Employer Size and Labor Turnover

by Todd Idson, Columbia University

Discussion Paper No. 673

75

Employer SizeAnd Labor Turnover

Todd L. IdsonColumbia University

November 1993

This work has benefitted from discussions with Joseph Altonji, David Bloom, Charles Brown, SherryGlied, Jacob Mincer, and James Rebitzer; and from seminar participants at Columbia University andthe University of Texas, Austin.

Abstract

This paper investigates the causes of higher tenure, and lower turnover, for workers at largeplants or firms. The primary hypothesis investigated is that long-term employment relationships inlarge plants and firms stem from (1) the greater capacity of large employers to provide jobopportunities within the enterprise and (2) the lower probability of business failure facing largeproducers. In addition to directly lowering turnover, these effects on average employment durationact to raise the expected returns from on-the-job training, resulting in higher levels of on-the-jobtraining that in turn inhibit turnover. Furthermore, in order to protect these training investments,remuneration is characterized by mobility inhibiting wage premia and fringe benefits, and firmsattempt to hire employees with lower mobility tendencies - both practices act to further reducemobility. In addition to empirically evaluating the above view of the manner in which size ofemployer influences mobility outcomes, three additional factors are investigated that may accountfor the link between size and mobility, (1) working conditions, (2) monitoring difficulties, and (3)union organization threats.

I. Introduction

Employer size wage differentials have puzzled economists for many years (Lester, 1967;

Masters, 1969), and have been the subject of intense empirical and theoretical scrutiny during the

past decade (Mellow, 1983; Oi, 1983; Weiss and Landau, 1984; Garen, 1985). Numerous theories

have been advanced to account for these wage differences, though empirical investigations have at

best been able to account for only about half of the observed differential, leaving a highly significant

wage premium after the introduction of a rather comprehensive set of controls (Brown and Medoff,

1989). While some theories, such as compensating differentials and union threat effects have

received little empirical verification, other explanations, such as labor quality differentials have

received some empirical support (Evans and Leighton, 1989; Idson and Feaster, 1990). The

underlying cause of the persistent residual employer size differential in wages, though, still remains

a mystery.

Given this wide interest in the effects of employer size on compensation, it is surprising that

relatively little attention has been directed towards size differentials in turnover. To date only two

studies have investigated this relationship, one focusing on tenure differentials using the 1979 May

Current Population Survey (Rebitzer, 1986) and the other addressing quit differentials using the

1973-1977 Quality of Employment Survey panel (Brown and Medoff, 1989). In fact, there are no

studies of employer size effects on layoffs, with the result that the relative contributions of voluntary

and involuntary turnover to observed size differentials in separations are unknown. Furthermore,

since the Quality of Employment Survey only reports plant size, extant research does not allow us

to distinguish between plant and firm size effects, leaving the question of the relative strengths of

plant and firm size on either quits or layoffs unanswered.

This study sets out to fill these gaps by providing a detailed analysis of the effects of both

plant and firm size on voluntary and involuntary turnover in order to discern exactly what attributes

of large employers lead them to establish longer term employment relationships than found in small

firms. The basic hypothesis advanced in this paper is that lower turnover in large firms results from

the inherently greater capacity that large employers possess to develop long-term relationships with

their employees. This differential capacity stems from two sources, (1) a greater ability to allow

employees to substitute intra-firm for inter-firm mobility, or alternatively stated, to develop their

careers and search for optimal matches (and desirable types of jobs) within the firm rather than

going to another employer, and (2) lower failure probabilities for large firms.1 Both factors cause

large firms to be more willing to invest (due higher expected returns) in long-term employment

relationships through higher levels of on-the-job training.2 As a result, larger employers will also

tend to pay mobility inhibiting remuneration, through both wage premia and fringe benefits, and tend

to hire workers with lower turnover propensities.

The structure of the paper is as follows. In Section I, a number of different data sets are

used to document the structure of the relationship between employer size and tenure, turnover, and

job mobility within the firm. Section II more fully develops the argument that a greater expected

duration of the employment relationship will lower turnover both directly through a substitution of

internal for external mobility, and indirectly through resulting differences in OJT, mobility inhibiting

compensation, and selectivity in hiring. Section III evaluates this explanation using a number of

different data sets. Section IV discusses and empirically evaluates three alternative explanations for

lower turnover in large firms, specifically (1) that working conditions inhibit quits, (2) that

differential monitoring difficulties affect layoff rates, and (3) that union threat effects impact most

strongly on larger employers. Finally, Section V summarizes the results and discusses directions for

failure probabilities for smaller firms may increase quits as workers search (and some find)alternative employment in anticipation of job loss through firm failure, while higher layoff rates may reflectemployment declines prior to failure and possibly greater sensitivity of small firms to cyclical fluctuations.

Hn addition, since small firms cannot as credibly guarantee long-term employment relationships, it is lesslikely that employment stability will follow from the use of delayed payment contracts (Lazear, 1981) in smallfirms (Idson and Valletta, 1993).

future research.

II. Employer Size Patterns in Tenure, Turnover, and Internal Mobility

Throughout this study the analysis is restricted to private sector, nonunion male employees

who are not employed in agriculture, and range in age from 16 to 64. The data sets used are (i) the

1979 May Current Population Survey (CPS), (ii) the 1973/1977 Quality of Employment Survey

(QES), (iii) the second wave (1982) of the employer survey of the Employment Opportunities Pilot

Projects (EOPP), and (iv) the National Longitudinal Survey of Young Men (NLS). The CPS is a

national household-based sample of the noninstitutionalized civilian population; the data used in this

study are from the May Pension Plan supplement that includes questions about employer size (and

fringe benefits, including pension plan coverage) that are matched to June earnings data and the

March income supplement file. The QES is a national sample of persons 16 years old or older who

were working for pay for 20 or more hours per week; 1455 respondents were interviewed in 1973

and then reinterviewed in 1977. The EOPP employer survey is a national sample of employers who

provided information on their firm and on their most recently hired employee before the interview;

this study uses the second wave of the employer survey that was conducted in 1982. Since EOPP

was designed to test the effects of an intensive job search program combined with a work and

training program, workers in the EOPP are primarily lower skilled. The NLS Young Men is a

national sample of 5,225 young men ages 14-24 in 1966 the first year of the panel (which ended in

1981); this study uses data from interview years 1976-1978 (no interview was conducted in 1977).

Use of a number of different data sets will facilitate tests of different hypotheses, and in some cases

will allow us to examine how robust the results are to different samples and specifications.

Table 1 provides definitions for the variables used in this study and the corresponding sources

from which they are derived. Tables 1.1-1.3 document the structure of mobility differentials by size

of employer. Looking at tenure patterns by employer size, in Table 1.1, the tenure-size profiles

reveal near monotonic increases with both plant and firm size in both the CPS3 and NLS, and a

similar pattern for plant size in the QES except for the 500-999 category (for which the sample size

is relatively small). In both the CPS and NLS we see that the increase in average tenure between

the 1-24 and 1,000+ size groups is greater for plant size than for firm size, though for both size

measures average tenure is approximately 50% higher in the 1,000+ size group than in the 1-24 size

group. For the QES the tenure-(plant) size profile is of similar magnitude to that in the CPS and

NLS, lying closer to their plant size profiles than to the firm size profiles. Apparently whatever

factors are producing these unstandardized size effects on tenure operate at both the plant and firm

level, yet these factors seem to carry more force at the plant level. Furthermore, we see that the

relative magnitudes of these effects are rather robust across distinct data sets.

Looking at the panel information in Table 1.2, a pattern of declines in separations, quits, and

layoffs is evident in both the QES and the NLS. As in the case of average tenure, plant size effects

are stronger than firms size effects, and this is true not only for total turnover, but also for both

voluntary and involuntary turnover. Furthermore, at both the plant and firm level, layoff rates

decline faster in percentage terms than do quits, though since quits are numerically greater than

layoffs the overall decline in separations rates with size appear to be predominately attributable to

the size pattern in quits. A similar pattern is observed at the plant level in the QES.

These results can be used to discern the relative contributions of quits and layoffs to

employer size patterns in overall turnover - these calculations are shown in Table 1.3. We see that

at both the plant and firm level, the percentage of total turnover due to quits rises with employer

size, and the percentage due to layoffs tends to decline (although the pattern is not monotonic).

3The patterns reported in Table 1.1 for the May 1979 CPS were also found in the May 1983 CPS data(results available on request).

This is as expected from the more pronounced decline in layoffs relative to quits with size.

Furthermore, using the figures in Table 1.2, in the QES the decline in quits (layoffs) between the

smallest and largest plant size categories accounts for 66.7% (33.3%) of the decline in separations

between these cells. Similar magnitudes are found in the NLS - quits (layoffs) account for 63.2%

(36.8%) of the decline in separations with plant size; quits and layoffs each account for 53.8% and

46.2%, respectively, of declines in separations with firm size.4 These results indicate that the decline

in quits with employer size constitutes the dominant factor accounting for the decline in total

turnover with size, and that this effect is strongest at the plant level.

III. Long-Term Employment Relationships

Understanding differences in labor markets outcomes in different size firms requires

distinguishing between factors that stem from differences in size per se, and those that are the

outcome of the behavior of different size firms. For example, observed higher levels of on-the-job

training (OJT) in large firms may help explain such phenomena as greater wage growth (Barron, et

al., 1987; Pearce, 1990) and lower turnover in large firms, yet still leaves unexplained why OJT is

greater in large firms. A complete explanation would account for OJT levels as an outcome of an

optimization process that is predicated upon factors that inherently vary with size. In this study I

posit that large firms are intrinsically better able to develop long-term relationships with their

employees due to a lower failure probability and a greater capacity to provide career growth and

alternative types of jobs within the organization, and that employer size patterns in remuneration,

calculations are based, as in Table 1.3, on defining separations as the sum of reported quits andlayoffs, thereby excluding separations reported in Table 1.2 that were not able to be classified as a quit orlayoff. If actual separations are used, then in the QES quits account for 64%, and layoffs 32% of separations,4% of the decline being left unclassified. Similarly, for the NLS quits account for 50% (39%) of the plant(firm) size decline in actual separations, layoffs account for 29% and 33%, respectively, with 21% (31%) ofthe decline with plant (firm) size unable to be classified into voluntary or involuntary turnover.

OJT, and the types of employees hired ultimately follow from these root causes.5

Developing this argument more fully, by virtue of their greater size large employers are more

able to allow workers who want to change jobs (say, lateral changes in the type of job performed

and/or the group of people worked with, career advancement, or the location of work in the case of

multi-plant firms) to do so within the firm, thereby allowing workers in large firms to move within

the firm rather than being forced to fulfill their job goals by changing employers. In addition to

higher levels of internal mobility, larger firms face a lower likelihood of failure. Both of these

factors act to raise expected tenure, and thereby to raise the expected returns from any given volume

of investment in on-the-job training.

These ideas can be seen by inspecting the present value of the profits (II) from training in

(1) below:

E(r, T) = fT[V(t) - W(f)]e-" dt (1)Jo

where r=the discount rate, T=expected tenure at retirement, V=the marginal revenue product of

labor, W=wages, and t=tenure (time employed). Assuming simple linear V and W schedules, say

V^Vo+Vjt and W=wo+w1t, then a sharing of costs and returns from OJT (assumed to be imperfectly

general in nature) requires that V<W during training (say up to t=t*) and V>W after OJT is

completed;6 this process yields parameter restrictions vo<wo and v1>w1. Since for any given volume

of OJT investment costs, rates of return will be increasing with the present value of the returns on

5One example of an alternative factor that has been posited as the basic cause of employer size effects isan inherently higher cost of monitoring worker performance in larger firms (Oi, 1983; Garen, 1985). In Oi'smodel higher monitoring costs are due to the greater opportunity cost of manager's time in larger firms, whilein Garen's model monitoring costs rise with size due to difficulties associated with policing larger,interdependent groups.

Similarly, W<Wa for t<t*and W>Wa for tet*, where Wa is the wage available to the employee inalternative employment.

the investment, we may evaluate the behavior of these returns:

R(r,T)= f*[V(t) -fFfl) ]« •*< ' - ' •>* (2)Jt*

From (2) it readily follows that returns to OJT are increasing in T, namely higher for workers with

longer expected tenure at the firm;7 using the linear functions for V and W yields:

§ )e-"T-''> > o (3)

Similarly, it also follows from (2) that returns to OJT are decreasing in r, namely that a lower

probability of business failure - reflected in a lower discount rate - will raise the anticipated returns

to OJT investments:

dr * * r2 r3

Hence, both longer expected tenure (higher T) and lower failure probabilities (a lower

discount rate, r) act to directly raise expected tenure, and to indirectly further raise expected tenure

(and lower both quits and layoffs to the extent that employers and employees share the investment

costs) as higher expected returns to OJT lead to greater training investments.

Furthermore, in order to protect OJT investments, large employers will pay mobility

inhibiting wage premia and structure compensation towards deferred compensation and nonvested

fringes. This conclusions follow from recognition that if larger employers have higher levels of

investments in their workers, it is consistent with profit maximization in competitive labor markets

for the employer to pay above opportunity cost wages (and/or fringes) in order to inhibit relatively

costly turnover, even in the face of queues of qualified workers willing to work at lower wages (see

7This is one of the reasons that larger employers tend to hire otherwise statistically identical people atyounger ages (see Idson, 1989).

Weiss, 1980). Letting H=training investments, V=the marginal product of labor (for simplicity

assumed to be constant), W=wages in the firm, Wa = wages available with other employers if the

worker quits, L=labor, and Q=the probability of a worker quitting, then as shown in Salop (1973)

and Calvo (1979) we may write firm profits as:

E = VL - WL - H Q (W - Wa) L (5)

where quits are assumed to be a decreasing function of the wage premium paid in the current firm

(Q'<0), and the last term captures training (and hiring) costs, paid for by the employer, that are lost

if the worker quits. Choosing a profit maximizing wage yields the condition:

1 / H = -Q ( W - Wa ) (6)

from which we see that profit maximization requires that greater training investments are matched

by larger wage premia, or other elements of the compensation package, in order to reduce relatively

more costly quits. These considerations help to explain not only the higher level of wages and higher

incidence of fringes in large firms (Mellow, 1983), but also the greater increase with size in measures

of total income from a job as opposed to just wages per se (Brown and Medoff, 1989).

In sum, I have argued that greater levels of intra-firm job changes and lower failure

probabilities constitute the underlying basis of size differentials in both wages and mobility. Direct

reductions in separations occur (i) as workers substitute internal for external mobility, reducing quits,

and as employers move workers into alternative positions rather than resorting to layoffs, and (ii)

as layoff rates decline due to lower actual and anticipated failure (similarly, quit rates may decline

to the extent that some quitting is due to fears of imminent failure). In addition, indirect reductions

in inter-firm mobility occur through any positive effect that these anticipated long-term employment

relationships have on training, wages and nonwage compensation, and selectivity in hiring.

IV. Evidence on Long-Term Employment Relationship Effects

As described above, larger establishments and/or firms have a greater capacity than small

enterprises to establish long-term employment relationships due to a greater ability to offer within

firm job mobility and lower failure probabilities. As a result of this aspect of size, workers in larger

firms will tend to undertake higher levels of OJT, receive wage premia and a higher proportion of

their compensation in the form of fringe benefits, and tend to exhibit lower pre-employment turnover

propensities. Although each of these factors constitutes different facets of a common employment

dynamic that differentiate larger from smaller enterprises, evidence on the impact of these factors

will be reported separately below for clarity of exposition.

(i) Internal Mobility

Table 2.1 investigates the effects of plant size on a number of alternative measures of intra-

firm mobility using information from both the QES and the NLS. Looking at the QES results in

panel (a),8 we observe a strong pattern of greater intra-firm mobility in larger plants for total

number of job changes with the current employer (NUMJOB), and both promotions (NPROMO)

and either demotions or lateral job changes (NDEMLAT); with a significantly stronger size effect

on promotions relative to demotions or lateral moves. We similarly see in the last column that time

in current position (JOBTEN), holding constant time with current employer, is shorter in larger

plants.9

8While these results provide insight into the relationship between employer size and internal job changes,we would realty like to have information on firm size since QES respondents were asked about their internalmobility behavior with their current employer, not in the current plant per se. For example, in a multi-plantfirm, employment in a small plant could correspond to a large firm, thereby clouding the relationship betweendifferent dimensions of employer size and internal mobility.

'It is essential to control for tenure in these regressions since longer average tenure levels at larger firmswill tend to produce longer time per se in their current position for employees of larger firms. For example,Brown and Medoff (1989) failed to control for tenure in similar regressions and as a result found positive,

It is interesting to note that a pattern of diminishing mobility within the organization at

higher levels of employment tenure is observed for the total number of job changes and for the

number of promotions (similar results are obtained when tenure is instrumented; see Idson, 1989).

These patterns are consistent with the job matching model of Javonovic (1979) where workers

initially relocate until they achieve a satisfactory match. What we may be observing, therefore, is a

process in larger organizations where workers are more able to engage in job search within the firm

rather than moving between firms.10 This is consistent with a pattern of substitution of internal for

external mobility predicated on a greater likelihood of finding a match within the larger organization,

with this effect most pronounced at relatively low tenure levels.

Of course these findings could simply be reflecting a greater tendency in larger organizations

to more finely define job titles. Yet supplementary evidence on employee's self-evaluation of

promotion likelihood (CPROMO) and the expected number of years until promotion (YPROMO)

both reveal greater anticipated promotions in larger establishments; in the first case we see that

workers in larger establishments feel that they are more likely to get a promotion, while in the latter

case they feel that they will receive a promotion sooner. Both of these findings indicate greater

opportunities for career advancement within larger organizations - results unlikely to be observed

if workers are simply shifting job titles. Furthermore, while more frequent job changes in larger

firms might be viewed as evidence of inferior jobs, with workers leaving undesirable job assignments

albeit insignificant, size effects for time in current job assignment. Similar results to those reported in Table2.1 (for the column labelled JOBTEN) are found when the dependent variable is coded as job tenure dividedby tenure (with tenure excluded as a regressor) - results available on request.

10While the greater number of jobs available at larger firms enables their workers to sample different jobswithout changing employer, this does not guarantee that the employee will choose to do so or that theemployer will allow the employee to do so. Yet, since labor turnover occurs when imperfect matches areformed and better matches are sought, lower search costs within the organization (as opposed to search acrossfirms) may induce the employee to first concentrate his search activity at the current firm. On the employersside, the higher levels of training provided at larger firms may induce larger employers to encourage theirworkers to search for a match in the firm (and in the process learn more about the employee before a long-term commitment is undertaken).

10

in search of better ones within the organization (which would support a compensating differential

explanation of employer size wage effects), the evidence on promotions shows rather that workers

are leaving their assignments more quickly due to their professional advancement in the firm.

Additional information on the link between employer size and intra-firm mobility is reported

in panel (b) for the NLS. Although the information on internal mobility in the NLS is far less

complete than in the QES, the NLS has the advantage of not only a much larger sample, but also

information at both the plant and firm level. These results generally confirm what was observed in

the QES, including the tenure patterns in internal mobility. Furthermore, as expected we see that

employer size effects operate predominantly through firm size, indicating that the plant size effects

seen in the QES are probably reflecting the positive correlation between plant and firm size.

Having documented the link between employer size and intra-firm mobility, Tables 2.2 and

2.3 report the results of maximum likelihood separation, quit, and layoff probits using the 1973-77

QES panel and the 1976-78 NLS panel. Looking first at Table 2.2, the first column reports runs

without internal mobility controls in order to estimate a baseline employer size effect. Since the

maintained hypothesis of this study is that employer size acts to generate lower turnover directly

through the effects of internal mobility and lower failure rates, and indirectly through resulting

variations in OJT, remuneration, and selection of employees with lower mobility tendencies, these

specifications do not include controls that are hypothesized to result from employer size (i.e. wages,

OJT, and so forth). The second and third columns add alternative measures of internal mobility in

order to see if they (i) are associated with reduced turnover, and (ii) reduce the estimated size effect.

Similar to the first specification, these regressions do not include factors that are assumed to vary

with size due to higher levels of internal mobility, and thereby estimates the gross effect of

differential intra-firm mobility on turnover.

We see initially from all three panels that the probability of an employee in a large plant

11

separating from his job is lower than for an employee in a small plant, and that this is true for both

voluntary and involuntary turnover. Looking at the effects of intra-firm mobility on turnover, we see

that the average number of job changes per year11 is negatively associated with the probability of

a separation,12 though taking into account the effect of intra-firm job change only reduces the size

effect by only approximately 4%. Similarly, the average number of promotions per year is

statistically insignificant at conventional levels (achieving significance at only a 15% level), leaving

the size effect essentially unchanged. In the case of quits, both internal mobility measures are

insignificant, although have the predicted negative signs, and only marginally reduce the size effect.

For layoffs similar conclusions follow, although the effect of controlling for total internal job change

reduces the size effect by approximately 11%. It is interesting to note that the effect of promotions

is weaker and less significant than total internal mobility for all turnover measures, a result consistent

with firms moving workers into different positions as a way to avoid layoffs.

The first two columns of Table 2.3 report a similar analysis using the NLS. As in the QES

we see that turnover is negatively associated with employer size, but that this is only evident at the

firm level, and neither size measure achieves conventional levels of significance in the layoff

regressions (though deleting plant size from the layoff regressions produced a "firm size" effect that

achieves 10% significance). The analysis of internal mobility effects yields results that are essentially

the same as in the QES, with insignificant internal mobility effects (using a dummy similar to

CHJOB in the QES also yielded an insignificant effect - results available on request) and essentially

no movement in either the plant or firm size effects.

uThese measures are calculated relative to current tenure since the external mobility inhibiting effects ofinternal mobility would be expected to differ if a given number of internal changes occurred over different timeintervals.

12 These results are consistent with Brown and Medoffs (1989) finding that the size effect on quits isstronger (and more significant) for workers who changed 3-digit occupational categories than for those whodid not; although a significant effect persisted for both groups.

12

Though these results produce little evidence in support of the direct effects of internal

mobility on turnover, an alternative approach was taken that allows all parameters to vary, especially

the employer size effects, by whether or not the employee has changed job with the current employer

- these results are reported in the last two columns of Table 2.3. We now see that firm size has a

significantly negative effect on separations only for employees who have moved internally at least

once, and that the effect is nearly three times as large as that for employee who have not changed

jobs internally. The results are even more dramatic for voluntary turnover, with an estimated firm

size effect for internal job changers that is over ten times the size of the effect for the group that

did not experience internal mobility. The results for layoffs follow precisely the opposite pattern,

exhibiting a significant firm size effect only for the no internal mobility group. Perhaps this result

reflects the effects of people either experiencing demotions, or else unsuccessful internal job search.

In sum, the results from this section indicate that internal mobility is greater in larger

organizations, and that both voluntary and involuntary turnover is lower for people employed by

larger employer, with indications from the NLS that this effect is predominantly a firm size effect.

Furthermore, both the QES and NLS results indicate that internal mobility is weakly associated with

lower separation probabilities but only acts at best to marginally reduce the size effect, so that the

measures of internal mobility used in this analysis do not by themselves account for these lower

turnover levels. Taken on face value these results indicate that while employees in larger firms do

exhibit greater intra-firm mobility than employees at small firms, large firm employees do not seem

to be choosing to move internally as a substitute for inter-firm mobility,13 though the results in

°One possible reason for the lack of a strong internal mobility effect on either turnover or the size effectsis that the current specification of the internal mobility measures do not take into account the "distance" ofthe job change, i.e. job changes and promotions that are relatively small may have little effect while largechanges may have stronger effects on both mobility decisions per se and the residual employer size effect.Alternatively stated, one employee may have relatively few job changes, but they may be significant changes,while another employee may have a large number of insignificant job changes. Although the latter employeehas a greater measured amount of internal mobility it may have less influence on mobility behavior than the

13

Table 2.3 based on splitting the sample on the basis of internal mobility are consistent with stronger

firm size effects on turnover for those who have moved within the organization than for those who

have not.

(ii) Survival Probabilities

Lower failure rates for larger firms have been documented in a number of studies (for

example, see Dunne et al., 1989; Mayo and Murray, 1991). As noted above, a greater expected

duration of operation will directly reduce turnover, and will exercise an indirect influence by raising

the expected returns to investment in workers. This secondary effect will produce higher volumes

of OJT in larger firms and accompanying practices to protect these investments such as wage premia,

provision of nonvested fringe benefits, and greater attention to employee selection with regard to

anticipated productivity in training (hence accounting for higher average schooling levels in larger

plants and firms) and the likelihood of remaining with the firm.

Since it is precisely these secondary effects that define the basis for long-term employment

relationships in larger firms, we would expect that introducing a measure of failure probabilities in

tenure (turnover) regression that only contain controls for employer size would act to reduce the size

effect, and similarly that in a simple regression of tenure (turnover) on failure probabilities the

introduction of employer size controls will act to reduce the estimated negative (positive) effect of

failure probabilities on tenure (turnover). In addition, controls for the attributes of employees

should further reduce the estimated effect of failure probabilities since controlling for attributes

yields a net effect of failure likelihood, i.e. the direct effect.

measured amount of internal mobility it may have less influence on mobility behavior than the formeremployee. This possibility would act to significantly weaken the ability of the current measures to test theinternal mobility hypothesis (in addition to likely measurement errors acting to downwardly bias our parameterestimates and hence lower estimated significance levels).

14

In order to empirically evaluate these propositions, failure probabilities are proxied by

industry failure rates taken from Dun and Bradstreet's annual publication, the Business Failure

Record, which provides information on the failure rate per 10,000 firms by 43 industry groups.14

Average failure rates over a number of years prior to the actual mobility decision were calculated

and matched to the respondents industry group (for the QES the average of 1970 through 1972 is

used, for the CPS the average of 1973 through 1978 is used, and for the NLS the average of 1970

through 1975 is used). Though prior failure rates were employed in an attempt to abstract from

contemporaneous contractions that might initiate turnover for reasons other than a lack of bonding

between workers and firms, results were largely similar when average failure rates were used that

included the years during which the turnover decisions were being undertaken (results are available

on request).

Tables 3.1 and 3.2 report the results of this analysis: Table 3.1 investigates the relationship

between employer size and failure rates, while Table 3.2 looks at the effects of failure rates on

tenure and turnover, and on the employer size effect on tenure and turnover. Looking at Table 3.1,

we observe in the first column a significantly inverse (unstandardized) relationship between failure

rates and both plant and firm size, with a generally stronger plant size effect than firm size effect.

When controls for the characteristics of respondents (and locational attributes such as region and

residence in an SMSA) are added, in an attempt to capture unobserved heterogeneity in firms as

reflected in the characteristics of the employees that they hire, we see that the plant size effect is

diminished in all three data sets, though the effects remain statistically significant. Further, in the

CPS the firm size effect is also diminished (though remains significant), while in the NLS the firm

size effect increases. A reduction in the employer size effect is what would be expected to the extent

14A more disaggregated vector of failure rates would clearly be desirable, but unfortunately failure ratesat the three digit industry level were only reported by Dunn and Bradstreet starting in 1984.

15

that higher levels of labor quality and productivity, as reflected by employee characteristics, in larger

plants and firms act to raise the survival likelihood of these organizations, though it is unclear what

is producing the anomalous effect evident in the NLS results. In the third column two-digit industry

controls are introduced and have a rather drastic effect on the employer size parameters, driving

plant size effects to insignificance in all three data sets, reducing the firm size effect in the NLS by

two-thirds, and actually changing the sign of the firm size effect in the CPS to a significantly positive

value. A portion of these effects may be due to the close association between the industry-based

failure rate measures and the industry dummies, and to the positive correlation between average

employer size and industry.15 Nevertheless, it does seem that within broad industry groups there

is no relationship between plant size and failure rates, though the inverse relationship between firm

size and failure rates does persist. The positive relationship between firm size and failure rates

observed in the CPS is difficult to understand, although part of the explanation might lie in increased

foreign competition as we entered the 1980's impacting most strongly on relatively larger and older

firms with outdated capital stocks.

Table 3.2 reports the effects of inclusions of failure rate probabilities on tenure, turnover,

and employer size effects. Looking first at tenure regressions in the CPS, industry failure rates have

a significantly negative effect on tenure, yet the significance level drops to 15% when SIC controls

are added, though this drop in significance is most likely due to collinearity between the industry

based failure rates and the SIC dummies.16 More importantly, inclusion of failure rates strongly

reduces the employer size effects between columns 1 and 3 (13.2% for plant size and 13.5% for firm

15Though, note that the precision of the employer size estimates increase upon the introduction of theindustry dummies, a result at variance with a multicollinearity based explanation.

16Note that between columns 5 and 6 the standard error on the failure rate effect nearly doubles and theparameter estimate only weakly declines; note that the -0.174 value in column 6 would still be significant at1% if the standard error in column 5 was used to compute P-values.

16

size), though when controls are included in the equation the reduction in the size effects are more

modest (7.1% and 2.3%, respectively, between columns 4 and 5). It is also interesting to note that

inclusion of employer size effects in column 3 reduces the failure rate effect by 31.7%, and that

controlling for respondent characteristics reduces the failure rate effect by 55.1% between columns

3 and 5 - indicating that even after size is taken into account, a substantial portion of the failure rate

effect on tenure is due to the association of failure rates with employee attributes that increase

tenure per se. The QES tenure regressions show a similar pattern of failure rates reducing the

employer size effects, yet unlike the CPS, failure rate effects are consistently insignificant and

actually increase when size effects and individual characteristics are taken into account, though we

cannot reject in any of these specifications the hypothesis that the failure rate effect is equal to zero.

Finally, looking at the NLS tenure regressions we observe patterns more closely akin to those in the

CPS: failure rate effects are significantly negative except when SIC controls are present, inclusion

of both employer size and individual characteristics reduce the failure rate effect, and inclusion of

failure rates generally reduce the effects of employer size (though the reduction is less pronounced

than in the CPS).

Looking at total separations we see that the failure rate has no effect on separations in the

QES, and essentially no influence on the magnitude of the employer size effect; similar patterns are

seen in the NLS, though the raw failure rate effect is significantly positive in column 2 and is driven

to zero when employer size effects are introduced. Distinguishing between voluntary and

involuntary turnover, results from the QES are rather surprising in that the failure rate effect is

negative and both the employer size and failure rate effects increase in magnitude when the other

effect is controlled for. Results from the NLS are more in accord with expectations in that the

failure rate has a significantly positive effect on quits, but as expected the effect is strongly reduced

when employer size is taken into account and is driven to zero when individual attributes are

17

controlled for. In addition, inclusion of failure rates moderately reduce the employer size effect.

Finally, looking at layoffs in the QES we observe significantly positive failure rate effects that

strongly decline with the introduction of both employer size and individual characteristics; similarly

the introduction of failure rates act to significantly reduce the estimated employer size effects. In

the NLS the results are less in accord with expectations since the failure rate appears to bear no

relationship to layoff probabilities, and acts to somewhat increase the employer size effect.

In sum, the results of this exercise provide somewhat mixed evidence across data sets, though

there is some support found for the predictions that failure rates act to reduce the estimated effect

of employer size on tenure and turnover and that standardizing for employee characteristics reduces

the estimated effect of failure rates on tenure and turnover.

(iii) On-the-Job Training

A number of studies have found higher levels of OJT (Barron, et al., 1987,1989) and search

investments (Barron, et al., 1985; Barron and Bishop, 1985) in large establishments. Numerous

explanations for these findings have been offered, including (1) the role of human and physical

capital complementarities (Griliches, 1969) in conjunction with the greater amounts of physical

capital per worker in larger plants,17 (2) more frequent introduction of new technology in larger

firms requiring recurrent OJT (Dunne, 1991) and the possibility that large firms use more customized

physical capital so that any given volume of OJT has a greater specific element (Oi, 1983), (3) a

lower level of risk associated with OJT investments in larger firms (Holtmann and Idson, 1991), and

(4) the role of internal labor markets in facilitating OJT (Idson, 1989), to mention only a few.

Regardless of the reasons for higher levels of OJT in large firms, so long as at least part of the

17Barron, et al. (1987), using the 1982 EOPP employer file, have shown that the cost of the most expensivemachine used by a worker is greater in larger firms; Idson and Feaster (1990) have shown with industry datathat larger firms have higher capital-labor ratios.

18

training is specific to the employer we know that workers and employers will share the costs of and

returns to the training, with a resulting decline in both quits and layoffs (Oi, 1962; Becker, 1975;

Parsons, 1972; Pencavel, 1972).

Tables 4.1-4.4 attempt to evaluate the influence of OJT on size differentials in mobility using

the QES, NLS and EOPP data sets. Looking first at the QES estimates in panel (a) of Table 4.1,

we see that the presence of an employer provided training program (TRN) is associated with higher

tenure and acts to reduce the plant size effects by 23%, driving the residual size effect to

insignificance (from significance at 1% to only 15% significance).18 Although these size results are

quite supportive of a training explanation of plant size effects on tenure, the results for turnover are

less supportive. For all three measures of turnover, TRN has the predicted negative sign but is not

significantly different from zero. Inclusion of TRN does reduce the employer size effects in all three

regressions, by 6.8%, 9.2%, and 10% for separations, quits, and layoffs, respectively, and reduces the

significance of the size effects for quit and layoffs from 5% to 10% levels, but in all three cases the

plant size effects remain significant at conventional levels.

Panel (b) of Table 4.1 employs the second wave (1982) of the EOPP employer survey to

investigate the effects on the employer size parameters of seven measures of training investments

in the last employee hired by the firm during his first three months on the job. We see that none

of the training variables significantly affect any of the turnover measures (this is also true for the

joint significance of the OJT vector - likelihood ratio test results available on request), except for

a positive effect of OJTE1 in the layoff regression, and have little effect on the size or significance

18In regressions not reported, the presence of an employer provided training program was significantly morelikely in larger plants even after controlling for personal characteristics, occupation, and industry (resultsavailable on request).

19

of the employer size effect.19 Although Barron, et al. (1987, 1989) demonstrate these OJT

measures, specified as dummies, are greater in larger establishments,20 the insignificant effects on

both turnover and the employer size effects might partly be due to (1) the fact the size distribution

of firms in the EOPP data is highly skewed toward small firms, and (2) the fact that the training

measures are only for OJT received during the first three months of employment - one of the effects

of long-term employment relationships in large firms is that training will be an ongoing process

through the worker's employment history, so that the positive effects of employer size on training

will be systematically understated by these EOPP measures.

Table 4.3 reports similar experiments using the NLS, which allow us additionally to look at

the differential impact of training measures on plant and firm size effects; Table 4.2 presents

evidence on the relationship between these NLS training measures and employer size. In Table 4.2

we see that the probability of receiving some on-the-job training in the year previous to the 1976

survey that was not provided through a "company training school", OJTN1, is weakly greater in larger

firms (at 16% significance) when no controls are used, yet becomes unambiguously insignificant

(though still positive) when controls are introduced. In columns 3 and 4, though, we see that the

probability of receiving some on-the-job training in the year previous to the 1976 survey that was

provided through a "company training school", OJTN2, is significantly, positively related to firm size.

Similarly, the amount of training provided, not simply the probability of receiving some training, is

greater in larger firms as seen in the last four columns; both hours per week spent on training,

^Alternatively measuring OJTE1-5 as dummies =1 if the OJT is positive does not affect the generalconclusions from these regressions.

authors argue for the use of dummies based on an assumed higher efficiency in the investmentprocess in larger firms yielding indeterminant predictions with regard to volumes of time investments. Whenspecified as levels, I found that, as predicted, these variables do not bear any relationship to employer size(results available on request).

20

OJTN3, and the number of weeks in training, OJTN4, are greater in larger firms.21

Looking at the turnover regressions in Table 4.3 we see that, unlike the QES and EOPP

results, the turnover regressions produce stronger support for a training interpretation of employer

size effects on turnover. Some of the OJT measures exert significantly negative effects on

turnover,22 the most consistently significant effects being training at a company training school,

OJTN2, and hours per week spent in an occupational training program during the last year, OJTN3.

Inclusion of OJTN1 and OJTN2 reduce the firm size effect on separations by 15% and reduce the

effect from 1% to 5% significance; these training dummies additionally drive down the firm size

effect on quits by 17% from significance at 5% to 15% significance and that on layoffs down by 14%

to insignificance. Similar, though far less dramatic results are found when the other training

measures, OJTN3 and OJTN4, are introduced.

It is also interesting to notice that the effect of worker participation in a training program

at a company training school, OJTN2, had a consistently larger and more significant effect, than

participation in an occupational training program that was provided elsewhere than a company

training school, OJTN1. As seen in Table 4.2, OJTN1 was not related to firm size, yet OJTN2 was

significantly and positively related to firm size.23 One might interpret these findings to simply

reflect economies of scale in provision of OJT in larger firms leading to provision of training on-site,

and not necessarily greater specificity in training at larger firms - although economies of scale

^Although the OJTN3 and OJTN4 regressions were estimated with OLS, the same results follow when atobit procedure is used (respondents who received no training in the previous year are assigned the value zero)the sign and significance pattern is essentially unaltered.

likelihood ratio tests for joint significance indicate that the training vectors in the separations regressionsare significant at 5% in all three columns, and for both quits and layoffs the first column of training measuresare jointly significant at 10%, but the training vectors in the second and third column specifications are onlyjointly significant at approximately 20% significance.

^Evidence from the Survey of Income and Program Participation also reflect these training patterns -Haber (1988) found that workers at small firms are more likely to receive training while on the job that isprovided at a different location than where they work.

21

considerations may facilitate provision of training with greater firm specificity. Nevertheless, the

results in Table 43, that OJTN2 significantly reduces total turnover likelihood, and both quits and

layoffs, indicates that training provided at company schools most likely reflects greater firm specificity

than training provided in another venue.

An alternative test of the influence of OJT on employer size patterns in mobility, that does

not rely on direct OJT measures, is to evaluate the differential impact of tenure in different size

plants and/or firms on turnover. To the extent that tenure proxies specific capital investments (see

Mincer and Jovanovic, 1981), greater training levels and/or greater specificity of training in larger

plants or firms would predict stronger inhibiting effects of tenure on mobility in larger firms,

producing a negative coefficient on the interaction of employer size and tenure. The results from

this approach are reported in Table 4.4 where an interaction term between tenure and establishment

size is introduced to allow for differential mobility-tenure profiles in different size plants and firms.

The consistently negative interaction term in panels (a) and (b) shows that tenure has a greater

(negative) effect on all dimensions of turnover in larger plants, for both the EOPP and QES samples,

with significant effects for layoffs in the QES and for separation and quits in the EOPP.

Furthermore, allowing the profiles to differ by establishment size acts to strongly reduce the additive

size effect in the QES, drive the effect on layoffs in the EOPP to insignificance, and actually cause

a change in the sign of the employer size effect for separations and quits in the EOPP (though

combining the additive size effect and the interaction effect yields a negative employer size effect

at tenure levels of about six months and greater).24 The results for the NLS reveal a similar pattern

of consistently negative interactions and corresponding reductions in the firm size effects

(interactions with plant size were experimented with, but were found to be consistently insignificant,

though standard errors increase when the interaction is introduced, indicating some degree ofcollinearity being induced, the additive size coefficients still consistently decline.

22

which is not surprising since the additive plant size effect was never found to be significant in the

NLS), though the interaction is only significant in the case of quits.

Thus, as with the results in Tables 4.1 and 4.3, the results in Table 4.4 do provide supportive

evidence for the important role played by on-the-job training in explaining employer size differences

in labor mobility.

(iv) Structure of Compensation

As discussed in Section III, higher levels of OJT lead employers to structure compensation

so as to inhibit costly mobility. Furthermore, to the extent that larger employers have a more

difficult time monitoring employees than do small employers, they may rely to a greater extent on

wage premia to raise the cost to the worker of being dismissed if caught shirking (similar

considerations would lead larger employers to more strongly backload pay and place a higher

proportion of pay in the form of nonvested fringe benefits). Thus, while linked to OJT and/or

monitoring costs, employer size differentials in both the level and structure of compensation may

constitute an independent factor that accounts for turnover differentials. Since we want to

investigate the effect of wage premia on turnover, in the empirical work below wages are entered

in conjunction with productivity (human capital proxies), so that the resulting wage effect gives the

influence of wages net of these factors, i.e. an estimated "wage premium" (similarly for fringe

benefits).

Table 5.1 reports the results of adding wage and fringe benefit variables in tenure regressions

for the CPS; Tables 5.2 and 5.3 similarly look at wage and fringe benefit effects in tenure and

turnover regressions for the QES and NLS, respectively. Looking first at the tenure results for the

CPS, we see that inclusion of wages reduces the plant and firm size effects by 22.6% and 26.1%,

respectively, though both effects remain statistically significant. Alternatively, adding two fringe

23

benefit availability measures (dummies for the presence of retirement plans and medical insurance)

also reduces the plant size effect (though less drastically), and drives the firm size effect to

insignificance. When wages and fringes are included together the results are similar - the plant size

effect falls by 25.3% and the firm size effect becomes insignificant. Similar patterns are evident in

panels (a) of Tables 5.2 and 5.3, though in the NLS both plant and firm size effects are driven to

insignificance (these results follow strictly from reductions in the estimated parameters, as the

standard errors remain essentially constant).

Looking at the turnover regressions for the QES in Table 5.2, we see that controlling for

earnings reduces plant size effect on separations, quits and layoffs by 11.3%, 10.3%, and 19.3%,

respectively, though except for a reduction in significance from 5% to 10% in the quit regressions,

the plant size effects maintain their initial significance levels. Fringe benefits have a more dramatic

effect on the employer size effects, reducing the effect on separations by 27.8% (and the significance

level to 5%), and driving the plant size effects on quits and layoffs to insignificance. Similar

patterns, though more pronounced, are observed in the last column when both wages and fringe

benefits are included.25 Comparable, though even more pronounced results are reported in Table

5.3 for the NLS. Inclusion of wages and fringe benefits acts to reduce both plant and firm size

effects to insignificance, and in this case this is true for total separations in addition to quits and

layoffs (similar patterns of individual and joint significance for wage and fringe benefit effects are

also observed).

In sum, the results reported in these tables indicate that both the level and structure of

compensation account for a good deal of employer-size differentials in turnover, a result consistent

^Likelihood ratio tests for the joint significance of the two fringe benefit variables indicate greater jointsignificance than individual significance; for the second specification in panels (a) and (b) the fringe benefitvector is jointfy significant at 5%, while in panels (c) and (d) joint significance is achieved at 10% (see Mitchell,1982, for a more extensive analysis of fringe benefits on quits per se, using the QES panel).

24

with both a desire to protect OJT investments and greater monitoring difficulties in larger plants and

firms inducing large employers to attempt to reduce worker malfeasance through raising the costs

to workers of dismissals.

(v) Heterogeneity

To the extent that large employers select workers with the intent of providing greater training

and establishing long-term employment relationships, these employers may attempt to hire workers

with lower (anticipated) turnover propensities. Similarly, workers who desire to undertake greater

volumes of OJT, and those who have a low tendency to move (say people who are particularly risk

averse and hesitant to change employers in search of a superior job) may have a preference for

employment at larger firms. Thus, a factor such as differential OJT patterns may generate

heterogeneity in the types of employees hired in small and large firms. To the extent that this is

true, employer size per se might not cause lower turnover, but may rather simply act as a screening

mechanism for more stable employees. Namely, if we placed a representative sample of large firm

workers at a small firm there may be little difference in turnover behavior when they are compared

to a comparable sample of workers who are employed by the large firm.

While personal characteristics such as education and experience levels may capture some of

this heterogeneity among worker, it is likely that they are insufficient proxies. A preferred approach

employed by Mincer and Jovanovic (1981) is to introduce into the turnover regressions a measure

of prior inter-firm mobility, along with individual characteristics, in an attempt to capture additional

elements of worker heterogeneity in turnover propensities. Results from this approach, using the

NLS, are reported in Tables 6.1 and 6.2; Table 6.1 investigates the relationship between two prior

mobility measures and both the plant and firm size at which the worker is currently employed, while

Table 6.2 investigates the effect of including these prior mobility measures in tenure and turnover

25

regressions.

Looking at Table 6.1 we see that as the level of prior mobility (either the total number of

inter-firm moves, or the number of inter-firm moves to different employers, between 1966 and 1976)

increases, the probability of being in a large plant or firm significantly decreases (with a stronger

effect being seen to operate at the firm level than at the plant level).26 These results are consistent

with large employers screening worker on mobility tendencies and choosing to hire those workers

who exhibit greater stability. While small employers might also prefer such employees, higher levels

of planned OJT and a desire to establish long-term employment relationships lead larger firms to

invest greater amounts in screening applicants (see Barron and Bishop, 1985 and Barron, et al., 1985

for evidence that larger firms more intensively and extensively screen applicants).27

Table 6.2 reports the effects of including each of the heterogeneity measures in tenure and

turnover regressions. To the extent that screening on turnover propensities by large firms is

producing a substantial portion of the observed positive (inverse) relationship between size and

tenure (turnover), and to the extent that the empirical proxies capture this aspect of employee

heterogeneity, then inclusion of the prior mobility variables should reduce the estimated size effects.

Looking first at the tenure regressions, we see that the prior mobility variables (1) have the expected

negative effect on tenure with the current employer, and (2) act to strongly reduce the magnitude

of the plant size effect, and drive the firm size effect to insignificance. While these results indicate

26The dependent variable is coded 0 through 4 in increasing order of size, and is thereby estimated witha maximum likelihood ordered probit procedure. Similar results were found when the dependent variable,plant or firm size, was coded as a continuous variable and the relationship was estimated by OLS (the effectsof prior mobility were largely unaffected by choosing either a linear or logarithmic specification of employersize - results available on request).

^It may be the case that part of the employer size effect on wages is a result of larger firms paying apremium for more stable workers, though to the extent that such workers also prefer the more stable workenvironment of larger firms, and also desire to invest in a long-term employment relationship, a wage premiummay not be necessary to attract these workers to large firms.

26

that worker heterogeneity in mobility propensities plays an important role in explaining the positive

relationship between employer size and tenure, we also see in panels (b)-(d) that the prior mobility

variables never achieve conventional levels of significance in turnover regressions, and essentially

have no effect on the estimated employer size effects.

In sum, though the results clearly indicate that larger employers are selecting workers with

lower mobility propensities, the evidence is mixed concerning the effects of employee selection on

long-term attachments between workers and large firms.

V. Working Conditions, Monitoring, and Threat Effects

A number of additional explanations have been advanced to account for employer-size wage

effects which similarly have implications for employer-size turnover effects, three of which will be

empirically evaluated in this section.

(i) Working Conditions

Researchers have shown that large and small employers differ with regard to nonpecuniary

aspects of their work environment (see Scherer, 1976; Dunn, 1980 and 1986; Kwoka, 1980; Brown

and Medoff, 1989), with the inference that higher wages in larger firms might constitute a

compensating differential for an inferior working environment.28 These studies generally find no

support for the compensating differentials explanation of employer size wage premiums, and, in fact,

find that many nonpecuniary aspects of work in larger firms are better than in small firms. Hence,

it is possible that these factors may be responsible for lower overall turnover by generating

differential quit rates, but could not account for lower layoff rates unless we believe that superior

Idson (1990) for an investigation of the effects of the work environment in different size plants onreported levels of job satisfaction, and see both Ehrenberg and Schumann (1982) and Idson and Robins(1990a) on the relationship between employer size and mandatory overtime rules.

27

working conditions lead workers to shirk less often either due to greater fear of job loss or greater

job satisfaction.

Table 7 reports the effects of nonpecuniary job attributes on turnover using the QES panel.

We see that although the work environment variables are all insignificant (a likelihood ratio test for

joint significance also failed to reject the hypothesis that the effects were jointly equal to zero),

inclusion of the vector of job attributes in a quit regression decreases the plant size effect by 12.7%,

reducing it from significance at 10% to statistical insignificance at conventional levels. These results

provide some support for the hypothesis that superior working conditions in larger firms act to

reduce quits, though a simple reduction in significance from a 10% to a 15% significance level does

not constitute very stark support for the working conditions hypothesis.

(ii) Monitoring

Difficulties in monitoring associated with larger organizations have been widely invoked as

an explanation of employer size differences in the types of employees hired and in the wages paid

these employees.29 Brown and Medoff (1989) found significantly positive size effects for piece rate

workers, which is rather strong evidence against monitoring explanations of size differentials in

wages; similarly, Kruse (1992) found little support for the monitoring explanation of employer size-

^Concerning the implications of monitoring costs for the types of workers employed at different size firms,Stigler (1962) speculated that the most dynamic and talented people should locate at smaller firms becausetheir talents and accomplishments may not be as easily recognized at larger firms. Oi (1983) and Garen(1985), though, predict that larger firms will hire more productive workers as a way to reduce the costs ofmonitoring, an argument that squares better with the observed higher average education levels and volumesof on-the-job training in large firms. The hypotheses of Oi and Garen imply that workers will be sortingthemselves across different size firms according to their comparative advantage in earnings (Willis and Rosen,1979), which implies in the context of selectivity corrected wage regressions positive signs for selection terms,while Stigler's argument would predict negative selection terms. Using a two-stage econometric procedure thattreats firm size as endogenous, Idson and Feaster (1990) find significant negative selectivity - patterns in thecovariances of the errors from the employer size selection equation and the wage equations are shown tosupport an explanation based on unobserved ability differentials which reconciles Stigler's ideas with observedhigher education and training levels in large firms.

28

wage effects - while reported frequency of supervision was negatively related to pay, it made no

difference to the employer size effect. Rebitzer and Robinson (1991), though, find that the employer

size effect on wages is considerably larger in primary than secondary sector jobs, yet still evident in

both sectors, which is consistent with the predictions of effort control models.

To the extent that collecting information on worker behavior is more difficult (and hence

costly) in large work environments, supervision models predict that large employers will structure

compensation to increase the cost of dismissals, thereby accounting for both higher wages and

(nonvested) fringe benefits in larger firms. Of course, the same size pattern in the structure of pay

is predicted by larger employers structuring compensation so as to protect higher levels of OJT

investments. The two models are therefore not cleanly distinguished by looking at size effects on

the levels or structure of compensation, or for that matter by the results reported above on the

effects of wages and fringe benefits on the employer size effect on tenure and turnover.

Nevertheless, to the extent that monitoring difficulties lead large employers to substitute

higher penalties for worker malfeasance in place of higher detection rates, one would expect to see

these practices reflected in lower layoff rates in larger firms as detection rates of worker malfeasance

decline with size (though, differential quit rates would also follow from wage premia induced by

monitoring difficulties). Of course, to the extent that dismissal threats are effective we might not

observe significant differences in employer size patterns in layoffs, i.e. a lack of employer size effects

on layoffs may be the result of monitoring factors, rather than evidence militating against the

likelihood of their importance. Hence, the mere presence of employer size effects on layoffs cannot

constitute an unambiguous test of the importance of information cost models.

An alternative approach to assessing supervision considerations is to look at differences in

plant and firm size effects on layoffs. To the extent that monitoring is performed mostly at the plant

level, monitoring explanations would predict that layoff rates will decline more strongly with plant

29

than with firm size. Of course, monitoring difficulties might not be strictly a function of plant size

since a one plant firm of, say, size 100 might have lower monitoring difficulties than a multi-plant

firm of size 100 with, say 10 plants of 10 workers each. Namely, monitoring difficulty may rise with

firm size, holding plant size constant, since larger firms (plant size constant) will tend to be positively

correlated with more plants, resulting in inter-plant monitoring problems. In fact, the finding in

Tables 1.1 and 1.2 that employer size patterns in tenure and turnover are stronger at the plant level

than at the firm level may be deceptive since the observed plant size patterns do not take into

account the size of firm that they are part of, and the firm size patterns do not take account of

whether or not we are observing a single-plant or multi-plant firm. The multivariate results in the

layoff regressions for the NLS reported above, though, control somewhat for this problem in that

both plant and firm size effects are estimated in the same regression so that the plant size

coefficients will not additionally reflect the influence of the size of firm that the plants are a part of.

Admittedly, though, these controls are not perfect since we still do not know how many plants

comprise a given firm, and no explicit controls are introduced for the presence of multi-plant versus

single-plant firms.

The NLS allows us to test some of these implications of the monitoring hypothesis, since

turnover is separately reported for voluntary and involuntary turnover and at both the plant and firm

level; Table 8 reports the results of this exercise. Looking at the first column of unstandardized

results we see that there is no independent plant size effect, but there is a significant firm size effect.

Similarly, when controls are introduced (in column 2) the plant size effect remains insignificant while

the firm size effect also becomes insignificant at conventional levels, but does achieve 15%

significance (though when the industry dummies were excluded from the regression, the firm size

effect did remain significant at 10%).

To test the idea that policing worker behavior may be at least as great a function of firm size

30

as plant size in cases where larger firms are composed of numerous plants, it would be desirable to

introduce a variable for the number of plants. In such a specification the (inter-plant) monitoring

hypothesis predicts that the firm size effect should decline (if not become insignificant), and the

number of plants variable should have a significantly negative effect on layoffs. Unfortunately this

variable is not available, so I attempt to proxy it by two variables (1) a multi-plant dummy (=1 if

firm size is reported to be greater than plant size) and (2) two alternative dummies that indicate if

the respondent is employed in a small plant relative to the size of the firm, which will generally

indicate employment in a setting that is not simply a multi-plant situation, but in which it is likely

that there are at least a few, and possibly many, plants comprising the firm. Columns 3 and 4 of

Table 8 report the results from this attempt to empirically investigate the influence of inter-plant

monitoring difficulties. Looking at columns 3 and 4 we see that while the multiplant dummy, and

both specifications of the "small plant - large firm" dummies, have the positive sign predicted by the

inter-plant monitoring hypothesis, in both specifications these dummies are individually insignificant

(though the multi-plant variable does achieve significance at the 15% level in column 3), and were

also found to be jointly insignificant even at the 20% level (likelihood ratio tests available on

request). We also see that the plant size effect remains insignificant, and that the firm size effect

increases drastically in magnitude and significance.

While these results cannot be seen as proof that supervision factors do not influence

employer size patterns in turnover (and wages), they are contrary to what we might expect to find

if inter-plant monitoring difficulties were the basis for the observed insignificant plant size effect and

significant firm size effects.

(iii) Threat Effects

It has also been hypothesized that large employers present more attractive targets for union

31

organizing drives, and as a result union threat effects are greater for large employers. Because of

this, large employers may be more likely than small employers to mimic the employment conditions

(both wages and employment stability) of union workers (see Antos, 1983; Podgursky, 1986). While

evidence reported in Brown and Medoff (1989) does not support this theory as an explanation of

size differentials in wages, it is still possible that greater employment stability is provided for workers

in order to preempt union certification.

Tables 9.1-9.3 investigate the union threat effect hypothesis in two ways, first by looking at

tenure patterns for groups with very small threat effects, and second by including a measure of the

percentage of the respondent's 3-digit industry that is unionized (see Freeman and Medoff, 1979)

in tenure and turnover regressions.

Following the approach used in Brown and Medoff (1989) to evaluate the threat effect

hypothesis for size differentials in wages, Table 9.1 presents tenure-size profiles for three groups that

ostensibly face low threats, (1) managers, (2) professional, technical, and kindred workers, and (3)

for people in 3-digit industry groups with less than 5% unionization rates. We see immediately that

both plant and firm size patterns in tenure are present for all three of these ostensibly low threat

groups. Further, if we compare the tenure-size profiles in Table 9.1 with those in Table 1.1 we see

that the firm size patterns are actually stronger for the low threat groups. Yet, since union

organization elections take place at the plant level, threats might tend to be more a function of plant

size than firm size, and we see that plant size effects on tenure are somewhat weaker among the low

threat groups, although a distinct plant size pattern is still evident. Hence, while the presence of

employer size patterns in tenure among the low threat groups is contrary to the predictions of the

union threat model, the weaker plant size pattern among low threat groups is consistent with the

predictions of the theory.

The last three columns of Table 9.2 extend this line analysis to the multivariate context.

32

Once controls for respondent characteristics, region, and industry effects are taken into account we

see that within the low threat groups the firm size effect disappears, except for managers where the

effect is larger than that for the entire sample (in the first column of the table). Further, contrary

to the patterns observed in the univariate profiles reported in Table 9.1, plant size effects are

uniformly higher among all three low threat groups than in the full sample, a result clearly at

variance with the predictions of the threat hypothesis.

As an alternative approach, columns 2 and 3 of Table 9.2 and Table 9.3 investigate the union

threat hypothesis of employer size differences in turnover by adding to the regressions a variable that

measures the percentage of the worker's 3-digit industry that is unionized, with the expectation that

plants and firms in more highly unionized industries are more greatly threatened by union

organization drives. In addition to allowing for an additive effect of this variable, interactions with

plant and firm size are included in order to allow the effect of plant and firm size per se to vary with

the level of unionization of the industry.

Looking first at the CPS tenure regressions in Table 9.2, we see in column 2 that tenure does

tend to be higher among workers employed in industries with a greater union presence (a result

consistent with the idea that union organization threats per se are affecting employer behavior), yet

inclusion of this measure in the regressions has almost no effect on the magnitude of the plant size

effect, and only causes a minor reduction in the firm size effect. When interactions are added in

column 3 we see that there is no differential effect of plant size on tenure in industries with a

differential union presence, yet consistent with the threat hypothesis we see that the firm size effect

on tenure is significantly higher in industries with a greater union presence.

Table 9.3 employs the same approach with turnover regressions using the QES. We see that

total separation and quits are lower in industries with a stronger union presence, a result once again

consistent with the general idea of union threats, though there does not appear to be any effect of

33

union presence on layoffs. These results seem to indicate that union threats do lead employers to

produce more desirable conditions of employment, but do not lead them to insulate their workers

more strongly from layoffs. We also see that while the plant size effect declines when we control

for union presence, the reduction is relatively small and the level of significance of the plant size

effects does not change in any of the turnover regressions - results inconsistent with the predictions

of union threats. Yet, the interaction effects reported in column 3 do indicate that plant size effects

on separations and quits are stronger when union presence in the industry is higher, yet only the

effect on quits is statistically significant at conventional levels (the interaction does, though, achieve

significance at the 15% level for total separations).

In sum, although some patterns in the data that are reported in this section are consistent

with the predictions of the union threat hypothesis, the preponderance of evidence does not support

a union threat explanation of employer size differences in either tenure or turnover.

VI. Conclusions and Directions for Future Work

Economists have long been interested in the structure and consequences of labor mobility

(see Parsons, 1977 for an early survey). Given the central role played by turnover in the efficient

allocation of labor market resources, and the costs to society of on-the-job training,30 employer size

effects on mobility are clearly of great potential importance and require detailed study. In fact,

investigation of the effects of employer size on labor mobility may be of particular importance today

given the structural shifts that have occurred in the U.S. economy over the last decade. While the

evidence is not definitive as to the relative share of small and large businesses in new job growth (see

Birch, 1981; Leonard, 1986; Brown, Hamilton, and Medoff, 1990; Davis, Haltiwanger, and Schuh,

1993), factors such as technological change, deregulation, and continued demographic changes in the

example, Mincer (1962) estimated that the average worker receives at least the equivalent of twoyears of college in the form of on-the-job training, with an annual cost of approximately 3% of GNP.

34

U.S. labor force (with attendant differential preferences for different types of work environments)

may well alter the size distribution of U.S. businesses and/or the demographic employment

distribution across different size businesses. The effect of different size employers on worker

mobility may therefore be an increasingly important factor in understanding the nature and efficiency

of the allocation of labor resources in the U.S. economy. This study has attempted to contribute to

our understanding of these complex issues.

The maintained hypothesis of this study is that large employers have an inherently greater

capacity to establish long-term relationships with their employees due to the larger job market within

the firm and their higher survival probabilities. As a result, the expected returns to on-the-job

training are greater, leading to higher levels of training, internal job mobility, employee selection with

an eye toward hiring people who exhibit both a capacity for training and relatively stable employment

histories, and mobility inhibiting wage premia and nonvested fringe benefits. The main findings in

the study are (1) although unstandardized mobility profiles indicate a stronger influence of plant size

than firm size, standardized estimates indicate a generally stronger and more significant firm size

effect, (2) the decline in turnover with employer size is predominantly produced by the inverse

relationship between employer size and quits, (3) internal mobility is greater in larger plants and

firms, though the firm size effect appears to dominate the plant size effect, (4) higher levels of

internal mobility are weakly associated with lower turnover, but do very little to account for size

differences in turnover, although the effects of firm size on separations and quits are only evident

for workers who have changed jobs with their current employer at least once, (5) higher levels of on-

the-job training in larger plants and firms are weakly inversely related to turnover in the QES and

EOPP data sets, though more strongly so in the NLS, yet the direct effects of higher levels of OJT

only slightly account for employer size differentials in mobility, (6) tenure profiles in turnover are

found to be generally steeper in larger plants and firms, a result consistent with greater volumes of

35

OJT occurring throughout the employment relationship in larger firms, (7) wage premia, and

especially the greater availability of fringes provided by large employer, largely account for employer

size differentials in tenure and turnover, (8) larger employers tend to select employees with lower

levels of prior inter-firm mobility, yet while prior mobility tendencies are negatively associated with

tenure and act to reduce the employer size effect on average tenure, they have little effect on either

turnover or the employer size effect on turnover, (9) nonpecuniary job attributes cause little variation

in the plant size effect on quits, (10) the dominance of firm size effects over plant size effects on

layoffs, combined with little support for an inter-plant monitoring difficulty explanation of these

patterns, cast doubt on a monitoring explanation of employer size patterns in mobility, and finally

(11) union threat effects are observed to influence tenure and turnover, though while some of the

results are mixed the general pattern of influence is not supportive of a union threat explanation of

employer size differentials in mobility.

Many additional questions arise surrounding the effects of employer size on labor market

dynamics. These questions include the effect of employer size on (1) transition rates to

unemployment, labor force withdrawal, and direct employer-to-employer moves without intervening

nonemployment states, (2) the likelihood of moves to different size employers (see Schiller, 1986)

and how these transition rates vary by the presence (and duration) of intervening states of

unemployment and/or spells out of the labor force, (3) the effect of employer size on the duration

of unemployment and periods out of the labor force, (4) how, if at all, the effect of employer size

on compensation, turnover patterns, and durations vary over the cycle, and how these patterns differ

between men and women and between different ethnic groups, and (5) lifecycle differences in the

size distribution of employment and the relationship between employee age and transitions across

different size firms.31

31A strong inverted U-shaped pattern between establishment size and age is present in the QES, suggestingthe possibility that workers begin their careers at relatively small firms, move to larger firms after some initialperiod of labor market experience, and then complete their careers at relatively smaller firms.

36

References

Antos, Joseph. "Union Effects on White-Collar Compensation." Industrial and Labor Relations Review36 (April 1983): 461-79.

Barron, John and John Bishop "Extensive Search, Intensive Search, and Hiring Costs: New Evidenceon Employer Hiring Activity." Economic Inquiry 23 (July 1985): 363-82.

, and W. C. Dunkelberg. "Employer Search: The Interviewing and Hiring of New Employees."The Review of Economics and Statistics 67 (February 1985): 43-52.

Barron, John, Dan Black, and Mark Loewenstein. "Employer Size: The Implications forSearch, Training, Capital Investment, Starting Wages, and Wage Growth." Journal ofLabor Economics 5 (January 1987): 76-89.

. "Job Matching and On-the-Job Training." Journal of Labor Economics 7 (January 1989): 1-19.

Barth, J., J. Cordes, and S. Haber. "Employee Characteristics and Firm Size: Are There SystematicEmpirical Relationships?" Applied Economics 19 (April 1987): 555-67.

Becker, Gary. Human Capital. Second Edition, The University of Chicago Press, Chicago, 1975.

Birch, David. "Who Creates Jobs?" Public Interest 65 (Fall 1981): 3-14.

Brown, Charles and James Medoff. "The Employer Size-Wage Effect." Journal of Political Economy97 (October 1989): 1027-1059.

, and James Hamilton. Employers Large and Small. Harvard University Press, Cambridge,Massachusetts, 1990.

Calvo, Guillermo. "Quasi-Walrasian Theories of Unemployment." A.E.R. Papers and Proceedings(May 1979).

Davis, Steve and John Haltiwanger. "Employer Size and the Wage Structure in U.S. Manufacturing."Working Paper, 1992.

, and Scott Schuh. Job Creation and Job Destruction in U.S. Manufacturing, forthcoming1993.

Dunn, Lucia. "The Effects of Firm and Plant Size on Employee Well-Being." in The Economics ofFirm Size. Market Structure, and Social Performance. John Siegfried (ed.), Washington D.C.,Federal Trade Commission, 1980.

Dunne, Timothy, Mark J. Roberts, and Larry Samuelson. "The Growth and Failure of U.S.Manufacturing Plants." The Quarterly Journal of Economics 104 (November 1989): 671-698.

Dunne, Timothy. "Technology Usage in U.S. Manufacturing Industries: New Evidence From theSurvey of Manufacturing Technology." Center for Economic Studies Discussion Paper 91-1,November 1991.

Evans, David, and Linda Leighton. "Why Do Smaller Firms Pay Less?" Journal of Human Resources24 (Spring 1989): 299-318.

Freeman, Richard, and James Medoff. "New Estimates of Private Sector Unionism in the UnitedStates." Industrial and Labor Relations Review 32 (January 1979): 143-74.

Garen, John. "Worker Heterogeneity, Job Screening, and Firm Size." Journal of Political Economy93 (August 1985): 715-39.

Grilliches, Zvi. "Capital-Skill Complementarity." Review of Economics and Statistics (November 1969).

Haber, S., J. Cordes, and J. Barth. "Employment and Training Opportunities in Small and LargeFirms." Report Prepared for the U.S. Small Business Administration, June 1988.

Holtmann, A. G. and Todd Idson. "Employer Size and On-the-Job Training Decisions." SouthernEconomic Journal (October 1991).

Idson, Todd. "Establishment Size Differentials in Internal Mobility." Review of Economics andStatistics 71 (November 1989): 721-24.

. "Establishment Size, Job Satisfaction, and the Structure of Work." Applied Economics 22(1990): 1007-1018.

, and Daniel Feaster. "A Selectivity Model of Employer Size Wage Differentials." Journal ofLabor Economics 8 (January 1990): 99-122.

Idson, Todd, and Robert Valletta. "The Effects of Sectoral Decline on the EmploymentRelationship." Columbia University Discussion Paper No. 654, May 1993.

Javonovic, Boyan. "Job Matching and the Theory of Turnover." Journal of Political Economy 87(October 1979): 972-90.

Kruse, Douglas. "Supervision, Working Conditions, and the Employer Size-Wage Effect." IndustrialRelations 31 (Spring 1992): 229-49.

Kwoka, John. "Establishment Size, Wages, and Job Satisfaction: The Tradeoffs." in The Economicsof Firm Size. Market Structure and Social Performance. John J. Siegfried (editor),Washington D. C: Federal Trade Commission, July 1980.

Lazear, Edward. "Agency, Earnings Profiles, Productivity, and Hours Restrictions." AmericanEconomic Review 71 (September 1981): 606-20.

Leonard, J. "On the Size Distribution of Employment and Establishments." NBER WorkingPaper, National Bureau of Economic Research, June 1986.

Lester, Richard. "Pay Differentials by Establishment Size." Industrial Relations 7 (October 1967): 57-67.

Masters, Stanley. "Wages and Plant Size: An Interindustry Analysis." Review of Economics andStatistics 51 (August 1969): 341-45.

Mayo, John and Matthew Murray. "Firm Size, Employment Risk and Wages: Further Insights on aPersistent Puzzle." Applied Economics 23 (1991): 1351-60.

Mellow, Wesley. "Employer Size, Unionism, and Wages." in Research in Labor Economics.Supplement 2, JAI Press, 1983.

Mincer, Jacob. "On-the-Job Training, Costs, Returns and Implications." Journal of PoliticalEconomy, (supplement) 1962.

, and Boyan Jovanovic. "Labor Mobility and Wages." in Studies in Labor Markets, (ed.)Sherwin Rosen, The University of Chicago Press (NBER), 1981.

Mitchell, Olivia S. "Fringe Benefits and Labor Mobility" Journal of Human Resources 18 (Spring1982): 286-98.

Oi, Walter. "Heterogeneous Firms and the Organization of Production." Economic Inquiry 21 (April1983): 147-71.

Parsons, Donald. "Specific Human Capital: An Application to Quit and Layoff Rates." Journal ofPolitical Economy 80 (December 1972): 1120-43.

Parsons, Donald. "Models of Labor Turnover: A Theoretical and Empirical Survey." in RonaldEhrenberg (editor), Research in Labor Economics. 1, JAI Press, 1977.

Pearce, James. "Tenure, Unions, and the Relationship between Employer Size and Wages." Journalof Labor Economics 8 (April 1990): 251-69.

Pencavel, John. "Wages, Specific Training, and Labor Turnover in U. S. Manufacturing Industries."International Economic Review 13 (February 1972): 53-64.

Podgursky, Michael. "Unions, Establishment Size, and Intra-Industry Threat Effects." Industrial andLabor Relations Review 39 (January 1986): 277-84.

Rebitzer, James. "Establishment Size and Job Tenure." Industrial Relations 25 (Fall 1986): 292-302.

, and Michael Taylor. "Employer Size and Dual Labor Markets." NBER Working Paper No.3587, January 1991.

Salop, Steven. "Wage Differentials in a Dynamic Theory of the Firm." Journal of Economic Theory(December 1973).

Scherer, F. M. "Industrial Structure, Scale Economics and Worker Alienation." in Essays in Honorof Joe S. Bain. R. T. Masson and P. D. Quails (eds.), Ballinger Publishing Co., 1976.

Schiller, Bradley. Early Jobs and Training: The Role of Small Businesses. Washington, D.C.,U.S. Small Business Administration, December 1986.

Stigler, George. "Information in the Labor Market." Journal of Political Economy 70 (October 1962):94-105.

Weiss, Andrew. "Job Queues and Layoffs in Labor Markets with Flexible Wages." Journal of PoliticalEconomy 88 (June 1980): 526-38.

and Henry Landau. "Wages, Hiring Standard, and Firm Size." Journal of Labor Economics 2(October 1984): 477-99.

Willis, R. and Sherwin Rosen. "Education and Self-Selection." Journal of Political Economy 87(October 1979): S7-S36.

Table 1: Variable Definitions

Variable Definition

1. Employer Size

P_Size plant size; number of employees (CPS, EOPP, NLS, QES)F_Size firm size; number of employees (CPS, NLS)

2. Turnover Variables

Tenure time employed by the respondent's employer at the time of the survey(CPS, EOPP, NLS, QES)

Separation = 1 if the respondent left his employer between surveys (for the QES thiscorresponds to an employer change between 1973 and 1977; for the NLS thiscorresponds to an employer change between the 1976 and 1977 surveys; for EOPPthis corresponds to whether or not the last employee hired prior to August 1981 isstill with the company); plant closures and seasonal and part-time jobs are excluded(EOPP, NLS, QES)

Quit = 1 if a separation occurs and a voluntary move is indicated (EOPP, NLS, QES)Layoff = 1 if a separation occurs and an involuntary move is indicated (EOPP, NLS, QES)

3. Internal Mobility

NUMJOB number of different jobs with current employer (QES)NPROMO number of different jobs with current employers that the respondent classifies as

promotions (QES)NDEMLAT number of different jobs with current employers that are not classified as promotions;

demotions or lateral job changes (QES)CPROMO how good the employee feels that the chances of promotion are (QES)YPROMO number of years until the employee expects to be promoted (QES)JOBTEN time on current job assignment with current employer (QES)CHJOB if changed jobs at least once with current employer; calculated as = 1 if tenure on

current job, or in current occupation, is less than tenure with current employer (NLS)OCCTEN months in current occupation with current employer (NLS)FRACOT fraction of tenure with current employer in current occupation (NLS)

4. On-the-Job Training

RQT number of weeks that the typical employee would require to become fully trained forthe job, if he has the requisite education (EOPP)

TRN = 1 if there is an employer provided training program (QES)OJTN1 =1 if participated in an occupational training program previous year that was not

given at a company training school (NLS)OJTN2 = 1 if occupational training program last year was at a company training school (NLS)

Table 1: Variable Definitions (continued)

Variable Definition

OJTN3 hours per week of occupational training program taken last year (NLS)OJTN4 weeks of occupational training program taken last year (NLS)OJTE1 hours spent away from normal work activities on orientation during the first three

months of employment (EOPP)OJTE2 hours spent away from normal work activities on training by watching others during

the first three months of employment (EOPP)OJTE3 hours spent away from normal work activities on formal training, such as self-paced

learning programs or training done by specially trained personnel during the firstthree months of employment (EOPP)

OJTE4 hours spent away from normal work activities by management and line supervisorsto give individualized informal training or extra supervision to the respondent duringthe first three months of employment (EOPP)

OJTE5 hours spent away from normal work activities by coworkers who are not supervisorsto give individualized informal training or extra supervision to the respondent duringthe first three months of employment (EOPP)

SCREEN total hours spent recruiting, screening, and interviewing all applicants for therespondent's position (EOPP)

5. Pay and Fringe Benefits

LNEARN log annual earnings (QES)LNWAGE log hourly wage (CPS, NLS)RET =1 if a retirement plan is available (CPS, NLS, QES)MED =1 if medical insurance is available (CPS, NLS, QES)

6. Nonpecuniary Job Attributes (QES): See Table 7

7. Heterogeneity in Mobility Propensity (NLS)

NEC the number of times the employee has changed employers, including returning to aprevious employer

NDE the same measure as NEC, but excludes returns to previous employers

8. Union Threats

%Union the percentage of the three digit industry that the respondent is employed in that isunionized; from Freeman and Medoff, 1979 (CPS, QES)

Table 1.1: Tenure-Size Profiles by Plant and Firm Size

(a) 1979 Current Population Survey

Size

Plant

Firm

1-24

7.67(7.70)[711]

7.04(6.91)[178]

25-99

8.73(8.52)[548]

7.52(8.35)[243]

100-499

9.87(9.16)[601]

7.22(7.05)[381]

500-999

10.46(9.13)[221]

9.24(9.14)[161]

1.000+

13.12(10.13)[564]

11.03(9.55)[1682]

% Change

71.05

56.68

(b) 1976 National Longitudinal Survey of Young Men

Size

Plant

Firm

1-24

3.47(3.49)[946]

3.29(3.61)[512]

25-99

4.22(3.59)[312]

3.56(3.40)[242]

100-499

4.75(3.78)[239]

4.02(3.56)[233]

500-999

5.27(4.28)[52]

4.52(3.70)[104]

1.000+

5.63(3.92)[87]

4.70(3.72)[535]

% Change

62.25

42.86

(c) 1973 Quality of Employment Survey

Size

Plant1=25.49

(8.24)[89]

10-49

5.89(7.39)[114]

50-99

5.94(7.96)[43]

100-499

7.26(7.68)[77]

500-999

5.94(4.70)[25]

1.000+

9.11(9.64)[58]

% Change

59.82

NOTE: Means are reported with standard deviations in parentheses. Sample sizes are listedin brackets. The % Change column calculates the percentage change in the means from the1-24 category to the 1,000 or more category, for panels (a) and (b), and between the weightedaverage of the two smallest categories (1-9 and 10-49) and the 1,000 or more category forpanel (c).

Table 1.2; Turnover-Size Profiles by Plant and Firm Size

(a) 1976 National Longitudinal Survey of Young Men

Plant

Separations

Quits

Layoffs

Firm

Separations

Quits

Layoffs

1-24(892)

0.429(0.50)

0.331(0.47)

0.077(0.27)

1-24(487)

0.474(0.50)

0.357(0.48)

0.092(0.29)

25-99(292)

0.312(0.46)

0.253(0.44)

0.058(0.23)

25-99(227)

0.436(0.50)

0.322(0.47)

0.106(0.31)

100-499(224)

0.308(0.46)

0.254(0.44)

0.045(0.21)

100-499(215)

0.335(0.47)

0.274(0.45)

0.042(0.20)

500-999(48)

0.188(0.39)

0.188(0.39)

0.00(0.00)

500-999(96)

0.250(0.44)

0.198(0.40)

0.052(0.222

1.000+(82)

0.183(0.39)

0.183(0.39)

0.00(0.00)

1.000+(503)

0.270(0.44)

0.239(0.43)

0.026(0.16)

% Change

-57.34

-44.71

-100.00

% Change

-43.04

-33.05

-71.74

(b) 1973-77 Quality of Employment Survey

Plant

Separations

Quits

Layoffs

1-9(89)

0.48(0.50)

0.31(0.47)

0.15(0.36)

10-49(114)

0.38(0.49)

0.30(0.46)

0.08(0.27)

50-99(43)

0.37(0.49)

0.33(0.47)

0.05(0.21)

100-499(77)

0.22(0.42)

0.21(0.41)

0.01(0.11)

500-999(25)

0.20(0.41)

0.20(0.41)

0.00(0.00)

1.000+(58)

0.17(0.38)

0.14(0.35)

0.03(0.18)

% Change

-59.52

-53.33

-72.73

NOTE: Means are reported with standard deviations in parentheses. Sample sizes arelisted in parentheses below each size category. For panel (a) the % Change columncalculates the percentage change in the means from the average of the 1-24 category to the1,000 or more category; for panel (b) the % Change is calculated between the average ofthe two smallest categories (1-9 and 10-49) and the 1,000 or more category.

Table 1.3: The Relative Contributions of Voluntary and InvoluntaryTurnover to Employer Size Patterns in Total Separations

(a) National Longitudinal Survey of Young Men, 1976-78

Plant

Firm

: Quits

: Layoffs

: Quits

: Layoffs

1-24

81.13

18.87

79.51

20.49

25-99

81.35

18.65

75.23

24.77

100-499

84.95

15.05

86.71

13.29

500-999

100.00

0.00

79.20

20.80

1.000+

100.0

0.0

90.19

9.81

(b) Quality of Employment Survey, 1973/1977

M 10-49 50-99 100-499 500-999

Plant: Quits 67.39 78.95 86.84 95.45 100.0

: Layoffs 32.61 21.05 13.16 4.55 0.0

NOTE: The cells contain the percentages of total separations (where the sum ofreported quits and layoffs is used as the measure of total separations) that are due toquits or layoffs, respectively.

Table 2.1: Effects of Employer Size on Internal Mobility

(a) 1973 Quality of Employment Survey (n=371)

ln(P_Size)

Tenure

Tenure2

NUMJOB

0.171"(0.044)

0.195"(0.035)

-0.004"(0.001)

NPROMO

0.128"(0.041)

0.171"(0.032)

-0.004"(0.001)

NDEMLAT

0.043"(0.018)

0.024c

(0.013)

-0.001(0.0005)

CPROMO

0.071b

(0.034)

0.003(0.027)

-0.0003(0.0009)

YPROMO

-1.127"(0.398)

0.339(0.317)

-0.004(0.011)

JOBTEN

-0.181c

(0.105)

0.756"(0.083)

-0.007b

(0.003)

(b) 1976 National Longitudinal Survey of Young Men (n=1,535)

CHJQB

In (P_Size)

In (FJSize)

Tenure

Tenure2

-0.062b

(0.030)

0.149"(0.025)

0.276"(0.028)

-0.009"(0.002)

OCCTEN

-0.034(0.040)

-0.092"(0.031)

0.887"(0.035)

-0.028"(0.003)

FRACOT

0.002(0.006)

-0.023"(0.005)

-0.046"(0.006)

0.001"(0.0004)

NOTE: Dependent variables are listed as column headings. All columns in panel (a) areestimated by OLS except for CPROMO which is estimated by a maximum likelihood orderedprobit (ranging from 0-3, indicating the respondents evaluation of how good the chances forpromotion are on his/her job). Each regression additionally includes a constant, quadratic inexperience, education, and dummies for SMSA, marital and health status, race, three regiondummies, 6 occupation dummies, and 11 industry dummies. Means and standard deviationsare: NUMJOB 1.129 (1.70), NPROMO 1.008 (1.57), NDEMLAT 0.121 (0.57), YPROMO13.836 (13.97), JOBTEN 4.252 (5.80), and CPROMO 1.712 (1.13). In panel (b), the CHJOBcolumn is estimated by ML probit, while the OCCTEN and FRACJT columns are estimatedby OLS. As in panel (a), each regression additionally includes a constant, quadratic inexperience, education, and dummies for SMSA, marital and health status, race, a dummy forresidence in the south, 7 occupation dummies, and 9 industry dummies. Means and standarddeviations are: CHJOB 0.375 (0.48), OCCTEN 2.567 (2.73), and FRACOT 0.780 (0.33). Inboth panels, superscripts a, b, and c denote significance levels of 1%, 5%, and 10%respectively.

Table 2.2:

In (P_Size)

Average Number ofJobs per Year

Average Number ofPromotions per Year

In (P_Size)

Average Number ofJobs per Year

Average Number ofPromotions per Year

In (P_Size)

Average Number ofJobs per Year

Average Number ofPromotions per Year

Effects of Intrafirm Mobility on Size EffectsQuality of Employment Survey (n=368)

(a) Separations

-0.1333

(0.043)

(b) Quits

-0.087b

(0.044)

(c) Layoffs

-0.150b

(0.076)

-0.1283

(0.043)

-0.130c

(0.076)

-0.084c

(0.044)

-0.072(0.064)

-0.134c

(0.079)

-0.656d

(0.421)

on Turnover

-0.130*(0.043)

-0.116d

(0.077)

-0.085c

(0.044)

-0.064(0.062)

-0.141c

(0.078)

-0.600(0.430)

NOTE: All regressions are estimated by maximum likelihood probit. Each regressionadditionally includes a constant, quadratics in tenure and experience, education, dummies forSMSA, marital and health status, race, three region dummies, nine industry dummies, and sixoccupation dummies. Parameter estimates are reported with standard errors in parentheses.Superscripts a, b, c, and d denote significance levels of 1%, 5%, 10%, and 15%, respectively.Average number of jobs or promotions per year are calculated as the total number of differentjobs, NUMJOB, or the number of these different jobs that the respondent indicates aspromotions, NPROMO, with the current employer up to the 1973 survey, divided by tenure in1973. The turnover variables are coded as =1 if the respondent left his 1973 employersometime between the 1973 and 1977 surveys. Means and standard deviations are: AverageNumber of Jobs per Year 0.497 (1.97), Average Number of Promotions per Year 0.449 (1.95).

Table 2.3: Effects of Internal Mobility on Size Effects on TurnoverNational Longitudinal Survey of Young Men (n=1,413)

In (P_Size)

In (F_Size)

CHJOB

FULL

-0.008(0.032)

-0.061a

(0.024)

SAMPLE

(a) Separations

-0.008(0.032)

-0.0603

(0.024)

-0.020(0.088)

CHJOB=0

-0.052(0.044)

-0.036(0.031)

CHJOB=1

0.036(0.048)

-0.0993

(0.041)

(b) Quits

In (P_Size)

In (F_Size)

CHJOB

In (P_Size)

In (F_Size)

CHJOB

0.021(0.033)

-0.039c

(0.024)

-0.063(0.056)

-0.057d

(0.038)

0.021(0.033)

-0.040°(0.025)

0.008(0.091)

(c) Layoffs

-0.064(0.056)

-0.054(0.039)

-0.065(0.141)

-0.022(0.045)

-0.008(0.031)

-0.052(0.076)

-0.082c

(0.049)

0.668(0.050)

-0.090b

(0.043)

-0.073(0.091)

-0.008(0.071)

NOTE: All regressions are estimated by maximum likelihood probit. Each regressionadditionally includes a constant, quadratics in tenure and experience, education, dummies forSMSA, marital and health status, race, a dummy for residence in the south, nine industrydummies, and seven occupation dummies. Parameter estimates are reported with standarderrors in parentheses. Superscripts a, b, c, and d denote significance levels of 1%, 5%, 10%,and 15%, respectively. The first two columns correspond to the entire (nonunion) sample, thethird column is restricted to respondents that did not change job or occupation with thecurrent employer since the last interview year, and the fourth column is restricted torespondents that did change job or occupation with the current employer since the lastinterview year. Mean (std. dev.) for CHJOB is 0.388 (0.487).

Table 3.1: The Relationship Between Employer Size and Failure Rates

(a) Quality of Employment Survey

In (P_Size)

In (P_Size)

In (F_Size)

In (P_Size)

In (F_Size)

-24.40"(6.14)

(b) Current

-21.65"(2.42)

-12.26"(2.97)

(c) National

-11.79"(2.92)

-9.76"(2.19)

-18.31"(6.80)

Population Survey

-16.35"(2.52)

-10.61"(2.94)

Longitudinal Survey

-6.40b

(2.92)

-12.68"(2.28)

3.84(4.97)

-1.28(1.66)

4.80"(1.87)

-1.66(2.38)

-4.07b

(1.85)

NOTE: All regressions are estimated by OLS. Parameter estimates are reported with standarderrors in parentheses. Superscripts a-c denote significance at 1%, 5%, and 10%, respectively.For each panel the first column gives raw employer size effects (the regressions include aconstant, but no controls), the second column includes a full set controls for personalcharacteristics of the respondents, including occupational dummies, and the last columnincludes the same set of controls as in the second column, but additionally includes controlsfor two-digit industry. The failure rate variables used are, (1) QES: the average for 1970, 1971and 1972, (2) CPS: the average for 1973 through 1978, and (3) NLS: the average for 1970through 1975.

Table 3.2:

In (P_Size)

In (FJSize)

FailureRate x 100

In (P_Size)

FailureRate x 100

In (P_Size)

FailureRate x 100

In (P_Size)

FailureRate x 100

In (P_Size)

FailureRate x 100

Failure Rate

0.795"(0.111)

0.444"(0.137)

0.428b

(0.189)

-0.152"(0.035)

-0.096"(0.036)

-0.171"(0.058)

Effects on 'Tenure, Turnover, and Employer Size

Current Population Survey: Tenure

-0.714"(0.086)

0.690"(0.112)

0.384"(0.137)

-0.488"(0.088)

0.533"(0.090)

0.399"(0.105)

0.495"(0.091)

0.375"(0.105)

-0.219"(0.069)

Oualitv of Employment Survey

-0.119(0.157)

0.015(0.027)

-0.034(0.029)

0.119"(0.038)

(a)1

0.396b

(0.193)

-0.134(0.159)

Tenure

0.560"(0.175)

(b) Separations

-0.154"(0.035)

0.008(0.028)

(c)

-0.106"(0.036)

-0.049c

(0.030)

(d)]

-0.146b

(0.060)

0.096"(0.039)

-0.134"(0.040)

Quits

-0.088b

(0.041)

Layoffs

-0.153b

(0.069)

0.521"(0.177)

-0.193(0.138)

-0.137(0.040)

-0.020(0.029)

-0.098b

(0.041)

-0.060c

(0.032)

-0.141b

(0.070)

0.068d

(0.044)

Effects (CPS, QES)

0.484"(0.094)

0.337"(0.106)

0.392b

(0.189)

-0.146"(0.044)

-0.101b

(0.045)

-0.141c

(0.077)

0.481"(0.094)

0.345"(0.106)

-0.174d

(0.109)

0.389b

(0.190)

0.064(0.204)

-0.145"(0.044)

-0.010(0.045)

-0.096b

(0.045)

-0.054(0.047)

-0.141c

(0.076)

0.071(0.079)

NOTE: Parameter estimates are listed with standard errors in parentheses. Superscripts a-ddenote significance levels of 1%, 5%, 10%, and 15%, respectively. Columns 1-3 contain nocontrols other than the listed regressors, columns 4 and 6 contain all standard controls except forindustry dummies, and columns 5 and 7 additionally contain industry controls (all regressionsinclude a constant).

Table 3.2

In (P_Size)

In (F_Size)

FailureRate x 100

In (P_Size)

In (F_Size)

FailureRate x 100

In (P_Size)

In (F_Size)

FailureRate x 100

In (P_Size)

In (F_Size)

FailureRate x 100

(continued)

0.346a

(0.076)

0.114a

(0.057)

-0.055b

(0.029)

-0.098a

(0.021)

-0.036(0.029)

-0.063a

(0.021)

-0.051(0.051)

-0.104a

(0.034)

: FailureNational

-0.2793

(0.068)

0.0753

(0.024)

0.073a

(0.025)

0.003(0.037)

Rate EffectsLongitudinal

(a)!

0.327a

(0.077)

0.129b

(0.058)

-0.160b

(0.069)

on Tenure, Turnover, andSurvey of Young Men

Tenure

0.232a

(0.072)

0.170a

(0.054)

(b) Separations

-0.051c

(0.029)

-0.0943

(0.021)

0.034(0.025)

(c)

-0.030(0.030)

-0.0593

(0.021)

0.048c

(0.025)

(d)J

-0.058(0.052)

-0.109a

(0.034)

-0.046(0.040)

-0.034(0.030)

-0.094a

(0.022)

Quits

-0.008(0.031)

-0.0693

(0.023)

Layoffs

-0.053(0.054)

-0.0863

(0.036)

0.221a

(0.072)

0.151a

(0.055)

-0.151b

(0.065)

-0.033(0.030)

-0.0913

(0.022)

0.021(0.026)

0.006(0.031)

-0.065a

(0.023)

0.031(0.027)

-0.060(0.054)

-0.0923

(0.036)

-0.052(0.042)

Employer

0.162b

(0.072)

0.097c

(0.056)

-0.023(0.030)

-0.081a

(0.023)

0.005(0.031)

-0.060*(0.024)

-0.069(0.055)

-0.069c

(0.037)

Size Effects

0.161b

(0.072)

0.094c

(0.056)

-0.076(0.080)

-0.023(0.030)

-0.081a

(0.023)

-0.008(0.033)

0.005(0.031)

-0.061a

(0.024)

-0.007(0.034)

-0.071(0.055)

-0.071c

(0.038)

-0.043(0.054)

NOTE: Parameter estimates are listed with standard errors in parentheses. Superscripts a-ddenote significance levels of 1%, 5%, 10%, and 15%, respectively. Columns 1-3 contain nocontrols other than the listed regressors, columns 4 and 6 contain all standard controls except forindustry dummies, and columns 5 and 7 additionally contain industry controls (all regressionsinclude a constant).

Table 4.1: Training Effects on Tenure and Turnover (QES, EOPP)

(a) Quality of Employment Survey (n=368)

In (P_Size)

TRN

Tenure

0.290d

(0.194)

1.350c

(0.748)

Separations

-0.124a

(0.044)

-0.147(0.169)

Quits

-0.079°(0.045)

-0.142(0.176)

Layofj

-0.135c

(0.78)

-0.293(0.303)

(b) Employment Opportunities Pilot Projects (n=820)

ln(P_Size)

OJTE1 x 100

OJTE2 x 100

OJTE3 x 100

OJTE4 x 100

OJTE5 x 100

RQT x 100

SCREEN x 100

Separations

-0.2593

(0.042)

0.197(0.375)

0.055(0.062)

0.068(0.162)

0.043(0.088)

-0.051(0.106)

0.061(0.120)

0.161(0.134)

Quits

-0.1733

(0.045)

-0.677(0.540)

0.008(0.068)

0.034(0.154)

0.088(0.095)

0.011(0.110)

-0.166(0.151)

-0.068(0.230)

Layoffs

-0.120c

(0.064)

1.265a

(0.432)

-0.043(0.094)

-0.024(0.223)

0.203d

(0.135)

-0.263(0.202)

0.008(0.181)

0.075(0.246)

NOTE: The first and second columns in panel (a) report OLS tenure regressions; all othercolumns report maximum likelihood probit estimates. Panel (a) regressions additionally includea constant, quadratics in tenure and experience, education, dummies for SMSA, marital andhealth status, race, three region dummies, nine industry dummies, and six occupation dummies;panel (b) regressions additionally include a constant, quadratics in tenure and experience, andeducation. Parameter estimates are listed with standard errors in parentheses. Superscripts a-ddenote significance levels of 1%, 5%, 10%, and 15%, respectively. Means (std. dev.) are: TRN.401 (.49), in panel (a); OJTE1 6.46 (14.3), OJTE2 57.68 (97.1), OJTE3 9.17 (38.3), OJTE450.93 (79.4), OJTE5 29.04 (64.8), RQT 28.20 (47.2), and SCREEN 14.96 (41.9) in panel (b).Baseline size effects (with all controls except the OJT measures, but restricted to a sample withnonmissing values for all size measures) in panel (a) are .377a (.19), -.133a (.04), -.087b (.04), and-.15b (.08) for tenure, separations, quits, and layoffs, respectively. Baseline estimates for panel(b) are -.249a (.04), -.183a (.04), and -.108° (.06) for separations, quits, and layoffs, respectively.

t-> co

© p.

go

d ©\

in ct\QO O\oq cn© o.

(jo vor vo\q tod p\

Os VOvo ooO <Nd P.

00 Ojd o

^H 00VO O

d p.

vo orH COo qd p.

o 6s

COO

d p\

(N CO ^O

p\ d p\

O O H Mi—I CO CO Mp p dpd d. d d.

.8

.2 ftfiH

£5 a «

III!

j ) O JH ^ *

S o <D •

O\I

VO

O N

rH

cf

c?

ICO

13

Is

CN3p

© P. o p.

o oTfr CO

d p.

f>. 00 r-l >OrH 00 ON »O,

© p, d p'

v o • nPP©;© p. ©

CO O NrH rHO OrH O

rH mCN COCN CNo o

00 ON

3 ©do

rH CO CO CN© O © O© P" © ©'

00 l>© CN

ooO

O P,CO rH

d p.S oo

CO

VO ON

© rH

© d

rH CO© ©© P.

I 1 "M u-> *-H oo mt~~ O N T—I i—T t CN rH rH

oo do

rH CO CO CN

© © © ©© P. © P.

© ooCO O N

CNCO i o CN

© © © ©© P. © ©,

© P.

2 87-i rH

oo

CN ©CO COCO rH

© ©

rf nO N v©CO ^

d p.

co mVO 00rH COd p.

ONCN

rH 00rH C -© rHd p.

CO O© rH© * P .

0)

^ ^ » W * * ^ ^ >

VO ^ VO 00CO O\ CO CNrH © r f ^-i

© ©, © ©, O ©,

0)

8 8

CO

co

8

8.O

coCO

CO

co

.P cra <D

<D

<U

(D

^fr|

.r'O

<L>

S1 -Ns

Table 4.4: Training Effects on TurnoverTenure-Size Interactions

SEPARATIONS QUITS LAYOFFS

(a) 1973-77 Quality of Employment Survey

In (P_Size)

In (P_Size)x Tenure

Tenure

-0.119b

(0.052)

-0.003(0.005)

-0.068c

(0.040)

-0.089c

(0.053)

-0.0003(0.005)

-0.082b

(0.042)

0.002(0.107)

-0.053c

(0.029)

0.281c

(0.147)

(b) 1982 Employment Opportunities Pilot Project (n=1,072)

In (P_Size)

In (P_Size)x Tenure

Tenure

0.379a

(0.083)

-0.6863

(0.080)

-0.339c

(0.208)

0.206a

(0.073)

-0.476a

(0.076)

-0.145(0.226)

-0.111(0.101)

-0.047(0.124)

-1.507a

(0.397)

(c) 1976-77 National Longitudinal Survey of Youth (n=1,448)

In (P_Size)

In (F_Size)

ln(FJSize)x Tenure

Tenure

-0.012(0.032)

-0.060b

(0.029)

-0.002(0.005)

-0.2283

(0.037)

0.021(0.032)

-0.017(0.030)

-0.009c

(0.005)

-0.166a

(0.037)

-0.067(0.056)

-0.086c

(0.047)

0.010(0.009)

-0.1793

(0.059)

NOTE: All regressions are estimated by a maximum likelihood probit procedure, andadditionally include the same controls as noted in previous tables. Parameter estimates arelisted with standard errors in parentheses. Superscripts a-d denote significance levels of 1%,5%, 10%, and 15%, respectively.

Table 5.1:

In (P_Size)

In (FJSize)

LNWAGE

RET

MED

Wage and Fringe Benefit EffectsCurrent Population Survey, May

0.391a

(0.094)

0.259"(0.106)

3.143"(0.341)

0.479"(0.095)

0.075(0.116)

1.619"(0.409)

1.613"(0.465)

on Tenure1979 (n=2,643)

0.377"(0.094)

0.066(0.114)

2.948"(0.343)

1.409"(0.404)

1.224"(0.461)

NOTE: All regressions are estimated by OLS, and additionally include aconstant, a quadratic in experience, education, SMSA, marital status, race,three region dummies, eleven industry dummies, and eight occupationdummies. Parameter estimates are listed with standard errors in parentheses.Superscript a denotes significance at 1%. The baseline employer size effects(based on regressions with all controls except wages and fringes, and using asample where wages and fringe benefits are not missing) are .505 (.095) and.307 (.108) for plant and firm size, respectively. Means and standarddeviations are: (WAGE 8.485 (4.52)), LNWAGE 2.014 (0.50), RET 0.823(0.38), and MED 0.893 (0.31).

In (P_Size)

LNEARN

RET

MED

In (P_Size)

LNEARN

RET

MED

Table 5.2:

0.266(0.190)

2.590s

(0.867)

-0.078c

(0.044)

-0.249(0.203)

Wage and Fringe Benefit EffectsQuality of Employment Survey

fa) Tenure

0.227(0.198)

2.1483

(0.839)

-0.217(1.000)

(c) Quits

-0.060(0.046)

-0.328c

(0.191)

-0.122(0.225)

0.174(0.197)

2.323a

(0.898)

1.861b

(0.839)

-0.703(1.009)

-0.057(0.046)

-0.161(0.211)

-0.311c

(0.192)

-0.088(0.229)

on Tenure and Turnover

-0.118a

(0.043)

-0.4883

(0.200)

-0.121d

(0.079)

-0.751b

(0.367)

(b) Separations

-0.096b

(0.045)

-0.372b

(0.182)

-0.261(0.215)

(d) Layoffs

-0.099(0.082)

-0.380(0.315)

-0.499(0.346)

-0.089b

(0.045)

-0.376c

(0.206)

-0.336c

(0.184)

-0.187(0.220)

-0.086(0.083)

-0.573d

(0.383)

-0.339(0.320)

-0.353(0.361)

NOTE: The tenure regressions are estimated by OLS, while the turnover regressions areestimated by a maximum likelihood probit procedure. All regressions additionally include aconstant, quadratics in tenure (excluding panel (a)) and experience, education, dummies forSMSA, marital and health status, race, three region dummies, nine industry dummies, and sixoccupation dummies. Parameter estimates are listed with standard errors in parentheses.Superscripts a-d denote significance levels of 1%, 5%, 10%, and 15%, respectively. Baselineemployer size effects on tenure, separations, quits, and layoffs are .377b (.19), -,133a (.04),-.087b (.04), and -.150b (.08), respectively. Means and standard deviations are: (EARN10,693.62 (6,140.32)), LNEARN 9.133 (0.55), RET 0.597 (0.49), and MED 0.757 (0.43).

In (PSize)

In (F_Size)

LNWAGE

RET

MED

In (P_Size)

In (F_Size)

LNWAGE

RET

MED

Table 5.3: Wage and Fringe Benefit Effects onNational Longitudinal Survey of Young

0.093(0.071)

0.043(0.056)

1.370"(0.242)

-0.018(0.033)

-0.039d

(0.025)

-0.298"(0.111)

(a) Tenure

0.115C

(0.071)

-0.031(0.060)

0.779"(0.216)

0.638"(0.239)

0.087(0.070)

-0.050(0.060)

1.175a

(0.245)

0.657"(0.215)

0.505b

(0.239)

(c) Quits

-0.017(0.033)

-0.017(0.027)

-0.235"(0.096)

-0.143(0.105)

-0.022(0.033)

-0.013(0.027)

-0.252b

(0.112)

-0.216b

(0.097)

-0.115(0.106)

TenureMen

and Turnover

(b) Separations

-0.008(0.032)

-0.055b

(0.025)

-0.343"(0.110)

-0.060(0.058)

-0.045(0.040)

-0.181(0.168)

-0.010(0.032)

-0.025(0.026)

-0.273"(0.093)

-0.244b

(0.104)

(d) Layoffs

-0.064(0.059)

-0.012(0.043)

-0.288c

(0.159)

-0.169(0.146)

-0.005(0.032)

-0.020(0.026)

-0.278"(0.111)

-0.252"(0.094)

-0.216b

(0.104)

-0.062(0.059)

-0.010(0.043)

-0.111(0.172)

-0.280c

(0.160)

-0.154(0.148)

NOTE: The tenure regressions are estimated by OLS, while the turnover regressions areestimated by a maximum likelihood probit procedure. All regressions additionally include thesame controls as in the NLS tables above. Parameter estimates are listed with standard errorsin parentheses. Superscripts a-d denote significance levels of 1%, 5%, 10%, and 15%,respectively. Baseline plant and firm size effects (based on regressions with all controls exceptwages and fringes, and using a sample in which wages and fringes are all not missing - samplesizes are 1,485 for the tenure regressions, and 1,371 for the turnover regressions) are: .127°(.071) and .090° (.056), -.015 (.032) and -.066" (.024), .011 (.033) and -.048b (.025), -.062 (.058)and -.051 (.040) for panels (a) - (d), respectively. Means and standard deviations are:(WAGE 5.772 (2.87)), LNWAGE 1.645 (0.47), RET 0.503 (0.50), and MED 0.764 (0.42).

Table 6.1: Effects of Prior Inter-Firm Mobility on Employer Size MatchNational Longitudinal Survey of Young Men (n=1,448)

Plant Firm

Number of EmployerChanges (NEC) x 10

Number of DifferentEmployers (NDE) x 10

-0.7293

(0.209)

-0.511b

(0.223)

-0.834a

(0.195)

-0.5643

(0.209)

NOTE: All regressions are estimated with a maximum likelihood ordered probit procedure.Each regression additionally includes a constant, a quadratic in experience, education,dummies for SMSA, marital and health status, race, three region dummies, eight industrydummies, and seven occupation dummies. Parameter estimates are listed with standarderrors in parentheses. Superscripts a and b denote significance levels of 1% and 5%,respectively. The inter-firm mobility (heterogeneity) measures are: NEC=the number oftimes the employee has changed employers, including returning to a previous employer,NDE=the same measure as NEC, but excludes returns to previous employers. Means(standard deviations) are: NEC 2.451 (1.627), NDE 3.214 (1.527).

ln(P_Size)

ln(FJSize)

NEC

NDE

ln(P_Size)

ln(F_Size)

NEC

NDE

Table 6.2: Heterogeneity EffectsNational

(a)1

0.118c

(0.064)

0.019(0.050)

-0.928a

(0.050)

(c)

0.015(0.032)

-0.044c

(0.024)

0.038d

(0.026)

I Longitudinal

Tenure

0.127b

(0.063)

0.042(0.049)

-1.0733

(0.051)

Quits

0.015(0.032)

-0.045d

(0.024)

0.043(0.028)

on Tenure and TurnoverSurvey of Young Men

(b) Separations

-0.012(0.031)

-0.0653

(0.024)

0.029(0.026)

-0.013(0.031)

-0.0653

(0.024)

0.035(0.028)

(d) Layoffs

-0.061(0.056)

-0.056d

(0.038)

-0.002(0.040)

-0.061(0.056)

-0.056d

(0.038)

0.009(0.042)

NOTE: Panel (a) is estimated by OLS, panels (b)-(d) are estimated by maximumlikelihood probits. Each regression additionally includes a constant, quadratics in tenure(excluding panel (a)) and experience, education, dummies for SMSA, marital and healthstatus, race, three region dummies, eight industry dummies, and seven occupationdummies. The inter-firm mobility (heterogeneity) measures are the same as in Table 6.1above. Parameter estimates are listed with standard errors in parentheses. Superscriptsa-d denote significance levels of 1%, 5%, 10%, and 15%, respectively. Baseline plantand firm size effects (based on regressions with all controls except the inter-firmmobility measures, and using a sample in which the inter-firm mobility measures are notmissing) are .162a (.07) and .097° (.06), -.013 (.03) and -.066a (.02), .015 (.03) and -.045b

(.02), and -.061 (.06) and -.056d (.04), for panels (a) - (d), respectively.

Table

Variables

ln(P_Size)

NFREE

NOSAY

FAST

HARD

NOTTME

EXCESS

REQOVT

REPET

NINTER

NSECUR

HOURB

DANGER

PHYS

7: Effects of Nonpecuniary Job Attributes on EmployerQuality of Employment Survey (n=358)

Variable Definitions

natural logarithm of the number of employees atthe locations where the respondent works

= 1 if there is little freedom as to how therespondent does his work

= 1 if respondent has little to say about whathappens on his job

= 1 if job requires that the respondent worksvery fast

= 1 if job requires that the respondent worksvery hard

= 1 if there is not enough time to get things doneon the job

= 1 if asked to do excessive amounts of work

= 1 if could not refuse to work overtime withoutbeing penalized in some way

= 1 if job often requires doing things that arevery repetitive

=1 if work is not interesting

= 1 if the job security is not good

= 1 if a problem with hours or days worked ismentioned (e.g. excessive hours, time slot)

= 1 if exposed to dangerous or unhealthy conditions

= 1 if physical conditions are not considered at allpleasant

Size Effects on

MeanCStd. Dev.)

4.312(2.058)

0.964(0.186)

0.633(0.483)

0.337(0.473)

0.376(0.485)

0.273(0.446)

0.102(0.303)

0.108(0.310)

0.519(0.500)

0.041(0.200)

0.472(0.500)

0.599(0.491)

0.414(0.493)

0.638(0.481)

Quits

Estimates

-0.069d

(0.047)

0.199(0.444)

-0.103(0.192)

0.131(0.181)

-0.033(0.187)

0.165(0.206)

0.101(0.277)

0.125(0.265)

-0.116(0.179)

0.043(0.410)

0.108(0.182)

0.227(0.181)

0.053(0.180)

0.090(0.183)

NOTE: Maximum likelihood probit estimates are reported with asymptotic standard errors inparentheses. The regression additionally includes a constant, ki(earnings), quadratics in tenureand experience, education, dummies for SMSA, marital and health status, race, three regiondummies, nine industry dummies, and six occupation dummies. Superscript c denotessignificance level 10%. The baseline employer size estimate (from a regression with allcontrols above except the job attribute controls, and using the same sample for which all jobattribute measures are not missing) is -.079° (.045).

log(P_Size)

log(F_Size)

Multiplant

SPLF1

SPLF2

Table 8: Plant and Firm Size Effects on Layoffs - Monitoring EffectsNational Longitudinal Survey of Young Men (n= 1,448)

-0.051(0.051)

-0.104*(0.034)

-0.061(0.056)

-0.056d

(0.038)

0.051(0.086)

-0.225b

(0.106)

0.378d

(0.245)

0.333(0.256)

0.060(0.115)

-0.203c

(0.125)

0.386(0.279)

0.240(0.318)

NOTE: The first column of regression results contains no controls except for plant size andfirm size. All other regressions are estimated by maximum likelihood probit, and include aconstant, quadratics in tenure and experience, education, dummies for SMSA, marital andhealth status, race, residence in the south, six industry dummies, and five occupation dummies.Parameter estimates are reported with standard errors in parentheses. Superscripts a, b, c, andd denote significance levels of 1%, 5%, 10%, and 15%, respectively. The listed regressors aredefined as follows: Multiplant is a dummy =1 if the respondent's firm size exceeds therespondent's plant size, SPLF1 (small plant, large firm) is a dummy =1 if the respondentworks in a plant of size 1-24 or 25-99 which is part of a firm or size 500-999 or 1,000, or aplant size of 100-249 that is part of a firm of size 1,000 or more, and SPLF2 (small plant,large firm) is a dummy= 1 if the respondent works in a plant of size 1-24 or 25-99 which is partof a firm of size 500-999 or 1,000.

Plant

Firm

1-24

8.47(8.30)[215]

9.80(9.34)[46]

Table 9.1: Union Threat EffectsTenure-Size Profiles,

25-99 100-499

(a) Managers

11.69 12.59(9.49) (9.85)[176] [139]

9.22 8.24(9.34) (7.15)[79] [126]

1979 CPS

500-999

15.21(9.53)[29]

10.33(9.52)[46]

1.000+

16.24(10.85)[92]

13.76(10.21)[354]

% Change

91.74%

40.41%

(b) Professional, Technical, and Kindred Workers

Plant 7.06 7.94 7.70 9.33 12.35 74.93%(6.22) (8.01) (8.44) (8.53) (9.73)[94] [110] [155] [84] [271]

Firm

Plant

Firm

7.69(6.15)[26]

6.84(7.21)[188]

6.49(5.87)[49]

6.41(6.71)[29]

(c) Industries

8.44(7.47)[115]

7.43(8.30)[58]

6.35(6.82)[71]

with Percentage

7.65(7.01)[89]

6.37(6.16)[65]

9.82(9.49)[50]

Unionized :

9.41(7.56)[34]

10.06(10.10)

[31]

10.29(9.20)[538]

12.30(9.22)[89]

9.22(7.98)[312]

33.81%

79.82%

42.06%

NOTE: Mean tenure levels are reported with standard deviations in parentheses and samplesizes in brackets. The % Change column calculates the percentage change in the means fromthe average of the 1-24 category to the 1,000 or more category.

log(P_Size)

log(F_Size)

%Union

log(P_Size)x %Union

log(F_Size)x %Union

0.491a

(0.093)

0.311a

(0.105)

Table 9.2: Union Threat EffectsTenure Regressions, 1979

Full Sample(n=2,722)

0.490a

(0.093)

0.299a

(0.105)

0.019c

(0.011)

0.576a

(0.135)

-0.022(0.153)

-0.086b

(0.041)

-0.005(0.004)

0.019a

(0.006)

Managers(n=651)

0.523a

(0.204)

0.488b

(0.217)

CPS

PT&K Workers(n=713)

0.588a

(0.174)

0.099(0.243)

% Union £5(n=515)

0.652a

(0.183)

0.111(0.190)

NOTE: All regressions are estimated by OLS, and include a constant, a quadratic inexperience, education, and dummies for SMSA, marital, race, region, industry and occupation.Parameter estimates are reported with standard errors in parentheses. Superscripts a, b, and cdenote significance levels of 1%, 5%, and 10% respectively. Mean (std. dev.) for %Union is22.9 (19.0).

In (P_Size)

%Union

In (P_Size)x %Union

In (P_Size)

%Union

In (P_Size)x %Union

In (P_Size)

%Union

In (P_Size)x %Union

Table 9.3: Union Threat Effects on TurnoverQuality of Employment Survey

-0.1333

(0.043)

-0.087b

(0.044)

-0.150b

(0.076)

(a) Separations

-0.1273

(0.043)

-0.015c

(0.008)

(b) Quits

-0.079c

(0.044)

-0.015c

(0.009)

(c) Layoffs

-0.150b

(0.076)

-0.006(0.015)

-0.051(0.067)

0.002(0.014)

-0.004d

(0.003)

0.019(0.069)

0.007(0.015)

-0.005c

(0.003)

-0.052(0.122)

0.015(0.024)

-0.006(0.006)

NOTE: All panels are estimated by maximum likelihood probit andadditionally include a constant, quadratics in tenure and experience,education, dummies for SMSA, marital and health status, race, region,industry, and occupation. Mean (std. dev.) for %Union is 22.3 (17.1).Parameter estimates are reported with standard errors in parentheses.Superscripts a- d denote significance levels of 1%, 5%, 10%, and 15%,respectively.

1992-93 DISCUSSION PAPER SERIESDepartment of Economics

Columbia University420 W 118th St., 1022 IAB

New York, NY 10027Librarian: Ms. Angie Ng

The following papers are published in the 1992-93 Columbia University Discussion Paper Serieswhich runs from July 1 to June 30. Individual discussion papers are available for purchase at$5.00 (U.S.) each for domestic orders and $8.00 (U.S.) for foreign orders. Subscriptions to theSeries are available at a cost of $185.00 (U.S.) per foreign subscription and $140.00 (U.S.) perdomestic subscription. To order discussion papers, please send your check or money orderpayable to Department of Economics, Columbia University to the above address. Please makesure to include the series number of the paper when you place an order.

612. Irreversible Choice of Uncertain Technologies with Network ExternalitiesJay Pil Choi

613. The Real Exchange Rate and U.S. Manufacturing Profits: A Theoretical frameworkwith Some Empirical SupportRichard H. Clarida

614. Cointegrauon. Aggregate Consumption, and the Demand for Imports: A structuralEconometric InvestigationRichard H. Clarida

615. Projecting the Number of New AIDS Cases in the U.S.David E. Bloom and Sherry Glied

616. Financial Markets for Unknown RisksGraciela Chichilnisky and Geoffrey M. Heal

617. Financial Innovation and Endogenous Uncertainty in Incomplete Asset MarketsGracieia Chichilnisky and Ho-Mou Wu

618. Arbitrage and Equilibrium in Economies with Infinitely Many Securities andCommoditiesGraciela Chichilnisky and Geoffrey M. Heal

619. Market Innovation and the Global EnvironmentGraciela Chichiinisky

620. Option and Non-Use Values of Environmental AssetsAndrea Beltratti, Graciela Chichilnisky and Geoffrey Heal

621. Competition among InstitutionsAndrew Caplin and Barry Nalebuff

622. Speculation on Primary Commodities: The Effects of Restricted EntryJohn McLaren

623. Why did Big Coffee seek regulation? A theory of dynamic monopsony pricingwithout commitmentJohn McLaren

624. Speculative Equilibria of "Managed" Primary Commodity MarketsJohn McLaren

625. Income Distribution. Political Instability, and InvestmentAlberto Alesina and Roberto Perotu

h26. The Political Economy of Growth: A Critical Survey of the Recent Literature and SomeNew ResultsAlberto Alesina and Roberto Perotn

627. The Term Structure of Forward Exchange Rates and the Forecastability of SpotExchange Rates: Correcting the ErrorsRichard H. Clarida and Mark P. Taylor

628. Why Homelessness? Some TheoryBrendan O'Flaherty

629. A Note on Heteroskedasticity IssuesPhoebus J. Dhrymes

630. Who Is Bearing the Cost of the AIDS Epidemic in Asia?David E. Bloom and Sherry Glied

631. Optimal Tariffs and the Choice of Technology: Discriminatory Tariffs vs. the "MostFavored Nation clauseJay Pil Choi

632. A Useful LemmaPhoebus Dhrymes

633 The New Homelessness in North America: Histories of Four CitiesBrendan O' Flaherty \ \

634 Bum-Outs: Fire Victims tn North Jersey, the Red Cross, and the Housing MarketBrendan O'Flaherrv

635. Labor and the Emerging World EconomyDavid E. Bloom and Adi Brender

636. Fiscal Policy, Income Distribution, and GrowthRoberto Perota

637. The Political Economy of Redistribution in a Federal SystemRoberto Perotn

638. A Note on Identification Test ProceduresPhoebus Dhrymes

639. The Optimal Income Tax ScheduleKelvin Lancaster

640 Strategies for Trade Liberalization in the Americas: A Report to ECLACGracieia Chichilnisky

641. Robustly Efficient Equilibria in Non-Convex EconomiesGracieia Chichilnisky and Geoffrey Heal

642. Financial Markets for Unknown RisksGracieia Chichilnisky and Geoffrey Heal

643. Price Uncertainty and Derivative Securities in a General Equilibrium ModelGracieia Chichilnisky, Jayasri Dutta and Geoffrey Heal

644 North-South Trade and the Dynamics of Renewable ResourcesGracieia Chichilnisky

645. Global Environmental RisksGracieia Chichilnisky and Geoffrey Heal

646. Chaotic Price Dynamics. Increasing Returns & the Phillips CurveGracieia Chichilnisky, Geoffrey Heal and Yun Lin

647. Notes on the Political Economy of NationalismRonald Findlay

648. After Maastricht: Public Investment, Economic Integration, and International CapitalMobilityRichard Clarida and Ronald Findlay

649. Markets. Arbitrage and Social ChoicesGracieia Chichilniskv

650. Limited Arbitrage is Necessary and Sufficient for the Existence of a CompetitiveEquilibriumGraciela Chichilnisky

651. Existence of a General Equilibrium with Price UncertaintyGraciela Chichilnisky

652. Existence of an Optimal Path in a Growth Model with Endogenous Technical ChangeGraciela Chichilnisky and Paul F. Gruenwald

653. Explaining Economic GrowthDavid Canning

654 The Effects of Sectoral Decline on the Employment RelationshipTodd L. Idson and Robert G. Valletta

655 Unemployment and the Economic of Gradualist Policy ReformMichael Gavin

656 Commodity-Price-Destabilizing: Commodity Price StabilizationJohn Mclaren

657. Executive Compensation and Agency EffectsTodd. L. Idson and Lawerence G. Goldberg

658. Will Free Trade With Political Science Put Normative EconomistsOut of Work?Brendan O'Flaherty and Jagdish Bhagwati

659 A characterization ot CointegrationPhoebus J. Dhrymes

660. The Production of Human Capital and the Lifecycles of EarningsJacob Mincer

661. Price Continuity Rules and Insider TradingPrajit K. Dutta

662. On Specifying the Parameters of a Development PlanPrajit K. Dutta

663. Bankruptcy and Expected Utility MaximizationPrajit K. Dutta

664 Moral HazardPrajit K. Dutta

665. Information Aggregation and Strategic Trading in SpeculationPrajit K. Dutta

666. Optimal Management of an R&D BudgetPrajit K. Dutta

667. Identification and Kullback Information in the GLSEMPhoebus J. Dhrymes

668. The Influence of Nonmarital Childbearing on the Formation ofFirst MarriagesNeil G. Bennett, David Bloom and Cynthia K. Miller

669 A Revealed Preference Approach For Ranking City Quality of LifeMatthew Kahn

Free Trade: Old and New ChallengesThe 1993 Harry Johnson LectureJagdish Bhagwati

1993-94 Discussion Paper SeriesDepartment of Economics

Columbia University420 W. 118 St., Room 1022

New York, N.Y., 10027Librarian: Angie Ng

The following papers are published in the 1993-94 Columbia University Discussion Paperseries which runs from November 1 to October 31. Domestic orders for discussion papers areavailable for purchase at $5.00 (U.S.) each and $140.00 (U.S.) for the series. Foreign orderscost $8.00 (U.S.) for individual paper and $185.00 for the series. To order discussion papers,please send your check or money order payable to Department of Economics, ColumbiaUniversity to the above address. Please be sure to include the series number for the paper whenyou place an order.

671. Investment in U.S. Education and TrainingJacob Mincer ( Nov. 1993)

672. Freer Trade and the Wages of the Unskilled: Is Marx StrikingAgain?Jagdish Bhagwati and Vivek Dehejia

673. Employer Size and Labor TurnoverTodd Idson

674. Less Crime May Be WorseBrendan O' Flaherty

675. Team Production Effects on EarningsTodd Idson

676. Language, Employment, and Earnings in the United States:Spanish-English Differentials from 1970 to 1990David Bloom and Gilles Grenier

677. The Impact of Performance Incentives on Providing Job Trainingto the Poor: The Job Training to the Poor: The Job Training PartnershipAct (JTPA)Michael Cragg

678. The Demands to Reduce Domestic Diversity among Trading NationsJagdish Bhagwati ^

679. Mass Layoffs and UnemploymentAndrew Caplin and John Leahy

680. The Economics of AdjustmentAndrew Caplin and John Leahy

681. Miracle on Sixth Avenue: Information Externalities and SearchAndrew Caplin and John Leahy

682. Arbitrage, Gains from Trade and Scoial Diversity: A Unified Perspective onResource AllocationGraciela Chichilnisky

683. Who should abate carbon emissions?Gracieia Chichilnisky, Geoffrey Heal

684. Believing in Multiple EquilibriaGraciela Chichilnisky

685. Limned Arbitrage, Gains from Trade and Arrow's TheoremGracieia Chichilnisky

686. International Emission Permits: Equity and EfficiencyGraciela Chichilniskv. Geoffrev Heal and David Starrett