01658998

Component

The Poisson-gamma (PG) and Poisson-lognormal (PLN) distributions are among the most popular sampling distributions used by safety practitioners and analysts to model crash data. Several studies have shown that the PLN offers a better alternative compared to the PG when data are skewed while the PG is a more reliable option, otherwise. However, it is not explicitly clear when the analyst needs to shift from the PG to the PLN – or vice versa, or what characteristics of data should be observed a priori when deciding between these two alternative distributions. In addition, in most research studies, the comparison between these two distributions or models and the subsequent Model Selection decisions has usually been accomplished using the Goodness of Fit (GoF) statistics or statistical tests. Such metrics rarely give any intuitions into why a specific distribution or model is preferred over another (addressing the classical issue of Goodness-of-Logic). This paper ponders into these topics by (1) designing characteristics based heuristics to select a logical distribution between the PG and PLN, (2) prioritizing the most important characteristics of the data under analysis to make a Model Selection decision between the PG and the PLN. The proposed heuristics allows the analyst to select a logical distribution between the PG and PLN, without any post-modeling efforts.

This paper was sponsored by TRB committee ABJ80 Standing Committee on Statistical Methods.

18-03035

English

5p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; References (7)

Crash data; Data analysis; Data quality; Mathematical models; Poisson distributions

Data and Information Technology; Highways; Safety and Human Factors

Transportation Research Board Annual Meeting 2018 Paper #18-03035

TRIS, TRB, ATRI

Jan 8 2018 10:43AM