Implementation and Evaluation of the Z-Score System for Normalizing Residency Evaluations

Assessment of clinical competence is essential for residency programs and should be guided by valid, reliable measurements. We implemented Baker's Z-score system, which produces measures of traditional core competency assessments and clinical performance summative scores. Our goal was to validate use of summative scores and estimate the number of evaluations needed for reliable measures.

We performed generalizability studies to estimate the variance components of raw and Z-transformed absolute and peer-relative scores and decision studies to estimate the evaluations needed to produce at least 90% reliable measures for classification and for high-stakes decisions. A subset of evaluations was selected representing residents who were evaluated frequently by faculty who provided the majority of evaluations. Variance components were estimated using ANOVA.

Principal component extraction from 8,754 complete evaluations demonstrated that a single factor explained 91 and 85% of variance for absolute and peer-relative scores, respectively. In total, 1,200 evaluations were selected for generalizability and decision studies. The major variance component for all scores was resident interaction with measurement occasions. Variance due to the resident component was strongest with raw scores, where 30 evaluation occasions produced 90% reliable measurements with absolute scores and 58 for peer-relative scores. For Z-transformed scores, 57 evaluation occasions produced 90% reliable measurements with absolute scores and 55 for peer-relative scores. The results were similar for high-stakes decisions.

The Baker system produced moderately reliable measures at our institution, suggesting that it may be generalizable to other training programs. Raw absolute scores required few assessment occasions to achieve 90% reliable measurements.