Validity and fairness of college admissions: Model choices and interaction with college characteristics
The current study employed two-level hierarchical generalized linear models (HGLM) to compare the admission validity and fairness with respect to gender, ethnicity, and first language between two admission models, and examined whether institutional characteristics can explain the variations across colleges. The two admission models included the Combined model, where students were admitted based on a multiple regression with HSGPA and SAT scores (SATM and SATV) to predict FRGPA. The other one is the Either-Or model, where students were admitted based on the regression of FRGPA on either HSGPA or SAT scores alone. The validity or fairness of admission was measured by log odds ratio (LOR), which was built in a logistic regression as the slope coefficient in HGLM. That was in the first level, logit of success probability for each individual was regressed on individual's admission status determined by the Combined model (or the Either-Or model). In the second level, the variation of the association between admission status and success status was regressed on institutional characteristic such as size, mean of SAT scores, mean of HSGPA, percentage of female, percentage of Non-Asian minority students, and percentage of Asian minority students alone. The admission fairness was also modeled based on an HGLM framework. In the first level, the logit of success probability (or admitted probability) of individual regressed on gender or other dichotomized subgroup indexes such as ethnicityI (Non-Asian minority students vs. White students), ethnicityII (Asian minority vs. White students) and first language (ESL or not), with the second level the same as that for the validity. ^ Based on the Empirical Bayesian estimates from Unit Specific and Population Average models under HGLM, the study found that the Combined model had slightly higher admission validity than the Either-Or model; the Either-Or model mimicked the association of success with ethnicityI and first language subgroups more closely than the Combined model, but not for the gender-success LOR nor the ethnicityII-success LOR. It was also found that institutional size, mean of SAT total score, mean of HSGPA, percentage of Asian minority students accounted for the variation of gender LORs significantly, but with negative coefficients. Institutional size, mean of SAT total score, and mean of HSGPA were significant predictors for the variation of gender fairness LORs and the variation of ethnicityI fairness LORs. Only the mean SAT total score accounted for the variation of ethnicityII fairness LORs to the positive direction. None of the institutional characteristics accounted for the variation of first language fairness LORs significantly. Interpretation of these findings, possible explanations and implications for practice were discussed, as well as limitations of the current study, and directions for future research. ^
Psychology, Psychometrics|Education, Higher
"Validity and fairness of college admissions: Model choices and interaction with college characteristics"
(January 1, 2007).
ETD Collection for Fordham University.