This study empirically tested the tenability of the hypothesis that tests developed based on the internal consistency principle for item selection tend to be systematically biased against ethnic groups with smaller or no representation in the test standardization sample. Two test construction models were developed under tight experimental controls: one with differential representation of ethnic groups (White, African-American, Hispanic, and Asian) in the test development sample, and the other with maximum representation of one ethnic group (100%) in the test development sample. The findings revealed that, under both test construction models, consistently there is no systematic bias against the group(s) with smaller or no representation in the test construction samples. The empirical results support the integrity of the sampling and item selection procedures widely used in psychometric practice.
 

This study assessed the effects of some potential confounding factors on some major descriptive model fit indices in structural equation modeling: data nonnormality, estimation method, and sample size. Based on a balanced experimental design, a total of 14,400 samples were fitted to a true and two misspecified SEM models. The major findings are: (a) relatively mild data nonnormality condition has little effect on the descriptive fit indices; (b) under misspecified models, estimation method (maximum likelihood vs. generalized least squares) has considerable influence on those descriptive model fit indices which belong to the category of incremental fit indices; and (c) some fit indices are more susceptible to the influence of sample size. The previous finding that SEM fit indices were consistent under different estimation methods may need to be revisited, because it was primarily based on true SEM models. Since SEM researchers rarely are certain whether they have correctly specified their models, it is critical that simulation studies are conducted in the presence of model misspecification.