A model for the correction for guessing on multiple-choice tests

Abstract

Multiple-choice tests are used widely because of obvious practical advantages, but inherent in them is the problem of dealing with the effects of guessing. The only widely used model for predicting effects of guessing is based on the primary assumption that, the probability of making a correct response on each test item is either 1.0 or the reciprocal of the number of available choices. The assumption that the subject either makes the correct response on each trial or guesses blindly is clearly too restrictive. Nunnally in [1] relates that because in most educational tests there is some opportunity for "narrowing" alternatives before guessing the standard correction for guessing tends to be an underestimate of the actual effects of guessing. The model proposed in this paper attempts to correct for this.