Prediction of Bridge Accident Rates

Abstract

A technique is developed to predict bridge accidents. These collisions are complex occurrances influenced by many factors. Previous research found as many as ten items contribute to a structure's potential for collisions. Four years of Texas data (1,000,000 accidents, 29,000 bridges and 100,000 roadway segments) were merged and searched for rural, two‐lane, two‐way bridge accidents. Screening and classification measures were used to purify the data leaving a uniform study sample of 2,849 cases for analysis. The relationship between accidents and predictor variables was investigated by manual, correlation, and regression analyses. The most important variables were: (1) Bridge relative width (bridge minus road); (2) average daily traffic (ADT); and (3) approach roadway width. Regression curves were developed to predict accidents using these factors as independent variables. A probability table was developed to predict accidents. Combinations of approach roadway width and bridge relative width indicate collision rate. Rates may be combined with ADT for particular structures to develop expected accidents per year. The statistical measures for effectiveness were very strong, indicating that the table is a good way to predict bridge accidents.