Technical Assistance for Modeling to the Champaign MPO

The study is being conducted on a task-order basis. The following tasks have been completed:

  • A review of the trip generation models currently used.
  • Address issues related to the small-scale household travel survey conducted for modeling purposes.
  • Implementation of alternative trip generation models.

Principal Investigator(s):

Paul Metaxatos

Participants:

Paul Metaxatos
Ashish Sen

Status:

01/01/2011

Objective:

To assist the staff of the Champaign, Illinois MPO in tasks related with enhancing the transportation models maintained in the region.

Strategy:

The study will review data requirements and model components and propose potential improvements.

Expected Results or Products:

Three specific issues related to the size of the household travel survey are addressed: identification of outliers, improving the reliability of small samples, and imputation issues in cases of missing information.

Identification of outliers: When sample sizes are large, estimates are generally not affected too much by a suspect observation. However, with small sample sizes, a ‘bad’ observation (due to a mistake in recording data or a respondent giving misleading information) can play havoc with the quality of estimates and have a profound effect on forecasts.  Such bad estimates need be of concern only when they are influential. Influential observations that also are typically outliers are also discussed.

Reliability of small samples: The small number of observations in the household travel survey has the consequence that cells representing several categories of household sizes and numbers of workers have only very few observations. In such a case, reliability can be enhanced combining table cells with like trip rates. A method for doing so is classification and regression tree (CART) analysis.  The procedure is well known in the statistical literature but has not been used much in transportation. It is used in this study to develop alternative trip generation models.

Imputation: Some cells during cross-classification analysis might not receive an observation at all – or the planner might find the observations too few or too unreliable. In that case, cell estimates might be imputed from observations in other cells. A procedure we have called row-column decomposition analysis is used as an imputation method. Imputation can also be used in lieu of combining of cells using CART.

NOTE: A report or paper from this research is not immediately available.

Contact:

Paul Metaxatos
Urban Transportation Center
University of Illinois at Chicago
412 South Peoria Street, Suite 340
Chicago, IL 60607
Voice: (312) 996-4713
Fax: (312) 413-0006
Email: pavlos@uic.edu

Sponsors:

Idot logo 2

 

 

 

Illinois Department of Transportation
Metropolitan Transportation Support Initiative