Tweedie Model for Analyzing Zero-Inflated Continuous Response: An Application to Job Training Data
Nabila Parveen
Department of Statistics, Biostatistics and Informatics, University of Dhaka, Bangladesh
Muhammad Abu Shadeque Mullah
Department of Statistics, Biostatistics and Informatics, University of Dhaka, Bangladesh
Mohammad Ahshanullah *
School of Business and Economics, United International University, Bangladesh
*Author to whom correspondence should be addressed.
Abstract
Continuous data with substantial proportion of zero values arise often in many disciplines. Modeling such zero-inflated data is always challenging. We use Tweedie model to analyze zero-inflated continuous outcome with a view to evaluate the effect of job training on future earnings, and also on the difference between pre- and post-training earnings. We further assess the effect of pre-training earnings on the post-training earnings. We used data from a job training program in the USA where 445 subjects were followed for three years. Results suggest that job training has statistically significant impact (p-value <0.05) on future earnings, as well as on the change in pre- and post-training earnings. The effect of pre-training earnings on post-training earnings is however not found to be statistically significant. We found the Tweedie model to be particularly suitable for analyzing zero-inflated data to make valid statistical inference.
Keywords: Zero-inflated data, Tweedie model, job training and future earning