In the past, the two-way left-turn lane (TWLTL) median treatment was frequently used in Florida to inexpensively improve motor traffic flow. A procedure to identify the factors that are influential in the safety experience of TWLTL sections was developed during this research. This procedure would also allow the identification of groups of TWLTL locations that present existing and future safety concerns and need further analysis to determine possible improvements. For this purpose, a three-year crash history database, from 1996 to 1998, including traffic crashes from totally 1688 TWLTL sections all over Florida was used in this research. Statistical models were used to determine the relationship between number of crashes per mile per year and several factors such as traffic volume, access density, posted speed and number of lanes. During the analysis, distribution fitting for the Poisson, Negative Binomial and Lognormal distributions was performed for the crash data. Then, a Negative Binomial regression model was developed to estimate the number of crashes per mile per year for the TWLTL sections. The regression parameters were estimated by using the maximum likelihood method. The goodness-of-fit for the developed model was evaluated based on Pearson's R-square and likelihood ratio index. In regard to the methodology to identify locations that need further study to determine if improvements are necessary, the procedure consisted of several steps. The steps included: plotting crash data to determine the distribution of the actual data for six different groups of locations based on posted speed and number of lanes; distribution fitting of crash data to determine the statistical distribution that better represents the actual data; determining percentile values for the average numbers of crashes from distribution fitting of the original crash database; estimating the average number of crashes per mile per year for sections with the same characteristics using the parameters of the crash predictive model; estimating critical values for the variables considered in the research with the percentile value of the average number of crashes and the curves plotted for number of crashes from the models; calculation of tables of critical annual average daily traffic (AADT) values based on the values of access density, number of lanes and posted speed for each one of the six groups; and the generation of a list of locations identified as critical according to the selected percentile value and the critical AADT based on access density, posted speed, and number of lanes. (Author/publisher)
Samenvatting