Investigated were the effects of two levels of penalty for incorrect responses on two dependent variables (a measure of risk-taking or confidence, based on nonsense items, and the number of response-attempts to legitimate items) for three treatment groups in a 2x3, multi-response repeated measures, multivariate ANOVA (Analysis of Variance) design. Subjects responded under one of three scoring-administrative rules: conventional Coombs-type directions and two variants suggested as mathematically more adequate. Results indicated significant differences both among groups and across conditions. The results were discussed with reference to the question of test validity in general, and the problems posed for criterion-referenced measurement. (Author)