A method to account for and estimate underreporting in crash frequency research

Area of Science:

• Transportation Engineering
• Econometrics
• Traffic Safety

Background:

• Underreporting of traffic crashes is a significant challenge in frequency research.
• Existing statistical methods often fail to adequately address this underreporting issue.
• Limited research has focused on statistical models capable of accounting for underreporting.

Purpose of the Study:

• To compare the performance of underreporting models with models that account for unobserved heterogeneity in crash frequency analysis.
• To assess the bias introduced by ignoring underreporting in statistical models.
• To evaluate the ability of underreporting models to predict non-reportable crashes.

Main Methods:

• Comparison of elasticities between negative binomial underreporting models and random parameters negative binomial models.
• Analysis of reported crashes versus police-reported but non-reportable crashes (e.g., property damage only).
• Utilized data from 21,340 segments of two-lane rural highways over an eight-year period (2005-2012).

Main Results:

• Statistical models that ignore underreporting yield biased regression coefficients when variables influence both crash frequency and reporting probability.
• The magnitude of bias in elasticities ranged from 0.00% to 16.79%.
• When variables affect only crash frequency, underreporting models align with methods that do not account for underreporting.

Conclusions:

• Ignoring crash underreporting can lead to significant biases in transportation safety research.
• Underreporting models provide a more accurate representation of crash frequency when underreporting is present.
• The findings underscore the importance of employing appropriate statistical methods to handle underreported crash data.