Multiple-cue probability learning can predict airline pilot training success
Résumé
Introduction : Could the multiple-cue probability learning paradigm (Brunswik, 1952) be used to assess the ability of pilot candidates to deal with unpredictable environments ? Three studies investigated this question at ENAC, the national French airline pilot training. Method : Each study comprised two parts. In part 1, the pilot candidates had to deal with an MCPL task that was added at the end of the psychological selection stage (N=401, N=448 and N=589). In part 2, we focused on the success/failure of the selected trainees (N=44, N=74 and N=80) during pilot training (5-10% of failures). The MCPL tasks consisted of predicting a continuous criterion from two or three continuous cues, with a mix of positive and negative cues (60 trials in limited time). Results : MCPL performance was dichotomized in poor/good based on final achievement and number of fulfilled trials. Good MCPL performance was a "sufficient" condition of training success in 100%, 93% and 97% of the cases. Poor MCPL performance was a "necessary" condition of training failure in 100%, 50% and 50% of failures. Moreover, the hypothetical use of the MCPL performance at the selection stage would have rejected only 2.7%, 6.9% and 8.6% supplementary applicants. Discussion : Traditionally, cognitive ability and personality measures can at best predict around 30% of the variance of pilot training success (e.g., Carretta, 2011). MCPL performance could be complementary of other measures. Conclusion : A poor final MCPL performance could be used to alert pilot training organizations at the selection stage.