The British Household Panel Study (BHPS) is a panel of about 13,000 individuals based on an original stratified random sample of 5,000 households throughout Britain. A joint CEPR workshop with the ESRC Research Centre on Micro-Social Change at the University of Essex on 24 November addressed labour market issues that can be investigated using this data set. `Labour Market Issues in the 1990s: Evidence from the BHPS', was organized by Alison Booth (University of Essex and CEPR) and formed part of the activities of CEPR's Labour Market Imperfections Group, which is supported by a grant from the former Employment Department. Additional financial support was provided by this ESRC Research Centre. Booth began the workshop by explaining the nature, purpose and possible uses of the BHPS. She noted that having started in 1991, three waves of the annual study were now available for analysis. The fourth wave, data from interviews conducted in 1994, would soon be available; the fifth wave was being conducted; and questions for the sixth wave were being tested and finalized.
The first paper, `Constraints on Desired Hours, Trade Unions and the Length of the Working Week for British Men' by Mark Stewart and Joanna Swaffield (University of Warwick), was a cross-sectional analysis of the constraints on the desired hours of work for male manual workers, using information on hours preferences from the 1991 BHPS. The motivation for the paper came from the findings that, on average, British men work more hours per week than their European counterparts, and that over a third of male manual workers in Britain work more than 46 hours a week, compared with just 5% elsewhere in the European Union. Using an econometric model, the authors estimated that manual men would prefer to work an average of 4.3 hours less than they actually do. Using a friction model, they found that the likelihood of an individual working more hours than desired is an increasing function of the unemployment rate faced by that individual. The commonly found age profile in actual hours is not apparent in desired hours and is the result of constraints imposed by the firm. Furthermore, higher minimum hours requirements are imposed on less well paid workers. Lastly, trade union status has no significant effect on preferred hours, minimum hours constraints nor the probability of working more hours than desired.
Steve Machin (University College, London) noted that the inequality in hours had increased over the last ten years. Amanda Gosling (Institute of Fiscal Studies) suggested that the unemployment variable used may also pick up other regional factors, adding that recent technological change may have affected hours worked. Wim Groot (Universiteit van Leiden) asked whether effort should be considered since reported actual hours may not be actual hours of effort; at the same time, he acknowledged that manual workers' effort can be monitored more easily. Barry McCormick (University of Southampton) argued that unemployment may affect individuals' hours of work in that they work as much as possible when work is available; such intertemporal substitution is not considered in the paper. Simon Burgess (University of Bristol and CEPR) agreed that a more dynamic framework could be used to analyse this problem using the BHPS, and commented that if workers were being forced to earn too much (by working longer hours than desired), there would be implications for consumption, investment and savings. McCormick wondered whether a second job, moonlighting or illegal work affected the interpretation by interviewees of the question used on desired hours. Andrew Oswald (University of Warwick) also questioned the desired hours question in the BHPS, circulating a handout comparing it with that used in the British Social Attitudes Survey (BSAS), which generates very different conclusions. Stewart replied by suggesting that the BSAS contains leading questions; he produced the European Community Labour Market Survey 1989, which has similar results to the BHPS.
In a paper written with Andrew Clark, `Satisfaction and Comparison Income', Andrew Oswald attempted to derive a utility measure from the job satisfaction questions in the 1991 BHPS, testing the hypothesis that utility depends on income relative to a reference or `comparison' level. Using data on 5,195 workers in the UK, he presented two main findings. First, the reported job satisfaction levels of workers are negatively related to their imputed reference wage rates. Second, job satisfaction declines sharply with educational achievement after allowing for income. Oswald hoped that the paper may be a beginning for an economics of job satisfaction, which would bring together elements of the public finance and labour economics literature.
The finding of a negative relationship between education and satisfaction led to much debate. Booth argued that wage profiles may affect people's expectations and the less well educated have flatter profiles, while Jonathan Gershuny (University of Essex) argued that this may result from `cognitive dissonance'. A link between the imputed reference income and education was a possible problem. Jonathan Haskel (Queen Mary and Westfield College, London, and CEPR) wondered whether this income measure may be a proxy for education, and Groot argued that if the fixed effects had not been fully accounted for, income may pick up some of the effects of education. McCormick argued that people place different values on leisure, but still wondered why the education and age effects were so strong. Tim Butcher (University of Essex) questioned the finding that `ignorance is bliss', and wondered whether the author supported the policy implications of this. John Ermisch (University of Essex and CEPR) argued that there may be constraints on employment so that job satisfaction cannot be seen as a measure of overall well-being. Andrew Henley (University of Kent) added that this would depend on whether the utility function was separable between various aspects of an individual's life. Gershuny suggested that instead of using the overall job satisfaction question, an index could be compiled from the questions on the various aspects of an individual's job.
Wim Groot presented a paper written with Hessel Oosterbeek, `The Incidence and Wage Effects of Incentive Pay', which analysed incentive pay and fixed salary methods of payment using a switching regression model. Firm and worker characteristics from the 1991 BHPS were used to determine the payment scheme; the wage differential between the two methods was decomposed into risk compensation, sorting and incentive effects; and productivity and salary profiles were deduced for each payment scheme. Groot found that higher average wages for workers on incentive schemes are completely attributable to effort effects; his results do not support sorting and risk compensation effects. He concluded that shirking-related and efficiency theories of the wage profile are not consistent with these results.
Wiji Arulampalam (University of Warwick) questioned the use of the wage variable used, since it included overtime, bonus, commission, etc.. Burgess questioned the interpretation and derivation of the risk premium measure as it included all measurement and misspecification errors. Jeff Frank (Royal Holloway College, London, and CEPR) added that the sorting effect was also in the error term as the characteristics were observable. Machin expressed concern about identification in the switching regression as many of the variables were used in both.
In `Semi-Markov and Markov Labour Histories', written with Ken Burdett, Mark Taylor (University of Essex) used information from the first three waves of the BHPS plus a 1992 lifetime employment history questionnaire, which yields information on the labour market activity of all those who had been of working age since 1946. He showed how Markov models can be used to identify and model the determinants of labour market flows between four states: employment, unemployment, non-participation and self-employment. The labour market transitions were estimated using a proportional hazard function in a competing risks framework. The effects of certain fixed individual characteristics were estimated separately by gender, and the estimated hazards were used to compute the steady state proportions. Due to the nature of the model, only fixed characteristics that are determined before the individual enters the labour market were used: highest qualification, ethnicity, father's activity state when the individual is 14, an innate ability index, and age. The results confirmed expectations: two-thirds of working age men but just under a half of women would be expected to be employed; 15% of men but only 5% of women would work for themselves; the unemployment rate would be 2.4% for men and 0.6% for women; and women were more likely to be non-participants – 45% compared with just 18% of men. The low estimate for unemployment was explained by poor recall of short periods in the past.
Machin commented that cyclical effects had not been taken into account although a positive simple time trend for all transitions, not reported, had been found. Several participants questioned the accuracy of recall for all states and transitions. Stewart questioned the use of a human capital index as returns to education and experience had changed considerably over time. He further commented that if the true hazard is downward, then if recall is biased (only longer spells are remembered), fitting the model presented will lead to the flat hazard found between moves from unemployment into employment. Alex Bryson (Policy Studies Institute) proposed the use of time-varying parameters though this would not be possible in the current model structure.
In `Employer Generosity and Free-riding: Union Wage Effects for Members and Non-members', Andrew Hildreth (University of Essex) used the first three BHPS waves to present longitudinal and cross-sectional estimates of the covered union member and covered non-member wage effects. He found that cross-sectional estimates yield a 10% premium for union members over other workers. He argued that these estimates are biased downwards as they do not allow for endogeneity, measurement error or fixed effects. By using longitudinal estimates, which allow for these problems, he deduced that the true union wage differential lies between 25–30%, but he could find no such robust estimate for covered non-members. He took this as evidence of employer generosity rather than free riding.
Henley suggested that the wage differentials to covered members and non-members over uncovered non-members should be tested to see if they are equal. Booth questioned the use of the instruments used for covered members and non-members (union dues and employer pension schemes respectively), and proposed a selection test on the probability of free riding instead. Arulampalam suggested using predicted membership values as instruments for the membership and non-membership effects. Gosling argued that instrumenting two periods back picks up membership and non-membership but not coverage, and that this will be a major problem if the wage and job change are endogenous. Stewart raised the point that the whole estimation process is driven by very small cell sizes as changes in union status are tiny, thus making fixed effects very difficult to discern. He also noted that coverage is endogenous, adding that rather than employer generosity being the only explanation, there could also be a closed shop effect or a promotion prospects effect.