Validation of the Dutch Airforce Test Battery Using Artificial Neural Networks

THEORETICAL INTRODUCTION The main selection criteria for individual tests and test batteries used to select military pilot applicants are the construct and criterion validity, the overall cost of testing and the time requirements. Naturally, the derivation of decisions from a test battery requires a sufficiently high correlation between the tests and the criterion variable. However, recent metaanalysis (cf. Burke, indicates that the correlation coefficients between a single test and the criterion measure do not exceed an absolute value of .30. There are a variety of causes for this, ranging from a lower reliability of the criterion or to the lack of symmetry between the generality of the predictor variables and the generality of the criterion variable. With regard to the later cause Wittmann and Süß (1997), Ajzen (1987) and Ree and Carretta (1996) pointed out that for more general and global criteria such as successful performance in a flight-simulator or an educational program, aggregate measures such as general ability (" g ") are better suited for prediction than more specific predictors. Thus one way to handle this problem is to combine the available information about an applicant to generate a prediction about his success. In general, one can resort to various methods of statistical judgment formation in order to do so. But classical methods of statistical judgment formation such as discriminant analysis or regression analysis are vulnerable to violations of their statistical assumptions and often lack stability in cross-validation in practical applications (cf. Bortz, 1999; Brown & Wickers, 2000). A promising alternative is the use of artificial neural networks. This statistical method has few requirements with respect to data characteristics and has proven to be a robust procedure for pattern recognition tasks evaluated artificial neural networks with regard to their ability to predict naval aviator flight grades in their primary phase of flight training using a test battery which primarily consisted of psychomotor tests. Griffin's results indicated that artificial neural networks resulted in a higher validity coefficient compared to the multiple linear regression analysis. However the difference did not reach statistical