1. Statewide Mode Choice Models for Tennessee
Stephen Tuttle (RSG) Vince Bernardin, PhD (RSG) Steven Trevino (RSG) Chin-Cheng Chen (RTCSNV)
May 17, 2017

2. Presentation Overview
Background and Model Development Plan
Mode Choice Models
Inputs
Estimation
Outputs/Validation
Conclusions

3. Background

4. Tennessee Statewide Model

5. Existing Passenger Mode Shares
2014 Census Journey to Work Shares
Geography
Auto
Transit
Walk/Bike
Tennessee
97.7%
0.8%
1.5%
Davidson County
(Nashville)
95.4%
2.3%
Hamilton County
(Chattanooga)
96.9%
Knox County
(Knoxville)
97.4%
0.9%
1.7%
Shelby County
(Memphis)
96.8%
1.6%
Automobile is dominant passenger mode
Some regional variation

6. Balancing Priorities
The Challenge: support evaluation of high level statewide scenarios for planning (e.g., more transit, more walkable growth) without requiring the user to develop detailed assumptions (e.g., sidewalks, bus routes)
INCLUDE
Sensitivity to factors affecting non-auto share
Simple, user-friendly inputs
SOV/HOV Split
AVOID
Large data maintenance requirements
Laborious scenario creation

7. Development Plan KEY FEATURES Zonal mode shares Aggregate inputs
Produce vector
No transit networks
Use aggregate funding
Service area
Simple Walk Index
Density, Diversity, Design
0-100 scale
Easy to edit
Zonal Mode Shares

8. What are we Sacrificing?
DISADVANTAGES
Less Precise
No zone-to-zone choices
No network level outputs (for non-auto modes)
Aggregation Bias
Zonal Inputs
No individual choices
Aggregate Model Inputs (HHSize)

9. Mode Choice Models

10. Transit Service Area SERVICE AREA DEFINTION
Percent of TAZ within about ¾ mile of stop
Directly compute from network (if available) or manually/visually approximate
Edit % for future year scenarios
Percent of Zone in Service Area

11. Transit Level of Service
LOS Variable
Agency funding normalized to population in service area
LOS only applies to % of zone in service area
Dial-a-ride handled through constant
Agency Funding
Agency Level of Service

12. Walk Inputs: What Generally Worked?
INTERSECTION DENSITY
Approach Density
Measure from “All Streets” network
Correlated with urban form conducive to walking (e.g., grid network)
Does not directly indicate amenities/attractions

13. Walk Inputs: What Partially Worked?
RETAIL DENSITY
Correlated with attractions/amenities and downtown areas
Also correlated with auto-dependent stores
Used interaction between retail and intersection density
Downtown
Superstore

14. Walk Inputs: What Didn’t Work?
Employment/Activity Diversity Measure
Large zones show significant “diversity” (aggregation bias – large statewide model zones)
Would expect high density zones to be most diverse
Employment Density
Zones with High Activity Diversity

15. Normalized Walk Measure
Index Features
Normalized to value (based on retail/intersection interaction variable)
Directly edited for future year scenarios
Walk Index

16. Simple Mode Choice Models
HBW Utility Equations
SOV =
𝐴𝑆𝐶 𝑆𝑂𝑉
HOV
+ 𝛽 1 ∗ 𝐻𝐻𝑆𝑖𝑧𝑒
Transit
𝐴𝑆𝐶 𝑇𝑟𝑎𝑛
+ 𝛽 2 ∗𝑇𝑟𝑎𝑛𝐿𝑂𝑆
+ 𝛽 3 ∗ 𝑉𝑒ℎ𝑖𝑐𝑙𝑒𝑠 𝑊𝑜𝑟𝑘𝑒𝑟𝑠
Active
𝐴𝑆𝐶 𝑊𝑎𝑙𝑘
+ 𝛽 4 ∗ 𝑊𝑎𝑙𝑘𝐼𝑛𝑑𝑒𝑥
+ 𝛽 5 ∗ 𝑉𝑒ℎ𝑖𝑐𝑙𝑒𝑠 𝑊𝑜𝑟𝑘𝑒𝑟𝑠
HBO Model
Similar to HBW model
Vehicles per HHSize

17. Estimated HBW Model HBW Utility Equations SOV = HOV −3.22 +0.40∗𝐻𝐻𝑆𝑖𝑧𝑒
HOV
−3.22
+0.40∗𝐻𝐻𝑆𝑖𝑧𝑒
Transit
−4.27
+1.38∗𝑇𝑟𝑎𝑛𝐿𝑂𝑆
−3.80 ∗ 𝑉𝑒ℎ𝑖𝑐𝑙𝑒𝑠 𝑊𝑜𝑟𝑘𝑒𝑟𝑠
Active
−3.62
+0.64∗ 𝑊𝑎𝑙𝑘𝐼𝑛𝑑𝑒𝑥
−3.07 ∗ 𝑉𝑒ℎ𝑖𝑐𝑙𝑒𝑠 𝑊𝑜𝑟𝑘𝑒𝑟𝑠
Observations:
6,713
Null log-likelihood:
Final:
Rho-square:
0.691

18. Model Results Distribution of Non-Auto HBW Share
Some regional variation; Relatively high share for some TAZs

19. HBW Model Validation Model vs. ACS and HH Survey HBW Mode Shares
Geography
Source
SOV
HOV
TRANSIT
ACTIVE
All TN
ACS
87.9%
9.9%
0.8%
1.5%
Model
88.0%
9.7%
0.7%
1.6%
Survey
86.2%
10.6%
2.4%
Nashville
84.7%
10.7%
2.3%
85.6%
8.9%
3.1%
Chattanooga
87.8%
9.1%
88.1%
9.4%
0.6%
1.9%
Knoxville
89.4%
8.0%
0.9%
1.7%
88.3%
9.3%
1.8%
Memphis
86.1%
84.8%
9.8%
3.6%

20. Transit Ridership Validation
Daily Transit Ridership .

21. Conclusions

22. Conclusions MAIN BENEFITS
Provides sensitivity to factors affecting non-auto share
Generally provides reasonable mode split
Relatively low transit and walk/bike data requirements
POTENTIAL ENHANCEMENTS
Disaggregate data
Employment diversity
0-Vehicle households
Adjust utility equations
Transform variables (e.g., log)
Add more terms
Explicit treatment of dial-a-ride transit service

23. Contacts www.rsginc.com Stephen Tuttle stephen.tuttle@rsginc.com
CONSULTANT
Vince Bernardin, PhD
DIRECTOR