Transcript Document

Examining Potential Demand of
Public Transit for Commuting Trips
Xiaobai Yao
Department of Geography
University of Georgia, USA
5 July 2006
Outline
•
•
•
•
•
The trend of public transit in the US
Objectives of the study
Methodology
Case study
Conclusions
Renaissance of Public Transit
in the US
•
•
•
•
Traffic congestion
Economic growth
Gas price vs affordable transit fare
Environment sustainability
Public transit
networks in the
city of Atlanta
Research on Public Transportation
•
•
•
•
•
Accessibility for special groups
Land use / transportation relationship
Cost, benefit, pricing
Network analysis
…?
Research objectives of the study
• Measure the potential need of public
transportation
• Identify and visualize clusters of high
potential needs areas
Methodology
• Identify Predictive Factors
• Identifying and Visualizing Potential
Demand Distribution
– The Need Index approach
– A data mining approach
• Case study
Data
Land-use, socioeconomic, and transportation
(trips by mode) data at TAZ level.
Identify Predictive Factors
Multiple Regression
k
R 
v
i
i
i 1
where R is the proportion of workers taking public transit as the
primary mode, vi ’s are the identified independent variables, and
k is the total number of these variables.
Identify Predictive Factors
- the Atlanta case
Independent variables:
•
Land-use characteristics
–
–
–
•
Population density
Employment rate
Percentage of home workers
- Average number of workers per HH
- Job density
Socioeconomic characteristics
–
•
Income
- Car ownership
Network structure
–
Density of bus stops in the TAZ
- Density of rail stations in TAZ
Regression Results
Predictive Variables
(Unstandardized)
Coefficients
B
(Constant)
Sig.
Std. Error
Collinearity Statistics
Tolerance
VIF
1.334
.824
.106
.008
.034
.816
.864
1.157
Percentage of workers below poverty line
(x1)
.074
.019
.000
.629
1.589
Percentage of workers with income from
100% to 150% of poverty line (x2)
.103
.026
.000
.679
1.474
Percentage of worker with 0 vehicle in the
household (x3)
.421
.017
.000
.510
1.961
Percentage of worker with 1 vehicle in the
household (x4)
.033
.010
.001
.552
1.812
Employment rate (x5)
-.045
.014
.001
.541
1.847
Average # of workers per household
-.007
.512
.989
.551
1.816
.036
.006
.000
.632
1.583
-.026
.002
.000
.336
2.974
Rail station Density
.098
.198
.623
.832
1.201
Bus stop Density
.080
.006
.000
.251
3.982
Percentage of home workers
Population Density (x6)
Job Density (x7)
Identifying and Visualizing Potential
Demand Distribution
1. The Need Index approach
2. A data mining approach – self-organizing
maps
1. The Need Index approach
n
R 

i 1
m
i
xi 

i
yi
i 1
yi ’s: variables accounting for the network structure and
level of service of transit systems
xi ’s: variables that are not about the transit systems.
R = NI + Net
NI = R-Net
Need Index for the Atlanta Case
NI ( i )  0 . 074 x1  0 . 103 x 2  0 . 421 x 3  0 . 033 x 4  0 . 045 x 5  0 . 036 x 6  0 . 026 x 7
Critique on the Need-Index approach
• Simple
calculation
• Easy
interpretation
• Possible to rank
and/or to
quantify the
difference
• Classification/Visu
alization Dilemma
(where are the
magic breaks)
• The validity of
linear relationship
assumption
2. The SDM approach : Selforganizing maps
<x1, x2, …. xn>
Self-organizing maps: how it works
N
d
j

 (y
i
( t )  w ij ( t ))
2
j 1
w ij ( t  1)  w ij ( t )   ( t )( x i ( t  1)  w ij ( t ))
SOM in this study
(weighted vector space )
  1 x1 ,  2 x 2 , ...  n x n 
  1 x1 ,  2 x 2 , ...  n x n 
7
8
9
4
5
6
1
2
3
Visualizing the SOM patterns
Critiques on the SOM approach
• No assumption on the
relationship
• Self-assigned clusters
• No quantitative
measure
• No ranking
Conclusions
• The integrative approach is successful.
• The Need Index approach and the spatial
data mining approach are complementary
and mutually confirmative.
• Confirmed by the other approach, the Need
Index approach provides an efficient and
effective solution to transportation planners.