as
described for ﬁxed effects on executive compensation. Panel B in Appendix Table 1.1
gives the means on changes in unionization for the 48 ﬁrms with all observations for this
variable in both 1980 and 1987. The mean change is a decline of 4 percentage points in
the proportion of workers covered. Fifty four percent of ﬁrms reported a decline in
unionization, while 8 percent of these reported an increase. While these numbers are
reasonably close to the changes in the executive sample, the number of ﬁrms that
changed categories decreases dramatically. Of the 47 ﬁrms, only 5 reported a change in
category. Thus, ﬁxed effects estimates are ignored for specifications using categorical
measures of unionization.

The estimate on changes in unionization for the linear measure is -.18 (s.e.=.17)
indicating that higher unionization results in lower per employee compensation.
Including the quadratic term changes the linear estimate to -.49 (s.e.=.42), while the

quadratic term equals .48 (s.e.=.60). The pattern of the results of the linear and quadratic

46

speciﬁcation, while not statistically significant, follow those of OLS estimates. The point
estimates suggest that the effect of unionization is to lower employee expenditures
initially, though further unionization eventually increases pay. The estimates are dubious,
however, given that they imply lower compensation expenditures for unionized ﬁrms
through the entire range of unionization. Again, the estimates are imprecise and likely to
suffer from substantial bias toward zero due to measurement error.

Fixed effects estimates, overall, do not provide precise estimates of the effect of
unionization on executive pay. Due to data problems, the ﬁxed effects estimates are
likely to be less reliable than OLS estimates. Nonetheless, in speciﬁcations which include
the quadratic term of unionization, the overall pattern of the estimates is the same as
those in OLS. Thus, the implication remains that among unionized ﬁrms additional
unionization eventually decreases the pay of executives, consistent with the prediction
that implicit regulators negatively impact the pay levels of chief executive ofﬁcers.
Similarly, unionization eventually increases the pay of average employees, thereby
decreasing the pay difference between employees and executives. In closing this gap,

unions seem to act analogously to regulatory agencies.

Conclusions

Other authors (Jensen and Murphy (1990), Joskow, Rose, and Shepard (1993) and
J oskow, Rose, and Wolfram (1996)) have argued that political considerations decrease
the pay of executive ofﬁcers and decrease the sensitivity of pay to ﬁrm performance. This
paper incorporates ﬁrm measures of unionization into the analysis of executive

compensation as a source of political variation between ﬁrms: economic research on

47

unionization frequently models unions as political organizations which pursue the
interests of their average members. Findings show that only the pay levels of executives
in highly unionized ﬁrms earn signiﬁcantly less in terms of annual salary plus bonus than
executives in nonunion and less unionized ﬁrms. .

Other estimates do not indicate that performance sensitivity for CEO salary plus
bonus declines with the level of unionization. In fact, point estimates for performance
sensitivity are the highest for the most unionized ﬁrms, though differences from less
unionized ﬁrms are not statistically signiﬁcant. All categories of unionized ﬁrms have
more performance-sensitive pay than nonunion firms, suggesting that the implicit
regulation hypothesis does not fully capture the preferences of unionized employees.
Findings also suggest that the executives in unionized firms have steeper eamings-tenure
proﬁles, the strongest indication that unions have an impact on pay analogous to that of
regulatory agencies found in J oskow, Rose, and Shepard (1993). Other results suggest
that the pay differential between executives and the average employee is substantially
narrowed by unionization, also consistent with previous ﬁndings in the literature. The
results suggest that unions can impact how much, and the way, executives are paid, but
they do not indicate unions have preferences for making executive pay less contingent on
ﬁrm performance.

Fixed effects estimates, however, seem to indicate that executive pay in unionized
ﬁrms is greater than the pay of executives in nonunion ﬁrms, though those estimates are
statistically imprecise and subject to several sources of measurement error. Despite the

imprecision, the results for some speciﬁcations are roughly consistent with OLS

48

estimates in that the most densely unionized firms have lower executive pay relative to
the executives in the low and moderately unionized ﬁrms.

Research of the effect of unions on ﬁrm performance has shown that unions
impact the performance of ﬁrms and the decisions that managers make (Hirsch, 1991).
This indicates possibility that individual productivity differences between executives may
account for the estimated differences in pay. While this is arguably less a concern than in
regulated industries where regulatory agencies can buffer ﬁrms from the effects of
relatively poor executive performance, it remains an important consideration as an
explanation of the negative impact of unionization on executive pay. The current data set
affords little opportunity to address these concerns, though future analysis should take
greater care in identifying instruments that will control for productivity differences

among executives.

49

Bibliography

Abowd, John M. (1989). “The Effect of Wage Bargains on the Stock Market
Value of the Firm.” The American Economic Review. Vol. 79, No. 4.

Abowd, John M. (1990) “Does Performance-Based Managerial Compensation
Affect Corporate Performance” Industrial and Labor Relations Review. Vol.43, No. 3.

Abowd, John M., George T. Milkovich and John H. Harmon. (1990) “The Effect
of Human Resource Management Decisions on Shareholder Value” Industrial and Labor
Relations Review. Vol.43, No. 3.

Akerlof, George, and Janet L. Yellen. (1990). “The Fair Wage Effort Hypothesis
and Unemployment.” The Quarterly Journal of Economics. May.

Becker, Brian E., and Craig A. Olson. (1987). “Labor Relations and Firm
Performance,” in Morris Kleiner, et. al., eds. Human Resources and the Performance of
the F irm. Madison, Industrial Relations Research Association.

Bell, Linda. (1989). “Union Concessions in the 19803.” Federal Reserve Bank of
New York Quarterly Review. Summer.

DiNardo, John, Kevin Hallock, and Jom-Steffen Pischke. (1997). “Unions and
Managerial Pay.” NBER Working Paper 6318. December.

Ehrenberg, Ronald G. and George T. Milkovich. (1987). “Compensation and Firm
Performance,” in Morris Kleiner, et. al., eds. Human Resources and the Performance of
the Firm. Madison, Industrial Relations Research Association.

Hamermesh, Daniel S. (1975). “Interdependence in the Labor Market.”
Economica. Nov.

Hamermesh, Daniel S., Randall K. Filer, and Albert E. Rees. (1996). The
Economics of Work and Pay, Sixth Edition. Harper Collins College Press.

Hirsch, Barry T. (1991). Labor Unions and the Economic Performance of Firms.
W.E. Upjohn Institute for Employment Research.

Hirsch, Barry T. and John T. Addison. (1986). The Economic Analysis of Unions:
New Approaches and Evidence. Allen and Unwin Press.

Holley, William H, and Kenneth Jennings. (1991). The Labor Relations Process.
The Dryden Press, FL.

Hsiao, Cheng. (1986) Analysis of Panel Data. Cambridge University Press.

50

Jensen, Michael C. and Kevin J. Murphy. (1990). “Performance Pay and Top-
Mangement Incentives.” Journal of Political Economy, Vol. 98.

J oskow, Paul, Nancy Rose and Andrea Shepard. (1993). “Regulatory Constraints
in CEO Compensation.” In Brookings Papers: Microeconomics I993. The Brookings
Institute.

Joskow, Paul, Nancy Rose and Catherine Wolfram. (1996). “Political Constraints
on Executive Compensation: Evidence from the Electric Utility Industry.” Rand Journal
of Economics. Vol. 27, No. 1.

Hall, Brian J. and Jeffrey Liebman. (1997). “Are CEOs Really Paid Like
Bureaucrats?” NBER Working Paper 6213, October.

Lazear, Edward P. (1979). “Why is There Mandatory Retirement?” Journal of
Political Economy. Vol. 87, No. 6.

Lazear, Edward P. (1989). “Pay Equality and Industrial Politics.” Journal of
Political Economy. Vol. 97, No. 3.

Lazear, Edward P. and Sherwin Rosen. (1981). “Rank-Order Tournaments as
Optimum Labor Contracts.” Journal of Political Economy. Vol. 89, No. 5.

Leonard, Jonathan S. (1990). “Executive Pay and Firm Performance.” Industrial
and Labor Relations Review. Vol.43, No. 3.

Levine, David I. (1989). “Cohesiveness, Productivity, and Wage Dispersion.”
Working Paper #14, The Institute of Industrial Relations, Berkeley.

Levine, David I. (1992). “Can Wage Increases Pay for Themselves? Tests With a
Production Function.” The Economic Journal.

Levine, David I. (1993). “Fairness, Markets, and Ability to Pay: Evidence from
Compensation Executives.” American Economic Review. Vol. 83, No. 5.

Main, Brian G.M., Charles A. O’Reilly HI, and James Wade. (1993). “Top
Executive Pay: Tournament or Teamwork.” Journal of Labor Economics. Vol. 11, No.4.

Mincer, Jacob. (1974). Schooling, Experience, and Earnings. National Bureau of
Economic Research.

Murphy, Kevin J. (1986). “Incentives, Learning, and Compensation: A

Theoretical and Empirical Investigation of Managerial Labor Contracts.” Rand Journal of
Economics. Vol. 17, No. 1.

51

Parsons, Donald. (1992). “The Internal Distribution of Union Rents: An Empirical
Test of the Voting Power Model.” Review of Economics and Statistics.

Pindyck, Robert S. and Daniel L. Rubinfeld. (1981). Econometric Models and Economic
Forecasts. Mcgraw Hill.

Rose, Nancy. (1987) “Labor Rent Sharing and Regulation: Evidence from the
Trucking Industry.” Journal of Political Economy. Vol. 95.

Rosen, Sherwin. (1986). “Prizes and Incentives in Elimination Tournaments.”
American Economic Review. Vol. 76, No. 4.

Rosen, Sherwin. (1992). “Contracts and the Market for Executives” in Lars
Wemin and Hans Wijkander, eds., Contract Economics. Oxford. Basil Blackwell.

Solow, Robert M. (1990). “The Labor Market as a Social Institution,” in The
Rover Lectures. Edited by John M. Lettuce. Basil Blackwell.

52

Table 1.1: Means (Standard Deviations) of Compensation Levels and Compensation
Growth for CEOs and Average Employees in Manufacturing Firms from 1980 to 1987

 

CEO salary + bonus Compensation expenditures

 

per employee
Levels %A Levels %A
All ﬁrms 707.6 .070 41.1 .005
(455.5) (.263) (10.5) (.080)
Sample size 1277 893 581 521
Nonunion ﬁrms 494.6 .055 39.2 .007
(303.6) (.206) (7.2) (.076)
Sample size 325 231 63 55 .
Percentage of employees unionized:
0 < unionization .<_ 40% 805.3 .068 38.3 .005
(389.2) (.211) (10.7) (.065)
Sample size 492 360 248 219
40 < unionization _<_ 70% 775.2 .089 41.3 .012
(457.5) (.355) (9.9) (.070)
Sample size 310 201 165 151
Unionization > 70% 709.0 .076 48.3 -.008
(709.2) (.324) (9.5) (.120)
Sample size 150 101 105 96

 

All values are adjusted to 1991 dollars using the Consumer Price Index for Urban Areas.

53

Table 1.2: Means (Standard Deviations) of Firm and CEO Characteristics

 

A. Firm Characteristics
Annﬂl sales Employees Total assets Market return
Levels %A Levels %A Levels %A Levels %A

All ﬁrms 5270.1 .005 32.8 -.004 4308.9 .028 .194 --
(13811) (.160) (71.7) (.152) (10451) (.203) (.384)

Nonunion firms 2506.5 .033 20.8 .023 2447.5 .061 .221 --
(8066) (.189) (57.4) (.175) (8485) (.236) (.459)

Percentage of employees unionized:
0 < U S 40% 6374.6 .009 34.1 .001 5083.0 .030 .200 --
(16231) (.150) (48.0) (.141) (11785) (.227) (.339)

40<US70% 4687.1 -.020 30.3 -.O32 3831.0 .003 .177 --
(5718) (.140) (32.6) (.130) (4860) (.150) (.389)
+..
U>70% 8772.4 -.018 57.7 -.019 6736.1 .008 .157 --
(22810) (.157) (158.8) (.168) (16013) (.129) (.329)

B. Executive Characteristics

 

TotzL Nonunion 0 < U S 40 40 < U S 70 U > 70

Tenure 9.0 11.6 8.0 8.6 7.2
Age at appointment 48.5 44.5 49.8 49.7 49.8
CEO ownership

1 S shares < 3% .132 .220 .093 .115 .104

3 _<_ shares < 9% .114 .198 .076 .076 .139

Shares 2 10% .103 .178 .072 .099 .052
Board Ownership

1 S shares < 3% .294 .156 .369 .346 .243

3 S shares < 9% .270 .370 .243 .256 .179

Shares 2 10% .188 .312 .109 .152 .249

 

Standard deviations of continuous variables are reported in parentheses. Assets and sales
are in millions of 1991 dollars and employees are thousands.

54

Table 1.3: Determinants of Annual CEO Salary + Bonus, OLS Estimates

 

 

1 2 3 4
Scale:
Log(annual sales) .263 .268 .262 .262
(.015) (.015) (.016) (.015)
Performance:
Market return .060 .059 .061 .059
(.050) (.050) (.050) (.049)
Human Capital:
Tenure .012 .012 .013 .012
(.007) (.007) (.010) (.007)
Tenure2x10’3 -.218 -232 -220 -.223
(.179) (.179) (.178) (.177)
Experience .032 .028 .024 .023
(.019) (.019) (.019) (.019)
ExperienceZX10’3 -.439 -.373 -339 -.318
(.271) (.275) (.274) (.274)
CEO common stock ownership:
1 - 3% -.009 -.008 -.020 —.017
(.060) (.060) (.060) (.059)
3 - 9% -.072 -.070 -.074 -.075
(.071) (.071) (.070) (.070)
>10% -.138 -.l37 -.156 -.161
(.080) (.080) (.080) (.079)
Board common stock ownership:
1 - 3% .037 .035 .024 .015
(.052) (.052) (.052) (.052)
3 - 9% .051 .048 .041 .038
(.057) (.057) (.057) (.056)
> 10% .057 .060 .060 .062
(.064) (.064) (.064) (.063)
(continued)

55

Table 1.3: Determinants of Annual CEO Salary + Bonus, OLS Estimates

 

 

(continued)
1 2 3 4
Less than 3 years .007 .008 .004 .009
board tenure before (.039) (.039) (.039) (.039)
becoming CEO
Unionization:
Union -- -.1 1 1 .399 --
(.078) (.238)
Union2 -- -- -.703 --
(311)
1 - 40% -- -- -— .070
(.050)
40 - 70% -- -- -- .028
(.056)
> 70% —- -- -- -.166
(.075)
R2 .695 .697 .702 .707

 

Sample size is 361. Standard errors are reported in parentheses. All speciﬁcations include
industry and year dummy variables and a dummy variable for the ﬁrst year of CEO

tenure.

56

Table 1.4: The Effect of Unionization on Per Employee Pay and the CEO-Average
Employee Pay Differential

 

 

Per Employee Difference Between Coefﬁcients
Compensation Expenditures of CEO and Per Employee Pay
1 2 3 4 1’ 2’ 3’ 4’
Log(annual sales) .050 .042 .048 .046 .213 .226 .214 .216
(.014) (.014) (.013) (.014) (.021) (.021) (.021) (.021)
Unionization:
Union -- .195 -.573 -- -- -.306 .972 --
(.076) (.235) (.109) (.334)
Union2 -— —— .949 -- -- -- -1.652 -—
(-276) (.416)
1 - 40% -- -- -- -.043 -- -- -- .113
(.058) (.077)
41 - 70% -- -- -- -.003 -- —- -- .031
(.063) (.084)
> 70% -- -- -- .118 -- -- -- -.284
(.071) (.103)
R2 .382 .407 .451 .415
N 169 169 169 169

 

Regressions in Columns 1 - 4 are for the log of per employee labor related expenditures.
They also include year and industry dummy variables and dummy variables indicating
the region of the ﬁrm’s headquarters. Table entries in Columns 1’ - 4’ are the difference
in the coefﬁcients from the corresponding OLS estimates in Table 3 and Columns 1 - 4.
Standard errors are reported in parentheses.

57

Table 1.5: Determinants of Annual CEO Salary + Bonus, Fixed Effects Estimates

 

 

1 2 3 4
Scale:
Log(annual sales) .357 .352 .351 .350
(.064) (.067) (.068) (.070)
Performance:
Market return .142 .140 .141 .145
(.059) (.060) (.060) (.060)
Human Capital:
Tenure .012 .012 .013 .014
(.010) (.010) (.010) (.010)
Tenure2x10'3 -.468 -.469 -.474 -.499
(.277) (.228) (.231) (.229)
Experience .022 .021 .020 .021
(.025) (.025) (.025) (.026)
Experience2x10'3 -200 -.195 -.186 -.l98
(.365) (.367) (.371) (.378)
CEO common stock ownership:
1 - 3% .093 .093 .090 .094
(.084) (.084) (.085) (.084)
3 - 9% .013 .012 .009 -.011
(.123) (.124) (.126) (.125)
>10% .129 .128 .125 .105
(.140) (.141) (.142) (.142)
Board common stock ownership:
1- 3% .130 .129 .129 .123
(.074) (.074) (.074) (.075)
3 - 9% .167 .167 .169 .160
(.089) (.090) (.090) (.093)
> 10% .213 .215 .220 .219
(.110) (.110) (.114) (.116)

(continued)

Table 1.5: Determinants of Annual CEO Salary + Bonus, Fixed Effects

 

 

(continued)
1 2 3 4
Less than 3 years .009 .011 .010 .014
board tenure before (.060) (.061) (.061) (.060)
becoming CEO
Unionization:
Union -- .041 .146 --
(.210) (.615)
Union2 -- -- -.138 --
(.761)
1 - 40% -- -- -- .158
(.112)
40 — 70% -- -- -- .109
(.132)
> 70% -- -- -- .105
(.178)
R2 .657 .654 .655 .656

 

Sample size is 361. Standard errors are reported in parentheses. All specifications include
industry and year dummy variables and a dummy variable for the first year of CEO

ICDUI‘C.

59

Table 1.A1: Changes in Unionization Between 1977 and 1987

 

A.) Executive compensation sample
Firms reporting:

 

 

Mean change Decline N 0 change Increase
l.) Unionization -.046 .480 .397 .123
Mean changes if ﬁrms report:
Decline N 0 change Increase
-.140 -- .166
2.) Union Categories: 1987
Nonunion 0 < U S 40 40 < U S 70 U > 70 Total
Nonunion 33 5 2 0 40 (.272)
1977 0 < U S 40% 3 49 0 0 52 (.354)
40 < U S 70% l 11 25 l 38 (.258)
U>70% 0 2 5 10 17 (.116)
Total 37 67 32 11 147
(.252) (.456) (.218) (.075)

Mean changes if ﬁrms change category:
Decline N 0 change Increase

category in category category
Decline Increase

-.231 -.098 .089 .263

(continued)

60

Table 1.A1: Changes in Unionization Between 1977 and 1987
(continued)

 

B.) Per employee compensation sample

Firms reporting:

 

 

Mean change Decline No change Increase
1.) Unionization —.042 .542 .375 .083
Mean changes if ﬁrms report:
Decline No change Increase
-.104 -- .175
2.) Union Categories: 1 87
Nonunion 0 < U S 40% 40 < U S 70% U > 70% Total
Nonunion 6 0 l 0 7 (.145)
1977 0<US40% l 19 0 0 20 (.417)
40 < U S 70% 0 2 10 0 12 (.250)
U> 70% 0 0 l 10 9 (.188)
Total 7 21 12 8 48
(.145) (.438) (.250) (.167)

 

Proportions of firms in that category are reported in parentheses.

61

Table 1.A2: Changes in Other Categorical Variables Between 1980 and 1987

 

Executive compensation sample

A.) Executive common share ownership:

 

 

1987
S<1 1SS<3 3SS<10 8210 Total
S < 1% 77 12 1 2 92 (.626)
1977 1S S < 3% 7 9 0 0 16 (.109)
3 S S < 10% 6 3 10 2 21 (.143)
S 2 10% 3 0 1 14 17 (.122)
Total 93 24 12 18 147
(.633) (.163) (.082) (.122)
B.) Board of directors’ common share ownership:
1987
S<1 1SS<3 3SS<10 8210 Total
S< 1% 25 8 0 1 34 (.231)
1977 1S S < 3% 11 17 6 l 35 (.238)
3 S S < 10% 4 18 19 2 43 (.293)
S 2 10% 2 4 10 19 35 (.238)
Total 42 47 35 23 147
(.286) (.320) (.238) (.157)

 

Proportions of firms in'that category are reported in parentheses.

62

CHAPTER 2

WIDENING PAY GAPS AND FIRM PRODUCTIVITY: DOES FAIRNESS MATTER?

Introduction

The pay difference between CEOs and average workers in the United States is
much larger than similar measures from other industrialized countries (Abowd and
Bognanno 1995). The wide differential does not escape public commentary, suggesting
that the relatively lucrative compensation arrangements enjoyed by CEOs in publicly
traded ﬁrms are unpopular with the media, many shareholders, and the general public.28
While not a particularly disinterested party when it comes to matters of pay, the AFL-
CIO has devoted a web page to “curtailing” what it deems “excessive” CEO
compensation packages, reﬂecting the skepticism that many people have for CEO pay
levels.29

This paper explores whether relatively high levels of CEO pay are related to
employee productivity. Evidence emerging from several recent surveys of compensation
directors and other human resource practitioners supports assumptions used in motivating
various efficiency wages models of the macroeconomy. Of particular interest in this
study is that variant of the efﬁciency wage model that argues wage rigidity and persistent

inter—industry wage differentials are caused in part by employee concerns with fairness in

 

28 Jensen and Murphy (1990) argue that public scrutiny likely suppresses CEO pay levels
and changes the way they are paid.

29 See the AFL-CIO’s “Executive Paywatch” web page at www.paywatch.org. See White
(1981) for a discussion of union motives in emphasizing worker-management pay
differentials.

63

relative earnings. Both Blinder and Choi (1990) and Campbell and Kamlani (1997) find
evidence that compensation specialists believe employees will respond with effort
reductions to unfair pay developments.

The specific focus of this paper is to determine if there is empirical evidence that
supports the fundamental assumption motivating the fair wage-effort hypothesis:30
namely, that workers reduce effort when wage differentials increase. At least two other
papers attempting to identify efﬁciency wage effects have used a framework similar to
the one used here.“ Both papers include a measure of relative wages in an augmented
production function and ﬁnd that firms paying wages that are high relative to their
industry also have higher output. This paper extends such analysis by including a ﬁrm-
specific measure of relative pay as an explanatory variable in a productivity equation.

The relative pay measure used here captures the difference in pay between chief
executives and average employees in the same ﬁrm. The intent is to determine if fairness
concerns, at least over this particular dimension of a firrn’s compensation system,
manifest themselves in per worker measures of sales or value added. As in the other
studies, the relationship between pay and performance estimated here is likely to be
biased due to the simultaneous determination of productivity and performance and
because adequate measures of worker quality have not been included. Attempts are made
below to address these issues.

In all, the results of this exercise do not provide solid support for the assertion that

effort reductions occur when pay differences increase. Neither do they point

 

30 See Akerlof and Yellen (1990). See Levine (1991) for a model incorporating similar
assumptions about employee behavior, and see Hamermesh (1975) for an early treatment
of these issues in a production function framework. Summaries are presented in table 1.
3' See Levine (1992) and Wadhwani and Wall (1991).

64

unequivocally to models of workplace tournaments, which suggest that pay differentials
will be positively associated with at least some measures of firm performance. Relatively
large contemporaneous differences in pay are associated with higher productivity, though
it is unlikely that the estimated relationship represents causality. One year lagged
measures are also positively associated with higher productivity, with the effect being
especially large. Using a two year lag, the effect of widening pay dispersion is negative
and frequently statistically significant. Estimates on lagged measures are difficult to
interpret theoretically, however. Instrumental variables results also fail to point
unambiguously toward a potential negative impact from increasing pay inequality.

It is worth noting that this paper analyzes the effect of increasing pay differences
on per employee output, as opposed to ﬁrm profitability. By focusing on productivity,
important considerations regarding ﬁrm ﬁnancial performance may be missed. For
example, a wage reduction imposed on employees may result in decreased productivity,
but the increase in proﬁts due to reduced labor costs may more than offset the reduction
in productivity. Therefore, one cannot conclude, a priori, that proﬁt maximizing ﬁrms
will not make negative adjustments in wages or relative wages even if such adjustments
lead to decreased output. Complementary analysis should therefore examine the effects
that wage inequality has on accounting and market rates of return, among other ﬁrm
performance measures.

The remainder of the paper proceeds as follows: section 11 reviews the relevant
literature linking pay dispersion and productivity. The review is wide ranging, covering
theoretical treatments of compensation practices both by economists and by researchers

outside of the profession. Section III presents the data and econometric specification

65

used and discusses some of the problems that arise in estimating the pay difference-
productivity relationship. Section IV presents the statistical results and analysis and,

Section V concludes the paper.

Pay, relative pay, and productivity: theory and evidence
Compensation systems and equity theory

A significant body of literature considers the question of whether pay dispersion
is associated with higher or lower productivity. Notably, this issue does not arise in the
textbook model of the ﬁrm, since standard neoclassical economic theory assumes that
proﬁt-maximizing ﬁrms pay individuals according to their observable marginal
contribution to production. Therefore, wage variation only reﬂects interpersonal
differences in productivity.32

However, in ﬁelds such as human resource management and industrial
psychology, pay is not regarded strictly in terms of market mechanisms. Compensation
systems are viewed as integral components of firms’ administrative policies, since they
are used, for example, to prompt work effort or to direct the ﬂow of employees through
internal labor markets. These elements of pay, which are somewhat familiar to
economists, also imply social relationships, since pay confers status among employees
and control over the pay of others entails issues of authority and conﬂict (White 1981).

In such a framework, social comparisons are likely to take place. According to

equity theory in industrial psychology,33 individuals derive utility based in part on the

 

32 See Filer, Hamermesh, and Rees (1996) for a discussion of the relationship between
ability and income distribution.

33 See Kenungo and Mendoca (1992) for a detailed discussion of equity theory. Other
theories of work motivation and job satisfaction abound, such as reinforcement theory,

66

relationship of their pay to the pay of others. Equity is presumably linked to the
workplace since the utility or disutility people feel regarding their relative pay affects
their morale and induces behavioral responses. Equity theory predicts that as wage
differentials increase, relatively underpaid employees will feel less committed to the
success of the ﬁrm, resulting in decreased work effort, lower output quality, increased
quit rates, and potential acts of sabotage. As Kahneman, et al put it, “the rules of fairness
deﬁne the terms of an enforceable implicit contract: ﬁrms that behave unfairly are
punished in the long run (1986, p.728, italics added)”

From a practical standpoint, the fact that the workplace is subject to sociological
and psychological phenomena does not help to explain how interpersonal comparisons
take place. Most obviously, the question arises of how comparison groups are selected.
Equity in the workplace has horizontal and vertical dimensions and these dimensions
have internal and external facets.34 Encinosa, Gaynor, and Rebitzer (1997), drawing on
experimental results in social psychology, emphasize “saliency” in the determination of
comparison groups. Saliency refers to the increasing effect of interpersonal comparisons
“with an individual’s similarity, proximity, and exposure to the reference group (p.6).”

Presumably, saliency will be strongest among workers doing similar tasks within

the same ﬁrm, though “similarity” is not the sole determinant of comparison groups.

 

expectancy theory and facet satisfaction theory. See Kenungo and Mendoca (1992) and
Harpaz (1982) for discussions.

34 Horizontal and internal dimensions of pay equity are generally emphasized by human
resource specialists (Milkovich and Neuman 1996, Gomez-Meija and Balkin 1992, and
Levine 1993). The internal aspects of horizontal equity refer to the dispersion of pay
among employees doing the same or similar work within the ﬁrm, while external
concerns refer to the relative levels of pay among workers in relevant labor markets
outside of the ﬁrm. Equity can conceivably have intertemporal facets, too. That is, an
individual may have concerns about her current wage relative to her wages in previous
years.

67

Given the nature of the variable used in this analysis, it is of interest to know whether
vertical comparisons matter in determining morale. Martin (1982) conﬁrms that workers
make comparisons to workers in superior positions. In a survey of blue-collar workers,
almost all made comparisons to those earning more than themselves, and the relative
earnings of supervisors were more important than other blue-collar workers’ earnings in
determining job satisfaction. In Campbell and Kamlani’s (1997) survey, 86 percent of
respondents thought workers would withdraw effort either a moderate amount (35
percent) or a great deal (51 percent) if the firm increased the pay of highly paid workers
without changing the pay of low wage workers. On the other hand, White’s (1981)
review of evidence suggests that at most only 20 percent of work stoppages or production
slow downs, measured by such incidents as strikes or employees’ systematic use of work
to rule strategies and overtime refusals, were attributable to conﬂicts over relative
earnings disparities.

Several authors have incorporated equity, per se, into economic models of
production and the macroeconomy. Notably, Akerlof and Yellen (1990), Levine (1989),
and Hamermesh (1975) have all motivated models of the macroeconomy with the
assumption that workers prefer greater wage equality and will withhold effort if
inequality increases. These models are frequently cited as the fairness variants of the
efﬁciency wage hypothesis, and have been used to explain a host of macroeconomic and
microeconomic phenomena that are difﬁcult to explain using conventional economic
theory. Examples of such real-world anomalies include persistent interindustry wage
differentials (Krueger and Summers 1986), the relatively small variance in pay within job

grades vis a vis the variation in individual productivity (Medoff and Abraham 1981),

68

involuntary unemployment (Akerlof and Yellen 1990, Levine 1989, Hamerrnesh 1975),
and tempered adjustments to increases in the minimum wage (Hamermesh 1975).

In Levine’s model, wage compression between groups of workers produces
cohesion.35 The optimal relative wage levels set the elasticity of cohesiveness with
respect to wage compression equal to the elasticity of output with respect to labor. That
is, the gains from increased cohesiveness are balanced against the marginal productivity
of an additional worker employed at the market wage. In equilibrium, some ﬁrms will
pay low-productivity workers wages above their marginal product. While the fair wage-
effort models offer a possible explanation of why ﬁrm or industry-level pay differences
arise and persist over long periods of time, they are motivated by assumptions about

potential economic consequences that have gone largely untested.

Pay variation in economics: the role of incentives

In economics, incentive effects of pay variation have been incorporated into
formal models. This is of particular importance here since models of workplace
tournaments (Rosen 1993 and Lazear and Rosen 1981), which emphasize the beneﬁcial
role of pay dispersion, may be especially applicable to executive labor markets. In
tournament models, pay hierarchy elicits effort from young employees who compete for
the pay raises and increased control of resources that go along with promotions. Winning
promotion into the next bracket of competition gives one the right to participate in

continued competition for the top position in the firm. Furthermore, the difference in

 

35 In Hamerrnesh (1975) market wage changes are treated as exogenous, while in Akerlof
and Yellen's (1990) model, wages are treated as endogenous and are adjusted as a
function of reference group wages and the competitive wage.

69

pay, that is the high payoff for promotion, assures that the most talented individuals
continue to compete for promotions.

This construction of organizational compensation helps to explain the distribution
of earnings among executives, where the pay difference between the chief executive and
the second in command is unlikely to be explained by their relative productivities. The
high pay associated with the chief executive position is justiﬁed on the grounds that the
compensation serves as an incentive for the most able executives to maintain effort as the
potential number of promotions dwindles. Skewness in the pay scale assures that the
most talented individuals do not opt out of further competition when they near the top of
the ﬁrm hierarchy.

Within the context of tournament models, however, there are instances when
increasing pay differentials could lower productivity (Lazear 1988). This occurs because
an agent can conceivably affect her relative performance in two ways: by increasing her
own productive work effort or by sabotaging the output of her competitors. High
potential payoffs from winning contests may induce contestants to engage in acts of
sabotage. With a ﬁxed probability of sabotage being detected, the marginal effect of
increased productive effort on winning a promotion eventually is lower than the marginal
impact from decreasing the output of opponents, accomplished by destructive effort.
Thus, if ﬁrms cannot detect the type of effort that employees engage in, raising the payoff
to winning a competition may have a net negative impact on productivity. In units of
production that rely on teamwork or that otherwise emphasize cooperation among

employees, excessive pay differentials are likely to lead to unproductive behavior.

70

One of the ostensive strengths of tournament models in explaining the function of
pay inequality is that they avoid the frequently arbitrary behavioral assumptions
regarding interdependence in utility functions that motivate models invoking fairness or
morale concerns. Lazear’s model, for example provides a non-equity based rationale for
pay compression. Whether Lazear’s model generalizes to situations where agents within
a ﬁrm are non-competing but potentially concerned about relative earnings is unclear. It
does suggest that care should be taken in identifying characteristics which are likely to
make individual incentives fail, and in that sense the model is complementary to those
emphasizing the social or psychological aspects of the employment relationship.

However, Lazear’s model does underscore troubling theoretical aspects of the fair
wage hypothesis. If ﬁrms can adjust relative wages to increase productivity, the question
becomes, why would ﬁrms choose suboptimal relative wage levels? The presumption in
economics is that profit-maximizing ﬁrms will employ the right amount of morale in
production. That is, firms are unlikely to choose a wage structure that will leave them
worse off than another possible set of wages. The assumption of proﬁt maximization
suggests that concerns about fair wages are theoretically “vacuous”.36

There may be circumstances when ﬁrms will choose the “wrong” combination of
wages. Managers may be able to make only near rational choices regarding the relative
wage structure. For example, managers may not know the magnitude of the response
their employees will have to changes in relative pay, or managers may overestimate their
ability to deter shirking. With respect to the pay measure used here, boards of directors

may choose the wrong compensation level for CEOs because they have been captured or

 

36 In his commentary on Katz’ (1986) discussion of efﬁciency wages, Weiss argues that
“the ‘sociological’ arguments (for efficiency wages)...appear vacuous (p.286).”

71

because they are not privy to all relevant information about executive performance. For
example, Joskow, Rose and Wolfram cite the work of Crystal which argues that
“compensation typically is controlled de facto by the CEO rather than by the board of
directors, and that many CEOs are overcompensated as a result (p. 166, 1996)” In such
instances, the relative wage structure may be near the optimal level, but measurable
effects from relative pay concerns may arise.

Implied in this analysis is the assumption that the maximization objectives of
chief executives do not coincide with those of owners (or those of the average employee).
Principal agency theory, among several other earlier theoretical treatments of managerial
behavior, suggests that executives and managers are likely to have different objectives
than owners. For example, chief executives may *** insert tirole stuff here *** Rosen
(1993) and Tirole (1990) present surveys of possible managerial objectives. Supposing
there is some optimal, profit-maximizing chief executive-average employee pay ratio,
one could not assume that it is the ratio of pay that the chief executive, or a “captured”

compensation committee, would choose.

Related Evidence
Empirical evidence on the role that pay inequality plays in the determination of
performance is sparse, though several compelling examples exist. In addition to direct

tests for the effects of relative wages on productivity and other performance measures,

72

there is a growing body of indirect evidence, though little of it derives from attempts to
sort out the implications of competing equity and incentive concerns.37

Evidence on equity concerns and the effects of pay variation can be divided,
roughly speaking, into two categories. The ﬁrst is characterized by surveys and
experiments which attempt to gauge attitudes regarding fairness and how these attitudes
are likely to impact economic behavior. The second category includes more data
intensive statistical analysis. This category includes explorations for evidence of
efﬁciency wages; studies of the effect of worker morale on output; and studies that
include wage variation as an explanatory variable in performance regressions. Table
2.2.1 provides a brief summary of these papers.

Kahneman, et a1 (1985), survey households about their attitudes regarding market
transactions in order to determine the “rules” that constitute fairness. They conclude that
attitudes about fairness pose binding constraints on the proﬁt-maximizing behavior of
agents in the marketplace. Gorman and Kehr (1992) ﬁnd that chief executives share, in
varying degrees, the same attitudes about fairness exhibited in Kahneman, et al’s
household survey.

As indicated above, Blinder and Choi (1990) and Campbell and Kamlani (1997)
surveyed ﬁrms to determine which of several efﬁciency wage assumptions, if any, are
consistent with practitioners views of the labor market and employment relationship.
Both ﬁnd relatively strong evidence that compensation directors are reluctant to make

wage changes that violate standards of fairness since they fear workers are likely to

 

37 On this latter point, this paper is no exception. See Encinosa, Gaynor, and Rebitzer
(1997) for a remarkable exception. In that paper they are concerned primarily with
demonstrating theoretically and empirically the role that group norms have in
determining optimal compensation practices.

73

 

respond by withdrawing effort. Concerns about adverse selection also were cited as
important reasons for not cutting wages during recessions.

Evidence presented by Levine (1993) from a survey of compensation executives
suggests that compensation policies generally are adjusted in response to changes in
market wages only gradually, and that compensation executives take care in maintaining
existing relative wage relationships when adjusting pay structures, though he ﬁnds less
concern with maintaining pay relativities across either narrowly or broadly deﬁned
occupational groups. In general, survey evidence reveals widespread agreement about
what constitutes fairness. Furthermore, these attitudes are perceived as binding
constraints on behavior in the employment relationship.

In an exceptional series of papers, Norsworthy and Zabala (1982, 1983, 1985,
1990) attempt to link an index of worker morale they create from information on wildcat
strikes, filed grievances, and other presumed measures of employee dissatisfaction to
productivity and production costs in the automotive manufacturing industry. Their
ﬁndings indicate that the slowdown of productivity growth during the 19603 and through
the 19708 is partly attributable to deteriorating worker morale, though Straka (1993)
questions the robustness of their ﬁndings.

In a model that is closely related to this study, Levine (1992) tests the efficiency
wage hypothesis using a measure of relative wages acquired in a survey of manufacturing

plants on the right hand side of a productivity regression. He shows that plants paying

74

relatively high wages38 successfully recoup the payments in increased per worker output,
though he is unable to distinguish between efficiency wage and rent-sharing explanations.

Wadhwani and Wall (1991) present similar results for a large panel of UK
manufacturing ﬁrms. They measure average ﬁrm-level pay analogously to the method
employed here, which divides total labor expenditures by the number of employees. This
variable is then divided by the industry average and included in production function
equation. They also ﬁnd that relatively high pay is associated strongly with higher ﬁrm
output, and that the estimate remains virtually unchanged when using instrumental
variables to correct for potential bias associated with the simultaneous determination of
wages and productivity.

In tests for the incentive effects of pay dispersion, Leonard (1990) examines
whether the steepness of the executive pay profile impacts firm ﬁnancial performance.
Using the same data, Main, O’Reilly, and Wade (1993) identify ﬁrms that presumably are
more team-oriented, and therefore more prone to the negative effects of sabotage, to
examine the effect of increasing executive wage dispersion. Leonard ﬁnds no evidence
that the steepness of the executive pay scale increases return on equity. Main, et al,
however, do ﬁnd a signiﬁcantly positive correlation between the coefﬁcient of variation
for executive compensation and return on assets, though they do not ﬁnd evidence of
sabotage occurring due to excessive pay dispersion.

Using a particularly rich data set, Pfeffer and Langton (1993) test the effect that

pay variation has on several performance measures. Speciﬁcally, using a large cross-

 

38 The relative wage measure was the respondent’s estimate of the firrn’s average wages
relative to its three closest competitors, with respondents instructed to hold the
characteristics of employees in comparison firms constant.

75

sectional survey of faculty in US college and university academic departments they find
that research productivity, job satisfaction, and participation in collaborative research are
all negatively affected by the department-level coefﬁcient of variation for compensation.
They also ﬁnd that faculty research output is higher the more closely it is tied to pay.
The results below present additional evidence on the effects of pay dispersion on
output. The analysis complements the above papers by attempting to identify whether the
relatively high levels of pay received by chief executives prompts an equity response
from employees, or if it is associated with higher productivity. The data used in this
exercise are far from ideal, though they compare favorably to data used in previous
analysis. The tests come from a panel of ﬁrm-level observations, so that effect of pay
dispersion is identiﬁed while controlling for unobserved, time-invariant firm
heterogeneity. In addition, care has been taken to eliminate bias associated with the
simultaneous determination of pay and productivity and to address bias associated with

the failure to control for employee characteristics.

Data and estimating framework
The relationship between executive-average employee pay inequality and
productivity is estimated using the following augmented translog production function,39
ln(Q,-f/L,-f,) = (10 + on ln(K,f,/L,-f,) + a2 ln(L,-f,) 4-

(X3 ln(K,-f,/L,-f,)xln(K,-f,/Liﬂ) + (X4 ln(L,-f,)xln(L,-f,) + a5 ln(L,-f,)xln(K,-f,/L;ﬁ) +

 

39 Alternatively, Cobb-Douglas or constant elasticity functional forms could be used. See
Appendix A for a discussion of results from these speciﬁcations. The source for the data
used in this analysis is Standard and Poor’s Compustat unless otherwise indicated. All
dollar measures used in this analysis are adjusted to 1991 values using the consumer price
index for urban areas.

76

016 ln(CEO,f, / LABiﬂ) + (17 ln(COV,-f,) + as ln(IND,-,) + 09 X], + 8,7,, (1)

where the subscripts, i, f, and t, represent industry, ﬁrm, and year, respectively. Q
represents output, consisting of total ﬁrm sales or value added. Value added is deﬁned as
total sales minus the cost of materials. Value added is recovered for this analysis by
subtracting the cost of goods sold (Compustat’s COGS) from total sales and adding back
in labor related expenditures (Compustat’s XLR).40 Both dependent variables are then
divided by the total number of employees, giving per worker measures of output

On the right hand side of the equation, capital is represented by K, and is
measured using the dollar value of property, plant, and equipment net of depreciation
(Compustat’s PPEN).41 Labor (Compustat’s EMP) input into production is represented
by L. X represents a vector of explanatory variables which presumably shift per worker
output, including time dummy variables and 2-di git industry dummy variables
representing the primary industry classification for the ﬁrm in its last reporting year.

The variables of special interest in this analysis measure inequality of pay. CEO
represents annual CEO salary plus bonus. Executive compensation data were collected
independently by the author from ﬁrm proxy statements ﬁled with the SEC and archived
by the Q-sup data company. Annual data on CEO compensation were collected for 12
years (1980-91) on 348 ﬁrms, with a mean of 9.7 years per ﬁrm available. In this
analysis CEO compensation is measured as the sum of annual salary plus bonus. It

therefore understates total annual executive pay by a substantial amount. A better

 

40 Technically, rental payments should also be added to profits. Unfortunately, that
variable was not among those collected for this analysis.

4’ Gross property plant and equipment and total assets were also used to measure capital
stock. Results using them are similar to the results presented here.

77

measure would also include the valuation of granted stock options and related incentive
pay, but reporting requirements for these elements of CEO compensation made the
collection of detailed information prohibitively difficult.42

LAB represents compensation expenditures per employee. The average employee
pay variable is constructed using labor related expenditures divided by the number of
employees. The measure is only available for a small proportion of ﬁrms due to the
relatively relaxed reporting requirements for this particular variable. This measure was
available from Compustat for 157 manufacturing ﬁrms, with a mean of 8 observations
per ﬁrm.

Matching the two sources of compensation information results in 1125 matches
for 152 ﬁrms. Sample sizes for the analysis below are smaller yet due to missing values
and the requirement that CEO pay be measured only in years when the CEO worked the
entire year. In some analysis below, the data are differenced so that sample sizes shrink
even further due to the elimination of a cross section of observations and because of
missing values in adjoining years. In addition, differencing reveals some changes in firm
size or output levels that represent rather large magnitudes. Such outliers were
eliminated by excluding the top and bottom one percent of observations on the following

variables: both dependent variables, changes in capital per employee, and changes in

 

42 Frequently, granted and exercised options during this period were reported over three
to ﬁve year spans, making it difﬁcult to identify the number of granted options in any
given years, especially given the large number of missing observations due to missing
microﬁche. Additionally, executives could be granted stocks from more than one
program, which presumably had different rules governing their exercise. Finally, granted
and exercised options frequently were reported in text, as opposed to being tabulated,
making it very time consuming to collect the information about granted and exercised
options. See Hall and Leibman (1997) for a discussion of the issues associated with
measurement of executive compensation and as an example of successful incorporation
of outstanding stock awards.

78

employment. The observations were eliminated from cross section and ﬁxed effects
estimates, as well, with little impact on estimates.43

Since high executive pay may be important in prompting work effort among
executives, the coefficient of variation (the standard deviation divided by the mean) for
the pay of the top ﬁve most highly paid executives is also included, represented in
equation 1 by COV. This information also was collected from ﬁrm proxy statements. Its
inclusion in the present analysis follows from models of firm hierarchies, which argue
that the productivity of the workers in the top of the hierarchy echo through subordinate
levels of the ﬁrm (Rosen 1992).44 If executive pay differentials effectively increase
executive effort levels, then one should expect to see higher average productivity in firms
where the pay differentials are higher.45 As Rosen states, hierarchical models imply that
“a little extra talent at the top can have enormous effects on total output (p.185, 1992).”

The variable is based only on five observations per ﬁrm per year, so it is likely to
be an imperfect proxy for the entire executive pay scale. The problem is exacerbated

because it is difﬁcult to determine if executives other than the chief executive worked the

 

43 The minimum (maximum) values for these before eliminating the tails were -.64 (1.17)
for value added per employee, -.67 (1.18) for sales per employee, -1.34 (.75) for
employment, and -.84 (1.14) for capital per employee. After eliminating the tails the
values were -.30 (.25), -.26 (.34), -.46 (.24), and -.23 (.36), respectively. The maximum
value for value added represents a 220 percent increase in reported output per employee
over one year, which is unlikely to be attributable to changes in employee morale.

44 It should be noted that Rosen presents these models to explain the relationship between
executive pay and ﬁrm size.

45 Main, O’Reilly, and Wade analyze the pay of executives in the top 5 levels of the
executive hierarchy. They include the coefﬁcient of variation for executives over these
levels in an analysis of ﬁrm ﬁnancial performance. There in no inexpensive way of
knowing how much of the senior executive hierarchy is represented by the ﬁve
executives in this sample, though it is probably safe to assume that they constitute only
the top two or three levels. Executives in sub-CEO positions in this sample generally had
job titles of president, chief operating officer, executive vice president, group vice
president, senior vice president, vice president, or vice chairman.

79

entire year. The variable should pick up relevant differences between the CEO and those
directly competing for the top position, and it should therefore provide useful information
about the incentive component of high chief executive pay. Pay variation over the
executive pay scale can also have equity effects, though Main, et al (1993) ﬁnd no
evidence of sabotage related to excessive variation in executive pay.

IND represents measures of pay variation from 2-digit industries. The variables
are from the Current Population Survey, and are derived by aggregating the wage and
compensation information from each month for the years between and including 1980
and 1991.46 They are included in this analysis as proxy variables for possible horizontal
developments in pay taking place in the firm. Since Groshen (1991) has shown that there
is substantial between-ﬁrm variation in pay levels within industries, the variables are
unlikely to serve as good proxies. In addition, Davis and Haltiwanger (1993) present
evidence that the variables will measure developments in ﬁrm-level pay with substantial
error. The variables are retained in some of the analysis below since their inclusion does
not materially affect the estimates on other variables and since they may be of general
interest independent of their usefulness as proxies for firm pay developments.

Included among IND are the following variables: the coefﬁcient of variation and
the median wage in industry i in year t measured separately for white collar workers
(including managerial, professional-technical, and sales occupations), blue collar workers
(including craft, Operative, and laborer occupations), and pink collar workers (including

clerical and service occupations).

 

46 The measure was constructed using all reported measures of pay available in each year
for all workers between and including the ages of 25 and 55. No adjustment were made
for worker characteristics.

80

Econometric problems

Several problems with the proposed estimating framework are apparent. Most
importantly, estimates on the relative pay variable from equation 1 will likely be biased
because of the simultaneous determination of pay and output and because of the omission
of important explanatory variables that also are correlated with the pay measures. It turns
out that neither of these problems can be resolved conclusively using the present data set,
though the problems will be addressed to the extent possible. The strategy used below is
to present a broad range of possible speciﬁcations and to discuss their relative merits
based on theoretical and econometric considerations.

The inclusion of firm-level measures of pay on the right hand side of equation 1 is
problematic since pay and productivity are simultaneously determined. In the present
case this source of bias may be especially problematic. Levine (1992), in a similar
exercise, discounts simultaneity as a source of bias in his analysis using between-ﬁrm
measures of relative wages as an explanatory variable in a production function. He
points out that ﬁrms may face short run positively sloped supply curves, implying that an
increase in product demand may prompt the firm to increase wages in order to attract
additional workers. He argues, however, that the persistence in relative pay levels across
plants over long periods of time makes it unlikely that transitory shocks account for his
ﬁnding. Furthermore, he demonstrates no substantial negative autocorrelation in the
growth of his relative wage variable, which he argues would indicate adjustments in
wages back to market levels.

Wadhwani and Wall also acknowledge the problem associated with the

simultaneous determination of pay and the output measures used here. Rent sharing

81

arrangements in high productivity firms, they point out, may increase the pay of
employees in those ﬁrms. Using lagged ﬁrm ﬁnancial variables as instruments, they ﬁnd
that their point estimate is virtually unchanged and that the estimate remains strongly
significant when instrumenting.

Given the dimension over which the relative pay variable is measured in this
analysis, it is unlikely that the issue of simultaneity can be dismissed. This is especially
the case for CEOs, since their pay is usually contractually tied to the performance of the
ﬁrm. Both dependent variables used in this analysis, per employee measures of value
added and sales, are constructed using total ﬁrm sales.

The potential positive bias associated with both the numerator and denominator of
the pay ratio makes the sign of the bias in the estimated effect difﬁcult to predict. By
increasing the denominator, positive shocks to productivity will result in a negative bias.
However, there is also a positive bias associated with the numerator, since positive
shocks to productivity will also increase executive pay. The “net” bias associated with
the variable is likely to depend on which measure of pay is most responsive to changes in
productivity.

The pay of executives is likely to be more responsive to productivity shocks than
the pay of average employees. This could be for a variety of reasons, such as the nature
of labor contracts for average workers. Contracts negotiated by unions generally
determine wage levels over relatively long periods. Furthermore, implicit contracts
between workers and firms may entail wage stability since ﬁrms may agree not to cut
wages during bad times in exchange for tempered wage demands during periods of

relative prosperity, insuring employees against negative income ﬂuctuations.

82

Empirically, Akerlof, Dickens and Perry (1996) demonstrate with several data
sets that the nominal pay of most employees is downwardly sticky. Blanchﬂower,
Oswald, and Sanfey (1996) show that average pay growth in manufacturing industries
increases with industry profitability, though the increases in pay come after a three year
lag. Thus for average employees there is some upward responsiveness in wages
associated with industry rents, though the response is delayed.

For executives, Rosen (1992) reviews both the studies establishing the theoretical
link between ﬁrm sales and executive compensation and the studies examining the
empirical relationship between them.47 Regarding the latter, he ﬁnds a relationship that is
statistically very strong whether the two variables are measured in levels or in changes,
with the strength and size of the estimate consistent across a wide range of data sets
representing different time periods. Furthermore, the relation of executive salary plus
bonus to other performance measures is also strongly established, with executive pay
being especially responsive to various accounting measures of profitability (Rosen 1992).
Thus, empirical evidence suggests that the pay of executives is more responsive to
exogenous productivity shocks, implying that the bias in the estimate on the pay ratio in
equation 1 should be positive.

The bias associated with endogeneity will be addressed in two ways. The ﬁrst
approach attempts to attenuate the endogeneity problem by including lagged values of the
measure instead of a contemporaneous measure. This solution is satisfying provided the
error term in equation 1 is not serially correlated, and to the extent that any of the

theoretical effects are persistent over time (as discussed below).

 

47 Sales can represent both a measure of scale, since it is highly correlated with measures
of capital stock and total assets, or a measure of performance. Rosen notes that recent

83

Second, the estimates from an instrumental variables approach are presented.
When using instrumental variables, the variable selected to identify the structural effect
must be correlated with the relative pay variable but unrelated to the error term in the
productivity equation. Even if the variable used in this exercise meets these strict
requirements, it does not provide unambiguous insight into the relationship between
productivity and pay inequality.

A second problem with the model presented in equation 1 which should be
addressed is related to unobserved variables. Namely, 00, may be biased in cross-
sectional estimation, even though the pay ratio is lagged, due to unobserved ﬁrm
characteristics, such as the quality of the work force or management, which are correlated
both with the pay ratio and the dependent variables. To get unbiased estimates,
measurements for these firm-level characteristics should be included in the analysis.

Due to the panel structure of the data set, they can be controlled for to the extent
that they are time invariant. Including firm-level dummy variables will control for time-
invariant ﬁrm heterogeneity. Unfortunately, the estimated 016 may still be biased for two
reasons. The first is due to the fact that important omitted variables, such as employee
quality, are unlikely to remain ﬁxed over time. For example, a ﬁrm cutting its workers’
wages, can expect its most productive employees to leave, since they are likely to have
the highest alternative wage. Thus, one would expect the wage cut and consequent
increase in the gap between executive and average employees to result in lower
productivity. By not controlling for the change in worker quality, one cannot identify an

independent equity or incentive effect.

 

analyses of executive compensation emphasize the “scale” interpretation of firm sales.

84

Results from attempts to address this problem are presented in the following
analysis. Levine (1992) also discusses the problem. Though the survey question
producing his relative pay measure instructs respondents to control for the relative quality
of ﬁrm employees, Levine acknowledges that respondents may not have accurately
estimated the relative quality of their employees. However, he argues that even if he
were able to control for observable differences in the human capital stock of employees,
the resulting estimate would not be substantially different than the one he presents.48
Wadhwani and Wall argue that differencing the variables controls sufﬁciently for
employee quality.

The second reason for bias in Q6 derives from the fact that ﬁxed-effects estimation
reintroduces the bias caused by simultaneity. This point is illustrated clearly by recalling
that the inclusion of ﬁrm dummies in equation 1 is equivalent to removing the ﬁrm mean
from each variable. Endogeneity is reintroduced in this process: payifu - meanpay is
correlated with em - means, since cov(meanpay, means) at 0. Note that the endogeneity
exists using this procedure regardless of the number of years the explanatory variable is
lagged. Lagging the pay variable one year and ﬁrst differencing equation 2 similarly

results in correlation between the differenced pay variable and the error term.”

 

48 To support this contention he cites an exercise he runs using two ﬁrm-level data sets
with detailed information on wages, ﬁrm characteristics, and human capital variables
(both data sets have observations of multiple individuals within ﬁrms. One is cross
sectional, the other is a panel). For each data set he regresses wages once on firm level
variables and once on the same firm variables and additional human capital variables. He
argues that the high correlation (greater than .9 for both data sets) between equations in
the estimates on firm-level variables imply that the high-wage ﬁrms do not pay high
wages primarily because they have higher quality workers (conditional on occupational
mix).

49 When ﬁrst differencing equation 1, the error term becomes 88 =8, - 8,-1, while the pay
variables (if lagged one year) become, Spay =pay,-1 - pay,-2, so that Spay is correlated

85

To control for the bias associated with simultaneity, first differencing equation 1
with second-order lagged measures of the endogenous variable:
[]n(Qif/Lift) ' ln(Qift-l/Lift-1)] = 0‘0 + a1[]n(Kifr/Lifr) - ln(Kift-I/ Lift-1)]

+ a2[ln(L,f,) - ln(L,-f,_ 1)] + a3[ln(K,-f,/ Lift)xln(K,f,/ Lm) - ln(K,-f,-1 /L,-f,-1)><ln(K.-f,-1 /L,-f,-1)]
+a4[ln(Liﬁ)XIn(Liﬁ) - ln(L,-f,- ,)><ln(L,-f,- 1)] + a5[ln(L,-ﬂ)xln(K,-ﬂ/ L,-f,) - ln(L,ﬁ-1)xln(K,~ﬁ-1/ Lm-1)]
+ 06 51n(CEO.-f/LABif)* + a7[ln(C0V,f,) - ln(C0V,f,-1)] + 0L3[X,-f, - Xm- 1] + [8,], - Ema], (2)
where 81n(CEO,~f/ [ABif)* is the difference between ln(CE0,-f,-2 / LAB,-ﬁ-2) and ln(CEO,-f. 3 /
LABm-3).50 Estimates of 016 from equation 2 will provide unbiased estimates of the effect
of increasing the pay gap within firms on future productivity growth; that is, G6, in
equation 2 represents permanent changes in the rate of productivity growth that are due to
past changes in the ratio of executive to average employee pay.51

Finally, to address potential problems in the estimated standard errors that are due
to heteroskedasticity, the tables below report robust standard errors. Standard errors are

calculated using version 5.0 of the Stata software package, which utilizes the method

developed by White (1980).52

 

with 58 since the covariance of 8,-1 and pay,-1 is not equal to zero.

50 Note that in the estimates below, results from contemporaneous and one year lagged
changes are also presented.

5' The implied model is related to the discussion in Pindyck and Rubinfeld (1981) on
distributed lags. Extending the length of the lag structure in this data set quickly becomes
costly in terms of loss of degrees of freedom, due to frequently missing information on
either of the compensation variables and because of the relatively short time frame for
which observations are available.

52 At this time, tests for heteroskedasticity have not been performed.

86

Predictions

The ﬁrm-level relative compensation measure used in this analysis emphasizes
vertical equity. Given the public ill will that greets the annual announcements of CEO
compensation, it is assumed that this dimension of pay inequality is relevant. Since CEO
pay is highly publicized and firm speciﬁc, two of Encinosa, et al’s criteria for saliency
are met.53 The dramatic growth in CEO pay that occurred during this period--when many
household living standards were stagnating-may make the use of CEO pay especially
appropriate.

Predictions about the relationship between pay dispersion and performance are
ambiguous. If the relationship between pay inequality and productivity is persistent over
the time period indicated, 06 < 0 is consistent with an equity response to increasing wage
gaps, provided there are sufﬁcient controls for worker quality. Alternatively, a positive
estimate on 016 is consistent with high executive pay prompting increases in average

productivity, provided there is not appreciable bias from simultaneity or worker quality.

Empirical results
Summary statistics

Table 2.2 presents summary statistics on compensation levels and growth for the
CEOs and average employees in this data set. The mean level of salary plus bonus for
chief executive ofﬁcers is $879 thousand per year. The minimum value for average

annual CEO compensation is $162 thousand, and the maximum exceeds $12 million.

 

53 Recall, the three elements of saliency they emphasize are similarity, proximity, and
exposure. The latter two are especially well met.

87

Elimination of the 1 percent tails resulted in a slight decrease in the mean level of CEO
pay, to roughly $850 thousand.

The average compensation expenditures per employee for the firms in this sample
is roughly $41 thousand per year. The range of values spans from $11 thousand to $99
thousand. Clearly, executives in this sample are making substantially more than the
average worker, even excluding the value of long term incentive awards. The ratio of
executive salary plus bonus to average employee compensation expenditures exceeds 21
to 1.

On average the ratio grew over the sample period, as reﬂected in the percentage
change measures presented in growth columns. Executive compensation grew at an
average rate of nearly 9 percent. Even after eliminating the 1 percent tales, the mean
growth rate in executive compensation approached 7 percent. On the other hand, pay for
average employees grew at only one percent per year. Both the levels and changes
presented here for salary plus bonus are similar to the results presented in Hall and
Liebman (1997).

In addition to the levels and changes, one can compare the incidence of pay cuts
and the distribution of pay growth. Average employees took cuts in real wages in 44
percent of cases, whereas chief executives took real pay cuts in their salaries and bonuses
in only 36 percent of cases. However, the pay cuts for executives were more likely to be
very large, with executives taking cuts in pay of 20 percent or more in 10 percent of

cases.54 For average employees, the largest pay cuts exceeded 7 percent.55 Executives

 

54 Again, these developments occur in the measure of salary plus bonus only, which does
not include losses due to ﬂuctuations in the valuation of long term incentive awards such
as stock appreciation rights and other granted options. Using their more comprehensive
measure of CEO wealth, Hall and Liebman (1997) report 24 percent of executives lost

88

also received large pay increases, with 25 percent of observations exceeding 12 percent.
Fully ten percent received raises that exceeded 35 percent. For average employees, 10
percent of employees experienced pay growth of 8 percent or more.

Specifying the pay ratio differently, dummy variables were constructed to indicate
whether the direction of the pay change for CEOs matched or diverged from the other
employees in their firm. In 36 percent of cases, the reported changes both increased,
though in nearly 50 percent of cases the two went in opposite directions. In 28 percent of
observations, the pay of the CEO increased, while average employee pay decreased. In
21 percent of cases, the pay of the CEO fell while the pay of the average employee
increased. In 16 percent of cases, pay levels decreased for both groups. Table 2.2
provides plenty of evidence at the firm level to suspect that violations of fairness
standards occurred during the 19808. Relative pay developments frequently diverged,
and the rate of income growth for chief executives greatly exceeded that for average
employees.

Table 2.3 presents summary statistics for the basic set of variables. The table
indicates that during the time period, annual sales growth averaged about 1 percent per
year,56 while value added increased at a slightly slower rate. The table also indicates a

wide range of firm size, although all ﬁrms are relatively large. The number of employees

 

money in 1994, with a median loss of $3 million. In this sample, the median loss
represents a reduction of $82 thousand, dramatically understating the downside risk to
CEOs for poor performance. However, given that the gains and losses in the CEO’s
portfolio of compensation beneﬁts are largely hidden, the more visible components are
probably most relevant to this analysis.

5 Bell (1989) reported that in the early 80s, union wage concessions resulted in
compensation reductions of an average of 6 percent in real terms.
56 Measures are of differences in logs, so that actual percentage growth will be higher.
The log growth in the ratio (see table 3) is close to that reported by Abowd and Bognanno
1995.

89

in the smallest firm is greater than 800, while the largest firm reports having over

800,000 employees.

Regression results

The ﬁrst set of regression estimates are presented in table 2.4. Results are from
augmented translog production functions where value added per employee and sales per
employee are regressed on measures of pay dispersion along with control variables for
capital, labor, the coefﬁcient of variation for the top five most highly paid executives,
industry—level measures of pay and pay dispersion, years, and 2-di git industries.
Estimates from OLS, ﬁrm fixed effects, and ﬁrst difference regressions are presented.

OLS estimates indicate that the relationship between the CEO-average employee
pay ratio and output is negative. The coefﬁcient is likely to be negatively biased due to
omitted worker quality measures. If workers are paid their marginal product, then ﬁrms
with the most productive workers will have higher pay on average. Higher average
employee pay will decrease the relative pay ratio, resulting in a negative correlation
between it and output.

Thus, OLS estimates are probably biased. Turning to the fixed effects and first
difference estimates, the estimated relationship between the pay ratio and output becomes
positive and, in the case of difference estimates, statistically significant. The estimates
for ﬁxed effects are also positive but are not statistically signiﬁcant. Notably, omitting
the coefﬁcient of variation for ﬁrm executives from fixed effects results in a larger and

statistically signiﬁcant relationship in both the value added and sales regressions.S7

 

57 See table A3 in appendix A for examples of regressions omitting the executive
coefﬁcient of variation.

90

Leaving the executive coefficient of variation out of the difference equations has no
effect on the size of the pay ratio estimate or its standard error.

Because of potential incentive and sabotage effects of pay dispersion among
executives, the coefficient of variation for them may have a positive or negative effect on
productivity. In fixed effects estimates the effect is positive, but in difference estimates it
is negative. None of the estimates for this variable is statistically signiﬁcant.

As argued above, the estimates on the pay ratio are likely to be biased upward due
to simultaneity. One of the proposed ways of dealing with this problem is to lag the
endogenous variable and difference the equation. Table 2.5 presents results from such a
procedure. Only estimates on the relative pay measure are reported. Panel A presents
results from a 1 year lag, panel B gives results using a two year lag, and panel C gives
results from regressions including the contemporaneous pay ratio and both lagged
measures.

Results indicate that the effect of the 1 year lag is generally positive. The effect is
strongest in the differenced value added regression, where it is signiﬁcantly positive.
Notably, as discussed above one would expect the estimate to be biased. However the
sign of the bias should be negative given that a positive shock in t-l decreases the error
term in equation 2 and increases the differenced pay ratio.58 On the other hand, bias on
ﬁxed effects estimates should be positive for similar reasons. The estimates for both

sales and value added are relatively close to zero for ﬁxed effects estimates.

 

58 Recall that the error term in the difference equation is S8 = [8, - 8,-1], while the growth
in the pay ratio is Spay = (payu - pay,-2). In fact, using the differenced pay variable from
t-2 as an instrument for the I year lag indicates that the bias is downward, though, as
argued above, the two year lag may not be a good instrument since it may affect
productivity.

91

The second year lag for the pay ratio indicates a negative relationship between the
pay ratio and output. In both ﬁxed effects and difference estimates for value added, the
estimates are negative and statistically signiﬁcant at the ten percent level. These provide
the strongest indication that increasing pay differences between chief executives and the
average employee will result in decreased output. However, in the speciﬁcations that
include all three pay ratio variables, difference equations indicate positive and signiﬁcant
effects for contemporaneous and 1 year lags, with no effect from two year lags.

As in table 2.4 these estimates generally indicate thatthe relationship between pay
differences and productivity is positive. The exception is for the 2 year lagged measure
of pay differences, which is negatively related to productivity. As outlined above,
lagging two years in a difference equation will help eliminate simultaneity bias. It also
changes the interpretation of the estimated relationship. Instead of measuring the effect
of a contemporaneous change on a change in a dependent variable, the relationship
becomes dynamic, with the effect measuring changes in productivity growth rates. While
such a speciﬁcation does not follow directly from the fairness models of efﬁciency wages
presented in table 2.2.1, anecdotal evidence does suggest that changes in morale and

labor relations tend to persist over long periods of time.

Industry pay variables

Results from industry-level variables are mixed. Since the measures represent a
substantially different level of aggregation than the dependent variable, they are likely to
be measured with substantial error. The direction of the resulting bias is not easily

predicted. Furthermore, the estimates are quite sensitive to changes in specification,

92

though their inclusion does not change the estimated effects for other variables. Fixed
effects estimates corresponding to table 2.4 (see appendix table A3) indicate that the
coefﬁcient of variation for all three groups is positively related to output, controlling for
the average quality of workers in the industry. However, in moving from ﬁxed effects to
difference estimates, the effect for pink collar workers changes from significantly
positive at the 5 percent confidence level to significantly negative at the 10 percent
conﬁdence level. The results for sales regressions exhibit even less consistency between
the ﬁxed effects estimates and difference estimates. Given their unreliability, these

variables are not included in further analysis.

Instrumental variables estimation

The problem addressed in this section stems from the simultaneous determination
of the pay ratio and productivity. It is likely that the strength of the incentive or equity
effect of a given relative pay development is strongest shortly after the change has taken
place. Thus, it is desirable to identify the productivity effect contemporaneously with the
relative pay development.

One can identify in principle a contemporaneous effect by using an instrumental
variables approach (Hsiao 1986). The conditions for identifying the structural effect are
that the instrument used for identiﬁcation be correlated with the endogenous variable but
uncorrelated with the error term in the original model.

The instrument used here is the proportion of the ﬁrm’s common stock owned by
the board of directors net of the proportion owned by the CEO. This information is

reported to ﬁrm shareholders in annual proxy statements and was collected with the

93

executive compensation data. Quite plausibly, ownership may have no relation to
productivity independent of its effect on the compensation ratio. Suppose executives are
paid according to easily observed outcomes, such as changes in the market valuation of
the ﬁrm, and otherwise for efforts that are more costly to observe and which don’t
materially impact the valuation of the ﬁrm. Secondly, suppose that reputation serves as
an important bonding mechanism for executives so that all executives engage in the
difficult to observe aspects of the job regardless of board monitoring.59 Executives in
high ownership boards may be observed and compensated more for these activities
especially if high ownership makes boards less fearful of objections by shareholders and
the public over relatively high executive pay. To the extent that hard to observe
behaviors don’t contribute to productivity based on their being observed, then ownership
will be related to compensation but independent of productivity60 In addition, high
ownership may impact the denominator of the pay ratio by deterring wasteful
compensation expenditures that are otherwise unrelated to employee performance.

Just as easily, however, the association of ownership with executive pay may
imply differences in productivity. As outlined above, in models of hierarchical ﬁrms
productivity in the highest positions “echoes” through the performance of subordinate
employees (Rosen 1992). Increasing the quality of the employees at the top of the

hierarchy will affect the productivity exhibited throughout the ﬁrm. Thus, one might

 

59 An executive’s concern about his how he is perceived professionally will likely ensure
that he expends a large amount of effort regardless of monitoring by board members. See
Rosen (1991) for a discussion of reputation as a bonding mechanism and its limitations as
such. As Hall and Liebman observe, “CEOs are a self-selected group of high-effort,
overachieving individuals, it may be that lack of effort is not a first-order agency issue in
this population (p.9 1997).”

60 Jensen and Murphy (1990) argue that public reactions to high executive pay packages
may decrease CEO pay levels.

94

argue that variables picking up productivity differences between executives should not be
excluded from the original model. High board ownership implies that a substantial
proportion of board members’ personal financial interests are vested in the success of the
ﬁrm, so that they may have greater incentive to recruit and retain CEO candidates with
higher average ability. As Rosen states,

If...competence has extraordinary marginal product for top management positions

in large ﬁrms, how incompetence is revealed and handled must be important.

Formally the job falls to boards of directors. Yet there is much opinion and some

evidence...that boards are controlled by the CEO (p.207).

Board ownership, to the extent that it implies board independence from executive control,
will presumably be related to productivity through its impact on the quality of the chief
executive, implying that it should not be excluded from the productivity equation.

Table 2.6 presents results from instrumental variables regressions. The ownership
variable is speciﬁed in two ways, in its continuous form and as a set of dummy variables
representing particular intervals of ownership.“ In either form, board ownership is
correlated with pay, though the ﬁrst stage t statistics for the linear measure are
insigniﬁcant at conventional levels in both the sales (t=1.44) and the value added (t=1.35)

samples. The F statistics for joint signiﬁcance of the categorical variables are both above

4, with p values of .002.

 

61 Reaser (1997) uses the variable in its categorical form in an analysis of the
determinants of executive compensation. Since neither of the forms constitutes a
“nested” version of the other, a Chow test measuring the difference between the different
models is not straightforward (and therefore not carried out here). As an approximation
of this, one can compare adjusted R2’s from first stage regressions, since adjusted Rz’s
give explained variation controlling for available degrees of freedom. The ﬁrst stage
adjusted R2 for the continuous variable in the regression for value added (sales) is .833
(.834), while for the categorical variables it is .836 (.836). Given these slight differences
and the fact that the second stage estimates vary somewhat depending on the variable
used, results using both forms are presented below.

95

A problem does arise in the use of the variable, however. Specifically, the
implied direction of the bias on the pay ratio estimate changes depending on the way the
ownership variable is speciﬁed. Table 2.6 presents results from three regressions each
for value added and sales. The first column gives non-IV ﬁxed effects estimates for the
contemporaneous measure of the ratio of CEO to average employee pay. This column
corresponds to the ﬁxed effect estimates in table 2.4, though sample sizes have changed
modestly due to missing values in the instrumental variable. The second gives the IV
ﬁxed effects estimate using the continuous form of the ownership variable as an
instrument, and the third gives estimates using the set of dummy variables for
ownership.62

Fixed effects results reﬂect those of earlier regressions. Using the continuous
form of ownership as an instrument, changes in the point estimates imply that estimates
in the non-IV ﬁxed effects are downward biased. That is, they imply that the ﬁxed
effects estimates understate the beneﬁts of relatively high executive pay. The nonlinear
speciﬁcation for ownership, however, indicates that the bias is in the other direction, with
the point estimate for both value added and sales becoming negative.

Given the sizes of the standard errors, neither the differences between the OLS
and IV estimates nor the differences between the IV estimates for each output measure

are statistically signiﬁcant. In addition, the point estimates on other variables seem to be

 

62 The four categories are for ﬁrms whose boards own less than 1 percent of outstanding
ﬁrm stock, greater than or equal to 1 and less than 3 percent, greater than or equal to 3
percent and less than 10 percent, and greater than or equal to 10 percent. The variables
originally were constructed so that the proportion of observations in each of the three
categories with greater than one percent ownership was roughly equal. In the present
analysis, roughly 25 percent of observations fall into the 1 to 3 percent category, 21
percent fall into the 3 to 9 percent category, and 16 percent fall into the greater than 10
percent category.

96

sensitive to the speciﬁcation of the instrument, though the differences are not statistically
precise. Whether the instrumental variable results indicate either an upward or
downward bias in the contemporaneous measure of pay inequality, they do not indicate a
signiﬁcant relationship between employee productivity and relatively high pay for chief

executives.

Firm size

Along with similarity, the saliency of a reference person is likely to be determined
by proximity and exposure (Encinosa, et al 1997). It stands to reason that ﬁrm size will
affect one’s proximity and exposure to the chief executive ofﬁcer. This will be true
especially if smaller ﬁrms have fewer production facilities and are less geographically
dispersed than large ﬁrms. Assuming this to be the case, the regression sample was
divided by the median number of employees, and production function estimates were
acquired for the separate samples. Even in the small-ﬁrm sample, ﬁrms are still rather
large, with the smallest ﬁrm having 800 employees and the largest having around 25000
employees.

For small firms, results of estimates on the ratio of CEO to average employee pay

 

 

are,
value added sales
1 2 3 1’ 2’ 3’
FE FE:IVI FE:IV2 FE FE:IVl FE:IV2
.037 .263 -.332 .020 1.464 -.062
(.032) (.872) (.226) (.033) (2.240) (.182)

97

For large firms, results are,

 

 

value added sales
1 2 3 1’ 2’ 3'
FE FE:IVl FE:IV2 FE FE:IVl FE:IV2
.061 .047 -.163 .063 -.032 -.241
(.036) (.262) (.154) (.032) (.221) (.156)

There is no indication of substantial or systematic differences by firm size. Fixed effects
estimates without instrumenting indicate that the effect of increasing the relative pay of
executives is larger in large ﬁrms. However, instrumenting for the likely simultaneous
determination of the pay ratio and output indicates results that are substantially the same
as in the pooled sample. The implied direction of bias for small ﬁrms depends on the
way that the instrumental variable is speciﬁed. In general, the results for categorical
variables in columns 3 and 3’ suggests that the direction of bias in ﬁxed effects without
instrumenting is positive, though the estimates do not indicate that the true effect is
signiﬁcantly negative. For small ﬁrms, estimates in columns 2 and 2’, using the linear
measure for board stock ownership, indicate ﬁxed effects estimates in columns 1 and 1’

understate the true effect, though the standard errors are inordinately large.63

Ability to pay

The speciﬁcations discussed above presume to measure the effects of fairness.
There may be alternative speciﬁcations that better represent developments in pay which
people would characterize as fair or unfair. Campbell and Kamlani (1997) ask

respondents speciﬁcally if employees will withdraw effort when wages increase for

 

63 The ﬁrst-stage t statistic on the linear version of the instrument for small (large) ﬁrms
is .74 (1.37), while for categorical variables the F statistic for small (large) ﬁrms is 3.33
with a p value of .02 (2.27 with a p value of .08).

98

highly paid groups but remain stagnant for other employees. They also ask if effort
reductions differ depending on ﬁrm proﬁtability. In both cases more than 85 percent of
respondents agreed that such pay developments would result in moderate to large effort
reductions. Kahneman, et al (1986) demonstrate that households view nominal
compensation decreases during times of firm profitability as unfair.

To examine whether a response by employees is dependent on various qualitative
developments in pay and proﬁtability, the four groups of relative pay developments
identiﬁed in table 2.2 are used in the production function to indicate various pay
developments. To recap, the first is constituted of observations where the CEO received
an increase in pay and the average employee received an increase in pay (36 percent of
observations); the second group is made up of observations where the CEO received an
increase in pay but the average employee received a cut in pay (28 percent); the third
group is where the CEO received a cut in pay but the average employee received an
increase (21 percent); and the fourth is where both the CEO and the average employee
took a cut in pay (15 percent). Note that these variables are speciﬁed such that ﬁrms can
fall into any of the categories depending on the relative wage developments that take
place in a given year.

Presumably, during periods when the executive gains while the average employee
takes a cut in pay, the negative effects of fairness should be the strongest. Productivity
should be the highest when the developments in pay follow one another or when pay
levels converge.

Table 2.7 presents estimates from difference regressions including the dummy

variables described above. The specifications include pay developments from the

99

preceding two periods. This speciﬁcation is chosen because in table 2.5 it is shown that
this is the period with the most negative relationship between relative pay and
performance. Using the pay-development variables from earlier periods is not more
illuminating, while at the same time it is complicated with issues of endogeneity.

The results suggest that per employee value added is the highest when both the
CEO and average employee took pay cuts. Surprisingly, productivity growth is lowest
when CEO pay decreased and average employee pay increased, signiﬁcantly different
than the highest group at the ﬁve percent level. When ﬁrms widened the pay gap
between executives and the average employee, they also experienced low productivity
growth. In terms of magnitude, the level of growth for these ﬁrms was similar to the
level when CEO pay and average employee pay converged. The fact that developments
which most people would regard as more fair resulted in still lower growth casts doubt on
the interpretation that the low growth in ﬁrms with the most unfair pay developments is
attributable to an equity effect.

It may be true, however, that firms make such adjustments in pay when they have
monitoring technology in place that allows them to deter shirking or sabotage. If that is
the case, then monitoring intensity is positively correlated with both wage inequality and
productivity, and the point estimates will be positively, suggesting the negative effect for
cases of diverging pay levels is understated. Thus, the relative effects presented here may
not reﬂect the difference in employee responses that would occur if one were able to
control for monitoring intensity.

An additional test of whether ability to pay matters incorporates financial

performance. If a ﬁrm is relatively proﬁtable, diverging pay developments may have

100

different effects than when the firm is performing poorly. If fairness is a constraint on
behavior, profitable ﬁrms that impose pay cuts on employees while increasing the pay of
the executives should have lower productivity than when pay developments mirrored one
another.

In results not reported the four categories of relative pay developments were
interacted with a variable indicating that the ﬁrm earned more than the median rate of
return (either market return or return on assets) in each of the two preceding years."4
Eight categorical variables are then generated by interacting the high proﬁt (HP) variable
and (l-HP) with the four pay categories. Seven of these variables were included in
regressions to determine if productivity growth rates differed depending on the
interaction of pay developments and proﬁtability.

Results indicate that in cases when one expects the effect to be the most negative
(times when CEO pay increased and average employee pay decreased during periods of
high proﬁts), the effect is negative with marginal statistical signiﬁcance. However, the
results for when CEO pay decreased and average employee increased in proﬁtable ﬁrms
are more strongly negative. Thus, arguments that productivity will fall if the rules of

fairness are violated do not seem to be supported by the data.

Employee quality
Typically the efficiency wage production function is specified so that labor input

is augmented with an effort function, e(-), which is a function of the wage rate,

 

64 Proﬁtability is measured as the firm being above the median rate of return for a sample
of roughly 300 manufacturing ﬁrms in each of the two years before the wage changes are
measured. That is, in years t-4 and t-3.

101

Q = A K6 [e(w)xL"] . (3)
Wadhwani and Wall (1991) specify the effort function in the following way,

e(w) = -a + b (w/w*)Y u”, (3a)
where w is a worker’s own wage, w* is her comparison wage, and u is the unemployment
rate.65 The negative intercept assures that ﬁrms cannot acquire positive effort at zero
wage, and the functional form ensures that the elasticity of effort with respect to the wage
declines with the wage. Substituting equation 3a into equation 3 and taking logs gives
approximately,

log(Q) = log(A) + 0 log(K) + n log(L) + or log(w/w*). (4)
When L represents homogeneous labor, proﬁt maximizing conditions imply that or will
equal 11, or that the output gain from increasing the wage will equal the effect of
employing an additional worker at the market wage. For the present exercise, the two
types of labor are quite different, and the above model is not appropriate.66

However, following Wadhwani and Wall, industry-level data can be used to help

clarify at least part of the relationship in question. Namely, the above framework implies
that one can include the ratio of average employee pay in the firm to the average pay in
the industry in the production function. In so doing, one can control for labor quality,
insofar as it is manifested in average compensation expenditures, and then attempt to

identify the effect of high executive pay. In this section the following equation is

 

65 The unemployment rate is disregarded here, though it is intended to capture the
incentive effect of unemployment, since workers caught shirking will pay a higher
p‘enalty when u is high according to some versions of the efﬁciency wage hypothesis.

Each of the three models summarized above are of ﬁrms that employ heterogeneous
labor who have interdependent wage concerns. Given the very speciﬁc compensation
measure used here and the fact that the high wage labor is composed of one person, those
models are not amenable to the present analysis.

102

estimated,
log(Q) = log(A) + 9 mm + n log(L) +
a1 log(LAB/w*) + or; log(CEO) + 0.67 (5)
All variables are deﬁned as above, except that the new variable, w*, measures the
average annual income for workers in ﬁrm i’s industry. The measure is from the Current
Population Survey and is equal to the mean wage in industry i in year t multiplied by the
average number of hours per week for full time workers. This number is then multiplied
by 52 to get annual income. The mean of the resulting measure is slightly over 75
percent of the mean ﬁrm-level compensation expenditure. The variables are correlated at
slightly over .50. Of special interest here is the effect of high executive pay when
controlling for labor quality or effort. The question of whether on represents rent sharing
arrangements, relative labor quality, or an efﬁciency wage effect is left unexamined.
Focusing on the coefﬁcients for CEO pay, one sees that the pattern in the

resulting estimates is largely the same as in previous tables:

 

 

value added sales
FE FE:IVl FE:IV2 FE FE:IVl FE:IV2
.123 .193 -.053 .099 .817 .025
(.017) (.345) (.135) (.019) (.698) (.138)

The ﬁxed effects estimates on CEO pay are strongly positive, with a magnitude that is
several times higher than when using CEO as the numerator in the relative pay ratio.
Using the same instruments as in table (5), the results of instrumental variables are

largely the same as seen earlier. The implied bias depends on the speciﬁcation of the

 

67 The simple Cobb-Douglas form indicated here is used. Estimates on the parameters of
interest from translog functions are not materially different. The estimates presented are
identical regardless of whether per employee or total output measures are used on the left
hand side. Year effects and the executive coefﬁcient of variation are also included.

103

instrument. However, it is worth noting that the relationship between the linear
speciﬁcation and the endogenous variable is relatively weak, with a t statistic that is
around 1.2. The F statistic for the categorical variables in ﬁrst stage estimates remains
high, with a value near 4 and a p value of less than 1 percent. Again, the conclusion from
instrumental variables estimates is that the productivity beneﬁt of relatively high
executive pay is indistinguishable from zero.

Additional notes on this specification are in order: the relative wage variable,
LAB/w *, has a coefﬁcient that is very close to the estimate on L. For example the point
estimate in column 1 for relative wages is .83, while for labor input it is .86. They are not
statistically distinguishable in this case, though in others, especially for sales, they do
diverge a little more. The estimate is consistent with efﬁciency wage theory, as well as
with labor heterogeneity and rent-sharing arrangements. Also, including LAB alone when
estimating equation 5 results in a coefficient that is similar to the relative wage ratio
without appreciably changing the estimates from the other variables, though the

difference between the LAB coefficient and the L coefﬁcient tends to increase.

Productivity levels as a function of changes in pay
A final test of the relationship between executive-average employee pay

differences and productivity relates changes in relative pay to output levels. That is,
Q=f(K. L. Spay). (6)
where Spay=log(CEO/LAB), - log(CEO/LABM, and Q, K, and L are measured in period t

and speciﬁed as in equation 3.

104

Focusing on fixed effects estimates, the results for the CEO-average employee
pay ratio are somewhat different than earlier speciﬁcations. Fixed effects estimates from
equation 6 suggest that increasing the rate of growth in the pay ratio will decrease growth
in value added and increase sales, though both effects are statistically insigniﬁcant and
economically small. The speciﬁcation does not eliminate the correlation of the
explanatory variable with the error term, so ideally an instrumental variables approach
could help identify if there is an effect. The instrument used in previous sections of this
paper, board ownership of the ﬁrm, is difficult to relate to the change in the pay ratio,
however. Measured in either levels or changes, the board ownership variables perform

poorly. Other instruments have not been identiﬁed.

Conclusions

The ratio of CEO to average employee pay is used to determine whether widening
income differences are associated with higher or lower productivity. Theoretically, pay
dispersion can have positive or negative effects on productivity, though the results
presented here do not indicate that increasing or decreasing pay differences between
average employees and chief executives have a straightforward effect on the productivity
of employees.

The strongest evidence of negative effects from increasing earnings inequality
come from pay variables that are lagged two years. Several reasons suggest that the two-
year lag represents the most reliable estimates presented here. First, contemporaneous
measures are likely to be biased toward ﬁnding a positive correlation between relatively

high CEO pay and measures of performance. Second, measures that are lagged only one

105

year remain correlated with the error term when variables are differenced to eliminate
unobserved, time invariant ﬁrm effects. Finally, the instrument used here to identify a
contemporaneous effect does not provide clear evidence for or against the argument that
increased pay inequality leads to lower productivity, and superior instruments have not
yet been identified. While it is difficult in context of existing theoretical models to
interpret the results from the two year lag, the econometric interpretation is
straightforward: increasing the current ratio of executive to average employee pay will
decrease future productivity growth.

It is difﬁcult to interpret this relationship in terms of existing theory, though there
is reason to believe that these represent the most reliable estimates.

While it may be the case that employee morale manifests itself in growth rates
over long periods of time, other evidence presented casts doubt on the relationship being
the result of perceived inequity. Namely there is no indication that productivity growth is
lower in ﬁrms experiencing the most unfair developments in pay. Contemporaneous and
l-year lagged measures produce positive and statistically significant results when
controlling for ﬁrm-level heterogeneity. Instrumental variable methods do not provide
clear indications of the direction, or existence, of bias in contemporaneous measures,
much less indicate significant positive or negative effects from increasing or decreasing
the relative pay of chief executives.

The ﬁrm-level pay variable used here measures changing pay practices over only
one dimension of the ﬁrm--a dimension which may not be the most relevant in these
rather large firms. A more satisfying model would incorporate detailed information on

the methods of pay, relative pay levels over other dimensions of the ﬁrm, and measures

106

of other determinants of employee morale. Discussions in papers presenting fairness
variants of the efﬁciency wage hypothesis seem to imply such a model: relative wages
are shorthand for referring to the potential productivity effects of employing a variety of
human resource tools that bolster employee morale.

An additional qualification of these results is that they do not indicate the
relationship between relatively high executive pay and proﬁtability. Theories examining
the connection between executive pay and performance are concerned with performance
as it is measured by ﬁrm proﬁtability. Principal-agent theory, for example, is concerned
with the mechanisms that align executive income with the owners’ interests, implying
that the most important relationship is the link between executive compensation and the
market valuation of the firm. While productivity and ﬁnancial performance are related,
they are not equivalent. Complementary analysis should evaluate the relationship of pay
inequality and profitability.

Relative wage concerns have many dimensions that are difficult to disentangle.
Both fairness and incentive concerns are likely to be important aspects of ﬁrms’
compensation systems. The lack of ﬁndings in this paper do not indicate that relative
wage concerns do not matter. For example, relative pay developments for workers doing
similar work within firms, or in ﬁrms attempting to employ more team-oriented
production technologies, may have more important consequences than developments
occuring between CEOs and average employees. Further research should attempt to
isolate firm-level relative pay developments and employee productivity in order to

determine the extent to which such concerns determine ﬁrm productivity.

107

Bibliography

Agell, Jonas, Per Lundborg. (1992). “Fair Wages, Involuntary Unemployment,
and Tax Policies in the Simple General Equilibrium Model.” Journal of Public
Economics, v47 n3, pp299-320.

Akerlof, George A.; William T. Dickens; and George L. Perry. (1996) “The
Macroeconomics of Low Inﬂation.” Brookings Papers on Economic Activity, 1996 nl.

Akerlof, George A. and Janet L. Yellen. (1990) “The Fair Wage-Effort
Hypothesis and Unemployment.” Quarterly Journal of Economics. v105, n2, pp256-283.

Bairam, Erkin I. (1994). Homogeneous and Nonhomogeneous Production
Functions. Ashgate Publishing, Ltd., Avebury.

Blinder, Alan S. and Don H. Choi. (1990). “A Shred of Evidence on Theories
of Wage Stickiness.” Quarterly Journal of Economics, v105 n, pp.1003-15.

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

Bowers, Mollie H. and Roger D. Roderick. (1987). “Two-Tier Pay Systems:
The Good, the Bad, and the Debatable.” Personnel Administrator, v32 n6, pp101-108.

Brown, Charles. (1990) “Firrn’s Choice of Method of Pay.” Industrial and Labor
Relations Review.

Campbell, Carl M. HI; and Kunal S. Kamlani. (1997). “The Reasons for Wage
Rigidity: Evidence from a Survey of Firms.” Quarterly Journal of Economics, pp.759-
789.

Caves, Richard E. and David R. Barton. (1990). Eﬂiciency in 0.5.
Manufacturing Industries. MIT Press, Cambridge.

Cooper, Christine L. and Bruno Dyck. (1992). “Improving the Effectiveness of
Gainsharing: The Role of Fairness and Participation.” Administrative Science Quarterly,
v32 n3, pp471-.

Doeringer, Peter B. and Michael J. Piore, (1971). Internal Labor Markets and
Manpower Analysis. Heath, Lexinton, MA.

Encinosa, William E., Martin Gaynor, and James B. Rebitzer. (1997) “The

Sociology of Groups and the Economics of Incentives: Theory and Evidence on
Compensation Systems.” Mimeo.

108

Green, Michael K. (1992). “Fairness in Hierarchical and Entrepreneurial Firms.”
Journal of Business Ethics, vll n11, p.877-.

Greene, William H. (1990). Econometric Analysis. MacMillan Publishing, New
York.

Gorman, Raymond F. and James B. Kehr. (1992) “Fairness as a Constraint on
Proﬁt Seeking: Comment.” American Economic Review, v82 n1 (March), pp.355-8.

Groshen, Erica L. (1991). “Sources of Intra-Industry Wage Dispersion: How
Much Do Employers Matter?” The Quarterly Journal of Economics, v106, n3, pp869-84.

Hall, Brian J. and Jeffrey B. Liebman. (1997). “Are CEOs Really Paid Like
Bureaucrats?” NBER Working Paper 6213.

Harpaz, Itzhak. (1983). Job Satisfaction: Theoretical Perspectives and a
Longitudinal Analysis. Libra Publishers, Roslyn Heights, New York.

Holley, William H. Jr. and Kenneth M. Jennings. (1991). The Labor Relations
Process, Fourth Edition. The Dryden Press, Orlando.

Hsiao, C. (1986). Analysis of Panel Data. Cambridge University Press,
Cambridge.

Kahneman, Daniel, Jack L. Knetsch, and Richard Thaler. (1986) “Fairness as a
Constraint on Proﬁt Seeking: Entitlements in the Market.” American Economic Review,

Kanungo, Rabindra N. and Manuel Manungo. (1992). Compensation: Effective
Reward Management. Butterworths, Canada Ltd.

Klaas, Brian S. and John A. McClendon. (1996). “To lead, Lag, or Match:
Estimating the Financial Impact of Pay Level Policies.” Personnel Psychology, v49 n1,
pp.121-42.

Kruse, Douglas L. (1993). Proﬁt Sharing: Does It Make a Difference? W.E.
Upjohn Institute for Employment Research, Kalamazoo.

Leonard, Jonathan S. (1990). “Executive Pay and Firm Performance.” Industrial
and Labor Relations Review,

Levine, David I. (1991). “Cohesiveness, Productivity, and Wage Dispersion,”
Journal of Economic Behavior and Organization. v15, n2 (March), pp237-55.

Levine, David I. (1992). “Can Wage Increases Pay for Themselves? Tests with a
Production Function,” Economic Journal. v102, n414 (September), pp1102-15.

109

Levine, David I. (1993). “Fairness, Markets, and Ability to Pay: Evidence from
Compensation Executives” American Economic Review. v83, n5 (December), pp124l-
59.

Main, Brian G.M. Charles A. O’Reilly III, and James Wade. (1993). “Top Executive
Pay: Tournament or Teamwork.” Journal of Labor Economics. v11 n4, pp.606-28.

Martin, Joanne. (1982) “The Fairness of Earnings Differentials: An Experimental
Study of the Perceptions of Blue-Collar Workers.” Journal of Human Resources, v17, n1
(Winter), pp. 1 10-22.

Mitchell, Daniel J .B., David Lewin, and Edward E. Lawler III. (1990).
“Alternative Pay Systems, Firm Performance, and Productivity.” in Alan S. Blinder, Ed.,
Paying for Productivity, Brookings Institute, Washington DC.

Pfeffer, Jeffrey, and Nancy Langton. (1993). “The Effect of Wage Dispersion on
Satisfaction, Productivity, and Working Collaboratively: Evidence from College and
University Faculty.” Administrative Science Quarterly, v38 n3, pp382-.

Pindyck, Robert S. and Daniel L. Rubinfeld. (1981). Econometric Models and
Economic Forecasts. McGraw Hill, New York.

Rees, Albert. (1993). “The Role of Fairness in Wage Determination.” Journal of
Labor Economics, v11, n3, pp243-252.

110

Table 2.1: Literature Review

 

A. Theoretical Treatments: Equity

Authors:

Daniel S. Hamerrnesh

George A. Akerlof and

Janet L. Yellen

David Levine

Paper:

“Interdependence in the

Labor Market,”
Economica, 1975.

“The Fair Wage-Effort
Hypothesis and
Unemployment,” QJE,
1990.

“Cohesiveness,

Model:

Conclusions:

Employees adjust effort Equity concerns affect

to changes in relative
wages:
Y=FlEr{8(W1.W2)l.Ezl;
gl>OrgZ<0

Employees supply
effort in proportion of
own wage to fair wage:
e=min(w/w*, 1);
w*=f(w,-,.]-, wc)

Firms increase

short-run responses by
ﬁrms to exogenous
market wage changes.
Firms may employ
more high-priced labor
when their wages
increase.

Equity concerns cause
involuntary
unemployment and
explains wage
compression.

For some ﬁrms, wages

Productivity, and Wage production by investing vary less than
Dispersion,” Journal of in cohesiveness through productivity as ﬁrms
Economic Behavior and wage compression:

Organization, 1991.

B. Theoretical Treatments: Incentives

Authors:

Edward P. Lazear and
Sherwin Rosen

Sherwin Rosen

Paper:

“Rank-Order
Tournaments as
Optimum Labor
Contracts,” JPE, 1981

“Prizes and Incentives
in Elimination
Tournaments,” AER,
1986

q = C(wl/wH) f(H,L)

Model:

Executive labor market
are best characterized
by contests where pay
is based on order of
ﬁnishing rather than
measured output.

Near the top of
hierarchies, contestants
in tournaments may

slack off or rest on their

laurels.

111

compress the wage
structure to utilize
cohesion.

Conclusions:

The high pay for ﬁrst
prize induces
productivity over an
executive’s career.

Concentration of
rewards in top ranks of
executive hierarchies
assures that incentive
effect is maintained
during ﬁnal stages of
the game.

B. Theoretical Treatments: Incentives (continued)

Authors:
Edward P. Lazear

Table 2.1: Literature Review (continued)

Paper:

“Pay Equality and
Industrial Politics.”
JPE, 1989

C. Theoretical Treatments: Equity and Incentives

Authors:

William E. Encinosa
HI, Martin Gaynor, and
James B. Rebitzer

Paper:

“The Sociology of
Groups and the
Economics of
Incentives: Theory and
Evidence on
Compensation

Systems,” Mimeo, Nov.

1997.

D. Empirical Evidence: Surveys

Authors:

Daniel Kahneman, Jack

L. Knetsch, and
Richard Thaler

David I Levine

Paper:

“Fairness as a
Constraint on Proﬁt
Seeking: Entitlements
in the Market,” AER,
1986

“Fairness, Markets, and
the Ability to Pay:
Evidence from
Compensation
Executives” AER, 1993

Model:

In context of
tournaments, wide

wage disparities may be

detrimental to output if
worker behavior is
ubobservable.

Model:

Economic analysis of
pay system
incorporating the
concept of “group
norms,” along with use
of incentive pay.

Model:

Uses answers to
questions administered
in a household survey
to analyze attitudes
regarding “fair
behavior” in product
and labor markets.

Survey of
compensation
executives.

112

Conclusions:

Contestants may be
able to increase their
chances for promotion
by engaging in
sabotage. Therefore,
wage compression
leads to more efﬁcient
behavior.

Conclusions:

“Group norms” effect
the pay system used by
medical partnerships.
Size of practice is
positively related to
use of incentive pay
systems.

Conclusions:

Finds wide consensus
for a “dual
entitlements”
conception of fairness,
that takes into account
the

Compensation
executives place more
emphasis on
maintaining the
relative wage
relationships in a ﬁrm
than on responding to
changes in labor
market conditions.

Table 2.1: Literature Review (continued)

E. Empirical Evidence: Production Function and Productivity

Authors:
David I. Levine

J .R. Norsworthy and
Craig A. Zabala

John W. Straka

Peter Capelli and Keith
Chauvin

Paper:

“Can Wage Increases
Pay for Themselves?
Tests with a Production
Function,” The
Economic Journal,

1992

“Worker Attitudes and
the Cost of
Production,” Economic
Inquiry, 1990

“Effects of Worker
Attitudes on Production
Costs and the Value of
Capital Input,”
Economic Journal,
1985

“Is Poor Worker
Morale Costly to
Firms?” ILRR, 1993

“An Interplant Test of
the Efﬁciency Wage
Hypothesis,” QJE,
1991

F. Empirical Evidence: Executive Pay Dispersion

Authors:
Brian G. M. Main,

Paper:
“Top Executive Pay:

Charles A. O’Reilly HI, Tournament or

and James Wade

Teamwork,” JOLE,
1993

Model:

Information about the
pay of employees in
manufacturing plants
relative to the pay
levels of similar
workers in competing
ﬁrms.

Proprietary information
on union-management
grievances and other
labor-relations
information. Industry
productivity
information.

Uses Norsworthy and
Zabala framework,
substituting industrial
relation information
from BLS.

Uses plant-level wage
information from a
geographically
dispersed multi-plant
ﬁrm.

Model:

Compensation
information for

executives in a panel of

large ﬁrms.

Conclusions:

Finds that relatively
high wages are
associated with more
productive employees,
and that the
productivity gain is
sufﬁcient to offset
higher wages.

Decreasing worker
morale accounts for
declines in
productivity and
productivity growth.

Finds that Norsworthy-
Zabala results are not
robust and could be a
result of the functional
form used.

Finds a fewer
disciplinary dismissals
when wages are high
relative to average
wages in the
geographic area of the
plant.

Conclusions:

Finds limited evidence
that higher pay
dispersion is
deleterious to ﬁnancial
performance in ﬁrms
more heavily reliant on
team production.

 

113

Table 2.2: Pay Levels and Pay Growth for CEO’s and Average Employees,

 

 

1980 - 1991
Chief Executive Average
Officer Employee
Mean pay levels 878.9 41.1
(597.7) (10.6)
Sample size 1002 1103
Mean pay growth .087 .009
(.5 15) ( .085)
Distribution of pay growth:
10th percentile -. 197 -.074
25th percentile -.051 -.027
50th percentile .039 .008
75‘h percentile .126 .044
90‘h percentile .351 .084
Proportion with positive pay growth .636 .560
Proportion of firms where:
CEO pay increases and average pay increases .358 --
CEO pay increases and average pay decreases .278 --
CEO pay decreases and average pay increases .202 --
CEO pay decreases and average pay .162 --
decreases
Sample size 717 895

 

Standard deviations are reported in parentheses. All dollar values have been adjusted
to 1991 levels using the consumer price index for urban areas. Levels represent
thousands of dollars, and growth measures represent real changes. Sample sizes are
smaller in the CEO column because observations for the first, incomplete year of
executive tenure are omitted. The minimum (maximum) value for CEO pay is 162.4
(12513.4), while for average employee pay it is 10.95 (98.97). The minimum
(maximum) value for CEO pay growth is -.528 (11.272), while for average employee
pay it is -.530 (.491).

114

Table 2.3: Summary Statistics, Means (Standard Deviations)

 

 

Levels Growth
Mean Min Max Mean Min Max
Log(sales/L) 5.054 3.760 6.963 .010 -.261 .336
(.523) (.082)
Sample size 767 689
Log(value added/L) 4.476 3.445 5.784 .009 -.299 .250
(.413) (.086)
Log(K/L) 4.022 1.793 6.527 .036 -.228 .359
(.892) (.091)
Log(L) 3.095 -.127 6.776 -.011 -.458 .243
(1.284) (.085)
Log(K/L)xlog(K/L) 16.973 3.216 42.599 .293 -2.623 2.918
(7.775) (.745)
Log(L)><log(L) 11.226 .000 45.918 -.090 -3.848 1.970
(8.219) (.558)
Log(K/L)><log(L) 12.381 -.604 30.080 .077 -1.245 1.996
(5.737) (.379)
Log(CEO/LAB) 2.999 1.403 4.778 .035 -.788 1.636
(.553) (.237)
Industry pay variables:
Log(MEch) 2.860 2.520 3.064 -.004 -.261 .238
(.109) (.046)
140g(C0Vwc) -.965 -l.293 -.568 -.003 -.525 .321
(.142) (.135)
Log(MEDbc) 2.468 1.760 2.873 -.010 -.107 .089
(.185) (.029)
Log(COVbc) -l.012 -1.135 -.676 .008 -.487 .252
(.123) (.097)
Log(MEDpc) 2.376 2.012 2.645 -.002 -.266 .289
(.131) (069)
Log(COVpc) -.988 -1.279 -.530 .002 -.550 .450
(.133) (.169)
Sample size 743 674

K

115

Table 2.4: The Effect of Pay Differences on Output

 

 

Log(value added/L) Log(sale/L)
OLS FE Dif OLS FE Dif
Log(CEO/LAB) -.049 .037 .037 -.041 .038 .034
(.028) (.025) (.017) (.030) (.023) (.014)
Log(COVem) .065 .015 -.029 .029 .014 -.013
(.032) (.029) (.017) (.035) (.028) (.015)
Log(K/L) .223 .112 .397 .217 .288 .191
(.165) (.230) (.222) (.155) (.192) (.198)
Log(L) -.178 -.308 -.216 -.249 -.478 -.396
(.143) (.107) (.150) (.048) (.100) (.136)
Log(K/L)XLog(K/L) -.006 .002 -.040 .008 -.031 -.028
(.020) (.026) (.024) (.019) (.021) (.022)
Log(L)><Log(L) .007 -.016 -.005 .019 -.021 -.004
(.003) (.011) (.016) (.005) (.011) (.014)
Log(K/L)xLog(L) .030 .057 -.011 .035 .091 .028
(.090) (.022) (.025) (.009) (.018) (.023)
Sample size 738 738 669 762 762 684

 

Robust standard errors are reported in parentheses. OLS and difference estimates also

include industry dummy variables and all specifications include year effects, and

industry-level pay variables.

116

Table 2.5: The Effect of Pay Differences on Output, Alternative Lag Structures

 

 

Log(value added/L) Log(sale/L)

OLS FE Dif OLS FE Dif

A. Log(CEO/LABM -.033 .015 .033 -.058 .005 .015
(.029) (.024) (.016) (.032) (.023) (.014)

Sample size 603 603 552 640 640 584

B. Log(CEO/LAB)(.2 -.O41 -.044 -.025 -.076 -.028 -.007
(.027) (.029) (.013) (.028) (.021) (.014)

Sample size 533 533 448 588 588 488

C. Log(CEO/LAB) .021 -.014 .055 .055 .032 .051
(.055) (.042) (.024) (.058) (.033) (.021)

Log(CEO/LAB)l-l .055 .014 .055 .016 —.014 .036
(.059) (.029) (.023) (.064) (.027) (.017)

Log(CEO/IAB)l-2 -.115 -.022 -.008 -.140 -.016 .007
(.047) (.027) (.020) (.055) (.024) (.016)

Sample size 378 378 341 384 384 347

 

Robust standard errors are reported in parentheses. Estimates are from augmented

translog production functions.

117

Table 2.6: The Effect of Pay Differences on Output, Instrumental Variables

 

 

Value Added Sales

FE FE:IV1 FE:IV2 FE FE:IVI FE:IV2

Log(CEO/LAB) .039 .120 -.195 .038 .505 -.055
(.026) (.301) (. 153) (.024) (.375) (.126)

Log(COVmC) .014 -.027 .130 .010 -.222 .057
(.029) (.152) (.080) (.028) (. 189) (.066)

Log(K/L) .137 .212 -.081 .287 .729 .199
(.242) (.387) (.280) (. 195) (.446) (.228)

Log(L) -.391 -.330 -. 180 -.470 -.677 -.429
(.109) (.183) (.151) (.099) (.221) (.120)

Log(K/L)xLog(K/L) .004 -.003 .024 -.028 -.069 -.020
(.027) (.034) (.029) (.023) (.037) (.024)

Log(L)xLog(L) -.014 -.012 -.020 -.O20 -.013 -.022
(.012) (.012) (.014) (.011) (.017) (.011)

Log(K/L)><Log(L) .051 .050 .052 .090 .087 .091
(.023) (.023) (.024) (.018) (.023) (.019)

 

Sample size for value added is 709 and for sales is 734. Robust standard errors are
reported in parentheses. A continuous measure of board ownership is used as an
instrument in columns IV], while categorical measures are used in 1V2.

118

Table 2.7: The Effects of Pay Differences on Output, Difference Estimates
Using 2-Year Lagged Relative Pay Developments

 

 

Log(value added/L) Log(sales/L)

CEO pay increased -.025 -.000
& ave pay decreased (.013) (.010)
CEO pay decreased -.028 -.016
& ave pay increased (.013) (.011)
Both CEO pay -.013 .002

& ave pay increased (.012) (.010)
Sample size 448 488

 

Robust standard errors are reported in parentheses. Relative pay developments are
between years t-2 and t-3.

119

APPENDIX A

This appendix presents the results from the estimation of alternative specifications
of production functions. It does so by estimating and comparing the results from three
different functional forms: Cobb-Douglas, constant elasticity of substitution (CES), and
translog production functions.68 The purpose of this appendix is threefold: ﬁrst, it brieﬂy
reviews the literature on production functions and some of the issues in productivity
analysis; second, it serves as an exploratory exercise to determine if any of the three
functional forms introduced above ﬁts the data best; and third, it checks for the
robustness of the results presented in the text, to determine if the results are sensitive to
the functional form used.

The estimation of production functions stems from the theory of production in
neoclassical economics. The production function illustrates the technical relationship of
how inputs are transformed into outputs:

Q = f(K, L), (A1)
where Q is output, K is capital stock, L is effective hours of labor input (measured here as
the total number of employees). It is assumed that f,->O and f),<0 for i=K,L. In terms of
the familiar Cobb-Douglas production function, equation (A1) can be expressed,

Q = A K" L", (A2)

where A represents a scale technology parameter. Dividing through by labor gives,
Q/L = A (K/L)° UN“ . (A3)

Taking the log of both sides and adding an error term gives a convenient estimating

equation:

 

68 Caves and Barton (1990) present the specific forms used here. Bairam (1994) presents

120

ln(Q/L) = 0(0 + or] ln(K/L) + a2 ln(L) + g", (A3a)
where (10 = logA, or] = 0, and or; = n+0—1.

A variety of other possible specifications for production functions have been
explored in the literature. One such equation is a Taylor expansion of a CBS equation:

ln(Q/L) = Bo + B, ln(K/L) + 62 ln(L) + [33 ln(K/L)2 + e, (A3b)
A third possible form is the translog production function:
ln(Q/L) = to + )1 mac/L) + 72 ln(L) + Y3 ln(K/L)2 +
y, ln(L)2 + 75 ln(K/L)xln(L) + w, (A3c)
The translog form is a “consistent second order approximation to any production function
that takes the form an = F (an, lnL) (Caves and Barton, 1990, p.21).”

CBS and translog functions are more ﬂexible representations of the production
process in that they relax assumptions implicit in the Cobb-Douglas form about the
homogeneity of production and the elasticity of substitution between inputs. Tests of the
parameters on second order inputs in (A3b) or (A3c) indicate whether Cobb-Douglas is a
sufﬁcient speciﬁcation of the production relationship.69

A noteworthy source of bias associated with the estimations of production
functions stems from the likelihood that ﬁrms simultaneously choose capital and labor
with their levels of output. Caves and Barton note that the estimation of single-equation
production functions implies the quite strong assumption that capital and labor inputs are
chosen, ex ante, and are uncorrelated with the error:

the production function is made stochastic with a disturbance representing purely
random elements, on the assumption that the entrepreneur cannot know the effect

 

a survey of various production functions.
69 See Greene (1990), Christensen, Jorgenson, and Lau (1973), and Bairam (1994) on
how to recover the elasticity estimates from regression parameters.

121

of the disturbance until after the quantities of the inputs have been preselected.

The procedure further assumes that the prices of output and inputs either are

known with certainty or are statistically independent of the production function

disturbance and that entrepreneurs maximize the mathematical expectation of

profit (p.25).

A possibly richer estimation of the effects of increasing inequality that would
circumvent the above assumption includes estimating a system of factor cost equations
which relate the cost of inputs to their share of production (Caves and Barton 1990, and
Norsworthy and Zabala 1990). Such an exercise is not carried out here due in part to the
implied endogeneity of wages in the model tested here. Economic theory assumes that
ﬁrms are price takers both in product and input markets. This assumption leads to duality
between proﬁt maximization and cost minimization for the ﬁrm, and it provides the
rationale for cost equation estimation. The implied endogeneity of wages complicates the
use of such a procedure here.70

On this point, Bairam (1994) argues that the cost minimization assumption may
be dubious for such reasons, so that the cost equation methods of estimating technical
efﬁciency in production are also suspect. He argues, consequently, that estimation of
output equations is superior to cost equation estimation. The relative validity of either
argument is left unresolved here, except insofar as this exercise is consistent with the
latter.

As is common among analyses of firm production, the speciﬁcations here are
augmented with other variables. Speciﬁcally, all speciﬁcations include year and 2 digit

industry controls, and table A.3, as in the text, includes relative compensation variables.

Before turning to the results in this section, a ﬁnal note on the use of “augmented”

 

70 See Agell and Lundberg (1992) for an analysis of general equilibrium conditions when
ﬁrms face “fair wage-effort” constraints.

122

production functions is in order. The estimating framework implies that widening pay
gaps affect production in a “factor neutral” manner. That is, the functional form will not
help determine if wage inequality affects output through “indirect” avenues, such as by
shifting the relative use of inputs. To illustrate this, consider the following model,

Y = f(K, L1(W1. W2), L2( W2», (A4)
where K is capital, L, can be thought of as effective effort for low-wage labor, L2 as effort
for high-wage labor, w, is the wage for L1, and w; is the wage for L2.71 Changes in
relative wages may impact the technical efficiency of producing Q both through its direct
impact and through its effects on the use of the other inputs, such as the relative use of L1
and L2. Hamermesh, for example, argues that the demand for high wage workers can
increase with increases in w; if low wage workers retract effort sufficiently in response to
their lower relative wage.72 The effects of relative wage changes on factor use or
substitution elasticities are not explored in this analysis. That is, by including them as
additive terms to the production function, the interactions are implicitly set to zero.73

Table A.1 presents results from regressing output on the various speciﬁcations for
capital and labor. Three procedures were used: OLS, ﬁrm ﬁxed effects, and ﬁrst
difference. The first three columns give results for the Cobb-Douglass speciﬁcations, the

third through sixth columns give specifications for a constant elasticity of substitution

(CES) speciﬁcation, and the last three columns presents results from translog regressions.

 

7' This model is essentially the same as Hamermesh (1975). See table 1 for examples of
roduction functions which reﬂect the model outlined here.

2 This is because the change in effort level changes the slope of the marginal rate of
substitution between high and low wage workers. Hamerrnesh’s model, however,
assumes that ﬁrms are wage takers. While the discussion is instructive, this assumption,
as discussed in the text, is not likely to be tenable.

73 See Norsworthy and Zabala (1982, 1985, 1990) for similar analysis that speciﬁes
production such that the interactive effects are identified.

123

Other control variables in table A.1 include year and 2-di git industry ﬁxed effects.
Sample sizes in these regressions are much larger than those presented in the text because
of the substantial portion of missing values in the compensation variables, especially for
the data available from Compustat. Similar tables are presented below for the text
sample.

Results reject the Cobb-Douglas production function in most cases. F-statistics
for the translog production function relative to the Cobb-Douglas imply rejection of the
hypothesis that Y3 = Y4 = 75 = 0 in 4 of 6 cases (2 of 3 cases for value added and 2 of 3 for
sales) For OLS regressions, the F-statistic exceeds 7 for both output measures. For fixed
effects estimates, the F-statistic is above 4. In all cases, p values are less than .01. For
difference estimates, the F-statistics do not reject the Cobb-Douglas speciﬁcation in any
case. Similarly the F-statistics imply that the translog function is preferred to the CES
function in 4 of 6 cases. CES is only preferred to Cobb-Douglass in one case.

Table A2 gives results that correspond to the sample used in the text but that
exclude the all compensation variables. The point estimates are similar to the results in
table A.1, as is the preference for tranlsog functions. In this case, both Cobb-Douglas
and CES functions are rejected in favor of the translog speciﬁcation in the same 4 of 6
cases. F—statistics for difference equations are moderately larger, though they still do not
imply rejection of the Cobb-Douglas specification.

Given the generally positive results for the translog function, it will be used for
regressions presented in the text. The regressions do tend to exhibit inordinately strong

decreasing returns to scale, however, with the estimates for ﬁxed effects and differenced

124

regressions being very large. Point estimates on the first-differenced translog equations
compare favorably to those presented in Kruse (1993).

Table A3 gives estimates that include the relative compensation variable. It is
clear that estimates on relative pay are not sensitive to which speciﬁcation is used.
Neither is there any indication of omitted variable bias attributable to the compensation

variable in estimates for the capital and labor variables.

125

Table A. 1, Production Function Estimates: Full Sample

 

 

Cobb-Douglas CES Translog

OLS FE Dif OLS FE Dif OLS FE Dif

A.) Log(value added/L)
Log(K/L) .281 .326 .071 .293 .515 .150 .139 .293 .123
(.015) (.027) (.029) (.105) (.112) (.134) (.114) (.138) (.178)
Log(L) -.026 -.117 -.230 -.026 -.127 -.232 -.165 -.183 —.294
(.006) (.019) (.031) (.006) (.022) (.031) (.035) (.073) (.124)
Log(K/L)xLog(K/L) -- -- -- -.002 -.023 -.010 .006 .010 —.009
(.013) (.013) (.017) (.013) (.014) (.019)
Log(L)XLog(L) -- -- -- -- -- -- .003 -.015 .007
(.003) (.008) (.013)
Log(K/L)XLog(L) -- -- -- -- -- -- .030 .037 .006
(.007) (.014) (.021)
Sample size 1345 1345 1345 1345 1345 1345 1345 1345 1345

B.) Log(sales/L)

Log(K/L) .366 .282 .113 .283 .257 .242 .286 .238 .248
(.011) (.015) (.017) (.046) (.051) (.060) (.046) (.055) (.064)
Log(L) -.022 -.099 -.229 -.021 -.098 -.229 -.083 -.138 -.213
(.004) (.015) (.019) (.004) (.015) (.019) (.017) (.042) (.050)
Log(K/L)xLog(K/L) -- -- -- .011 .004 -.018 .008 -.003 -.020
(.006) (.006) (.008) (.006) (.007) (.008)
Log(L)XLog(L) -- -- -- -- -- -- .008 -.017 -.005
(.002) (.005) (.006)
Log(K/L)><Log(L) -- -- -- -- -- -- .008 .029 .001
(.005) (.008) (.010)
Sample size 4063 4063 4063 4063 4063 4063 4063 4063 4063

 

126

Table A2: Production Function Estimates, Text Sample without Relative Compensation Variables

 

 

Cobb-Douglas CES Translog

OLS FE Dif OLS FE Dif OLS FE Dif

A.) Log(value added/L)
Log(K/L) .281 .286 .065 .406 .360 .353 .251 .073 .403
(.021) (.053) (.040) (.153) (.290) (.185) (.162) (.235) (.225)
Log(L) -.028 -.165 -.267 -.029 -.168 -.278 -.193 -.345 -.200
(.007) (.037) (.047) (.007) (.042) (.048) (.044) (.108) (.151)
Log(K/L)xLog(K/L) -- -- -- -.016 -.009 -.036 -.008 .003 -.039
(.019) (.025) (.024) (.020) (.026) (.025)
Log(L)XLog(L) -- -- -- -- -- -- .007 -.012 -.007
(.003) (.011) (.016)
Log(K/L)xLog(L) -- -- -- -- -- -- .031 .064 -.009
(.009) (.022) (.026)
Sample size 743 743 674 743 743 674 743 743 674

B.) Log(sales/L)

Log(K/L) .404 .286 .062 .425 .750 .347 .241 .329 .217
(.020) (.043) (.038) (.150) (.189) (.174) (.154) (.197) (.202)
Log(L) -.008 -.215 -.293 -.008 -.239 -.305 -.265 -.483 -.370
(.009) (.034) (.044) (.009) (.035) (.045) (.049) (.099) (.137)
Log(K/L)xLog(K/L) -- -- -- -.003 -.056 -.036 .006 —.039 -.029
(.019) (.023) (.022) (.019) (.022) (.023)
Log(L)xLog(L) -- -- -- -- -- -- .020 -.019 -.005
(.004) (.011) (.015)
Log(K/L)XLog(L) -- -- -- -- -- -- .035 .093 .025
(.010) (.018) (.024)
Sample size 767 767 689 767 767 689 767 767 689

 

127

Table A3: Production Function Estimates, Text Sample with Relative Compensation Variables

 

 

Cobb-Douglas CES Translog

OLS FE Dif OLS FE Dif OLS FE Dif

A.) Log(value added/L)
Log(K/L) .274 .303 .053 .379 .383 .347 .225 .123 .387
(.022) (.054) (.039) (.153) (.213) (.180) (.162) (.231) (.222)
Log(L) -.023 -.l91 -.266 -.024 -.195 -.277 -.182 -.325 -.215
(.010) (.035) (.047) (.009) (.040) (.047) (.043) (.105) (.150)
Log(K/L)xLog(K/L) -- -- -- -.013 -.010 -.037 -.006 .001 -.039
(.020) (.025) (.023) (.020) (.026) (.024)
Log(L)xLog(L) -- -- -- -- -- -- .006 -.015 -.006
(.003) (.011) (.016)
Log(K/L)xLog(L) -- -- -- -- -- -- .031 .059 -.007
(.009) (.021) (.025)
Log(CEO/LAB) -.023 .051 .030 -.021 .052 .030 -.025 .047 .030
(.026) (.021) (.015) (.025) (.021) (.015) (.025) (.020) (.015)

Industry pay variables
Log(MEch) .429 .236 .102 .425 .232 .102 .374 .221 .102
(.240) (.146) (.080) (.240) (.145) (.080) (.238) (.142) (.080)
Log(COVwc) .253 .185 .036 .249 .180 .036 .224 .155 .036
(.158) (.081) (.029) (.158) (.082) (.029) (.161) (.082) (.029)
Log(MEDbr) .213 -.005 -.024 .202 -.002 -.009 .156 -.121 -.008
(.281) (.188) (.116) (.282) (.187) (.117) (.281) (.188) (.117)
Log(COVbc) -.001 .114 .068 .005 .121 .069 -.012 .105 .069
(.141) (.084) (.030) (.145) (.084) (.030) (.142) (.082) (.030)
Log(MEDpC) .097 .051 .107 .097 .055 .110 .139 .070 .109
(.180) (.116) (.050) (.182) (.118) (.050) (.180) (.114) (.056)
Log(COVpc) .160 .129 -.034 .159 .127 -.032 .158 .130 -.032
(.076) (.042) (.018) (.076) (.043) (.017) (.077) (.044) (.018)
Sample size 743 743 674 743 743 674 743 743 674

128

Table A3: Production Function Estimates, Text Sample with Relative Compensation Variables

 

 

(continued)
Cobb-Douglas CES Translog
OLS FE Dif OLS FE Dif OLS FE Dif
B.) Log(sales/L)

Log(K/L) .397 .304 .057 .399 .721 .336 .215 .296 .180
(.020) (.044) (.036) (.151) (.190) (.170) (.153) (.192) (.197)
Log(L) -.001 -.244 -.294 -.001 -.266 -.305 -.254 -.491 -.397
(.011) (.034) (.044) (.011) (.035) (.045) (.049) (.097) (.136)
Log(K/L)xLog(K/L) -- -- -- -.0003 -.050 -.036 .009 -.032 -.027
(.019) (.023) (.022) (.019) (.021) (.022)
Log(L)xLog(L) -- -- -- -- -- -- .019 -.020 -.004
(.005) (.011) (.014)
Log(K/L)xLog(L) -- -- -- -- -- -- .035 .093 .030
(.010) (.017) (.023)
Log(CEO/LAB) -.029 .050 .032 -.029 .055 .032 -.031 .046 .032
(.028) (.021) (.013) (.027) (.020) (.013) (.027) (.019) (.013)

Industry pay variables
Log(MEch) .473 .091 .038 .473 .081 .042 .403 .046 .043
(.297) (.140) (.081) (.298) (.138) (.081) (.296) (.132) (.081)
Log(COVwr) .220 .168 .015 .220 .145 .014 .179 .110 .015
(.146) (.067) (.024) (.146) (.068) (.024) (.143) (.065) (.023)
Log(MEDbc) -.069 .160 -.177 -.069 .171 -.166 -127 -.001 -.170
(.327) (.171) (.101) (.328) (.167) (.101) (.325) (.169) (.102)
Log(COVbc) -.251 -.174 .011 -.251 -.143 .011 -.249 -.170 .008
(.151) (.087) (.028) (.152) (.085) (.028) (.149) (.081) (.028)
Log(MEDpc) -.208 -.110 .102 -.208 -.093 .104 -.139 -.067 .107
(.212) (.114) (.048) (.213) (.111) (.047) (.202) (.104) (.047)
Log(COVpc) .085 .103 -.033 .085 .094 -.030 .072 .095 -.031
(.081) (.038) (.015) (.084) (.038) (.015) (.081) (.037) (.015)
Sample size 767 767 689 767 767 689 767 767 689

 

129