Admission Guidelines Attendance Enrollment Procedure for ...
Enrollment Forecasting Approaches for Open Admission Institutions
description
Transcript of Enrollment Forecasting Approaches for Open Admission Institutions
1
Enrollment Forecasting Approaches for Open Admission Institutions
R. Ty JonesDirector of Institutional ResearchColumbia Basin College
PNAIRP Annual ConferencePortland, Oregon November 7, 2012
Links
2
If you would like to follow along with the data and techniques and presentation, here are the links.
http://dl.dropbox.com/u/9234919/2012_3Way.xlsx
http://dl.dropbox.com/u/9234919/2012_MP.xlsx
http://dl.dropbox.com/u/9234919/SPSSEnrollment.sav
http://dl.dropbox.com/u/9234919/SPSSMLR.sav
http://dl.dropbox.com/u/9234919/2012_Forecast_Data.xlsx
http://dl.dropbox.com/u/9234919/20121106_Forecast_Workshop.pptx
3
OverviewApproximate Timeline
Rational and pragmatic philosophy to enrollment forecasting (5 Minutes)Forecasting basics (5 Minutes)Linear Regression approaches (SLR) (15 minutes)Fitted Curve approaches (CLR) (10 Minutes)Multivariate Linear Regression (MLR) (20 minutes) Autoregressive–moving-average models (ARIMA) (20 minutes)Data imputation (10 minutes)Mixed methods (10 minutes)Other approaches (5 minutes)Forecast weighting (5 minutes)Presenting the data (5 minutes)Conclusion, questions and answers (10 minutes)
That’s 120 minutes plus a break to fit into 90 minutes! So, lets go!!!!
Philosophy
4
• Predicting the future is hard! • Forecasting is easy.• There is no such thing as a perfect forecast.• A forecast is only as good as the data that goes into it.• All forecasting methods have strengths.• All forecasting methods have weaknesses.• E pluribus unum!• If it doesn’t make sense, don’t use it.• If you can’t explain it, don’t use it.• Prepare, prepare, prepare!
5
Basics
“Forecasting is the process of making statements about events whose actual outcomes (typically) have not yet been observed. A commonplace example might be an estimation of some variable of interest at some specified future date.” - Wikipedia
Forecasting requires process and estimation. Anything else is WAG!
The processes chosen by institutional research must be founded on statistical and/or mathematical principles. That means data must be at its core to have any validity.
Linear Regression
6
Linear regression uses the process of least squares to model the relationship between a dependent variable and an explanatory variable.
Strengths:• Robust• Minimal data requirements• Easily explained
Weaknesses:• Variances make short and long
term estimates difficult• Tends to over simplify trends
7
Fitted Curve
Fitted curve regression operates on a similar basis as linear regression. Instead, transformations to the data optimize the least square process to fit an equation line dictated by the transformation.
Strengths:• In many cases, curve fitting better fits time series
data.• Provides stronger explanation than linear models.Weaknesses:• Variances can force large margins of error in
making estimates.• Some curve fitting may be significant, but not make
actual sense.
Multivariate Linear Regression
8
Multivariate linear regression uses the process of least squares to model the relationship between a dependent variable and multiple explanatory variables.
Strengths:• Robust• High explanatory value• Once model is established, allows a lot of different “what if”
scenarios to be looked at.Weaknesses:• Extending the model for significant independent variables into the
future can be difficult.• Interactions can make model interpretation difficult.• Resulting models can be very complex.
9
ARIMA
An autoregressive–moving-average model uses a combination of data smoothing and regression in time series data. Unlike true regression approaches, uses only dependent data to estimate future outcomes.
Strengths:• Often better reflects cyclical dependent data.• Lack of dependence on explanatory factors allows
sbetter long term projections.Weaknesses:• Getting the correct model can be very difficult• Explaining the model can be difficult.• Sometimes, no model can be generated.
Imputation
10
There are a variety of data imputation techniques. All aim at filling holes or extending estimates. All use various formulae to look for patterns in existing data to estimate missing data.
Strengths:• Not as effected by variances so short term and long term
estimates are more consistent.• Mathematically more straight forward.Weaknesses:• Can miss cyclic patterns.• Using the wrong imputation for the data can result in
large out of range errors.
11
Mixed Methods
Mixed method models use a combination of forecasting approaches to arrive at estimations.
Strengths:• Mixing methods may provide data smoothing to highly
variable data.• May allow access to estimates that a single model
approach would not allow.Weaknesses:• Can result in amplified error and variance of estimates.• Explaining the model can be difficult.• Measuring confidence in the model is difficult
Other Issues
12
Other forecasting methods•Bayesian estimate models•Hot-Decking•Random Wandering Models
Forecasting weighting
13
Presenting The Data
2012 2013 2014 2015 2016 2017 2018 2019 2020 20217000
7500
8000
8500
9000
9500
77707913 7960
80188050
81348220 8247
83158286
2012 Longterm Forecast
Enro
llmen
t
Finish
14
Conclusion, questions and maybe some answers…
Thank you for participating!