A new method for estimating effectiveness of road safety features

Abstract

From a Bayesian perspective, historical data from past similar studies can be very helpful in interpreting the results of the current study. According to Haseman, Huff and Boorman(1984), historical data can be useful when control tumor rates are low and when marginal significance levels are obtained in a test for dose effects. For example, that 4 of 50 animals in an exposed group develop a specific tumor, compared with 0 of 50 in a control group. This difference is not statistically significant (p=0.12, based on Fisher’s exact test). However, the difference may be biologically significant if the observed tumor type is known to be extremely rare in the particular animal strain being studied.