Investigating the Impact of Lack of Motorcycle Annual Average Daily Traffic Data in Crash Modeling and the Estimation of Crash Modification Factors

Chapter 6. Conclusions and Recommendations

Data on traffic volumes are vital to the development of SPFs required for effective implementation of strategies to improve the safety of road networks. Mitigating motorcycle crashes can be especially challenging in this regard because few jurisdictions collect motorcycle traffic volume data systematically. To address this challenge, the project team conducted several analyses to explore (1) how much predictive power for an SPF is lost when motorcycle volumes are unknown and how this lack of information may affect a study of motorcycle countermeasures, and (2) alternative methods for deriving accurate predictions of motorcycle crashes or motorcycle volumes. The research investigated and demonstrated methods and the mathematical models that can be applied by jurisdictions that lack motorcycle volumes when undertaking the development and the evaluation of motorcycle-related safety countermeasures to estimate CMFs.

The project team investigated two groups, or avenues, of methods. The methods for avenue A focused on investigating (1) the difference in predictive performance for motorcycle SPFs calibrated with motorcycle AADT versus total AADT, (2) relating total crash SPFs to motorcycle crash SPFs so jurisdictions without motorcycle volumes could predict motorcycle crashes using total crash SPFs, and (3) methods to predict segment-level motorcycle AADT. The methods for avenue B focused on the differences in CMF estimates when using motorcycle AADT versus total AADT when applying before-after or cross-sectional regression CMF estimation methods.

For developing the avenue A models, data were collected from Florida and Pennsylvania. Both States had a large number of locations with an estimated motorcycle AADT and could provide linkable roadway inventory, traffic, and crash data. The project team also acquired data from Virginia to validate the models developed.

Table 84 summarizes the avenue A methods, with a final column on conclusions from the analysis. In addition to the analyses depicted, an assessment was conducted of how well EB estimates derived from the model type A1 models predict future motorcycle crashes for high-crash locations typically of interest in countermeasure applications that form the basis for future CMF development. The results of that assessment show that the models using total AADT and those using motorcycle AADT perform similarly, although the EB estimates from the total AADT models of the crashes at the high crash sites are marginally closer to the actual future crash counts than those based on the motorcycle AADT models.

The methods applied in avenue B make use of simulated data. Simulating data creates a database with many locations and with assumed relationships between roadway geometry or other countermeasures and motorcycle crashes. This tests the ability to accurately measure this true relationship when motorcycle volumes are and are not used in the process. The fixed relationships affecting motorcycle crashes were determined considering a likely range of values based on existing safety knowledge.

To investigate the impact of the lack of motorcycle AADTs on the estimation of CMFs, two CMF estimation approaches were investigated: model type B1, the EB before-after approach, and model type B2, cross-sectional generalized linear models. The project team chose the approach for avenue B because it will provide a direct measure of how the lack of motorcycle AADT affects CMF estimation by replicating the process of estimating CMFs.

For the EB before-after approach, a countermeasure was assumed with a known value of its CMF. The project team divided the simulated database into two time periods and adjusted by the value of the CMF in the after period the expected crash means for each location. The Poisson distribution generated the new after period counts. The project team then applied the EB approach to the data for these treated sites, using the remaining sites as a reference group. The project team completed this once using the motorcycle AADTs and once total AADT. The project team then made a comparison to see how the lack of motorcycle AADT affected the estimate of the CMF. The project team performed this entire process, beginning with the simulated data, multiple times and with multiple sample sizes and assumed CMF values.

For the cross-sectional regression model approach, an assumed CMF relationship based on logical considerations and related research was defined and added to the SPFs developed in model type A1. The project team used this modified SPF to simulate the data. The project team then used GLM to re-estimate the SPF, including the fictional variable, with and without motorcycle AADTs. The project team then made a comparison to see how the lack of motorcycle AADT affected the estimate of the CMF. The project team performed this entire process, beginning with the simulated data, multiple times with varying sample sizes and assumed CMF values.

The avenue B analyses used the roadway inventory, total AADT, and motorcycle AADT collected for the avenue A methods in Florida and Pennsylvania. For motorcycle crashes, the project team simulated the crash counts using the SPFs developed in the model type A1 models as a starting point.

The results for avenue B, which investigated the EB before-after approach, indicate that there was relatively little difference (0.05 or less) between the CMFs estimated using the motorcycle AADT SPF versus using the total AADT SPF. However, the estimated CMFs between simulation runs can vary considerably due to the relatively low frequency of motorcycle crashes.

The results for avenue B, which investigated the cross-sectional regression approach for estimating CMFs, also showed that the estimated CMFs were close when using either the motorcycle or total AADT as an exposure measure but that the variability in CMF estimates between simulations was large.

The findings of both the avenue A and avenue B modeling indicate that when motorcycle volumes are not known, using total AADT on its own is sufficient for developing SPFs and CMFs. The potential bias due to missing motorcycle-specific AADT is sufficiently negligible where it exists so as not to preclude SPF and CMF development.

The project team concluded that attempting to predict motorcycle volumes is not possible using typically available roadway and county-level data. Improvement could possibly be found in trip generation type modeling at a disaggregate scale, although given the success of the SPFs using total AADT, such an effort may not be worthwhile.

A more significant issue in developing motorcycle crash SPFs and CMFs is working with a crash type that is relatively rare. The analysis did not develop SPFs for all motorcycle crash types or site types. More evidently, in the estimation of CMFs, the CMF value varied significantly between simulation runs due to the low frequency of motorcycle crashes.