Electronic Thesis and Dissertation Repository


Doctor of Philosophy




Audra J. Bowlus

2nd Supervisor

Lance J. Lochner

Joint Supervisor


This thesis consists of three related papers. The first paper examines whether informal sector jobs are a source of training for young less-educated workers. Controlling for worker and job characteristics, it is found that in the early years of workers' careers in Mexico, wage growth in the informal sector is higher than in the formal sector. This result is consistent with general human capital investment on-the-job if the informal labor market is more competitive than the formal labor market due to frictions generated by labor regulations. These results motivate a deeper analysis of the informal labor market which is presented in the second paper.

The second paper examines two roles that informal sector jobs play in the early stages of a worker's career: informal jobs may (i) provide the opportunity to accumulate skills, and (ii) act as a screening device that enables employers to learn a worker's ability. This paper develops a matching model of the informal and formal sectors that can accommodate both roles. Implied hazard rates from informal to formal sectors as a function of tenure are shown to differ depending on whether the role of informal sector jobs is human capital accumulation or screening. Using the ENOE, a longitudinal employment survey from Mexico, hazard functions are estimated for less-educated workers. The estimated hazard functions suggest the informal sector plays an important role by screening less-educated workers in the early stages of their careers. The estimation results also imply that employers would only learn the ability of 14% of their workers after one month of employment. This finding suggests that employers' capacity to select workers is limited in government employment programs requiring employers to provide permanent positions to a predetermined fraction of workers after a short period of time.

The duration data used for estimation in the second paper is obtained from the stock of individuals employed in the informal sector at a given point in time. It is known that duration data obtained from a given stock of individuals can fail to observe those with relatively short spells. Accounting for this sample bias requires constructing a conditional likelihood function, which in turn requires knowledge of the exact starting times of each spell. Unfortunately, it is common in duration data to have coarse measures for starting times, complicating the resolution of sampling bias. The third paper investigates several alternatives for overcoming coarseness by imputing interval-censored starting times and performing a Monte Carlo analysis. The results indicate that imputed interval midpoints outperform the alternatives.