Using non-linear methods to investigate the criterion validity of traffic-psychological test batteries

Abstract

In several countries in Europe (among others Germany and Austria) persons who have lost their drivers licence have to undergo a psychological test in order to regain their licence. The article discusses the validity of two test batteries of the Expert System Traffic using standardized driving tests [Schuhfried, G., 2005. Manual Expert System Traffic (XPSV). Schuhfried GmbH, Mödling]. A global evaluation of the respondents' performance in a standardized driving test was used as criterion measure in order to divide the subjects into drivers, who were classified as relatively safe or unsafe according to their performance in a standardized driving test. Artificial neural networks were used to calculate the criterion validity. This yielded superior classification rates and validity coefficients compared to classical multivariate methods such as a logistic regression. The stability and generalizability of the results was empirically demonstrated using a jack-knife validation, an internal bootstrap validation and an independent validation sample which completed the test batteries and the standardized driving test as part of a so-called traffic-psychological assessment which is compulsory in Austria in all cases, where the driver's licence has been withdrawn, e.g., when caught driving under the influence of alcohol. Moreover, the procedure allows calculating incremental validities which enable the assessment of the relative importance of the individual predictor variables. This contributes to the transparency of the results obtained with artificial neural networks. In summary it can be said that the results provide empirical evidence of the validity of the traffic-psychological tests batteries used in this study. The practical implications of the results for traffic-psychological assessment are described.