Robust crash modification factor estimation with case-control method

Quantitative accident analysis methods are being used alongside qualitative ones to enhance transportation safety. One key measure to assess safety improvement actions is the Crash Modification Factor (CMF). CMFs are estimated using various methods, including case-control. This method defines two groups from data: the case group, which includes instances of crashes, and the control group, which includes instances without any reported crashes. In the case-control method, a sample is drawn from the control group to match the number of observations within the case. Consequently, the estimates from the case-control method are based on a single estimation of the logit binary model and do not utilize all the available data. To address this limitation, this research's main objective is to add robustness to estimated CMF by resampling from the control group and comparing the results with the standard, simpler methods. Additionally, we will evaluate the impact of U-turns on road safety, considering their placement and geometric characteristics.

In this research, factors other than the presence of a U-turn are referred to as intervening factors, as we aim to estimate the CMF of midsection U-turn elimination. We resampled from the control group for one thousand iterations, creating a distribution for the CMF and providing a robust estimate. Some intervening factors, such as traffic volume, are continuous; therefore, we used two different discretization methods in addition to keeping traffic volume as a continuous variable in the descriptive model.

We show that the CMF distribution is skewed and has a wide range, which makes it inaccurate to use a single value in safety analysis instead of considering the entire distribution. Furthermore, we verify the effect of stratifying algorithms based on statistical inference of robust CMFs. We also show that CMF values derived from different sets of intervening variables result in statistically significant differences, with more than half of comparison pairs showing significantly variation. The results also show that different stratification methods lead to significantly different robust CMF values.

The robust case-control, i.e. the method used in this research, is recommended for use by safety analysts. Generating a distribution of CMF values allows for a statistical comparison of obtained CMF values with the value of one, thereby indicating whether a safety action is ineffective.