It may have been useful to determine if human type factors provided better explication of accidents under those conditions. The researchers did not state that type of sample design employed. The sampling in this work is problematic as observed in the survey stage as the majority of the population did not return the questionnaire. While it was not stated the interview process produced high levels of non-response, and hence non-response error may introduce systematic errors, that may impact on later choices. For the data collection from the intersections again the sample design employed was not mentioned. The reader is informed that 274 intersections are chosen from 33 counties, there is no information on the population size or how the sample is procured. So that a proper critique of the sample designs appropriateness is difficult. However based on the researchers comments the sample size appeared inadequate for the research. The researchers noted that they were unable to assess the impact of type and location of lighting and this may have been because of inadequate sample size (Neal et al., 2008, p.39). Another weakness centered on critical assumptions that were made for which the researchers provided no empirical evidence. They decided that 150 feet was the cut off point for accidents no rationale was provided for that determination. Secondly even though they noted that for many of the night time accidents they were uncertain as to the cause but assumed that lighting was involved. It would seem prudent to provide greater empirical support for critical decisions in the study.

Data Analysis

Bayesian statistical analysis was employed to meet the research objectives. Several models were developed based on the variables of concern and the models tested using a hierarchical Bayesian modeling utilizing a Poisson distribution. The initial testing of the two models, one for day and the other for night, did not produce statistically significant results. After an adjustment of the models by the removal of some variables, and the reclassification of lighting as present or absent, they were able to determine that intersections with no lighting are 2.01 times more likely to have accidents than those with lighting. The researchers were unable to successfully assess the impact of other low cost measures....

This is not surprising because as Gelman (2008) points out that "Bayesian statistics is about making probability statements, frequentist statistics is about evaluating probability statements" (p.2). While the decision to use Bayesian analysis appears to have been a reasonable choice it limited the tools available to the researchers. The researchers may have found greater fecundity using a frequentist analysis of the data. It is possible that ANOVA and multiple regression could have been employed to great success in this study (Wortman et. al., 1972).
Limited and justifiable conclusions

The researcher should be commended for their fidelity to the data in this study. The conclusions were congruent with the results of the data analysis. They concluded that the use of lighting may be effective in high crash areas. They were unable to make any conclusive statement on the use of other low cost non-lighting measures. Thus the study did not allow the researcher to provide practical guidance to the state on the removal or installation of lighting.