The speed-error trade-off problem in psychometric testing

The standard procedure for scoring speeded psychometric tests (guessing corrected number right in fixed time) produces scores which depend both on the speed and accuracy of candidates' responding. These in turn depend both on a candidate's ability to perform the task and the compromise between speed and accuracy which is adopted. This paper uses computer simulations based on McClelland's (1979) cascade model to investigate the extent to which the scores produced reflect ability, as they purport to do, and the extent to which they are contaminated by speed-error strategy. The standard scoring procedure is compared with some alternative methods for combining speed and accuracy. It is concluded (a) that the standard scoring procedures give scores which are strongly affected by the candidate's compromise between speed and error, (b) that alternative scoring formulae may be less affected and (c) that tests for the sensitivity of scores to speed accuracy compromise should be added to standard test development procedures.