| Description |
This work uses linear and nonlinear models in order to predict student success and pathways in higher education in the state of Utah. Postsecondary Grade Point Average is used as a metric for success in higher education. Pathways are identified using clustering analyses, which group observations according to various distance measures. Eleven institutional pathways were found for the 2008 cohort of Utah high school graduates. Analyzing enrollment data from 2000{2013, four major trends emerged as distinct patterns of post-secondary attendance (or lack thereof) for the cohort: no attendance, early peaking attendance, late peaking attendance, and completion. Under-represented Racial Minority (URM) students, low income students and geographically mobile students were over represented in the group of No-Goers. After identifying pathways, a Random Forest (RF) algorithm was used in order to predict which pathway a student might take based on high school characteristics. Important variables in the random forest prediction include High School GPA, college courses taken in High School, ACT scores, High School proportion of Low Income students, and High School proportion of URM students. Linear models also showed that demographic variables such as race, mobility, and Low Income status, in addition to Pell Grant status carry significant predictive weight for Postsecondary GPA. Attending schools with high proportions of Low Income or URM students adversely a fected students' likelihood of succeeding in college. Pell Grants may have the ability to partially compensate for the disadvantage faced by low income students; Pell Grant eligibility and reception were strongly associated with increased grade point average and retention. The RF algorithm also predicted college attainment correctly at a higher rate than the logistic model. The RF algorithm outperformed Ordinary Least Squares regression models and random coefficients hierarchical linear models in predicting student Grade Point Averages from a test set. The RF algorithm had smaller mean squared error than linear models, as well as smaller absolute medians and quartile values. Due to its accuracy, simplicity and intuitive nature, RF analysis is recommended for future critical quantitative research in higher education. Substantially, the author advocates for an expansion of the Pell Grant program in order to combat the signifcant postsecondary disadvantages faced by high achieving low income students. Mobility is identifed as an underresearched area with high potential impact in postsecondary attainment. Addressing inequities in education pathways, particularly the lasting effects of inferior schools which low in- come and URM students often attend, will be necessary in the development of the United States as a free, democratic society. Only through critical, equitable, education can members of society liberate themselves from various forms of oppression and reach their full humanity. |
