A general Bayesian approach to job search models: Methodology and applications

This dissertation outlines a method for analyzing a broad range of job search models using Bayesian techniques and uses those techniques to compare various permutations of the basic job search model. The method that is outlined is a Gibbs sampling algorithm with a multivariate rejection sampler built into each iteration. The algorithm allows investigators to sample from the posterior distribution of a wide range of job search models. With the ability to sample from a wide range of models made possible by the sampling algorithm, the various job search models can be compared to determine which is most likely to have been the model that generated a given data set. This is done using Bayes factors. In this dissertation it is shown that, for the models examined, job search models in which optimality is not imposed are generally preferred to those for which it is imposed. A new type of optimality that approximates the traditional definition is shown to be preferred in certain situations. In addition, it was found that the simplest job search model is often preferred to models that are only slightly more complex. All of these results, however, were found to depend upon the simulated labor market conditions from which the data came. In addition, the results are sensitive to the choice of prior distribution and the definition of optimality that is used.