A Conditionally Parametric Probit Model of Micro-Data Land Use in Chicago Daniel McMillen Maria Soppelsa
Dec 14, 2015
A Conditionally Parametric ProbitModel of Micro-Data Land Use in Chicago
Daniel McMillenMaria Soppelsa
Overview
• Residential v. Commercial/Industrial Land Use in Chicago, 2010
• A conditionally parametric (CPAR) approach produces smooth estimates over space
• Target points chosen using an adaptive decision tree approach (Loader, 1999)
• Interpolation from 182 target points to all 583,063 individual parcels in the data set
Estimation Procedures
• Case (1992). Special From for W• McMillen (1992). EM Algorithm• Pinkse and Slade (1998). GMM for spatial
error model.• LeSage (2000). Bayesian approach• Klier and McMillen (2007). Linearized version
of GMM probit/logit for spatial AR model.
GMM Probit
• ,
β, ρ to minimize
Linearized GMM Probit
1. Standard probit: 2. 2SLS regression of e on on and , where
3. . Requires inversion of
CPAR Probit
• = kernel weight function, distance between observation j and target point.
• Straightforward extension of “GWR” – a special case of locally weighted or locally linear regression.
• Applications:– McMillen and McDonald (2004)– Wang, Kockelman, and Wang (2011)– Wren and Sam (2012)
Spatial AR v. LWR
Data
• Individual parcels in Chicago, 2010• Major Classes:1. Vacant Land (33,139)2. Residential, 6 units or fewer (728,541, 539,975 after
geocoding)3. Multi-Family Residential (11,529)4. Non-Profit (316)5. Commercial and Industrial (50,508, 43,088 after
geocoding)6. “Incentive Classes” (1,487)
Explanatory Variables
• Distance from parcel centroid to:1. CBD2. Lake Michigan3. EL line4. EL stop5. Rail line6. Major street7. Park8. Highway
Rogers Park
Adobe Acrobat Document
Descriptive StatisticsVariable Mean Std. Dev. Min Max
Residential Lot 0.926 0.262 0.000 1.000
Distance from CBD 7.518 3.433 0.022 17.006
Distance from Lake Michigan 4.116 2.716 0.005 12.321
Distance from EL Line 1.358 1.277 0.001 6.265
Distance from EL Stop 1.214 1.081 0.001 6.265
Distance from Rail Line 0.428 0.294 0.001 1.997
Distance from Major Street 0.080 0.057 0.000 0.508
Distance from Park 0.233 0.153 0.000 2.999
Distance from Highway 1.476 1.027 0.011 4.809
Probit Models, Probability Residential Standard Probit CPAR Probit
Variable Coef. Std. Error Mean Std. Dev.Intercept 0.061 0.046 0.351 1.008Distance from CBD 0.132 0.007 0.101 0.266Distance from Lake Michigan -0.095 0.007 -0.086 0.308Distance from EL Line 0.002 0.013 -0.423 1.168Distance from EL Stop -0.091 0.013 0.511 1.263Distance from Rail Line 0.626 0.014 0.649 0.686Distance from Major Street 8.748 0.070 11.570 6.427Distance from Park -1.099 0.020 -0.881 0.994Distance from Highway 0.212 0.007 0.048 0.351Log-likelihood -131518.9 -120714.1Pseudo-R2 0.144 0.215
Probability of Residential Land Use: Standard Probit
Probability of Residential Land Use: CPAR Probit, 10% Window Size
Difference, CPAR Probability – Standard Probit Probability
Kernel Density Estimates for CPAR Coefficients
Adobe Acrobat Document
LWR Estimates of CPAR Coefficients
Adobe Acrobat Document
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Marginal Probabilities
Rogers Park
Adobe Acrobat Document
Rogers Park, n = 3,193 Standard GMM CPAR
Coef Std. Err. Coef Std. Err. Mean Std. dev.Intercept 49.979 11.999 42.977 12.592 0.025 2.445CBD -1.804 0.462 -1.549 0.480
Lake Michigan -7.621 1.672 -6.555 1.814 -0.726 5.314
EL Line -3.324 0.651 -2.901 0.723 -4.449 9.934
EL Stop 3.127 0.654 2.698 0.739 6.593 9.706
Rail Line 1.906 0.395 1.659 0.428 1.675 4.059
Major Street 7.123 0.837 5.992 1.346 15.900 9.561
Park -1.797 0.514 -1.594 0.525
Highway -7.207 1.743 -6.197 1.809
Metra Stop 0.038 0.216 0.024 0.178ρ 0.155 0.167pseudo-R2 0.084 0.084 0.343
Correlations, Predicted Probabilities
Standard GMM CPAR
Standard 1 0.57 0.99
GMM 0.57 1 0.57
CPAR 0.99 0.57 1
Standard Probit Probabilities
Adobe Acrobat Document
CPAR Probit Probabilities
Adobe Acrobat Document
Standard Probit: Southwest
Adobe Acrobat Document
CPAR – Standard: Southwest
Adobe Acrobat Document
Standard Probit: Southeast
Adobe Acrobat Document
CPAR – Standard: Southeast
Adobe Acrobat Document
Standard Probit: Northwest
Adobe Acrobat Document
CPAR – Standard: Northwest
Adobe Acrobat Document
Standard Probit: Northeast
Adobe Acrobat Document
CPAR – Standard: Southeast
Adobe Acrobat Document