.1 1'. an r um.--

r-1'

”v...

I...

w

1'1

4““—

 

.‘. J.- .-

 

    
 
   
   
   
     
      
  
  

    

 
  
   

 
     

  
   

    
          

     

      
 
 
  

   

    
  
  
 
 
 
  
   

   
   
   

  

 
    
         
     
    
   
  
  
   
    

      
       
  
      

 
 
  
 

   
   
 
 
 
 
  

 
      

 

   
      

  

     

I
1.333”: 3.3“ 1
“JV l '
U M

     
 
 
  
  

  
 
 
 
 
 

W95.
4::

.3 .1: .
i o

x:
' "‘J..—"‘J‘""':.
TI}:

""JI 1'31.-

" -‘,
o-«vn
1

    

I. .r. ' . VI I
‘ :IISI'I 3 :J J“ "~“ . :-
1'. 3 | . .3. .J. ‘ I|.,I.:_ um. .14.... J .. 3 1-.-
| ' ; I; 3'. I 5 r .1 , ._-a ‘ , ,
J ..'.J. 7:. J :JJ J'JJIJ‘J . ' 13 ‘-‘:v . . 'E‘Zi‘flf '21} ' " 2.-
';II:*; .:' I -. I I» I .... m, .u . ~ 3:... I“ -. w»
('0 .‘Il .J "I”! I I In" 3 '3'3J3 '. if}: 3.2.3 JII33I3; 33.}. 3‘ 31"” .3333 '45: 1““ (3:33" “1.0313932: ...: 3 3 fizhfihfi‘ "éhts:
J JJJJ ‘JUJJ‘J'L‘J‘ J" ‘IJ.‘:' ‘ JI'J LJ J; 5‘ .JE' 'IJJJJJIIII'_1J"J.J 2 . :1::.JJJ;" IJ;:1|!'_I3_IJ§$' '3:Jf‘i}IJ!43—.‘;; @le .
. - -. ' I "1 »-'I.».. ‘ .I .~.. ' t':~"‘r.';r
W; I.-’ 31‘» . .1' I ‘ 1.; 33' It- 'II'. .1 r. .~.~.I WI .. .I —. ..,.:»:I:'4I.131’3.%
- .,,-. I .3. I... ., ..., V... .I .1: .. 2'; ; . -I. :35,” ; 3:41 ‘1‘ ,,._-.-
I 3.1, ; 14' 3. ..1_. "l mu.- .. JJ-,J .f'l' ‘JI 33th 4...,“ fly; . I.» .. ganja.»
:" 57:31:13 J ‘i' I""‘.""I'.érll.!' J‘" «1;:3'3533; 31.x J ”Ema-4”“:1I‘I'
0; {II J JJ:j'.3. .' ‘ J' J‘.’"1I “#1.?!“ .1: ; ‘1J'r;:‘iJ_J:LJJ'.';"‘-‘.3 EJ ' ' 30 . {I 13.22? Jfleéfi
.3 .3" '. I... 3:3 .3“ 7- 3:3.“ .3 '3:-X3I ." . ‘2; ' U? 3‘3 I 2"; gar-3:34:32 = 11%,?
3.1 '. .1. .I. 3‘.J'?.' 3 J3J ' .‘I’JJ. IJ .L 3?." ‘I“".‘J ' 'H 33‘ ; .‘Jl gfi:".‘?.!!e ;.
3 1H3 , 3 . 3.,1 .. .3. .. 53.? "'1‘.” 331' 3.. in“. 3 A _. “3. ”lg-'1 ,§.J?;.,.3E.;.
III-IzIz-w . ;I :1. I .11..» .H’u man-«w
'3 I‘J .lu‘» 3.” 3’.‘i}..;,f'1"‘.=x ~34: .qu‘ug‘dwzgi . {gfiu' .-
J‘ "‘ " ‘ ‘ ”5" -~ 3.39. IJIf.I,JII.I-II , JJ 3'.’ 'Li‘: " .335?
9 L, ' ‘3. J ' .' ’
J.
{'9‘ '

   
   
  

 

 
 
 
  

A:
“IE...
3.3.

.3

i waive-1 4‘

        
       
   

--.
.
0|.
. ~ 353:3.-
. at,

  
  

..I- ;. .
.j'
.l
.51
"'

  

~

- b- J."-

 
 
 
 
  
    

 
 
    
  
  

v
a... .

   
    
   

“1
iii. Jan 0.
h . _3

3.,"-
. ‘1'

n.
u.

"3.“:
w

 

1 .
III?

  

 

 
   
       

...-

“Pr

  
    

 
  
    
  

 
 

“2:3 if, i. 335'
333'?! I. J; 3 31.3.3. .

    

“ ' 37.”: 3"..7..

       
   

   

   
 
 
  

       
   
     

    

 

   

I . ‘J 3 ’ .r‘ J' .
I » J ;. J 1": .3333':3:’1J‘3. ‘ it»:
.IJ JJ JIJJJ!‘ J. 1%.;.5: I“ .lJJEjM: .3:ng {fig
.1 ' 5." ' I t “av-31.33.11.355: *I 1: :33.
351' ' " I II. JJJI'IJEI % .t I; JQJEI‘JE ‘i; '
.IJJ. . J | JJJJJ ..‘|:J:J.: J {ITJ 131333. .3... 33133 3
LI “I J, J ['3' . HJ.‘ HI .I, .‘ JP.“ 3 :. I .3333 .5‘
JI“"‘!"' ..‘J.“;I.I :‘;'3= Jafi‘ii .
1:21.1' I I I... «.1 . J1:
JIJJJ E3J1. .JJ'JJ I. J

J. I J':.J :'.'3. .1-

ml -9-

THESiS

,/

SITY LIBRARIE

||||Hllllll|lllllllllllllllllllllllllllllll || Ill“!

3 1293 015815

LIBRARY
Michigan State
University

 

 

 

 

 

 

 

 

 

 

 

This is to certify that the
dissertation entitled

THE IMPACT OF INCENTIVE PLANS 0N PRODUCTIVITY,
WORKER QUALITY, AND THE EXTENT OF SUPERVISION

presented by

Daniel G. Hansen

has been accepted towards fulfillment
of the requirements for

Ph.D; degree in Economics.

06% I

v Major professor

 

Date April 30. 1997

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

PLACE IN RETURN BOX to remove thie checkout from your record.
To AVOID FINES return on or before date due.

DATE DUE DATE DUE DATE DUE

       

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU ie An Afflnnetive Action/Equal Opportunity inetitmion
Wm:

THE IMPACT OF INCENTIVE PLANS ON PRODUCTIVITY, WORKER QUALITY,
AND THE EXTENT OF SUPERVISION

By

Daniel G. Hansen

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Economics

1997

ABSTRACT

THE HVIPACT OF INCENTIVE PLANS ON PRODUCTIVITY, WORKER QUALITY,
AND THE EXTENT OF SUPERVISION

By

Daniel G. Hansen

Incentive plans such as profit sharing plans or gain sharing plans have become an
important part of the economy. This dissertation explores three distinct aspects of them.

The first chapter uses monthly, individual-level data collected from two units
within a large US. financial corporation to estimate the individual response to the
introduction of a group incentive plan. The findings indicate a large and statistically
significant response to the plan. A closer look at the data reveals that the incentive plan
caused performance to converge to a standard. The initially least productive workers
improved greatly, while the workers who were initially performing well did not change
their behavior. However, the incentive plan was successful in increasing the average
level of productivity.

The second chapter examines whether higher-quality workers are attracted to (or
retained by) profit sharing firms. Studies of the effect of profit sharing on productivity
do not account for changes in the composition of the work force. Therefore, if turnover
leads to a more productive labor pool for a firm, one cannot necessarily make the claim
that the current studies demonstrate that profit sharing plans increase the performance of
individual workers. Some findings from this study, indicating that workers entering

profit sharing firms are “better” than those entering non-profit sharing firms, are

consistent with this bias. However, attrition out of profit sharing firms is no different
than for non-profit sharing firms.

The final chapter explores whether compensation policies -- efficiency wages,
profit sharing plans or performance-based bonuses -- act as substitutes for supervision.
The results show that the shirking model of efficiency wages is not relevant for the
majority of workers. However, evidence consistent with this theory is found for a
sample of supervisors and for a sample composed of piece rate and commission workers.
In addition, workers are not more likely to be offered incentive plans as the extent of
supervision is reduced. However, supervisors are more likely to be offered performance-

based bonuses as the number of workers supervised increases.

For my parents.

ACKNOWLEDGMENTS

This work could not have been completed without the help of many people.
David Neumark has been an invaluable asset. As an advisor, I appreciated the quality
and speed of his feedback on my work, and that he helped to generate the motivation that
I sometimes lacked. In addition, working for him taught me more about research
methodology than any class could have. Many of the techniques that I used in this work
came directly from the work that I have done for him. It is truly difficult to imagine a
better advisor.

Several other faculty members have helped along the way. Harry Holzer
provided excellent insight and an amazing knowledge of other literature that related to
this work. Robert LaLonde helped me to look at my work with a healthy skepticism and
more critical eye, forcing me to think about what it all meant. Finally, Daniel
Hamermesh deserves a word of thanks for seeing me through the very beginning of this
work, a time when patience for stupid questions is most needed.

I certainly could not have made it through the program without my fellow
graduate students. Complaining has never been as much fun as it has been here. There
were too many people to list everyone, but I would like to thank two of them specifically.

Paul, for worrying about the details that drove me crazy, and Jess, for having the good

sense to work in the same area that I do so he could summarize articles for me that I
never knew existed.

Finally, I would like to thank my parents, Glenn and Connie, for supporting me
through this time. They provided encouragement and support (and care packages) when

I needed it.

vi

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................ viii

LIST OF FIGURES ............................................................................................... x

CHAPTER 1

INDIVIDUAL RESPONSES TO A GROUP INCENTIVE ................................... 1
Introduction ................................................................................................ 1
The Data and Setting .................................................................................. 3
The Model and Empirical Results .............................................................. 7
Comparison to Other Studies ..................................................................... 15
Summary and Conclusion .......................................................................... 17
References ................................................................................................. 25

CHAPTER 2

WORKER QUALITY AND PROFIT SHARING: DOES UNOBSERVED
QUALITY BIAS FIRM-LEVEL ESTIMATES OF THE PRODUCTIVITY

EFFECTS OF PROFIT SHARING? ..................................................................... 26
Introduction ............................................................................................... 26
Theory ....................................................................................................... 29
Data and Methodology ............................................................................... 31
Results for Job Changers Leaving Non-Profit Sharing Jobs ........................ 35
Results for Workers in Firms that Adopt Profit Sharing ............................. 40
Results Relaxing the Sample Restrictions ................................................... 44
Conclusions ................................................................................................ 46
References .................................................................................................. 55

CHAPTER 3

SUPERVISION, EFFICIENCY WAGES AND INCENTIVE PLANS:

HOW ARE MONITORING PROBLEMS SOLVED? ........................................... 56
Introduction ............................................................................................... 56
Data ........................................................................................................... 6O
Re-Examining Evidence on the Shirking Model of Efficiency
Wages for Non-Supervisory Workers ......................................................... 63
Evidence on the Shirking Model of Efficiency Wages for
Supervisory Workers .................................................................................. 67
Evidence on the Use of Incentives .............................................................. 70
Conclusions and Discussion ........................................................................ 77

vii

References ............................................................................................. 92
Appendices

Appendix A ................................................................................ 94
Appendix B ................................................................................ 95
Appendix C ................................................................................ 96

viii

CHAPTER 1

Table 1:

Table 2:

Table 3:

Table 4:

CHAPTER 2

Table 1:

Table 2:

Table 3:

Table 4:

Table 5:

Table 6:

CHAPTER 3

Table 1:

Table 2:

LIST OF TABLES

Means of the Data by Year ............................................................ 19
Fixed Effects Estimates of the Effect of the Incentive Plan ........... 20
Fixed Effects Estimates Testing for Sample Attrition Bias ............ 21
Fixed Effects Estimates of the Separate Samples ........................... 22
Means of the Data for Job Changers .............................................. 49
Estimated Differences Between Workers Entering Profit

Sharing Firms and Non-Profit Sharing Firms by Worker
Characteristic ................................................................................ 50

Means of the Data for Workers Already in Profit Sharing
Firms ............................................................................................ 5]

Estimates of the Differences in Characteristics for Workers
Already in Profit Sharing Firms .................................................... 52

Means of the Data for Workers in Firms Adopting Profit
Sharing ......................................................................................... 53

Estimates of the Differences in Characteristics for Workers
in Firms Adopting Profit Sharing .................................................. 54

Summary of Previous Studies on the Shirking Model of
Efficiency Wages .......................................................................... 81

Descriptive Statistics ..................................................................... 82

ix

Table 3:

Table 4:

Table 5:

Table 6:

Table 7:

Table 8:

OLS Estimates of the Effect of Supervision and Firm Size
on Wages for Workers (Not Piece Rate or Commission) ............... 83

OLS Estimates of the Effect of Supervision and Firm Size
on Wages for Piece Rate and Commission Workers ...................... 84

Estimates of the Effect of Supervision and Firm Size on
Wages for Supervisors, Broadly Defined ...................................... 85

Estimates of the Effect of Supervision and Firm Size on
Wages for Supervisors Responsible for Pay or Promotion ............ 86

Estimates of the Effect of Supervision and Firm Size on
Wages for Supervisors Defined by Occupation Codes .................. 87

Logit Estimates of the Relationship Between the
Supervisor-to-Staff Ratio and the Presence of
Performance-Based Bonuses and Profit Sharing Plans .................. 88

LIST OF FIGURES

CHAPTER 1
Figure 1: Average Minutes Per Call by Month .............................................. 23
Figure 2: The Change in Individual’s Minutes Per Call ................................. 24
CHAPTER 3
Figure 1: Inter-Industry Wage Differences and Supervision for the

Sample of Workers ........................................................................ 89
Figure 2: Inter-Industry Wage Differences and Supervision for the

Piece Rate and Commission Workers ............................................ 90
Figure 3: Inter-Industry Wage Differences and Supervision for the

Sample of Supervisors Identified by Occupation Codes ................. 91

xi

Chapter 1
INDIVIDUAL RESPONSES TO A GROUP INCENTIVE

Introduction

It is well known that firms have difficulty in getting the desired level of
productivity out of their workers. Piece rate wages and individual incentive plans can
lead to higher individual effort, but present problems of their own. First, cooperation is
not encouraged in these systems. For example, a bonus that pays only to the worker with
the highest sales volume puts the members of the sales force in competition with one
another, providing an incentive to sabotage a co-worker's efforts. Second, it may not be
possible to implement an individual incentive because the output of individual workers is
often difficult to measure. For example, a team of workers may be responsible for
developing a software application. In the end, the application is produced, but it would
be difficult to determine what was contributed by any one worker.

An alternative is to offer incentives based on group productivity in the form of
either profit sharing or gain sharing plans. A profit sharing plan simply shares a fraction
of the profits the company earns with the workers, usually on an annual basis. Gain
sharing plans are based on more specific goals, with workers sharing in the financial gain
that comes from achieving these goals. Several variants of gain sharing have been used,
some with very specific compensation formulas.1

Ideally, gain sharing plans are linked to quantifiable goals (such as fewer

programming errors in the example above), but the workers share equally in the gain.

 

I For example, in the Rucker Plan, the worker's share is calculated by finding value added, computing
from this the allowed labor costs (by multiplying value added by an historical percentage), and
subtracting from this the actual labor cost. The remainder is shared between the workers and the
managers.

2

Unlike with individual incentives, the workers are not pitted against one another, so any
potential gains from cooperation should be realized. However, group incentives are
plagued by the free rider problem. Under individual incentives, if there are N workers in
the firm, and one employee improved her productivity by $100, this would (could) lead
to a $100 reward. But if the gains are shared with the work group, this $100 gain in
productivity only leads to a $lOO/N gain to the worker. As the size of the group grows,
the individual incentive to expend more effort decreases. Furthermore, if it is assumed
that the marginal cost of effort increases with effort and workers have identical cost-of-
effort functions, the hardest workers are the least likely to respond to the incentive plan.
That is, the workers who are expending the highest levels of effort prior to the
implementation of the incentive plan need a larger reward in order to elicit increases in
their effort.

The majority of the previous studies on the effects of these incentive plans have
used firm-level data, typically using value added as the dependent variable. Weitzman
and Kruse (1990) provide a thorough survey of the literature.2 They conclude that the
"mean estimated effect of profit sharing on productivity for 'average' amounts of profit
sharing is 7.4 percent, with a median estimate of 4.4 percent.”

In contrast, this paper examines individual responses to the introduction of a gain
sharing plan. The data come from two units within American Express Financial
Advisors, a subsidiary of the American Express Corporation. The only comparable study
appears to be Weiss (1987).4 He studies three plants within a large US. electronics
manufacturer. At these plants, newly hired workers face an individual incentive, but

once they have reached a performance threshold--usually occurring after three months--

 

2 Weitzman and Kruse consider profit sharing and gain sharing plans to be fundamentally the same, and
therefore examine them together.

3 Weitzman and Kruse (1990), p. 138.

4 A recent paper by Shearer (1996) uses individual-level data from a firm using a group incentive plan.
However, the goal of that paper is to estimate a tenure-productivity profile controlling for the incentive
effects, and not to directly estimate the impact of the group incentive.

3

the workers' pay is proportional to the output of their work group.5 Weiss then estimates
the change in individual output when the workers change incentive schemes. He finds
that "Almost all of the workers whose first month performance was above the median
decreased their output between months 1 and 4, ...[but] only half of the low performing
workers decreased their output. "6 In addition, "Among workers whose first month output
was more than 10% below the median, 79% increased their output. Out of the 208
workers whose first month output was more than 10% above the median only one worker
increased his output. "7 In other words, the group incentive failed to motivate the
majority of the workers, and workers' performance converged to a standard. Weiss
believes that the most able workers are held back both by "trivial financial incentives"
and pressure from other workers.8 The conclusion generally reached from his study is
that group incentives can fail where work groups are large.9

The findings in this study indicate that, as in Weiss (1987), the workers who are
initially the most productive are not motivated, and that performance converges to a
standard. However, the incentive plan was effective at improving the mean level of

productivity and therefore should be considered a success.10

The Data and Setting

The two units to be examined comprise the service representatives for the
discount brokerage area of American Express Financial Advisors. The units perform
identical tasks, and are physically adjacent. One unit is responsible for clients in the

Western half of the United States, while the other covers the remainder of the country.

 

5 The work groups studied in Weiss (1987) average 126 members in size. As shown in Table 1. the
groups studied in this paper are substantially smaller, averaging about 10 workers per group.

Weiss (1987), p.140.
7 Weiss (1987), p.140.
8 Weiss (1987), p. 143.
9 Brown (1990), p. 174-5.
'0 It cannot be determined if the incentive plan was, strictly speaking, a financial success for the firm.
While the payouts to workers can be calculated, it is not possible to place a monetary value on the change
in worker behavior that occurred.

4

Each unit has its own supervisor, who reports to still another manager who oversees both
units. The units are rated jointly to determine the size of their bonus. Although they
could operate in isolation, in practice there is a great deal of cooperation between the
units, so a joint rating is not inappropriate.

The workers are primarily responsible for data entry, bookkeeping and receiving
calls from clients.“ There are two distinct phone lines to which a worker may be
assigned: the service line and the trading line. New transactions (the actual buying and
selling of securities) take place on the trading line. As this is a discount brokerage firm,
it is not the worker’s job to give advice or promote sales. Because of this, conversations
on the trading line follow a standard process of information gathering (finding out how
many shares to trade, etc.). The service line handles a wider variety of calls, ranging
from simple questions about the status of a trade to a complex solution to a trade error.
All of the workers are qualified to work on the service line, but only some of them have
obtained the licensing required to work on the trading line.

This study examines only performance on the service line. This is limiting in that
it cannot be determined whether any measured response in this area comes at the expense
of the other tasks.12 As is noted below, the specific goals of the incentive plan focus on
improving performance on this line, and in fact constitute the only quantifiable goal.
Therefore, an analysis of the impact of the plan using telephone performance data should
be sufficient for determining whether or not the incentive plan succeeded.

The telephone performance data do not depend on individual reporting. In order
to take calls off of the queue, workers log onto a system, after which the number of calls
answered, the amount of time spent on calls, the amount of time available to take calls

(i.e. logged onto the system), and the number of calls transferred are all recorded. The

 

“ The information regarding details of the incentive plan, work conditions and responsibilities of the
workers was obtained while I was employed with this firm. I was a temporary employee, however, and
therefore not involved in the incentive program.

‘2 One advantage Weiss (1987) may have relative to this study is that the performance measure he uses
may be a better measure of a worker's total output on the job.

numbers are totaled for each worker on a weekly basis and sent to the supervisor. Only
monthly data were available for use here. The average minutes spent per call in a month
(abbreviated "minutes per call" hereafter) is used as a proxy for productivity. Workers
who take fewer minutes per call can handle a higher volume of calls, and are therefore
considered to be more productive. There are some potential problems with this
interpretation, however, which will become apparent below.

The distribution of incoming calls across workers is random, so in principle the
variety among types of calls should not affect the average minutes per call across
workers. However, a worker can choose to transfer a problem to someone who is
considered more able. Although any call that a worker receives from the queue is
included in the data, a call received via transfer from another worker is not recorded.
Since transferred calls are likely to be short, there may be a bias in minutes per call,
associating less able workers with lower minutes per call. In the empirical analysis, this
is taken into account by adding the percentage of calls that a worker transfers to other
workers as a control variable.

Workers differ substantially in their methods. Some prefer to call back clients
instead of leaving them on hold, some are more thorough (and therefore lengthy) in their
descriptions, etc. These differences make it difficult for supervisors to judge relative
performance in this area. If a worker is taking more time per call it could mean that
better service is being given, fewer problem calls are being transferred to other workers,
or that he or she simply likes to chat with clients.

Given these problems with the data, one must be careful in interpreting what the
variable minutes per call can say about worker performance. One cannot necessarily
compare two workers in one month, observe that one's minutes per call is higher, and
conclude that he or she is less productive. It is true that fewer calls can be answered, but

if more complex problems are dealt with, it could be that this worker is more valuable.

6

Although the level of minutes per call may not be as informative as one would
like, the change in minutes per call for a given worker when the incentive plan is
implemented does indicate whether effort was exerted to fulfill the established goals.
That is, the primary goal of the plan was to reduce the amount of time that a client must
hold before speaking to a representative. From an individual standpoint, the only way to
achieve this is to spend less time on each call, thereby getting to the calls in the queue
faster. Therefore, if workers take less time per call after the implementation of the
incentive, they have responded positively to the incentive.

The details of the incentive plan are as follows. It was conceived in periodic
meetings between managers, supervisors and workers in late 1991, and the performance
was to be evaluated during the following year. The meetings were used to discuss the
most pressing needs (or shortcomings) of the department, and the best means of
addressing them. Following this, the specific goals were agreed upon, put in print, and
distributed to all of the concerned parties.

The amount of the bonus was based on a unit rating given by upper management,
and all members of the unit, including supervisors, were eligible for the bonus. Workers
hired during 1992 were eligible for a pro-rated bonus. The awards were to be paid in
early 1993, with the highest rating paying each worker $1,150, the second highest $800
and the third highest $600.13 There were two lower ratings that did not merit bonus pay.
In addition, all of the workers were had to meet a minimum individual performance
rating (given annually by the supervisors) in order to be eligible for the awards. The
requirements for meeting this minimum were not likely to have been stringent enough to

preclude the possibility of free riding, however, as the lowest ratings were the equivalent

 

'3 Since the reward schedule is specified in advance, the nature of the free rider problem in this firm is
somewhat different than that described in the introduction. That is, any gains in individual productivity
are not directly shared with the other workers, but contribute toward the team goal. Free riding is still an
issue, however, as a worker who contributed little to an otherwise successful plan would receive the same
reward as the other workers.

7

of being placed on probation and were rarely given. All of the workers in the sample
were given performance ratings the exceeded the minimum.

As stated above, the plan focused on reducing the amount of time callers had to
wait to receive service. This is measured at the unit level by the "ASA", or average
speed of answer. 14 In June 1991, the ASA was about 79, meaning that on average callers
waited 79 seconds before speaking to anyone. The targets for the plan were as follows: a

reduction to 35 for the highest rating, 55 for the second and 65 for the third.15

The Model and Empirical Results

The dependent variable in all of the following specifications is the monthly
average minutes spent per call for each worker. The independent variables control for
changes in the workplace conditions such as supervisory policy and demand conditions.
Data are not available to determine if a change in supervisors occurred in the sample
period, but information on supervisory policies do exist. That is, there are two actions
that a supervisor may take to reduce a client’s time on hold: hire more workers, or
schedule more telephone time for the existing workers. In fact, as Table 1 shows, the
number of workers did increase over the sample period, from about ten workers per unit
in 1991 to thirteen workers per unit in 1992.16

A change in the allocation of a worker’s time is indirectly measured by the
percentage of a worker’s time spent logged on the phone (i.e. available to take a call), but
not speaking to a client (abbreviated PERCTIME). Telephone time was scheduled for

each day, and supervisors may have increased the amount of time scheduled in order to

 

‘4 Because the ASA is only available at the unit level, but the analysis is conducted at the individual
level, minutes per call is used as the proxy for productivity.
‘5 Other, more qualitative goals were mentioned. As an example, "Actively develop a pro-active vs. re-
active listening approach. " Given the difficulty in measuring any progress towards such a goal, it is easy
to see why managers would focus on changes in the ASA as the definition of success or failure of the
lan.
6 This count includes all of the workers in the units, including those not in the sample.

8

help meet the goal of reducing the ASA. PERCTIME increased from 1991 to 1992, a
finding one would expect to observe if more time was allocated to the telephones.

The monthly volume of calls received by each unit controls for changes in
behavior due solely to fluctuations in the amount of business received. Table 1 shows
that the volume of business was, on average, increasing over the sample period,
presenting the possibility that workers may have hurried through calls simply because
more calls were coming into the queue.

An examination of the workers' characteristics in Table 1 reveals that there were
slightly more women than men, the vast majority were college educated and workers
averaged about seven years of pre-company experience.” The average nominal wage
was between $12 and $13 per hour.

The data are monthly, from June 1991 through November 1992. Although a total
of forty-two people were employed in the two units during this time period, only twenty-
one of them are in the sample both before and after the implementation of the incentive
plan.18 The average number of workers per unit in Table 1 refers to the actual number of
workers per unit. This variable, and not the number of workers in the sample, is used in
the analysis because it is the actual number of workers that could affect the behavior of
the workers. The average number of workers in the sample is also shown in Table 1,
revealing that on average there are fewer workers in the sample in 1992 than in 1991.

The basic form of the model that is estimated is:

ln(minutes per call),t= 2, or, I, + B1(incentive plan)t + [32(workplace controls),t + 8,,

 

17 Pre-company experience was estimated by age-tenure-education-6, where a high school degree was
equivalent to twelve years, an associate‘s fourteen and a bachelor's sixteen.

‘8 Complete data on the entire work force were not made available, preventing a study including all of
the workers.

9

where i indexes workers, t indexes months and I, are dummy variables for individuals.
The model is run both with and without the workplace controls. The "incentive plan"
variable is a dummy variable equaling zero for months in 1991 and one for months in
1992. This specification of the incentive plan may be considered a shortcoming of this
study because other changes affecting the productivity of workers may have occurred
during this time. Under these circumstances, the dummy for the incentive plan may be
picking up these other effects.

Several measures have been taken in order to minimize the possibility that the
results are contaminated in this way. First, the data do not begin until June of 1991
because of a reorganization. It was at this time that the workers split into two units
handling accounts from different regions of the country. No other reorganization
occurred during the sample period. Second, the analysis controls for the monthly volume
of calls to account for any demand fluctuations. Third, variables controlling for
supervisory behavior are included. Specifically, variables are included to reflect the fact
that more workers were hired and that the time within the work day was reallocated. 19

The first set of regression results is reported in Table 2. Fixed effects estimation
is used throughout. An F -test of the joint significance of the individual dummies
produces an F-statistic of 11.22, rejecting the null hypothesis that the pooled model is
efficient.

Column 1 shows the effect of the incentive plan with no control variables
included. On average, minutes spent per call went down by about thirteen percent in

1992. Column 2 adds the workplace controls, increasing the estimated effect of the

 

'9 The supervisors could also increase the level of their monitoring, but this cannot be measured. A
possible problem with the analysis is that the supervisory policies that can be included (hiring and
reallocating time) may lead to higher minutes per call for existing workers, while the excluded action
(increased monitoring) should reduce minutes per call. This may bias the results toward finding a large
independent response from workers to the incentive plan. The entire analysis has been done excluding all
supervisory behavior (dropping PERCTIME and number of workers in the unit from the regressions).

The results from these regressions show a stronger estimated response in most specifications, with
decreases in the estimates (in magnitude and significance) occurring only when a time trend is included.

lO

incentive plan. The estimates of the workplace controls are generally highly significant,
and consistent across models.

A potential problem arises from the fact that the panel of workers is unbalanced,
raising the possibility that attrition from the panel may be affecting the results. That is,
one may falsely observe that the incentive plan was successful if less productive workers
are in the sample prior to the beginning of the plan, but leave soon after implementation.
To test for this, the sample is divided into two groups: those who performed better than
and worse than the median level in 1991.20 A worker’s presence in the panel is
represented by two variables, the number of months present in the data in 1991
(“LENGTH91”) and the number of months present in the data in 1992
(“LENGTH92”).21 A test for the equivalence of means between the two groups is done
for these variables. In both cases, the test does not reject the hypothesis that the means
for the two groups are equal.22 Since neither group is greatly over-represented either
before or after the incentive plan is introduced, one cannot make the claim that an
improvement may be observed only because inferior workers were weeded out of the
sample.

A second test for sample attrition bias is conducted by observing how the results
change as the sample period is shortened. The results (using the specification in column
2 of Table 2) are contained in Table 3. Notice that, regardless of when the sample is cut
off, the qualitative finding of a strong response to the incentive plan remains. In
addition, the way the estimated effect of the incentive plan evolves in relation to the
sample size shows that sample attrition bias is not driving the results. In February 1992,

the sample size is at its largest level, containing all twenty-one workers. If the data are

 

20 This is done based on the fact that the estimated response is driven primarily by the workers who
performed worse than the median in 1991.

I There are 18 months of data, beginning in June of 1991. So, for example, a worker who was in the
data from November of 1991 through March of 1992 would have LENGTH91 = 2, LENGTH92 = 3.
22 The means are as follows (workers better than median, workers worse than median): LENGTH91
(59,63) and LENGTH92 (7.5,7.7).

11

cut off at this point, the estimated effect of the incentive plan is about an eleven percent
improvement. Five workers leave the sample from February to March, the largest drop
between any two months. Yet, the estimated effect of the incentive plan using data
through March is almost identical to the February estimate. Furthermore, no additional
workers are lost through May of 1992, yet the estimated effect of the incentive plan
increases to about fifteen percent. Given these facts, it does not appear that the estimated
improvements in minutes per call are due to attrition from the sample.

Columns 4 and 5 of Table 2 refine the previous estimates by correcting for serial
correlation and including a time trend. Quasi first-differencing is performed in order to
correct for serial correlation using a separate value of p for each worker.23 As shown in
column 3, the finding that the incentive plan was a success is robust to the correction for
serial correlation. Furthermore, while the addition of a linear time trend reduces both the
magnitude and significance of the estimated effect of the incentive plan, the estimate is
still sizable and significant at the ten percent level.

It seems that the finding of a significant response to the incentive plan is fairly
robust. Given that the variable representing the incentive plan is merely an indicator
equaling one for months in 1992, this result shows that the average minutes per call
dropped significantly across years, but gives no indication about the immediacy of the
impact. Figure 1 graphs the pooled average minutes per call across months. This shows
clearly that it took several months for a large effect to occur. There is no excluded
variable known to the author that would cause a large change from March through

May.24 The volume of business was increasing, as was the number of employees, but

 

2" The quasi first-differencing is performed as follows: p is estimated as the coefficient on the residuals
from month t-l regressed on the residuals from month t (separately for each worker). The data are then
transformed by: xt - p(x,_,). The first observation for each individual and any observations following gaps
in the data series within individuals are dropped in this process.

24 Changing the specification of the incentive plan dummy variable to begin one to two months before
and after the actual implementation produces the following coefficients (using the specification from
column 2 of Table 2): November 1991: -.135, December 1991: -.l39, January 1992 (actual start date): -
.184, February 1992: -.158, March 1992: -. 160. Note that the largest estimated effect of the incentive

12

this has been controlled for in the regressions. One possible explanation is that free
riding occurred in the time shortly after the implementation of the incentive plan. It may
be rational for the workers to attempt to free ride early on, as an immediate response is
not required for the incentive plan to be considered a success. This could be taken as
evidence that the free rider problem may be overcome in a repeated game setting, as was
theorized in Weitzman and Kruse (1990).

The next issue to explore is whether the workers who were initially more
productive responded as strongly to the incentive plan. To do this, the average minutes
per call is calculated for each worker and for the total unit for 1991. The sample is then
divided into those who started better and worse than the median level of minutes per
call.25 Table 4 shows the estimates for these two groups. The first two columns can be
compared to column 2 of Table 2, while the second two columns employ the
specification used in column 4 of Table 2. These estimates show that the initially slow
workers responded strongly to the incentive plan, but the initially fast workers showed no
response at all.

Figure 2 shows the change in the productivity of individual workers graphically.
The horizontal axis measures the average minutes per call for each worker in 1991, while
the vertical axis measures the change in minutes per call from 1991 to 1992, calculated
so that a positive number indicates an improvement (reduction in minutes per call). The
vertical line is drawn at the 1991 median level of minutes per call. Of the ten workers
who were better than the median in 1991 (to the left of the vertical line), six showed a
decline in performance while only three improved.26 However, all of the ten workers

who performed worse than the median in 1991 improved their performance.

 

plan occurs at the actual start date, as one would expect if the estimates are picking up the effect of the
incentive plan, and not some other event.

25 The worker at the median is dropped from the sample.

26 Only one of these differences is significantly different from 0, however.

13

There are several possible explanations for the fact that the initially faster workers
did not improve in 1992. First, this finding could support Weiss’ (1987) result that group
incentives do not motivate the "best" employees. Second, it could be that there is a
minimum amount of time that is required to provide service, so that those who were
already performing well had very little room to improve.

Third, given that the calls are initially randomly distributed, it is possible that this
effect merely reflects a regression to the mean. That is, the workers who were slow in
the early months may have been as productive as other workers, but were “unlucky” in
that they happened to draw long calls of the queue. The following test is done in order to
see if this may be affecting the results. A dummy variable is created for those who were
slower than the median worker from July through September of 1991. A regression is
then run using the previous specification, but including the newly created dummy, for the
months of October through December of 1991. The use of only 1991 data ensures that
the incentive plan is not affecting the test. This regression produces a coefficient of .529
(with a standard error of .093) on the dummy variable for the initially slow workers,
showing that workers who began the sample by taking more time on calls continued to be
slow in the following months. Given this, the results do not seem to be affected by
regression to the mean.

Finally, an Akerlof-type gift exchange model might explain why the better
workers seemed to ignore the incentive plan. If they felt that the slower workers were
made to look bad by the disparity in performance, they may have attempted to lessen the
difference by waiting for the slower workers to improve before attempting such an
improvement themselves (Akerlof, 1986). According to Akerlof, this represents a gift
exchange between workers motivated by the fact that, "in working together, workers
acquire sentiment for each other. "2’ The situation here is somewhat different than

Akerlof describes, however. The workers that he describes are performing in excess of

 

27 Akerlof (1986), p. 74.

14

the established standard as a gift to the firm. In exchange, the firm must not raise the
standard, ensuring that the marginal workers meet the required performance standard.

In contrast, the "better" workers studied here are theorized to be refraining from
raising the standard, or helping the less efficient workers by handling the more difficult
calls (via transfer, which would not be recorded in the data). In exchange, the "worse"
workers are responsible for increasing their level of performance enough to meet the
goals of the incentive plan. The firm receives the gift of faster service, and in exchange
must pay all of the workers for any improvements. Frequent feedback regarding the
progress made toward lowering the ASA (recall that telephone performance data were
available each week) makes this exchange among workers possible. If, after a few
months, the goal appeared to be unattainable under the arrangement, the faster workers
could simply speed up and achieve the goal. Therefore, the gift that the faster workers
are attempting to give may not jeopardize their chances for the bonus pay.

The final issue to be addressed here is whether the observed change in minutes
per call was purely cosmetic, or a real improvement on the part of the workers. In
principle, minutes per call could easily be manipulated by answering the phone, taking a
message, and then returning the call. Only the short initial call would be recorded as
data, while the total time spent with the client may have increased. However, it is
unlikely that this occurred, since the managers are not directly concerned with a change
in minutes per call, but in the average speed of answer (ASA), which is not as easy for a
worker to manipulate without a real improvement.

There is one way that ASA can be manipulated, but managers were not only
aware of this possibility, but actually tried to use it to their advantage the previous year.
As one might expect, this job faces a peak-load problem. During peaks, wait times for
callers may increase dramatically. In the Spring of 1991, the “release valve” was created
in an attempt to solve this problem. One person would log onto the telephone queue and

only take brief messages for the calls answered. The messages were then distributed to

15

the workers so that the calls could be returned when business slowed. Note that the use
of this method may result in a reduction of the ASA, even though the clients do not
receive service any sooner (nor, possibly, any more efficiently). The release valve was
considered to have benefits in spite of this because it alleviated the frustration clients felt
from remaining on hold for an extended period of time.

Although the theory behind the release valve was sound, it ultimately failed
because business rarely slowed enough to allow for time to return the calls. Because of
this, the method was abandoned within months of its inception. Its existence, however,
signals that the managers knew how the ASA could be manipulated. The fact that, with
this knowledge, they chose to award bonuses suggests that the incentive plan caused a

real improvement in service.

Comparison to Other Studies

Given that this paper finds that a group incentive can increase the average level of
productivity but Weiss (1987) does not, it is instructive to compare the two environments
that are studied. First, the relative size of the work groups studied are quite different. A
total of 42 employees worked in the two units studied here over the sample period, versus
an average work group of 126 in Weiss' paper. Given that the free rider problem should
worsen as work groups increase in size, this may be an important reason for the
difference in findings.

There is, however, some evidence suggesting that size may not be negatively
correlated with the success of an incentive plan. For example, Schuster (1984) uses
longitudinal plant-level data to examine the effect of a Scanlon Plan. This variant of gain
sharing uses committees of workers to evaluate suggestions from other employees, and a
bonus formula to share gains in productivity among the workers. In spite of the fact that

there were 890 non-supervisory production employees in a plant,28 Schuster finds that the

 

28 The size of the work groups is not reported.

16

plan had a significant impact on productivity, and gave much of the credit for its success
to employee participation in decision making. In addition, in his survey of the literature
Schuster concludes that "firm size correlated positively with rated Plan success and Plan
retention".29 This may be evidence that the specific form of the incentive plan, and not
the size of the work groups, is what led to the difference in the measured success between
this study and Weiss (1987).

Second, the form of the incentive plan examined here differs from the one in
Weiss (1987) in several ways. In the plan studied here, specific goals were set, so that
the success or failure of the incentive plan could be easily determined. In Weiss (1987),
the group incentive is not a goal-based incentive, but an everyday compensation formula,
in which an individual is placed in a group, and each worker receives pay in proportion
to the output of his or her group. In addition, the firm studied in this paper used
committees consisting of workers and managers to set the goals. These could be
important differences, and form a common theme in the industrial relations literature on
group incentives.

Strauss (1990) lists some of the reasons that goal-setting and employee
participation may raise productivity. They include: "Participation may result in better
decisions..."; "Participation may improve communication and cooperation..."; and
"Participation frequently results in the setting of goals. There is considerable evidence

that goal setting is an effective motivational technique.”30

Schuster (1984) provides an
example of the empirical support for these assertions, in that the use of these methods

resulted in an increase in productivity.

 

3" Schuster (1984), p. 25.
’0 Strauss (1990), p. 5.

17

Summary and Conclusion

This paper uses individual level data from the discount brokerage area of a large
firm to examine the impact of a group incentive plan. It shows that there was a
significant unit level response, about a seventeen percent improvement from the pre-plan
average. Given the imperfection of the proxy for productivity and the fact that only one
dimension of performance is examined, there is some question as to whether this can
strictly be thought of as an increase in productivity, but there can be little doubt that at
least a change in behavior occurred.

Although the group incentive did increase the average performance for the units,
those workers who were initially the best may not have been motivated by the plan.
Some of these workers did not merely show less improvement than those who were
initially worse, but actually declined in performance after the introduction of the
incentive. Here too, the limitation of the dependent variable makes the conclusion
tenuous. That is, if those who were initially taking very little time per call were
subsequently asked to handle more time consuming problems or provide better service,
they may have improved in a way that the data cannot measure. Whatever this
researcher's judgment is regarding the success or failure of the incentive, the managers
who evaluated the plan deemed it successful enough to merit the second highest rating,
giving each worker $800.31

The finding that performance converges to a standard was also observed in Weiss
(1987). Given the many differences between the environments, this is a surprising
similarity, and may indicate a pattern in workers’ responses to group incentives.

However, Weiss (1987) can also be interpreted as showing that group incentives

may be ineffective when work groups are large, whereas this study found an

 

3’ You may recall that, in order to be given this rating, the incentive plan called for a reduction in the
ASA from 79 seconds to 55, a 41 percent reduction. It may seem that the 17 percent reduction observed
in individual minutes per call could not account for this, but the two statistics are not comparable in this
way. Remember that ASA is a unit-level measure, and would also be affected by such things as hiring
more workers and the allocation of more of each employee's day to the telephones.

18

improvement in mean performance. Evidence from other studies has been presented
suggesting that the design and implementation of the incentive plan, and not the
difference in the size of the work groups, led to this disparity.

This paper therefore presents the possibility that a well designed group incentive
can increase the mean level of productivity in a variety of settings. An important caveat
to this is that the initially most productive workers may be largely unaffected (and

possibly adversely affected) by the incentive.

Table 1
Means of the data by year

 

 

 

 

 

1991 1992
Minutes per call 4.86 4.22
Percent of worker’s time logged .22 .24
on the phone, but not on a call
Percentage of the unit’s volume .09 .05
taken by the worker
Percentage of calls transferred .26 .21
by the worker
Average number of workers 9.4 13
in the unit
Average number of workers 18.7 15.8
in the sample
(two units combined)
Average monthly volume of calls 5,411 6,277
received by the unit
Male .43 .43
High School .10 .10
Associate’s .10 . 10
Bachelor’s .80 .80
Tenure (months) 11.9 18.9
Pre—company experience (years) 7.2 7.2
Nominal wage (hourly) 12.41 12.92

 

l9

 

Table 2
Fixed Effects Estimates of the Effect of the Incentive Plan

 

 

 

 

 

 

(1) (2) (3) (4)
Incentive plan -.134*"‘ -.l84** -.174** -. 100*
(.031) (.043) (.043) (.061)
Serial Correlation No No Yes Yes
Correction
Trend --- --- --- -.019**
(.009)
Percent of worker's time --- -.620** -.547** -.467**
logged on the phone, but (.113) (.117) (.120)
not on a call
Percentage of the unit's --- -.979** -.925** -l.085**
volume taken by (.361) (.366) (.370)
the worker
Percentage of calls --- -.613** -.726** -.781**
transferred by the worker (.290) (.300) (.304)
Number of workers --- -.007 —.0004 .015
in the unit (.010) (.012) (.014)
Log(Volume of calls --- .115 .114 .038
received by the unit) (.083) (.077) (.085)
N 290 290 266 266

 

 

 

Notes: Standard errors are in parentheses. Log(minutes per call) is the dependent variable. Therefore,
lower numbers mean higher productivity. All models include a constant. The incentive plan variable is a
dummy variable equaling one for months in 1992. Observations are lost in the correction for serial
correlation because of breaks in the data series for three of the workers. ** denotes significance at the five
percent level. * denotes significance at the ten percent level.

20

Table 3
Fixed Effects Estimates Testing for Sample Attrition Bias

 

Sample Cut-off Date

 

 

 

 

 

Febru_ary 1992 March 1992 May 1992 August 1992 Novem_ber 1992
Incentive plan -.109* -. 106* -. 149** -.151** -.184**
(.059) (.057) (.054) (.048) (.043)
Number of workers 21 16 16 14 11
in the sample at the
cut-off date
N 169 185 216 260 290

 

 

 

 

 

 

Notes: Standard errors are in parentheses. Log(minutes per call) is the dependent variable. Therefore,
lower numbers mean higher productivity. All models include a constant. The incentive plan variable is a
dummy variable equaling one for months in 1992. The specification used is the same as the one used in
column 2 of Table 2. ** denotes significance at the five percent level. * denotes significance at the ten

percent level.

21

 

Table 4
Fixed Effects Estimates of the Separate Samples

 

No Serial Correlation Correction

Corrected for Serial Correlation

 

 

Initially better Initially worse Initially better Initially worse
workers workers workers workers
Incentive plan .007 -.339** -.014 -. 173*
(.051) (.065) (.067) (.095)
Trend --- --- .001 -.033**
(.01 1) (.014)
Percent of worker's time -.358** -.889** -.309** -.818**
logged on the phone, (.121) (.187) (.113) (.224)
but not on a call
Percentage of the unit's -.332 -2.060** -.090 -2.103**
volume taken by (.390) (.625) (.385) (.633)
the worker
Percentage of calls -.645* -.419 -.884** -.315
transferred by the worker (.339) (.431) (.345) (.488)
Number of workers -.009 -.019 .002 .015
in the unit (.012) (.016) (.016) (.023)
Log(Volume of calls .048 221* .031 .140
received by the unit) (.100) (.125) (.107) (.135)
N 134 140 122 129

 

 

 

 

 

 

Notes: Standard errors are ill parentheses. Log(minutes per call) is the dependent variable. Therefore,
lower numbers mean higher productivity. All models include a constant. The incentive plan variable is a
dummy variable equaling one for months in 1992. Observations are lost in the correction for serial
correlation because of breaks in the data series for three of the workers. ** denotes significance at the

five percent level. * denotes significance at the ten percent level.

22

Minutes per call

Figure 1
Average Minutes Per Call by Month

5.28 "

 

3.26 ‘

Jun 3991 Jan 3992
Month

 

Nov 1952

23

Change in minutes per call (1991 - 1992)

 

Figure 2
The Change in Individual’s Minutes Per Call

 

 

 

O O O O
O
o O
o O
l l l I I I l
2 3 4 7 8

Minutes per call in 1991

24

BIBLIOGRAPHY

BIBLIOGRAPHY

Akerlof, George A. 1986. “Labor Contracts as Partial Gift Exchange.” In George A.
Akerlof and Janet L. Yellen, eds, Efiiciency Wage Models of the Labor Market.
Cambridge: Cambridge University Press, pp. 66-92.

Brown, Charles. 1990. “Firms’ Choice of Method of Pay.” Industrial and Labor
Relations Review. 43, pp. 165S-828.

Greene, William H. 1990. Econometric Analysis. New York: Macmillan Publishing
Company, pp. 440-441.

Schuster, Michael. 1984. “The Scanlon Plan: A Longitudinal Analysis.” The Journal
of Applied Behavioral Science. 20, pp. 23-38.

Shearer, Bruce. 1996. “Piece-Rates, Principal-Agent Models, and Productivity Profiles:
Parametric and Semi-Parametric Evidence from Payroll Records.” The Journal
of Human Resources. 31, pp.275-303.

Strauss, George. 1990. “Participatory and Gain-Sharing Systems: History and Hope.”
In Myron I. Roomkin, ed., Profit Sharing and Gain Sharing. Metuchen, NJ:
IMLR Press/Rutgers University, pp. l~45.

Weiss, Andrew. 1987. "Incentives and Worker Behavior: Some Evidence." In Haig R.
Nalbantian, ed., Incentives, Cooperation, and Risk Sharing. Totowa, NJ:
Rowman and Littlefield, pp. 137-150.

Weitzman, Martin L. and Douglas L. Kruse. 1990. "Profit Sharing and Productivity."

In Alan S. Blinder, ed., Paying for Productivity. Washington, DC: The
Brookings Institution, pp. 95-140.

25

Chapter 2
WORKER QUALITY AND PROFIT SHARING:
DOES UNOBSERVED WORKER QUALITY BIAS FIRM-LEVEL ESTHVIATES
OF THE PRODUCTIVITY EFFECTS OF PROFIT SHARING?
Introduction
It has been suggested that profit sharing plans may be used to increase the

productivity of workers. While theoretical work on this issue is inconclusive (largely
because of the free-rider problem), the empirical evidence has generally been favorable
towards profit sharing plans. Kruse (1993) reviews 265 estimates of the effect of profit
sharing and finds that, "8.7 percent take on negative values, and nearly all of these are
within the range of sampling error [relative to zero], while 57.4 percent take on positive

"1 Weitzman and

values where random sampling error can be ruled out as an explanation.
Kruse (1990) conduct a meta-analysis of the (then existing) firm-level studies, and
conclude that the "median productivity difference associated with profit sharing was 4.4
percent."2 Supporters of profit sharing plans interpret this figure as reflecting the
incentive effects profit sharing has on worker behavior.

However, it is possible that some of the estimated productivity difference

between profit sharing and non-profit sharing firms is not due to individual workers

becoming more efficient, but instead is observed because the composition of the firms’

 

' Kruse (1993), p.55.
2 Weitzman and Kruse (1990), p.138-9.

26

27
labor has changed. This possibility is most easily considered in the context of cross-
sectional studies of firms, in which it is difficult to determine if observed differences
between firms are due to the effects of profit sharing plans, or worker selection into
profit sharing firms. For example, if firms with more productive workers tend to adopt
profit sharing plans, or if higher-quality workers sort into firms with profit sharing plans,
the estimated productivity premium for profit sharing firms in a cross-section includes
this difference, and does not entirely reflect an improvement in the performance of
workers.

Longitudinal studies of firms are less prone to this form of bias, but are not
immune. Because data are available both before and after the adoption of a profit
sharing plan, any characteristics of the firms that remain constant through the regime
change can be made irrelevant by the use of fixed effects estimation. So, for example, if
firms with more productive workers adopt profit sharing, and the same workers are
present in the firm both before and after the adoption of profit sharing, the within-firm
change in productivity captures the effect that the profit sharing plan has on the
productivity of the workers.

However, this result only occurs if the composition of the work force is identical
before and after the adoption of a profit sharing plan. This assumption is almost
certainly not true, but its violation is only relevant if the incoming workers change the
average quality of the firm’s work force (and, as in previous studies, no information is
available to explicitly account for this change). Therefore, this study does not explore

the average differences between workers in profit sharing and non-profit sharing firms,

28
but instead examines the effect that profit sharing plans have on the quality of workers
entering and leaving firms.

Worker selection into profit sharing firms could have policy implications, for if
the previous estimates of productivity changes are partially due to worker selection, then
mass adoption of profit sharing plans would not lead to as large of a gain for society as
the current estimates suggest.3 For example, in the extreme case in which worker
selection accounted for the entire estimated productivity difference, the adoption of profit
sharing plans causes high quality workers to change location (though the magnitude of
the sorting will dissipate as more firms adopt the plans), but the nationwide output per
worker remains constant.

A longitudinal firm-level study by Kruse (1993) provides some indirect evidence
on worker selection into profit sharing firms. If worker selection into profit sharing
biases the estimated changes in productivity, one would expect that firm-level
productivity would continue to rise following adoption, as the pre-adoption workers are
gradually replaced by more productive workers through turnover. However, Kruse
(1993) finds no significant trend effect for profit sharing firms.4 Even this may not solve
the problem entirely, as a linear time trend may not reflect the turnover pattern following

adoption. That is, if profit sharing brought about a large amount of turnover shortly after

 

3 That is, the productivity effect would be overstated. If, as Weitzman (1983) asserts, the national level
of employment is stabilized, there may be some benefit. This paper will not speak to this issue, but
Kruse (1995) uses the same data to address it.

4 Shepard (1994) also estimates the effect of profit sharing using longitudinal firm-level data. He
includes a time trend, but does not interact it with the profit sharing indicator. The trend that is estimated
is not significantly different from zero.

29

adoption and very little thereafter, a trend effect may not be observed, but worker
selection could be affecting the results.5

This paper directly examines the issue of worker selection into profit sharing
firms using recent data from the National Longitudinal Survey of Youth (NLSY).
Workers are compared using three characteristics: years of education, scores on the
Armed Forces Qualifications Test (AF QT) and wages. Both selection into and attrition
out of profit sharing firms are examined. The findings indicate that workers entering
profit sharing firms are “better” than those entering non-profit sharing firms, but attrition
out of profit sharing firms is no different than that of non-profit sharing firms. The
estimates indicate (under assumptions specified below) a .6 to .85 percent annual
improvement in firm-level productivity due to the fact that higher-quality workers enter
profit sharing firms. Given to the 4 percent increase in productivity found in Kruse
(1993), the majority of the estimated effect of profit sharing plans represents a pool of
workers becoming more efficient, rather than a shift in the composition of the pool

toward more productive workers.

Theory
Although theory provides no clear answer regarding worker selection into and out

of profit sharing plans, there are some reasons to believe that higher-quality workers may

 

5 Wilson and Peel (1991) find that quits are significantly lower in profit sharing firms than in non-profit
sharing firms. However, this does not preclude worker selection from affecting the productivity
estimates for these firms, as the level of quits is still positive (around 5 percent per year), and the period
just following adoption is not observed (because none of the firms in the panel adopt profit sharing during
the sample period).

30
prefer this form of compensation. For example, they may be attracted to profit sharing
firms if they believe that profit sharing more accurately rewards them for their
productivity. Alternatively, lower-quality workers may be attracted to profit sharing
firms if they believe that they can benefit from the skill and effort of their co-workers
through free riding.

Risk is another issue that the workers must face. Because compensation is tied to
firm performance, profit sharing may increase the variability of a worker’s income. If
workers are assumed to have concave and identical utility functions, those with higher
income should be more attracted (or less averse) to profit sharing plans. If it is further
assumed that high-quality workers are paid more than low-quality workers, profit sharing
may tend to attract (or retain) higher-quality workers.

Alternatively, workers may be attracted to a characteristic of firms that is
correlated with the presence of a profit sharing plan. This has the same impact on firm-
level productivity studies as the case in which workers are attracted to the compensation
system itself, but may not require very strong assumptions about the agents. As an
example, Kruse (1993) finds that a strong predictor of profit sharing adoption is an
increase in the profit margin in the previous year. If successful firms tend to adopt profit
sharing plans, these firms may be able to select the “best” workers from a queue.
Workers are attracted to the firm because of the possibility of sharing in the firm’s rent,
allowing them to earn wages in excess of their marginal product.

The theories presented above are meant only to present the possibility that better

workers may choose to enter profit sharing firms. This paper does not address which of

31
the above theories (or the many conceivable theories that are omitted) dominates a
worker’s decision about where to work. Rather, its task is to determine whether patterns
in the characteristics of workers moving into and out of profit sharing firms are

consistent with selection on the basis of worker quality.

Data and Methodology

This study uses data from the National Longitudinal Survey of Youth (NLSY)
from 1988 through 1993.6 The survey dates back to 1979, but beginning in 1988
respondents were asked if profit sharing was made available by their current or most
recent employer. Using this measure, three methods are employed to examine the issue
of worker selection and profit sharing. First, workers changing jobs and entering a profit
sharing firm are compared to those who change jobs and enter a non-profit sharing firm
(both groups having left jobs in non-profit sharing firms). This is an attempt to explore
whether workers who choose to enter (or simply tend to enter) profit sharing firms are
different from those who do not.7

Job changers are of interest because they enter the firm following the adoption of

a profit sharing plan. The alternative of including all of the workers who change profit

 

6 The percentage of US. firms making profit sharing payments was fairly constant over this period at
about 22 percent (Kruse 1993, p. 8).

7 Kruse (1995) uses the same data with the primary intent of examining the association of profit sharing
“with the disposition of a job over a five-year period, focusing on the risk of layoff.” A portion of that
work bears a surface similarity to this study: Kruse tracks job changers “to examine the change in pay
associated with joining or dropping out of profit sharing (p. 16-7).” This study is concerned with a
different measure: comparing the wages (and other characteristics) of workers when they report the same
profit sharing status (but later differ), thus providing a proxy for productivity by not including any
difference in wages due to profit sharing.

32

sharing status is not correct if the results are to have any bearing on a longitudinal firm-
level study because workers who remain in a firm as it adopts a profit sharing plan do not
affect fixed-effects estimates of the change in firm-level productivity. A second reason
for interest in job changers is that they are more likely to be revealing some preference
for the profit sharing environment. A worker who remains in a firm that subsequently
offers a profit sharing plan may not be particularly fond of their situation, but may face
mobility costs that exceed any disutility suffered due to the change in the compensation
system.

The second part of the analysis examines workers who leave firms that adopt a
profit sharing plan. To do this, the sample is reduced to include only those who held a
non-profit sharing job, then stayed in that job as the firm adopted profit sharing. The
workers who leave the firm in the following year are then compared to those who stay.
This approach has the advantage of controlling for firm selection into profit sharing.
Remember that for worker selection to have a bearing on a longitudinal firm-level study,
turnover in the firm must change the average quality of the workers. Therefore, workers
leaving profit sharing firms must be compared to the workers who stay in these same
firms. Because the sample is reduced to only those workers in firms that adopted profit
sharing, the workers leaving profit sharing firms are compared to the appropriate group.

The final method is similar to the second, but relaxes the sample restrictions. In
this section, workers are categorized using two years of observations (rather than three).
In the first year, only those workers who report that profit sharing is made available are

included in the sample. Their status in the second year forms the categories (by second

33
year profit sharing status and whether or not the worker remained with the same firm).
The advantage of this method is that it may provide evidence of the effect of turnover
further after the adoption of a profit sharing plan. Because of the construction of the
sample, the second method examines turnover immediately following adoption. In
contrast, the third method potentially includes turnover at many points following profit
sharing adoption. Therefore, while this study cannot speak to the quantity of turnover
occurring at different points following adoption, it may be able to show a temporal
pattern in the quality of the workers leaving firms with profit sharing plans.

Several proxies for worker quality are used. Years of education and the score on
the Armed Forces Qualifications Test (AF QT) are used to approximate the respondent's
skill level. While these measures should provide a good indication of the skill level of the
workers, they may not be as effective in capturing less tangible qualities such as effort
and the ability to work well in groups. The third proxy, wages, may be superior to the
others because it conveys information on all productivity-related traits (as long as the
correlation of wages with these traits has the expected sign).

Those reporting their industry as public administration in the current or previous
year are removed from the sample. There should be no instances of profit sharing in
public administration, as government work does not generate profits.8 In addition, those
who report working less than 35 hours per week, less than 6 years of education, an

hourly wage in excess of $100 per hour or below one half of the minimum wage, are

 

8 In spite of this, 7 percent of those workers reporting themselves as working in public administration also
report that profit sharing was made available (compared to 30 percent of the workers in the private
sector).

34
excluded. Lastly, if a worker meets the requirements for entry into a sample more than
once, one observation is randomly selected and used for the analysis. This is done so that
workers with an increased propensity to change jobs do not receive greater weight.

The profit sharing variable is an indicator equaling one if profit sharing is
reported as being made available by the employer in that year. This is not an ideal
variable in that there are many types of profit sharing plans, and a worker’s reaction may
differ across types. For example, a deferred plan (which pays into a retirement fund)
should provide a smaller incentive than a cash plan (which pays in the same year).9 This
concern is lessened by the fact that the prediction for all of the plan types is in the same
direction; only the strength of the incentive differs.

The fact that the profit sharing indicator depends on self-reporting raises the
potential for measurement error. Kruse (1995) employs the same data, and notes that this
response may be a bad indication of profit sharing status if workers report profit sharing
as “not being made available” in years that a firm earns no profits, and therefore the
value of the profit share per worker is zero. 10 On the other hand, the self-reporting of
profit sharing status may allow for a better measure of a worker’s response if it captures
the perception that the employer provides profit sharing. After all, in order for there to
be an incentive effect, a worker must believe that she is, in fact, eligible to receive a

share of the profits, regardless of the firrn's classification of her status.

 

9 There is some evidence that the NLSY workers frequently report a profit sharing plan that contains at
least some deferred component, in that 41% of those reporting that they have a retirement plan (other
than Social Secmity) also report the presence of a profit sharing plan, while only 10% of those who do
not report having a retirement plan report having a profit sharing plan.

‘0 Kruse (1995), p. 7.

35

Results for Job Changers Leaving Non-Profit Sharing Jobs

Table 1 shows the mean years of education, AF QT scores and hourly wages for
the workers who changed jobs, classified by the profit sharing status of their new job.11
The AP QT score reported in this table (and used in the regressions that follow) has been
modified in two ways. First, each respondent’s score has been rescaled to standard
deviations from the mean. Second, the score is adjusted to reflect the fact that the
respondents were administered the exam at different ages. All of the respondents were
asked to take the AF QT in 1980, at which time their ages ranged from 15 to 23.
Potentially, therefore, differences exist that reflect the fact that less schooling had been
acquired at the time the test was taken. To correct for this, the AF QT scores are
regressed on age dummies, and the residuals from this regression are used in the analysis
that follows.

The means in Table 1 indicate that, relative to workers entering non-profit sharing
firms, workers entering profit sharing firms score .124 standard deviations (4.9%)12
higher on the AF QT, have .41 (3.2%) more years of education and are paid 27 cents
(3.2%) more per hour prior to changing jobs.13 This wage measure, as opposed to the

wage obtained after changing jobs, is used to avoid including any differences in wages

 

’1 Wages are expressed in 1983 dollars.

’2 The base used in the calculations is the weighted average for both groups. The percent difference for
the AFQT scores is calculated using the raw scores on the exam (63.31 for those entering profit sharing
versus 60.29 for those not entering profit sharing).

’3 Job changers are defined as those reporting less weeks on the current job than weeks between
interviews.

36
due solely to the presence of a profit sharing plan (because the sample is limited so that
all of the workers are in non-profit sharing firms before changing jobs).

Table 2 presents OLS regression results with each of the three characteristics as
dependent variables. The first row of columns 1 and 2 simply regresses the years of
education and AF QT scores (respectively) on a constant and a dummy for workers
entering a profit sharing firm. Columns 3 and 4, which present estimates using wages as
the dependent variable, proceeds differently. An attempt is made in these regressions to
net out any differences in wages due to discrimination or region, while retaining
variation due to differences in “productive” characteristics such as tenure. For example,
if profit sharing workers tend to have more tenure or experience than non-profit sharing
workers, one does not want to remove these differences by including these variables in a
wage regression as one typically would. However, it may also be inappropriate to
exclude these variables entirely. If, for example, men have a higher tendency than
women to enter profit sharing firms, but also tend to have more experience, the estimate
of the wage difference for those entering profit sharing firms is biased downward if a
dummy variable for men is included but experience is not.

In order to account for this, the wage regressions include two variables that are
constructed as the residuals from regressions of male and white dummies on tenure and

its square, experience and its square, education and marital status.14 This technique

L

’4 The inclusion of marital status is based on the Korenman and Neumark (1991) finding that married
workers are more productive, and therefore may deserve the premium they are observed to earn.

37
should remove the influence of differences in “productive” characteristics from the sex
and race indicators.

In addition, in column 3 the residuals from a regression of wages on AF QT scores
and years of education are used as the dependent variable in attempt to derive a result
that is distinct from those obtained in columns 1 and 2. The estimates from the first
wage regression (row 1 of column 4), show that the wages prior to changing jobs were
almost seven percent higher for those entering profit sharing firms versus those entering
non-profit sharing firms. ’ 5

Examining simple differences (as above) is sufficient if profit sharing is
distributed equally across occupations within firms. Unfortunately, this does not seem to
be the case. While Kruse (1993) finds that the occupational mix within profit sharing
firms is not significantly different from other firms, the coverage within profit sharing
firms does vary by occupation. Specifically, he finds that “[w]ithin profit-sharing firms,
production and service workers are somewhat less likely to be covered (75.8 percent)
than are clerical / technical (86.5 percent) and professional / administrative employees
(87.6 percent)” ’6 This is important, for if one finds (as is the case here) that workers
who report having a profit sharing plan are more educated on average, this could simply
reflect that, within a firm more white collar workers are offered profit sharing, and not

that profit sharing firms attract higher-skilled workers. Given this, it may be more

 

1’ Region (northeast, south and west, with north central excluded) and year dummies are also included in
the wage specifications.
'6 Kruse (1993), p. 19.

38
appropriate to look at the differences in characteristics controlling for occupation. The
second row Table 2 presents these results.

Row 2 of Table 2 shows that the inclusion of occupation dummiesl7 reduces the
estimated differences between workers in each of the specifications. Still, the finding
that profit sharing firms attract better workers persists in two of the four columns. Also,
note that the inclusion of occupation dummies may be over-controlling if firms change
their occupational mix following the adoption of profit sharing. Firms may have the
incentive to substitute away from low-skilled workers and into capital in order to avoid
sharing profits. If this results in an increase in the hiring of high-skilled workers, the
average skill level for the firm increases, but not necessarily within occupations at the
firm. This, combined with the fact that profit sharing is not distributed equally across
occupations within firms, suggests that rows 1 and 2 bound the correct estimate of the
difference between workers entering profit sharing and non-profit sharing firms.

There is another issue to be addressed before one can conclude that these results
have any bearing on a longitudinal firm-level study. That is, even if the workers entering
profit sharing firms are “better” on average than those who do not, the workers entering

the firm must be better than the average worker already in the profit sharing firm. In an

 

’7 There are eleven categories: Professional, technical and kindred; Managers, officials and proprietors;
Sales workers; Clerical and kindred; Craftsmen, foremen and kindred; Operatives and kindred; Laborers,
except farm; Farmers and farm managers; Farm laborers and foreman; Service workers, except private
household; and private household. Those reporting private household as their occupation are excluded
from the sample.

39
effort to account for this, various measures of firm size are used as proxies for the labor
quality at the firm the worker enters. ’8

Rows 3 and 4 of Table 2 show the results including the proxies for firm size (of
the firm into which the worker switched). The proxies are the log of the number of
workers at the respondent’s site, a dummy indicating a firm with multiple sites, and an
interaction between the multiple-site dummy and an indicator for more than 1,000
employees at the other sites combined. These proxies are included with and without
occupation dummies, so that row 1 can be compared to row 3, and row 2 to row 4.
Although several coefficients remain significant following the introduction of these
proxies, the fact the magnitude of the estimate is reduced seven of the eight specifications
may indicate that the inclusion of better information about existing labor quality would
reduce the estimates further.

This section has provided somewhat mixed evidence about workers entering
profit sharing firms. Although it seems that the differences in AF QT scores and wages
net of education and AF QT scores are negligible, more significant differences between
workers exist for years of education and prior wages (not net of the other characteristics).
To provide some perspective, the estimates show that if the entire work force turned over
in the year of adoption, one would expect to see about a 3.5 to 5 percent increase in firm
level productivity.19 The next section shifts the focus to workers leaving profit sharing

firms.

 

’8 This is based on the Brown and Medoff (1989) finding that larger firms pay higher wages in part
because they have higher labor quality.

’9 This number is based on the estimated wage difi’erences in rows 3 and 4 of column 4 of Table 2. The
alternative method of adding the first three columns of these rows (and assuming a return of .051 to

40

Results for Workers in Firms that Adopt Profit Sharing

This section presents a different attempt to control for the existing level of worker
quality by studying workers in firms that adopt a profit sharing plan. While the previous
analysis focused on workers who tend to enter profit sharing firms, this section examines
those who leave firms following the adoption of a profit sharing plan. No attempt is
made to differentiate between voluntary and involuntary separations, as either could
contribute to a change in the composition of the firm’s labor.

Three years of observations are required to categorize the respondents. For three
of the groups, the respondents are required to have been in a non-profit sharing firm in
year 1, then be observed in the same job in year 2, but reporting that a profit sharing plan
is offered. Workers are placed in the groups based on their response in year 3: the
respondent remains in the job and still reports having a profit sharing plan (PS-PS),
remains in the job and now does not report having a profit sharing plan (PS-NPS), or the
respondent leaves the firm (PS-LV). The remaining two groups are composed of
workers who never report the presence of a profit sharing plan (from 1988-1993). NV-
LV workers are observed to leave their job after holding it at least one year (to match the
selection criterion for the profit sharing workers), and NV-ST workers remain in their

jobs for at least two years.

 

standardized AFQT scores taken from a “kitchen sink” wage equation, and a 7 percent return for a year
of education) yields the same estimate.

41

The observations meeting the requirements (from data series beginning in 1988,
1989, 1990 and 1991) are pooled to form the data set. In the rare case that an individual
meets these qualifications more than once over the sample period, one observation is
selected randomly. Years of education, standardized AF QT scores and wages (from year
1) are used to compare the workers.

The use of wages in this section is somewhat problematic. That is, if the wage
from year 3 is used, a comparison is made between workers with a different profit
sharing status, which is misleading if institutional differences exist in the setting of
wages between the profit sharing and non—profit sharing sectors. The wage from year 2
may also be a bad proxy because of potential correlation between the selection categories
and the error term. That is, a comparison is made between workers who are about to
leave the firm with those who will be staying. It is difficult to distinguish whether
observing a low wage draw for those leaving the firm reflects a lower level of
productivity, or the worker leaving because of the perception that she is underpaid.
Because of these problems, the analysis instead uses the wage from year 1. Due to the
high correlation between the year 1 and year 2 wages (.64), this may not solve the
endogeneity problem entirely, but should be less problematic than the alternatives.

Table 3 reports the means of the data for four categories of workers: PS-LV, PS-
PS, NV-LV and NV—ST. PS-NPS workers are dropped from the sample because of the
possibility that one of the two observations on profit sharing status is incorrectly

reported. That is, in order to be classified as PS-NPS, a worker must report that her firm

42
adopted and then abandoned a profit sharing plan in the span of one year.20 Therefore,
the group containing workers remaining in profit sharing plans (in row 2) includes only
workers classified as PS-PS.

A comparison of the first two rows of Table 3 shows that the workers remaining
in profit sharing firms have significantly higher wages, slightly higher (though not
significantly) AF QT scores, and essentially the same number of years of education
(although the mean is slightly higher for workers leaving profit sharing firms). Although
this comparison of means indicates that profit sharing firms retain workers of slightly
higher quality, a simple comparison of the means may be misleading. That is, turnover
in any firm may leave it with a higher quality work force, regardless of the profit sharing
status of the firm. Because of this, a differences-in-differences estimator is introduced.
The model that is used is:

Traiti = 0‘1 + BI(PS'LV)1 + 52(NV‘LVh + B3(NV‘ST)i + '1" xi + 8i
where the trait is either the standardized AF QT score, years of education or the log of the
hourly wage;21 X is included in the wage specifications and is a vector containing year
dummies, region dummies and the transformed dummies for male and white (see
previous section). PS-PS is the excluded group.

Table 4 reports the OLS estimates: row 1 reports [3,, the difference between
leavers and stayers for profit sharing firms, and row 2 reports the differences-in-

differences estimator: B, - (Bz - B3). Focusing first on the columns that do not include

 

20 It should be noted that the relative size of this group (400) is disturbing if one believes that its size is an
indication of the extent of the measurement error in the profit sharing indicator.

2’ As in the previous section, two wage variables are used: one is the residual from a regression on
education and AFQT scores while the other is simply the reported wage.

43
occupational controls, there is no significant difference between PS-LV and PS-PS
workers in years of education or AF QT scores, but the wages for workers leaving profit
sharing firms are significantly lower. However, once the “leaver” effect is removed
using the differences-in-differences estimator (shown in row 2), this difference vanishes.
Therefore, the differences between future leavers and stayers in profit sharing firms are
no different than they are in non-profit sharing firms.

It should be emphasized that the labor quality proxies (the firm-size variables)
used in the previous section are not necessary here because the PS-PS and PS-LV
workers were all in firms that adopted a profit sharing plan. Therefore, if a below
average worker is observed to leave their firm, it is now certain that the average quality
of the remaining pool of profit sharing workers is higher.

The occupational controls are also not vital here, but may have some use. That is,
because all of the workers in this sample report that their firm adopted a profit sharing
plan, there is no longer the problem of comparing groups composed of different
percentages of white collar workers. However, the use of occupational controls may be
useful in determining whether firms change their occupational mix following adoption.
To illustrate this, consider the case in which a firm shifts its occupational mix toward
high-skilled workers. A larger fraction of low-skilled workers will then leave the firm (if
the low-skilled workers are forced out), making the estimated difference between the
stayers and leavers appear to be large simply because workers in low-wage occupations
with lower educational requirements are leaving more frequently. If one controls for

occupation, however, the differences between leavers and stayers should disappear if the

44

firm changes the ratio of high-skilled to low-skilled workers without changing the
average level of quality within each category. For example, one would expect to see
such a change in the estimated differences if secretaries that leave are no different from
the secretaries that stay, but more secretaries than executives leave. Table 4 shows that
this is not the case, however. Comparing the wage regressions in row 1 (containing the
only the statistically significant estimates) shows that the estimated differences are
unaffected by the inclusion of occupation dummies.

Of course, whether a change in productivity is due to lower quality workers
leaving the firm (across all occupations) or a shift away from low-skilled labor is not
important to the central issue of this paper. Either response leads to a change in the
estimated firm-level productivity without an increase in the performance of individual
workers. Because of this, the most informative estimates are those from the regressions

without occupational controls.

Results Relaxing the Sample Restrictions

This section uses the same estimation method as the previous section, but relaxes
the requirements for inclusion in the sample. While the previous section required three
years of observations in order to categorize respondents, this section uses only two. For
three of the groups, the first year includes only workers reporting the presence of a profit
sharing plan. The observation in the second year is used to create the same categories as
in the previous section: PS-PS, PS-NPS, and PS-LV. As before, the PS-NPS workers are

dropped from the sample. In addition, the NV-LV and NV-ST groups are formed, but in

45
this case, “leavers” are workers who left the firm in year 2 (in the previous section, both
groups remained in the same job from year 1 to year 2).

This method has two benefits. First, the sample size of profit sharing workers is
greatly increased, allowing for more precise estimates of the differences. Second, it may
allow an examination of turnover further distant from the adoption of a profit sharing
plan. By construction, the second methodology examined workers immediately
following the adoption of a profit sharing plan (which may be considered an advantage
of the second method). In contrast, this section may be studying workers at any point
beyond adoption, creating the possibility of observing qualitative differences in turnover
across time.

Table 5 presents the means for this sample. A comparison of this table with
Table 3 shows that the sample sizes for the profit sharing groups are three to four times
larger, and that there are larger differences between the PS-PS and PS-LV groups in the
new sample. Table 6 presents the results using the same specifications as in the previous
section. Row 1 of Table 6 shows that there are consistent differences between future
leavers and stayers from profit sharing firms. However, the differences-in-differences
estimates, shown in row 2, show that these differences are the same for workers staying
in or leaving non-profit sharing firms. Furthermore, a comparison across the columns
shows that the occupational controls move the estimates in the expected direction,
although frequently the difference is not substantial.

The previous two sections have attempted to observe whether attrition out of a

profit sharing firm improves the average quality of its workers. The findings indicate

46
that, while some differences are observed between leavers and stayers, these differences
are the same in non-profit sharing firms. Therefore, the adoption of a profit sharing plan

does not appear to affect attrition out of a firm.

Conclusions

Previous studies of the effects of the adoption of profit sharing plans on worker
productivity employ firm-level panel data, and may not adequately control for changes in
the quality of labor due to turnover. If the quality of the work force improves following
the adoption of a profit sharing plan, one may not be able to interpret results from these
studies as fully representing incentive effects due to profit sharing plans. This paper uses
individual-level data in an attempt to find if the quality of the work force in a profit
sharing firm improves because higher-quality workers enter the firm and lower-quality
workers leave.

The results show that higher quality workers tend to enter profit sharing firms
versus non-profit sharing firms, which increases the average quality of the workers in
profit sharing firms. However, profit sharing plans have no effect on attrition out of the
firm, as the estimated differences between leavers and stayers is the same for profit
sharing and non-profit sharing firms.

The magnitude of the differences between workers entering profit sharing firms
can be interpreted as follows: assume that the number of employees at the firm is

constant (so that everyone who leaves the firm is replaced by one worker), and that the

47
effect of higher quality workers entering the firm creates a 3.5 to 5 percent improvement
per worker (see footnote 19). These assumptions, in conjunction with the
(approximately) 19 percent of profit sharing workers changing jobs per year, produce a
.67 to .95 percent annual improvement in firm-level productivity due a change in worker
quality.22 Given that Kruse (1993) estimates a 4 percent improvement in the year
following the adoption of a profit sharing plan, the exercise shown above suggests that
less than one-fourth of the increase in firm-level productivity is due to a change in the
quality of the firm’s labor.

It is possible, however, that this fraction is larger. First, given that the profit
sharing status is self reported, there may be a significant amount of measurement error in
the selection categories, which biases the estimates toward zero,23 reinforcing one’s
confidence that the observed differences exist.

The 19 percent annual turnover figure may also understate the actual conditions
because it is possible that turnover is higher in the first year of the new compensation
system. That is, if workers select jobs based on the form of compensation that is used,
one might expect that turnover would increase when the method of compensation
changed.

While the size of the impact on the firm-level productivity studies may be in

doubt, it seems more certain that higher quality workers are more likely to enter profit

 

22 The estimate of annual turnover (19 percent) is from the NLSY sample of full-time workers. Although
this estimate may be high because this is a sample of young workers, it is modest in magnitude relative to
the estimates contained in Anderson and Meyer (1994), which finds a 17 percent quarterly permanent
separation rate.

2’ Freeman (1984).

48
sharing firms. Further research may better indicate why these workers prefer the profit

sharing environment.

Table 1
Means of the Data for Job Changers1

 

Standardized Years of Wage in the N
AFQT Score Education Previous Year

 

 

 

Entered a profit .096 13.15 8.53 728
sharing firm (.037) (.09) (.20)
Entered a non- -.028 12.74 8.26 2,477
profit sharing firm (.020) (.05) (.11)

 

 

 

 

 

' All workers in the sample were in a non-profit sharing job and then changed jobs. They are classified
by “year 2” status. Standard errors are in parentheses. Wages are expressed in 1983 dollars.

49

 

Table 2

The Diference Between Workers Entering Profit Sharing Firms and Non-Profit
Sharing Firms by Worker Characteristic

 

 

 

 

 

 

 

Dependent Variable
Specification Education AFQT Scores Log (W age)l Net of Log (W age)l
Other Traits
No Controls .422** .119** .035* .067**
(.103) (.042) (.020) (.021)
Including Occupation .178** .047 .029 .042**
Controls (.087) (.039) (.019) (.020)
Including Firm Size .350** .089** .024 .050**
Variables (.105) (.043) (.020) (.021)
Including Firm Size and .180** .033 .021 .033

Occupation Controls (.089) (.040) (.020) (.020)

 

 

 

 

 

N=3.205 in rows 1 and 2 of columns 1 and 2. N=3,077 in rows 1 and 2 of columns 3 and 4. N=2,999 in
rows 3 and 4 of columns 1 and 2. N=2,880 in rows 3 and 4 of colums 3 and 4. l The wage variable is
first regressed on AFQT scores and years of education, with the residuals from this regression used in the
specifications of this table. Standard errors are in parentheses. All specifications include geographic

region dummies. Wages are in 1983 dollars. ** denotes significance at the five percent level. * denotes

significance at the ten percent level.

50

 

Table 3
Means of the Data for Workers Already in Profit Sharing Firms1

 

 

 

 

 

 

Standardized Years of Wage in N
AFQT Score Education Year 1
Left a firm with a profit .012 13.21 7.43 173
sharing plan (PS-LV) (.065) (.18) (.27)
Stayed in a firm with a .058 13.15 8.17 560
profit sharing plan (PS-PS) (.041) (.09) (.15)
Left a non-profit sharing -.097 12.98 7.00 542
firm for another non-profit (.045) (. l l) (.21)
sharing firm (NV-LV)
Stayed in a non-profit .034 13.04 7.43 2,115
sharing firm (NV-ST) (.021) (.05) (.09)

 

 

 

 

 

‘ The workers in rows 1 and 2 report profit sharing jobs in “year 1.” They are classified by their status the following
year. Workers in rows 3 and 4 never report having a profit sharing plan. Standard errors are in parentheses. Wages
are expressed in 1983 dollars.

51

 

Estimates of the Differences in Characteristics for

Table

4

Workers Already in Profit Sharing Firms

 

 

 

 

 

Years of Standardized Log (Wage) Net Log (W age)1
Education AFQT Scores of Other Traitsl
Occupational No Yes No Yes No Yes No Yes
Controls

Leave PS - Stay in PS .062 -.069 -.046 -.087 -.112** -.101** -.114** -.112**
(.208) (.172) (.086) (.079) (.036) (.035) (.039) (.036)

(L - S)” - (L - Shops .122 -. 130 .085 .008 -.030 -.025 -.019 -.029
(.238) (.196) (.098) (.090) (.041) (.040) (.045) (.041)

 

 

 

 

 

 

 

 

 

N = 3,390. ’ The wage used is from “year 2,” when all of the workers are in a profit sharing firm. Two variables

are included in the wage regressions in order to remove the effects of discrimination. These variables are

constructed by forming the residuals from regressions of male and white on tenure, tenure-squared, experience,
experience-squared, education and marital status. This is done so that any variation in wages due to differences in
tenure, et a1. by group are retained in the estimates. Standard errors are in parentheses. .. denotes significance at the
ten percent level. ' denotes significance at the five percent level.

52

 

Table 5
Means of the Data for Workers in Firms Adopting Profit Sharing2

 

 

 

 

 

 

Standardized Years of Wage prior to N
AFQT Score Education PS adoption
Left a firm with a profit -.009 12.97 9.11 723
sharing plan (PS-LV) (.033) (.08) (.19)
Stayed in a firm with a .128 13.28 10.90 1,635
profit sharing plan (PS-PS) (.024) (.05) (.13)
Left a non-profit sharing -. 188 12.49 7.84 1,021
firm for another non-profit (.031) (.08) (.18)
sharing firm (NV-LV)
Stayed in a non-profit .0003 13.02 9.49 2,422
sharing firm (NV-ST) (.020) (.05) (.11)

 

 

 

 

 

‘ The workers in rows 1 and 2 first report a non-profit sharing job, remained in the job the following year, but
reported that profit sharing is offered. They are classified by their status the following year. Workers in rows 3 and
4 never report having a profit sharing plan. Standard errors are in parentheses. Wages are expressed in 1983
dollars.

53

 

Table 6
Estimates of the Differences in Characteristics for
Workers in Firms Adopting Profit Sharing

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Years of Standardized Log (Wage) Net Log (W age)1
Education AFQT Scores of Other Traitsl
Occupational No Yes No Yes No Yes No Yes
Controls
Leave PS - Stay in PS -.302** -.159* -.137** -.098** -.159** -.145** -.184** -.161**
(.104) (.089) (.044) (.041) (.019) (.018) (.020) (.019)
(L - S)ps - (L - S)N,,ps 228* .035 .051 -.006 -.007 -.015 .011 -.013
(.136) (.116) (.057) (.053) (.024) (.024) (.026) (.024)
N = 5,801. I The wage used is from “year 1,” when all of the workers are in a non-profit sharing firm. Two

variables are included in the wage regressions in order to remove the effects of discrimination. These variables are
constructed by forming the residuals from regressions of male and white on tenure, tenure-squared, experience,
experience-squared, education and marital status. This is done so that any variation in wages due to differences in
tenure, et a1. by group are retained in the estimates. Standard errors are in parentheses. .. denotes significance at the
ten percent level. . denotes significance at the five percent level.

54

 

BIBLIOGRAPHY

BIBLIOGRAPHY

Anderson, Patricia M. and Bruce D. Meyer. 1994. “The Extent and Consequences of
Job Turnover.” In Brookings Papers on Economic Activity, Martin Neil Bailey et
al, eds. Washington, DC: Brookings Institution: 177-236.

Brown, Charles and James Medoff. 1989. “The Employer Size-Wage Effect,” Journal
of Political Economy 97: 1027-1059.

Freeman, Richard B. 1984. “Longitudinal Analyses of the Effects of Trade Unions,”
Journal of Labor Economics 2: 1-26.

Korenman, Sanders and David Neumark. 1991. “Does Marriage Really Make Men
More Productive?” Journal of Human Resources 26(2):282-3 O7.

Kruse, Douglas L. 1993. Profit Sharing: Does It Make a Diference? Kalamazoo, MI:
WE. Upjohn Institute.

Kruse, Douglas L. 1995. "Profit Sharing and the Demand for Low-Skill Workers,"
Unpublished working paper.

Shepard, Edward M. HI. 1994. “Profit Sharing and Productivity: Further Evidence from
the Chemicals Industry.” Industrial Relations 33: 452-66.

Weitzman, Martin L. 1983. "Some Macroeconomic Implications of Alternative
Compensation Systems," Economic Journal 93: 763-83.

Weitzman, Martin L. and Douglas L. Kruse. 1990. "Profit Sharing and Productivity."
In Paying for Productivity: A Look at the Evidence, Alan S. Blinder, ed.
Washington, DC: Brookings Institution: 95-140.

Wilson, Nicholas and Michael J. Peel. 1991. “The Impact on Absenteeism and Quits of

Profit-Sharing and Other Forms of Employee Participation.” Industrial and Labor
Relations Review 44: 454-68.

55

Chapter 3
SUPERVISION, EFFICIENCY WAGES AND INCENTIVE PLANS:
HOW ARE MONITORING PROBLEMS SOLVED?
Introduction

Many models of the employment relationship assume that monitoring worker
behavior is costly. Examples include the shirking model of efficiency wages in Shapiro
and Stiglitz (1984) and the bonding model in Lazear (1981). A common characteristic of
these models is that firms employ a compensation mechanism in order to extract effort
from workers. This paper examines the use of three such mechanisms: efficiency
wages,1 perforrnance-based bonuses and profit sharing plans.

In theory, any one of the three schemes could be observed where monitoring is
difficult (costly). In the shirking model of efficiency wages, for example, as monitoring
becomes more costly, the firm benefits by substituting away from supervision and instead
pays its workers a higher wage.2 A similar motivation may lead to the use of both
performance-based bonuses and profit sharing plans.3 Because these plans allow the
workers to share in productivity gains, they may no longer find it optimal to shirk, as

shirking now has an opportunity cost associated with it. Because all three compensation

 

’ Throughout the paper, the term “efficiency wages” refers to the Shapiro and Stiglitz (1984) variant of
the efficiency wage model. Other versions, such as Akerlof (1986), do not necessarily imply the same
tradeoff between supervision and wages that is contained in the shirking model of Shapiro and Stiglitz.

2 See Shapiro and Stiglitz (1984) or the final section of this paper for the mechanism through which this
affects worker effort.

3 The difference in the definition of these plans can be seen on p. 62.

56

57

mechanisms have the potential to perform the same task, a common prediction is that,
ceteris paribus, each one is more likely to be observed as monitoring becomes more
difficult.

It should be noted that this study explores the prevalence of these plans vis-a-vis
the level of supervision, but cannot directly determine the eflectiveness of these three pay
schemes at reducing shirking. However, assuming that firms would not knowingly adopt
ineffective methods, prevalence and effectiveness should be closely related.

The majority of the previous work in this area has focused on the relationship
between wages and supervision as tests of the shirking model of efficiency wages. Table
1 summarizes these studies, which tend to follow one of two methods. The first type of
study isolates a source of previously unexplained wage dispersion (e. g., the firm-size
wage effect in Brown and Medoff, 1987, or the inter-industry wage differential in Neal,
1993), and then introduces a proxy for monitoring in an attempt to explain the wage
dispersion. Other studies, such as Kruse (1992) or Groshen and Krueger (1990), take a
more direct approach by regressing wages on a measure of supervision (typically the
supervisor-to-staff ratio or the frequency of supervision) in order to test whether wages
increase as the extent of supervision decreases.

The results of these studies have provided mixed evidence on the shirking model.
Studies using the supervisor-to-staff ratio as a proxy for monitoring difficulty (Leonard,
1987; Fitzroy and Kraft, 1988; Groshen and Krueger, 1991) find almost no support for
the shirking model. However, many of the studies using alternative proxies, including

Krueger (1991), Kruse (1992), Arai (1994), Osterman (1994) and Rebitzer (1995) find

58

evidence of a tradeoff between wages and the extent of supervision. The support among
such studies is not universal, though, as Brown and Medoff (1989) and Neal (1993) do
not observe this tradeoff.

Groshen and Krueger (1990) provide a possible explanation for the difference in
the results depending on the form of the monitoring proxy that is employed. If
supervisors and workers are substitutes in production, as the wages of workers increase,
one expects a shift in the composition of the work force toward supervisors (now the
relatively cheaper input). This creates a positive relationship between the worker’s wage
and the supervisor-to-staff ratio, biasing studies using this proxy against finding the
prediction of the efficiency wage model. This “substitution bias” is potentially
important in this paper because the data contain a proxy for monitoring that is
functionally identical to the supervisor-to-staff ratio -- the number of employees that the
respondent’s supervisor supervises (the inverse of this variable is used in the analysis so
that the interpretation of the coefficients is consistent with studies utilizing the
supervisor-to-staff ratio).

A second bias, based on unobserved worker ability, may also affect these studies.
As a way of illustrating this bias, assume that supervisors monitor high-ability workers
less closely. If the data cannot identify these workers, one would expect a negative
correlation between the extent of supervision and the wages of these workers, biasing the
estimated effect of supervision on wages downward. Therefore, this bias makes it more
likely to falsely observe the prediction of the efficiency wage model. Alternatively, the

bias works in the opposite direction if high-ability workers want to ensure that they are

59

rewarded for their productivity, and therefore seek out jobs in which they are closely
supervised. This study attempts to mitigate the effect of unobserved ability bias by
including standardized test scores in otherwise standard specifications from the literature.
Although test scores may be a controversial proxy for ability (and are certainly
imperfect), the results indicate that unobserved ability does not play a role in estimating
the relationship between wages and supervision.

Omitted firm characteristics may also bias the results. Rebitzer (1995) provides
an example of the effect of omitted firm characteristics by focusing on the screening of
employees. As is shown below, the effect that the omitted variables have on the
estimates is determined by whether the variable is a substitute or a complement for
supervision. Because in general it is not possible to determine the degree of
complementarity, the sign of this bias is difficult to determine.

The study then goes beyond the usual wage-supervision analysis by focusing on
the extent to which performance-based bonuses or profit sharing plans are used as
substitutes for monitoring. Although the proxy for the level of supervision is the same as
above, the substitution bias that may affect the study of a wage-supervision tradeoff
should not be present. There is no compelling reason to believe that a firm would
substitute away from workers and toward supervisors following the adoption of an
incentive plan. Provided that the incentive is linked to productivity, the cost of the input
(workers) should not increase without a concomitant increase in its productivity,

eliminating the need for substitution away from the input. In addition, the potential for

6O

omitted ability bias may be lessened, as the relationship between ability and incentive
plans is less certain than its relationship with wages.4

An additional innovation in this study is its ability to separately examine the
relationship between monitoring and compensation schemes for supervisors, as well as
those supervised. Proxies are available for both the monitoring difficulty that the
supervisors face (the number of workers reporting to them) and the level of monitoring to
which supervisors are subjected (the number of other people their manager oversees).
The distinction between workers and supervisors proves to be important, as it appears
that only supervisors are offered incentives as work groups grow larger.

The findings may be summarized as follows: wages are related to the extent of
supervision for one of three samples of supervisors and for piece rate and commission
workers, but not for the majority of workers. In addition, neither performance-based
bonuses nor profit sharing plans are more likely to be given to workers as it becomes
more difficult to monitor them. However, the results do show a strong relationship
between supervisory bonuses (but not profit sharing plans) and the number of employees

that a supervisor monitors.

Data
The data are from the National Longitudinal Survey of Youth (NLSY). Although
this survey has been taken since 1979, the length of the panel used here is substantially

reduced because some of the questions of importance are only included in the survey in a

 

4 This theory is borne out in the data, as AF QT scores are not a significant determinant of either form of
incentive plan studied here, but have a statistically significant impact on wages.

61

limited number of years. Supervisors and workers are examined separately, as defined
by responses to questions about supervisory duties and the size of work groups.
Appendix A displays the text of the questions used to identify the samples.

Three definitions are used to form the samples of supervisors. The broadest of
these includes anyone who reports that they “supervise the work of others, or tell them
what work to do, on a day-to-day basis.” A follow-up question on the respondent’s
power to set pay or determine promotions is used to narrow the sample to ensure that it is
composed of supervisors with real authority (so that supervisory monitoring, and not
inter-worker monitoring, may be examined). The final sample of supervisors is formed
using three-digit 1980 Census of Population occupation codes. The criterion used for
inclusion is that the word “supervisor” is in the description of the occupation (the specific
occupations selected are shown in Appendix B). This method should exclude most
managers (those higher-up in the ladder of authority) and include only the “front-line”
supervisors. Therefore, this method may provide a cleaner test of the traditional
interpretation of the shirking model: the supervision of workers involved in the
production of a good or service (as opposed to the supervision of other supervisors).
Regardless of which definition is best, looking at three definitions of supervisors should
provide a sense of the robustness of the results.

The sample of workers consists of those responding to the question, “[b]esides
yourself, for how many other people does/did your supervisor serve as the immediate

supervisor on a day-to-day basis?” This question is only included in the survey in 1990.

62
It should be noted that, because many supervisors have their own supervisor, this sample
includes some respondents who are also in the supervisory samples.5
The profit sharing indicator is taken from the question “Does/Did your employer
make available to you...profit sharing.” The performance-based bonus indicator comes
from the question:
The earnings on some jobs are based all or in part on how a person
performs on the job. On this card are some examples of earnings that are
based on job performance. Please tell me if any of the earnings on your job
are/were based on any of these types of compensation.
1. Piece rate
2. Commissions
3. Bonuses (based on job performance)
4. Stock options
5. Tips
The indicator equals one for those responding that earnings are based on bonuses (part
3), and zero otherwise. Parts 1 and 2 of this question are used to further define the
sample. For the primary analysis, those workers and supervisors reporting pay based on
a piece rate or commission are excluded because these forms of compensation may
themselves be substitutes for supervision. The use of these methods of pay may signal
either a greater need for monitoring on these jobs, or that any monitoring problem that
may have existed is solved by these pay schemes (without the use of efficiency wages or
other incentive plans). Because of this, it may be useful to examine these workers
separately.

The ability proxy comes from standardized test scores on the Armed Services

Vocational Aptitude Battery (ASVAB), administered to the sample in 1980. Four of the

 

5 The qualitative results do not change when supervisors are excluded from this sample.

63

ten test scores from the ASVAB are used to form the Armed Forces Qualifications Test
(AF QT) score.6 All of the respondents were asked to take the AF QT in 1980, at which
time their ages ranged from 15 to 23. Because scores may differ because less schooling
had been acquired at the time the test was taken, the AF QT scores are regressed on age
dummies, and the residuals from this regression used in the analysis that follows.

Finally, those reporting their industry as public administration,7 working less than
35 hours in the survey week, in a union, reporting a wage less than one half of the
minimum wage, or greater than $100 per hour are excluded from the samples.8 Table 2

shows the means for the samples to be examined.

Re-Examining Evidence on the Shirking Model of Efficiency Wages for Non-
Supervisory Workers
The methodology used to test for the shirking model of efficiency wages follows
previous attempts in the literature. A wage regression of the following form is used:
ln(wage) = X13 + E5 + Sy + AA + e
where X is composed of the standard controls,9 E is the log of the establishment size

(from a respondent’s estimate), S is the supervisor-to-staff ratio, and A is the

 

6 The AP QT score is calculated by adding the scores from arithmetic reasoning, word knowledge,
aragraph composition and one-half of the numerical operations exams.

Those reporting public administration as their industry are removed from the analysis because of the
frequent difficulty in determining either the level of productivity or the monetary value of increases in
productivity in this industry. Support for this restriction comes in Arai (1994). He separately examines
public and private sector workers, and finds that the efficiency wage model is consistent with the results
for the private sector, but not public sector workers.

8 All monetary figures are converted to 1983 dollars.
9 The dummies that are included are: male, white, married, northeast, north central, south, a dummy for
multiple-site firms, an interaction of this with a dummy for multiple-site firms with more than 1,000

o4

standardized AF QT score. First, this model is run with the restriction that y = 0 and A. =
0. The restrictions are then removed in stages.

When examining the samples of workers, three hypotheses may be tested. First,
y<0 -- showing that workers are paid more as they are supervised less closely -- is
consistent with the shirking model. Alternatively, finding 'y>0 suggests that workers are
paid a compensating differential for having to endure closer supervision, or that the
worker substitution bias has overwhelmed the true estimate. Second, it has been
theorized that large firms (or establishments) pay higher wages in part because
monitoring worker behavior is more difficult in that setting. 10 If this is true, the
inclusion of a proxy for monitoring difficulty moves the estimate of 5 toward zero.
Third, researchers have hypothesized that the wage variation across industries may be
due to systematic monitoring differences (Krueger and Summers, 1988; and Neal, 1992).
An examination of the industry indicators before and after the addition of the supervision
proxy provides a test of this theory.

Table 3 shows the estimates from OLS wage regressions for workers who are not
paid by piece rate or commission. The magnitude of the establishment size-wage effect
in column 1 (.034) is consistent with the findings from individual-level data in Brown
and Medoff (1989), which range from .013 to .038. The supervisor-to-staff ratio,

included in column 2, has a coefficient that suggests a negative relationship between

 

employees, two-digit industry (see Appendix C) and one-digit occupation. The other variables included
are: years of education, experience (actual) and its square and tenure and its square.

’0 See Brown and Medoff (1989) for a discussion of this. These data are consistent with the theory that
monitoring is more difficult in large firms, as the correlation between the establishment size and the
supervisor-to-staff ratio is negative in all of the samples examined.

65

wages and supervision, but the magnitude is small. A one standard deviation increase in
the supervisor-to-staff ratio decreases a worker’s wage by less than one percent.
Furthermore, the inclusion of the monitoring proxy does not affect the estimate of the
establishment size-wage effect. According to these findings, supervision is not strongly
related to wages for these workers.

An attempt is made to eliminate the omitted ability bias by including standardized
AF QT test scores.11 If it is true that higher-ability workers are paid more and do not
need to be supervised as closely, the inclusion of a proxy for ability should increase the
estimate of y. However, column 3 shows that, while the respondent’s score on the AF QT
is a significant determinant of wages, the coefficient on the supervisor-to-staff ratio is
essentially unchanged. ‘2

The focus shifts to employees paid a piece rate or by commission in Table 4. 13
Working under these compensation systems presumes that at least some dimension of a
worker’s productivity is easily determined (i.e. one has to be able to count the pieces in
order to pay a piece rate). Therefore, it may be that the number of workers per
supervisor does not restrict the level of monitoring. For example, if all one has to do to

monitor perfectly is add up the daily production of widgets, it may be as easy to monitor

 

1’ Firm size, establishment size (see Brown and Medoff, 1989) and education may also act as proxies for
ability.

’2 An attempt was made to correct for substitution bias by instrumenting with the average supervisor-to-
staff ratio for the other three regions in the same two-digit industry and for the other regions in the same
one-digit industry and one-digit occupation. The use of these instruments actually increased the
estimates of y, however, suggesting that either the substitution bias is in the opposite direction than others
have theorized, or that the instruments are not valid.

’3 The hourly wage used for the majority of these workers may actually represent hourly earnings. That
is, only 19 percent of the piece rate and commission workers report their wage on a per hour basis, but
over 40 percent of the non-piece rate and commission workers report an hourly wage.

66

one hundred workers as it is to monitor one worker. The results do not bear out this
prediction, however, as the estimate of y is negative and statistically significant (in
column 2), and the estimate of 5 (the establishment size-wage effect) is reduced by the
inclusion of the supervisor-to-staff ratio.

There are two possibilities for reconciling this result with existing theories. First,
recall that y<0 is consistent with an alternative theory, that wages increase as supervision
falls because of unobserved worker ability. Support for this theory applying to piece rate
workers comes from Lazear (1996), who finds that the variance in the ability of a firm’s
work force increases with the use of piece-rate pay relative to hourly pay. Therefore,
concerns about the quality of the good or service produced (for example) may lead firms
to supervise inferior workers more closely when they are paid per unit produced. If this
is the case, and ability is not accounted for in the regression, the supervision variable will
not provide an estimate of the relationship between supervision and wages for workers of
equal ability. This line of reasoning does not hold up when AF QT scores are added to
the regression (in column 7), as the coefficient on the supervisor-to-staff ratio is
unaffected. However, it should be noted that the coefficient on the AF QT scores is not
significantly different from zero, pointing to the possibility that ability, as it applies to
these jobs, is not captured by AF QT scores.

The second explanation for finding y<0 for piece rate and commission workers is
that these may be workers who are particularly in need of monitoring, and more than one
method of alleviating this problem is used simultaneously. That is, paying these workers

a commission or piece rate may be used in combination with an efficiency wage or a

67

higher level of supervision in order to prevent them from shirking. An example for each
type of worker may illustrate the point: piece rate workers may have the incentive to
shirk on quality in order to increase the quantity produced, while workers paid by
commission may benefit in the short-term by using tactics such as high-pressure sales,
which may reduce the profitability of the firm in the long run. In either case, closer
supervision is required to elicit the desired behavior from the worker.

These results, as in other studies that use the supervisor-to-staff ratio as a proxy
for monitoring, show little or no relationship between supervision and wages for
workers. The exception to this may be that piece rate and commission workers are paid

an efficiency wage.

Evidence on the Shirking Model of Efficiency Wages for Supervisory Workers

The analysis now turns to an examination of the relationship between wages and
monitoring for supervisors. Tables 5 through 7 present the similar specifications to those
estimated in Tables 3 and 4, but instead using the three samples of supervisors defined
above. Columns 1 and 2 take advantage of the fact that panel data are available for this
sample (except when the manager-to-supervisor ratio is included) by including individual
fixed effects in the estimation, which should eliminate bias due to fixed unobserved
ability. ’4 A surprising finding comes from the inclusion of fixed individual effects, in

that it completely eliminates the establishment size-wage effect across all three samples,

 

’4 Hausman tests strongly support the use of fixed effects, as reflected in the p-values reported in Table 5
through 7.

68

while Brown and Medoff (1989) show at most a 45 percent reduction. 1’ A negative and
significant coefficient on the supervisor-to-staff ratio in Table 5 is less surprising, as the
interpretation of this coefficient differs from the regressions examining the samples of
workers.

For supervisors, y<0 is consistent with two theories. First, it may reflect a
compensating differential paid to supervisors because of a distaste for having to oversee
more employees. Second, it may be due to the fact that supervisors and workers are
substitutes in production. As the wages of the supervisors are increased, more workers
are hired simply because they have become a relatively cheaper input.

The inclusion of the supervisor-to-staff ratio does not provide a test of the
shirking model, however. This is tested in columns 4 and 5, which explore the relevance
of the shirking model for supervisors by including the manager-to-supervisor ratio, for
which a negative coefficient is consistent with the shirking model. This variable
represents the level of monitoring to which the supervisor is exposed (fewer supervisors
per manager is interpreted as each supervisor being monitored more closely). ’6
Unfortunately, information about the manager-to-supervisor ratio is only available in
1990, so that fixed effects estimation can no longer be used to eliminate omitted ability
bias. Column 3 re-estimates the specification in column 1 without individual fixed

effects and using only 1990 data. The supervisor-to-staff ratio and manager-to-

 

” It should be noted that this result comes from data on supervisors, who are not examined ill Brown and
Medoff (1989). If the establishment size-wage effect is estimated using all NLSY workers from 1988 to
1993, the OLS coefficient on ln(establishment size) is .028, and the fixed-effects estimate is .011. Both
of these estimates are consistent with the fmdings of Brown and Medoff (1989).

’6 This variable is taken from the same survey question as the supervisor-to-staff ratio for the worker
sample. See Appendix A for details.

69

supervisor ratio are included in column 4 of Tables 5 through 7, producing mixed results.
The findings for supervisors defined by occupation codes (Table 7) are consistent with
the shirking model, but the other two samples show no relationship between wages and
supervision. Column 5 adds AF QT scores in attempt to eliminate omitted ability bias,
but, as in the case with the sample of workers, this does not affect the other coefficients.

The difference in the results by sample may make sense. That is, the supervisors
defined by occupation codes should be those who directly monitor the workers who
produce goods or services, as opposed to managers overseeing supervisors. Therefore,
shirking on the part of these supervisors could lead to an increase in shirking among non-
supervisory workers, who have the most direct effect on the product of the firm (less
intensive monitoring leads to a lower probability of detection, lowering the expected cost
of shirking). It may therefore be wise for the firms to focus the use of efficiency wages
on this subgroup.

Recall that the regressions performed in Tables 3 through 7 may be used to test
another hypothesis regarding efficiency wages. Previous studies (Neal, 1993, for
example) have explored whether the inter-industry wage differential can be explained by
monitoring differences across the industries.

Figures 1 through 3 display the coefficients on the industry dummies from the
samples examined in Tables 3 through 7. Figure 1 shows the coefficients from model

estimated in column 2 of Table 3, with and without the supervisor-to-staff ratio. ””8 So,

 

’7 See Appendix C to match industries with the codes used in the figures.
’8 The coefficients are expressed as deviations from the sample mean (there is no omitted industry),
following the procedure described in Suits (1982).

70

for example, wages in the construction industry (coded “CON”) are about 15 percent
higher than average wages, even after introducing a control for the level of supervision.
The coefficients on the industry dummies for the piece rate and commission workers are
shown in Figure 2 (based on the regression from column 2 of Table 4). Finally, the
results for the sample of supervisors are shown in Figure 3 (from the regression in
column 4 of Table 7). The three figures tell a similar story. Clearly, the variation in
wages across industries cannot be explained by monitoring differences. This conclusion
is reinforced in Neal (1993), who uses a different proxy for supervision, but similarly
finds that “controls for frequency of supervision do not affect estimates of industry wage
premiums.”l9

Taken together, the results from the previous two sections do not provide support
for the widespread applicability of the shirking model of efficiency wages. At most,
piece rate and commission workers and front-line supervisors are offered this form of

incentive to reduce shirking. The next section explores alternative methods that may be

used to solve monitoring problems.

Evidence on the Use of Incentives

A lack of evidence pointing toward the widespread applicability of the efficiency
wage model may not mean that extracting effort is costless, however. It could be that
other pay schemes are used to accomplish this task. Incentive plans, whether they are

based on individual or group goals, may be able to prevent workers from shirking by

 

'9 Neal (1993), p. 416.

71

giving them a share of the value of increases in their productivity. For example, profit
sharing plans linking a worker’s compensation to the profitability of the firm have been
proposed as a solution. While research on profit sharing plans indicates that its adoption
increases firm-level productivity,2O there is a reason to be skeptical of the ability of the
plans to provide a financial incentive to individuals. The problem is that, as the number
of workers in a firm increases, the incentive for any one worker to enhance profits is
reduced because any gain that a worker creates must be shared with the rest of the
workers. Because of this, profit sharing plans may not be as effective as other incentives
are at providing a substitute for supervision.

There are a variety of other incentive plans that may act as a substitute for
supervision. These include group incentive plans such as gain sharing plans,21 in which
the incentive pay is based on well-defined goals instead of profits, and individual
incentives, which can range from sales goals to more subjective performance ratings on
which to base pay.

There are two incentive-plan variables available for use here: a self-report of
whether profit sharing is “made available by the employer,” and the presence of bonuses
based on job performance. The profit sharing indicator has two potential flaws. First,
workers may report that profit sharing is not “made available” for cases in which a profit

sharing plan is in place, but there are no profits to be distributed.22 Second, many pay

 

2° Kruse (1993) finds a 4 percent increase in firm-level productivity due to the adoption of a profit
sharing plan using longitudinal data. This finding is consistent with studies using other (usually cross-
sectional) data sets.

2’ Case-study evidence on the success of a gain sharing plan can be found in Hansen (1997).

22 This notion is introduced in Kruse (1995).

72

schemes that are referred to as “profit sharing plans” may have very little to do with
profits, but instead add a fixed percentage of a worker’s contributions to a 401(K) plan
(for example). In addition, a distinction exists between cash and deferred profit sharing
plans for which this study cannot account. Since cash plans involve more immediate
rewards, they are predicted to have a larger effect on behavior. While it is unfortunate
that the data cannot determine which form of profit sharing plan is being reported, the
predicted direction of the impact on behavior is the same across plan types, with only the
magnitude of the effect differing.

The performance-based bonus indicator could represent a number of specific
incentives. However, the question from which this indicator comes (see page 62 for the
text of the question) separately lists piece rate pay, commissions, bonuses (based on job
performance), stock options and tips, so it may be expected this indicator does not
represent the other forms of pay listed in the survey question because it is defined to refer
to bonuses only.

The dependent variable in the following specifications is an indicator for the
presence of the incentive plan, with separate regressions run for performance-based
bonuses and profit sharing plans. Because the dependent variable is binary, the logit

model is used for estimation:

X+S
613 r

Prob[incentive plan = l] =
1 + cm i 87

73

where X includes the same controls as in the wage equations, adding the log of the
hourly wage, and removing the region dummies.23 S represents the supervisor-to-staff
ratio.

The explanatory variables are intended to control for variation in the provision of
incentive plans based on the position of the employee in the firm (occupation indicators,
tenure, experience, and wages) or in the technology involved in the job (through industry
and firm size indicators). In addition, some demographic characteristics (male, white and
married) are included to control for the possibility that some groups are given preference
in the provision of incentives.

The key hypothesis is this: if incentive pay plays a role in reducing monitoring
problems, the presence of profit sharing plans or performance-based bonuses should be
associated with a lower supervisor-to-staff ratio, or y<0. Once again, separate analyses
are performed for the supervisor and worker samples.

Table 8 displays the results of this model estimated using both profit sharing
plans and performance-based bonuses as the dependent variable, and for each of the five
samples that have been formed.24 Only 7 is reported in Table 8, and each cell represents

a different regression. As the first two rows show, there is very little relationship

 

2" The inclusion of the region dummies does not affect the results.

2" For the samples of supervisors, where panel data are available, fixed effects logit models are estimated.
In all cases, a Hausman test cannot reject the hypothesis of no heterogeneity bias. For the broadly
defined supervisors and supervisors responsible for pay or promotion, the coefficients from the fixed
effects logit model are quite similar to the coefficients from the logit model. For example, the fixed
effects logit coefficient corresponding to row 3 of column 1 is -.340 with a standard error of .116.
However, because of the small sample size, the fixed effects logit coefficients for the supervisors defined
by occupation codes are very imprecisely estimated (the fixed effects logit coefficient corresponding to
row 5 of column 1

is -1.791 with a standard error of 1.428).

74

between incentives and supervision for the worker samples. The only statistically
significant result is found for workers (not piece rate or commission) and the presence of
profit sharing plans. However, the magnitude of this result is small. A movement from
one standard deviation below the mean of the supervisor-to-staff ratio to one standard
deviation above the mean only decreases the probability of reporting a profit sharing plan
from 32.8 percent to 29.5 percent.

These findings are consistent with Osterman (1994), who uses a dummy variable
for little or no supervision as a proxy for the extent of monitoring, and finds that non-
supervisory workers are not more likely to be offered profit sharing plans or group
incentives in the absence of supervision.

This study goes beyond Osterman (1994) by looking at the relationship between
incentives and monitoring for supervisors. Although there is no relationship between
monitoring and profit sharing plans, rows 3 through 5 of the first column show that
performance-based bonuses are strongly related to supervision across all three samples.25
Recall that these specifications may be less biased than the wage specifications presented
in the previous sections. That is, the provision of a bonus should not lead to substitution
between the factors of production (supervisors and workers) in the same way that wage
changes do. This may lead one to be more confident of the results from Table 8 relative
to the wage specifications in earlier tables.

The results in the performance-based bonus specifications for supervisors are not

only statistically significant, but the magnitude is also large enough to represent a

 

2’ The manager-to-supervisor ratio (not shown in Table 8) has no significant relationship with report of
profit sharing plans of performance-based bonuses.

75

meaningful effect. The weakest effect is for the broad definition of supervisors (in row
3), for which an increase from one standard deviation below the mean of the supervisor-
to-staff ratio to one standard deviation above the mean reduces the probability of
reporting a performance-based bonus from 21.4 percent to 14.1 percent. The strongest
relationship is seen for the supervisors defined by occupation codes (row 5), for whom
the aforementioned change in the supervisor-to-staff ratio decreases the probability of
reporting a bonus from 29.3 percent to 12.6 percent.26

Taken together, the results from the previous three sections indicate that
supervisors, and not the workers themselves, are given the incentive to reduce the level
of shirking as monitoring becomes more difficult. This is seen in the fact that front-line
supervisors show evidence of earning efficiency wages, and that all of the samples of
supervisors are more likely to be offered a performance-based bonus as they monitor
more workers.

Rebitzer (1995), however, provides a potentially important criticism of studies
such as this. That is, in estimating the relationship between supervision and wages (or
the provision of incentive plans), one typically cannot account for other human resource
practices -- or other firm characteristics -- that may also affect productivity.27 An

example of this is employee screening. As more resources are devoted to weeding out

 

2" Omitted ability bias does not appear to be an issue in this section, as the addition of the AFQT scores
does not significantly impact the results of Table 8. This is also reflected in the fact that Hausman tests
reject the need for individual fixed effects where panel data are available.

27 There is some evidence that film heterogeneity may not strongly affect the results. Wage regressions
are run without firm size or industry indicators. Then firm size variables are added, followed by one-
digit industry and then two-digit industry indicators. There is little change in the coefficient on the
supervisor-to-staff (or manager-to-supervisor) ratio across the samples. For example, for supervisors
identified by occupation codes, the coefficient evolves as follows: -.027, -.025, -.032, -.043.

76

shirkers, less effort is required to monitor the behavior of workers. Rebitzer states that
this omitted variable problem “will lead to underestimates of the trade off between wages
and supervision.”28 However, this is only true if the omitted human resource practices
are associated with lower wages (they are substitutes for efficiency wages) and, more
importantly, the human resource practices are substitutes for supervision. To illustrate
the point, consider the following model:29

ln(wage) = X13 + HRS + 87 + 8
where X is composed of the standard regressors, HR represents resources devoted to
screening, etc. and S is the supervisor-to-staff ratio. Recall that the efficiency wage
model predicts that y<0. If it is assumed that screening acts as a substitute for
supervision and it is omitted from the regression, y is biased upward. In this case, as
Rebitzer suggests, it is more difficult to observe a wage-supervision trade off.

However, it seems equally plausible that human resource practices and
supervision act as complements. For example, if a firm is very concerned about quality
(e. g. it manufactures computer processors), it may screen employees heavily during the
hiring process and supervise work closely. In this case, the omission of the screening
variable biases y downward, in favor of finding the prediction of the efficiency wage
model.

The results from this paper can therefore be interpreted given two scenarios: HR

and S are substitutes, or HR and S are complements. If Rebitzer is correct, and HR and S

 

2“ Rebitzer (1995), p. 111.

29 The discussion in Rebitzer (1995) focuses on the relationship between supervision and wages.
However, the same arguments may be made about the relationship between supervision and the provision
of incentive plans.

77

are substitutes, then the findings from the wage regressions may be called into question.
That is, little evidence of a relationship between wages and supervision is found, but this
may be because the true relationship is masked by omitted variable bias. However, the
direction of this bias further supports the most interesting finding from the incentive plan
regressions, showing a strong relationship between the size of the supervisor’s work
group and the provision of a performance-based bonus.

Alternatively, one can examine the case in which employee screening and
supervision are complements. Under this assumption, the findings from the wage
equations are reinforced. That is, there is little evidence in favor of the efficiency model
even when the statistical model is biased toward that finding. However, the results for
the provision of incentives may be called into question, although the magnitude of the

coefficient on the supervisor-to-staff ratio and its statistical significance are reassuring.

Conclusions and Discussion

The goal of this study is to determine if compensation policies -- efficiency
wages, profit sharing plans or performance-based bonuses -- act as substitutes for
supervision. According to the results, the shirking model of efficiency wages is only
relevant for piece rate and commission workers and front-line supervisors. An attempt is
made to remove bias due to omitted worker ability, but it does not produce evidence that
is more favorable towards the shirking model for the large sample of workers.

In addition, it does not appear that workers are offered incentive plans as a

substitute for supervision. However, the evidence does support the idea that supervisors

78

are more likely to have a performance-based bonus as a part of their compensation
package when they supervise a larger group of workers. This may be an economically
sound approach by firms, as offering only supervisors a bonus for increases in
productivity may involve a smaller total cost to the firm than offering an incentive (or
higher wage, in the case of the efficiency wage model) to every employee. This assumes,
of course, that supervisors are equal to the task, and are able to monitor effectively
enough to prevent their subordinates from shirking.

There may be another reason for the fact that the strongest results are found for
the supervisory samples; it may be that supervisors are more difficult to monitor than
workers. That is, workers frequently produce output that is observable, whereas
supervisors, who exert effort in the form of increased monitoring, have less observable
output. If this makes it more costly to monitor supervisors, then it is reasonable for
managers to substitute toward other methods in order to prevent them from shirking.

To get a sense of the macroeconomic implications of these results, one can
incorporate supervisory bonuses into the shirking model. The bonuses for supervisors
can be imposed on the model exogenously by affecting the probability that workers are
caught shirking. This is seen in the no-shirking condition as modeled in Shapiro and
Stiglitz (1984):

w>=w' +e+e(a+b+r)/q
where w = wage, w ’ = unemployment benefit, e = worker effort, a = probability of re-
employment, b = exogenous separation rate, r = interest rate, and q = probability of being

caught shirking.

79

In this model, a single firm has the incentive to increase wages above the market-
clearing level in order to create a cost of losing one’s job, preventing workers from
shirking. Because the model contains identical firms, if one firm has the incentive to
increase wages, all of the others do as well. The fact that economy-wide wages exceed
the market-clearing level creates involuntary unemployment (an appealing aspect of the
model to some). In the end, workers are motivated by a fear of suffering through a
period of unemployment.

Providing supervisors with the incentive to monitor more closely (i.e. firm
performance is increased by increased monitoring, so the supervisor monitors more
closely in order to receive a bonus) increases q for workers. Notice that in the equation
above, an increase in q reduces the size of the wage premium that is paid to workers.

The reasoning for this is as follows: the higher probability of job loss can be offset by an
increase in aggregate demand, which reduces involuntary unemployment in the economy.
In other words, because the wage premium is smaller, aggregate demand is

reduced less than it is in the absence of bonuses paid to supervisors. Therefore,
involuntary unemployment is reduced by the use of the bonuses. In fact, it can be shown

that as q approaches one, involuntary unemployment goes to zero.30

 

’0 It could therefore be true that a tradeoff exists between wages and supervision for workers, but this
study does not observe it because it cannot control for the presence of supervisory bonuses (i.e., there are
fluctuations in q controlling for the supervisor-to-staff ratio). One attempt is made to approximate this,
however. The percentage of supervisors reporting a performance-based bonus is calculated for each two-
digit industry. This is then compared to the industry wage differential from a standard wage equation
using the worker sample. The correlation of these two variables is .06, showing that high-wage industries
offer more supervisory bonuses. This contradicts the idea that supervisory bonuses act as substitutes for
worker’s wages.

80

It is also interesting to note the lack of a relationship between supervision and the
presence of profit sharing plans. This is in accordance with much of the theoretical
discussion on profit sharing plans, in that the free rider problem inherent in these
schemes may limit the financial incentive to monitor oneself. This suggests that profit
sharing plans may be adopted (or maintained) for reasons other than alleviating

monitoring difficulties (e.g., the desire to reduce turnover).

Table 1
Summary of Previous Studies on the Shirking Model of Efficiency Wages

 

 

 

 

 

 

 

 

 

 

 

 

Author(s) and Year Monitoring Proxy Data Findings
Leonard (1987) Supervisor-to-stafi’ ratio 1982 survey of 200 plants in the Inclusion of supervision
high-technology sector of one variable does not explain the
state dispersion of wages; does not
support shirking model.
Fitzroy and Kraft Supervisor-to-staff ratio 65 West German metalworking Supervision is not related to
(1988) firms wages.
Brown and Medoff Examines piece rate workers, BLS Industry Wage Survey Establishment size-wage effect
(1989) who should require less exists for piece rate workers,
monitoring. and therefore cannot be
explained by monitoring
difficulty.
Groshen and Krueger Supervisor-to-stafi’ ratio BLS 1985 Hospital Industry Wages fall as supervision
(1990) Wage Survey of 300 hospitals increases for 1 of the 4
occupations studied.
Krueger (1991) Franchised vs. company-owned 1982 National Institute for Work Supports shirking model; pay is
outlets and Learning higher in company-owned
outlets, where monitoring is
more difficult
Kruse (1992) Frequency of supervision. Uses 1980 Survey of Job Supports shirking model; wages
continuous and categorical Characteristics from the Center increase as supervision
variables. for Survey Research decreases.
Neal (1993) A categorical response to: “How 1977 Supplement to the PSID Industry wage premiums are not
often does your supervisor affected by inclusion of the
check up on your work?” supervision variables.
Arai (1994) Measure of “autonomy"; a Swedish Level of Living Survey Supports shirking model for
combination of the possibility 1981 private sector workers, but not
of varying work effort, the use for public sector workers.
of a punch card and flex time.
Osterman (1994) Dummy variables for little or no 1992 survey of 875 private- Both proxies support the
supervision and whether or not sector establishments with 50 or shirking model.
the worker has discretion over more employees
method.
Rebitzer (1995) Indicator for contractors having 1990 survey of contract Supports shirking model.

 

to report safety problems to the
host plant management

 

personnel employed in the
petrochemical industry.

 

 

81

 

Table 2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Descriptive Statistics
Workers (Not Piece Rate and Supervisors Supervisors
Piece Rate or Commission Supervisors Responsible Defined by
Commission) Workers for Pay or Occupation
Promotion Codes
Supervisor-to- .189 .173 .354 .269 .194
Staff Ratio (.233) (.226) (.307) (.279) (.211)
Manager-to- --- --- .l 73 . l 91 . 141
Supervisor Ratio (.214) (.244) (.183)
Bonus .144 .295 .191 .257 .241
(.351) (.457) (.393) (.437) (.428)
Profit Sharing .337 .370 .343 .388 .417
(.473) (.484) (.475) (.487) (.493)
Hourly Wage 7.49 8.34 8.02 9.23 7.88
(4.02) (4.85) (4.51) (5.21) (4.06)
Education (Years) 13.00 13.03 13.26 13.75 12.89
(2.58) (2.31) (2.61) (2.56) (2.27)
Tenure (Years) 3.37 3.00 3.55 3.97 4.23
(3.30) (3.07) (3.10) (3.20) (3.44)
Male .564 .690 .585 .610 .677
(.496) (.463) (.493) (.488) (.468)
White .730 .795 .763 .813 .777
(.444) (.404) (.425) (.390) (.417)
N in 1988 --- --- 1,938 834 207
N in 1989 --- -- 1,892 845 277
N in 1990 3,329 332 1,770 624 238

 

 

 

 

 

 

Note: Standard deviations are in parentheses.

82

 

Table 3

OLS Estimates of the Effect of Supervision and Firm Size on
Wages for Workers (Not Piece Rate or Commission)

 

 

 

 

 

 

 

 

 

(1) (2) (3)

Supervisor-to-Staff Ratio -- -.006 -.007
(.005) (.005)
Standardized AFQT Score --- --- .042**
(.008)
Establishment Size .034** .033** .032**
(.004) (.004) (.004)

N 3,329 3,329 3,329

 

 

Note: All variables are in log form. Standard errors are in parentheses. *
denotes significance at the ten-percent level. ** denotes significance at the
five-percent level. Other independent variables are: tenure, tenure-squared,
experience (actual), experience-squared, and dummies for male, white,
married, northeast, north central, south, two-digit industry and one-digit
occupation.

83

Table 6
Estimates of the Effect of Supervision and Firm Size on Wages

for Supervisors Responsible for Pay or Promotion

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimation Method Fixed Effects OLS
(1) (2) (3) (4) (5)
Supervisor-to- --- -.01 1 --- .007 .008
Staff Ratio (.011) (.015) (.015)
Manager-to- --- --- -- -.005 -.006
Supervisor Ratio (.012) (.012)
Standardized --- --- --- --- .005
AFQT Score (.020)
Establishment Size -.006 -.009 .037** .038** .038"
(.008) (.008) (.010) (.010) (.010)
p-value from .00 .00 --- --- ---
Hausman Test
N 2,303 2,303 588 588 588

 

Note: All variables are in log form. Standard errors are in parentheses. * denotes significance at

the ten-percent level. ** denotes significance at the five-percent level. Other independent

variables are: tenure, tenure-squared, experience (actual), experience-squared, and dummies for
male, white, married, northeast, north central, south, two-digit industry and one-digit

occupation.

86

 

Table 8

Logit Estimates of the Relationship Between the Supervisor-to-Staff Ratio
and the Presence of Performance Based Bonuses and Profit Sharing Plans

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable: Performance-Based Bonus Profit Sharing Plan
Sample Coefficient on the Supervisor-to-Staff Ratio
Workers -.061 -.066*
(Not Piece Rate (.046) (.036)
or Commission) [-.007] {-.014]
N=3,289 N=3,329
Piece Rate Workers -.115 .016
and Workers Paid (.145) (.158)
on Commission {-.022] [.004]
N=309 N=280
Supervisors -.215** -.030
(.034) (.030)
{-.031] {-.006]
N=5,600 N=5,595
Supervisors Responsible for -.281** .007
Pay or Promotion (.048) (.044)
{-.049] [.002]
N=2,293 N=2,300
Supervisors Defined by -.480** .061
Occupation Codes (.108) (.092)
[-.076] [.014]
N=685 N=717

 

 

Note: All variables are in log form. Standard errors are in parentheses. Marginal changes in probability
are reported in brackets. * denotes significance at the ten-percent level. ** denotes significance at the
five-percent level. Other independent variables are: ln(wage), tenure, tenure-squared, experience,
experience-squared, years of education, establishment size, and dummies for male, white, married,

multiple-site firm, an interaction between multiple-site firm and more than 1,000 total employees, two-
digit industry and one-digit occupation.

88

 

 

 

J
.4 am H... m MTHJ JV
uaav namm can mm Ina avail
mmmm mmmnwmmmmmmmmmmmmmmmmmmmmmm
mud-
as-
m m.

 

 

 

 

 

 

 

 

 

 

 

 

sea assessoaeasm menses-
eeaeoea as .380 35

 

 

 

mNd

.Cunnnq Lumptq at]; no mapmaoo

 

 

€8.83 mo 295m ofi com
832695 was magenta owe? homegrown:
u «Ema

89

 

JOHJ

INTI

SHEJ

HIVJX

TRIM

MON

8110

11m

WOO

(100M
HHI’I
881121
ILLEIJ
WEIRD
NIXJ
HJVJ
'1qu
X1].
([01.
(IOOaI
NOD
MIN
9"

Oil 8.1.
DEI'ITI
HZ)

him
OLOHJ

 

 

 

med-

 
 

 

 

 

.r was

 
 

1. mad

 

.1 mm...

.. mbd

i mad

 

 

 

 

possessesoatoeew masses-
:ommmtoanw com 35:00 628

 

 

 

[mama armpit; all; no mapmaog

 

 

Eco—83 :oamESoU can 8nd 805 2: com
:oEansm was 888055 owe? DEESEBE
N PEER

90

 

808d

1N8

S888
8IV¢18

1.88

NON

80(1
WOO
NV8.I.
OLOHJ
Oil 8.1.
D’J’IEI
HDVW
'IJMI
NOLS
N8fl.iI
(IOOM
88.08
8L'ild
WEIRD
N188
8JV¢I
188V
80.].
(100.!
N00
NIW
9V

 

 

 

med-

1.. WAVI

.r mnd-

 

. . mad-

 
 
 
 
 

...
..v

n.”

. .

.o. a. r e
.u.

...

...

“on

4..
.o.
.J
H. l I I
c
o .
u.
to.
.3
s.
.

.. 2.

 

 

 

 

 

mad
ovum uofltonsmcfiuowacaz $5.205-

 

 

:oEEonfi—m com .8280 oZE

 

Kununq [armpit] at“ no mapmaog

 

 

mouoo nogasooo 3 womueog 883.895 mo oaaam 2: com
Sarcasm can Bazooka ”.me Emzuflififi
m oaswE .

91

BIBLIOGRAPHY

BIBLIOGRAPHY

Akerlof, George A. 1986. “Labor Contracts as Partial Gift Exchange.” In George A.
Akerlof and Janet L. Yellen, eds, Efliciency Wage Models of the Labor Market.
Cambridge: Cambridge University Press, pp. 66-92.

Arai, Mahmood. 1994. “Compensating Wage Differentials versus Efficiency Wages:
An Empirical Study of Job Autonomy and Wages.” Industrial Relations 33: 249-
62.

Brown, Charles and James Medoff. 1989. “The Employer Size-Wage Effect.” Journal
of Political Economy 97: 1027-1059.

Fitzroy, Felix and Kornelius Kraft. 1988. “Efficiency Wages and Supervision in the
Firm: A Direct Test with FRG Microdata.” Draft. December.

Groshen, Erica L. and Alan B. Krueger. 1990. “The Structure of Supervision and Pay in
Hospitals.” Industrial and Labor Relations Review 43: 134-S - l46-S.

Hansen, Daniel G. 1997. “Individual Responses to a Group Incentive.” forthcoming in
Industrial and Labor Relations Review.

Krueger, Alan B. 1991. “Ownership, Agency, and Wages: An Examination of
Franchising in the Fast-Food Industry.” Quarterly Journal of Economics 106: 75-
102.

Krueger, Alan B. and Lawrence H. Summers. 1988. “Efficiency Wages and the Inter-
lndustry Wage Structure.” Econometrica 56: 259-93.

Kruse, Douglas L. 1992. “Supervision, Working Conditions, and the Employer Size-
Wage Effect.” Industrial Relations 31: 229-49.

Kruse, Douglas L. 1993. Profit Sharing: Does It Make a Diference? Kalamazoo, MI:
W.E. Upjohn Institute.

Kruse, Douglas L. 1995. "Profit Sharing and the Demand for Low-Skill Workers."
Unpublished working paper.

Lazear, Edward P. 1981. “Agency, Earnings Profiles, Productivity, and Hours
Restrictions.” American Economic Review 71: 606-20.

92

93

Lazear, Edward P. 1996. “Performance Pay and Productivity.” NBER Working Paper
5672.

Leonard, Jonathan S. 1987. “Carrots and Sticks: Pay, Supervision, and Turnover.”
Journal of Labor Economics 5: $136-$152.

Neal, Derek. 1993. “Supervision and Wages Across Industries.” The Review of
Economics and Statistics 75: 409-17.

Osterman, Paul. 1994. “Supervision, Discretion, and Work Organization.” American
Economic Review 84: 380-84.

Rebitzer, James R. 1995. “Is there a trade-off between supervision and wages? An
empirical test of efficiency wage theory.” Journal of Economic Behavior and
Organization 28: 107-29.

Shapiro, Carl and Joseph E. Stiglitz. 1984. “Equilibrium Unemployment as a Worker
Discipline Device.” American Economic Review 74: 433-44.

Suits, Daniel B. 1984. “Dummy Variables: Mechanics v. Interpretation.” The Review of
Economics and Statistics 66: 177-180.

Weitzman, Martin L. and Douglas L. Kruse. 1990. "Profit Sharing and Productivity."
In Paying for Productivity: A Look at the Evidence, Alan S. Blinder, ed.
Washington, DC: Brookings Institution: 95-140.

APPENDICES

Appendix A

The broadest sample of supervisors is composed of those responding “yes” to:
“(Do/Did) you supervise the work of other employees, or
tell them what work to do, on a day-to-day basis?”
The supervisor-to-staff ratio for the supervisor samples is derived from the follow-up
question:
“About how many people (do/did) you supervise on a day-
to-day basis?”
The second sample of supervisors is composed of those responding “yes” to another

follow-up question:

“(Are/W ere) you responsible for deciding their rate of pay
or promotion?”

In 1988 and 1989, the possible responses are “Yes” and “No.” In 1990, the responses are
expanded to: “Yes - Full Responsibility,” “Yes - Partial Responsibility,” and “No.” The
two forms of “Yes” are combined for the purpose of this study. The results are not
affected if only the 1988 and 1989 data are used.
All of the above questions appear in the survey from 1988 through 1990.
The supervisor-to-staff ratio for the worker samples, and the manager-to-supervisor ratio
for the supervisor samples is taken from the question:

“Besides yourself, for how many other people (does/did)

your supervisor also serve as the immediate supervisor on a

day-to-day basis?”

This question is only asked in 1990.

94

Appendix B
1980 Census of Population Occupation Codes Used to Identify Supervisors

243 Supervisors and proprietors, sales occupations

303 Supervisors, general office

304 Supervisors, computer equipment operators

305 Supervisors, financial records processing

307 Supervisors; distribution, scheduling, and adjusting clerks
413 Supervisors, firefighting and fire prevention occupations
414 Supervisors, police and detectives

415 Supervisors, guards

433 Supervisors, food preparation and service occupations
448 Supervisors, cleaning and building service workers

456 Supervisors, personal service occupations

477 Supervisors, farm workers

485 Supervisors, related agricultural occupations

494 Supervisors, forestry, and logging workers

503 Supervisors, mechanics and repairers

553 Supervisors; brickmasons, stonemasons, and tile setters
554 Supervisors; carpenters and related workers

555 Supervisors; electricians and power transmission installers
556 Supervisors; painters, paperhangers, and plasterers

557 Supervisors; plumbers, pipefitters, and steamfitters

558 Supervisors; n.e.c.

613 Supervisors, extractive occupations

633 Supervisors, production occupations

803 Supervisors, motor vehicle operators

843 Supervisors, material moving equipment operators

863 Supervisors, handlers, equipment cleaners, and laborers, n.e.c.

95

AG

CON

FOOD
TOB
TEX

PAPR
PRIN
CHEM
PETR
RUBR
LTHR

woop
FURN
STON
MTL
MACH
ELEC
TR EQ
PHOTO

TRAN
COM
UTIL

DUR
NON

FIRE
RPAIR
PERS
ENT
PROF

Appendix C
Two-Digit Industries Used in the Figures and Regressions

Agriculture, Forestry, and Fisheries
Mining
Construction

Manufacturing, Nondurable Goods
Food and kindred products

Tobacco manufactures

Textile mill products

Apparel and other finished textile products
Paper and allied products

Printing, publishing, and allied products
Chemicals and allied products

Petroleum and coal products

Rubber and miscellaneous plastics products
Leather and leather products

Manufacturing, Durable Goods
Lumber and wood products, except furniture

Furniture and fixtures

Stone, clay, glass, and concrete products

Metal industries

Machinery, except electrical

Electrical machinery, equipment, and supplies
Transportation equipment

Professional and photographic equipment, and watches

Transportation, Communicationsfiand Other Public Utilities
Transportation

Communications

Utilities and sanitary services

Wholesale Trig
Durable goods
Nondurable goods

Retail Trade

Finance, Insurance, and Real Estate
Business and Repair Services

Personal Services

Entertainment and Recreation Services
Professional and Related Services

96