USING PREDICTIVE ANALYTICS TO UNDERSTAND HOUSING ENROLLMENTS Heather Kelly, Ed.D., University of Delaware Karen DeMonte, M.Ed., University of Delaware Darlena Jones, Ph.D., EBI MAP-Works
Predictive Analytics: A variety of statistical techniques that analyze current and historical facts to make predictions about future events. Study the past, if you would divine the future Confucius
Why Retention? Retention affects Enrollment numbers Institutional image Institutional rankings Finances (revenue and funding) Retention is often part of our institutional planning. Retention is essential to student and institution success. Because retention strongly impacts the institution, it makes sense to use predictive analytics to better understand retention
Why Predict Retention Risk? Faculty/staff can t accurately predict students perceptions of program or risk of retention Most faculty are focused on Research Course load Advising Most staff are focused on Putting out fires Working with students who self-identify issues Working with students with judicial issues Not a good strategy?
Why is Predicting Difficult? Many factors come into play. The impact of any one factor can be affected by other factors. Some factors are exacerbated by others. Some factors are mitigated by others. Time frame is long. Our judgment may be affected by: Anecdotes Outlier data points Issues we care about Institutional initiatives Other issues
Our best predictor is not a sum of issues It s fairly obvious that Travis Gatlin is at risk It s less obvious that Jessica Anderson, is equally at risk.
NATIONAL PREDICTIVE MODELS ACUHO-I/EBI Resident Assessment
Users of Predictive Modeling Administrators CHO Hall Staff Researchers Understand what drives success for improvement and accreditation review Identifies predictors for program to focus resources in these areas Identifies predictors for their area/hall (may be different from program) to focus work Understand the impact of housing on housing and institutional enrollments
Statistical Methods EBI s Reporting Research Automatically generated by institution Reported in written and online formats Uses linear multi-variant regression Descriptive Analysis Correlations Linear or Logistic Regression Discriminant Analysis Classification Trees Measurement and Path Models Neural Networks Simulation Models
Predicting Residence Hall Retention Factor Description Predictor Status Contribution to the Total Variance Factor Performance Overall Resident Satisfaction 1st Predictor 0.151 Moderate Overall Learning Experience 2nd Predictor 0.032 Moderate Satisfaction: Room Assignment or Change Process 3rd Predictor 0.031 Moderate Climate: Sense of Community 4th Predictor 0.024 Good Climate: Fellow Residents are Respectful 5th Predictor 0.020 Moderate NOTE: Data from 2011 ACUHO-I/EBI Resident Assessment, 271 institutions, 317,000 responses, R 2 =.173
Predicting Satisfaction Improving these predictors should improve resident satisfaction for future populations NOTE: Data from 2011 ACUHO-I/EBI Resident Assessment, 271 institutions, 317,000 responses
Predicting Learning Experience Improving these predictors should improve resident learning for future populations NOTE: Data from 2011 ACUHO-I/EBI Resident Assessment, 271 institutions, 317,000 responses
Limitations of a National Model Data definitions Using data definitions developed for a national model can make it difficult for some institutions to align with their internal data data Outlier Institutions Some institutions housing operations are outside the average thus making a national model less applicable Outlier Populations Some student populations are served by specialized institutions (e.g., HBCU) which is outside the average Solution: Institutional Models
INSTITUTIONAL PREDICTIVE MODEL University of Delaware
Enrollment Projections: The Variables New Students + Continuing Students Freshmen Transfers Readmitted Total number enrolled from prior semester minus (number of graduates + number of withdrawals)
Enrollment Projections: The Model Historical Data Predicted Data FALL 2009 FALL 2010 FALL 2011 History SPRING 2010 SPRING 2011
Enrollment Projections: The Output
Persisters Report TABLE 1: ENROLLMENT, DROPOUT RATES AND GRADUATION RATES FOR FIRST-TIME FRESHMEN ON THE NEWARK CAMPUS (Total) Enrollment and Dropout Rates Graduation Rates Entering 1st 2nd 3rd 4th 5th 6th within within within Fall Term Fall Fall Fall Fall Fall Fall 3 yrs 4 yrs 5 yrs Total 1995 N 3154 2673 2439 2355 599 113 21 1721 2219 2344 % enrollment 100.0% 84.7% 77.3% 75.3% 19.0% 3.6% 0.7% 54.6% 70.4% 74.3% % dropout 0.0% 15.3% 22.7% 24.7% 26.4% 26.1% 1996 N 3290 2804 2585 2489 606 108 22 1825 2302 2427 % enrollment 100.0% 85.2% 78.6% 76.3% 18.4% 3.3% 0.7% 55.5% 70.0% 73.8% % dropout 0.0% 14.8% 21.4% 23.7% 26.1% 26.7% 1997 N 3180 2766 2523 2436 581 117 27 1827 2284 2401 % enrollment 100.0% 87.0% 79.3% 77.5% 18.3% 3.7% 0.8% 57.5% 71.8% 75.5% % dropout 0.0% 13.0% 20.7% 22.5% 24.3% 24.5% 1998 N 3545 3080 2830 2762 653 118 22 2079 2621 2727 % enrollment 100.0% 86.9% 79.8% 78.5% 18.4% 3.3% 0.6% 58.6% 73.9% 76.9% % dropout 0.0% 13.1% 20.2% 21.5% 22.9% 22.7% 1999 N 3513 3126 2871 2757 526 83 31 2193 2632 2684 % enrollment 100.0% 89.0% 81.7% 79.4% 15.0% 2.4% 0.9% 62.4% 74.9% 76.4% % dropout 0.0% 11.0% 18.3% 20.6% 22.6% 22.7% 2000 N 3128 2738 2524 2453 496 85 24 1884 2297 -- % enrollment 100.0% 87.5% 80.7% 79.2% 15.9% 2.7% 0.8% 60.2% 73.4% % dropout 0.0% 12.5% 19.3% 20.8% 23.9% 23.8% 2005 2 nd Fall Retention Rate: 90.3% 2005 Cohort Graduation Rates: Within 4 yrs: 64.1% Within 5 yrs: 76.1% Within 6 yrs: 78.4% 2010 2 nd Fall Retention Rate: 92.5% 2001 N 3358 2976 2746 2674 472 0 31 2138 -- -- % enrollment 100.0% 88.6% 81.8% 80.6% 14.1% 0.0% 0.9% 63.7% % dropout 0.0% 11.4% 18.2% 19.4% 22.3% 0.0% 2002 N 3399 3055 2866 2787 0 0 42 -- -- -- % enrollment 100.0% 89.9% 84.3% 83.2% 0.0% 0.0% 1.2% % dropout 0.0% 10.1% 15.7% 16.8% 0.0% 0.0% 2003 N 3433 3035 2808 0 0 0 -- -- -- -- % enrollment 100.0% 88.4% 81.8% 0.0% 0.0% 0.0% % dropout 0.0% 11.6% 18.2% 0.0% 0.0% 0.0% 2004 N 3442 3064 0 0 0 0 -- -- -- -- % enrollment 100.0% 89.0% 0.0% 0.0% 0.0% 0.0% % dropout 0.0% 11.0% 0.0% 0.0% 0.0% 0.0%
Student Retention Benchmarks
Accuracy of Enrollment Projection Model
THE IMPACT OF RETENTION ON HOUSING ENROLLMENTS University of Delaware
The Issue 22
23 Why the concern? Impact of student outcomes and success Facilities under-utilized Loss of revenue
24 What is our Goal? Identify areas for improvement to help reverse the trend of second year students moving off-campus and reestablish acceptable occupancy rates.
25 Methodology Utilize existing information Student Extracts National Survey of Student Engagement (NSSE) Educational Benchmarking, Inc. (EBI) Undergraduate Student Satisfaction Survey Develop new instrument for further exploration Housing Retention Survey developed by UD administered by Campus Labs
26 EBI 2010 ACUHO-I/EBI Resident Study Overall Results 6.0000 5.5000 5.0000 4.5000 4.0000 3.5000 University Experience Residential Experience Overall Residential Value Source: 2010 ACUHO-I/EBI Resident Study
Impact on Overall Program Effectiveness 27 Which factors predict students overall perception of the Full Residential Experience? Regression Variables Performance Factor R 2 β Mean Descr. Learning Outcomes: Personal Interactions Top Predictor 0.663 0.302 5.29 Good Satisfaction: Room Assignment or Change Process 2nd Predictor 0.239 5.12 Good Satistaction: Room/Floor Environment 3rd Predictor 0.153 5.04 Good Satisfaction: Dining Services 4th Predictor 0.152 4.55 Good Learning Outcomes: Manage Time, Study, Solve Problems 5th Predictor 0.139 4.85 Good Climate: Sense of Community 6th Predictor 0.098 5.63 Excellent Satisfaction: Res Hall Student Staff 7th Predictor -0.039 5.76 Excellent Source: 2010 ACUHO-I/EBI Resident Study
28 Impact on Cost to Quality Rating Which factors predict students rating of the overall value of their residence hall experience? Regression Variables Performance Factor R 2 β Mean Descr. Satisfaction: Dining Services Top Predictor 0.372 0.303 4.58 Good Learning Outcomes: Manage Time, Study, Solve Problems 2nd Predictor 0.125 4.58 Good Satistaction: Room/Floor Environment 3rd Predictor 0.089 5.04 Good Learning Outcomes: Personal Interactions 4th Predictor 0.120 5.38 Good Satisfaction: Services Provided 5th Predictor 0.096 5.17 Good Climate: Sense of Community 6th Predictor 0.068 5.64 Excellent Satisfaction: Safety and Security 7th Predictor -0.058 5.83 Excellent Satisfaction: Room Assignment or Change Process 8th Predictor 0.048 5.12 Good Source: 2010 ACUHO-I/EBI Resident Study
29 Impact on Intent to Live On Campus Which factors predict students degree of their intent to live on campus the following year? Regression Variables Performance Factor R 2 β Mean Descr. Satisfaction: Dining Services Top Predictor 0.116 0.210 4.58 Good Learning Outcomes: Personal Interactions 2nd Predictor 0.175 5.38 Good Satisfaction: Facilities 3rd Predictor -0.085 5.14 Good Satisfaction: Safety and Security 4th Predictor 0.088 5.83 Excellent Learning Outcomes: Diverse Interactions 5th Predictor 0.065 5.36 Good Climate: Fellow Residents are Respectful 6th Predictor -0.065 5.27 Good Source: 2010 ACUHO-I/EBI Resident Study
30 What does it all mean? Where should we focus our efforts? What areas do we simply monitor? What further questions do we ask?
31 Next Steps Establish student needs Discuss results with our Stakeholders Dining, Residence Life, Facilities, etc. Investigate and Communicate perceived value and student satisfaction factors Focus Groups for in-depth discussion Enhance marketing efforts to parents
32 Changes to Date & Future Recommendations New marketing messages Approval to purchase new Housing Administration System
33 Changes to Date & Future Recommendations Develop Resident Life Living Learning Communities (LLC)
34 Changes to Date & Future Recommendations Rodney Complex facility improvements
35 Changes to Date & Future Recommendations East Campus Freshmen Residence Complex and Dining Hall Projects Planning and Recommendations Source: New dorms to open in 2013, The Review, February 22, 2011.
FINAL THOUGHTS
Predicting Retention Group behaviors are easier to predict Individual student behaviors are impossible to predict Student decisions strongly impact departmental budgets Limitations National data doesn t factor in institutional characteristics Departmental/institutional data may not have large enough populations Accuracy of our predictive models is important, but we must also focus on getting people to use the data. Don t forget this is about informing and motivating behavior
Want to Learn More? Visit the EBI / MAP-Works booth in the exhibit hall Attend one of EBI s free educational webinars (www.webebi.com/community) Heather Kelly, Ed.D., hkelly@udel.edu Karen DeMonte, M.Ed., kdemonte@udel.edu Darlena Jones, Ph.D., Darlena@webebi.com