Assessing economic performance and aviation accidents using zero-inflated and over-dispersed panel data models

This paper explores the link between air transport safety and profitability. Traditionally, the empirical literature has used Poisson regression models to estimate the expected number of accidents given the profitability of the airline. However, there are two major deficiencies in this analysis related to the statistical properties of the data. First, the equi-dispersion assumed in Poisson models hardly holds in the airline data. Second, accidents are rare cases, so the data has an excess of zeros. In this paper, we propose the use of a zero-inflated negative binomial model to deal with these shortcomings. Our results show several interesting facts. On the one hand, they show that airlines with higher levels of profitability are less likely to have an accident. On the other hand, when an accident occurs, there is a higher expected number of accidents in airlines with higher profitability. Finally, the severity of an accident has an inverse relationship with profitability.