Estimating New Freshmen Enrollment

1
Estimating New Freshmen Enrollment

• Agatha Awuah, Eric Kimmelman, Michael Dillon
• Office of Institutional Research
• Binghamton University
• AIRPO
• June 11-13, 2003

2
Admissions Process

• Set new freshmen targets.
• Make offers of admission.
• Build wait list.
• Collect deposits.
• Estimate enrollment based on deposits received.
• Make offers to the wait list if needed.

3
Previous Method

• Required to estimate enrollment
• Yieldlast years enrollment (1,000) divided by
  last years offers (3,000).
• Est. Yield1,000/3,000
• .33
• Target for current year (2,000).
• Est. Offers Needed2,000/.33
• 6,000

4
Previous Method-Results
5
Yield by SAT Score-Fall 2002
SAT Score Admits Enrolled Yield
LE 1150 1433 543 37.89
1160-1230 1623 481 29.64
1240- 1280 1321 356 26.95
1290-1360 1648 325 19.72
GE 1370 1454 194 13.34
Total 7479 1899 25.39
6
Logistic Regression

• Dichotomous dependent variable.
• Estimates conditional probability of enrollment
  controlling for multiple independent
  variables-yield.
• Available in most statistical packages.

7
The Data

• Five fall semesters -1998 to 2002.
• Only matric freshmen admits (35,796) included.
• Enrollment of admitted applicants 9,811.
• Yield rate (9,811/35,796)10027.4.

8
Steps to Building Model 1

• Estimate baseline model using 5 years of data
  (intercept only), estimate enrollment, then
  calculate absolute prediction error by semester.
• Add additional variables and calculate new
  absolute prediction error.

9
Steps to Building Model 2

• Compare prediction errors. If the second
  prediction error is smaller than the first, keep
  new variable in the model. If not, drop it from
  the model.
• Continue process until smallest possible
  prediction error is attained.
• Predict enrollment for each year in the sample
  with data from other 4 years.

10
Step One-Baseline Model
Year Offers Est. Enr. Act. Enr. Abs. Diff.
1998 7004 1920 1909 11
1999 6765 1854 1943 89
2000 6761 1853 1834 19
2001 7787 2134 2226 92
2002 7479 2049 1899 151
Total 361
11
Step Two-Add SAT and HS Avg. 1
Variable Est. Coeff. Std. Dev Chi Sqr. Pr. gt Chi Sqr.
Intercept 8.730 0.295 878.335 0.001
SAT -0.003 0.000 1157.089 0.001
HS Avg. -0.061 0.003 322.130 0.001
12
Step Two-Add SAT and HS Avg. 2
Year Offers Est. Enr. Est. Yield Act. Enr. Act. Yield
1998 7004 1925 27.48 1909 27.26
1999 6765 1890 27.94 1943 28.72
2000 6761 1854 27.42 1834 27.13
2001 7787 2171 27.88 2226 28.59
2002 7479 1971 26.36 1899 25.39
13
Step Two-Add SAT and HS Avg. 3
Year Offers Est. Enr. Act. Enr. Abs. Diff.
1998 7004 1925 1909 16
1999 6765 1890 1943 53
2000 6761 1854 1834 20
2001 7787 2171 2226 55
2002 7479 1971 1899 72
Total 216
14
Full Model 1-Academics
15
Full Model 2-Inqs/Demo
16
Full Model 3-Inst.
17
Full Model Performance
Year Pred. Enr Low 95 High 95 Act. Enr. - Admits Pred. Error
1998 1880 1807 1952 1909 29
1999 1919 1846 1991 1943 24
2000 1861 1789 1933 1834 27
2001 2191 2113 2268 2226 35
2002 1961 1886 2036 1899 62
Total 177
18
Full Model Evaluation
Year Pred. Enr Low 95 High 95 Act. Enr. Diff.
1998 1872 1799 1944 1909 37
1999 1910 1837 1983 1943 33
2000 1870 1798 1942 1834 36
2001 2183 2104 2260 2226 43
2002 1974 1900 2049 1899 75
Total 221
19
Estimating Quality of Regular Admits Fall 2002
Estimated Actual Prediction Error
Mean SAT Score 1231 1238 -7
Mean HS Average 92 92 0
20
Additional Applications

• Predict retention.
• Identify Hot Prospects.
• Identify potential donors.
• Evaluate recruitment efforts.

21
Logistic Regression

• Berge, D.A. Hendel, D.D. (2003, Winter).
  Using Logistic Regression to Guide Enrollment
  Management at a Public Regional University. AIR
  Professional File, 1-11.
• Thomas, E, Dawes, W. Reznik, G. (2001,
  Winter). Using Predictive Modeling to Target
  Student Recruitment Theory and Practice. AIR
  Professional File, 1-8.
• Aldrich, J.H. Nelson, F.D. (1984). Linear
  Probability, Logit and Probit Models. Sage
  University Papers Quantitative Applications in
  the Social Sciences, 07-045. Newbury Park, CA
  Sage Publications

22
Estimating New Freshmen Enrollment

• Agatha Awuah, Eric Kimmelman, Michael Dillon
• Office of Institutional Research
• Binghamton University
• AIRPO
• June 11-13, 2003
• Website http//buoir.binghamton.edu