A 5-hour workshop taught by Serge Herzog, Ph.D.
An overview of our predictive analytics workshop
The purpose of this workshop is to teach institutional research, assessment, and evaluation professionals how to effectively build and implement a predictive model to identify students at risk of dropping out using standard regression methods with SPSS. Instruction will be delivered in a hands-on format, offering an interactive step-by-step model-building process that allows participants to develop their own prediction model, using preloaded data that mimics information available with the typical college enrollment matriculation system.
Expected outcomes of our predictive analytics workshop
By the end of the workshop, participants should know how to do the following:
- Develop a conceptual understanding of how predictive models developed can improve institutional effectiveness with a focus on student retention;
- Set up a matriculation system (or census warehouse) data file in IBM-SPSS that can be used to develop a predictive statistical model to identify students at risk;
- Use historical data to ‘automatically’ develop predictor coefficients to estimate (score) the dropout risk for students in future cohorts; and
- Translate the student dropout risk into a relative percentile risk score to assist student support services with ‘actionable’ information in a timely fashion.
Who should attend?
The target audience is educational researchers who are familiar with logistic regression, and wish to use it to develop prediction models to estimate student dropout risk or other student or educational outcomes that are categorical in nature. This is an applied course, so no advanced math skills are required beyond an understanding of logistic regression and its associated statistical output and model fit indicators (which will be explained in the workshop).
Attendees should be proficient in the basic use of and have access to IBM-SPSS software in order to participate in hands-on exercises to develop a prediction model with furnished data and syntax files.
- Introduce the power of predictive analytics (including examples of forecasting data used in improving student success and college operations).
- Examine elements needed and available at start/middle of semester to predict student dropout at end of semester (including pre-college academic preparation, student socio-demographic data, income/financial aid profile, semester course data, on-campus housing and campus engagement data).
- Conduct exploratory data analysis: Discussion of variable selection, variable coding, missing data imputation, composite variable construction to achieve maximum model parsimony.
- Discuss regression model: Prediction versus variance explanation in logit analysis.
- Develop training dataset using historical data to generate predictor coefficients for future data (outcome estimation for future cases)
- Estimate outcome probability for future data (cohorts) Identify statistical outliers and develop ROC curve to maximize correct classification rate.
- Chose model with optimal balance in correct classification.
- Transform outcome probability for each case (student) into ‘actionable information’.
- Discuss how predictive analytics improves organizational productivity and outcomes
About the instructor
Serge Herzog, Ph.D., has been the Director, Office of Institutional Analysis, and Consultant, Center for Research Design & Analysis, University of Nevada, Reno since 2001. His research has been covered in the Chronicle of Higher Education, the University Business Magazine, and Campus Technology Magazine among others. Most recently, he co-edited (with Nicolas Bowman) Methodological Advances and Issues in Studying College Impact (New Directions for Institutional Research) San Francisco: Jossey-Bass, 2014.