Many safety researchers feel that a properly designed before-after study is more likely to provide reliable estimate of the safety effectiveness of a treatment, i.e., a crash modification factor (CMF). When before-after studies are not feasible, cross-sectional studies are usually performed to estimate CMFs. These models are usually assessed using goodness of fit measures that provide insight into how well a model fits the data. However, how well a model fits a data does not provide insight into whether a particular coefficient can be used to reliably estimate the CMF associated with a treatment. Some researchers have used real-world data sets to compare the effectiveness of different statistical methods. However, in a real-world database, the true underlying relationships between crashes and roadway elements are unknown. Hence, simulated data can serve as a testbed to compare the performance of different modeling approaches. This paper (based on research funded by FHWA) will describe the results of an evaluation that was conducted using an artificial realistic dataset (a simulated dataset) to compare the performance of different cross-sectional modeling approaches.