2017 Theses Doctoral

Mining Transactional Student-Level Data to Predict Community College Student Outcomes

Lenchner, Erez

A longitudinal analysis of transactional data for an entire college cohort was mined from administrative student records systems to identify individual student behaviors and establish correlations between individual students’ behaviors and academic outcomes. Conducted at one large urban community college, this study determined curricular peer association behavior between individual students, and also evaluated late registration and course schedule change behaviors. Findings demonstrated a strong correlation between these three behavioral patterns and a lasting influence on academic outcomes, such as: semestrial GPA and cumulative GPA, credit accumulation, persistence and graduation rates. Finding also indicated a correlation among the three behaviors themselves. Furthermore, conducting a longitudinal analysis of individual students made it possible to identify the temporal tipping-points which differentiated at-risk behavior from otherwise benign behavior. The intrinsic factors associated with individual students’ behaviors were followed over a period of thirteen consecutive semesters. Mining Transactional Student-Level Data at the scale achieved in this study, when compared to traditional methods of data collection, provided the precision needed to determine the actual proximity among specific peers, and the identification of registration behavior patterns. The extraction of transactional data from the records of each student in an entire cohort resulted in a method of inquiry immune to the negative effects of student’s non-response or selection bias. Complimenting previous research, this study provides a detailed descriptive analysis of those behaviors not only at the semestrial level, but also cumulatively across consecutive semesters.
This study demonstrates that curricular peer association can be measured directly from common, ubiquitous, transactional records. The rates of Peer Association among individual students was very dynamic: While the majority of students had some peer associations while enrolled, in the aggregate two thirds of students had no peer association (were soloists) at some point in time, while more than a quarter of all students were soloists for at least half of their entire enrollment period.
Soloists differed from students with peer associations. They were likely to be older, international students, African Americans, transfer students, or those entering fully prepared for college level coursework (no remedial coursework). Peer association was positively correlated, both in the semester in which it occurred and cumulatively, with: GPA, credits earned, and retention or graduation rates. These correlations to academic outcomes varied with the number of peer associations established, and the intensity of peer encounters.
The study revealed that nearly a quarter of all students practiced late registration at least once; and more than 10 percent have registered late multiple times during their studies. Nearly three quarters of students made modifications to their course schedule at least once after the semester began. Overall, two fifths of students changed their initial schedule every semester. These behaviors were unrecorded in previous studies that were limited in the evaluation of longitudinal behaviors, used subsets of students and were subject to non-response bias. Late registration and student schedule changes was correlated with lower semestrial and cumulative academic outcomes. Late registration behavior subsequently increased the likelihood of a student being a soloist. When compared to previous studies, the analysis conducted here not only accounted for academic, demographic and financial variables at baseline, but went on to perform updates at key points in time each semester to reflect changes over time. The exhaustive revisiting of the covariates each semester provided enhanced control to the ‘order of time’ influence. All covariates were re-measured each semester allowing to better evaluate the correlation of student behavioral indicators for a given semester, and cumulatively. This enhanced the study’s ability to account for common unobserved variables inherent to academic, demographic and financial attributes that might influence student outcomes correlated with peer association, late registration and schedule changes.
This study contributes to the literature by showing that peer association can be evaluated in the setting of an open admission commuter institution, and that peer association has consistent and positive correlation with academic outcomes. It provides new insights regarding the magnitude of late registration and schedule changes, as well as their negative immediate and longitudinal correlation with student outcomes. Further implications to community colleges’ faculty, administrators, researchers and policymakers, as well as future directions for research employing transactional level data are discussed.