In this paper we study the effect of subsidised on-the-job training, classroom training for the unemployed and pure wage subsidies to employers (i.e. without a compulsory training content). We analyse the effect of these programmes on the probability of leaving an employer. The empirical analysis is based on a sample of workers who were hired from the unemployment pool. This paper has several objectives. First, it intends to throw some light on the debate about 'job insecurity'. Second, it aims at improving the available understanding of the functioning of the labour market. Finally, it extends the literature on job tenure by dealing with selection caused by unobserved heterogeneity and by endogenous sampling.
The feeling that jobs have become more insecure or precarious is nowadays widespread in developed countries (see OECD, 1997). Job tenure is clearly one of the main objective indicators related to this feeling. In this context, it is worthwhile to raise the question whether active labour market policies favour job tenure. Dealing with this widespread concern is a first aim of this paper.
Studying the effect of labour market programmes on job turnover can also improve our understanding of the functioning of the labour market. In particular, it can throw some light on the relative importance of human capital and matching theories in explaining job tenure and on the importance of skill mismatch as an explanation of unemployment.
According to human capital theory, the extent to which job tenure is increased crucially depends on the nature of the training investment. If the training is firm-specific, job tenure will increase more than if it is general (in a classroom). As the acquired productivity increase is specific to the firm, the wage will rise relative to the alternative wage. The probability of leaving an employer will therefore fall.(1) In addition, employers will be less likely to lay off workers in whom they have invested in specific skills. If, on the other hand, training is general, the increased productivity is useful to all firms and the worker has less or no incentives to stay on the same job. The same distinction holds mutatis mutandis as far as weakly transferable,(2) and widely transferable training are concerned. If these arguments are correct, one would expect that on-the-job training favours more job tenure than classroom training for the unemployed, the latter being more transferable. Testing this hypothesis is a second objective of the paper.
Matching theories (Spence, 1973; Stiglitz 1975; Jovanovic, 1979a, 1979b, 1984) provide an alternative explanation for the relationship between training and job tenure. According to these theories, training does not enhance the productivity of the worker but merely improves the quality of the match between the worker and the firm. The quality of the match is improved because participation in the training programme signals information about the productivity of the worker that previously was private information. If participation signals that the trainee is more productive than the average employee, training will typically reduce the rate of job turnover. On the other hand, if the selection of trainees favours workers whose characteristics are negatively correlated with productivity (see e.g. Burtless, 1985, or Dubin and Rivers, 1993) and if these characteristics are not readily observable to employers, then participation in the training programme discloses that the worker is of low productivity, increasing therefore the job turnover of trainees.
According to matching theories, the relationship between training and job tenure can therefore be positive or negative. Irrespective of the direction of the relationship, on-the-job training programmes as compared to classroom programmes are likely to intensify the relationship, since obviously more information will be disclosed if it is the employer himself who operates the programme (Barron et al., 1989). A third objective of the paper is to test whether matching theories or human capital theories turn out to be more in accordance with our data.
Until now we have been silent about the potential impact of pure wage subsidies to employers on job turnover. We now turn to a discussion of this effect. It will appear that it can provide evidence on the relative importance of skill mismatch as an explanation of unemployment.
Even if it is widely accepted that employment is enhanced during the period in which the subsidy is granted, one might question whether a temporary subsidy can have a long run impact on employment. On the one hand we have authors, such as Layard et al. (1991), who emphasise the role of genuine duration dependence in any explanation of the evolution of European unemployment exit rates during the last decades. Genuine duration dependence occurs if, for example, an increase in unemployment duration creates some loss of accumulated human capital or generates a discouraged worker effect. In this case, a temporary wage subsidy can induce permanent effects as it inverts the process of depreciation and discouragement. On the other hand, authors such as Sneessens and Shadman-Mehta (1995) argue that the causes of unemployment are more structural to the disadvantage of low skilled workers. In view of these models of unemployment, a temporary subsidy programme with little training content cannot produce permanent effects, as there exists a fundamental mismatch between the characteristics of jobs and the qualifications of the unemployed. It is therefore possible to discriminate between these two theories to the extent that we find a significantly positive effect of a temporary wage subsidy on the length of the job spell after the subsidised period. The available data do unfortunately not disentangle the period during which the worker is subsidised from the after-subsidy spell. Hence, the results provided by this paper will necessarily be tentative.
One of the major problems in estimating the effects of the above-mentioned labour market policies is to correct adequately for the selection bias. The latter arises whenever participants to the labour market programme are non-randomly selected. For, in this case a differential impact of treatments as compared to controls may merely reflect differences in (un)observed characteristics.
Most commonly one tries to control for the selection bias in a non-experimental setting by modelling jointly the participation decision and its impact on the variable of interest, such as earnings or the exit rate out of (un)employment. Recent research (LaLonde, 1986; Fraker and Maynard, 1987) has thrown doubt on the capacity of non-experimental methods to correct for selection bias. Estimates are found to be sensitive to model specification and estimation method. This line of research asserts that the selection bias can only be controlled for if one disposes of data emerging from an experiment, be it controlled or natural (cf. Angrist, 1992; Meyer, 1995).
This paper remains in the tradition of the non-experimental literature. Heckman and Hotz (1989) defend this approach to the extent that adequate specification tests allow to discard the inappropriate models. In this paper we follow this approach and test for the sensitivity of the results to the model specification. However, due to data limitations and to the computational complexity of the optimisation problem, the sensitivity analysis is limited in nature.
Despite the widespread use of these active labour market policies, there exists relatively little micro-econometric evidence on the effects of these measures. Most progress has been made in the evaluation of training programmes. However, most studies concentrate on the effect of training on subsequent earnings or on reemployment probabilities. Only recent studies have considered the effect on the length of the subsequent employment spell as well (cf. e.g. Kaitz, 1979; Ridder, 1986; Card and Sullivan, 1988; Ham and Lalonde, 1991; Gritz; 1993, Bonnal, et al.; 1994, Torp, 1994; Zweimuller and Winter-Ebmer, 1996). One should distinguish employment duration from the length of job tenure, the variable of interest in this paper. To the extent that there are job-to-job transitions without an intervening spell out of employment, the two concepts differ. Lynch (1991) and Elias (1994) estimate the effect of on-the-job training on the probability of job termination, the variable of interest of this paper. Booth and Satchell (1994) do the same in the case of apprenticeships. These authors do not correct for selection bias, however. As far as we know, there is no empirical analysis of the effect of pure wage subsidies on job tenure.(3)
We now turn to a brief discussion of the empirical evidence. We focus on the literature studying job tenure rather than employment duration. It should be emphasised that the literature has not yet adopted an homogeneous vocabulary. In this overview, job-related and private sector (respectively, off-the-job and government sector) training programmes have been assimilated with on-the-job (respectively, classroom) training programmes. Lynch (1991), Elias (1994), …