Auto Ownership Model For Southeast Florida Models Southeast Florida FSUTMS Users Group Meeting Ft. Lauderdale, FL May 16, 2008 Corradino
Why? Life Style Gen Requires Number of HHs Current model input is: At each auto ownership level (0, 1, 2, 3+) For HH with and HH without children Current model input is: Vehicles in HH’s with children Vehicles in HH’s without children Disaggregated by a series of polynomials Lack of consistent and logical forecasting methodology Most models either grow vehicles in proportion to HH’s, or Perhaps adjust to meet county aggregate total May 16, 2008 Corradino
Past Work For District 4 – Paper by Li, Ewing & Gross (late 1990’s?). Other major urbanized areas throughout the US. Encouraged by USDOT Most models use some combination of: Persons Workers Density Children Income – Almost all require this one **** Accessibility May 16, 2008 Corradino
Since FSUTMS Doesn’t Use Income and Previous Work Required Income Persons/HH Workers/HH Children With Without Density (HH, Emp, Area) Accessibility (Time & Trip Attractions) May 16, 2008 Corradino
Model Form Multinomial Logit Sx = Exp(ux)/[Exp(u0)+ Exp(u1)+ Exp(u2)+ Exp(u3)] Where Sx = share of HH’s at each auto ownership level (0,1,2,3) ux = utility for level x = f(persons or workers/HH, density, accessibility) May 16, 2008 Corradino
Density Calculation Density = (Households + Total Employment)/(net area) May 16, 2008 Corradino
Accessibility Calculation Inputs Travel Time Skim Matrix (for GM) Trip Attractions (from “trial” Gen run) Friction factors For each TAZ: Accessibility= Sum of: FF(t)*Attractions for each purpose Over all destinations May 16, 2008 Corradino
Best Model Households without children Rho-squared: 0.342683 Persons Density Accessibility Rho-squared: 0.342683 May 16, 2008 Corradino
Best Model Households with children Rho-squared: 0.288637 Workers Density Accessibility Rho-squared: 0.288637 May 16, 2008 Corradino
Other Notes Gross and Ewing found a rho-squared of 0.4704 for their model 4. For the 0, 1, 2+ models, comparable to Gross and Ewing, our models have rho-squared of 0.39 and 0.50 for without and with children respectively. These models are at least as well as Gross and Ewing. For the 0, 1, 2, and 3+ models, the proposed models have rho-squared of 0.34 and 0.29 for without and with children respectively. These are still good models, but as expected, the task is more difficult when trying to estimate 4 levels. The signs in all models are correct, and all variables included in the equations are significant at the 0.05 level and beyond for most variables. May 16, 2008 Corradino
Remaining Work County Biases will be tested Life-Style Gen will have to be modified to use the new model Actual model implemented in Matrix May 16, 2008 Corradino
Questions? May 16, 2008 Corradino