Multi-Institutional Data Predicting Transfer Student Success Denise Nadasen Anna Van Wie Institutional Research University of Maryland University College Outcomes for this Session • You will learn about the: – Goals for this grant and the research project – Process for integrating a multi-institutional data base – Research questions, methods, and findings – Lessons learned and next steps 2 Goals of the grant • • • • • • • Collaborate with the community colleges Define research questions and variables Build a dataset for transfer students Explore predictor/outcome variables Predict student success Report the results at national conferences Use the results to inform policy and practice to better serve transfer students 3 Collaborative Partners • UMUC is an online institution that enrolls over 90,000 diverse students each year worldwide • Prince George’s Community College is located within two miles of UMUC’s Academic Center and enrolls over 37,000 diverse students. • Montgomery College is located within 10 miles of UMUC’s largest regional center, and enrolls over 35,000 diverse students. 4 The Team • PI– President, Provost • Sponsor – Institutional Research • Partners – Montgomery College and Prince George’s Community College – Undergraduate retention and data mining specialist – External evaluators • Researchers: – Cheoleon Lee, Jing Gao, Futoshi Yumoto, Husein Abduhl-Hamid • Data Mining Specialists – Stephen Penn, The Two Crows 5 The Student Population • Students enrolled at UMUC between 2005 and 2011 • PG and MC transfer students – Direct compare (32,000) – National Student Clearinghouse (12,000) – UMUC records (8,000) 6 Merging Multi-Institutional Data • Protect this data! • Balance institutional-specific protocols with research-based definitions • Address data anomalies • Distinguish student level vs. course level • Define LMS data – Limits on data extract 7 KDM • Integrates student data – Community College and UMUC SIS – Demographic – Courses – Performance – Classroom behavior (LMS) • 300 source and derived variables • Gather from disparate sources • One time snapshot 8 MC-PGCC-UMUC Transfers UMUC students who transferred from MC or PGCC and were matched in the BASE file PGCC students and class data MC students and class data Prior Work derived data for transfer students Base Extract UMUC undergrad students enrolled between Spring 2005 and Spring 2011 WT extract Classroom activity Data Warehouse UMUC students from PeopleSoft Daily Update PeopleSoft WT Live SIS with UMUC students Online Classroom Question • What barriers would your institution face in merging multi-institutional data? 10 Research Goals • • • • Define outcome variables Define predictor variables Model the student lifecycle Determine the success and failure factors • Develop and implement interventions • Impact outcomes 11 Outcome Variables • • • • • Successful course completion (percent) First term GPA (dichotomized) Reenrollment in next term (Y/N) Retention (12 month window – Y/N) Student Classification (Slackers, Splitters, Strivers, and Stars) 12 Transfer Student Progressions Transfer Four-Year Institution cc Demog and Other Academic Work First Semester Semester 2 Last Semester Graduate School cc Transfer 13 Research Studies 14 Which variables contribute to the prediction of online course success? The data: • 4,558 new, undergraduate, first bachelordegree seeking enrollments in 15 UMUC online gateway courses in Spring 2011. • Transfer data on students from partner institutions, Montgomery College (MC) and Prince George's Community College (PGCC). Methodology • Exploratory factor analysis (EFA) was used to identify key covariates. • Logistic regression was used to predict course success. Findings • Total number of transferred credits is the best predictor of course success – pseudo R2 value around .12 • GPA from transferred credits is the second best predictor of course success – pseudo R2 around .11 • Semester course load contributes less to course success than other covariates. Findings • Four of five predictors derived from online student behavior show a strong contribution to successful course completion. Final Predictive Model Total number of transfer credits Summary of students’ week 0 behavior prior to the first day Significant Variables GPA from transferred credits Semester course load Amount of time since students attended the last institution Read a conference note Significant Online Entering a class behaviors Created a conference note Created a response note Which variables predict retention in an online environment? • The same data set for the prediction of course success • Add in retention status from Summer 2011, Fall 2011, and Spring 2012. Methodology • Logistic regression • Preliminary analysis focused on the evaluation of covariates (as identified in the previous analysis) predictors based on the students’ coursework behavior, and course success. Findings • The covariates and student behavior variables made less of a contribution to this model than the prediction of course success. • These results indicate that course success may be a good predictor of retention. What is the relationship between prior academic coursework and UMUC first semester gateway course on reenrollment? The population: • Students new to UMUC in Fall 2008 to Fall 2010 • Took ACCT220, BMGT110, CMIS102, GVPT170, or PSYC100 in their first semester. Methodology • Association algorithm Apriori to determine relationships between courses in previous academic work and re-enrollment rates. • The algorithm indicates when a certain condition is found another condition can be expected. Findings Significant Course Relationships Community College Course Disciplines First UMUC Course Math or Business ACCT 220 Math or Business CMIS 102 Business or Science BMGT 110 Science or communication GVPT 170 Math or communication PSYC 100 We cannot assume causality Current Studies 26 Examine CC Courses • Explore relationship between CC courses and first term GPA • Identify courses of interest • Developmental Ed sequencing • Successful completion of CC course • Mixing course level and student level data 27 Predicting First Term GPA • What CC variables predict first term GPA of 2.0 or higher? • Course Efficiency • CC courses – English, Math, Speech, Computer, Honors, Online Course, Remedial • Demographics – Age, Gender, Race, Marital Status, Cohort, Community College Origin, Terms skipped 28 The Population • 9,063 students from MC and PGCC • Mostly Single, African-American, and female • Most do not skip terms • Most get A’s and B’s at CC • Most have >2.0 at UMUC • Only PGCC offered online courses 29 Predictors of Success Age Gender Race Marital Status C.C. Courses English Math Speech Success @ UMUC Computer Honors On-line Remedial Course Efficiency Logistic Regression 30 Question • What CC variables do you think are good predictors? 31 Findings • Predictive variables: – Age, marital status, and under-represented minorities have predictive power – Math and Honors courses have positive effects – Remedial and Online have a negative effect – Course efficiency has a positive effect 32 Predicting Student Clusters • Dataset includes all PGCC and MC students who transferred • Student level derived variables • Cluster students based on retention and first term GPA at UMUC • Predict clusters from prior CC work and demographic variables 33 Retention No Retention Yes Success Quadrants Strivers Stars Slackers Splitters GPA < 2.0 GPA > 2.0 34 Stay Tuned …. • Data mining continues – So far, Stars appears to have distinguishing features • Focus on top 50 CC courses and combinations of courses as predictors • Focus on performance in gateway courses at UMUC as outcomes 35 Summary of Findings • Positive effects – Transfer credit, prior GPA, math, honors, course efficiency, online activity, age, marital status – Course success can predict retention • Negative effects – remedial, online, minority status 36 Interventions • Identify areas of risk • Collaborate with CC • Develop intervention strategies – Advising – Messaging – Learning community – Course development 37 3 Projects Synergizing PAR Kresge Civitas 38 Next Steps • • • • Examine course success at the CC Implement/evaluate interventions Update KDM with more data Develop, understand, and explain predictive models to identify at risk students at the CC 39 Lessons Learned • • • • Long term plan for data up front Get a project manager Manage expectations Communicate progress far and wide 40 Questions • Anna anna.vanwie@umuc.edu • Denise Denise.nadasen@umuc.edu 41