You are on page 1of 39

Multilevel Model for Predicting Future Enrollment and FTEs

By Ron Polland Assistant Director Office of Institutional Research

The Basic Enrollment Model


The three primary functions of an enrollment model are to predict the number of new students who will be entering UNF, to predict the number of current students who will return to UNF and to predict the number of current students who will not return to UNF.

Primary Sources of New Students


Prior High School Graduates

Florida Community colleges

grad

HS enrollment

other colleges

Current High School graduates

Other transfers

FL CC transfers w/o AA

FL CC transfers with AA

The Application and Admission Process


Current High School graduates Other transfers FL CC transfers w/o AA FL CC transfers with AA

number of applications

number of applications

number of applications

number of applications

# admitted

# admitted

# admitted

# admitted

The Enrollment Process (undergraduates only)


number of applications

number of applications

number of applications

number of applications

# admitted

# admitted

# admitted

# admitted

new freshmen

new sophomores

new juniors

new seniors

Retention and Class Progression


new freshmen new sophomores new juniors new seniors

retained freshman

retained sophs

retained juniors

retained seniors

grad

drop

Predicting Future Trends from Historical Data


The starting point for this prediction model was an examination of enrollment and FTE data at UNF for the previous ten academic years. The overall growth in enrollment has been fairly stable at a rate of about 3 percent per year.

Overall Growth In Enrollment


FALL HEADCOUNTS (1990 - 2001)
14,000

13,000

12,000
HEADCOUNTS

11,000

10,000

9,000

8,000
1990 1991 1992 1993 1994 1995 YEAR 1996 1997 1998 1999 2000 2001

Enrollment Trends by Student Class


While the growth in total enrollment can be represented by a simple, straight line, such is not the case for the different classes of students. The numbers of new juniors and post-bacs who enrolled at UNF, and the numbers of juniors and seniors who remained at UNF varied greatly over the past ten years.

Composition of Students
FIGURE 2: HEADCOUNT DISTRIBUTION BY CLASS Fall, 1995 to Fall, 1999 199508 11.6% 7.6% 28.3% 21.6% 14.0% 11.9% 4.4%

199608

10.4%

8.0%

23.7%

28.1%

12.6%

11.9%

4.6%

Term

199708

13.8%

7.9%

26.4%

22.8%

12.3%

10.7%

5.5%

199808

15.4%

9.7%

26.8%

20.8%

12.2%

9.7%

4.5%

199908

17.1%

10.6%

29.6%

17.3%

11.1%

9.5%

4.1%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Percent of Total Headcount

Freshman

Sophomore

Junior

Senior

Post-Bacs

Masters

PhDs

Special Students

Changes in FTE Distribution by Class Level


Changes in the distribution of enrollment by class is also reflected in changes in the distribution of total FTEs by class.

The proportion of FTEs attributable to juniors, seniors, and post-bacs has steadily declined over the past five years.

Percent of Annual FTEs by Student Class


PERCENT OF ANNUAL FTE BY CLASS 1996/97 to 2000/01
100% 90% 80% 70% 60% 50% 32.3% 40% 30% 20% 9.6% 10% 10.5% 0% 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 14.6% 14.0% 17.3% 18.2% 24.1% 12.8% 10.2% 12.5% 12.7% 29.6% 29.6% 34.6% 2.3% 0.6% 10.6% 7.6% 2.5% 0.5% 9.2% 7.5% 2.1% 0.6% 8.6% 7.2% 1.8% 0.5% 8.8% 6.4% 1.8% 0.5% 10.1% 5.1% 18.9% SPECIAL DOCTORAL MASTERS POST-BAC 32.6% SENIOR JUNIOR SOPHOMORE FRESHMAN

20.2% 26.0% 25.5%

Changes in FTE Distribution by Course Level


Changes in the distribution of enrollment by class level are also reflected in changes in the distribution of FTEs by course level (i.e., LOWER, UPPER, GRAD I, GRAD II). For this reason, we must look at how FTEs of different course levels are partitioned across different class levels.

FTEs by Course Level and Student Level


LEVEL2 LOWER CLASS FRESHMAN SOPHOMORE JUNIOR SENIOR POST-BAC MASTERS SPECIAL LOWER Total UPPER FRESHMAN SOPHOMORE JUNIOR SENIOR POST-BAC MASTERS SPECIAL UPPER Total GRAD I FRESHMAN SOPHOMORE JUNIOR SENIOR POST-BAC MASTERS SPECIAL GRAD I Total GRAD II DOCTORAL GRAD II Total Grand Total 1996/1997 1997/1998 1998/1999 1999/2000 2000/2001 635.2 908.8 912.7 1160.7 1279.9 522.4 542.6 702.6 700.8 733.2 489.3 470.1 550.7 510.4 537.2 245.5 195.4 159.3 115.0 106.9 66.8 77.4 74.8 71.8 53.5 11.1 10.7 9.4 6.8 4.8 65.1 91.3 85.6 88.9 91.0 2035.4 2296.4 2495.1 2654.4 2806.4 17.4 26.0 26.2 46.1 54.1 72.6 111.2 132.2 188.4 199.1 1008.0 1428.2 1430.3 1737.9 1855.7 1892.7 1467.4 1542.1 1281.2 1270.5 286.3 273.9 251.6 221.0 181.7 52.7 52.3 44.0 38.1 40.7 70.4 64.0 50.2 37.3 42.9 3400.0 3422.9 3476.5 3550.0 3644.7 0.2 0.2 0.2 0.4 0.4 0.4 0.2 9.8 6.5 5.1 11.4 7.7 120.4 131.4 155.2 149.8 136.7 590.7 525.6 524.0 567.3 694.3 8.0 5.7 3.1 0.8 1.0 729.0 669.6 687.9 729.7 840.1 36.2 34.6 37.2 31.5 38.4 36.2 34.6 37.2 31.5 38.4 6200.7 6423.5 6696.6 6965.5 7329.6

Relationship between Headcount and FTE


Since FTEs are generated by student headcounts, there is a direct, linear relationship between headcount and FTE. This linear, or straight-line, relationship has remained fairly stable and consistent over time such that it would be fairly easy to estimate FTEs from headcounts.

Relationship Between Fall Headcounts and Annual FTEs


(1996/1997 2000/2001 )
FALL HEADCOUNTS BY ANNUAL FTES
17000 16000 15000
HEA DCOUNTS

14000 13000 12000 11000 10000 9000 8000 7329.689 7575.797 7935.416 8291.511 FTEs 8665.613 9036.677 9406.033

FTEs Grouped by Class Level and Student Level


Because FTE allocations vary among class levels and course levels, a one size fits all formula for predicting future FTEs will not suffice. A more accurate approach is to generate formulas that represent each of the different historical trends.

Each FTE group has a different pattern over time:


ANNUAL LOWER FTES BY STUDENT CLASS (1996/1997 - 2000/2001)
1400.0 100.0 90.0 1200.0 80.0 1000.0 70.0 60.0 50.0 600.0 40.0 30.0 20.0 200.0 10.0 0.0 1996/1997 1997/1998 1998/1999 YEAR 1999/2000 2000/2001 0.0

FRESHMAN

HEADCOUNTS

800.0

HEADCOUNTS

SOPHOMORE

JUNIOR

SENIOR

POST-BAC

MASTERS

400.0

SPECIAL

Each FTE group has a different pattern over time:


ANNUAL UPPER FTES BY STUDENT CLASS (1996/1997 - 2000/2001)
2000.0 1800.0 1600.0 60.0 1400.0
SOPHOMORE

80.0 70.0

HEADCOUNTS

1200.0 1000.0 800.0 600.0

HEADCOUNTS

50.0 40.0 30.0 20.0

JUNIOR
SENIOR
POST-BAC
FRESHMAN
MASTERS
SPECIAL

400.0 200.0 0.0 1996/1997 1997/1998 1998/1999 YEAR 1999/2000 2000/2001 10.0 0.0

Automated Curve Fitting Program


Because there are so many different patterns in the data some of which are non-linear finding the proper mathematical formula to represent each of them would be a daunting task, if not for the availability of a new, curve-fitting program called Table Curve 2D.

Table Curve 2D
Table Curve 2D, produced by the makers of SPSS statistical software, analyzes a users data set and generates thousands of possible mathematical solutions in a matter of seconds. Then, the user can evaluate these solutions according to how well they model or fit the data.

Table Curve 2D
Table Curve 2D uses a procedure called least squares to derive a single function or formula that describes the data with minimum error. The process is similar to linear regression which finds a single, straight line that best represents the relationship between two sets of data.

Linear Regression
An example of a linear model to predict FTEs from headcounts:

y = 0 + 1x +
y = FTEs x = HEADCOUNTS 0 = y-intercept 1 = slope of the line = error variable
y
0 and 1 are unknown, therefore, are estimated from the data.

Rise 0 Run 1 = Rise/Run

Finding the Best-Fitting Straight Line


Q: Which straight line fits the data best? A: The one that minimizes the distances between the data points and the lines x

Finding the Best-Fitting Curved Line


As previously shown, much of the FTE and headcount data cannot be adequately represented by linear, or straight lines. They can, however, be represented by nonlinear, curved lines.

Table Curve 2D fits a curved line to each FTE grouping:


FRESHM AN - LO W ER
Rank 3 Eqn 2425 y=a+bx^(1.5)+ce^x+d(lnx)^2
r^2=0.99916919 DF Adj r^2=0.99667674 FitStdErr=14.443221 Fstat=400.8792 a=-253.13307 b=909.60992 c=-9.6162787 d=-2782.4683
1300
1300

1200

1200

1100

1100

1000

1000

FTE

900

900

800

800

700

700

600

600

YEAR

FTE

Table Curve 2D fits a curved line to each FTE grouping...


SENIO R - LO W ER
Rank 2 Eqn 2206 y=a+bxlnx+cx^(1.5)+dlnx/x^2
r^2=0.99999683 DF Adj r^2=0.99998731 FitStdErr=0.20553739 Fstat=105075.91 a=-567.70308 b=-1041.7348 c=813.22139 d=-535.52173 250 250

225

225

200

200

FTE

175

175

150

150

125

125

100 1 2 3 4 5

100

YEAR

FTE

...AND then predicts what the future FTE values will be...

Predicted Lower Level FTE for Seniors, 2001/2002

Predicted FTEs and Headcounts are then used to calculate a FTE/Headcount ratio for each class...
Projected FTEs 2001-2002 584.165 829.463 818.017 552.815 111.237 41.930 3.749 0.000 90.382 Projected Headcounts 2001-2002 Projected FTE ratios 2001-2002

First-Time Freshmen Continuing Freshmen Sophomore Junior Senior Post-baccalaureate Masters Doctors Special

First-Time Freshmen Continuing Freshmen Sophomore Junior Senior Post-baccalaureate Masters Doctors Special

1,558 957 1,528 3,971 2,211 1,059 1,430 108 315

First-Time Freshmen Continuing Freshmen Sophomore Junior Senior Post-baccalaureate Masters Doctors Special

0.375 0.867 0.535 0.139 0.050 0.040 0.003 0.000 0.287

...and create a matrix for all student classes and levels:


LEVEL CLASS CONT. FRESHMAN SOPHOMORE JUNIOR SENIOR POSTBAC MASTERS DOCTORAL SPECIAL NONDEGREE 0.287 0.007 0.059 0.155 0.478 0.568 0.134 0.026 0.000 0.128 0.390 0.008 0.072 0.160 0.480 0.529 0.136 0.028 0.000 0.164 0.531 0.008 0.088 0.163 0.479 0.496 0.136 0.031 0.000 0.209 0.714 0.009 0.108 0.165 0.476 0.467 0.140 0.036 0.000 0.264 0.953 0.010 0.132 0.167 0.471 0.442 0.125 0.042 0.000 0.329 1.266 0.011 0.164 0.167 0.464 0.418 0.104 0.050 0.000 0.406 2001-2002 2002-2003 2003-2004 2004-2005 2005-2006 2006-2007 0.375 0.867 0.535 0.139 0.050 0.040 0.003 0.369 0.884 0.521 0.137 0.056 0.040 0.002 0.367 0.886 0.505 0.135 0.057 0.040 0.000 0.365 0.887 0.491 0.133 0.059 0.041 0.001 0.361 0.885 0.478 0.130 0.061 0.037 0.002 0.358 0.882 0.466 0.128 0.063 0.031 0.004

LOWER FT FRESHMAN

UPPER FT FRESHMAN
CONT. FRESHMAN SOPHOMORE JUNIOR SENIOR POSTBAC MASTERS DOCTORAL SPECIAL NONDEGREE

GRAD I CONT. FRESHMAN


SOPHOMORE JUNIOR SENIOR POSTBAC MASTERS DOCTORAL SPECIAL NONDEGREE GRAD II DOCTORAL 0.004 0.145 0.458 0.000 0.003 0.000 0.418 0.004 0.168 0.496 0.000 0.003 0.000 0.422 0.003 0.193 0.537 0.000 0.005 0.000 0.425 0.003 0.221 0.580 0.000 0.006 0.000 0.427 0.003 0.251 0.624 0.000 0.008 0.000 0.428 0.003 0.289 0.652 0.000 0.010 0.000 0.428

These ratios are then used to relate headcounts to FTEs


Projected F E all nrollment

Number of Students

15,000 10,000 5,000

13,137

13,573

14,059

14,598

15,190

15,831

P rojected Annual F s TE 10,000 9,000 51.8 879.1 55.2 909.3 58.5 939.7 61.9 952.9

2001

2002

2003

2004

2005

2006
FTEs

8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0

45.2 819.4

48.5 849.0

3679.5

3770.2

3865.7

3973.2

4080.2

4202.8

3031.8

3267.7

3494.8

3727.9

3958.2

4188.5

2001-2002

2002-2003

2003-2004 LOW ER

2004-2005
GRADI

2005-2006
GRADII

2006-2007

UPPER

Predicting Retention and Class Progression


new freshmen new sophomores new juniors new seniors

retained freshman

retained sophs

retained juniors

retained seniors

grad

drop

Predicting Retention and Class Progression


The approach to predicting retention and class progression is quite different (and less complex) than curve-fitting. The approach uses a probabalistic procedure to estimate the likelihood that a student, of any given class, will stay at that class level, move to a higher class level, or leave the university.

Predicting Retention and Class Progression


This approach calculates the cumulative probability of an event occurring across successive time periods. For example, the probability that a student will be retained to the next semester is a function of the combined probabilities for being retained for all prior semesters. It is similar to a decisiontree or Markov chain

The Transition Matrix

Fall 2000 enrollment

Fall 2001 enrollment

Transition Matrix

Number of First-time Freshmen in Fall, 2000 who became Sophomores in Fall, 2001

Transition Matrix

Probability of First-time Freshman in Fall, 2000 becoming Sophomore in Fall, 2001

Or, 48.5% of First-Time Freshman in 2000 go on to become Sophomores in 2001

Transition Probabilities Fall, 2000 to Fall, 2001

New Transition Matrix Fall, 2001 to Fall, 2002

You might also like