Conference presentation from the Texas Association of Graduate Admissions Professionals (TxGAP) 2012 Professional Development Conference.
Author:
Jeancarlo Bonilla
Director of Graduate Enrollment Management
Polytechnic Institute of New York University
Description:
Learn how to use predictive modeling techniques and apply them to the area of graduate enrollment management.
For more information, visit www.txgap.com.
Running the Numbers: Improving Your Position for Enrollment Planning and Forecasting - Jeancarlo Bonilla
1. Running the Numbers: Improving Your
Position for Enrollment Planning and
Forecasting”
TxGAP – Summer Conference
July 20th, 2012
JeanCarlo (J.C) Bonilla
Director of Graduate Enrollment Management
New York University, Polytechnic Institute
2. The Plan for the Session:
• Overview of predictive modeling & optimization for enrollment
management
• Case 1: Do cycles have memory? The case of the 3-yr adjusted yield (4
examples)
• Case 2: Am I making my class? Modeling for scenarios forecasting (3
examples)
• Case 3: The magic ball, ranking and an enrollment predictor (2 examples)
• Case 4: Opps, I ran out of time, but this is a very cool model
Worksheets download at EnrollmentAnalytics.com
5. Where is the industry today with the
idea of business analytics &
intelligence?
6. Degree of intelligence
Standard
Ad
hoc
Report
Query
Alert
Reports
“how
many,
how
“what
is
exactly
“what
actions
are
“what
happened”
often,
where”
the
problem”
required”
Descriptive Analytics
7. Degree of intelligence
Statistical
Randomized
Predictive
Optimization
Model
testing
Model/Forecast
“what
is
the
best
“why
is
this
“what
happens
if
“what
will
that
can
happening”
we
try
this”
happened
next”
happened”
Predictive Analytics
9. So, what do we know so far about
predictive modeling for Enrollment
Management?
10. Ad
hoc
Report
Standard
Reports
“how
many,
how
“what
happened”
often,
where”
SUSPECTS> PROSPECTIVE> APPLICANTS> ADMITS> DEPOSITS> NEW
Query
Alert
TACTIC “what
is
exactly
the
“what
actions
are
problem”
required”
11. Predictive Analytics
Predict
&
Statistical
Model
Random
Testing
Optimization
Forecast
“why
is
this
“what
happens
if
we
“what
is
the
best
that
happening”
try
this”
“what
will
happened
can
happened”
next”
SUSPECTS> PROSPECTIVE> APPLICANTS> ADMITS> DEPOSITS> NEW
TACTIC
12. Examples of Enrollment Predictive Modeling
• Case 1: North Dakota University
– Type of Model: inquiry model using geo-demographic
– Predictive Power: 36% of students who will enroll &
97% of student who will not enrolled
13. Examples of Enrollment Predictive Modeling
• Case 2: University of Minnesota
– Type of Model: application generation model using, ACT and geo-
demographic information
– Predictive Power: 85% of applicants to a “large
research university” are from within the same state
or form a neighboring state
14. Examples of Enrollment Predictive Modeling:
• Case 3: State University of New York
– Type of Model: lead modeling using geo-demographic, academic data,
and financial aid data
– Predictive Power: 45.67% of applicants predicted to enroll
did in fact matriculate and 82.16% who where predicted not
to enroll did not matriculate
15. Better predictive power with students who
do not matriculate than with model that forecast actual
students enrollments
16. The “technique” is used in other consolidated markets...
if it works for them, it should work for us!
17. It requires quantitative analysis of past
student characteristics to predict probabilities of
future results
18. Your predictive modeling team should have
people who are confortable doing:
The modeling guy:
1. Regression Analysis (logistic regression)
2. Business analytics
The computer guy:
1. Database architecture & design
2. Database querying
3. Data aggregation & integration
4. Data reporting
30. CASE#1:
Do cycles have memory? The case of the 3-
yr adjusted yield
Predictions through the admissions funnel
Download worksheets at: www.EnrollmentAnalytics.com
31. Recommendations
1. Rapid “back-of-the-envelop” modeling
2. You can go “up” or “down” the funnel
3. Need for historical data (static snapshots of cycles)
4. Student characteristics add more resolution to the model
5. Use of adjusted 3-year cycles are useful for historical
modeling
6. Historical validity: account for new initiatives
32. CASE#2:
Am I making my class?
New Student forecaster
Download worksheets at: www.EnrollmentAnalytics.com
33. CASE#3:
The magic ball
ranking and an enrollment predictor (2 examples)
Download worksheets at: www.EnrollmentAnalytics.com
35. Example: say that you have 20k leads in your
cycle and only 300 matriculate, then you have a
2% conversion rate
36. Now, you “observe” that 200 out of your 300
new students presents a subset of 5% of your
prospective students pool.
This means that 1000 prospective students (5%
of 20k) converted into 200 enrollments, which
means that your conversion rate for this subset is
20%
37. 300 new students
33%
of 20 %
on rate or p redictability 67%
5%
New conver
si
95%
20,000 prospective students
38. FT-‐Dom,
FT-‐Dom,
5%
20%
FT-‐Int'l,
40%
PT,
40%
FT-‐Int'l,
55%
PT,
40%
300 new students
20,000 prospective students
39. ..and if you get really good at
understanding your students...
40. Student
Name
Status
Predictor
Inquiry
11/16/10
App.
N/A
Hall,
Joy
Adm.
N/A
0.4
Conf.
N/A
Enr.
N/A
Inquiry
12/22/10
App.
12/24/10
Li,
Xiao
Adm.
3/23/11
0.6
Conf.
4/2/11
Enr.
N/A
Inquiry
12/5/10
Build a model that Lopez,
Jose
App.
Adm.
Conf.
1/5/11
1/29/11
3/16/11
0.2
does the following:
Enr.
Inquiry
App.
N/A
12/20/10
2/3/11
Mitchell,
Tamara
Adm.
N/A
0.2
Conf.
N/A
Enr.
N/A
Inquiry
1/26/11
App.
1/28/11
Smith,
John
Adm.
4/16/11
0.4
Conf.
5/5/11
Enr.
N/A
Inquiry
12/13/10
App.
N/A
Troy,
Bryan
Adm.
N/A
0.9
Conf.
N/A
Enr.
N/A
41. So, how can I build a model like that
predicts enrollments?
42. Student Uncertainty & Variance
Behavior &
Personal Life Academic
Demographi
Financial
cal
Geographic
al
FT vs intl vs PT
43. Recommendations for Advanced Models
1. It gets complicated
2. Its “easy” to model for student characteristics, but
complexity increases when accounting for student behavior
3. Models are better at predicting student who do NOT register
4. Every school is different, so every model is also different
5. Trust your instincts! No one knows students better than
you... Your job is then trying to articulate and generalized
characteristics and behavior
44. CASE#4:
Opps, I ran out of time, but this is a very
cool model
Download worksheets at: www.EnrollmentAnalytics.com
45. Final Recommendations
1. Plan for good, bad, and what you think is going to realistic
2. Avoid predictions but give options
3. Its about resource allocation
4. Work with other groups in your institution
5. Trust your GEM instincts
6. Its earsier to account for student characteristics, but
modeling and forecasting behavior is very complex