Union membership and coverage: a study using the nested multinomial logit model.
This article seeks to examine the link between union membership and union recognition. It considers if an individual's decision of whether to join a trade union is conditioned by whether or not there is a union present which is recognized by the management for the purposes of bargaining. This takes up Disney's (1990) criticism of micro-econometric analyses of union membership at the level of the individual which adopt a single equation specification (see, for example, Bain and Elias, 1985; Booth, 1986; Cregan and Johnston, 1990; Guest and Dewe, 1988). He argues that the determinants of union membership should "be analysed in two stages. At the first stage, there are certain factors which determine whether there is a recognized union or staff association within the establishment for the individual to join. Second, there is the question of whether an individual chooses to join a union which is recognised. This can be termed the 'determinants of individual membership', conditioned on coverage" (Disney, 1990, p. 171). The solution to this problem suggested by Disney (1990), and the approach adopted in the articles of both Green (1990) and Wright (1994), is to use a bivariate probit model in order that union membership and coverage can be modelled jointly. Both Green (1990) and Wright (1994) find that unconditioned estimates appear to be considerably biased in relation to the conditioned model. While both these articles provide coherent and well-defined explanations of union recognition and individual union membership, they do not explicitly model the utility maximizing response of the individuals concerned. This article examines the link between coverage and individual union membership using a different econometric framework - the nested multinomial logit model (NMNL).
Modelling individual decisions
The nested multinomial logit model
The dominant method of analysing the determinants of individual union membership is the univariate probit model. More recently, the bivariate probit model has been used to model whether an individual is in a covered institution and whether the individual is a member of a union jointly by allowing the error terms in the membership and coverage equations to have a bivariate normal distribution. Having obtained the joint density of individual membership and coverage it is then possible to address questions such as "what is the probability of an individual joining a union conditional on there being a union which represents them for the purposes of bargaining?". Such an approach does not, however, explicitly ground the behaviour of the individuals concerned within a utility maximizing framework. The restrictions that the bivariate probit model implies on the range of allowable activity therefore remains uncertain. A potential advantage of logit models, as opposed to probit models, is that a large literature has built up relating to how these formulations may be derived from individual utility maximization, stemming from the work of McFadden (1974). Similar models have been developed for univariate probit models but not, as yet, for bivariate probit models (see Wright, 1994, chapter 7, for a discussion of this possibility). It is argued that this grounding makes them somewhat less arbitrary than the bivariate probit model and therefore more satisfactory from a theoretical point of view.
Application to modelling union membership
When the NMNL model is applied to the question of union membership it is necessary to specify the source of the utility which an individual derives from membership. Models of union membership fall into two broad classes: incentive private goods models and social custom models. Both are attempts to explain why workers still choose to pay subscription fees and join trade unions even though they are not usually excluded from the fruits of collective action if they are non-members. Incentive private good models argue that, in addition to the public goods of wages and work conditions, the union also provides a private good for which the individual is willing to pay. This might be a reduction in the risk of unemployment or a reduction in the risk of arbitrary dismissal by management (Blanchflower et al., 1989; Booth and Chatterji, 1989; Denny, 1988).
Social custom models of union membership (Booth, 1985; Naylor, 1989; Naylor and Cripps, 1993) argue that the union provides a non-material private good to its members in terms of belonging to a collective to which their colleagues also belong. As such social custom models are, by their very nature, concerned with the issue of what determines the density of union membership within an establishment, since they argue that an individual's decision cannot be considered in isolation from the decisions of other workers in the plant.
Other theoretical models concerned with the determinants of density argue that membership is likely to be higher: first where the potential for wage gains is higher (Stewart, 1983) either due to imperfect product market conditions or due to the existence of quasi-rents arising from sunk cost technology or legal licensing agreements (Connolly et al., 1989); second, where the cost of organizing collective action is low (Crouch, 1982); and finally, where management opposition is low (Naylor and Rauum, 1993).
The nested logit results
The dataset which is used in this study is the 1986 Social Change and Economic Life Initiative (SCELI). This is a randomly selected dataset of 6,110 individuals from six travel-to-work areas in the UK (namely Aberdeen, Coventry, Kirkcaldy, Northampton, Rochdale and Swindon) with the sample stratified on residential location within each labour market. From this dataset interviews were conducted with the employers of 2,005 individuals. It is this subset which is used in the analysis.
The choice of explanatory variables is dependent on two main criteria: first, variables were selected if there were strong theoretical priors for their inclusion; second, variables were included if they had appeared in previous econometric work. So as to be comparable with most earlier unionization studies, this subset of variables is further divided into blue and white collar workers.
It is noted that some theoretical models are easier to test than others. There is little direct data concerning union provision of private goods and so this theory must be tested indirectly. This can be done since it has been argued that the need to provide private benefits varies with both workplace and job characteristics and these should therefore prove to be significant in any membership regression. The main problem with such an indirect method of testing a theory is that many of the other theoretical models of union membership would predict a similar response. For example, it is a tenable argument that both the ease of organization and the costs of organizing might be related to job or workplace characteristics such as the percentage of part-time workers or the presence of casual staff. Some authors have also argued that the nature of the job influences the strength of social customs (e.g. Kerr and Siegal (1954) for the mining industry).
Another problem relating to the direct testing of the private-good theory relates to the inclusion of the wage mark-up, which several previous studies have included. Since the level of wages is likely to be a function of whether or not the union is recognized for the purposes of bargaining and/or the level of union density, a potential endogeneity problem arises. Since there are very few theoretical priors to act as guides with regard to exclusion restrictions (and so they will essentially be arbitrary), reduced form relationships are estimated.
The variables which are used to judge the importance of social customs in determining the likelihood of union membership are responses to the questions "how favourable is your husband/wife/partner to trade unions?" and "how favourable was your father to trade unions at about the time when you first got a job?" It is hoped that these variables capture the social pressure on an individual to join a trade union to some extent, though it is recognized that subjective answers of this kind may be self-justifying and therefore possibly endogenous. Area dummies also serve to pick up differing social traditions between regions, though it is recognized that they may also be picking up a variety of other area specific effects, such as the state of the local labour market.
Workers were also asked their opinion of their employer's attitude to trade unions. This variable may be seen as being either indicative of the social pressures facing an individual or a measure of employer resistance to trade unions. As a more direct measure of employer attitude, the responses that managers gave to a the question "what would you say was your firm's approach towards trade union membership?" were also included. As an additional measure of employer resistance to trade unions, a dummy was included if the employer provided an alternative employee organization for its workers, as this may be viewed as an attempt by the employer to pre-empt the necessity of a union organized by the workers themselves.
Controls for recent workplace changes were also included in order to take account of the view that unions may be acting as "voice" mechanisms for workers' grievances, and that these grievances are likely to be higher during periods in which workplace practices are undergoing rapid change (Freeman and Medoff, 1984). Finally, variables were also included to control for the personal characteristics of the workers.
White collar workers
All of the selected variables from the SCELI dataset were initially allowed to influence both whether or not the establishment is covered and also the workers' join/not-join decision. The model was then progressively simplified as variables which proved to be insignificant were eliminated from the specification. Three sets of results are presented. The first model shows the "parsimonious equation". This is the nested logit formulation in which insignificant variables have been progressively eliminated from all of the constituent equations. The second, on the other hand, has a common subset of variables for both the "join when not covered" and the "join when covered" equations, so that a strict comparison across the subgroups may be made. The third model then extends the common sub-group of variables to the coverage equation. Looking at the results in detail, white collar workers are more likely to be covered in publicly owned firms, with the area in which the firm is located also playing an important role. The attitude of management to unions also plays an important part in determining whether the plant is covered, as does the worker's perception of the employer's attitude. Finally, workers on temporary contracts are less likely to be in covered establishments.
Looking at the "join when covered" results, membership is likely to be lower in plants which operate merit-related pay schemes, a result which is perhaps not surprising given their non-collective nature. It is also interesting to note that recent workplace changes fail to show up in the membership equation which, taken together with the fact that workers employed on temporary contracts are less likely to join a union, suggests that union membership may be a longer run decision. These results concur with the results obtained in Wright (1994) using a bivariate probit specification.
A worker's perception of the employer's attitude to unions is also found to have a strongly positive influence on the likelihood of membership as well as coverage, with the strength of this effect declining monotonically as the employer becomes less favourable. Other measures of the strength of social custom, reflecting the individual's partner's and father's attitude to unions, prove to be insignificant however.
Personal characteristics also influence individual choice, with both increased age and schooling making a worker less likely to join a union (though with schooling these effects are diminishing in their influence). Divorced individuals, on the other hand, are more likely to be in unions, as are individuals with more work experience. It is also noted that the presence of stewards strongly increases the likelihood of union membership, as was found to be the case previously in bivariate probit specifications (Wright, 1994).
The hypothesis that the NMNL model may be simplified to the multinomial logit model is rejected. This lends support to the view that the union membership follows the type of decision structure implied by the nested logit model.
Blue collar workers
Looking at the results for coverage it is found that, as with white collar workers, firm and job characteristics are important in determining the likelihood of coverage (size of firm, area, operation of shift workers, use of homeworkers). Interestingly, however, whether or not the firm is in the public sector appears to have no independent influence. It is again found that if the management encourages unions it is more likely to be covered, an effect which can again be distinguished from the worker's perception of the management's attitude.
For blue collar workers, if the firm provides an alternative voice mechanism to the union then the firm is less likely to recognize the union - a result which might be expected but which was not found for either blue or white collar workers in either the univariate or bivariate probit specifications used by Wright (1994). More perplexing is the fact that personal characteristics show up in the coverage equation. Those living with their parents or their partner's parents are more likely to be in covered establishments, an effect which is enhanced if the individual's father is favourable to unions. Quite why this might be is uncertain, though it is suggested that individuals are selecting, or being selected into, the covered establishments according to their personal attitudes and/or situations. It may be for instance that, prior to the decision process being modelled in this article, workers chose whether or not to work in a heavily unionized sector. Such assumptions are common in the US context (Grossman, 1983), though they have also been applied to the UK labour market (Booth, 1984).
With regard to membership, firm and workplace conditions again have an important influence on the propensity to join (area, type of work contract, employment of homeworkers and size), as does the management's attitude to the union and the employee's perceptions of it. A general satisfaction with the job was also found to make a worker less likely to join a union. Another interesting finding is that, as was the case with white collar workers, more experienced workers are more likely to join the union.
It is interesting to note that, in neither the white nor the blue collar membership equations, does the sex of the individual affect the probability of their membership. It is surmised that this may be a result of controlling for the presence of a recognized trade union since those articles which have introduced similar controls also find no sex effect (Green, 1990; Wright, 1994). On the other hand, articles based on single equation techniques generally find the sex of the individual to be statistically significant. The implication is that females, given the opportunity to join a union, have the same propensity to membership as males. As was the case for white collar workers, the hypothesis that the nested decision structure implied by the NMNL can be simplified to a multinomial logit structure is rejected.
Summary of results and suggestions for future work
This article adopted a nested multinomial logit approach to model the union membership decision. The main advantage of this model, as opposed to the bivariate probit model, is that it may be derived explicitly from a random utility framework and so in this sense it is less arbitrary. The results obtained using this framework indicate that it is important to model the supply of unionism, and models based on univariate specifications will yield biased results. The detailed results of the estimations offer support for both the private goods model and the social custom model, and suggest that the influences on an individual are a combination of these two effects. However, as was discussed earlier, there are considerable difficulties involved in disentangling these effects. The idea that unions provide a "voice" for discontent about current working practices or recent workplace changes received little support, and it seemed that in general union membership was a longer run decision. A particularly strong feature of the data, both in determining coverage and the probability of individual membership, was the importance of the employer's attitude to unions, and the workers perception of this attitude. Future work may seek to model the determinants of this attitude more explicitly.
Bain, G.S. and Elias, P. (1985), "Trade union membership in Great Britain: an individual level analysis", British Journal of Industrial Relations, Vol. 23, pp. 71-92.
Blanchflower, D., Crouchley, Estrin and Oswald (1989), "Unemployment and the demand for unions", Centre for Labour Economics, London School of Economics, London.
Booth, A. (1984), "A public choice model of trade union behaviour", The Economic Journal.
Booth, A. (1985), "The free-rider problem and a social custom model of union memberships", Quarterly Journal of Economics, Vol. 100, pp 253-61.
Booth, A. (1986), "Estimating the probability of trade union membership: a study of men and women in Britain", Economica, Vol. 53, pp. 41-61.
Booth, A. and Chatterji, M. (1989), "A theory of the open shop trade union", discussion paper 319, University of Essex.
Connolly, R.A., Hirsch, B.T. and Hirschey, M. (1989), "Union rent seeking, intangible capital and the market value of the firm", Review of Economics and Statistics, Vol. 68, pp. 567-77.
Cregan, C. and Johnston, S. (1990), "An industrial relations approach to the free-rider problem: young people and trade union membership in the UK", British Journal of Industrial Relations, Vol. 28, pp. 84-104
Crouch C. (1982), Trade Unions, the Logic of Collective Action, Fontana, London.
Denny, K. (1988), "Modelling trade union membership: open shops and private goods", Nuffield discussion paper 36.
Disney, R. (1990), "Explanations of the decline in trade union density in Britain: an appraisal", British Journal of Industrial Relations, Vol. 28, pp. 165-77.
Freeman and Medoff (1984), What Do Unions Do?, Basic Books, New York, NY.
Green, F. (1990), "Trade union availability and trade union membership in Britain", Manchester School, Vol. 53, pp. 378-94.
Grossman, G. (1983), "Union wages, seniority and unemployment", American Economic Review, Vol. 73, pp. 277-90.
Guest, D.E. and Dewe, P. (1988), "Why do workers belong to a trade union? a social psychological study in the UK electronics industry", British Journal of Industrial Relations, Vol. 25, pp. 178-201.
Kerr, C. and Siegel, A. (1954), "The inter-industry propensity to strike", in Kornhauser, A., Durbin, R. and Ross, A.M. (Eds), Industrial Conflict, McGraw-Hill, New York, NY.
McFadden, D. (1974), "Conditional logit analysis of qualitative choice of behaviour", in Zarembka, P. (Ed.), Frontiers in Econometrics, Academic Press, New York, NY, pp. 105-42.
Naylor, R.A. (1989), "Strikes, free riders and social customs", Quarterly Journal of Economics, Vol. 104, pp. 771-85.
Naylor, R.A. and Cripps, M. (1993), "An economic theory of the open shop trade union", European Economic Review, Vol. 37, pp. 1599-1620.
Naylor, R. and Rauum, O. (1993), "The open shop, wages and management opposition", Warwick Economic Research Papers, 332.
Stewart, M.B. (1983), "Relative earnings and individual union membership in the United Kingdom", Economica, Vol. 50, pp. 111-25.
Wright P.W. (1994), "Unions, unionisation and imperfectly competitive labour markets - a theoretical and empirical analysis", unpublished PhD, University of Warwick, Coventry.
Peter Wright, University of Nottingham, Nottingham, UK
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||EMRU Conference Papers|
|Publication:||International Journal of Manpower|
|Date:||Feb 1, 1995|
|Previous Article:||Skilled and unskilled employment in UK manufacturing over the 1980s.|
|Next Article:||Is pay discrimination against young women a thing of the past? A tale of two cohorts.|