The effect of training to improve letter-naming speed on beginning reading skill was evaluated with 39 first-grade students. Subjects were randomly assigned to training and control groups and were administered curriculum-based measures of reading at pretraining, post-training, and follow-up. Trained students demonstrated significantly faster letter-naming speed at post-training assessment as compared to untrained peers. However, the difference in letter-naming speed did not lead to a significant difference in overall measured reading skill as evaluated by an analyses of covariance. Scores for one of the four reading measures (Fall passage oral reading rate) demonstrated a significant difference at post-training, but were not significant at follow-up. In two of the three remaining measures of beginning reading skill, differences between the mean scores of trained and untrained students were greatest at points where the assessment of skills most closely matched the content of classroom instruction. However, these differences were not statistically significant.

Recent research designed to investigate construct bias in the Kaufman Assessment Battery for Children served the additional purpose of illustrating several useful methods for detecting construct bias. The purpose of this reanalysis was to illustrate a new, powerful method for testing the similarity of the constructs measured across groups (and, thus, detecting construct bias): multi-sample confirmatory factor analysis (MCFA). Our reanalysis suggested that the K-ABC measures indistinguishable abilities for Black and White children at ages 7 through 8. Those analyses also showed differences in the attributes measured for Black and White children for ages 9 through 12. Follow-up demonstrated that the differences among that latter group were due to trivial aspects of measurement rather than critical differences in the constructs measured by the test across groups. We argue that MCFA provides a more organized, direct, objective, and complete method for detecting construct bias than do other methods.

No abstract