01337997

Component

Before-and-after studies of safety measures introduced at locations that have a low mean number of accidents often encounter the problem of zero accident counts. This is a problem for three reasons. First, it is highly implausible that the true long-term mean number of accidents at any site is zero. Second, if results from several sites are combined by means of the inverse-variance (i.e. log-odds) technique of meta-analysis, zero counts must be adjusted when estimating the statistical weight to be assigned to each result. Third, if a zero count is taken at face value, it suggests that the effects of a safety treatment could be either a hundred percent accident reduction (if there was a positive count before and a zero count after) or an infinite increase in the number of accidents (if there was a zero count before and a positive count after), both of which are highly implausible. This paper explores the use of techniques for continuity correction to adjust zero counts. A technique proposed in epidemiology is applied to a fictitious data set. A simple method derived from the empirical Bayes (EB) method is proposed for implementing continuity corrections to the count of accidents after treatment in EB-studies.

01329018

11-0133

English

17p

2011

Transportation Research Board 90th Annual Meeting

DVD

References (24)

Epidemiology; High risk locations; Highway safety; Intersections; Meta-analysis; Roundabouts; Variance

Road Safety Audit Program (Federal Highway Administration)

Empirical Bayes method; Research methodology

Data and Information Technology; Highways; Safety and Human Factors; I73: Traffic Control

Transportation Research Board Annual Meeting 2011 Paper #11-0133

TRIS, TRB

Feb 17 2011 5:20PM