Modeling Semi-continuous Response by Proportional Odds Models: Investigating the Effect of Job Training on Future Income

Abstract

Continuous outcome with numerous zero values is often known as semi-continuous response. Analyzing such data by using typical continuous distribution models may sometimes be misleading. With the aim of investigating the effect of job training program on the future earnings, we run into a response variable (future earnings) which is zero-inflated (semi-continuous). We convert the semi-continuous response into ordinal data by appropriately categorizing it. We then apply proportional odds models to study the association between job training and future income. We also evaluate the effect of job training on the difference in earnings before and after the training program. The association between pre- and post- training earnings has also been examined. It is found that job training has statistically significant impact on the post training earnings as well as on the change in earnings.

Keywords: Zero-inflated Variables, Proportional Odds Models, Job Traning