Results today based primarily on three data sources…...SHRPII C04: TEG Meeting, Washington, DC -...
Transcript of Results today based primarily on three data sources…...SHRPII C04: TEG Meeting, Washington, DC -...
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Results today based primarily on three data sources…
Seattle 2006 household travel survey (RP)
Seattle 2006, San Francisco 2007, Los Angeles 2009 congestion pricing SP surveys
Seattle 2006 Traffic Choices “experimental RP” data
Comparable experiences with many other data sets
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Seattle region (PSRC) RP data
2006 household travel survey
Used 2-day place-based diary
4700 HH, 90,000 trips
HW network times for SOV and HOV 5 periods in the day (AM,MD,PM,EV,NT)
17 periods in the day (mostly 1-hr long)
Separate skims of time on highly congested links
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Models estimated on Seattle RP
Time of day choice (1 hour periods)
Mode choice (6 modes)
Joint time of day & mode choice
Two purpose groups: HBW, HBO
Two decision levels: trip-based, tour-based
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
General RP analysis approach
1. Time and cost only
2. Test specification of time variables
3. Add cost segmentation (income, occ.)
4. Add time segmentation
5. Add other explanatory variables
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Findings- basic TOD tests
Based on time coefficient(s) only- No cost difference across TOD alternatives
Using departure time from home works better than using arrival time at work
Using arrival time back home works better than departure time from work
Using a restricted set of alternatives (6-8 hours) works better than all hours of day
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Findings- basic TOD tests (2)
Variable of extra time on very congested links is highly correlated with travel time. Works best as a shift variable related to extra time in the worst hour (peak of the peak)
Similar findings from Sacramento data
May be proxy effect for variability
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Mode choice nesting structure
Auto Pure Transit Non-motorized
Drive Shared Shared Drive to Walk to Walk/
alone ride 2 ride 3+ Transit Transit bike
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Joint mode/TOD nesting structure (trip-based)
Tested seven different structures
HBW Bottom level- nesting of one hour periods into
broader time of day periods
Middle level – nesting of modes into groups
Top level – joint decision across mode groups and TOD groups (logsum close to 1.0)
HBO Bottom level – nesting of modes into groups
Mode groups nested under one hour periods
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Income and car occupancy included several ways…
Mode-specific dummy variables
Modifiers to cost effect
Modifiers to travel time effect
Time of day shift variables
Correlations between variables can cause instability
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
HBW- Imputed VOT by income and occupancy
Income quartile SOV HOV2 HOV3+
Lowest 5.09$ 11.19$ 12.47$
Second 5.94$ 14.51$ 16.75$
Third 6.45$ 16.88$ 19.99$
Highest 9.44$ 42.61$ 70.13$
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
HBW joint mode and TOD-tour-based
8 departure hours from home x 7 arrival hours at home x 6 modes = 336 alternatives
Same mode nesting structure
No conclusive results yet on TOD nesting
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
RP analysis- tasks remaining
Add travel time variability measures for Seattle data
Add more “time pressure” variables to tour-based models – demonstrate value of activity-scheduling approach to influence value of travel time
Test transferability of models to Bay Area (BATS 2000) RP data
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SP data analysis
Three main data sets…
Seattle : General toll scenarios:
Free vs. tolled routes, Peak vs. off-peak
San Francisco: Downtown cordon pricing:
Before, during or after peak, or transit
Los Angeles: HOT/Express lane scenarios:
Free vs. tolled routes; before during or after
peak, or transit (express bus via HOT lane)
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SP analysis approach
Same general approach as for RP, but..
More restricted… which alternatives and variables to include is largely pre-defined by the survey experiment
Tests mainly on segmentation and covariates
Allows “dynamic” analyses, such as departure time switching
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Nesting for Seattle SP data
TOD
nests
Off-peak
period
Peak
period
Free
route off-
peak
Toll
route off-
peak
Free
route-
peak
Toll
route-
peak
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Nesting for San Francisco SP
Mode
nests
Auto
mode
Transit
mode
Auto
before
peak
Auto
during
peak
Auto
after
peak
Public
transit
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Nesting for Los Angeles SP
Level 1: Modes
Auto Transit
Level 2: Periods
Before peak Peak After peak Transit
Level 3: Routes
Toll Free Toll Free Toll Free Transit
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Summary of SP Results- Value of SOV In-
vehicle Time ($/hour)
02468
1012141618202224
Lowest
income
quartile
Second
income
quartile
Third
income
quartile
Highest
income
quartile
Seattle-work
Seattle-non-work
San Fran-work
San Fran-non-work
Los Ang- work
Los Ang-non-work
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Occupancy results-SP
In-vehicle time coefficient for HOV relative to
SOV Work Non-work
Seattle 1.03 1.39
San Francisco 1.12 1.90
Los Angeles 1.28 1.34
No consistent effects on cost (?!)
SP responses may not reflect values of other
vehicle occupants
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Departure time shift disutility- Work - AM peak
-
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
90 80 70 60 50 40 30 20 10 0 10 20 30 40 50 60 70 80 90
Earlier Time shift (min) Later
Equ
iv.
min
ute
s in
-ve
hic
le t
ime
Actual 6-7amActual 7-8amActual 8-9amActual 9-10 am
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Departure time shift disutility- Non-work - AM peak
-
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
90 80 70 60 50 40 30 20 10 0 10 20 30 40 50 60 70 80 90
Earlier Time shift (min) Later
Equ
iv.
min
ute
s in
-ve
hic
le t
ime
Actual 8-9amActual 9-10amActual 10-11amActual 11-12am
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Departure time shift disutility- Non-work - PM peak
-
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
90 80 70 60 50 40 30 20 10 0 10 20 30 40 50 60 70 80 90
Earlier Time shift (min) Later
Equ
iv.
min
ute
s in
-ve
hic
le t
ime
Actual3pm-4 pmActual4pm-5 pmActual5pm-6 pmActual6pm-7 pm
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Reliability results- SP
Seattle – each additional percent chance of a
15+ minute delay is equivalent to about 0.4
minutes travel time, for both work and non-
work
San Francisco – each minute of extra delay at
10% probability is equivalent to about 0.15
minutes of travel time for work, and 0.5
minutes for non-work. Not as statistically significant as Seattle results.
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SP analysis remaining
Very little – possible refinements to previous models
Compare to results of other SP-based studies
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Seattle Traffic Choices data
“Experimental RP” – respondents given an amount of money and then charged by the mile for using main roads
Price varied by time of day/week and facility type
500 vehicles with GPS units, experiment lasted more than a year >>> 750,000 trips
Some data collected during no-toll period
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Traffic Choices analysis approach
Determine PSRC TAZ’s for trip ends
Skim best freeway and non-freeway paths for various times of day and attach to trip records
Analyze toll links actually used to determine the type of path chosen
Analyze toll distance and amount actually paid to confirm path choice
Estimate joint time of day / path type models, separately for HBW and HBO trips
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Initial results
Based on very many observations
Toll variable very insignificant – high
correlations between toll & travel time
Toll/distance or log(toll) gave better results
Nesting of route type under TOD – stronger
for non-work than for work
Freeway route type constant increases with
the fraction of path distance on freeway
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Coeff. T-stat. Coeff. T-stat. Coeff. T-stat.
Nesting routes under TOD 0.735 16.6 0.46 8.1 0.793 19.7
Highway time (min) -0.0272 -30 -0.0251 -24.1 -0.0259 -28.8
Toll (cents) -1.10E-04 -1.2
Toll per mile (cents) -0.0221 -7
Log of toll -0.272 -10.8
Constant-freeway route -0.277 -6.7 -0.233 -5.4 -0.225 -5.5
Fract. Distance on freeway 1.2 13.4 1.62 14.1 1.36 15.6
Initial results (partial) HBW
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Initial results (partial) HBO
Coeff. T-stat. Coeff. T-stat. Coeff. T-stat.
0.173 17.5 0.25 0.25 Nesting routes under TOD
-0.0289 -43.5 -0.0283 -43 -0.0284 -42.9 Highway time (min)
0.0023 18.5 Toll (cents)
-0.0163 -11.2 Toll per mile (cents)
-0.145 -5.4 Log of toll
-0.125 -4.7 -0.138 -5.3 -0.125 -4.8 Constant-freeway route
1.15 18.9 1.75 28.4 1.61 26.4 Fract. Distance on freeway
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Traffic Choice travel time reliability measures
Analyze all GPS travel times between pairs of
network nodes
Determine mean, st.dev., 80th and 90th % time for each node pair/hour period
Aggregate node pairs into TAZ pairs and
calculated weighted average of mean, st.dev.,
80th and 90th % time
For each path type, apply st.dev./mean ratio to the network-based times
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Results using reliability Standard deviation of travel time negative and
significant for HBW – similar to coefficient for
travel time
But… wrong sign for HBO trips.
St.deviation per mile gave similar results
Toll coefficient now negative (included
observations from post-tolling period)
Imputed VOT quite low ($5/hr)
Based on relatively few observations
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Additional results HBW
VOT ($/hr) 5.29$ 4.50$ 4.56$
Coeff. T-stat. Coeff. T-stat. Coeff. T-stat.
Highway time (min) -0.0529 -7.2 -0.0435 -5.7 -0.0479 -6.5
St. dev. Highway time (min) -0.0402 -5.4
St. dev. HW time/distance -0.156 -5.3
Toll (cents) -0.006 -9.1 -0.0058 -8.8 -0.0063 -9.5
Constant-freeway route -3.31 -12.5 -3.28 -12.3 -3.41 -12.8
Fract. Distance on freeway 4.72 11.1 4.57 10.7 4.87 11.5
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Additional results HBO
3.55$ 5.26$ 4.09$ VOT ($/hr)
Coeff. T-stat. Coeff. T-stat. Coeff. T-stat.
-0.0231 -6.9 -0.0342 -8 -0.0266 -7.6 Highway time (min)
0.0355 4.6 St. dev. Highway time (min)
0.0984 3.9 St. dev. HW time/distance
-0.0039 -5.6 -0.0039 -5.7 -0.0039 -5.6 Toll (cents)
-1.01 -7.2 -1.01 -7.2 -0.957 -6.8 Constant-freeway route
2.45 8.4 2.41 8.2 2.36 8.1 Fract. Distance on freeway
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Traffic Choices analysis remaining
Try to extend reliability measures to more node pairs and zone pairs… Use a lower threshold of number of
observations needed
Use only for periods with the most demand (or aggregate across other periods)
Use observed pairs to estimate synthesis equation
Match in household variables (income)
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Summary comparison –absolute values of beta(time)
-0.14 -0.12 -0.1 -0.08 -0.06 -0.04 -0.02 0
HBW mode & tod, tour-based
HBW mode & tod, trip-based
HBO mode & tod, trip-based
HBW mode choice, trip-based
HBW tod choice, trip-based
Traffic Choices- HBW 1
Traffic Choices- HBW 2
Traffic Choices- HBO 1
Traffic Choices- HB0 2
Seattle SP- HBW
Seattle SP- HBO
San Francisco SP- HBW
San Francisco SP- HBO
Los Angeles SP- HBW
Los Angeles SP- HBO
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Conclusions (1)
Consistent willingness to pay for travel time savings across RP and SP data sets and types of models, but…
With important systematic differences by travel purpose, income and car occupancy
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Conclusions (2)
For meaningful analysis of congestion pricing, time of day choice models are essential
Ideally, such models take advantage of data on current departure time or preferred departure or arrival time > time of day shift models
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Conclusions (3)
Consistent evidence about hierarchy of types of travel decisions across RP & SP data sets… Path type choice (toll/non-toll) is at the
lowest level,conditional on other choices
For commuting, similar time periods nested underneath mode choice, but broader time of day periods generally above mode choice (similar to Dutch results)
SHRPII C04: TEG Meeting, Washington, DC - January 14, 2010
Conclusions (4)
The effect of travel time variability is important, above and beyond the mean travel time
For OD-level analysis, std. deviation works better than 80th or 90th pctile, particularly when divided by distance
For application, simulation approaches best in the longer term (L04), but other approaches (link-based measures or synthesized OD measures) may be more feasible in the shorter term