|
" Improving Year-to-Year Bicycle Traffic Flow Prediction for Health and Safety Behavior and Other Public Health Applications "
Martin, Eric Vance
Torabi, Mohammad R.
| Document Type
|
:
|
Latin Dissertation
|
| Language of Document
|
:
|
English
|
| Record Number
|
:
|
1104127
|
| Doc. No
|
:
|
TLpq2193766589
|
| Main Entry
|
:
|
Martin, Eric Vance
|
|
|
:
|
Torabi, Mohammad R.
|
| Title & Author
|
:
|
Improving Year-to-Year Bicycle Traffic Flow Prediction for Health and Safety Behavior and Other Public Health Applications\ Martin, Eric VanceTorabi, Mohammad R.
|
| College
|
:
|
Indiana University
|
| Date
|
:
|
2019
|
| student score
|
:
|
2019
|
| Degree
|
:
|
Ph.D.
|
| Page No
|
:
|
232
|
| Abstract
|
:
|
Policy makers would like to increase physical activity through bicycling and to reduce bicyclist exposure to injury and air pollution. Bicycle traffic flow can be used to study these problems and design interventions; however, it is typically observed sparsely. Statistical bicycle facility demand (BFD) models, grounded in the land use theory of traffic flow, can predict it citywide, enabling analyses. Typically, BFD modelers' exploratory model building methods incorrectly assume that observations are independent. This can lead to prediction error, Type I error in covariate selection, and poor transferability of results. The purposes of this dissertation were to assess the impact of assumption violations, to demonstrate better methods for clustered, temporal, and spatial bicycle count data, and to test rigorously whether improvements in bicycle facilities may increase bicycle traffic flow. This exploratory study re-analyzed a published dataset of unbalanced, longitudinal, annual bicycle counts and candidate covariates. In both main analyses, data were fit to multilevel regression-based growth curve and spatial models. Re-analysis of a published BFD model showed that it likely retained several covariates in error. In addition to the expected explanation of the spatial distribution of the outcome, a new growth curve model-based BFD model built with forward selection indicated that the within-location addition of bicycle facilities was associated with increasing bicycling, the first study to find this in a longitudinal framework. In validation, measured predictors performed well, but kriging predictions of the model error did not. Unlike previous studies, the results are presented with high confidence that non-independent residuals did not affect covariate selection or effect estimation. Results demonstrated the importance of statistical control of correlated residuals due to clustering and spatial autocorrelation. The literature review suggested a framework for rigorous BFD modeling and improved transferability. Future studies should use random samples and statistically control clustering and spatial autocorrelation.
|
| Subject
|
:
|
Behavioral psychology
|
|
|
:
|
Public health
|
|
|
:
|
Transportation
|
| |