September 15, 2016
This thesis comprises three essays. The first two make use of individual-level data on British workers from the British Household Panel Survey to study different aspects of nonstandard employment. The first essay, co-authored with Mark Bryan, presents estimates of the implicit monetary value that workers attach to non-standard work. We employ and compare two alternative methods to measure workers’ willingness to pay for four non-standard working arrangements: flexitime, part-time, night work, and rotating shifts. The first method is based on job-to-job transitions within a job search framework, while the second is based on estimating the determinants of subjective well-being. We find that the results of the two methods differ, and relate them to conceptual differences between utility and subjective wellbeing proposed recently in the happiness literature. The second essay builds on economic theories of consumption and saving choices to investigate whether workers expect temporary work to be a stepping stone towards better jobs, or a source of uncertainty and insecurity. The evidence provided shows that temporary work entails both expected improvements in future earnings, and uncertainty. Households’ consumption and saving choices are used to assess which of these two effects is prevailing, providing an alternative empirical approach to measure the consequences of temporary work for workers’ welfare. The results suggest that a stepping stone effect towards better jobs is present and, more importantly, is perceived by individuals and internalized in their behaviour. Finally, the last essay has a specific focus on econometric methods. A Monte Carlo experiment is used to investigate the extent to with the Poisson RE estimator is likely to produce results similar to ones obtained using the Poisson FE estimator when the random effects assumption is violated. The first order conditions of the two estimators differ by a term that tends to zero when the number of time periods (T), or the variance of the time-constant unobserved heterogeneity (V), tend to infinity. Different data generating processes are employed to understand if this result is likely to apply in common panel data where both characteristics are finite. As expected, the bias of RE estimates decreases with T and V. However, the same does not hold for the estimated coefficient on the time invariant dummy variable embedded in the conditional mean, which remains substantially biased. This raises a note of caution for practitioners.