Enrollment Management Predictive Modeling Simplified
description
Transcript of Enrollment Management Predictive Modeling Simplified
![Page 1: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/1.jpg)
Enrollment Management Predictive Modeling Simplified
Vince Timbers, Penn State University
![Page 2: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/2.jpg)
Overview
• Common Enrollment Management Uses
• Basic Principles of Predictive Modeling
• Penn State Predictive Models
![Page 3: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/3.jpg)
What is Predictive Modeling?
• Predicting future behavior of a population based on the past behavior of a similar population
![Page 4: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/4.jpg)
Common Uses of Predictive Modeling in Enrollment Management
• Retention projections
• Applicant enrollment projections
• Accepted student enrollment projections
• Suspect/prospect application projections• Recruitment and retention strategies and activities
• Budget and resource planning
![Page 5: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/5.jpg)
Predictive Modeling Basics• Past behavior is a good predictor of future
behavior
• Similar groups tend to behave in a similar manner, under similar circumstances
• Model effectiveness depends on the ability to identify similar groups and similar circumstances
• Always test new models on historic data
![Page 6: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/6.jpg)
Model Building Steps• Identify what is being predicted
• Identify the population
• Identify predictors
• Select data sources
• Select a modeling technique
• Build and Test - Rebuild and Retest
![Page 7: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/7.jpg)
Penn State Projection Models
• Retention Projections
• Accepted Student to Enrollment Projections
• Accepted Student Probability of Enrollment
• Paid Deposit to Enrollment Projections
![Page 8: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/8.jpg)
Retention Projections• Retention
• Enrolled students
• College, semester standing
• Official enrollment data
• Contingency table approach
• Build and Test - Rebuild and Retest
![Page 9: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/9.jpg)
Retention ProjectionsContingency Table Approach
• Aggregated prior data to the appropriate level
• Calculate retention rates
• Aggregated current data to the appropriate level
• Apply prior retention rates to current data to calculate the retention projection
![Page 10: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/10.jpg)
University Park Retention ProjectionsCollege Semester
StandingFall 2010Enrolled
Fall 2010RetainedTo Fall 2011
Retention Rate
Fall 2011Enrolled
Projected 2012Retention
AG 01 176 154 87.50% 172 150.50
AG 02 38 31 81.58% 45 36.71
AG 03 220 209 95.00% 200 190.00
AG 04 170 140 82.35% 197 162.24
AG 05 279 259 92.83% 352 326.77
AG 06 174 131 75.29% 178 134.01
AG 07 230 74 32.17% 255 82.04
AG 08 140 18 12.86% 152 19.54
AG 09 61 8 13.11% 76 9.97
AG 10 25 1 4.00% 19 0.76
AG 11 9 2 22.22% 8 1.78
1,114.32
![Page 11: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/11.jpg)
Retention Projection Results
University Park Retention
• 2011 Projection 24,662• 2011 Actual 24,761• Under Projected .5%
• 2012 Projection 24,851• 2012 Actual 25,046• Under Projected .8%
Change of Campus to University Park
• 2011 Projection 3,617• 2011 Actual 3,540• Over Projected 2.1%
• 2012 Projection 3,459• 2012 Actual 3,380• Over Projected 2.3%
![Page 12: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/12.jpg)
Accepted Student Enrollment Projections (Contingency Table)
Model Variables
• Semester
• Application Pool
• Residency
• College Group
• Academic Performance
![Page 13: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/13.jpg)
Accepted Student Probability of Enrollment
Logistic Regression
• Explain the relationship between a discrete outcome (enrollment) and a set of explanatory variables
• Logistic Regression produces a set of coefficients (model) used to predict the outcome (enrollment) for similar populations
![Page 14: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/14.jpg)
Probability of Enrollment (Logistic Regression)
• logit= 0+ 1*X1 + 2*X2…… + k*Xk
InterceptApp Date Out of State
HS GPA Verbal Math Writing Predicted PSU GPA
Age Logit Probability
Variable Coefficient 2.13396 -0.00687 -1.14124 -0.24361 -0.00115 -0.00006 -0.00321 0.26485 0.04767
Value 30 0 3.0 700 700 700 3.0 18
2.13396 -0.2061 0 -0.73083 -0.805 -0.042 -2.247 0.79455 0.85806 -0.244360.439212
2
![Page 15: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/15.jpg)
Probability of Enrollment Results (Logistic Regression)
Probability Range
Accepted Paid Deposit Yield
0 (0 - .049) 2852 112 3.930.1 (.05 - .149) 7096 718 10.120.2 3475 713 20.520.3 2192 662 30.20.4 1620 638 39.380.5 1219 610 50.040.6 1051 608 57.850.7 943 671 71.160.8 789 602 76.30.9 580 486 83.791 88 85 96.59
![Page 16: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/16.jpg)
Probability of Enrollment Results (Logistic Regression)
Probability Range
Accepted Paid Deposit Yield
0 (0 - .049) 2852 112 50.1 (.05 - .149) 7096 718 150.2 3475 713 250.3 2192 662 350.4 1620 638 450.5 1219 610 550.6 1051 608 650.7 943 671 750.8 789 602 850.9 580 486 951 88 85 96.59
![Page 17: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/17.jpg)
Paid Deposit to Enrollment Projections
Model Variables (Contingency Table Approach)
• Semester
• Residency
• Placement test completion
![Page 18: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/18.jpg)
Fall 2012 Paid to Enrollment ResultsAs of 5/15/2012
Without Test Completion in Model
• Deposited 8,415• Projected 7,640• Actual 7,574• Difference +59
With Test Completion In Model
• Deposited 8,415• Projected 7,570• Actual 7,574• Difference -4
Test completion=78%
![Page 19: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/19.jpg)
Paid Deposit to Enrollment ResultsAs of 5/29/2012
Without Test Completion in Model
• Deposited 8,342• Projected 7,625• Actual 7,590• Difference +35
With Test Completion In Model
• Deposited 8,342• Projected 7,486• Actual 7,590• Difference -104
• Test completion=88%
![Page 20: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/20.jpg)
Paid Deposit to Enrollment ResultsAs of 7/31/2012
Without Test Completion in Model
• Deposited 8,098• Projected 7,619• Actual 7,632• Difference -47
With Test Completion In Model
• Deposited 8,098• Projected 7,431• Actual 7,632• Difference -201
• Test completion=96%
![Page 21: Enrollment Management Predictive Modeling Simplified](https://reader036.fdocuments.in/reader036/viewer/2022081513/568161b7550346895dd183ae/html5/thumbnails/21.jpg)
Model Building Steps• Identify what is being predicted
• Identify the population
• Identify predictors
• Select data sources
• Select a modeling technique
• Build and Test - Rebuild and Retest