Objective
Determine what variables, if any, are most closely correlated with and predictive of admissions yield of universities within the Integrated Postsecondary Education Data System (IPEDS) data from the Institute of Education Sciences National Center for Educational Statistics.
1. Objective
Determine what variables, if any, are most closely correlated with and predictive of admissions
yield of universities within the Integrated Postsecondary Education Data System (IPEDS) data
(available here) from the Institute of Education Sciences National Center for Educational
Statistics.
The Integrated Postsecondary Education Data System (IPEDS), conducted by the NCES, began in
1986 and involves annual institution-level data collections. All postsecondary institutions that
have a Program Participation Agreement with the Office of Postsecondary Education (OPE) of the
U.S. Department of Education are required to report data using a web-based data collection
system.
Predictors of the Admissions Yield
2. Why Admission Yield Matters
• Universities that don’t enjoy high demand have to worry about their admissions yield
policy. While some universities have the prowess to admit as few as 5 percent of
applicants, even these universities must consider admissions yield before they can
set such low admissions standards. Even elite universities have to balance the books
and enrollment is usually the largest portion of what pays the bills. For the thousands
of universities without demand high enough to decline 90+% of applicants yearly,
knowing the predictors of admissions yield are especially important.
University Admissions Yield Analysis
3. Briefly discuss the predictive
models used in your project.
Predictive Models Used (SAS):
• Normal Linear Regression
• Linear Regression (Stepwise)
• Neural Network using Variable Selection
• Neural Network on Stepwise results
• Decision Tree
• Neural Network with no variable
selection
• AutoNeural
University Admissions Yield Analysis
4. Comparing predictive models
Using
Average
Squared Error
as the
comparison
measure, the
most accurate
model for
predicting the
Admission
Yield was the
Ensemble
model.
University Admissions Yield Analysis
5. Decision Tree Results
After the Ensemble model, the decision tree is the #2 most important model for its ASE level
Decision tree variables in order of importance:
The #1 & #2 most powerful variables were
1. “Total price for in-state students living on campus 2013-14”
2. “Percent of freshmen receiving federal student loans”
University Admissions Yield Analysis
6. Decision Tree Continued:
The most powerful decision tree variable was
“Total price for in-state students living on campus 2013-14”
Values greater than $37,695 resulted in an average Admission Yield value of 26.28%
Values less than $37,695 (or missing) resulted in an average Admission Yield value of 38.09%
University Admissions Yield Analysis
7. Predictive models used
The negative &
positive
relationships in the
stepwise regression
variables werevery
interestingto
observe. For
instance, for every
$8 decrease in
in-state tuition
there was a 1
percentage point
increase in
admission yield
University Admissions Yield Analysis
8. Findings
• “Total price for in-state students living on campus 2013-14” was the most powerful
predictor of admissions yield across multiple models.
• “Percent of freshmen receiving federal student loans was the 2nd most powerful
predictor.
Universities with a lower than desired admissions yield improve yield by focusing on:
By decreasing their in-state tuition by $8, Universities can gain a 1 percentage point
increase in admission yield.
• Decreasing the “total price for in-state students living on campus”
• Boosting the “percent of freshmen receiving federal student loans”
University Admissions Yield Analysis