NBER WORXING PAPER SERIES

HUMAN RESOURCE POLICIES AND

UNION-NONIJNION PRODUCTIVITy DIFFERENCES

Steven C. Allen

Working Paper No. 2744

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
October 1988

This paper was presented at the Academy of Management meetings in Anaheim,

August 9, 1988. I have received helpful comments from
Barry Cerhart, John
Fossum, Dan Hamermesh, Barry Hirsch, and Brian Bemmels. This paper is part
of NBER's research program in Labor
Studies. Any opinions expressed are
those of the author not those of the National Bureau of Economic Research.
 NBER Working Paper #2 744
October 1988

HUMAN RESOURCE POLICIES AND

UNION-NONUNION PRODUCTIVITY DIFFERENCES

ABSTRACT

Many researchers in both economics departments and business schools

recently
have become interested in examining how much of an effect human resource

decisions and policies have on firm performance. This paper surveys the

literature on unionism and productivity and discusses its implications for

future research on more general issues. The main focus is on (I) conclusions

as to whether unions raise or lower productivity and (2) procedures used to

identify the channels through which unions affect productivity.

The studies of unions and productivity have documented

large productivity
differences between seemingly comparable union and nonunion establishments. In

many cases unionism is associated with higher productivity, especially when

unionized firms are in a competitive environnent. However, the mechanisms

responsible for union-nonunion productivity differences in each study remain

poorly understood, either because detailed information on how unions affected

company decisions was not available or because the available information

produced inconclusive results. These conclusions suggest that human resource

policies can have a very large effect on financial outcomes, but our ability to

estimate the magnitude of that effect for a particular policy is currently very

limited.

Steven C. Allen
Department of Economics and Business
North Carolina State University
Box 8110
Raleigh, NC 27695-8110
 In mainstream economic theory, unions are viewed mainly as
wage-raising
institutions. Managers respond to higher wages by cutting back or shutting

down operations that are no longer profitable. Beyond this, they take steps to

offset the cost of higher wages by substituting capital for labor and

increasing the skill level of the work force through adjustments in training

and hiring procedures. The result is greater observed productivity in union

than nonunion establishments. Despite the productivity gains, econonists view

union-generated wage increases as socially undesirable because in theory the

cost of higher wages is never completely offset by additional productivity,

resulting in greater unit costs and reduced output. Unions further reduce

output and productivity through strikes and restrictive work practices.

This itandard view has been challenged over the last dozen years by work

initiated by Professors Richard Freeman and James Medoff of Harvard University.

Drawing from Hirschman's (1970) "exit-voice" model, Freeman and Medoff (1984)

argued that unions should be viewed as institutions that give workers voice at

the workplace. Union voice can result in increased productivity through a

variety of mechanisms - - reduced turnover, more informal training and

information sharing among workers, better communication between workers and

management, and improved morale. It can also have a "shock effect" on

management, the result of which is that decision making becomes more sensitive

to worker reactions (Slichter, Healy and Livernash, 1960).

There is now a sizable empirical literature examining the effect of unions

on productivity. This literature is highly relevant for understanding the

impact of human resource policies on firm performance for two reasons. First,

despite the sharp drop in union organizing success, the question of what is

likely to happen to a f inn if it becomes unionized remains a very important

issue. Faced with the threat of unionization, managers must judge whether the
 2

costs of becoming organized are greater than the costs of attempting to remain

nonunion. Today in the private sector it is quite clear that almost all

managers believe the costs of becoming organized are greater. However, there

is evidence that managerial views on this question were quite different in the

l940s (Kochan, Katz, and McKersie (1986)) and one should not totally discount

the possibility that these views may change again in the future.

Second, many of the studies in the literature on unions and productivity

(Ui') attempt to identify the key mechanisms causing any observed productivity

differences, a central issue in the new literature on human resources and firm

performance (HRFP). Because the focus is almost always on human resource

practices, the UP studies end up asking exactly the same types of questions as

the HRFP studies, as one can see from the surveys of the HRFP literature in

Kleiner j. (1987). Not only does the UP literature provide a set of

"preliminary results," but also it points out the difficulties in obtaining

useful conclusions about the impact of human resource practices.

After a suary of the key mechanisms through which unions affect

productivity, this paper surveys the UP literature. The survey focuses on (1)

conclusions as to whether unions raise or lower productivity and (2) studies

that attempt to identify channels through which unions affect productivity.

The last section of the paper assesses what is currently known about unions and

productivity and discusses useful future research directions for both the UP

and HRFP literatures.

 3

I. HOW UNIONS INFLUENCE PRODUCTIVITY

One important route through which unions influence productivity has

already been noted in the introduction: higher union wages create an incentive

to find substitutes for labor. This is but one of many channels through which

a union can influence the productivity of an establishment. A newly unionized

establishment is likely to witness across-the-board changes in its human

resource policies, especially those dealing with hiring, promotion, layoffs,

training, and planning. The adjustments in compensation practices will be much

more complex than a simple change in the overall wage level. Unionized firms

tend to have relatively small occupational and geographical wage differentials.

They also tend to spend a larger share of their compensation package on

employee benefits and a smaller share on pay linked to individual or group

performance. Management behavior also changes under unionism. In the short

run, one must consider the initial shock effect of union organization, the

constraints spelled out by the current contract, union recourse under grievance

procedures, and the desire for peaceful settlement of future contracts. Over a

longer time horizon, policies toward union cooperation or union avoidance often

turn out to be a key element in corporate strategy.

Some of the most salient features of unionism that are believed to

influence productivity are summarized in Table 1. This table is not intended

to be all inclusive, nor does each factor listed in the table apply to every

organization. The table should be viewed as merely a reflection of how today's

economists think unions affect productivity. Although there is much more to

this thinking than the simple textbook tale about factor substitution, it is

useful to emphasize the distinction between productivity changes arising from

higher wage levels and productivity changes arising from sources other than
 4
wage levels. This is done in the table by indicating whether the main source

of a particular channel of union influence is wage (W) or nonwage (N). Because

one would naturally expect unions to raise productivity through higher wage

levels, every UP study attempts to control for wage-induced changes in capital

intensity and skill levels. HRFP studies will also have to make this

distinction whenever the HR practice under examination is correlated with the

wage level.

Both the wage and nonwage channels of union influence should result in a

workforce with an upgraded skill level. Higher union wage levels give managers

the incentive to raise hiring standards and cut back on unskilled labor, With

the higher wage there is also generally a queue of workers wanting jobs,

enabling firms to make these adjustments without extra recruiting expenditures.

Higher union wages reduce turnover, thereby increasing the incentive to spend

more on training. Freeman (1980a) has shown that the overall reduction in

turnover under unionism is much larger than one would expect from higher union

wages. He attributes this turnover reduction to the greater voice unions give

employees in workplace decision making. With voice at the workplace, union

members are less likely to quit their jobs when they are unhappy with working

conditions or human resource policies, leading to even more training under

unionism. Another factor leading to more investments in training under

unionism is the continued reliance in sectors such as construction on an

occupational mix molded by craft traditions, with apprenticeship remaining the

key route of entry into the craft. A final reason to expect workers in

unionized establishments to be more highly skilled than their nonunion

equivalents is the greater cost of false positive errors in hiring decisions.

 S

Managers are likely to raise their hiring standards in order to offset union-

imposed restrictions on job assignments, layoffs and dismissals.

If work effort is a simple matter of carrots and sticks, then one would

naturally expect unions to be a negative factor. The use of seniority in

making promotion decisions in unionized establishments is well known.

Many
nonunion establishments base their promotions on seniority as well (Abraham and

Medoff, 1985). This does not mean that union and nonunion carrots for

promotion are identical. Compensation packages under unionism tend to have

highly compressed occupational wage differentials (Freeman, 1980b, 1982). The

spread in compensation among different jobs is further compressed by the large

share of the package which goes to health insurance, retirement plans, and paid

time off (Freeman, 1981). Another factor that may diminish the size of the

union carrot is the less frequent use of merit pay and other incentive pay

mechanisms, although this requires the controversial and empirically

questionable (Ehrenberg and Milkovich (1987)) assumption that such pay systems

actually influence employee behavior in the desired fashion.

Two other key mechanisms through which unions are believed to reduce work

effort are work rules and grievance procedures. Work rules, both written and

unwritten, result in too many jobs and too much break time, accompanied by

restrictions on output levels and management's ability to get the job done in

the most efficient manner. Crievances undermine productivity by reducing the

penalty for shirking and by protecting incompetent and dishonest workers.

Once one goes beyond this no-carrots/no-sticks scenario to examine other

factors believed to influence work effort, a different picture emerges. Some

of the very factors listed above as decreasing productivity under unionism from

a carrots/sticks framework turn out, when viewed from a different perspective,

 6

also to be factors leading to increased productivity. For instance, compressed

pay structures and promotion and layoff by seniority reduce the incentives for

rivalry in the workplace and make it easier to promote cooperation among

workers. Limitations on the ability of supervisors to discharge make employees

feel more like long-term stakeholders, thereby encouraging them to be more

committed to their employer.

The "exit-voice" model is also relevant for work effort. Freeman and

Medoff (1984) argue that unions are essential if employees are to have a true

voice in the work place. One obvious reason that unions are viewed as

essential for voicing worker concerns is fear. In a nonunion setting, a worker

who speaks out about workplace issues at a minimum risks being labelled as

noncooperative and at a maximum risks losing his job. Another reason is that

working conditions, human resource policies and the behavior of managers are

what economists call "public goods," conditions that everyone in the workplace

"consumes." No public good, whether it be national defense or occupational

safety, will be provided in adequate amounts by voluntary, individual behavior

because the individual bears all the costs and must share the benefits with

everyone else. The economically rational individual will instead wait for

someone else to speak out about public goods at the workplace, the result being

an inadequate flow of information to management about worker concerns. Workers

will speak out only when the costs of doing so are offset by the expected

benefits to them as individuals. The collective institution of unionism can

get around this problem in theory by sharing the costs of voice behavior

equally across all workers, changing the decision rule to a tradeoff between

group benefits and group costs.

 7

Unions send messages to managers through informal day-to-day interaction

as well as through formal grievance procedures and contract negotiations. Many

nonunion firms have attempted to put their own grievance systems into place,

but unless such systems are subject to outside arbitration or a balanced review

panel containing both workers and managers, workers still may view such systems

as an extension of management and use them infrequently. As a result, managers

in unionized workplaces are likely to be better informed, which should increase

productivity. Workers are also more likely to make suggestions for

productivity improvements when bargaining allows the gains to be shared between

the firm and the workforce. Grievances and contract negotiations also give

workers a voice in how the workplace is run. If more voice at the workplace

raises morale, productivity is even further enhanced. Needless to say, many

managers dispute this theory, pointing out cases where unions have created

conflict where none previously existed and other cases where dissatisfaction

with the union resulted in reduced efficiency.

The response of management to unionization is a critical factor in

determining the influence of unions on productivity. Slichter, Healy, and

Livernash and Clark (l9BOb) note that in many cases there are often dramatic

changes in managerial practices when a plant becomes covered by a collective

bargaining agreement. There is a greater reliance on company or plant policy

in making decisions and less reliance on the judgment of any individual

manager. When policies are not applicable or when developing policies,

management becomes more likely to think through the consequences of any

decisions and make appropriate adjustments. The emphasis on policy and

accountability often spreads to other areas of decision making, such as

production scheduling and quality control. There are also usually some
 8

personnel changes; frequently the plant manager is replaced and supervisors are

either replaced or retrained.

Any productivity gains obtained from more professional managerial

techniques can be offset by the reduction in flexibility which inevitably takes

place under unionism. Union and nonunion managers alike almost always claim

that the biggest advantage to remaining nonunion is flexibility. Occupational

jurisdictions and contract provisions limit the ability of managers to

optimally assign workers to jobs. Fluctuations in the demand for labor result

in further inefficiencies as unions limit managerial discretion in layoffs,

overtime, and subcontracting. However, many nonunion firms voluntarily follow

policies that are just as "inflexible" as those followed in their unionized

counterparts. These include promises to workers that they will not be laid off

and de facto seniority rules for promotions.

Unions can also influence the amount of resources the firm allocates to

managerial tasks. For instance, lower turnover under unionism reduces

recntiting, hiring, and training costs; the union even performs a number of

personnel functions in industries such as shipping and construction. However,

higher union wages also imply a queue of applicants that must be screened more

carefully because of union-imposed restrictions on the right to dismiss or

reassign.

A final factor behind any observed union-nonunion productivity differences

is technology. In a static context, the direction of union impact is unclear.

Managers have an incentive to use a less labor-intensive technology to offset

the higher wages, but union work rules may limit this type of adjustment. Over

a longer time horizon, managers must trade off the gains from R&D oriented
 9

toward further reductions in labor intensity against the risk that the union

will be able to expropriate such gains at the bargaining table.

As the above discussion and the accompanying table indicate, it is hard to

draw any general conclusions about the overall impact of unions on productivity

from a purely theoretical discussion. The research question that various

studies have tried to address is to determine the net effect of the nonwage

channels of union influence discussed above, If they show no change or a

reduction in productivity, then the institution of unionism is clearly

undesirable on economic grounds (although not necessarily on the grounds of

equality and social cohesion). If the influence is positive, it is important

to learn whether the productivity gains are large enough to offset the output

losses associated with higher wages. Regardless of whether the effect is

positive or negative, it is also important to determine whether the sources of

the productivity gains can be isolated so the forces that determine

productivity at the workplace can be better understood.

II. EVIDENCE ON OVERALL UNION IMPACT

To estimate the impact of unions on productivity through channels other

than increased wages, the procedure most commonly used in the UP literature is

to estimate a Cobb-Douglas production function where the intercept is allowed

to vary with unionization. The usual specification is an OLS regression of the

log of output per hour or employee on the log of the capital-labor ratio, some

control variables for labor quality, and a unionization variable. Two types of

data have been used in these studies. Brown and Medoff (1978) and Allen (1984)
 10

use Census data broken down by two-digit industry and state or region,

supplemented by data on percentage union members and labor quality from the

public use tapes for the Current Population Survey. This type of data set has

been widely used by economists to study such issues as the effect of education

on productivity and the substitutability of labor for capital. The advantage

of using such data is that they are comprehensive, reasonably accurate, and

readily available in the government documents section of the campus library.

All other studies use some type of micro data, usually collected from

establishments. These data sets generally contain information on more

variables than the Census reports and the results are free of any possible

aggregation bias created by arbitrary state by industry classifications of

heterogeneous establishments.

The productivity measure used in the earliest studies is value added

(dollar value of output minus expenditures for materials) per hour or employee.

Most studies done in the l980s have used some physical measure of output. In

theory the value added concept allows the researcher to compare output and

productivity levels across different industries or within an industry where

there is considerable product differentiation.

In practice there are two difficulties with using value added as an output

measure in a UP study. The main difficulty is that prices of homogeneous goods

and hedonic price functions of heterogeneous goods vary over establishments,

geographic location, and time. This requires some type of price deflation to

make sure that the variation in value added reflects output variation rather

than price variation. A related concern is that value added not only equals

the value of output to buyers, but also equals the cost of labor and capital to

producers. Higher union wages translate into higher labor costs as long as
 11

labor demand is sufficiently insensitive. Critics of the UP studies often

claim that the studies are actually reporting wage equations, not productivity

equations.

In practice this claim seems to be totally unsubstantiated. The union-

nonunion wage difference, generally believed to be in the neighborhood of 15

percent, is much smaller than the productivity difference reported in many of

the studies. Also, some studies report results for both value added and some

physical measure of output. The results show no tendency to obtain

systematically larger estimates of productivity gains under unionism when

output is measured in terms of value added. This concern about possible

systematic upward bias seems to have unnecessarily limited the scope of recent

UP studies to markets where output can be measured in physical terms.

Turning from methodology to results, there seem to be three main

conclusions from the private sector UP studies summarized in Table 2. The most

striking conclusion is that many studies tend to find productivity to be higher

under unionism, even after one controls for the greater capital and skill

intensity of unionized workplaces. This implies that there are sizable

productivity gains associated with the voice and shock effects discussed above

and that the productivity losses associated with union work rules, reduced

management flexibility, and no-carrots/no-sticks are relatively small.

Brown and Medoff (1978) was the first published study to show productivity

to be higher under unionism. They found that in 1972, productivity in

unionized manufacturing establishments was as much as 25 percent higher than in

nonunion settings. Work done by Jonathan Leonard reported in Freeman and

Medoff (1984) replicated this finding for 1977. Further substantiating

evidence was provided in Frantz's (1976) study of the furniture industry. All
 12

of these studies used value added as their output measure. Clark (1980a,b)

found similar conclusions in the cement industry in both a cross section study

based on 1973-1976 data and a before-after comparison of productivity in six

plants unionized between 1953 and 1976. Clark was the first to report results

where output was measured in physical units (tons of Portland cement).

Two manufacturing studies found no union-nonunion productivity difference:

Clark's (1984) study of lines of business in 902 large corporations and Kaufman

and Kaufman's (1987) study of 37 auto parts manufacturing plants. Almost all

of Clark's sample consists of Fortune 1000 companies, whereas other studies

examine firms of all sizes. If the union productivity advantage found in other

studies is limited to smaller organizations, then his results could still be

consistent with studies finding a positive overall union productivity effect.

Another possible explanation for the difference between Clark's results and

those of Brown and Medoff is that tiP studies are very sensitive to different

aggregation techniques. It is instructive to note that Clark's study also

finds that unions reduce profit rates in unconcentrated industries but not in

concentrated industries, exactly opposite to the pattern found in other studies

of unionization and profitability using Census and stock market data. Further

work is needed to determine the role played by aggregation and sample

restrictions based on firm size in generating these two very different sets of

results for manufacturing. It is more difficult to pinpoint possible

explanations for the findings of Kaufman and Kaufman, except to note that there

is no reason to expect unions to be associated with higher productivity in

every setting.

Finally, Bemmels (1988) finds productivity to be 32 percent lower in

unionized plants in manufacturing. This finding is based on 46 responses

 13

generated from a survey of 1,000 firms, raising questions about the

representativeness of the sample. The results also imply an implausibly low

estimate of capital's share of output. Despite these limitations, this is the

only UP study based on recent data.

The second main conclusion is that the impact of unionism on productivity

varies with product market structure. This is shown very strikingly in my

studies (1986a, 1986b, l988b) of various types of construction projects. In

private sector work, the productivity of union contractors is much greater than

that of nonunion contractors, whereas there is no significant union-nonunion

productivity difference in projects completed for the public sector. These are

the same unions and the same contractors. In the hospital and nursing home

sample, they are the same types of structures. Two key reasons for the

difference in the results seem to be inadequate pressures for cost minimization

in the public sector and prevailing wage laws that effectively prevent

competition from nonunion contractors in many markets.

The third main conclusion is that the impact of unionism on productivity

can change dramatically over time. Connerton, Freeman, and Medoff (1983) found

that in 1965 productivity of union mines in the bituminous coal industry was 25

to 30 percent greater than that of nonunion mines. A mere ten years later,

productivity in union mines was 20 to 25 percent lower. They attribute this

largely to changes in the leadership of the United Mine Workers (UMW) union.

Until 1960 the union was run by John L. Lewis. Operations were centralized and

internal dissent was not tolerated. After Lewis' retirement there was frequent

turnover in union leadership (including the conviction of union president Tony

Boyle for ordering the murder of his opponent Joseph Yablonski) and a breakdown

in discipline, reflected in a rapid increase in the number of wildcat strikes.

 14

The union productivity advantage has also declined substantially in the

construction industry. My l988a study shows that for the entire industry union

productivity was about 20 percent higher than nonunion productivity in 1972,

but a statistically insignificant 6 and 8 percent higher in 1977 and 1982. A

key factor behind this change seems to be the rising share of union members

working for nonunion contractors, which has offset the lack of training

possessed by most nonunion workers. In both bituminous coal and construction,

the market share of unionized establishments has declined substantially as one

would expect when higher wages are no longer offset by higher productivity.

Only one study has been done of the impact of unions in the service

sector. Craddy and Hall (1985) compared productivity of the 30 banking

establishments unionized in 1982 with e matched sample of 30 nonunion banks.

They found productivity to be 11 percent lower in the unionized banks.

Even in cases where unionism is associated with higher productivity, the

"bottom line" question is whether this efficiency gain is offset by higher

union wages. The most direct way of examining this question is to estimate a

cost function, a relationship linking total costs to input prices and output.
The cost function evidence is mixed. My 1987 study finds that average cost per

square foot is about the same for office buildings constructed by union and

nonunion contractors, with union contractors having a cost advantage on large

office buildings and nonunion contractors having an advantage on smaller

projects. The same study found nonunion contractors had a cost advantage in

school and hospital construction. In retail construction, my 1988b study found

roughly equal costs for union and nonunion contractors. Cost studies have also

been done for hospitals. Salkever (1982) and Sloan and Adamache (1984) both

find unionism associated with higher costs.

 15

Indirect cost estimates can be obtained by comparing the union effects on

productivity and wages as long as information is available on labor's share of

total costs. Estimates of the impact of unions on wages almost always come

from data on individual workers, whereas the productivity results are based on

establishments or Census aggregates of establishments. Lewis concludes that

estimates of union-nonunion wage gaps based on aggregated data "substantially

overestimate" (1986, p. 45) the true union impact. Without knowing the impact

of aggregation on the productivity gap estimates, comparisons of wage and

production equation results must be interpreted with some care. These

comparisons generally find that when unions are associated with higher

productivity, the union-nonunion wage gap is slightly larger than the

productivity gap, implying that union and nonunion costs are comparable. Of

course, in studies finding zero or negative union productivity effects, higher

union wages necessarily imply higher costs.

The most relevant "bottom line" measure from the firm's point of view is

profits. Studies of the effect of unions on profitability are surveyed in

Freeman and Medoff (1984, ch. 12) and Addison and Hirsch (1989). The profit

studies all reach the same conclusion: unions reduce profits. Whether this

reflects a transfer from owners to union members or a loss in efficiency

remains a subject of ongoing research, which is discussed very well by Addison

and Hirsch.
 16
III. EVIDENCE ON SOURCES OF UNION IMPACT

The evidence sununarizeci above shows that unionization is often correlated

with higher productivity and that the impact of unionization on productivity

varies across different sectors of the economy and over time within each

sector. Yet it does not reveal any information about the mechanisms through

which these productivity adjustments take place. In fact, many of the studies

summarized above make no effort to account for the sources of the reported

union-nonunion productivity difference. In these studies unionism is nothing

more than a label on a black box, inside of which are hidden the true

mechanisms of adjustments at the workplace (Lewin and Feuille, 1983).

In defense of the "black box" studies, including Allen (1986b. 1988b),

three points should be noted. First, the reaction of most academic economists

to studies that claimed to find higher productivity under unionism (holding

skills and capital-labor ratios constant) can be best characterized as

incredulous. This meant that the authors of the earliest studies had to devote

a considerable amount of research effort to establishing the robustness of

their results to all conceivable theoretical and econometric objections. These

efforts seem to be much more appreciated within the economics profession than

in other academic circles. Second, many of the studies have relied on

secondary data sources and in many cases the party collecting the data was not

at all concerned with UP issues. In such situations it is extremely difficult

for the researcher to derive any statistical conclusions about what is going on

inside the black box. Third, it can be argued that the black box criticism

applies across all research areas in labor economics, not just the UP

literature. Why are education and experience correlated with earnings? Why

are real wages procyclical? What determines when a person will retire? Many
 17

other fields of research in labor economics have not advanced very far beyond

the black box stage.

One simple approach used in many UP studies to establish the mechanisms

through which unions influence productivity is to add interaction ternis between

unionization and other variables or, equivalently, to split the sample into

smaller groups and compare the union coefficients of each group. For instance,

Clark (1980a) finds that the union coefficient in cement manufacturing is

greater in the Southwest than in the South or North Central regions and is

positive for new plants but near zero for old plants. Noting the prevalence of

nonunion plants in the Southwest, Clark points out that union plants in that

region may be responding to greater competitive pressure. Another example of

this approach is my 1988a study, in which the union coefficient varies with the

percentage of union construction workers employed by nonunion contractors. The

results showed that the union productivity advantage is much smaller in areas

where a large share of union workers are employed in the open shop. This

suggests that worker characteristics such as training and experience are key

sources of the union productivity advantage in that sector. Clark (1984)

tested whether the union coefficient varied with the market share of the firm

or with percentage of the firm unionized but found that neither factor had any

impact on the size of the union coefficient. Graddy and Hall tested the impact

of the size of the bargaining unit and the maturity of the labor-management

relationship on productivity in their union sample, but once again neither

variable had any explanatory power. Other researchers have no doubt tried this

approach as well and come up emptyhanded. Estimating interaction terms thus

far has yielded a couple of useful clues about possible sources of union-

nonunion productivity differences but no hard evidence.

 18

An approach that would seem to be much more promising is to collect data

on the mechanisms believed to cause union-nonunion productivity differences and

add these variables to the model. This is based on the premise that the union

productivity effect varies within the sample and that some of this variation

reflects differences across both union and nonunion firms in such causal

variables as turnover, job satisfaction, communication, and management. In

theory one could account for how much of the union productivity advantage

results from a particular variable by observing the change in the union

coefficient when that variable is added to the model. For instance, suppose

that differences in job satisfaction fully account for the union-nonunion

productivity difference in a particular sample. Then when this variable is

added to the model, the union coefficient should become zero.

This approach was first used by Brown and Medoff to determine how much of

their estimated union productivity advantage could be explained in terms of

reduced turnover. When they added the quit rate to their model, the union

coefficient dropped from .205 to .160, indicating that lower turnover accounts

for 22 percent of the overall effect and that other factors (communication,

morale, management, etc.) account for the remainder.

In their study of the coal industry, Connerton, Freeman, and Medoff

compared the growth in strikes across different states to productivity growth,

finding that productivity declined the most in states where strikes increased

most rapidly.

I have also used this approach in two of my studies of the construction

industry. In my 1984 study using Census data, I added the ratio of production

workers to total employment and the fraction of workers who have completed

apprenticeships to the model. The first variable was meant crudely to adjust
 19

for changes in the occupational structure (a control for observable labor

quality associated with experience and education was already in the model)1

whereas the second reflected training practices. These two variables accounted

for 15 to 27 percent of the union effect on productivity.

The data set analyzed in Allen (1986a) was collected by the Bureau of

Labor Statistics to measure and analyze productivity in the construction

industry. A likely source of some of the union productivity advantage in

office building construction was lower supervisory requirements. To test this,

I added the ratio of supervisor to total hours and found this accounted for 10

percent of the union effect.

A unique aspect of this data set was that the interviewers asked the

general contractors questions about the impact of seven different factors on

productivity in their project: weather, strikes, building codes, apprenticeship

programs, prefabricated components, standardized components, and supply of

skilled workers. Although this battery of questions ignores some obviously

important factors, one would think ex ante that some of this information would

help identify the underlying sources of higher union productivity.

Ex post this information turned out to be nearly useless. There turned

out to be very little difference between the responses of union and nonunion

contractors. "No effect" was by far the most common response for every factor

on the survey. The factor that seemed to have the biggest impact on

productivity in the contractors' minds was the weather, especially among union

contractors (for reasons unknown). Complicating matters further was the

finding that when each factor was added to the model, its coefficient almost

always turned out to be zero. The only reportable result I obtained from this

analysis was a finding that the greater tendency among union contractors to
 20

report that standardized components raised productivity accounted for as much

as 8 percent of the overall effect. Combining this with the supervisor ratio

and two other variables (ratio of unskilled and semiskilled hours to total

hours and dollar volume of office building construction), I was able to

"explain" 26 percent of the overall union effect.

Kaufman and Kaufman also tried this approach and came up emptyhanded.

Their study probed into eight different personnel policies: grievances,

promotions, layoffs, restrictive work rules, job reassignment during slack

periods, work by supervisors, job posting systems, and profit-sharing. They

found union plants made greater use of seniority in promotions and layoffs and

were more likely to have restrictive work rules, grievance systems, and job

posting systems. Union plants were less likely to have profit sharing plans

and managers in those plants had less ability to reassign workers or use

supervisors. Ho one should find any of these conclusions especially

surprising. What is a bit shocking, however, is that none of these policy

variables were at all related to productivity. Granted the sample size was

small (30 plants for this part of the analysis), but it is naive to think that

anyone looking into these issues is going to have the CPS-esque luxury of

thousands of observations.

The only other way to determine the sources of union productivity effects

besides statistical probing is that old, familiar business school standby -.

the case study. The case study will rarely produce quantitative evidence, but
it holds some promise in regard to its ability to determine whether a

particular channel of union effect has played an important role in a particular

situation. To explain his before-after finding that unionization raises

productivity by 6 to 8 percent, Clark (l980b) interviewed union officials and

 21

management. His interviews successfully isolated changes in personnel

practices. When they were nonunion, plants tended to have no formal grievance

procedures (and thus very few grievances); promotions, layoffs, and recalls

were made by foremen using unspecified criteria. This all changed in a

predictable fashion after unionization.

Clark found substantial changes in management. In each case a new plant

manager was brought in and foremen were either replaced or retrained. In most

plants, formal methods of organizational control were introduced. This usually

involved setting production targets, followed by reviewing and assessing actual

performance. One plant manager noted that ".. .before the union this place was

run like a family; now we run it like a business." Clark concludes that these

changes were one of the "key adjustments to unionization."

Clark's findings about changes in workers' behavior point out another

potential limitation of the case study approach. Unable to collect before-

after data on turnover, absenteeism, worker attitudes, or discipline problems,

Clark had nothing to draw on except interviews with union and management

representatives. He found no solid evidence of any change in turnover or

absenteeism, whereas union and management views concerning changes in morale

often differed.

IV. ASSESSMENT AND FUTURE DIRECTIONS

The UP literature has documented large productivity differences between

seemingly comparable union and nonunion establishments. The sign and the
 22

magnitude of the estimated differences vary considerably across different

studies. Evidence indicates that unionized establishments are most likely to

have a productivity advantage when they must compete on equal terms with the

nonunion sector. The two studies that have examined unionism and productivity

from a dynamic standpoint have both found the estimated union-nonunion

difference to have changed dramatically over time in a direction unfavorable to

the competitive position of unionized establishments.

Faced with this evidence, it would seem very difficult to deny that

differences in the human resource practices and policies must be at least

partly responsible for these productivity differences. This is a very

encouraging signal for the HRFP literature. The prospect that human resource

management can generate productivity changes of as much as 25 percent should be

sufficient to motivate and support HRFP research.

Nonetheless, I think it would be fair to say that most academic labor

economists today remain far from convinced that such a conclusion can be drawn

from the UP literature. A detailed critique is provided by Addison and Hirsch.

Here I will focus on what I believe to be the two most serious contentions.

One reason behind this skepticism is that information about labor quality in

the UP literature is usually limited to occupational mix, schooling, and

experience. If this were all employers really needed to know about the

potential contribution of a job applicant, they would not have to spend any

resources on screening. Of course, we all know that employers learn much more

than this about all applicants and employees. This information, known to the

employer but not to the person analyzing the data, is usually referred to as

unobservable labor quality. The problem this poses for the UP literature is

that under the greater competitive restraint imposed by higher wages, managers
 23

in the union sector should tend to hire more productive individuals than those

in the nonunion sector. Thus, some of any estimated union productivity

advantage is likely to be attributable to unobservable labor quality.

The issue is "how much?" The estimated union-nonunion productivity

difference has not been at all sensitive to the inclusion of observable labor

quality control variables in any UP study. If the estimates are not sensitive

to factors such as schooling, experience, and occupation, it is very hard to

believe they will be more sensitive to some missing variable. Observable human

capital variables generally explain as much as half the interpersonal variation

in wages. Are we to believe that unobservables explain a much larger fraction

of interestablishment differences in productivity?

Selectivity bias is another rationale given by those who discount the UP

literature. Unions presumably are successful in raising wages whenever they

organize an establishment. Some organizations adapt successfully. The data

sets used in the UP studies generally do not contain information on those that

do not. As a result, the estimated productivity level of unionized

establishments is biased upward.

This criticism can be questioned on both theoretical and empirical

grounds. Theoretically, selection bias can operate in either direction. The

case outlined above is one of positive selection bias where the "best plants"

are the ones that get organized. Yet from a logical standpoint, one must also

admit the possibility of negative selection bias where the "worst plants" get

organized, in which case the union-nonunion productivity difference is

underestimated by OLS regressions. Most experts on union organizing campaigns

will tell you that today it is the "worst plants" that stand the best chance of

getting organized. Whether this was true fifty years ago when CIO-affiliated
 24

unions were organizing the likes of U.S. Steel and Ford is more questionable.

The observed distribution of union and nonunion establishments in any data set

is a byproduct of selection processes from different historical periods. Their

overall direction is uncertain.

Empirically, note that Clark obtained exactly the same results in simple

cross section and longitudinal analysis in his studies of cement industry

productivity. This is difficult to reconcile with pronounced selectivity bias.

More fundamentally, many studies of the union wage effect have attempted to

isolate a positive selectivity bias in OLS wage equations generated by the same

process that supposedly has biased the UP literature. These studies,

summarized in Lewis (1986), have failed to produce any agreement about not only

whether the bias is small or large but also whether it is positive or negative.

Even if one accepts the conclusion that the UP literature has firmly

established that human resource management practices can generate large

productivity differences, the one discouraging signal for the HElP literature

is the extremely limited knowledge that has been obtained about the mechanisms

through which human resource policies and outcomes affect productivity. The UP

literature has produced quantitative evidence that productivity is related to

supervision and training in construction and turnover in manufacturing. In

each case, however, most of the union-nonunion productivity difference remains

unexplained. All other studies that have attempted to obtain quantitative

evidence about sources of union-nonunion productivity differences have been

unsuccessful. Clark (1980b) obtained some useful conclusions about the impact

of unionization on management from interviews, but even here one is left with

the question of which was more important - - changes in managerial techniques or

 25

changes in managerial personnel? Recall also that the interview approach was

poorly suited for learning about changes in worker behavior.

To some extent this dearth of conclusions can be attributed to faulty

experimental design. One can point in some cases to the use of secondary data

sources; in others, to inadequate resources for primary data collection or the

inherent limitations of interviews. If one accepts this line of argument, it

would then be sensible to conclude that one way to increase the odds that

future studies will be more successful is to provide those doing such studies

with more resources and to direct those resources toward collecting more

quantitative data on a wider array of variables from more establishments.

This very well may be the ticket, but after reading Katz, Kochan, and

Keefe (1987), I have reservations. Katz, Kochan, and Keefe analyzed

productivity differences across 53 plants of a major automobile manufacturer in

the United States and Canada. They were able to examine how four productivity

measures (supervisors per 100 production workers, labor hours per vehicle,

adjusted labor hours, and a product quality index) were related to grievances,

absence rates, relative wage levels, local unemployment rates, and responses to

a 48-question survey on managerial discretion, pace of work, worker and union

participation, and the use of teams. The data sets used in even the best UP

studies seem spartan in comparison.

Did "more data" break open the black box? The first new challenge the

authors faced was a surplus of data, more variables than observations napped

with multicollinearity. To deal with this, they resorted to principal

components analysis to reduce 48 survey responses to four variables. This

raises the hackles of economists, who always go for Colonel Sanders and pass on

the McNuggets. Focusing on the results rather than the processing method, one
 26

sees that even with massive amounts of information, it is hard to establish

quantitative links between human resource management variables and "bottom

line" outcomes. The principal component reflecting management discretion and

pace of work was the only variable to be consistently related to the

productivity variables. Except for the absence rate, which was related to two

productivity variables, the other variables were either only weakly correlated

or uncorrelated with productivity.

The message here is not to belittle the final product. The findings on

managerial discretion and work pace are a very important contribution to the

literature. The point here is simply that new, bigger data sets are not a

guaranteed route to success. The key methodological message I received from

Katz, Kochan, and Keefe's study is that the linkages between human resources

and productivity are very difficult to sort out and that models that try to do

the whole job in a single step (equation) may not be up to the job.

An alternative approach to getting inside the black box is that used by

studies that focus on a specific mechanism through which unions affect

productivity but do not attempt to go further and show how that mechanism is

related to productivity in the particular sample being examined. Summaries of

selected studies are reported in Table 3; Freeman and Medoff (1984) contains a

more comprehensive set of references. These studies have shown that unions are

associated with lower turnover, higher absenteeism, less management flexibility

in staffing, more work effort, less overall job satisfaction, reduced wage

dispersion, greater employee benefits, and less investment in R&D. The

findings at this stage seem very robust; each is derived from a number of

different data sets using a variety of specifications.

 27

One way of integrating the results of the UP studies with those of studies

focusing on intervening variables is illustrated in Figure 1. The figure

demonstrates a simple model of how human resource policies affect firm

performance. It is not overly dissimilar front the approach used in some

textbooks on human resource management. Companies make choices about how they

hire, train, pay, and manage their workers. This in turn determines human

resource outcomes, as exhibited by easily measured variables such as labor

costs and turnover and by not so easily measured variables such as effort and

cooperation. The "bottom line" depends on the interaction of human resource

outcomes with other firm decisions (e.g., marketing) and the external

environment.

The best UP studies attempt to link productivity directly to changes in

human resource policies associated with unionization. The studies that focus

directly on the intervening variables through which unions influence firm

performance in effect attempt to link human resource outcomes to policies. One

obvious gap is that our knowledge of how these intervening variables are

related to firm performance indicators is very limited. Some progress has been

made along these lines in the studies of turnover, absenteeism, and work rules

cited in Table 3. The studies of grievance procedures summarized in Ichniowski

and Lewin (1987) are also noteworthy in this regard.

Even though the UP studies indicate that human resource policies and

outcomes have a large influence on firm performance, the challenge facing the

HRFP literature is to develop testable hypotheses that will advance the theory

well beyond what is presented in Figure 1 and to develop data sources

appropriate for testing them. Economic theories about labor markets and

employment contracts often focus on decision variables involving selection,

 28

compensation, supervision, and training. Sometimes the models are really

designed for welfare analysis (e.g., share economy) or to explain phenomena

that seem paradoxical to economists (e.g., wage rigidity). New models or

perhaps even the retooling of some existing models could lead to new insights

into how firms make hiring and compensation decisions.

In other cases, the empirical implications of the model are quite clear,

but appropriate data for empirical testing are not available. More data

collection funded by the government, as suggested by Hamermesh (1988), is one

route that shows promise. The development of closer linkages between academic

economists and human resource practitioners could also help in developing

better case studies of single firms and possibly assembling panels of

establishment data.
Table 1. HoR Productivity Chanlee Under Unionism

Type of
effect
(Nsage, Impact on
Variable Union Effect N=nonsage) Productivity

A. Skill levels
1. Hiring standards •Raised to offset increased labor costs N Positive
•Raised to offset restrictions on N Positive
managemenVe ability to re-assign
or terminate

2. On—the-job •Longer tenure Of union sorkers H, N Positive

training ratees payoff to firm From training
•
Apprenticeahi p programs encourage N Positive
broader training in certain trades
3. Occupational •Nore use of skilled labor in response H, N Positive
to increased labor costs
and compressed sage structure
•Cratt traditions in some trades N Positive
•More use of skilled labor reduces N. N Positive
eupervisor-sorker ratio

B. Effort Level,

1. Incentives for •Compreeeed sage Structure reduces N Negative

pay and incentives to advance
promotion •Pay raises rarely based on merit N Negative
•Seniority rule, restrict ability N Negative
to promote on merit
•Incentive pay plans resisted by N Negative
many unions

2. Cooperation •Seniority rules for promotion, N Positive

among sorkers job-based pay systems eliminate
excessive rivalry and encourage
informal on-the—job training

'C
Table 1. (continued)

Type of
effect
(Nwae. Impact on
Variable Union Effect Nnonwage) Productivity

3. Hork rules •Excessive staffing, requirements of N Negative

unnecessary work, output restrictions.
exeseive break time

1*. Discipline •Grievance procedures used to protect N Negative

inoompetent or dishonest workers
•Supervisors unable to discipline or N Positive
discharge workers on arbitrary basis

5. Labor—manage— •Negotiations, grievances open channels N Positive

ment communica- for corkers to send messages to
tion management
•Suggestions more likely when bargaining N Positive
allows gains to be shared between
firm, workers
•Norkers more willing to use chennele N Positive
in union establishment
b. Job •Norkors have greater voice in how N Positive
satisfaction workplace is run, increasing morale
•Diesatiefaotion with union interferes N Negative
with work
•Union creates dissatisfaction where N Negative
none previously existed

C. Management

1. Professionalism •Greater emphasis on policy in making N Positive

day—to-day decisions
'Establish system of production goals and N,H Positive
targets, accompanied by regular reviews
and meetings with supervisors
 Table 1. (continued)

Type of
effect
(Rnage, Impact On
Variable Union Effect Nnonwage) Productivity

1. Profeeeionaiiem •Replao.ment of plant manager, N Frequently

(continued) replacement or re—training positive
of eupervisore

2. Flexibility •Dnion juriedictiona restrict ability N Negative

to reassign personnel during slack
periods
•Seniority rules interfere nith staffing N Negative
especially during layoffs
•Contracts limit subcontracting, overtime N Negative

3. Personnel •Longer tenure reducee recruiting R,N Positive

functions and hiring expenditures
•Applioatione must be screened more carefully R,N Negative
•Hiring halls organize labor market uhen N Positive
morkere not attached to a single
employer

D. Technology •Higher •ages encourage more capital- N Positive

intensive methods, inveet.ent in
labor-saving RID
•Nork rules prevent uee of most N Negative
eftioient technologies
•RID reduced out of fear union nih N Negative
capture all benefits
Table 2. EvidenoS on Union-Nonunion Productivity Difference. in the Private Sector

Output
Sample Type of Data Measure Results Source

Manufacturing, State by industry Value added 1972:10 to 25% Bronn and Kedoff (1978)
1972—1977 aggregates 1977: 31 Freeman and Nedoff (198k)

Manufacturing, Linen of business Value added —2 Clark (19814)

1970-1980 in large corpora-
tions

Manufacturing, Establishments Value added —32 Bemmels (1988)

1983

Nooden household Establishments Value added 15 Frants (19Th)

furniture, 19Th

Cement, 1973-1976 Establishments Tons 7 to 10 Clark (1980.)

Cement, 1953-1976 Before-after Tons 6 to 8 Clark (1980b)

comparisons of six
establishments

Luto parts, 1982 Establishments Value added 2 kaufman and Kaufman

(1987)

Bituminous coal, Establishments Tons 1q65: 33 to 38 Connerton, Freeman and

1965—1980 1970: —U to 8 Nedoff (1983)
1975: —20 to —17
1980: —18 to —14

Construction, State by industry Value added 1972: 17 to 22 Allen (19814a, 1988a)

1972-1982 aggretates deflated by 1977: 8
price or cost 1982: 6
index

Commercial office Structures Square footage 36 to 38 Allen (1Q86.)

buildings, 19714
Table 2. (continued)

Output
Sample Type of Date Measure ResultS SoUrce

Elementary and Structure. Square footage 1 to $ Allen (1986a)

secondary school.,
1972

Private hospital. Structures Square footage 23 Allen (1986b)

and nursing
homes, 1976

Public hospitals Structures Square footage 2 Allen (1986W

and nursing
ho..., 1976

Retail store. and Structures Square footage 51 Allen (1988b)

shopping centers,
197?

Commercial banking, Establishment. Lending products —11 Graddy and Hall

1978 (1985)
T.ble 3. Evjd.nc. on Sourc•s Di' Union—Nonunion Productivity Differences

Variable Results Source

Turnover Voluntary quits are 50 to 100 percent lower Free.an (1980a)

among union members.

Absence rate Absence rates are 30 to 40 percent higher Allen (1984b)

among union members.

Nanagement Union end nonunion establishments in aanu- Freeman and Hedoff (1982)
flexibility racturing have equal flexibility in adjusting
capital—labor ratios, but union establishments
have loss flexibility in adjusting ratio of
production to nonproduction labor.

Union and nonunion construction contractors Allen (19860)

and subcontractors have equal flexibility in
substituting capital and materials for labor,
but union contractors have less flexibility
in substituting supervisory and unskilled
labor for skilled labor. Results in excess
staffing of 3%, excess cost8 of 2%.

Work effort Union members report greater work effort, more Duncan and Stafford (1980)
difficulty in getting a couple of hours oft'
work, less choice in how they can do work,
and less ability to refuse overtime without
penalty.

,Job satisfaction Union members report more satisfaction with Kochan (1979),
bread—and—butter aspects, but less satis— Freeman (1978)
faction with promotions, supervisors, job
content, and resource adequacy. Overall
levels of job satisfaction are lower for
union workers.
Table 3. (continued)

Variable Results Source

Rage dispersion Union cage policies reduce cage dispersion Freeman (1980b, 1982)
cithin and across establishmente.

Employee benefits Unionhted establishments more likely to have Freeman (1981)

pension., health insurance, overtime premiums,
.nd shift differentials, chore.. they are
les, likely to have bonuses.

Retired union corkers receive larger benefits Allen and Clark (1986)
upon retirement and larger benefit increases
after retire.ent. Union iorkers retire
earlier, allocing them to collect benefits
over a longer period.

RID Firms in highly unionized industries invest Connolly, Hirsch, and

lees intensively in RID nd RID adds Rtrschey (1986)
relatively less to the market value of'
such firms.

Investment Unionized firms invest less in capital Bronars and Deere (1987),
Hirsch (forthcoming)
 36

Human resource Human resource Financa

policies outcomes outcomes

Selection Performance PrOL'its

Employee voice Turnover Stock value
Training Attendance Costs
Compensation Flexibility I Productivity
Monitoring Job satisfaction Growth
Labor relations Labor costs Market share

External Other firm

influences policies

Labor supply Production

Collective bargaining Marketing
Legal environment Finance
Competition IS & U
Consumers Information

Figure 1. Model of human resource policies and firm performance

 37

REFERENCES

Abraham, Katharine G., and Medoff, James L. "Length of Service and Promotions
in Union and Nonunion Work Groups." Industrial and Labor Relations Review 38
(April 1985): 408-420.

Addison, John T., and Hirsch, Barry T. "Union Effects on Productivity,

Profits, and Growth: Has the Long Run Arrived?" Journal of Labor Economics 7
(January 1989): forthcoming.

Allen, Steven C. "Unionized Construction Workers Are More Productive."

Quarterly Journal of Economics 99 (May 1984): 251-274 (a).

"Trade Unions, Absenteeism, and Exit-Voice." Industrial and Labor

.

Relations Review 37 (April 1984): 331-345 (b).

"Unionization and Productivity in Office Building and School

.

Construction." Industrial and Labor Relations Review 39 (January 1986): 187-

201 (a)

"The Effect of Unionism on Productivity in Privately and Publicly Owned

.

Hospitals and Nursing Homes." Journal of Labor Research 7 (Winter 1986): 59-
68 (b).

. "Union Work Rules and Efficiency in the Building Trades." Journal of

Labor Economics 4 (April 1986): 212-242 (c).

"Declining Unionization in Construction: The Facts and the Reasons."

.

Industrial and Labor Relations Review 41 (April 1988): 343-359 (a).

"Further Evidence on Union Efficiency in Construction." Industrial

.

Relations 27 (Spring 1988): 232-240 (b).

Allen, Steven C., and Clark, Robert L. "Unions, Pension Wealth, and Age-
Compensation Profiles." Industrial and Labor Relations Review 39 (July 1986):
502-517.

Bemmels, Brian. "Managerial Efficiency, Unions, and Productivity at

Manufacturing Plants." Mimeograph. Edmonton, Alberta: University of Alberta,
1988.

Bronars, Stephen C., and Deere, Donald R. "Union Threats, Unionization, and
Firm Behavior: A Decision Event Study." Mimeograph. Santa Barbara, Cal.:
University of Claifornia, 1987.

Brown, Charles, and Medoff, James L. "Trade Unions in the Production Process."
Journal of Political Economy 86 (June 1978): 355-378.

Clark, Kim B. "Unionization and Productivity: Micro-Econometric Evidence."

quarterly Journal of Economics 95 (December 1980): 613-639 (a).
 38

"The Impact of Unionization on Productivity: A Case Study." Industrial

.

and Labor Relations Review 33 (July 1980): 451-479 (b).

. "Unionization and Firm Performance." American Economic Review 74

(December 1984): 893-919.

Connolly, Robert A.; Hirsch, Barry T.; and Hirschey, Mark. "Union Rent Seeking,
Intangible Capital, and Market Value of the Firm." Review of Economics and
Statistics 68 (November 1986): 567-577.

Connerton, Marguerite; Freeman, Richard B.; and Medoff, James L. "Industrial

Relations and Productivity: A Study of the U.S. Bituminous Coal Industry."
Mimeograph. Cambridge, Mass.: Harvard University, 1983.

Duncan, Greg J., and Stafford, Frank P. "Do Union Members Receive Compensating
Wage Differentials?" American Economic Review 70 (June 1980): 355-371.

Ehrenberg, Ronald C., and Milkovich, George T. "Compensation and Firm

Performance." In Human Resources and the Performance of the Firm, edited by
M.M. Kleiner j. Madison, Wis.: Industrial Relations Research Association,
1987.

Frantz, John R. "The Impact of Trade Unions on Production in the Wooden

Household Furniture Industry." Senior honors thesis, Harvard College, 1976.

Freeman, Richard B. "Job Satisfaction as an Economic Variable." American

Economic Review 68 (May 1978): 135-141.

"The Exit-Voice Tradeoff in the Labor Market: Unionism, Job Tenure,

.

Quits, and Separations." Quarterly Journal of Economics 94 (June 1980): 643-

674 (a).

"Unionism and the Dispersion of Wages." Industrial and Labor Relations

.

Review 34 (July 1980): 3-23 (b).

"The Effect of Unionism on Fringe Benefits." Industrial and Labor

.

Relations Review 34 (July 1981): 489-509.

"Union Wage Practices and Wage Dispersion Within Establishments."

.

Industrial and Labor Relations Review 36 (october 1982): 3-21.

Freeman, Richard B., and Medoff, James L. "Substitution between Production

Labor and Other Inputs in Unionized and Nonunionized Manufacturing." Review of
Economics and Statistics 64 (May 1982): 220-233.

. What Do Unions Do? New York: Basic, 1984.

Craddy, Duane B. , and Hall, Gary. "Unionization and Productivity in Commercial

Banking." Journal of Labor Research 6 (Summer 1985): 249-262.

Hamermesh, Daniel S. "Data Difficulties in Labor Economics." National Bureau

of Economic Research Working Paper No. 2622. Cambridge, Mass.: NBER, 1988.
 39

Hirsch, Barry T. "Innovative Activity, Productivity Growth, and Firm

Performance: Are Labor Unions a Spur or a Deterrent?" In Advances in ADDlied
Microeconomics, Vol. 5, edited by Albert N. Link and V. Kerry Smith.
Greenwich, Conn.: JAI Press, forthcoming.

Hirscbman, Albert 0. Exit. Voice, and Loyalty. Cambridge, Mass.: Harvard, 1970.

Ichniowski, Casey, and Lewin, David. "Grievance Procedures and Firm

Performance." In Human Resources and the Performance of the Firm, edited by
M.M. Kleiner . Madison, Wis.: Industrial Relations Research Association,
1987.

Katz, Harry C.; Kochan, Thomas A.; and Keefe, Jeffrey H. "Industrial Relations
and Productivity in the U.S. Automobile Industry." BrookinEs Papers on Economic
Activity (1987:3): 685-728.

Kaufman, Robert S., and Kaufman, Roger T. "Union Effects on Productivity,

Personnel Practices, and Survival in the Automotive Parts Industry." Journal of
Labor Research 8 (Fall 1987): 333-350.

Kleiner, Morris M.; Block, Richard N.; Roomkin, Myron; and Salsburg, Sidney W.,
eds. Human Resources and the Performance of the Firm. Madison, Wis.: Industrial
Relations Research Association, 1987.

Kochan, Thomas A. "How American Workers View Labor Unions." Monthly Labor
Review 102 (April 1979): 15-22.

Kochan, Thomas A.; Katz, Harry C.; and McKersie, Robert B. The Transformation
of American Industrial Relations. New York: Basic, 1986.

Lewin, David, and Feuille, Peter. "Behavioral Research in Industrial

Relations." Industrial and Labor Relations Review 36 (April 1983): 341-360.

Lewis, H. Gregg. Union Relative Waze Effects: A Survey. Chicago: University of

Chicago Press, 1986.

Salkever, David S. "Unionization and the Cost of Producing Hospital Services."

Journal, of Labor Research 3 (Summer 1982): 311-33.

Slichter, Sumner H.; Healy, James J.; and Livernash, E. Robert. The InDact of
Collective Bargaining on Manaement. Washington: Brookings Institution, 1960.

Sloan, Frank A., and Adamache, Killard W. "The Role of Unions in Hospital Cost
Inflation." Industrial and Labor Relations Review 37 (January 1984): 252-262.