Trade Unions and Quits: Australian Evidence*

MARK WOODEN MEREDITH BAKER

National Institute of Labour Studies,

Flinders University of South Australia, Adelaide, Australia

* This study uses data from the recent Australian Workplace Workplace Industrial Relations

Survey to test the union voice hypothesis that unions reduce quits. Unlike the U.S., however,

it is argued that union voice may not be directly correlated with union membership as a result

of the protections afforded trade unions by the unique Australian industrial relations system.

It is found that, while unions are inversely associated with quit rates, this effect is strongest

where union membership is supplemented with a more direct indicator of what unions

actually do in the workplace.

1

Trade Unions and Quits in Australia

I. Introduction

A key prediction of the union voice hypothesis proposed in the seminal work of Freeman and

Medoff (1979, 1984) is that union members will be less likely to voluntarily separate from

their employer (i.e., quit) than their nonunionized counterparts. Drawing on the work of

Hirschman (1970), who distinguished between the use of “exit” and “voice” to express

dissatisfaction in relationships, Freeman and Medoff argued that nonunion workers will rely

heavily on the exit mechanism (quitting) to obtain desired job conditions because of the fear

of victimization should they attempt to “voice” their concerns. In contrast, unions provide

workers with a mechanism for relaying their preferences to management without fear of

retaliation so employees are less likely to have to quit in order to secure better working

conditions. Moreover, given the public goods nature of many working conditions, the

incentive for an individual employee to attempt to secure changes to plant-wide conditions

and policies is very small. Collective action, via unions, therefore, offers a more efficient

mechanism for the expression of worker preferences, enabling the firm to choose a more

optimal remuneration and human resource package and thereby discourage quits.

Empirically, this hypothesis receives strong support from U.S. studies (e.g., Brown and

Medoff, 1978; Leigh, 1979; Freeman, 1980a, 1980b; Blau and Kahn, 1981, Long and Link,

1983). In a recent paper, however, Drago and Wooden (1991) questioned whether such

findings can be generalized to other industrial relations settings. In particular, they point to

the Australian case, arguing that the nature of the highly centralized industrial relations

systems that predominate there, in tandem with a system of unionism which is still largely

structured along occupational lines, may attenuate the effective provision of voice by unions

at some Australian workplaces. Indeed, in subsequent work they argue that the provision of

voice will depend on the idiosyncratic features of workplaces rather than on the mere

2

presence of union representation (Drago et al., 1992, p. 48). As a test of this hypothesis they

suggest replacing union membership with more direct indicators of union activity within the

workplace. This paper provides such a test using data from the recent Australian Workplace

Industrial Relations Survey.

We first summarize previous Australian research on the relationship between unions

and quits and, drawing heavily from Drago and Wooden (1991), explain why union voice

should be explicitly measured in the Australian context. In Section III the conceptual

framework, data and empirical model are presented and discussed. Sections IV and V contain

the results and conclusions.

II. Union Voice within the Australian Setting

To date, two published studies have examined the impact of unions on quit rates using

Australian data. The first, by Miller and Mulvey (1991), adopts an approach which closely

mimics most of the U.S. studies. Using individual-level data for a sample of Australian male

youths, they include union membership along with a number of other variables designed to

control for characteristics of both individuals and jobs which might affect worker mobility in

a probit model to explain the incidence of quits over a one-year period. They find that union

membership is associated negatively with quit behavior, but the relationship is relatively

weak (significant at only the ten percent level).1 Further, Miller and Mulvey report that

unions have a much stronger restraining effect on layoffs, indicative of the monopoly face of

trade unionism in Australia.

The second study, by Drago and Wooden (1991), is relatively unique in its emphasis on

the need to augment measures of union membership with more direct indicators of union

voice.2 This need stems from their claim that mere union presence does not necessarily

guarantee effective voice provision. While union membership may be an adequate proxy for

voice in the U.S. context where unions are dependent on the level of service provided to

3

members in order to survive, Drago and Wooden argue that in Australia the protections

afforded trade unions undermine the incentive to provide voice (p. 237).

Drago and Wooden point to five factors which lead them to expect a relatively weak

relationship between unionization levels and voice provision.

i) Trade union organization in Australia continues to reflect occupational, rather than

industry or enterprise, distinctions. Consequently, the membership of the average union

is spread over a large number of workplaces, reducing opportunities for contact

between employees and union officials and increasing the likelihood of agency

problems. Further, occupational unionism gives rise to multiple unions in the

workplace which tends to fragment union voice.

ii) Australian industrial relations has been characterized by a reliance on highly

centralized bargaining processes (in part, a result of the predominance of

occupationally-based union structures) which, by mitigating high levels of union

activity at the workplace, have further exacerbated agency problems.

iii) Closed shop agreements, though on the wane, remain widespread in Australia, and as

Drago et al. (1992, p. 47) observe, such agreements significantly reduce the incentive

for unions to respond to the wishes of their members (retention of members is not

contingent on the quality and level of union services provided).3

iv) Because of the “conveniently belong to” provision of the Industrial Relations Act –

which states that the registration of a new union must be refused if another union, to

which employees might more conveniently belong, is already registered – the freedom

of employees to express their dissatisfaction with their current union by joining an

alternative union is curtailed. Furthermore, adherence to rigid union demarcation (or

jurisdictional) lines has typically precluded existing unions from competing for

4

members. The result is again weakened incentives for unions to be responsive to the

interests of their membership.

v) Whereas in the U.S. it is the existence of formal and effective grievance procedures in

the union sector that distinguishes it most from the nonunion sector (Freeman, 1980a),

such procedures have historically been absent from the Australian industrial relations

setting (Dabscheck and Niland, 1981, pp. 71-72).

The Drago and Wooden study can be also be differentiated from previous research in

its use of workplace-level data. Previous studies of quit behavior have typically utilized

either individual- or industry-level data, with only one other study (Wilson et al., 1990)

relying on information collected from workplaces. While individual data would seem most

appropriate given that the decision to quit is made by individual workers, there are many

potential influences on quits which may not be easily observed with data collected from

individual workers. Further, the workplace is the obvious unit from which to assess union

activity and voice, especially given the practice of “blanket” award coverage, whereby the

conditions set down in union negotiated contracts (or “awards”) are automatically extended

to cover nonunion members at most workplaces where union members are present.4

Using data collected from over 300 Australian workplace managers, Drago and

Wooden find labor turnover rates unrelated to workplace unionization rates. An inverse

relationship with the number of workplace union delegates per union member, however, was

discerned, providing support for their hypothesis that union voice effects are present in

Australian workplaces but are not well captured by differences in unionization rates. But

Drago and Wooden’s analysis is deficient for at least two reasons. First, the data that underlie

their analysis were not collected from a representative sample of Australian workplaces.

Instead, the sample was constrained to workplaces operated by member companies of the

Business Council of Australia, which tend to be relatively large and concentrated in the

manufacturing sector. Second, the dependent variable does not explicitly distinguish between

5

voluntary and involuntary separations, and, as the work of Miller and Mulvey (1991)

demonstrates, unions are likely to have opposite effects on these two types of turnover.

III. Model Specification and Data

Conceptual Framework. While a large number of different theoretical models of labor

turnover and quit behavior exist within the economics literature, the most common

approaches rest on the related theories of job search and human capital. According to this

framework, individuals are assumed to maximize their expected discounted lifetime utility

from employment, net of search costs and other costs of quitting. In this context, quit

behavior is a function of the expected discounted utility in the current job, expected

discounted utility in the best alternative employment or activity, and a range of factors

hypothesized to be related to the cost of mobility (e.g., Freeman, 1980a; Borjas, 1984).

Union voice can therefore be conceived as a means by which the match between workers and

jobs is enhanced, increasing average worker productivity and thereby increasing expected

utility in the current job relative to other now less productive alternatives.

Operationalizing such models empirically is invariably difficult. Most empirical

analyses of quit behavior, for example, include measures of wage income. The attractiveness

of the current job, however, depends not only on wages, but also on other nonwage benefits

and nonpecuniary job characteristics, factors which have generally received much more

attention from industrial psychologists than from labor economists. Even more difficult is

locating measures to control for expected changes in future income. As noted above,

theoretical considerations suggest that it is expected lifetime utility which is important, and

this will be influenced by such factors as promotion opportunities and the likelihood of

layoff. Finally, and unlike current wages, wages in alternative employment are difficult to

observe and hence are typically proxied by personal and job characteristics (e.g., Freeman,

1980a; Miller and Mulvey, 1991). Such approaches, however, give rise to a number of

problems, not the least of which is that the proxies included are unlikely to completely

control for alternative wages.

6

Most empirical studies, therefore, utilize relatively simple specifications which

typically consist of current wages, job characteristics (such as occupation and industry),

personal characteristics (such as education, age, experience, and gender), a labor market

variable which is expected to be related to the availability of alternative opportunities, and a

union membership variable. In algebraic terms, such models can be expressed as follows:

Qi = α0 + α1Wi + α2Ui + α3Zi, (1)

where, for observation i, Qi denotes the quit rate; Wi is the wage in the current job; Ui is an

indicator of union presence; and Zi is a vector of other variables that may affect turnover.

This type of specification is also at the heart of this study, though we include a more

extensive range of controls than most previous studies.

The AWIRS Data. The data for this analysis were collected from Australian workplaces

during 1989-1990 as part of the Australian Workplace Industrial Relations Survey (AWIRS).

Described in more detail in Callus et al. (1991), the principal sample covers workplaces with

at least 20 employees and was randomly selected from the register of establishments

maintained by the Australian Bureau of Statistics, after stratification by location, size, and

industry.5 In total, the sample numbered 2004 observations, though information on quits was

only received from 1,747 of these.6 Furthermore, following the convention with large sample

cross-sectional data, wherever non-response appeared on a variable of interest, that

observation was purged from the data set, leaving a total of 1,301 usable observations.

Finally, the data are weighted using factors provided with the data set to correct for biases

arising out of the stratified nature of the sample.7

As already noted, while the workplace may not be the obvious unit of observation

given that the decision to quit is made by individuals, workplace data are ideal for testing the

union voice hypothesis within the Australian industrial relations context, since differences in

union voice are only likely to be apparent across workplaces and firms rather than across

7

individuals. Similarly, given that many aspects of the employment relationship are to a large

extent a reflection of management policies, workplace data offers the opportunity to better

test the influence of these factors. Nevertheless, it has to be admitted that workplace data, by

forcing aggregation across individuals, are less than ideal when it comes to the measurement

of variables like wages, tenure, and other personal characteristics.

Dependent Variable and Estimation Method. The dependent variable is proxied by

QUIT, the ratio of the number of permanent employees who voluntarily resigned from the

workplace during the year prior to survey administration to the number of permanent

employees at the time of the survey. Since this variable is truncated at zero, the Tobit model

might be considered the most appropriate estimation method. However, as Maddala (1988, p.

286) points out, Tobits should only be used where the dependent variable can, in principle,

be negative, which is not so in the case of quits. Instead, Maddala recommends the use of

sample selection models, selecting on non-zero quit rates. As it turns out, however, there are

relatively few zero cases, suggesting that these might be discarded and ordinary least squares

estimation employed.8

Note that while the dependent variable does not have an obvious upper limit (as a result

of the presence of high turnover jobs), an inspection of the data indicates that the variable

distribution is highly skewed. We, therefore, opt for a log-linear specification of quit rates.

Finally, it could be argued that two-stage least squares would be a more appropriate

estimation method given the possibility of simultaneity between both wages and quits and

unions and quits. However, estimation of a system of equation wherein quits, wages and the

union variables are treated as endogenous had little affect on the results implying, perhaps

surprisingly, that wages and the union-voice variables can safely be treated as exogenous.

Explanatory Variables. Table 1 lists the main variables considered along with their expected

influence on quit rates. As noted earlier, most research has used indicators of union

membership to capture the effects of union voice. At the workplace level this effect is best

8

represented by the proportion of employees who are members of a union (DENSITY). We

also include a dummy variable for nonunion workplaces (NUWP) to control for the

possibility that any relationship between union membership and quits is nonlinear. To test the

argument that union voice effects are better represented with more direct measures of union

activity within the workplace, we follow Drago and Wooden by including the number of

union delegates (shop stewards) per employee (UDELEG).9

Current wages are proxied with the group equivalent for the worker’s current wage –

average workplace weekly earnings (WAGES). The attractiveness of the current job,

however, depends on not just current wages, but on other nonwage benefits including

nonpecuniary work characteristics. Moreover, future expected earnings and job

characteristics are also relevant. A number of additional variables are therefore suggested.

Most straightforwardly, we include the proportion of the work force covered by an employer-

sponsored pension (PENSION). This variable should have a negative sign, especially given

that under many pension schemes the value of the benefit is reduced when an employee

changes jobs.

An important variable in the work of Long and Link (1983) is market structure. They

argue that monopoly power in product markets generates economic rents which will mean

higher wage and nonwage returns to employment. Given the effects of market structure on

wages are controlled for with the inclusion of the wages variable, inclusion of a market

structure variable can only be justified on the grounds that the measures of nonwage returns

are imperfect. The latter assumption seems justified. In addition, as Parsons (1972) argues, in

more highly concentrated industries, fewer firms bid for that component of a worker’s skills

which are industry specific, and hence attractive alternative employment prospects are

reduced. This will be reinforced by collusive non-poaching agreements which are more

likely in highly concentrated industries. We, therefore, include a variable, COMPETE, which

is constructed from responses to a question asking the General Manager to describe the

intensity of competition for the main product or service produced at the workplace.10

9

Expectations of future earnings, which depend on expectations about future

employment prospects, should also be relevant to the quit decision. In turn, these

expectations will hinge on perceived promotion prospects and job security. We argue that

more favorable perceptions of promotion prospects are likely in firms with well-developed

internal labor markets. Following Drago (1992), this phenomenon is represented by a

variable measuring the proportion of managers promoted to their current position from

within the workplace (PROMOTE). With respect to job security, two proxies are suggested,

the first of which is the proportion of workers employed on a contract basis (CONTRACT).

We argue that workplaces which rely heavily on subcontracting will provide relatively more

secure employment for the remaining “permanent” staff (the brunt of labor adjustment will

be borne by the contract work force and our dependent variable only concerns quits amongst

the “permanent” workforce). Second, we include a dummy variable signalling whether

product demand had been rising or not in the year prior to the survey (DEMAND). Rising

demand will presumably be associated with increased workplace activity which in turn will

enhance job security for the incumbent employees. Additionally, we also include a dummy

variable to distinguish private sector employers from public sector employers (PRIVATE)

given that the latter typically provide much greater security of employment.

The attractiveness of the current job will also depend on working conditions and job

characteristics. A measure of shift work (SHIFT) is therefore included because shift work is

often perceived as an adverse working condition, especially once wages are taken into

account. Work conditions are also likely to vary with workplace age, with pleasant working

environments likely to be more characteristic of relatively new workplaces. The influence of

workplace age is controlled for with the variable NEW, a dummy variable indicating if the

workplace is less than five years old.

Potentially important influences on working conditions are workplace and firm size.

Work environments within large organizations are likely to be more alienating due, for

example, to a greater degree of bureaucracy, more extensive division of labor, and the

remoteness of the worker from decision making (Shorey, 1980). Relatedly, large

10

organizations may be associated with poor communication networks. All of these factors are

likely to promote low levels of job satisfaction which may be reflected in high levels of quits.

On the other hand, large firms are more likely to to be able to devote substantial resources to

human resource management policies and practices designed to alleviate the turnover

problem. Furthermore, large firms have more well-developed internal labor markets which

will create incentives for both employers and employees to pursue long-term relationships.

To control for these effects, measures of workplace and firm size are included. Workplace

size is measured with the continuous variable, WPSIZE, while firm size is represented with a

series of dummy variables (FIRMSIZE2 to FIRMSIZE7).

Management policies and practices can influence quit behavior. Wilson et al. (1990),

for instance, find evidence of inverse relationships between quit rates and the presence of

both employee participation mechanisms and profit sharing arrangements, which they argue

results because of the greater degree of worker identification with the firm that such schemes

induce. Profit sharing can be simply proxied by the proportion of nonmanagerial employees

who received any pay from a profit-sharing scheme in the previous year (PRSHARE).

Measurement of employee participation, on the other hand, is rather more difficult and it

proved impossible to derive an adequate measure from these data.11 We do, however,

include the variable PBR, which measures the proportion of non-managerial employees who

receive pay which is contingent on their performance. Given that performance-related pay

increases both uncertainty about, and the variability of, weekly income, and that the average

worker is risk averse, we expect PBR to be positively associated with quits.

An important influence on both alternative employment opportunities and the transactions

costs of mobility is the state of the labor market. Indeed, time-series studies invariably find

that quit rates are pro-cyclical, rising when labor markets strengthen and declining when they

weaken (e.g., Parsons, 1973; McCormick, 1988). When labor markets are weak, alternative

employment opportunities are fewer and the cost of quitting in terms of the duration of

unemployment while searching for a new job will be higher. We proxy the state of the labor

market with the regional unemployment rate (UR).

11

Human capital theory suggests a lower incidence of quits where firms and employees

invest heavily in firm-specific human capital. A worker who quits foregoes expected future

benefits accruing to his or her investment. Similarly, the firm fails to reap the full

productivity benefits from its investment in the employee’s training. Investment in firm-

specific human capital is traditionally proxied by measures of employee tenure (e.g., Blau

and Kahn, 1981). In these data we include four dummy variables representing the proportion

of the work force who have been employed at the workplace for more than ten years

(TENURE2 to TENURE5).12 Somewhat differently, we also include a dummy variable

which equals one if it takes at least a year before a new recruit (working in the numerically

most dominant job classification in the workplace) can expect to achieve the standard

expected of other employees in that same job classification. This variable should, therefore,

be closely correlated with the level of on-the-job training and hence its label, OJT.

Finally, as is common in other studies of quits, we include a number of variables

representing various worker characteristics often thought to be associated with job mobility

(gender, age, part-time employment, ethnicity, and occupation). We also include a series of

industry dummies to help control for any other unmeasured inter-firm differences associated

with quit rates.

IV. Results

The empirical results from the OLS estimation of our model of quit rates are provided in

Table 2. In all specifications there is evidence of heteroscedasticity, as evidenced by the

large values on the Breusch-Pagan statistics. This is not unexpected and has been found to be

problematic in other studies utilising the AWIRS data (e.g., Balchin and Wooden, 1992;

Crockett et al., 1992). To correct such problems it is usual to employ weighted least squares,

particularly where the data are collected at the workplace-level. Dickens (1990), however,

demonstrates that the weighting procedure will be unsuccessful in eliminating

heteroscedasticity from grouped data where individuals within the group (that is, the

12

workplace) share common unobserved determinants, which is likely the case here. Thus we

accept that the estimators may not be the most efficient, and use White’s (1980) method to

obtain a heteroscedastic consistent estimator for the covariance matrix.

In other respects the models perform quite well. The overall explanatory power is

good, with the adjusted R-squared values approaching 0.4, and there is no evidence of

misspecification, as evidenced by the low values on the RESET test. More importantly, the

signs on most of the control variables are in accord with expectations. Of greatest

significance, the results suggest that the most critical factors in explaining workplace

differences in quit rates are labor market conditions, promotion prospects, job security (as

proxied by both CONTRACT and DEMAND), workplace age, firm size, investment in firm-

specific human capital, the presence of incentive pay, employer-sponsored pension schemes,

and various worker characteristics (especially age, hours of work, and ethnicity).

Most of these results are relatively straightforward and require little further

explanation. Moreover, as a result of the log-linear specification used, size effects are easily

gauged. Thus, in the case of specification (1), the coefficient of -7.395 on UR implies that a

10 percent increase in the unemployment rate will be associated with a 4.7 percent reduction

in quit rates (the coefficient multiplied by the mean of UR multiplied by 10). Requiring

somewhat more explanation, however, are the firm size results. The coefficients on the

various dummies indicate that the relationship is not linear. Instead, quit rates are largest for

medium-sized firms (500 to 10,000 employees) and lowest for both very small and very large

firms. This can be explained as the outcome of different and opposing forces operating on

quits. Small organizations will be conducive to low rates of quitting as a result of closer

communication between workers and management. However, in very large firms, these scale

diseconomies are outweighed by the presence of internal labor markets and the larger

resources that such organizations can bring to bear in helping alleviate and prevent turnover

problems.

13

Perhaps somewhat disquieting is the generally insignificant influence that wages and

pension schemes are found to exert on quit rates. These results fly in the face of standard

economic theories and is a point to which we shall return below.13

Turning our attention to the union variables, a significant difference in quit rates

between union and nonunion workplaces is found, with the coefficient on NUWP reported in

column (1) implying a quit rate which is almost 25 percent higher in nonunion workplaces,

and hence provides some evidence to support the union voice hypothesis. This result,

however, is not all that important given that less than 20 percent of workplaces in the sample

are nonunion, the majority of which are very small. Of greater interest is whether quit rates

vary with the level of union membership within the workplace and as expected, the relevant

measure, DENSITY, does attract a negative sign. As in Miller and Mulvey (1991), however,

the significance of this variable is relatively weak. Specification (2), therefore, augments the

model with UDELEG, the ratio of union delegates to employees. This too attracts the

expected inverse sign, but in contrast to DENSITY achieves statistical significance at the

conventional five percent level.14 These results, therefore, suggest that, as in the work of

Drago and Wooden (1991), the only unions in Australia which provide significant voice

benefits are those which are active in the workplace; mere numeric representation is not

sufficient. The importance of this result, however, should not be overstated. The estimated

elasticity (at the mean) is quite small (0.043). For example, in a workplace of average size an

additional union delegate is estimated to be associated with a 2.1 percent reduction in quit

rates.

Theoretical considerations as well as some empirical evidence suggests that it may not

be appropriate to treat the union variables as exogenous. Unions might, for example,

influence promotion prospects, the level of fringe benefits, the ability of firms to subcontract,

and the presence of pay incentives. Omission of the variables representing these influences

(PROMOTE, PENSION, CONTRACT, PBR, and PRSHARE) had little impact on the

coefficient attached to UDELEG but the size of the coefficient on DENSITY does increase in

absolute size. In particular, after some experimentation, we found the value of the coefficient

14

on DENSITY was maximized when PROMOTE, PENSION, and PRSHARE are excluded.

The results of this specification are presented in column (3). Thus part of the effects of union

density on quits may be operating through promotion prospects and the presence of

employer-sponsored pension schemes and profit sharing arrangements, which in turn imply

that in workplaces with higher levels of union density, promotions from within, employer

contributions to pension schemes, and profit sharing are all much more common. While the

first two relationships are not surprising, the positive relationship between profit sharing and

union density is, especially given previous Australian evidence indicating that unions

militate against such schemes (see Drago et al., 1992, p. 216).

Finally, we also tested the impact of omitting the occupation and industry controls.

These results are presented in column (4) and reveal little change to our central finding that

union voice effects may be better captured with measures of union activity on the shopfloor

than union membership. However, specification (4) results suggest a major difference in how

the impact of the wages variable should be interpreted, for it is now highly significant and

attracts the expected inverse sign. This finding together with our earlier results suggests that

the effects of wages are completely captured by inter-industry and -occupation differences

and are suggestive of the effects of Australia’s award structure in facilitating a high degree of

uniformity in wage structures within occupations and industries. The effects of market

structure (as proxied by COMPETE) and shift work would also appear to be partly captured

by the occupation and industry dummies, with the coefficients on both COMPETE and

SHIFT increasing substantially in size once these controls are omitted.

V. Conclusions

This paper set out to test the union voice hypothesis for Australia using data from the

AWIRS and found clear evidence to support the union voice hypothesis. Most obviously,

nonunion workplaces had much higher quit rates than unionized ones. Complete absence of a

union presence, however, is not usual at the average Australian workplace. Indeed, nonunion

workplaces are atypical and tend to be small and concentrated in the service sector,

15

especially wholesale and retail trade. More interesting, therefore, is how quit rates vary with

differences in the extent of union presence, and, as in the earlier work of Drago and Wooden

(1991), the relationship between average workplace unionization rates and quit rates was

found to be quite weak. This does not mean that union voice effects are absent, but rather

that better measures of union presence are needed. Drago and Wooden suggest that one such

measure, though still highly imperfect, is the ratio of the number of union delegates to

employees, and this variable is indeed found to be inversely and significantly associated with

quit rates.

We, however, are wary of concluding that this necessarily implies that such measures

are preferred to union density measures. Indeed, there is evidence to indicate that part of the

effect of variations in union density is being exerted via other variables; namely, promotion

prospects, employer-sponsored pension schemes and profit sharing. Furthermore, the

magnitude of the coefficients suggest that a 50 percent increase in union density will be

associated with a greater reduction in quit rates than a 50 percent increase in the delegate-

employee ratio. This, of course, is not a valid comparison since unlike union density, the

average delegate-employee ratio observed in Australian workplaces is quite small (less than

0.02 in this sample). At greater levels of union activity the effect on quit rates may be much

greater. Moreover, the presence of delegates also does not imply active union representation.

Future research, therefore, might consider the possibility of constructing variables based

more closely on what unions actually do.

Overall, however, it is difficult to escape the conclusion that irrespective of what

measure is used, the effects of unions on quit rates at the average Australian workplace are

probably quite small. Callus et al. (1991), for example, argue that the evidence from the

AWIRS suggests that workplace union activity is “quite patchy” and in many workplaces

“the union plays essentially a monitoring role, reacting to change rather than initiating it” (p.

210). As Drago and Wooden conclude, if unions are to have a significant impact on quits,

they must provide meaningful voice and this will require expenditure of resources and

energies within the workplace in greater amounts than has been characteristic of the past.

16

REFERENCES

Balchin, Jeffrey and Mark Wooden. “Absence Penalties and the Work Attendance Decision.”

Paper presented to the 1992 Australian Conference of Economists, University of

Melbourne, July, 1992.

Blau, Francine D. and Lawrence M. Kahn, L.M. “Race and Sex Differences in Quits by

Young Workers.” Industrial and Labor Relations Review 34 (July, 1981): 563-77.

Borjas, George J. “Race, Turnover, and Male Earnings.” Industrial Relations 23 (Winter,

1984): 73-89.

Brown, Charles and James L. Medoff. “Trade Unions in the Production Process.” Journal of

Political Economy 86 (June, 1978): 355-78.

Callus, Ron, Agnes Morehead, Mark Cully, and John Buchanan. Industrial Relations at

Work: The Australian Workplace Industrial Relations Survey. Canberra: Australian

Government Publishing Service, 1991.

Chapman, Bruce J. “An Economic Analysis of Quit Behavior: A Case Study of Young U.S.

Males.” Unpublished PhD dissertation, Yale University, 1981.

Crockett, Geoffrey, Peter Dawkins, Paul Miller, and Charles Mulvey. “The Impact of Unions

on Workplace Productivity in Australia.” Australian Bulletin of Labour 18 (June, 1992):

119-41.

Dabscheck, Braham and John Niland. Industrial Relations in Australia. Sydney, George

Allen and Unwin, 1981.

17

Dickens, William T. “Error Components in Grouped Data: Is it Ever Worth Weighting?”

Review of Economics and Statistics 72 (May, 1990): 328-33.

Drago, Robert. “Do Efficiency Wages Explain Dismissals?” National Institute of Labour

Studies, Working Paper Series No. 118, National Institute of Labour Studies, Flinders

University, Adelaide, 1992.

Drago, Robert and Mark Wooden. “Turnover Down Under: Trade Unions and Exit

Behaviour in Australia.” Journal of Industrial Relations 33 (June, 1991): 234-48.

Drago, Robert, Mark Wooden, and Judith Sloan. Productive Relations? Australian Industrial

Relations and Workplace Performance. Sydney: Allen and Unwin, 1992.

Freeman, Richard B. “The Exit-Voice Tradeoff in the Labor Market: Unionism, Job Tenure,

Quits, and Separations.” Quarterly Journal of Economics 94 (June, 1980a): 643-73.

Freeman, Richard B. “The Effect of Unionism on Worker Attachment to Firms.” Journal of

Labor Research 1 (Spring, 1980b): 29-61.

Freeman, Richard B and James L. Medoff. “The Two Faces of Unionism.” Public Interest

No. 57 (Fall, 1979): 69-93.

Freeman, Richard B and James L. Medoff. What Do Unions Do? New York: Basic Books,

1984.

Hirschman, Alfred O. Exit, Voice and Loyalty. Cambridge, Harvard University Press, 1970.

Leigh, Duane E. “Unions and Nonwage Racial Discrimination.” Industrial and Labor

Relations Review 32 (July, 1979): 439-50.

18

Long, James E. and Albert N. Link. “The Impact of Market Structure on Wages, Fringe

Benefits, and Turnover.” Industrial and Labor Relations Review 36 (January, 1983): 239-

50.

McCormick, Barry. “Quit Rates Over Time in a Job-rationed Labour Market: The British

Manufacturing Sector, 1971-1983.” Economica 55 (February, 1988): 81-94.

Maddala, G.S. Introduction to Econometrics. New York: Macmillan, 1988.

Miller, Paul W. and Charles Mulvey. “Australian Evidence on the the Exit/Voice Model of

the Labor Market.” Industrial and Labor Relations Review 45, (October, 1991): 44-57.

Parsons, Donald O. “Specific Human Capital: An Application to Quit Rates and Layoff

Rates.”, Journal of Political Economy 80, (November, 1972): 1120-43.

Parsons, Donald O. “Quit Rates Over Time: A Search and Information Approach”, American

Economic Review 63 (June, 1973): 390-401.

Shorey, John. “An Analysis of Quits Using Industry Turnover Data.” Economic Journal 90,

(December, 1980): 821-37.

White, Halbert. “A Heteroskedasticity Consistent Covariance Matrix Estimator and a Direct

Test for Heteroskedasticity”, Econometrica 48 (May, 1980): 817-38.

Wilson, Nicholas, John R. Cable, and Michael J. Peel. “Quit Rates and the Impact of

Participation, Profit-Sharing and Unionization: Empirical Evidence from UK Engineering

Firms.” British Journal of Industrial Relations 28 (July, 1990): 197-213.

19

Wooden, Mark. “Compulsory Unionism and the AWIRS: Redrawing the Map.” Economic

and Labour Relations Review 3 (December, 1992): 180-98.

20

Table 1 Variables: Summary Statistics (n=1,269)

Variable

name Description Expected

sign Mean Std. dev.

QUIT Workplace quit rate 0.275 0.410 DENSITY Workplace unionization rate - 0.554 0.345 NUWP Nonunion workplace + 0.182 0.386 UDELEG Number of union delegates per employee - 0.018 0.023 ln WAGES Log of average workplace weekly wage - 6.129 0.273 PENSION % of work force with entitlements under an

employer sponsored pension scheme -

0.720

0.352

COMPETE Index of intensity of competition for main product or service

+

2.866

2.038

PROMOTE % of managers promoted from within - 0.534 0.406 CONTRACT % of work force employed on contracts - 0.048 0.146 DEMAND Product demand rising - 0.545 0.498 PRIVATE Private sector firm + 0.664 0.472 SHIFT Shift work + 0.115 0.319 NEW New workplace - 0.080 0.271 WPSIZE Workplace size ? 115.5 241.7 FSIZE Firm size ? a a PRSHARE % of employees in profit sharing scheme - 0.032 0.158 PBR % of employees receiving performance-

related pay

+

0.133

0.272 UR Regional unemployment rate - 0.064 0.015 TENURE % of work force with more than 10 years tenure - a a OJT On-the-job training requirement of average

job - 0.086 0.280

FEMALE % of females in the work force + 0.402 0.276 YOUTH More than 25% of work force aged under 20

years

+

0.010

0.300 PART % of workforce employed part-time + 0.203 0.263 NESB More than 10% of work force born overseas

in a non-English-speaking country

+

0.235

0.424

a. Series of dummy variables used to proxy effect of this influence.

21

Table 2

OLS Estimates of the Determinants of Quit Rates (Dependent variable = ln QUIT; Heteroscedastic consistent t-ratios in parentheses)

Variable (1) (2) (3) (4)

Intercept 0.163 (0.20)

0.131 (0.16)

0.045 (0.05)

1.545 (2.05)**

DENSITY -0.223 (1.84)*

-0.197 (1.63)

-0.300 (2.45)**

-0.067 (0.57)

NUWP 0.239 (2.54)**

0.221 (2.34)**

0.191 (1.99)**

0.228 (2.43)**

UDELEG -2.365 (2.16)**

-2.511 (2.25)**

-2.641 (2.29)**

ln WAGES -0.145 (1.14)

-0.133 (1.06)

-0.147 (1.16)

-0.465 (4.15)***

PENSION -0.171 (2.11)**

-0.169 (2.09)**

-0.183 (2.30)**

COMPETE 0.026 (1.28)

0.027 (1.34)

0.025 (1.24)

0.055 (3.05)***

PROMOTE -0.269 (4.06)***

-0.268 (4.03)***

-0.283 (4.22)***

CONTRACT -0.421 (2.99)***

-0.420 (2.95)***

-0.350 (2.46)**

-0.482 (3.37)***

DEMAND -0.110 (2.18)**

-0.110 (2.18)**

-0.092 (1.82)*

-0.910 (1.79)*

PRIVATE 0.214 (2.36)**

0.191 (2.09)**

0.204 (2.22)**

0.225 (2.72)***

SHIFT 0.118 (1.33)

0.124 (1.40)

0.117 (1.32)

0.199 (2.28)**

NEW -0.316 (3.13)***

-0.321 (3.19)***

-0.288 (2.79)***

-0.307 (3.04)***

WPSIZE (x1000) -0.063 (0.62)

-0.088 (0.85)

-0.071 (0.71)

-0.085 (0.85)

FSIZE2 0.109 (1.42)

0.107 (1.40)

0.118 (1.53)

0.087 (1.15)

FSIZE3 0.323 (3.27)***

0.334 (3.38)***

0.340 (3.43)***

0.310 (3.25)***

FSIZE4 0.306 (3.44)***

0.316 (3.57)***

0.273 (3.14)***

0.265 (3.12)***

FSIZE5 0.265 (2.18)**

0.290 (2.37)**

0.226 (1.85)*

0.229 (1.90)*

FSIZE6 0.176 (1.22)

0.189 (1.31)

0.150 (1.02)

0.104 (0.75)

FSIZE7 0.021 (0.21)

0.042 (0.41)

-0.019 (0.18)

-0.196 (2.12)**

PRSHARE -0.211 (1.59)

-0.202 (1.53)

-0.131 (0.98)

PBR 0.236 (2.80)***

0.229 (2.72)***

0.179 (2.14)**

0.218 (2.61)***

22

Table 2 (cont.)

Variable (1) (2) (3) (4)

UR -7.395 (3.99)***

-7.576 (4.09)***

-7.377 (3.93)***

-6.499 (3.54)***

TENURE2 -0.169 (2.29)**

-0.162 (2.20)**

-0.159 (2.14)**

-0.166 (2.27)**

TENURE3 -0.141 (1.73)*

-0.131 (1.61)

-0.161 (1.97)**

-0.150 (1.80)*

TENURE4 -0.514 (5.83)***

-0.509 (5.79)***

-0.522 (5.91)***

-0.515 (5.76)***

TENURE5 -0.445 (3.91)***

-0.436 (3.82)***

-0.495 (4.39)***

-0.462 (3.95)***

OJT -0.289 (3.61)***

-0.286 (3.52)***

-0.291 (3.59)***

-0.374 (4.50)***

FEMALE 0.121 (0.85)

0.111 (0.78)

0.156 (1.10)

-0.012 (0.10)

YOUTH 0.547 (5.33)***

0.539 (5.25)***

0.548 (5.40)***

0.546 (5.46)***

PART 0.461 (2.97)***

0.466 (2.95)***

0.473 (2.95)***

0.542 (3.73)***

NESB 0.159 (2.31)**

0.154 (2.23)**

0.157 (2.25)**

0.189 (2.86)***

Occupation dummies

Yes

Yes

Yes

No

Industry dummies Yes Yes Yes No Adjusted R squared 0.379 0.380 0.369 0.357 F statistic 17.815*** 17.551*** 17.874*** 23.698*** Breusch-Pagan 115.146*** 117.824*** 109.350*** 80.417*** RESET test 1.729 1.861 2.001 0.255

***, ** and * indicate significance at the one, five and ten percent levels, respectively. In the case of t-ratios, these tests are two-tailed. In the case of diagnostics they are one-tailed.

23

NOTES

* The authors thank the Australian Commonwealth Department of Industrial Relations

for the data used herein, Robert Drago, Bruce Chapman, and an anonymous referee for

helpful comments and advice, and attendees at the Australian Labour Market Research

Workshop, February 1993, Perth, at which an earlier version of this paper was

presented.

1. Part of the explanation for the weakness of this relationship, however, may lie in the

youth of the sample (males aged 16 to 25 years).

2. Wilson, Cable, and Peel (1990), in their study of British engineering workplaces,

however, apparently experimented with a number of different indicators of union voice

including the existence of union-management committees and the number of union

representatives within the workplaces. They report that these indicators were out-

performed by a composite measure dominated by closed shop presence, number of

unions, and the percentage of employees belonging to a trade union.

3. Wooden (1992) has estimated, using the same data set analyzed here, that closed-shop

arrangements, both formal and informal, still cover 40 percent of all union members in

Australia.

4. As a consequence, according to data collected by the Australian Bureau of Statistics,

while only 41 percent of Australian wage and salary earners were members of a union

in 1990, 80 percent were covered by “awards”.

5. Workplaces in the agriculture, forestry, fishing and hunting, and defence industries,

however, are excluded.

6. Interviews with managers and union delegates (where applicable) were conducted at all

2,004 workplaces. The quits data, however, were collected via a questionnaire

administered by mail leading to some nonresponse.

7. In particular, large workplaces are over-represented in the sample.

8. Estimation of a sample selection model indicates that selectivity bias is not imparted to

the results by the exclusion of these cases (which only number 32 after data purging).

24

9. A number of other measures were experimented with but none performed better than

UDELEG.

10. The concept of “market competition” was meaningless for workplaces classified as

operating on a noncommercial basis. To minimize observation loss, we have, therefore,

assigned such workplaces a zero value in the construction of this variable.

11. A dummy variable signalling the presence of formal structures for facilitating

employee involvement in decision making was included in preliminary equations but

was never close to significance.

12. As Blau and Kahn (1981) note, the expected inverse relationship between tenure and

quits will be strengthened by any institutional arrangements conferring seniority-

related benefits such as long-service leave and non-vested pensions.

13. As Chapman (1981) points out, studies of quit behavior which include a simple wage

variable are likely to underestimate the size of the negative quit-wage elasticity

because of their failure to adequately control for alternative wages. Nevertheless, as

noted above, inclusion of the residual from a wage equation as a proxy for relative

wages has little affect on the results.

14. Omission of the nonunion workplaces in the sample leaves our findings qualitatively

unchanged.