Availability bias in road safety systematic reviews and its impact on the meta-analysis findings.

Abstract

Meta-analyses, which present the best source of information on the effectiveness of interventions, are influenced by several biases. One category relates to the convenience of selective inclusion of those primary studies, which are more easily available than others. This availability bias includes bias from excluding the grey literature, bias from excluding non-English literature, and bias from excluding older studies. Existing studies are not conclusive about the impacts of this bias; in addition, none of them focus on road safety meta-analyses. To fill this gap, the present paper consisted of two studies: (1) exploring the presence of availability bias in road safety meta-analyses, and (2) demonstrating the impact of availability bias in several example meta-analyses. Based on an analysis of 80 existing meta-analyses, the first study concluded that compared to the medicine meta-analyses, the road safety meta-analyses use a longer time range, are more often restricted in terms of language, and less often involve the grey literature. The second study utilized selected unrestricted data samples to demonstrate the impact of availability bias in seven meta-analyses. The differences in intervention effectiveness in terms of crash frequency changes between unrestricted and restricted scenarios were identified. This shows that the search restrictions clearly lead to availability bias, which influences the differences in meta-analysis results.