Printer Friendly

Trade unions and financial performance.

1. Introduction

There is now a large body of empirical work indicating that trade unions are associated with lower profitability or financial performance. This finding emerges in studies based on microeconomic data at establishment- or firm-level, from work that uses aggregate industry-level data, and in studies based on data from various countries.(1) The British work that exists on this issue focuses on union effects in the late 1970s and early 1980s and there is currently no work that uses data after the mid-1980s. This is important as the British labour market, especially with respect to the role of unions, underwent many changes in the 1980s. Unionization, however measured,(2) fell very dramatically after the election of the Conservative government in 1979 and many have argued that this resulted in a reduction in union power. The same time period saw the introduction of several pieces of anti-union legislation that were designed to limit the scope of trade unions in affecting various economic outcomes.(3)

It is worth stressing that the major part of the 1980-90 fall in unionization occurred after 1984. Take the case of trade union recognition. According to the Workplace Industrial Relations Surveys (WIRS) of 1980, 1984, and 1990 the proportion of British establishments (private and public sector) that recognized unions for collective bargaining purposes was 0.64 in 1980, 0.66 in 1984 and 0.53 in 1990 (Millward et al. 1992, Table 3.7). Furthermore, the fact that moves to make the closed shop, which is where most agree that unions were traditionally at their strongest, legally unenforceable occurred primarily from 1984 onwards with the provisions specified in the Employment Acts of 1982 and 1990, with the use of the former not coming into effect until 1984.

The fact that much of the union decline seems to have occurred after 1984 (and extremely sharply when the decline set in) makes a more recent re-evaluation of the relationship between unions and the 'bottom line' of corporate performance an important one. This is what we provide in this paper. We consider the relationship between trade unions and financial performance in 1990 using establishment-level data from the Workplace Industrial Relations Survey of that year.(4) Our first issue of interest concerns whether, following the large upheavals of the 1980s, there still exists a relationship between union presence and financial performance. The second issue is to identify conditions where a union effect exists.

Our results suggest roughly a halving of the overall union effect between 1984 and 1990 and evidence of a less widespread union effect in 1990 (as compared with our earlier work: see Machin and Stewart 1990). However, there are still situations where unions seem to be negatively associated with financial performance. With respect to union structure, unions are seen to exert a negative influence on financial performance only where the closed shop (or a pseudo-closed shop supported by management - see below) is still in place. Over and above this we report results suggesting that unionized establishments have lower financial performance only where there are closed shop arrangements and the establishments have some product market power and that this effect is greater where managerial freedom to allocate tasks is limited by union work rules. This latter combination occurs in only about one in ten of the workplaces in our sample.

The layout of the rest of the paper is as follows. In Section 2 we describe trends in union presence and corporate performance in the 1980s, and provide a brief overview of existing work on the relation between unions and performance (broadly defined) in this time period. In Section 3 we describe the nature of the data that we use, and in Section 4 present econometric models of the determinants of financial performance in 1990. In Section 5 we attempt to identify situations where one can detect a union-performance relationship, and finally Section 6 concludes.

2. Unions and performance in Britain in the 1980s

2.1. What happened to unions in the 1980s?

It is well known that there was a very sharp decline in union activity in Britain in the 1980s. The percentage of private sector establishments with recognized union(s) in the 1980 Workplace Industrial Relations Survey was 53; by 1984 this fell to 48%; by 1990 it had dropped to 39%. Other indices of union activity paint a similar picture: however one chooses to measure union presence (e.g. collective bargaining coverage; membership density; incidence of the closed shop) there was a sharp fall in unionization over the 1980s.

Several features underpin this decline. Those that seem to have been pin-pointed as potential explanations are: compositional changes in the labour market (i.e. the move away from traditionally unionized areas of employment); increased management opposition; anti-union legislation; a reduced demand for union representation; and unfavourable public opinions towards unions. Recent examinations (Disney et al. 1995; Miliward 1994) have stressed that the decline appears to be due to a failure to organize new establishments set up in the 1980s, most of which operate in an environment free of unions.

The 1980s also saw the introduction of a number of pieces of anti-union legislation, following the election of the Conservative government in 1979. The 1980 Employment Act placed additional restrictions on picketing, made unions liable to be sued for damages, enlarged the grounds on which to claim unfair dismissal for refusal to join a closed shop, and repealed the statutory recognition procedure. The 1980 Social Security Act reduced the benefits paid to strikers' families and the codes of practice on picketing and closed shops introduced in 1980 imposed additional restrictions on the numbers and conduct of pickets and on the operation of closed shop agreements.

The 1982 Employment Act reduced the range of disputes considered lawful, increased trade unions' liability for damages, and introduced a requirement of 80% support in a ballot to legalize a closed shop (although this latter did not come into effect until 1984). The 1984 Trade Union Act required secret ballots before industrial action, executive elections every five years by secret ballot, and political fund ballots every ten years. The 1986 Public Order Act introduced a criminal offence in relation to picketing.

The 1988 Employment Act made action to preserve a post-entry closed shop unlawful, placed further restrictions on industrial action and election ballots, required unions to compensate members disciplined for not complying with majority decisions, opened union finances to inspection, and prevented unions from paying the fines of officials and members.

In the 1990s, and after the date of the data used in this paper, the 1990 Employment Act made it unlawful to refuse to employ non-union members (i.e. effectively undermined the pre-entry closed shop), made all secondary action unlawful, and made selective dismissal of strikers taking unofficial action possible. Most recently, the 1993 Trade Union Reform and Employment Rights Act allows individuals to seek injunctions against unlawful action, requires all strike ballots to be postal, and requires written consent for check off to be reobtained every three years.

It is evident that these legislative measures, though difficult to quantify, may well have had an impact on the way in which we intend to measure unionization in our empirical work. For example, researchers have in the past considered heterogeneity in estimated union effects on economic outcomes, and one specific issue that has been considered is whether one is able to identify a closed shop effect over and above any basic union effect (e.g. Stewart 1987, for union wage effects; or Machin, 1991, for union productivity effects). These legislative changes (at least those before our sample period) may well have altered the role of the closed shop. This should be borne in mind when we define the union variables that we consider.

2.2. What happened to corporate performance in the 1980s?

At the same time as the union decline and the introduction of the anti-union legislation, corporate performance showed a marked improvement. Figure 1 plots an aggregate trading profit margin, defined as the gross trading profits of companies and financial institutions divided by gross domestic product, between 1980 and 1990 and, whilst it displays a procyclical pattern (as in Machin and Van Reenen's, 1993, firm-level study), aggregate profitability appears to be higher in 1990 than in 1980. As these two points are (broadly) at similar phases of the business cycle this seems to suggest an improvement in the profit performance of UK companies that occurred during the 1980s.

There is some interesting work on the 1980s behaviour of UK profitability. For instance, Haskel (1993) uses industry data and concludes that the fall in unionization was a large factor underpinning the rise in profitability. Haskel and Martin (1994) have also looked at these trends and remarked that the 1980s saw British industry become 'leaner and fitter'. An obvious question that emerges is whether the performance improvement has been made easier by the move away from collective bargaining arrangements in workplaces operating in sectors across the economy. We hope to shed some light on this below.

2.3. Evidence on unions and performance in the 1980s

There is a small body of British evidence suggesting that unionized establishments and firms have lower levels of financial performance than their non-union counterparts. Blanchflower and Oswald (1988) and Machin and Stewart (1990) have presented establishment-level findings based on data from 1980 and 1984 suggesting that financial performance is lower where unions are recognized for collective bargaining purposes. In a similar vein, Machin (1991) and Cable and Machin (1991) use (quite small) samples of firms in the early to mid-1980s and show that unionized firms have lower accounting profit margins than do non-union firms. These negative union profit effects are entirely in line with the large body of US evidence on this issue (Belman 1992).

There is also some evidence that unionized establishments and firms improved their relative economic position in the 1980s. Several pieces of work point to the notion that productivity growth was higher in union firms, both at the start of the decade (Nickell et al. 1992) and at the end (Gregg et al. 1993). There is also evidence that unionized establishments were more prone to organizational changes in the early 1980s (Machin and Wadhwani 1991) and that this may well have contributed positively to their overall performance. Finally, Menezes-Filho (1994) uses company-level panel data and reports some results indicating that the negative union effect on firm profit margins diminished through the 1980s.

Hence, as the 1980s progressed we have a picture of reduced (and perhaps weakened) union presence in Britain, together with a simultaneous improvement in corporate performance. In the remainder of this paper we study this in more detail and attempt to identify the circumstances where there may be a link, if any, between unions and financial performance using establishment-level data for 1990.

3. Data description and modelling procedure

3.1. The workplace industrial relations surveys

The Workplace Industrial Relations Surveys are nationally representative surveys of around 2,000 British establishments that were conducted in 1980, 1984, and 1990. They contain very rich information on industrial relations issues, and some (more limited) information on the economic performance of the establishments in the surveys. The surveys have been used extensively to analyze various issues of interest to labour economists and industrial relations researchers, ranging from empirical studies of the union wage mark-up (Stewart 1987, 1990, 1991), to the determination of union status (Disney et al., 1995) and on to issues associated with 'non-unionism' or 'human resource management'.(5)

3.2. Establishment-level financial performance

For the purposes of this paper the key question asked of managers in private sector establishments is the following:

How would you assess the financial performance of this establishment compared with other establishments in the same industry? Would you say it was

. . . better than average . . . below average . . . or about average?

Respondents in the better than average and below average categories were then asked:

'Is that a lot (below) or a little better (below)?: (i) lot; (ii) little.'

Mean responses to this question from the three Workplace Industrial Relations Surveys are reported in Table 1 and union/non-union differences in Table 2. The first point to note is that a greater proportion of managers feel they are doing better than average. This does raise some questions regarding the suitability of this variable as an index of establishment performance. We have articulated elsewhere (Machin and Stewart 1990) that we feel this variable is of considerable use and provides a valuable alternative to measures like accounting profits, price-cost margins, or stock market valuations, that have very clear limitations of their own. Most importantly it actually reflects what managers think financial performance is, and therefore does not suffer from the measurement problems that plague the other, more conventional, measures of financial performance.

Responses to the financial performance question from the three
Workplace Industrial Relations Surveys

                              1980          1984          1990

A lot above average           0.477         0.241         0.254
A little above average                      0.224         0.303
About average                 0.464         0.468         0.373
A little below average        0.059         0.037         0.054
A lot below average                         0.029         0.016
Number of establishments   1,172/1,386   1,077/1,142   1,105/1,080


1. Based on all private sector establishments without missing
on the financial performance question.

2. Weights are based on the Census of Employment three years prior
to the relevant survey and due to the deliberate oversampling of
larger establishments to guarantee their presence in the sample
Daniel and Millward 1983; Millward and Stevens 1986; and Millward
al. 1992 for more details on each of the three surveys).

3. In 1980 managers were only asked if their establishment's
financial performance was above, about or below average relative to
their industry.


A second reason why we think the variable contains important information is that, when we test whether one observes systematic differences in the reporting of this variable (by a method described below), we are unable to reject the null hypothesis that there is no systematic reporting difference by union and non-union establishments or across other observable characteristics of the establishment.

Thirdly, we can also supply more supportive evidence regarding its suitability. In the 1984-90 trading panel of establishments contained in the Workplace Industrial Relations Surveys one can identify establishments that shut down in the 1984-90 time period (see Machin 1995; or Millward 1994). Tabulation of plant closure 1984-90 against the reported relative financial performance in 1984 gives the following (weighted) proportions:


The summary statistics(6) illustrate that those establishments reporting below average financial performance vis-a-vis their competitors in 1984 are significantly more likely to have closed down between 1984 and 1990. This significant correlation with closure clearly adds more weight to the validity of this particular measure of financial performance.

It is, therefore, our view that this variable is a useful indicator of the financial performance of establishments who are sampled in the Workplace Industrial Relations Surveys. Considering the descriptive statistics on this variable, reported in Table 1, there appears to have been a slight shift over time, indicating higher levels of establishment performance in 1990 than in the earlier years. In 1990, 56% reported above average performance, as compared to 46% in 1984 and 48% in 1980. Similarly, in 1990 7% reported their financial performance to be below average relative to their competitors, whilst comparable percentages for 1980 and 1984 were 6% and 7% respectively.

Turning to the average union/non-union differences in Table 2, one can see a clear pattern illustrating that, in all years, unionized establishments are more likely to have below average performance levels and less likely to have above average financial performance. For example, in 1980 unionized establishments were just over 9% less likely to have above average performance, whilst in 1984 a comparable percentage was 10%, and by 1990 this fell slightly to 8%. One potentially interesting change that is masked in this comparison concerns the 1984-90 fall in the union related difference between reporting well above average performance as compared to a little above average. We investigate this in more detail below.

3.3. Modelling procedure

The financial performance variable is a categorical indicator defined in terms of ordered responses. We estimate our models using an Ordered Probit estimator. Denoting the (relative) financial performance variable for establishment i as F[P.sub.i] the basic specification we are interested in estimating is of the form

F[P.sub.i] = [X[prime].sub.i][Beta] + [[Epsilon].sub.i] (1)

where X is a set of independent variables and [Epsilon] an error term.

Consider the nature of the dependent variable. It is clear that F[P.sub.i] is related to the qualitative indicator of financial performance that we actually observe, say [[Pi].sub.i], as follows

[[Pi].sub.i] = 0 if F[P.sub.i] [less than] [[Omega].sub.1]

= 1 if [[Omega].sub.1] [less than or equal to] F[P.sub.i] [less than] [[Omega].sub.2]

= 2 if [[Omega].sub.2] [less than or equal to] F[P.sub.i] [less than] [[Omega].sub.3]

= 3 if [[Omega].sub.3] [less than or equal to] F[P.sub.i] [less than] [[Omega].sub.4]

= 4 if [[Omega].sub.4] [less than or equal to] F[P.sub.i] (2)

Equation (2) relates the qualitative indicator of performance, [[Pi].sub.i], to the actual underlying indicator of performance, F[P.sub.i], where the intervals between each performance category are determined by the (unobserved) thresholds [[Omega].sub.j] (j = 1, 2, 3, 4). One can think of the levels of F[P.sub.i], the underlying continuous measure of performance, being distributed along the real line and the thresholds dividing this line into the five categories we observe, with the top and bottom categories being open-ended.

The Ordered Probit estimator that we use gives Maximum Likelihood estimates of the [Beta] parameters and of the [[Omega].sub.j].(7) The estimator maximizes the following likelihood function:

[Mathematical Expression Omitted]

where [Phi]([center dot]) denotes the standard normal distribution function (so that [[Omega].sub.0] = - [infinity], [[Omega].sub.5] = [infinity]).

We are careful in our empirical work to test the suitability of the estimated models by considering a number of diagnostics, designed specifically for Ordered Probit models (these are described in more detail in Machin and Stewart 1990, and are modified versions of the score tests based on pseudo or generalized residuals for limited dependent variable models as in Chesher and Irish 1987, or Gourieroux et al. 1987). The basic diagnostics we present are tests of standard hypotheses relating to misspecification via functional form or heteroskedasticity, and of the assumption of normality.

We also formulate a test to examine whether the thresholds [[Omega].sub.j] differ systematically with any of the variables contained in X (or with any other variables). This is important for the reasons discussed above pertaining to the suitability of the dependent variable. To be more specific, this heterogeneity in thresholds test is formulated as a test of the null hypothesis that the thresholds do not vary over the observations: i.e. it is a test of whether [Theta] = 0 if the thresholds are allowed to differ with observable establishment characteristics, say [q.sub.i], via the parameterization [[Omega].sub.ji] = [[Omega].sub.j] + [q[prime].sub.i][Theta]. Clearly if [Theta] = 0 we are unable to reject the hypothesis that the thresholds that delineate performance levels are not systematically related to the observable characteristics in [q.sub.i]. As will be seen below, the statistics relating to this test generate more confidence in the validity of the financial performance variable, particularly as they do not differ systematically with the measures of union presence that we consider.

4. Econometric models of financial performance in 1990

4.1. Basic models

Table 3 presents a set of estimates of financial performance equations for a sample of 892 private sector establishments in 1990. The four specifications [TABULAR DATA FOR TABLE 3 OMITTED] reported differ in their specification of the union variable and in whether or not they include a set of control variables. Column 1 simply includes a dummy variable indicating whether or not manual union(s) are recognized for collective bargaining purposes, whilst column 2 additionally includes a dummy variable indicating whether closed shop arrangements for manuals are present or whether management recommends that manual workers join a union.(8) This latter variable is included as it seems likely that it indicates situations of greater union strength. We feel that this variable will pick up the presence of a strong union as the recommendation itself is probably a recognition of the de facto nature of closed shops in the presence of high membership. It is worth pointing out that the period 1984 to 1990 saw a dramatic decline in the incidence of the closed shop, but no decline in the number of establishments where management recommends union membership. Thus, this latter category increased as a proportion of establishments with manual union recognition (Stewart 1995). Given the legislative changes, there is now relatively little to distinguish remaining closed shops from management recommendation situations and Stewart (1995) finds that, ceteris paribus, the wages paid in these two situations are not significantly different from one another. Columns 3 and 4 are the column I and 2 specifications with a set of controls describing establishment characteristics additionally included.

In column 1 the coefficient on the union recognition variable is estimated to be negative and significantly different from zero, indicating, as in Table 2, that unionized establishments have lower financial performance levels. In column 2 we also include the closed shop/management recommends dummy variable. Its inclusion causes the coefficient on the basic recognition variable to drop considerably and lose statistical significance. The coefficient on the closed shop/management recommends variable is estimated to be negative and strongly significant.

In columns 3 and 4 we add in the control variables and, focusing on the statistically preferred specification in column 4, it is clear that any union effect on financial performance in 1990 is confined to those unionized establishments which have closed shop arrangements or where management recommends that manual workers be union members.(9) This is an interesting result as it suggests that, by 1990, unions only reduced financial performance in a sub-set of unionized establishments, namely those where they were stronger.(10) This contrasts with our earlier work based on data for 1980 and 1984 where we identified a significant negative effect associated with manual union recognition, and no additional effects from the closed shop/management recommends membership variable.(11) The closed shop or management recommends variable does not appear to be acting simply as a proxy for union density: when a variable measuring union membership density among manual workers at the establishment is added to the specification in column 4, the closed shop or management recommends coefficient changes to -0.346 (0.105). The manual unions recognized coefficient remains statistically insignificant as is that on the added density variable.(12) We explore further heterogeneities in the estimated union effect in more detail in the next section of the paper.

Looking at the diagnostics reported in the table, one can see that the equations seem to perform reasonably well. One cannot reject the null hypotheses of homoskedasticity and normality and the equations do not seem to suffer from functional form misspecification (from a RESET-type test based on adding second, third, and fourth powers of the fitted value). Probably of most interest is the test for heterogeneity in the thresholds: for the preferred model in column 4 one cannot reject the null hypothesis that the thresholds do not differ with the right-hand side variables (the score test statistic is 34.55 as compared to a 5% critical value of 40.12). Even more importantly one can set up a test of whether the estimated thresholds differ with specific variables: [[Chi].sup.2](3) statistics for the hypothesis that the thresholds differ with manual recognition and the closed shop/management recommends variable were 1.30 and 1.26 respectively, both of which lie well below the 5% critical value of 7.82. This is clearly reassuring as it suggests that the thresholds between performance categories do not differ systematically with the measures of union presence that we consider.

The specifications in Table 3 also point to superior financial performance in those establishments that are operating at full capacity, and in those where the value of sales of the main products or services of the establishment had been rising over the 12 months prior to the interview date. There is also evidence that UK owned establishments have higher relative financial performance. The other two interesting results are the negative effect of 1-digit industry union recognition,(13) and the positive impact of having a bigger employment share within the 4-digit industry in which the establishment operates (although this latter is of marginal significance when included in this additive form).(14)

4.2. Robustness of results

The union effect identified in column 4 of the table, operating through the closed shop/management recommends membership variable, seems to be a robust finding. In this sub-section we describe a number of additional experiments:

(i) inclusion of a set of one-digit industry dummies in place of the industry-level union recognition variable resulted in coefficient estimates (standard errors) on the union recognition and closed shop/management recommends variables of 0.009 (0.100) and -0.275 (0.095). As compared to the model in column 4 of Table 3 a [[Chi].sup.2](7) statistic of 13.57 (5% critical value = 14.07) suggested that the inclusion of the industry dummies did not add significantly to the explanatory power of the equation. To put it another way, we cannot reject the hypothesis that 1-digit industry can be simplified to the industry-level recognition variable;

(ii) inclusion of a more disaggregated set of 48 2-digit industry dummies produced coefficients and standard errors of -0.064 (0.107) on manual union recognition and -0.222 (0.102) on the closed shop or management recommends union membership variable;

(iii) the only size variable included thus far is the relative size variable (i.e. the employment share above threshold variable), but inclusion of an additional set of five absolute size dummies (for 50-99, 100-199, 200-499, 500-599, and 1,000 + employees compared to a reference group of 25-49) made little difference to the results. They were jointly insignificant ([[Chi].sup.2](5) = 5.39, 5% critical value = 11.07) and the coefficients (standard errors) on the manual union recognition and closed shop/management recommends variables were 0.009 (0.100) and -0.313 (0.095) respectively;(15)

(iv) we also included a set of six additional relative employment size dummies with little effect. They were jointly insignificant ([[Chi].sup.2](6) = 11.89, 5% critical value = 12.59) and the coefficients (standard errors) on the manual union recognition and closed shop/management recommends variables were 0.008 (0.099) and -0.278 (0.094) respectively; and

(v) we also defined a wider set of control variables including several variables relating to workforce characteristics (manual, skilled, female, part-time proportions), although for a smaller sample of 820 establishments. Again, the model remained robust to this, with the estimated coefficient and associated standard errors on manual recognition and closed shop/ management recommends of 0.032 (0.104) and -0.274 (0.100).

4.3. Union/non-union probability differences

The negative effect of unions on financial performance seems to be confined to establishments which have relatively strong unions, as manifested by some form of compulsory or management recommended union membership arrangement. In terms of the magnitude of this union effect we can convert the Ordered Probit coefficient estimates into ceteris paribus union/non-union probability differences. The preferred specification in Table 3 is from an econometric model of the form

F[P.sub.i] = [[Alpha].sub.1]U[N.sub.i] + [[Alpha].sub.2]C[S.sub.i] + [Z[prime].sub.i][Gamma] + [[Epsilon].sub.i] (4)

where UN denotes manual union recognition, CS the existence of closed shop or management recommends membership arrangements, and Z the other controls included in the financial performance equation.

The difference in the probability of a given performance level between a unionized establishment with a closed shop or management recommended membership arrangements and a comparable non-union establishment is given by

[Delta][P.sup.cs][[FP.sub.i] = j] = {[Phi][[[Omega].sub.J + 1] - ([[Alpha].sub.1] + [[Alpha].sub.2] + [Z[prime].sub.i][Gamma])] - [Phi][[[Omega].sub.j] - ([[Alpha].sub.1] + [[Alpha].sub.2] + [Z[prime].sub.i][Gamma])]}

- {[Phi][[[Omega].sub.J + 1] - [Z[prime].sub.l][Gamma]] - [Phi][[[Omega].sub.j] - [Z[prime].sub.i][Gamma]]} (5)

This expression gives the probability difference for a given [Z.sub.i]. The mean probability difference is given by evaluating it at the means of the Z-vector. There is, as usual, an issue as to which means should be used for this purpose. In the calculations reported below, we use the means of the establishments with a closed shop or management recommended membership arrangement.

The probability difference for an establishment with manual union recognition but no closed shop or management recommended membership is similarly given by

[Delta][P.sup.un][[FP.sub.i] = j] = {[Phi][[[Omega].sub.J + 1] - ([[Alpha].sub.1] + [Z[prime].sub.i][Gamma])] - [Phi][[[Omega].sub.j] - ([[Alpha].sub.1] + [Z[prime].sub.i][Gamma])]}

- {[Phi][[[Omega].sub.J + 1] - [Z[prime].sub.i][Gamma]] - [Phi][[[Omega].sub.j] - [Z[prime].sub.i][Gamma]]} (6)

and the mean probability difference can be calculated equivalently. In this case we use the means of the unionized establishments without a closed shop or management recommendation. The above two sets of probability differences are calculated from the model given in column 4 of Table 3. We also use the results in column 3 to calculate overall mean union probability differences in an analogous way. For both these and those defined earlier we also present equivalent estimates for 1984.

The overall mean union probability difference, evaluated at the union recognition (weighted) means from column 3 of Table 3 and the equivalent model for 1984 are as follows:
Mean union probability differences

                                 1990        1984

A lot above average             -0.043      -0.095
A little above average          -0.017      -0.031
About average                    0.035       0.082
A little below average           0.015       0.023
A lot below average              0.010       0.021

Thus in 1990 establishments with manual union recognition are 6% less likely to report above average financial performance and 2.5% more likely to report below average. This shows a considerable reduction compared with 1984 when they were 12.6% less likely to report above average financial performance and 4.6% more likely to report below average. Thus the overall impact of recognition shows a marked decline over the period, roughly halving the effect.

Turning to the model in which we distinguish closed shop and management recommends establishments from the others with recognition (column 4), the estimated probability differences are as follows:
Mean union probability differences

                           Closed shop/man.       Union recognition
                             recommends             but no cs/mr

                           1990       1984       1990          1984

A lot above average       -0.083   -0.099      -0.005        -0.087
A little above average    -0.043   -0.035      -0.001        -0.027
About average              0.067    0.086       0.004         0.076
A little below average     0.034    0.025       0.001         0.020
A lot below average        0.025    0.023       0.001         0.018

Thus in 1990 establishments with a closed shop or management recommendation of union membership were 12.6% less likely to report above average financial performance and 5.9% more likely to report below average than a corresponding non-union establishment. These are similar to the figures for 1984. However, the probability differences for establishments with manual union recognition but no closed shop or management recommendation of union membership are all negligible in 1990, reflecting a dramatic decline from those in 1984.

Thus overall we may conclude that the overall impact of unions on financial performance has roughly halved. This change is made up of relatively little change for establishments with a closed shop or management recommendation, but a complete collapse of the effect for establishments with manual union recognition but no closed shop or management recommendation of union membership.

5. Where do unions have an effect on financial performance?

The results of the previous section point to the existence of a negative union effect on financial performance in 1990, though the effect does not apply to all union establishments, and is confined to situations where stronger unions are present. In this section of the paper we consider possible heterogeneities in the union effect in more detail by explicitly focusing on two particular sources of potential union effects, related to product market structure and to the existence of union work rules.

5.1. Product market structure

The first source of heterogeneity that we consider relates to the relative size of establishments. We consider variations in the union effect by relative size in order to test the notion that unions are more likely to reduce financial performance in establishments which have greater product market power. We found this to be the case in our earlier work (Machin and Stewart 1990) and other authors have reported results indicating that unions reduce profitability by more in such situations (Machin 1991, is a British example; Freeman 1983, a US example). There is also evidence that union wage differentials are higher in less competitive circumstances (Stewart 1990).

In this sub-section we examine this notion by letting the union effects ([[Alpha].sub.1] and [[Alpha].sub.2] in eq. (4)) differ between high relative share establishments ([REL.sub.i] = 1) and low relative share establishments ([REL.sub.i] = 0), by using the specification

[[Alpha].sub.1] = [[Alpha].sub.11] [REL.sub.i] + [[Alpha].sub.12](1 - [REL.sub.i]) (7)

[[Alpha].sub.2] = [[Alpha].sub.21] [REL.sub.i] + [[Alpha].sub.22](1 - [REL.sub.i])

Hence we can test whether the basic union recognition effect differs between high and low relative share establishments by testing whether [[Alpha].sub.11] and [[Alpha].sub.12] are significantly different from one another, and likewise for closed shop/management recommends effects by considering [[Alpha].sub.21] and [[Alpha].sub.22].

We report estimates of the union effect stratified by relative size in this way in Table 4. Column 1 is the column 4 Table 3 specification which is reproduced to facilitate comparison. Column 2 is the interactive model for the 1990 data, whilst columns 3 and 4 give the results for the same models estimated on data from the earlier 1984 survey.


In column 2 the union effect is estimated to be more negative in higher relative share establishments, but only if there is a closed shop or management recommendation. In addition the negative closed shop/management recommends effect is confined to high employment share establishments in 1990. The other three coefficients are jointly and individually insignificant ([[Chi].sup.2](3) statistic = 1.33). In addition the restricted model including the variable for the closed shop/management recommends and high relative employment share category, but excluding the other three, dominates column 1 in likelihood terms.(16)

In columns 3 and 4 we therefore report specifications comparable to columns 1 and 2 estimated from data in 1984.(17) Whilst it is also the case that the interaction with relative employment shares only matters for the closed shop/management recommends variable, there is also a negative effect associated with recognition irrespective of relative employment shares. It appears that there is a less widespread union effect in 1990, though the result that negative union effects are more likely to be present in high relative share establishments, which we take to proxy product power, remains.

The mean probability differences between establishments with both a closed shop or management recommendation and higher relative employment share and comparable non-union establishments are given below. They are based on the estimates in Table 4, column 2 and are evaluated at the means of the former group.
Mean probability differences

A lot above average              -0.094
A little above average           -0.052
About average                     0.076
A little below average            0.040
A lot below average               0.030

5.2. Limits to managerial freedom

The industrial relations literature (and some work in economics) has often argued that one way in which union strength emerges is via an ability of unions to define job tasks, via job demarcation rules or by controlling work practices. The 1990 Workplace Industrial Relations Survey contains a (2-part) question related to this:

In practice is management here able to organize work as it wishes among non-managerial employees, or are there limits to the way it can organize work?: (i) management able to organize as it wishes; (ii) limits to way management organizes work, and What limits the way management can organize the work here?: (i) opposition from groups of ordinary union members; (ii) opposition from groups of workers who are not union members; (iii) opposition from shop stewards or representatives; (iv) formal agreements with trade unions

In this section we consider whether union effects on performance differ with the presence of a union that is able to affect the managerial allocation of job tasks. We define a variable LIMIT equal to I if management responded that there are limits to the way it can organize work and if the opposition is from ordinary union members, shop stewards/representatives, or via formal trade union agreements. The omission of the second category in the second question ensures that we are considering a union limits variable.

The methodology is the same as in the previous section and we parameterize [[Alpha].sub.1] and [[Alpha].sub.2] as a function of limits to managerial freedom ([LIMIT.sub.i] = 1), or no limits ([LIMIT.sub.i] = 0) in

[[Alpha].sub.1] = [[Alpha].sub.13][LIMIT.sub.i] + [[Alpha].sub.14](1 - [LIMIT.sub.i])

[[Alpha].sub.2] = [[Alpha].sub.23][LIMIT.sub.i] + [[Alpha].sub.24](1 - [LIMIT.sub.i])

Table 5 reports a set of estimates of the determinants of financial performance, including interactions between the union variables and the [LIMIT.sub.i] variable. Column 1 is the basic model from before, column 2 presents a four-way breakdown of the two union variables and the limits variable, whilst columns 3 and 4 present some simplified versions.

The estimates in column 2 point to an important interaction between the union effect and the presence of a union which can affect the process of work organization though, as in the case of the relative share interactions, it is confined to the closed shop/management recommends variable. It is easy in statistical terms to simplify the model to one which only includes the closed shop/management recommends variable and its interaction with [LIMIT.sub.i] (the appropriate [[Chi].sup.2](2) statistic being 0.05, which lies well beneath the 5% critical value of 5.99) and this model is reported in column 3.

The model in column 3 suggests that unions reduce financial performance by more where a closed shop (or pseudo closed shop recommended by management) exists and unions can limit the ability of management to freely organize work. There is a weak negative effect (significant at the 10% but not at the 5% level) associated with the closed shop/management recommends effect where there are no union imposed limits to work organization.

The final column of the table reports the results for the specification excluding the closed shop/management recommends and no limits variable. Testing against column 2 gives a [[Chi].sup.2](3) statistic of 3.75 (compared with a 5% critical value of 7.82). There is a strongly negative union effect where there is a closed [TABULAR DATA FOR TABLE 5 OMITTED] shop (official or management recommended) and where unions can have an impact of the organization of work. This is true in about one-eighth (weighted) of unionized workplaces.

In terms of the magnitude of the effect, we can again convert the coefficient estimates into ceteris paribus probability effects. Doing so produces the following results that indicate differences in the probability that establishments with closed shops or management recommends union membership and where unions affect managerial freedom have a given level of financial performance as compared to other establishments:
Mean probability differences

A lot above average          -0.091
A little above average       -0.071
About average                 0.065
A little below average        0.051
A lot below average           0.046

Ceteris paribus, establishments with closed shops and union induced limits to managerial freedom are about 10% more likely to have below average financial performance and 16% less likely to have above average financial performance. This amounts to a sizable effect, though, as noted above, it is confined to only about one in eight of the unionized workplaces in the sample.

5.3. General specifications

We finally considered a general model, based on including a full set of interactions between the union variables and the relative employment and limits variables. Consideration of which effect dominates in statistical terms points to the employment share variable. The interaction terms with the manual union recognition variable are individually and jointly insignificant. Column 1 of Table 6 gives the results for the specification with them excluded, i.e., with the interactions with the closed shop variable only included. (The likelihood ratio test against the general 8-way split gives a [[Chi].sup.2](4) statistic of 5.23.) It is clear from column 1 that the effect in establishments with low relative employment share is negligible whether or not there are limits to managerial freedom. These two variables are therefore excluded in the specification presented in column 2. The test of their joint significance gives a [[Chi].sup.2](2) statistic of 0.007.

Both the high relative employment share effects are significantly negative, with that where there are also limits to managerial freedom being roughly double that where there are not. The specification in column 3 excludes the smaller effect, while that in column 4 combines the two effects into a single high relative share effect. The two effects in column 2 are significantly different from one another at the 10% level, but not at the 5% level. Although the simplifications of column 2 to columns 3 and 4 are both rejected (at the 5% level), neither is rejected against the general 8-way split referred to above. Despite this, column 2 is regarded as the preferred specification and probability differences are calculated on the basis of it.(18) These are given for establishments with a closed shop or management recommendation of union membership and a high relative employment share. Mean differences are given for establishments with and without limits to managerial freedom. In both cases the means used are those for that particular group.
Mean probability differences

                           Limits to           No limits to
                      managerial freedom     managerial freedom

A lot above                 -0.138                  -0.076
A little above              -0.072                  -0.027
About                        0.111                   0.063
A little below               0.057                   0.025
A lot below                  0.042                   0.016

Ceteris peribus, establishments with closed shops or management-recommended membership and high relative employment share are about 21% less likely to have financial performance above average if there are union-induced limits to managerial freedom, and about 10% less likely if there are not, compared with other comparable establishments. Correspondingly they are about 10% more likely to have below average financial performance if there are such limits and 4% more likely if there are not. The former group constitute about 10% of those establishments in which unions are recognized for manual workers and the latter group about a further 15%.

6. Concluding remarks

In this paper we have investigated the relationship between trade unions and financial performance using establishment-level data from the 1990 Workplace Industrial Relations Survey, a nationally representative survey of British establishments employing 25 or more workers. The large upheavals that occurred in the British labour market in the 1980s, especially with respect to the role of unions and their economic effects, make this an important topic for investigation.

We estimate that the overall impact of manual union recognition on financial performance in 1990 is roughly half what it was in 1984 and that, by 1990, there was evidence of a less widespread negative association. It would seem that unions are less successful in extracting a share of the quasi-rents for their [TABULAR DATA FOR TABLE 6 OMITTED] members than they used to be. Within this overall average we find the effect in establishments with a closed shop or where management recommends union membership to have remained roughly constant, but that the effect in the remainder of unionized establishments has collapsed completely. Our interpretation of the results is that stronger unions can still extract a share of the rents, but the weaker unions no longer can in the face of the management/legislative counter-attack of the 1980s.

In addition to a closed shop or management recommendation, we find that a high relative employment share of the four-digit industry, taken to reflect product market power, is also required for there to be an effect. Within this group we then find the effect to be roughly double if unions are able to limit managerial freedom to allocate tasks. This latter group (with a closed shop or management recommendation, a higher relative employment share and union limits on managerial freedom) constitute about one in ten of the establishments with manual union recognition.

1 For UK work based on micro-data see Blanchflower and Oswald (1988), Cable and Machin (1991), Machin (1991), and Machin and Stewart (1990); for an industry-level analysis, see Conyon and Machin (1991); for a review of the UK work see Metcalf (1994); and for the cross-country comparison, see the large body of US work (examples are Freeman 1983, or Karier 1985, and the US literature is reviewed in Belman 1992).

2 For example, between 1980 and 1990 recognition of manual or non-manual trade unions for collective bargaining over pay and conditions in British establishments fell by almost 20% (Millward et al. 1992); the proportion of workers covered by a collective agreement fell from 0.71 in 1984 to 0.54 in 1990 (Millward et al. 1992). The fall in aggregate union density was from 54% to 38%.

3 Indeed, the assault on trade unions has continued into the 1990s: for example, the 1993 Trade Union Reform and Employment Rights Bill contained a number of actions designed to further limit the ability of unions to affect economic outcomes.

4 A recent firm-level study by Menezes-Filho (1994) considers the relationship between profitability and unionization using firm-level panel data over the 1980s.

5 See Millward (1993) for a survey of research based on the Workplace Industrial Relations Survey data.

6 All the summary statistics reject the null hypothesis of no relation between closure and 1984 financial performance. The [Gamma] and [[Tau].sub.b] statistics are non-parametric statistics for ordinal data as implemented in STATA 3.1 ([Gamma] and [[Tau].sub.b], are Goodman and Kruskall's gamma statistic and Kendall's tau respectively: more details on methods of computation are in the STATA 3.1 user's manual).

7 The J thresholds can only all be identified if the constant is omitted. (If it is included, only (J - 1) can be). [Epsilon] is taken to have a standard normal distribution (with unit variance).

8 The precise categories are: all or some manual workers have to be members of unions to get or to keep their jobs (closed shop); management strongly recommends that all or some manual workers be members of a union (management recommends).

9 One cannot reject the null hypothesis that the closed shop and management recommends categories be lumped together: if entered separately the approximate [[Chi].sup.2](1) statistic testing whether they are different was 0.08 compared to a 5% critical value of 3.84. This accords with the finding for wages in Stewart (1995).

10 This finding is robust to the exclusion of the industry-level recognition variable. The estimated coefficient on the manual unions recognized variable remains statistically insignificant, whilst that on closed shop management recommends changes only slightly: to -0.303 (0.094).

11 For instance, in a specification comparable to column 4 of Table 3 estimated for a sample of 957 private sector establishments in the 1984 Workplace Industrial Relations Survey (reported below in column 3 of Table 4) the estimated coefficients (standard errors) on union recognition and closed shop/management recommends were -0.285 (0.102) and -0.050 (0.095).

12 If five density group dummies are used in place of the continuous variable, the picture is very similar. The coefficient on the closed shop or management recommends variable is -0.330 (0.108), while that on recognition is insignificant and the five density dummies are individually and jointly insignificant.

13 This variable is insignificant in a 1984 equivalent specification as are 1-digit industry dummies when included.

14 This latter is the effect associated with the employment share being above a threshold that is estimated by combining dummies that are defined for unit width intervals of the log of (establishment employment/4-digit industry-level employment) in a data consistent way. 35% (weighted) of the establishments are classified as high share (employment above threshold) under this definition. A further experiment based on a more detailed consideration of relative size dummies is reported below.

15 A referee suggested that we consider a further robustness test of this kind by simply including establishment size and industry dummies (i.e. dropping the other controls). Inclusion of five size dummies and eight 1-digit industry dummies produced an estimated coefficient (standard error) on manual union recognition of 0.014 (0.101) and on closed shop/management recommends of -0.296 (0.095); including five size dummies and 48 2-digit industry dummies produced coefficients (standard errors) of -0.090 (0.108) on recognition and -0.226 (0.102) on closed shop/management recommends.

16 There is also a question in WIRS that asks managers about the number of competitors that they face. Three responses are possible: none, few competitors (five or less) and many (more than five). We also defined a dummy variable indicating few or no competitors and broke down the union effects by this variable. These results were inferior in that they did not add any explanatory power to the model. The estimated coefficients and standard errors were as follows: union recognition, few competitors -0.026 (0.121); union recognition, many competitors -0.055 (0.117); closed shop/management recommends, few competitors -0.285 (0.129); closed shop/management recommends, many competitors -0.323 (0.139).

17 The diagnostic tests indicate evidence of heteroskedasticity and threshold heterogeneity in the models for 1984, or more generally of model misspecification. One such important misspecification is the combining with other categories of the pre-entry closed shops, which Machin and Stewart (1990) find to stand apart. This contrasts with the findings for 1990.

18 Machin and Stewart (1990) focused on the manufacturing sector because certain variables included in the specification were only available for manufacturing. If one allows the two effects in column 2 to differ between the manufacturing and non-manufacturing sectors, one cannot reject column 2 (the likelihood ratio test gives a [[Chi].sup.2](2) statistic of 1.96).


We would like to thank EMRU for financial assistance, Mike Townsend, Pauline Crichton, Dave Wilkinson, and Steve Woodland for research assistance, two anonymous referees, Richard Blundell, Alan Carruth, Richard Disney, Neil Millward, Andrew Oswald, and participants in a Centre for Economic Performance seminar for helpful comments on an earlier draft.


BELMAN, D. (1992). 'Unions, the quality of labour relations and firm performance', in L. Mishel and P. Voos (eds), Unions and Economic Competitiveness, M. E. Sharpe Inc., New York.

BLANCHFLOWER, D. and OSWALD, A. (1988). 'Profit related pay: prose discovered', Economic Journal, 98, 720-30.

CABLE, J. and MACHIN, S. (1991). 'The relationship between union wage and profitability effects', Economics Letters, 37, 315-21.

CHESHER, A. and IRISH, M. (1987). 'Residual analysis in the grouped and censored normal linear model', Journal of Economics, 34, 33-61.

CONYON, M. and MACHIN, S. (1991). 'Profit determination in UK manufacturing', Journal of Industrial Economics, 39, 369-82.

DISNEY, R., GOSLING, A., and MACHIN, S. (1995). 'British unions in decline: An examination of the 1980s fall in trade union recognition', Industrial and Labor Relations Review, 48, 403-19.

FREEMAN, R. (1983). 'Unionism, price-cost margins and the return to capital', Working Paper No. 1164, National Bureau of Economic Research.

GOURIEROUX, C., MONTFORT, A., RENAULT, E., and TROGNON, A. (1987). 'Generalised residuals', Journal of Econometrics, 34, 5-32.

GREGG, P., MACHIN, S., and METCALF, D. (1993). 'Signals and cycles: Productivity growth and changes in union status on UK companies, 1984-89, Economic Journal, 103, 894-907.

HASKEL, J. (1993). 'Why has manufacturing profitability risen over the 1980s?', Empirica, 20, 51-67.

HASKEL, J. and MARTIN, C. (1994). 'Is UK manufacturing leaner and fitter?' Discussion Paper No. 309, Queen Mary and Westfield College, London.

KARIER, T. (1985). 'Unions and monopoly profits', Review of Economics and Statistic, 62, 34-42.

MACHIN, S. (1991). 'Unions and the capture of economic rents: an investigation using British firm-level data', International Journal of Industrial Organization, 9, 261-74.

MACHIN, S. (1995). 'Plant closures and unionization in British establishments', British Journal of Industrial Relations, 33, 55-68.

MACHIN, S. and STEWART, M. (1990). 'Unions and the financial performance of British private sector establishments', Journal of Applied Econometrics, 5, 327-50.

MACHIN, S. and VAN REENEN, J. (1993). 'Profit margins and the business cycle: Evidence from UK manufacturing firms', Journal of Industrial Economics, 41, 29-50.

MACHIN, S. and WADHWANI, S. (1991). 'The effect of unions on organizational change and employment', Economic Journal, 101, 835-54.

MENEZES-FILHO, N. (1994). 'Unions and profitability over the 80s: Some evidence on union-firm bargaining in the UK', Discussion Paper No. 94-17, University College London.

METCALF, D. (1994). 'The transformation of British industrial relations', in R. Barrell (ed.), The UK Labour Market, Cambridge University Press, Cambridge.

MILLWARD, N. (1993). 'Uses of the Workplace Industrial Relations Surveys by British labour economists', Discussion Paper No. 146, Centre for Economic Performance, LSE.

MILLWARD, N. (1994). The New Industrial Relations?, PSI Publishing, London.

MILLWARD, N., STEVENS, M., SMART, D., and HAWES, W. R. (1992). Workplace Industrial Relations in Transition, Dartmouth Publishing, Aldershot.

NICKELL, S., WADHWANI, S., and WALL, M. (1992). 'Productivity growth in UK companies, 1975-86', European Economic Review, 36, 1055-91.

STEWART, M. (1987). 'Collective bargaining arrangements, closed shops and relative pay', Economic Journal, 97, 140-56.

STEWART, M. (1990). 'Union wage differentials, product market influences and the division of rents', Economic Journal, 100, 1122-37.

STEWART, M. (1991). 'Union wage differentials in the face of changes in the economic and legal environment', Economica, 58, 155-72.

STEWART, M. (1995). 'Union wage differentials in an era of declining unionisation', Oxford Bulletin of Economics and Statistics, 57, 143-66.
COPYRIGHT 1996 Oxford University Press
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1996 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Machin, Stephen; Stewart, Mark
Publication:Oxford Economic Papers
Date:Apr 1, 1996
Previous Article:Individual rights and social evaluation: a conceptual framework.
Next Article:Are Nash bargaining wage agreements unique? An investigation into bargaining sets for firm-union negotiations.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters