Predicting performance in anatomy using antecedent academic criteria

The performance of two multivariate prediction models (equal and differential weights) at forecasting the outcome of the comprehensive examination in preclinical anatomy is reported. The models were devised by regression analysis, using scores in two antecedent examinations which correlated highly with the criterion variable, namely the science aggregate at the Joint Matriculation Examination and the premedical science aggregate. Five years of data were considered and the scores for each year were computed using the formula derived from scores of the preceding year. The average failure rate for the period covered by the study was 31.28%. The correlation between actual and predicted scores was positive and moderately high (Pearson r = 0.47 and 0.49 for the equal- and differential-weights models respectively). On average, the equal-weights model correctly predicted 33.9% of failures, with a false alarm rate of 28.4%, compared with 44.3% and 38.9% respectively for the differential-weights model. In predicting candidates whose scores would fall in the bottom half and bottom third of the class, no statistically significant difference was noted between the hit rates achieved by both models. The equal-weights model is simple to formulate and efficient in operation. Although its hit rate at the pass/fail boundary was lower than that of the differential-weights model, it yielded a significantly lower false alarm rate (28.4% vs 38.9%). The relevance and application of performance prediction to the planning of remedial instruction are discussed.