Consistency of test behaviour and individual difference in precision of prediction

Ghiselli ((1956, 1960) argued that the precision of prediction on the basis of a test may vary for different individuals. To quantify the individual precision of prediction he compared the observed criterion scores with the expected criterion scores estimated on the basis of the total scores on a predictor test. Using these difference scores as a moderator variable, predictor tests were developed to identify persons with large errors of prediction. One of the drawbacks of Ghiselli's methods is that it is very situation specific. As an alternative, Hulin, Drasgow & Parsons (1983) proposed to identify persons with inconsistent or unscalable test responses by means of appropriateness measurement (or person‐fit) statistics. In the present study a person‐fit statistic is used to identify persons with unexpected test responses on selection data. It is shown that, in general, persons with inconsistent test responses are less predictable than those with consistent test responses. Both persons with lower criterion scores as well as persons with higher criterion scores than predicted could be identified. Because person‐fit statistics are relatively easy to calculate, they may be used to improve the precision of measurement in selection settings. Implications for the selection practice are discussed.