Upload
hudsonvalleyreporter
View
217
Download
0
Embed Size (px)
Citation preview
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
1/32
Whither KLSD?A Discussion of Enrollment Forecasts, Their Uses,
Requirements and Limitations
Claire Aldrich and Satya NittaAcknowledgments: Chris Sommers
1Disclaimer: The analysis performed in the following pages is to be considered as opinions expressed by the authors only
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
2/32
Executive Summary
Demographers application of CSR method is invalid
CSR approach is inappropriate for cyclical situations
Historical data reveals enrollment could be
recovering soon
Multivariable linear regression may be a more useful
tool
Data requirements persist to complete analysis
Data thus far suggest the risk and cost of being
wrong are still too high
2
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
3/32
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
4/32
Outline
Critical review of the Grip report and conclusions
Review of cyclicality in enrollment
Multivariable linear regression
Observations and discussion
4
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
5/32
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
6/32
...which reflects the major limitations of the CS method for this situation, further exacerbated
by its flawed application.
Summary of C-SR method
2 parts to a standard Cohort Survival Ratio method.
Cohort survival method
KG enrollment prediction
C-SR Method: Start with current enrollments, by grade, then advance students one
grade for each year of the forecast. Thus, the current years kindergarteners
become next years first graders. The current years first graders become nextyears second graders, and so on.
KG Enrollment Prediction: The cohort survival method needs some way to obtain
future kindergarten enrollments. The study done for KLSD assumes that the recent
trend in kindergarten enrollments will continue.
Ex: If kindergarten enrollment declined by 5% annually during the last five years,kindergarten enrollment is assumed to continue to decline by 5%.
Mr. Grip used a variation. Average 3 years of birth data, used a fixed (1.25x) multiplier
based on 6 yr BK ratio taken during the recent downturn of the Great Recession.
6
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
7/32
Major Issues With Demographers Report
Does not consider growth in home sales which is the single
biggest source of increased enrollment in KLSD (statistical
proof later in this document)
Fixed births
Fixed birth to survival ratio of 1.25: - B-K survival ratios
could not be computed for the entire 20 year period b/c
KLSD geocoded data was not available. As a result, Dr. Grip
notes he could only use 5 and 6 year averages (p 12, Nov
2013 report). This dramatically skews the forecast
downward.
Misleading error projections. Claims enrollment can be
predicted within 1% for 5 years.
Implied suggestion that housing growth, macroeconomic
conditions etc are all being predicted to BETTER than 1% for 5
years
Very few first principles models in the physical sciences comeanywhere near this number, leave alone social sciences
Cannot model human behavior accurately (yet)
C-S method not appropriate for cyclical variations
irrespective of size of districts
Misunderstood our SAC/HU ratio historyhas been as high
as 1.25 (a fact overlooked by Dr G, but on the BOE website)
Chip Timing Models
5% accuracy outstanding
7
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
8/32
Two major disadvantages:
The first problem is that the method for calculating future kindergarten enrollments is likely
to be problematic when the trend in kindergarten enrollments is changing.
The second major problem with the K Linear Cohort Survival method is that it will poorly
forecast enrollments if migration patterns are changing. The most likely situation for this to
occur is when there is a substantial change in housing
Disadvantages of the Cohort-Survival Method
Berk & Associates, 2008
its accuracy as a long-term predictive
tool is poor to questionable CSR is only able to accurately forecast
for a very short interval into the future
Beyond one to two years, the
projections tend to be highly suspect since
CSRs major assumption is a linear trend
which may not hold true after a few years
Ref: Grip and Young, Planning and
Changing, 1999
Berk & Associates, 2008
8
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
9/32
Could this Be Us?Case Study: Washington State Mis-projection using Cohort-Survival method
Above graph shows the kindergarten trend in the State during the two time periods. Note that
kindergarten enrollments were declining during the 1997-2002 period, but rose (swiftly and
strongly) during the 2003-2007 period. Continuing the trendunderstated future kindergartenenrollments.
9
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
10/32
Outline
Critical review of the Grip report and conclusions
Review of cyclicality in enrollment
Multivariable linear regression
Observations and discussion
10
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
11/32
Two Major Questions
Is cyclicality a relevant issue for us to factor in?
If the C-SR method is not reliable, what is?
Lets consider our cyclicality
11
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
12/32
Based on Superintendent Kreutzers graph, it looks
like enrollment is going down..
12
0
500
1000
1500
2000
1990 1995 2000 2005 2010 2015
Elementary
Elementary
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
13/32
Based on preliminary data, however, a full history suggests we
are in a trough that could be on the verge of recovery
0
500
1000
1500
2000
2500
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
KLSD Elementary Enrollment
Elementary Enrollment
ElementaryEnrollment
Year 13
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
14/32
Interestingly, Chappaqua follows very much the same
pattern
0
500
1000
1500
2000
2500
1975 1980 1985 1990 1995 2000 2005 2010 2015
Chappaqua
Katonah
1. Chappaqua, 2013: surprise Elementary enrollment increase of 39 students this year (58 total
across all grades). Attributed to existing home sales
2. KLSD 1998: Surprise increase of 91 students in grades 1-6 from the prior year's K-5, due to
home sales.
3. KLSD 1999: Additional 80 students
Chappaqua vs Katonah, Elementary Enrollment
14
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
15/32
So, if CSR is not reliable for cyclical
situations, What is?
The goal of this study was to construct a regression model tomore accurately predict enrollments than calculated by either theCohort-Survival Ratio method or by moving averages for a timeframe of one to seven years into the future
Although this *CSR+ is very accurate in short-term planning, its accuracy
as a long-term predictive tool is poor to questionable CSR is only able to accurately forecast for a very short interval into the
future
Beyond one to two years, the projections tend to be highly suspectsince CSRs major assumption is a linear trend which may not hold trueafter a few years
In a separate study to test whether regression analysis was abetter method of prediction than the Cohort-Survival Ratio,Webster (1970) found a regression model to be significantly moreaccurate than the CSR.
Source: R.S. Grip & J. W. Young, Planning and Changing (1999).15
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
16/32
Outline
Critical review of the Grip report and conclusions
Review of cyclicality in enrollment
Multivariable linear regression
Observations and discussion
16
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
17/32
How common are Regression algorithms?
One of the most popular projection algorithms
Used everywhere:
Epidemiology: Ex: Relation of smoking to mortality
Finance: Ex: Capital pricing asset models
Economics: Ex: Predict consumption spending
Real Estate: Ex: Predict housing prices
Physical Sciences: Ex: Complexity
Behavioral Sciences: Ex: Athletics world records
17
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
18/32
Multivariable Linear Regression: Inputs to the model
Housing
Birth rates
18
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
19/32
Housing strongly correlates with Enrollment
R = 0.6816
0
500
1000
1500
2000
2500
0 100 200 300 400 500
Total Elementary Enrollment
Total Elementary Enrollment
Linear (Total Elementary
Enrollment)
19
0
50
100
150
200
250
300
350
400
450
500
0
500
1000
1500
2000
2500
1975 1985 1995 2005 2015
Enrollment vs Housing
Elementary
EnrollmentHouses Sold
C-SR method applied during a growth cycle
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
20/32
Multivariable Linear Regression:
How good is this model?
0
500
1000
1500
2000
2500
2000 2002 2004 2006 2008 2010 2012 2014
Actual
Predicted
Limitations:Fits training data very well if data is linear (R2~ 0.9), but
Need independent variable trends
Accuracy improves with dataif cyclicality is considered, need more mathematical rigor (and
perhaps use polynomial regression or normalization)
Independent variables should not be strongly correlated (sufficient to establish housing and
birth rates are not highly correlated)
Enrollment = f(Housing, Birth Rates)
20
0
500
1000
1500
2000
2500
1990 1995 2000 2005 2010 2015
Elementary
Projections
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
21/32
So - What factors might drive a recovery home
sales forecast?
Economic indicatorsstock market up, unemployment claims down, GDP up..
Affluent Echo boomers nearing peak move to suburbs with children age
(KLSD) Empty nester pent up supply (needs more research- can we geocodelocations? Work w. R/E brokers to do surveys?)
Early signs of life in KLSD: In our townfirst grade class up 7%; Preschool classes also seeing growth
Home sales up 12%, now at 200 annual run rate
Last cyclehome sales went from 200 to 300 in one year and remained onaverage over 300 for the next 13 years...
Other notes:
FYI - Our SAC/HU ratio historyhas been as high as 1.25 (a fact overlooked by DrG, but on the BOE website.). Stock ratio, likely higher for new home sales.
Any secular concerns need to be addressed with hard data, not hunches oranecdotes
21
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
22/32
Linear Regression Projections
Capacity footnotes: BottomLine (A) 1380 = capacity with all 61 standard classrooms
Top line (B) 1440 = capacity in event of unlucky surprise: +60 students, across 3 grades and 2 schools need +6
extra classrooms in 2 schools. Note also violates ideal school size of 400 as shown in literature.
22
0
50
100
150
200
250
300
350
400
450
500
0
500
1000
1500
2000
2500
1970 1980 1990 2000 2010 2020
Chart Title
Elementary Enrollment
Enrollment
Enrollment
Enrollment
Grip Projections
Houses Sold
A*
B*
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
23/32
Our forecast thus far
Our forecast thus far:
Housing assumptions
Conservative: Housing stays flat, 210 sales a year
Moderate: 210300 growth over 5 years
Strong Recovery: 210330 growth over 5 years
Historical data show that home sales rose from 200 to 300 in one year
and remained on average above 300 for the next 13 years
23
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
24/32
Simple math and The Perfect Storm
(You tell me housing, Ill tell you enrollment)
Additional housing comes on line (Falcon
Ridge, Oakridge), housing market recovers, we
get a surprise increase K, 1stand 2ndgraders,
we need 3 extra classrooms at 1, 2, or 3schools.
Class sizes rise from 200 to 300 annually
enrollment reaches 1450 in 3 yrs (300 275 250200 200 200)
School size rises above 450+
24
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
25/32
Outline
Critical review of the Grip report and conclusions
Review of cyclicality in enrollment
Multivariable linear regression
Observations and discussion
25
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
26/32
What if were wrong?
What if were wrong? (How do we pass the What were you thinking
test?)
Is it appropriate to make decisions in the trough?
Especially where young children and their education is involved?
Would closing a school now be the equivalent of shorting the market?
If Jamie Dimon could not time it, and Alan Greenspan couldnt either
why should we be able to?
As we all know, its virtually impossible to time the market.
The only way to win long term is to stay invested and not to be going in
and out of the market. Capacity issues - caution re squeezing. MBA 101: NEVER run your plant
over 85% capacity
26
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
27/32
Other issues
Slippery slope risk: Need to keep our school system attractive to attract future
buyers and keep our community vibrant. MBA 101: Successful businessesmaintain and invest in their brands. They are sacred.
Impact of school sizeA survey of the literature shows that 400 students is the
ideal school size- could it be that all of our elementary schools were designed with
this principle deliberately in mind?
Tax cap issuesto reopenor grow when necessary
Execution riskthoughtful redistricting, physical simulation of bus routes still to be
done, as well as many other transition plans
27
Recap
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
28/32
Recap Demographers enrollment projections invalid.
Not appropriate for cyclical situations - ignored housing growth as a key predictor
Our district, like our peers, has cyclic enrollments. We are in a trough, and may berecovering soon.
We showed housing is a strong predictor
We built a multi-variable regression model. Predicted the past accurately. Needs work.
Initial regression model suggests strong correlation with housing
Over-predicting test data
A recovery could be underwaybut no one can accurately predict housing growth.
Chappaqua shows strong and sudden growth in enrollment and housing. We are very similar (recall
comparison)
In the absence of a credible enrollment forecast, there is no case to close a school.
At a minimum, we need more time to study and collect data
In a broader sense, even after further analysis, it may not be appropriate to
recommend closing a school in a period of cyclical recovery: (Not if, but when)
Respectfully submit that in faithfully upholding its stewardship of young children
and their education, it is not the province of a school administration (or any of us)
to be speculating on the future of housing. We are a school system, not a hedge
fund. 28
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
29/32
Appendix
Claire Aldrich Holds a BA in Economics and Political Science from Yale
University and a CFA. Claire has worked as a credit
analyst on Wall Street for 30 years.
Satya Nitta
- Holds a Ph. D in Chemical Engineering from Rensselaer
Polytechnic Institute. Master Inventor at IBM T J Watson
Research center. 15 years of experience in Nanoelectronicsinventing and developing future chip technologies. Recent
focus is on applications of Cognitive Computing.
29
i i l i d
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
30/32
Hypothesis:
Parameters:
Cost Function:
Goal:
Linear Regression Explained
30
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
31/32
0
1
J(0,1)
31
8/13/2019 Katonah-Lewisboro Parent Report in Response to Proposed School Closure
32/32
Gradient descent algorithm
Correct: Simultaneous update Incorrect:
32