Hamner enrollment prediction model: Transition to college for individuals with access needs
Perlow, Ellen Jeanne Ilana
MetadataShow full item record
This Institutional-Review Board-approved study expanded application of the evidence-based enrollment prediction model by Hamner used to forecast enrollment at Texas Woman's University (2002-present) to the population of individuals with access needs transitioning to 2-year or 4-year U.S. colleges. The sampling frame used for this SAS®-based endeavor was the entire National Longitudinal Transition Study-2 (NLTS2, Waves 1-5, 2000-2012, n = 11,270 as rounded to the nearest ten; n > 15,000 variables), analyzed under a restricted use data license number 10100011 with the U.S. Department of Education. In compliance with U.S. Department of Education requirements, NLTS2 sample sizes and derived values were rounded to the nearest ten. Within the context of cross-validation, analyses performed for each model included comprehensive data and documentation exploration, theme and variable construction, chi-squared tests of independence, logistic regression, interpretation of Beta coefficient and odds ratio parameter estimates, and predictive accuracy calculations. Fashioning a thematic categorical approach facilitated singling out from NLTS2 a manageable set of literature-cited themes and component independent variables that evidenced substantial association with the 2-year or 4-year college enrollment outcomes. Moreover, this study introduced two major innovations. Selection of a limited number of independent variables for each model (2-year model: 7; 4-year model: 6) was exacted for association with each model’s dependent variables, current 2-year or 4-year college enrollment during a particular NLTS2 wave (time). Finally, these models showed excellent predictive accuracy (2-year-model, 2.1% and 4-year-model 2.0%) when sample sizes were rounded to or up to the nearest ten) on the testing dataset in the cross-validation analysis. Future research directions include replication using Bayesian methods, model testing using more well-occupied college applicant databases inclusive of both civilian and military veteran populations with access needs. Another direction is design of college applications that (legally) collect data potentially indicative of individuals having access needs, for example, applicant inquiries about [also assistive] technology use and personal transportation availability. Such data collection demonstrates promise for revealing colleges’ future institutional access needs, for example, for public transit and increased on-campus housing and parking accommodations.