End-of-Chapter Application Case Keeping Students on Track with Web and Predictive Analytics What makes a college student stay in school? Over the years, educators have held a lot of theoriesfrom student demo- graphics to attendance in college prep coursesbut theyve lacked the hard data to prove conclusively what really drives retention. As a result, colleges and universities have struggled to understand how to lower dropout rates and keep students on track all the way to graduation. That situation is changing at American Public University System (APUS), an online university serving 70,000 distance learners from the United States and more than 100 countries. APUS is breaking new ground by using analytics to zero in on those factors that most influence a students decision to stay in school or drop out. By leveraging the power of predictive analytics, we can predict the probability that any given student will drop out, says Phil Ice, APUS director of course design, research, and development. This translates into actionable busi- ness intelligence that we can deploy across the enterprise to create and maintain the conditions for maximum student retention.

Rich Data sources As an online university, APUS has a rich store of student information available for analysis. All of the activities are technology mediated here, says Ice, so we have a very nice record of what goes on with the student. We can pull demo- graphic data, registration data, course level data, and more. Ice and his colleagues then develop metrics that help APUS analyze and build predictive models of student retention. One of the measures, for example, looks at the last time a student logged into the online system after starting a class. If too many days have passed, it may be a sign the student is about to drop out. Educators at APUS combine such online activity data with a sophisticated end-of-course survey to build a complete model of student satisfaction and retention. The course survey yields particularly valuable data for APUS. Designed around a theoretical framework known as the Community of Inquiry, the survey seeks to understand the stu- dents learning experience by analyzing three interdependent elements: social, cognitive, and teaching presence. Through the survey, Ice says, We hone in on things such as a students perception of being able to build effective community. increasing accuracy It turns out that the students sense of being part of a larger communityhis or her social presenceis one of the key variables affecting the students likelihood of staying in school. Another one is the students perception of the effectiveness of online learning. In fact, when fed into IBM SPSS Modeler and measured against disenrolment rates, these two factors together accounted for nearly 25 percent of the overall sta- tistical variance, meaning they are strong predictors of stu- dent attrition. With the adoption of advanced analytics in its retention efforts, APUS predicts with approximately 80 percent c ertainty whether a given student is going to drop out. Some of the findings generated by its predictive mod- els actually came as a surprise to APUS. For example, it had long been assumed that gender and ethnicity were good predictors of attrition, but the models proved otherwise. Educators also assumed that a preparatory course called College 100: Foundations of Online Learning was a major driver of retention, but an in-depth analysis using Modeler came to a different conclusion. When we ran the numbers, we found that students who took it were not retained the way we thought they were, says Dr. Frank McCluskey, pro- vost and executive vice president at APUS. IBM SPSS predic- tive analytics told us that our guess had been wrong. strategic course adjustments The next step for APUS is to put its new-found predictive intelligence to work. Already the university is building online dashboards that are putting predictive analytics into the hands of deans and other administrators who can design and implement strategies for boosting retention. Specific action plans could include targeting individual at-risk students with special communications and counseling. Analysis of course surveys can also help APUS adjust course content to better engage students and provide feedback to instructors to help improve their teaching methods. Survey and modeling results are reinforcing the univer- sitys commitment to enriching the students sense of commu- nitya key retention factor. Online courses, for example, are being refined to promote more interactions among students, and social media and online collaboration tools are being deployed to boost school spirit. We have an online student lounge, online student clubs, online student advisors, says McCluskey. We want to duplicate a campus fully and com- pletely, where students can grow in all sorts of ways, learn things, exchange ideasmaybe even booksand get to know each other. smart Decisions While predictive modeling gives APUS an accurate picture of the forces driving student attrition, tackling the problem means deciding among an array of possible intervention strategies. To help administrators sort out the options, APUS plans to implement a Decision Management System, a solu- tion that turns IBM SPSS Modelers predictive power into intelligent, data-driven decisions. The solution will draw from Modelers analysis of at-risk students and suggest the best intervention strategies for any given budget. APUS also plans to delve deeper into the surveys by mining the open-ended text responses that are part of each questionnaire (IBM SPSS Text Analytics for Surveys will help with that initiative). All of these data-driven initiatives aim to increase student learning, enhance the students experience, and build an environment that encourages retention at APUS. Its no coincidence that achieving that goal also helps grow the universitys bottom line. Attracting and enrolling new students is expensive, so losing them is costly for us and for students as well, McCluskey says. That is why we are excited about using predictive analytics to keep retention rates as high as possible.

What additional analysis are they planning on conducting? Can you think of other data analyses that they could apply and benefit from?