Details:
-
Creators:
-
Corporate Creators:
-
Contributors:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Corporate Publisher:
-
Abstract:In developing accidents-flow-roadway design models, the R-squared goodness-of-fit measure has been used by traffic safety engineers and researchers for many years to (1) determine the quality and usability of a model; (2) select covariates (or explanatory variables) for inclusion in the model; (3) make a decision as to whether it would be worthwhile to collect additional covariates; and (4) compare the relative quality of models from different studies. Through computer simulations, this study demonstrated the pitfalls of using R-squared to make these decisions and comparisons. Other goodness-of-fit criteria such as the Akaike Information Criterion, scaled deviance, and Pearson's X-squared statistics were also introduced and evaluated. Based on limited simulation results, one of the alternative criteria called R-squared-alpha was recommended for evaluating and comparing the quality of accident prediction models when sample size is large. Finally, the interrelated and complementary nature of two approaches that have traditionally been used to develop the relationship between run-off-the-road accident frequency and roadside hazards (i.e., accident-based approach and encroachment-based approach) were studied and demonstrated using data from a Federal Highway Administration and Transportation Research Board roadway cross-section design data base. It was suggested that exploring the complementary nature of these two approaches could be a viable avenue to reduce data collection cost.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: