Sex and ethnic differences in aptitude indicator measurement models/
Measurement error in construct indicators is known to bias structural model regression coefficients, and differences in regression coefficients when two groups are measured with differential error. The validity and reliability of six aptitude tests administered to high school seniors in the High School and Beyond (HSB) national longitudinal study were investigated for white, black and Mexican-American males and females. The tests were found to be valid measures of the same constructs across groups, but test reliability coefficients were found to differ between males and females, between whites and blacks, and between whites and Mexican-Americans. For unspeeded tests, reliability coefficients were consistently lower for females than males, and lower for blacks and Mexican-Americans than for whites.
The effect of different test reliabilities on structural model regression coefficients, and differences in coefficients when two groups are compared, was assessed. The coefficient of determination (R) was attenuated to a greater extent and regression coefficients were more biased for blacks and Mexican-Americans than for whites. Coefficient differences were modestly biased when two groups with different test reliabilities were compared.
Maximum likelihood covariance structure analysis procedures were used to estimate, fit and compare measurement and structural models. The computer program used was LISREL IV. Ordinary least squares regression coefficients were compared to LISREL regression coefficients in assessing the extent of bias in regression coefficients.