Enhancing Statistical Methodologies For Highway Safety Research – Impetus From FHWA
CHAPTER 3. FUTURE STEPS AND OTHER CONSIDERATIONS
A number of institutional suggestions, technical approach suggestions, and other considerations were identified during and following the technical experts meeting. On the basis of these suggestions, some future steps are outlined that might foster the effective implementation of the suggestions. The suggestions are categorized below as institutional or technical.
Institutional Suggestions
The following is a set of considerations for institutional initiatives to advance the development of new statistical methodologies that might improve how SPFs and CMFs are developed, by engaging researchers highly qualified in statistical methods, and by fostering high-level research in general:
- FHWA may consider hiring statisticians to work alongside transportation engineers. This may require special funding and longer-term commitment to such a program.
- FHWA may consider an active role and more regular participation in the annual meetings of the ASA and other statistical communities, and in conducting workshops, developing sessions, and making presentations.
- FHWA may consider creating opportunities for sabbatical and internship programs in areas of statistics at the Turner-Fairbank Highway Research Center. This could be an extension of similar programs in other fields.
- FHWA may consider promoting the enhancement of statistics course requirements for civil engineering students to increase the comprehension of statistics by graduates working in road safety analysis. At present, a graduate course in statistics is typically not compulsory.
- Transportation safety data might be made available by facilitating free download to interested statisticians and advertising the availability of these data. Specifically, these data would pertain to the information required to develop SPFs and CMFs. The databases used for the FHWA ELCS project might be tapped for this purpose. Highway Safety Information System data are already readily available and might be used for SPF development research. Machine Learning and Chance magazines were suggested as means to advertise the availability of these data to graduate students and other researchers in statistics.
- Short courses and workshops might be arranged at the ASA and other related annual meetings on current highway safety research practices/products in order to introduce these topics, to discuss the needs for new methodologies, and to introduce data resources and capabilities in sparking creativity and innovation in both researchers and decision makers.
- FHWA may conduct a workshop with the Transportation Research Board (TRB) committees ABJ80 (Statistical Methods) and ANB20 (Safety Data, Analysis and Evaluation) for effectively marketing the need for statistical innovation in highway safety analysis applications. There have been sporadic attempts at putting on such workshops, the most recent being a hands-on workshop before the January 2014 annual meeting conducted by Ezra Hauer on developing SPFs. Consideration may be given to developing a workshop that would be led by statisticians or by both statisticians and transportation engineers highly skilled in statistics.
- FHWA may consider sponsoring a graduate student competition for working with transportation safety data to solve a set of problems related to highway safety methodology development. Graduate students in statistics should be encouraged to participate.
- FHWA may publish articles in Public Roads magazine to communicate the increasing need for statistical expertise in highway safety application to practitioners.
- FHWA, in coordination with ASA, might make presentations and hold workshops for new highway safety research products in selected schools of statistics nationwide. This is to communicate to/with faculty and students and introduce highway safety research methodology needs and sources of data to spark statistical research inspiration.
- Consideration may be given to future and alternative sources of data, e.g., Strategic Highway Research Program data, and data available from mobile devices, in developing new statistical methods.
Technical Approach Suggestions
Below is a summary of specific technical approaches that could be pursued in high-level research aimed at improving CMF and SPF estimation. Specifically, it is suggested that future researchers engage in the following:
- The application of a multivariate hierarchical Bayes modeling approach for estimating SPFs. Among other strengths, this approach allows for borrowing of information across outcome and predictor variables to make stronger inferences.
- The modeling of spatial correlation and the analyzing of locations as a network in estimating SPFs.
- The use of prior information in the estimation of CMFs/SPFs through the hierarchical Full Bayes approach. For example, the prior estimation may pertain to previously estimated SPFs and CMFs of relevance.
- The exploration of meta-regression for developing CMFunctions. It is expected that this approach will be pursued in NCHRP Project 17-63.
- The use of multilevel modeling where treatment effects are heterogeneous and nested. For example, results from individual sites may be nested by county of origin.
- The continued pursuit of the use of surrogate measures estimated from simulation by seeking better simulation models and better models to relate crashes to surrogate measures.
- The exploration of data imputation methods to address missing data instead of discarding those observations. This would pertain both to data for treatment and non-treatment reference/comparison sites.
- The use of propensity scores for comparison or reference sites matching to reduce bias between treated and non-treated groups.
Other Considerations
A number of items for potential DCMF program next steps were identified during and after the technical experts meeting. These are as follows:
- Experts might be convened on a regular (perhaps annual) basis for follow-up and discussion of fresh ideas. A desirable format would be to utilize a combination of "new" experts and some from the most recent meeting.
- Research needs statements for guiding future research in new statistical methodologies might be created and disseminated through appropriate channels, e.g., TRB. Since TRB committees are charged with creating these statements, appropriate committees such as ABJ80 (Statistical Methods) and ANB20 (Safety Data, Analysis and Evaluation) and ANB25 (Highway Safety Performance) could be instrumental in leading this effort.
- Projects might be funded through the DCMF contract to pursue the application and demonstration of identified methodologies using existing databases. Some projects could involve statisticians only and some could be joint with the DCMF project team, other prominent road safety researchers, and statisticians.
- Results of research using new methods resulting from this white paper might be disseminated through conferences such as the TRB and ASA annual meetings. Research in progress could be presented at appropriate TRB committee meetings and the ASA transportation interest group.
- The ASA transportation interest group might be engaged to foster the encouragement of research using the new methods, the presentation of not only the results of this research but also the details of research in progress.