Testing for the structural invariance of the Ohio ninth-grade proficiency test across sex, race, and year

The Ohio ninth grade proficiency test has been validated based on its content, response process, relations to other variables since it was first administered in the fall of 1990. The lack of structural investigation of the high stakes test, however, drew the researcher's attention because examination of external and consequential aspects of a test does not become fully meaningful until one can first ensure its content and structural validity. Furthermore, unless one can show that the test measures the same knowledge or skills with the same accuracy for all populations, testing for mean differences between groups is inappropriate (AERA, APA, & NCME, 1999; Bejar, 1980; Labouvie, 1980; Rock et al., 1978).; In this study, the internal relationships of the writing, reading, math, citizenship, and science areas of the Ohio ninth grade proficiency test at the subscale level was validated by examining the hypothetical five-factor model to ensure its structural validity. Further, the invariance of the structural composition of the test across sex, race, and year was examined with samples of 713 and 735 eighth graders who took the test in spring 1999 and 2000 in a large urban school district in Ohio. To examine the invariance of the test (1) loadings, (2) variances, (3) covariances, and (4) measurement errors were individually constrained and the respective model fits were compared with those of the baseline models.; Analyses of the internal structure of the test indicated fairly good degrees of model fit for all of the samples from sex, race, and year. Findings confirmed the suggested construct of the Ohio ninth grade proficiency test on which the test score interpretation is based. Investigation of the structural invariance across sex, race, and year, however, revealed several parameters unequally functioning across groups. While the variances and the covariances were relatively invariant, the factor loadings and the measurement errors were equitable across neither groups nor years. Regarding the structural invariance across test administrations, none of the parameters hold for the two years' test scores. Based on the findings, it was suggested that caution needed in interpreting the test scores from the different populations.