Test Driving a Small-Area Population Forecasting Model: Seeking Additional Horsepower through Updated Engineering and Non-Demographic Fuel Additives Paul R. Voss and Guangqing Chi Applied Population Laboratory Center for Demography and Ecology University of Wisconsin – Madison BSPS Annual Conference 2006 September 2006 The University of Southampton Support provided by the Wisconsin Agricultural Experiment Station (Hat ch project no. WIS04536)
29
Embed
Paul R. Voss and Guangqing Chi Applied Population Laboratory Center for Demography and Ecology
Test Driving a Small-Area Population Forecasting Model: Seeking Additional Horsepower through Updated Engineering and Non-Demographic Fuel Additives. Paul R. Voss and Guangqing Chi Applied Population Laboratory Center for Demography and Ecology University of Wisconsin – Madison - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Test Driving a Small-Area Population Forecasting Model:
Seeking Additional Horsepower through Updated Engineering and Non-Demographic Fuel Additives
Paul R. Voss and Guangqing ChiApplied Population Laboratory
Center for Demography and EcologyUniversity of Wisconsin – Madison
BSPS Annual Conference 2006
September 2006
The University of Southampton
Support provided by the Wisconsin Agricultural Experiment Station (Hatch project no. WIS04536)
Motivating Questions• What can be done to improve the abysmally atheoretical
nature of small-area population forecasts?• In particular, what about a regression approach?• Especially, what if we step outside our disciplinary confines
and incorporate variables from other fields that, at face value, must be predictors of population growth?
• nature of the land (ground cover, wetlands, hydrography, slope)• accessibility (transportation infrastructure, highways, airports, etc.)• developability (high/low growth potential)• desirability (natural and built amenities)• livibility (potential quality of living)
• And, surely, should we not begin immediately to adopt some of the spatial econometric approaches long effectively employed by quantitative geographers and regional scientists?
• Broaden our thinking regarding the relationships between population change and the host of factors influencing such change – some drawn from demography but many others from disciplines not normally involved in formal population forecasting efforts
• Categorize and integrate these factors in an effective way (construct indexes)
• Incorporate spatial process effects into the model• Carry out the forecasting at a sufficiently fine geogra
phic level that environmental and geophysical effects on population change can be better captured and modeled
Proposed Regression Approach
Strategy
• Assemble all necessary data for 1990 base year
• Forecast populations for 2000
• Compare 2000 forecasts with 2000 census results
Preview of Findings…
It didn’t work
Our Region1,837 minor civil divisions in state of Wisconsin, U.S.
Our Datacensus data
satellite imagery
other data from several federal and state statistical agencies
Population
Demographics
AccessibilityDevelopability
Livability Desirability
Temporal
Spatial
Population Change Conceptual framework
Population
Demographics
AccessibilityDevelopability
Livability Desirability
Temporal
Spatial
Population Change Conceptual frameworkLocal demographic characteristics----------------------------------------------Population densityAge: the young and the oldMinority: black and HispanicInstitutional population (college)Education attainment: HS and Bchl.Geographic mobilityPovertySeasonal housingSustenance organization: retail and agricultural industrial structure
Population
Demographics
AccessibilityDevelopability
Livability Desirability
Temporal
Spatial
Population Change Conceptual framework
Transportation infrastructure--------------------------------------Residential preferenceHighway infrastructureAccessibility to airportsAccessibility to highwaysAccessibility to workplaces
Population
Demographics
AccessibilityDevelopability
Livability Desirability
Temporal
Spatial
Population Change Conceptual framework
The potential for land conversion & development-----------------------------------WaterWetlandsSlopeTax-exempt (protected) landsBuilt-up lands
Population
Demographics
AccessibilityDevelopability
Livability Desirability
Temporal
Spatial
Population Change Conceptual framework
Natural & built amenities desirable for living--------------------------ForestsWaterLakeshore/riverbank/ coastlineGolf coursesslope
Standard regressionPartial spatio-temporalregression
Full spatio-temporalregression
Dependent variables: population change, population density, population density changeIndices generating methods: PCA, coefficients, coefficients and correlations
Projections using indices
Population forecast adjustments
Evaluation and comparison
Projection using individual variables
Select the best one
Select the better one
Regression projection
Forecasting and Evaluation
Model 1: Extrapolation projection
P P G2 0 0 0 9 0 1 0
GP P P P P P
9 0 8 0 9 0 7 0 9 0 6 0
1 0 2 0 3 03
Model 2: Standard regression
L nP
PL n
P
PX
9 0
8 0
8 0
7 08 0
L nP
PL n
P
PX
0 0
9 0
9 0
8 09 0
Model 3: partial spatio-temporal regression(incorporating spatial population effects)
L nP
PL n
P
PX W L n
P
P neighbor
90
80
80
7080
80
70
L nP
PL n
P
PX W L n
P
P neighbor
0 0
9 0
9 0
8 09 0
9 0
8 0
L nP
PL n
P
PX W L n
P
PW X
neighborneighbor
9 0
8 0
8 0
7 08 0 1
8 0
7 02 8 0
( )
L nP
PL n
P
PX W L n
P
PW X
neighborneighbor
0 0
9 0
9 0
8 09 0 1
9 0
8 02 9 0
( )
Model 4: full spatio-temporal regression(incorporating spatial population effects
and other neighbor characteristics)
Four Finalized Population Forecasting Models
So… How did it turn out with all this re-engineering and fancy fuel additives?
Population projections to 2000 with adjustments at the MCD level
Summary• Things just didn’t turn out as we hypothesized (and
hoped) they would• Our fancy spatio-temporal model outperformed
simple regression in the estimation stage of the analysis (but who cares?)
• But, to our dismay, in the forecasting stage, the a-theoretical, simple extrapolation model outperformed the regression models in all comparisons but one
• In only one set of MCDs did the fancy model outperform all others: MCDs of fewer than 250 people. We launched this project in the belief that non-demographic variables might perform best in very small areas, and this finding may suggest that we explore that possibility further