Introduction
-Vehicle ownership is a long-term decision that directly impacts daily mode choice. The choice is influenced by household factors like income and number of workers, but also by where people choose to live. The auto-ownership model allows the TRMG2 to be sensitive to these factors and respond to changes in the future.
+Auto ownership is a long-term decision that directly impacts daily mode choice. The choice is influenced by household factors like income and number of workers, but also by where people choose to live. The auto ownership model allows TRMG2 to be sensitive to these factors and respond to changes in the future.
+Auto ownership is predicted using a discrete choice model. For more details on general model form, click here.
Model Structure
-The auto-ownership model in the TRMG2 makes use of variables from the synthetic population and zonal accessibility to make predictions. The model structure is a simple multinomial logistic (MNL) regression model with five alternatives:
+The auto ownership model in TRMG2 makes use of variables from the synthetic population and zonal accessibility to make predictions. The model structure is a simple multinomial logistic (MNL) regression model with five alternatives:
- 0 Vehicles
- 1 Vehicle @@ -278,76 +285,79 @@
Coefficients
TermCoefficients
Coefficients
The coefficients all have the right sign and the relative sizes are intuitive. One particularly encouraging result of this model is that households with strong walk and transit accessibility are less likely to own a vehicle and even less likely to own multiple vehicles. This adds another dimension of model sensitivity to transit investments. New transit routes will affect long-term household decisions about auto ownership, which further influence their daily decisions about mode choice.
-Looking in more detail, the trend for coefficients across alternatives is also intuitive. Large numbers of workers in a household has a small positive impact on the utility of owning 1 auto, but a large impact on owning 2 or more. High income households are more likely to own more vehicles. Children make it less likely to own 3 or more cars, because there are fewer drivers. Finally, the model’s Rho^2 shows strong predictive power.
+Household income categories are defined here. Accessibility terms are discussed in detail here.
+As an example, the 1-vehicle coefficients in the table above translate into a utility formula that starts like this:
+\(U = 4.88 + 0.7 * Workers + 0.1 * Seniors - 0.4 * WalkAccess + ...\)
+The coefficients all have the right sign and the relative sizes are intuitive. For example, higher-income households are more likely to own more cars. One particularly encouraging result of this model is that households with strong walk and transit accessibility are less likely to own a vehicle and even less likely to own multiple vehicles. This adds another dimension of model sensitivity to transit investments. New transit routes will affect long-term household decisions about auto ownership, which further influence their daily transportation decisions.
+Looking in more detail, the trend for coefficients across alternatives is also intuitive. Large numbers of workers in a household have a small positive impact on the utility of owning 1 auto, but a large impact on owning 2 or more. High income households are more likely to own more vehicles. Finally, the model’s Rho^2 of 0.42 shows strong predictive power.
Calibration
-Even before calibration, the estimated coefficients produced results that closely matched the survey as shown in the table below.
+The estimated coefficients produce results that closely matched the survey as shown in the table below.
--0.0706 +-0.0852 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
-0.1128 +0.0483 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
-0.1454 +0.1047 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
--0.4734 +0.0197 |
Directionality factors are not calculated for commercial vehicles and trucks. Instead, they are assumed to be evenly split by direction.
Distribution
-Once generated, the commercial vehicle and truck trips are distributed among zones using a gravity model. Unfortunately, the CV survey did not contain information on both ends of the trip, which meant it could not be used to estimate model parameters. Model parameters were borrowed from the travel model in Blacksburg, VA and are shown below.
+Once generated, commercial vehicle and truck trips are distributed among zones using a gravity model. Unfortunately, the CV survey did not contain information on both ends of the trip, which meant it could not be used to estimate model parameters. Model parameters were borrowed from the NRVMPO travel model and are shown below.
The table below shows the resulting average trip lengths in the model.
+The table below shows the resulting average commercial vehicle trip lengths in the model.
-7.3 +7.1 |
-9.0 +8.9 |
The charts below show the trip length frequency distribution graphs (in miles).
+CV
+ +SUT
+ +MUT
+ +Internal/External
Caliper Corporation
-March 01, 2022
+April 07, 2022
Internal/External
Introduction
-Travel from outside the Triangle model’s geographic boundaries to locations inside the model’s boundaries are referred to as “external-internal” or “EI” trips; movements from inside the boundary to outside are “internal-external” or “EI” trips; and movements through the region are referred to as “external-external” or “EE” trips. Together, these movements are referred to here as “IEEI” travel. This document discusses the development of an IEEI component of the regional model.
+Travel from outside the Triangle model’s geographic boundary to locations inside the boundary are referred to as “external-internal” or “EI” trips. Movements from inside the boundary to outside are “internal-external” or “EI” trips. Movements through but not stopping within the region are referred to as “external-external” or “EE” trips. Together, these travel patterns are referred to here as “EE” and “IE/EI” travel. This document discusses the development of these components of the regional model.
Available Data Sources
-The two best available data sources for movements through the modeling region are (i) StreetLight Data extracted by ITRE and (ii) a sub-area extraction from the North Carolina Statewide Travel Model (NCSTM). Our analysis suggests that, for our purposes, the NCSTM data is superior. The reason for this is that there appears to be an error with either the StreetLight Data or its extraction. The map below shows the “attractions” for the IE/EI trips from the StreetLight data, assuming the “E” end of the trip is the production end and the I end of the trip is the attraction end, and aggregated to a 21-district geographies defined by ITRE. The red dots show data extracted using the CAMPO zone system and the blue dots show data extracted using the Durham MPO zone system. The map shows an illogical outcomes of attractions clustured in select locations near the external stations. Given these odd results, we move forward with the NCSTM data, which, as demonstrated in the balance of the document, revealed logical results.
+The two best available data sources for external-related travel are (i) StreetLight Data provided by the client, and (ii) a sub-area extraction from the North Carolina Statewide Travel Model (NCSTM v.4.3a). The analysis suggests that the NCSTM data is superior because there appears to be an error with either the StreetLight Data or its extraction. The map below shows “attractions” for the IE/EI trips from StreetLight. The “E” end of the trip is asserted to be the production end and the “I” end of the trip is the attraction end. Finally, the data was aggregated to a 21-district geographies defined by ITRE. The red dots show data extracted using the CAMPO zone system and the blue dots show data extracted using the DCHC MPO zone system. The map shows an illogical outcome of attractions clustered in select locations near the external stations. Nearly all the DCHC trip ends were in Chatham County while CAMPO trip ends were predominantly in Franklin County.
StreetLight Attractions
@@ -291,7 +289,7 @@StreetLight Attractions
EE
-The first step in the analysis is to determine the share of traffic at each external station that can be attributed to EE movements. This will allow us to attribute the balance of the travel to internal-external (IE) and external-internal (EI) movements. The NCSTM data contains flow estimates for each of the model’s 97 EE station pairs. We use these flows to estimate the share of travel at each of the external stations using the ADT estimates for the external stations derived from the NCSTM. In addition, the share of commercial vehicles are segmented by single-unit trucks and multi-unit trucks. The table below summarizes these outcomes.
+The first step is to determine the share of traffic at each external station that can be attributed to EE travel. This allows the balance of trips to be attributed to IE and EI travel. The NCSTM data contains flow estimates for each of the model’s 97 external stations. These flows are used to estimate the share of travel using the AWDT estimates derived from the NCSTM. In addition, the share of commercial vehicles are segmented by single-unit trucks and multi-unit trucks. The table below summarizes these outcomes.
EE Traffic Shares
Station | --Avg Daily Traffic + | +Avg Weekday Traffic |
Pct EE Automobile
@@ -322,1020 +320,1020 @@ EE Traffic Shares | 3235 | --3523.9 + | +3,524 | -30.1 +30 | -20.8 +21 | -3.5 +3 | -17.3 +17 | ||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
3236 | --927.2 + | +927 | -24.2 +24 | -12.8 +13 | -4.9 +5 | -7.8 +8 | ||||||||||||||||||
3237 | --2176.4 + | +2,176 | -22.4 +22 | -8.6 +9 | -2.7 +3 | -5.8 +6 | ||||||||||||||||||
3242 | --1701.1 + | +1,701 | -24.0 +24 | -13.0 +13 | -4.0 +4 | -9.1 +9 | ||||||||||||||||||
3243 | --1567.7 + | +1,568 | -38.4 +38 | -8.7 +9 | -3.5 +3 | -5.2 +5 | ||||||||||||||||||
3245 | --3931.1 + | +3,931 | -15.3 +15 | -2.0 +2 | -1.8 +2 | -0.2 +0 | ||||||||||||||||||
3248 | --1902.2 + | +1,902 | -26.8 +27 | -27.6 +28 | -3.8 +4 | -23.7 +24 | ||||||||||||||||||
3252 | --25645.1 + | +25,645 | -53.5 +54 | -13.8 +14 | -3.0 +3 | -10.8 +11 | ||||||||||||||||||
3253 | --2559.5 + | +2,560 | -9.0 +9 | -16.1 +16 | -4.9 +5 | -11.2 +11 | ||||||||||||||||||
3254 | --3656.1 + | +3,656 | -7.9 +8 | -9.4 +9 | -2.5 +3 | -6.9 +7 | ||||||||||||||||||
3257 | --7315.2 + | +7,315 | -8.7 +9 | -14.9 +15 | -1.8 +2 | -13.1 +13 | ||||||||||||||||||
3260 | --4087.5 + | +4,087 | -44.8 +45 | -8.1 +8 | -3.4 +3 | -4.7 +5 | ||||||||||||||||||
3261 | --1775.7 + | +1,776 | -51.7 +52 | -9.9 +10 | -3.7 +4 | -6.2 +6 | ||||||||||||||||||
3263 | --2057.6 + | +2,058 | -9.0 +9 | -10.4 +10 | -3.9 +4 | -6.4 +6 | ||||||||||||||||||
3265 | --1198.9 + | +1,199 | -56.5 +56 | -11.5 +11 | -1.9 +2 | -9.5 +10 | ||||||||||||||||||
3267 | --1682.4 + | +1,682 | -18.5 +18 | -12.5 +12 | -5.5 +5 | -7.0 +7 | ||||||||||||||||||
3272 | --1171.9 + | +1,172 | -4.1 +4 | -11.5 +12 | -6.8 +7 | -4.7 +5 | ||||||||||||||||||
3273 | --15209.7 + | +15,210 | -31.0 +31 | -14.9 +15 | -2.4 +2 | -12.5 +12 | ||||||||||||||||||
3274 | --1027.1 + | +1,027 | -2.6 +3 | -7.9 +8 | -4.6 +5 | -3.3 +3 | ||||||||||||||||||
3275 | --2141.6 + | +2,142 | -69.8 +70 | -5.0 +5 | -2.9 +3 | -2.1 +2 | ||||||||||||||||||
3276 | --106.6 + | +107 | -76.3 +76 | -3.0 +3 | -3.0 +3 | -0.0 +0 | ||||||||||||||||||
3277 | --20162.0 + | +20,162 | -38.0 +38 | -15.0 +15 | -2.1 +2 | -12.9 +13 | ||||||||||||||||||
3280 | --2328.7 + | +2,329 | -0.0 +0 | -4.5 +4 | -2.3 +2 | -2.2 +2 | ||||||||||||||||||
3281 | --26150.0 + | +26,150 | -83.0 +83 | -16.3 +16 | -3.9 +4 | -12.4 +12 | ||||||||||||||||||
3282 | --3483.4 + | +3,483 | -34.2 +34 | -9.7 +10 | -3.0 +3 | -6.6 +7 | ||||||||||||||||||
3283 | --4173.8 + | +4,174 | -37.3 +37 | -7.3 +7 | -3.9 +4 | -3.3 +3 | ||||||||||||||||||
3289 | --15915.0 + | +15,915 | -21.9 +22 | -6.5 +6 | -2.4 +2 | -4.0 +4 | ||||||||||||||||||
3291 | --2201.1 + | +2,201 | -14.0 +14 | -7.8 +8 | -3.2 +3 | -4.6 +5 | ||||||||||||||||||
3295 | --12093.1 + | +12,093 | -39.1 +39 | -9.1 +9 | -2.1 +2 | -7.0 +7 | ||||||||||||||||||
3296 | --1917.4 + | +1,917 | -46.8 +47 | -8.3 +8 | -3.8 +4 | -4.5 +5 | ||||||||||||||||||
3297 | --964.8 + | +965 | -10.1 +10 | -7.4 +7 | -2.0 +2 | -5.4 +5 | ||||||||||||||||||
3298 | --37430.7 + | +37,431 | -66.3 +66 | -16.4 +16 | -3.4 +3 | -13.1 +13 | ||||||||||||||||||
3299 | --3600.0 + | +3,600 | -16.1 +16 | -6.3 +6 | -5.7 +6 | -0.6 +1 | ||||||||||||||||||
3301 | --2107.3 + | +2,107 | -0.0 +0 | -1.5 +1 | -1.1 +1 | -0.3 +0 | ||||||||||||||||||
3302 | --2683.5 + | +2,684 | -0.0 +0 | -3.4 +3 | -2.2 +2 | -1.2 +1 | ||||||||||||||||||
3303 | --6359.3 + | +6,359 | -36.6 +37 | -7.4 +7 | -3.8 +4 | -3.6 +4 | ||||||||||||||||||
3305 | --14573.4 + | +14,573 | -19.5 +19 | -6.7 +7 | -2.6 +3 | -4.0 +4 | ||||||||||||||||||
3306 | --2101.5 + | +2,102 | -1.4 +1 | -9.7 +10 | -4.1 +4 | -5.6 +6 | ||||||||||||||||||
3308 | --17249.9 + | +17,250 | -12.0 +12 | -11.0 +11 | -2.0 +2 | -9.1 +9 | ||||||||||||||||||
3309 | --3699.7 + | +3,700 | -24.7 +25 | -10.0 +10 | -2.4 +2 | -7.6 +8 | ||||||||||||||||||
3311 | --1567.9 + | +1,568 | -2.6 +3 | -6.4 +6 | -2.4 +2 | -4.0 +4 | ||||||||||||||||||
3313 | --10491.2 + | +10,491 | -15.1 +15 | -12.9 +13 | -1.8 +2 | -11.1 +11 | ||||||||||||||||||
3317 | --1716.9 + | +1,717 | -65.3 +65 | -20.5 +21 | -4.6 +5 | -15.9 +16 | ||||||||||||||||||
3318 | --1950.3 + | +1,950 | -1.6 +2 | -5.9 +6 | -4.3 +4 | -1.6 +2 | ||||||||||||||||||
3319 | --545.3 + | +545 | -2.2 +2 | -18.4 +18 | -9.1 +9 | -9.2 +9 | ||||||||||||||||||
3321 | --10281.3 + | +10,281 | -27.4 +27 | -2.4 +2 | -1.4 +1 | -1.0 +1 | ||||||||||||||||||
3322 | --7153.6 + | +7,154 | -42.6 +43 | -2.5 +2 | -2.2 +2 | -0.3 +0 | ||||||||||||||||||
3323 | --68297.2 + | +68,297 | -37.8 +38 | -12.4 +12 | -2.4 +2 | -10.0 +10 | ||||||||||||||||||
3326 | --2710.5 + | +2,711 | -4.0 +4 | -6.9 +7 | -4.0 +4 | -2.9 +3 | ||||||||||||||||||
3328 | --2870.9 + | +2,871 | -42.8 +43 | -11.3 +11 | -3.4 +3 | -7.9 +8 | ||||||||||||||||||
3329 | --1352.2 + | +1,352 | -37.5 +38 | -12.5 +13 | -2.4 +2 | -10.2 +10 |
-Term - | --Productions - | -
---|---|
-Productions - | --NA - | -
-Population - | --0.676 - | -
-Employment - | --0.187 - | -
The positive correletions move us to the next step of estimating attraction models for the internal end of the IE/EI movements, assigning the external station the production end.
-The number of IE/EI trips are controlled by the volumes at the external station after external travel has been subtracted. These trips are distributed using a gravity model. For this model form, attractions and gamma parameters must be estimated.
+IE/EI Attraction Model Estimation
-The NCSTM data was aggregated to the 43 district system to support model estimation. A handful of specifications were tested to determine a useful attraction model, including segmenting the attraction models for Freeway and non-Freeway external stations. The preferred model generated positive coefficients on the population and employment variables. It does not segment the stations by roadway type and asserts a Y-intercept of zero.
-Preferred
+The NCSTM data was aggregated to the 43-district system to support model estimation. A handful of specifications were tested to determine a useful attraction model, including segmenting the attraction models for freeway and non-freeway external stations. These more complicated models generated coefficients with wrong signs. The final, simple model generated positive coefficients on the population and employment variables. It does not segment the stations by roadway type and asserts a Y-intercept of zero.
-0.0059 +0.0045 | -0.0648 +0.0492 | -0.9487 +0.9610 |
-0.1551 +0.1559 | -2.8129 +2.8185 | -0.0076 +0.0075 |
-0.5013 +0.5008 |
NA
@@ -1458,332 +1412,199 @@ Preferred |
A scatter plot of the NCSTM and estimated attractions by the 43-district system is shown in the plot below.
Freeway Model
+NCSTM and Estimated IE/EI Attractions
+ +Distance Assessment
+The chart below plots the trip length frequency, with the trip length measured from the external station to the internal zone segmenting trips that enter the region at freeways from those at other locations. The plot shows a logical and expected outcome: trips entering the region at freeway external stations travel longer distances than those entering at non-freeway locations.
+IE/EI Trip Length Frequency
+ +The above chart suggests the IE/EI distribution model should be segmented by freeway and non-freeway stations. The estimated gravity parameters are shown below.
-Variable +Type | -Estimate +a | -t-statistic - | --p value +b |
---|---|---|---|
-intercept - | --1950.0099 - | --2.1975 - | --0.0342 - | -
-Employment - | --0.0657 - | --1.8059 - | --0.0789 - | -
-Population - | --0.0438 +Freeway | -1.7447 +5 | -0.0891 +0.25 |
-Adjusted R-squared - | --0.4699 +NonFreeway | -NA +5 | -NA +0.60 |
The model estimated distances for freeways and non-freeways are plotted against the observed data in the two following figures.
+IE/EI Observed and Estimated Trip Length Frequency for Freeway Stations
+ +IE/EI Observed and Estimated Trip Length Frequency for Non-Freeway Stations
+ +A table summarizing the mean trip lengths is presented below.
Non-Freeway Model
+IE/EI Average Trip Lengths
-Variable - | --Estimate +Category | --t-statistic + | +Source | -p value +Mean Distance (miles) |
---|---|---|---|---|
-intercept - | --4411.9350 +Freeway | --3.1756 + | +Estimated | -0.0030 +21.6 |
-Employment - | ---0.0321 +Freeway | ---0.5641 + | +Observed | -0.5760 +26.0 |
-Population - | --0.0231 +Non-freeway | --0.5867 + | +Estimated | -0.5609 +14.5 |
-Adjusted R-squared - | ---0.0427 +Non-freeway | --NA + | +Observed | -NA +15.9 |
Combined with Intercept
+Final adjustments
+During validation, The IE/EI model was enhanced to allow additional control of trip lengths. This is due to the existence of some town along the border. Stations near those towns need even shorter trip lengths. Two additional sets of gravity parameters were created and the terminology was changed as shown below.
-Variable +Type | -Estimate +a | -t-statistic - | --p value +b |
---|---|---|---|
-intercept - | --6361.9449 - | --3.1567 - | --0.0031 - | -
-Employment - | --0.0335 - | --0.4063 - | --0.6868 - | -
-Population - | --0.0669 +Long | -1.1727 +5 | -0.2482 +0.25 |
-Adjusted R-squared - | --0.1222 +Medium | -NA +5 | -NA +0.60 |
A scatter plot of the NCSTM and estimated attractions by the 43-district system is shown in the plot below.
-NCSTM and Estimated IE/EI Attractions
- -Distance Assessment
-In the chart below, we plot the trip length frequency, with the trip length measured from the external station to the internal zone, for the NCSTM trips, segmenting trips that enter the region through external stations at Freeways from those at other locations. The plot shows a logical and expected outcome: trips entering the region at Freeway external stations travel longer than those entering at non-freeway locations.
-IE/EI Trip Length Frequency
- -The above chart suggests the IE/EI distribution model should be segmented by station type, freeway or non-freeway. The external stations identified as freeways are as follows: 3298, 3252, 3305, 3277, 3308, 3281, 3273, 3313, 3323.
-Using the friction factors from the existing IE/EI model, the model estimated distances for freeways and non-freeways are plotted against the observed data in the two following figures.
-IE/EI Observed and Estimated Trip Length Frequency for Freeway Stations
- -IE/EI Observed and Estimated Trip Length Frequency for Non-Freeway Stations
- -A table summarizing the mean trip lengths is presented below.
-IE/EI Average Trip Lengths
--Category - | --Source - | --Mean Distance (miles) - | -
---|---|---|
-Freeway - | --Estimated +Sort | -21.6 - | -
-Freeway - | --Observed +5 | -26.0 +1.30 |
-Non-freeway - | --Estimated +Very Short | -14.5 - | -
-Non-freeway - | --Observed +5 | -15.9 +1.60 |
The first two options (“Long” and “Medium”) are the same coefficients as estimated. The final two coefficients were arrived at through iterative trials during highway validation.
IE/EI Time of Day
diff --git a/docs/img/cv/cv.png b/docs/img/cv/cv.png new file mode 100644 index 00000000..353a45a2 Binary files /dev/null and b/docs/img/cv/cv.png differ diff --git a/docs/img/cv/mut.png b/docs/img/cv/mut.png new file mode 100644 index 00000000..e144444c Binary files /dev/null and b/docs/img/cv/mut.png differ diff --git a/docs/img/cv/sut.png b/docs/img/cv/sut.png new file mode 100644 index 00000000..4c7a2cca Binary files /dev/null and b/docs/img/cv/sut.png differ diff --git a/docs/img/mc/asc_calibration.png b/docs/img/mc/asc_calibration.png new file mode 100644 index 00000000..76d1011f Binary files /dev/null and b/docs/img/mc/asc_calibration.png differ diff --git a/docs/img/mc/fta_path_constants.png b/docs/img/mc/fta_path_constants.png new file mode 100644 index 00000000..7967246e Binary files /dev/null and b/docs/img/mc/fta_path_constants.png differ diff --git a/docs/img/mc/fta_visibility.png b/docs/img/mc/fta_visibility.png new file mode 100644 index 00000000..3bbdb460 Binary files /dev/null and b/docs/img/mc/fta_visibility.png differ diff --git a/docs/img/mc/full_tree.png b/docs/img/mc/full_tree.png new file mode 100644 index 00000000..04ca8f09 Binary files /dev/null and b/docs/img/mc/full_tree.png differ diff --git a/docs/img/mc/lr_adjustment.png b/docs/img/mc/lr_adjustment.png new file mode 100644 index 00000000..c5d40186 Binary files /dev/null and b/docs/img/mc/lr_adjustment.png differ diff --git a/docs/img/mc/n_hb_k12_1.png b/docs/img/mc/n_hb_k12_1.png new file mode 100644 index 00000000..c3e2ca68 Binary files /dev/null and b/docs/img/mc/n_hb_k12_1.png differ diff --git a/docs/img/mc/n_hb_k12_2.png b/docs/img/mc/n_hb_k12_2.png new file mode 100644 index 00000000..9d088ff4 Binary files /dev/null and b/docs/img/mc/n_hb_k12_2.png differ diff --git a/docs/img/mc/n_hb_od_long1.png b/docs/img/mc/n_hb_od_long1.png new file mode 100644 index 00000000..4e839518 Binary files /dev/null and b/docs/img/mc/n_hb_od_long1.png differ diff --git a/docs/img/mc/n_hb_od_long2.png b/docs/img/mc/n_hb_od_long2.png new file mode 100644 index 00000000..373cd971 Binary files /dev/null and b/docs/img/mc/n_hb_od_long2.png differ diff --git a/docs/img/mc/n_hb_od_long3.png b/docs/img/mc/n_hb_od_long3.png new file mode 100644 index 00000000..531f5ac1 Binary files /dev/null and b/docs/img/mc/n_hb_od_long3.png differ diff --git a/docs/img/mc/n_hb_od_long4.png b/docs/img/mc/n_hb_od_long4.png new file mode 100644 index 00000000..21510408 Binary files /dev/null and b/docs/img/mc/n_hb_od_long4.png differ diff --git a/docs/img/mc/n_hb_od_short1.png b/docs/img/mc/n_hb_od_short1.png new file mode 100644 index 00000000..dfd4822e Binary files /dev/null and b/docs/img/mc/n_hb_od_short1.png differ diff --git a/docs/img/mc/n_hb_od_short2.png b/docs/img/mc/n_hb_od_short2.png new file mode 100644 index 00000000..3a2fe43b Binary files /dev/null and b/docs/img/mc/n_hb_od_short2.png differ diff --git a/docs/img/mc/n_hb_od_short3.png b/docs/img/mc/n_hb_od_short3.png new file mode 100644 index 00000000..286284b0 Binary files /dev/null and b/docs/img/mc/n_hb_od_short3.png differ diff --git a/docs/img/mc/n_hb_od_short4.png b/docs/img/mc/n_hb_od_short4.png new file mode 100644 index 00000000..6976c448 Binary files /dev/null and b/docs/img/mc/n_hb_od_short4.png differ diff --git a/docs/img/mc/n_hb_ome1.png b/docs/img/mc/n_hb_ome1.png new file mode 100644 index 00000000..ce4835cd Binary files /dev/null and b/docs/img/mc/n_hb_ome1.png differ diff --git a/docs/img/mc/n_hb_ome2.png b/docs/img/mc/n_hb_ome2.png new file mode 100644 index 00000000..6c5039d9 Binary files /dev/null and b/docs/img/mc/n_hb_ome2.png differ diff --git a/docs/img/mc/n_hb_ome3.png b/docs/img/mc/n_hb_ome3.png new file mode 100644 index 00000000..7d4371c7 Binary files /dev/null and b/docs/img/mc/n_hb_ome3.png differ diff --git a/docs/img/mc/n_hb_ome4.png b/docs/img/mc/n_hb_ome4.png new file mode 100644 index 00000000..6c20681e Binary files /dev/null and b/docs/img/mc/n_hb_ome4.png differ diff --git a/docs/img/mc/n_hb_omed1.png b/docs/img/mc/n_hb_omed1.png new file mode 100644 index 00000000..428b4447 Binary files /dev/null and b/docs/img/mc/n_hb_omed1.png differ diff --git a/docs/img/mc/n_hb_omed2.png b/docs/img/mc/n_hb_omed2.png new file mode 100644 index 00000000..ffcca854 Binary files /dev/null and b/docs/img/mc/n_hb_omed2.png differ diff --git a/docs/img/mc/n_hb_omed3.png b/docs/img/mc/n_hb_omed3.png new file mode 100644 index 00000000..eb98c623 Binary files /dev/null and b/docs/img/mc/n_hb_omed3.png differ diff --git a/docs/img/mc/n_hb_omed4.png b/docs/img/mc/n_hb_omed4.png new file mode 100644 index 00000000..3b48c288 Binary files /dev/null and b/docs/img/mc/n_hb_omed4.png differ diff --git a/docs/img/mc/nhb_full_nest.png b/docs/img/mc/nhb_full_nest.png new file mode 100644 index 00000000..e783e39d Binary files /dev/null and b/docs/img/mc/nhb_full_nest.png differ diff --git a/docs/img/mc/transit_sub_other1.png b/docs/img/mc/transit_sub_other1.png new file mode 100644 index 00000000..dfd9354d Binary files /dev/null and b/docs/img/mc/transit_sub_other1.png differ diff --git a/docs/img/mc/transit_sub_other2.png b/docs/img/mc/transit_sub_other2.png new file mode 100644 index 00000000..11687307 Binary files /dev/null and b/docs/img/mc/transit_sub_other2.png differ diff --git a/docs/img/mc/transit_sub_other3.png b/docs/img/mc/transit_sub_other3.png new file mode 100644 index 00000000..a9313dde Binary files /dev/null and b/docs/img/mc/transit_sub_other3.png differ diff --git a/docs/img/mc/transit_sub_shop1.png b/docs/img/mc/transit_sub_shop1.png new file mode 100644 index 00000000..2cc84154 Binary files /dev/null and b/docs/img/mc/transit_sub_shop1.png differ diff --git a/docs/img/mc/transit_sub_shop2.png b/docs/img/mc/transit_sub_shop2.png new file mode 100644 index 00000000..7985b37b Binary files /dev/null and b/docs/img/mc/transit_sub_shop2.png differ diff --git a/docs/img/mc/transit_sub_work.png b/docs/img/mc/transit_sub_work.png new file mode 100644 index 00000000..57de8617 Binary files /dev/null and b/docs/img/mc/transit_sub_work.png differ diff --git a/docs/img/mc/transit_sub_work2.png b/docs/img/mc/transit_sub_work2.png new file mode 100644 index 00000000..1f6aa240 Binary files /dev/null and b/docs/img/mc/transit_sub_work2.png differ diff --git a/docs/img/mc/transit_sub_work3.png b/docs/img/mc/transit_sub_work3.png new file mode 100644 index 00000000..6b3ff75d Binary files /dev/null and b/docs/img/mc/transit_sub_work3.png differ diff --git a/docs/img/mc/transit_sub_work4.png b/docs/img/mc/transit_sub_work4.png new file mode 100644 index 00000000..06b93249 Binary files /dev/null and b/docs/img/mc/transit_sub_work4.png differ diff --git a/docs/img/mc/w_hb_o1.png b/docs/img/mc/w_hb_o1.png new file mode 100644 index 00000000..66fe3208 Binary files /dev/null and b/docs/img/mc/w_hb_o1.png differ diff --git a/docs/img/mc/w_hb_o2.png b/docs/img/mc/w_hb_o2.png new file mode 100644 index 00000000..dfc924c9 Binary files /dev/null and b/docs/img/mc/w_hb_o2.png differ diff --git a/docs/img/mc/w_hb_o3.png b/docs/img/mc/w_hb_o3.png new file mode 100644 index 00000000..934e596f Binary files /dev/null and b/docs/img/mc/w_hb_o3.png differ diff --git a/docs/img/mc/w_hb_o4.png b/docs/img/mc/w_hb_o4.png new file mode 100644 index 00000000..ce2e5bf3 Binary files /dev/null and b/docs/img/mc/w_hb_o4.png differ diff --git a/docs/img/mc/w_hb_o5.png b/docs/img/mc/w_hb_o5.png new file mode 100644 index 00000000..616eba4b Binary files /dev/null and b/docs/img/mc/w_hb_o5.png differ diff --git a/docs/img/mc/w_hb_w1.png b/docs/img/mc/w_hb_w1.png new file mode 100644 index 00000000..7fc79e93 Binary files /dev/null and b/docs/img/mc/w_hb_w1.png differ diff --git a/docs/img/mc/w_hb_w2.png b/docs/img/mc/w_hb_w2.png new file mode 100644 index 00000000..b0cd4826 Binary files /dev/null and b/docs/img/mc/w_hb_w2.png differ diff --git a/docs/img/mc/w_hb_w3.png b/docs/img/mc/w_hb_w3.png new file mode 100644 index 00000000..9a80ad75 Binary files /dev/null and b/docs/img/mc/w_hb_w3.png differ diff --git a/docs/img/mc/w_hb_w4.png b/docs/img/mc/w_hb_w4.png new file mode 100644 index 00000000..9e1f7a61 Binary files /dev/null and b/docs/img/mc/w_hb_w4.png differ diff --git a/docs/index.html b/docs/index.html index 7699a6ab..92333503 100644 --- a/docs/index.html +++ b/docs/index.html @@ -7,7 +7,7 @@ - +Triangle Regional Model G2
Caliper Corporation
-2022-02-22
+2022-03-30
On-board Survey Processing
Introduction
-On-board surveys ask transit riders a series of question while they are on the bus in order to obtain detailed information about demographics, trip purpose, modes used, and other variables. These targeted surveys are critical for understanding transit users in the region. Due to the small size of the transit market in the Triangle and the small household survey sample size, the household survey does not contain enough transit trips to estimate the behavior of these travelers.
-In the Triangle, a survey of GoTriangle and DCHC transit was conducted in 2014 (excluding Duke Transit). CAMPO agencies were surveyed in 2015 along with new GoTriangle routes. These are used to gain insight into the transit market.
+Due to the small size of the transit market in the Triangle and the small household survey sample size, the household survey does not contain enough transit observations to estimate behavioral models for transit users. On-board surveys are critical to fill this gap. They are intercept surveys, which means they are conducted on the bus itself. They ask transit riders a series of questions in order to obtain detailed information about demographics, trip purpose, modes used, and other variables.
+In the Triangle, a survey of 9,231 GoTriangle and DCHC area transit riders was conducted in 2014 (excluding Duke Transit). 5,390 CAMPO area transit riders were surveyed in 2015 along with new GoTriangle routes. These are used to gain insight into the transit market.
Combining surveys
-Unlike the household surveys, the transit surveys were collected by different companies. Combining the surveys required translating the different field codes from each into a combined set of values. The surveys collected the same basic information, which made this process easier. Additionally, because different routes were collected in each survey, no routes were double counted and no adjustments to the sample weights were needed.
+Unlike the household surveys, the transit surveys were collected by different companies. Combining the surveys required translating different field codes into a combined set of values. The surveys collected the same basic information, which made this process easier. Additionally, because different routes were collected in each survey, no routes were double counted and no adjustments to the sample weights were needed.
Geocoding
-As with the household survey, the transit trip latitude/longitude values were translated into TRMG2 TAZs. For a transit trip, this includes the home, origin, destination, boarding, and alighting locations. A sample of this is shown in the table below:
+As with the household survey, the transit trip latitude/longitude values were translated into TRMG2 TAZs. For a transit trip, this includes the home, origin (which could be somewhere other than home), destination, boarding, and alighting locations. A sample of this is shown in the table below:
These differences are large, but manual path checking did not uncover any errors. This test assignment was performed on a network with a rough initial guess of travel times, which likely contributed to these differences. Further differences may be because the surveys were from 2014 and 2015 while the transit network is 2016. Any changes in route alignments or offerings would lead to higher discrepancies.
-More attention will be paid to agency- and route-level ridership comparisons during model validation. It may be that fares or other route attributes received from stakeholders need to be adjusted. One positive result of this early test was that all routes received ridership, which indicates no critical accessibility flaws in the network.
+These differences are large, but manual path checking did not uncover any errors. This test assignment was performed on a network with a rough initial guess of travel times, which contributed to these differences. Further differences may be because the surveys were from 2014 and 2015 while the transit network is 2016. Any changes in route alignments or offerings would lead to higher discrepancies.
+During model development, much more accurate roadway times were calculated from actual model assignments. Further, the transit network settings were also dialed in to achieve final results that look much more reasonable.
Parking Model
Introduction
-The parking model component of the TRMG2 determines where auto travelers will park and either walk or take transit to their ultimate destination. The model is only applied to CBD and campus areas where parking is generally limited and paid for.
-In the TRMG2, the results of the parking model influence both mode and destination choices, which means travelers react realistically to changes in both parking availability and price.
+The parking model component of the TRMG2 determines where auto travelers will park and either walk or take transit to their ultimate destination. The model is only applied to CBD and university areas where parking is generally limited and requires payment.
+In TRMG2, results of the parking model influence both mode choice and destination choice, which means travelers react realistically to changes in parking availability and price.
Design
-The parking model is applied at the end of the demand model stream, just before assignment. Trip matrices from the preceding demand models are combined across purposes maintaining only segmentation between trips on work tours and trips on non-work tours (and maintaining SOV and HOV forassignment classes – but applying the same models to them).
-The model is adapted from the one estimated as part of the 2016 Triangle Region Parking Behavior Study but reformulated for application in the regional model. The estimated model did not include the lowest level choices of parking zone, which Caliper has added. In addition, the TRMG2 model only includes “Park and Walk” and “Auto Intercept” options (“Drive and Park” and “Drive, Park & Shuttle” in the 2016 study). The regional model handles transit upstream in the mode choice model. However, for the auto intercept mode, the modeled transit system is used directly for path building. This ensures travelers see accurate options rather than asserted average headways and fares.
-This design is a departure from the previous generation of the TRM in grouping the Auto Intercept mode as part of the parking choices at the end of the model stream rather than as part of the main mode choice model. The 2016 study supports this structure as evidenced by nesting parameters less than one for its Auto nest which included both the “Drive and Park” and the “Drive, Park & Shuttle” modes. This structure and original nesting coefficient in the 2016 study imply that travelers view the auto intercept mode as a parking strategy and not a main mode choice.
+The parking model is applied at the end of the demand model stream, just before traffic assignment. Trip matrices from the preceding demand models are combined across purposes maintaining segmentation between trips on work tours and non-work tours (SOV and HOV are also maintained for assignment classes, but the same models are applied to them).
+The TRMG2 parking model is adapted from the one estimated for the 2016 Triangle Region Parking Behavior Study conducted by RSG. It has been reformulated for application in the regional model. Several differences exist between the model estimated by RSG and the TRMG2 model. Cailper added the lowest level choices of parking zones. In addition, the TRMG2 model only includes “Park and Walk” and “Park and Transit” options. For “Park and Transit”, the G2 uses the actual transit network for path building. This ensures travelers see accurate options and make appropriate decisions.
+The G2 design is a departure from the previous generation of the TRM, which included “Park and Walk” (called “Auto Intercept”) as a primary mode in the mode choice model. G2 effectively treats it as a subchoice under the auto mode. The structure and nesting coefficient in the 2016 RSG study imply that travelers view the auto intercept mode as a parking strategy and not a main mode choice.
-The applied parking model in the TRM is a two-level, nested logit combined parking mode and destination choice model. Various coefficient values and implied values of time differ from those in the 2016 report for a number of reasons including that the RP to SP scaling has been applied here, the model has been calibrated, and because the values of time reported in the 2016 report were generally scaled in comparison to willingness to pay transit fare rather than parking costs.
-The model was calibrated to the 2016 survey in two steps. The model’s AM period was calibrated since the survey only asked about the first trip of the day downtown / to campus and was strongly skewed towards AM trips. First, the lower level, parking destination choice was calibrated by scaling the non-size terms to match the observed average walk time from the survey for the park and walk mode which was 7.8 minutes for CBD areas and 11.25 minutes for campus areas. The park and shuttle coefficients were scaled consistent with park and walk rather than calibrated separately both in order to biasing the mode choice and because the 2016 survey would have required significant additional processing to obtain the time from parking to destination for the park and shuttle mode. Second, the upperl level parking mode choice was calibrated by adjusting the alternative specific bias constant for park and shuttle to match observed shares for the mode which were 6.9% for CBD areas and 14.6% for campus areas.
-A parameter was also added to allow the model to better reflect satellite parking lots like the Friday Center at UNC. In the absence of actual counts of parking at the Friday Center, the parameter was simply asserted to produce a roughly reasonable amount of parking of about 250.
- +The parking model in TRMG2 is a two-level, nested logit model that combines parking mode and destination choice. Various coefficient values and implied values of time differ from those in 2016 for a number of reasons. This includes Caliper’s scaling from Stated Preference to Revealed Preference, calibration of the model, and because the values of time reported in the 2016 report were generally scaled to transit fares rather than parking costs.
+The model was calibrated to the 2016 survey in two steps. The model’s AM period was calibrated since the survey only asked about the first trip of the day to downtown / campus and was strongly skewed towards AM trips. First, the lower level, parking destination choice was calibrated by scaling the non-size terms to match the observed average walk time from the survey for the park-and-walk mode which was 7.8 minutes for CBD areas and 11.25 minutes for campus areas. The park-and-shuttle coefficients were scaled consistent with park and walk rather than calibrated separately. This was both in order to bias the mode choice and because the 2016 survey would have required significant additional processing to obtain the time from parking to destination for the park-and-shuttle mode. Second, the upper level parking mode choice was calibrated by adjusting the alternative specific bias constant for park and shuttle to match observed shares for the mode which were 6.9% for CBD areas and 14.6% for campus areas.
+A parameter was also added to allow the model to better reflect satellite parking lots like the Friday Center at UNC. In the absence of actual parking counts at the Friday Center, the parameter was simply asserted to produce a reasonable amount of about 250 parked vehicles.
+Resident Destination Choice
Intro
- Nested Logit (NL) models were estimated for the home-based trip purposes while multinomial logit (MNL) models were estimated for non-home-based trip purposes. NL models were attempted for the latter set of purposes but did not provide justifiably better models.
- All models were estimated using Larch (Newman, 2021). @@ -359,26 +354,32 @@
- Size term: This is the natural logarithm of the attractions at a destination zone, which itself is a linear combination of employment variables and their coefficients. These coefficients are simultaneously estimated with other parameters. Since it is required that these coefficients be positive, the utility formulation specifies coefficients that are then exponentiated. Successful estimation requires that one of these attraction variable coefficients is fixed. The estimation also provides a coefficient for the size variable, which theoretically should be between 0 and 1. -
- Mode Choice Logsums: Using mode choice root logsums typically presents a problem in destination choice. The root logsum value is dominated by the auto nest which washes out any effects from the non-dominant modes. For example, drastically improving the transit accessibility to a particular zone will not affect the probability of choosing that zone, if the transit logsums are dwarfed by the auto logsums. To circumvent this issue, separate auto, transit, and non-household auto logsums from the respective nests are used in the TRM specifications. The logsums are segmented by vehicle sufficiency market segments for added explanatory power. The work logsums are further distinguished by high- and low-income segments. Finally, note that there is no auto logsum component for the zero-vehicle households market segment. -
- 4D measures: Several 4D measures such as transit accessibility, hospital accessibility, walk accessibility, employment densities are considered. -
- Time Coefficient(s): The auto times corresponding to the latest model skims are attached depending upon the time of day the respondent made the trip. A time coefficient is estimated in each specification. -
- Intra-Cluster effects: Intra-Cluster coefficients are estimated if the home and the chosen zone are within the same cluster (nest). -
- Intra-Zonal effect: A coefficient is estimated to boost the utility of choosing the same zone as the origin. -
- Cluster Nest coefficients: A nest coefficient for each of the 12 clusters is estimated. -
- Cluster ASCs: Likewise, ASCs are estimated for each cluster. +
- 4D measures: Several 4D measures such as transit accessibility, hospital accessibility, walk accessibility, employment densities are considered. +
- Time Coefficient(s): The auto times corresponding to the latest model skims are attached depending upon the time of day the respondent made the trip. A time coefficient is estimated in each specification. +
- Intra-Cluster effects: Intra-Cluster coefficients are estimated if the home and the chosen zone are within the same cluster (nest). +
- Intra-Zonal effect: A coefficient is estimated to boost the utility of choosing the same zone as the origin. +
- Cluster Nest coefficients: A nest coefficient for each of the 12 clusters is estimated. +
- Cluster ASCs: Likewise, ASCs are estimated for each cluster.
- ASCs were estimated for all but one of the clusters. Only the significant ASCs were kept for each model purpose. The ASCs with poor significance generally had values close to zero. Further, these ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intrazonal effects are strong for a few of the purposes -
- Intra-Cluster effects are strong implying that there is a tendency to choose locations that belong in the same cluster as the home zone. -
- Using mode choice logsums for the auto model (and for vehicle sufficient segments) caused unrealistic signs as well as much poorer model fit. This can be explained by the fact that these logsums are primarily driven by auto skims, which are already a part of the utility equation. Hence most specifications do not have an auto logsum. -
- ASCs were estimated for all but one of the clusters. Several of them had values close to zero with poor significance. These were dropped and only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intrazonal effects are strong. -
- Home cluster effects are also strong, with residents of Cary, Durham, Chapel Hill, Garner often living and working in the same cluster. Similar effects are also true for the peripheral clusters toward the northeast and south. -
- Mode choice logsums are significant for workers from zero-auto households, capturing a coupling with access to public transit modes. -
- Using mode choice logsums for the other auto sufficiency markets (with or without the presence of congested time) caused unrealistic signs as well as much poorer model fit. This can be explained by the fact that these logsums are primarily driven by auto skims, which are already a part of the utility equation. -
- Hospitals being a major source of employment in the region, accessibility to hospital jobs helps drive work destination attractiveness. Transit and walk accessibilities are also drivers of a zone’s attractiveness for work tours. -
- The cluster nest coefficients are mostly significantly different from 1, so the nested structure is justified. Research Triangle Park is a special cluster (very low residential land use, very high technology employment, etc.) and has an MNL-type sub-nest. -
- The size variables are stratified by income. The percentage of low and high paying jobs in each zone was used to split the total attraction into income-specific attractions. -
- ASCs were estimated for all but one of the clusters. Several of them had values close to zero with poor significance. These were dropped and only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intrazonal effects are strong. -
- Travel time is more important up to the first 30 minutes. Excess time beyond this threshold is perceived as being less onerous. -
- Home cluster effects are also strong, with residents of Cary, Garner, Raleigh, Southern Durham and the southeastern periphery often stopping in the same cluster for their intermediate stops. -
- The public transit mode choice logsum is significant. -
- The number of households is a major factor in the calculation of the size variable, reflecting the inclusion of personal visits in this trip purpose. Logically, the retail and service employment also help drive trips to specific destinations. Income-based stratifications of employment did not make a difference. -
- The cluster nest coefficients are mostly significantly different from 1, so the nested structure is justified. Research Triangle Park is a special cluster (very low residential land use, very high technology employment, etc.) and has a nearly MNL-type sub-nest. -
- School location choice is driven by travel time. -
- Attractions are driven by school enrollment. -
- Some of the cluster nest coefficients are significantly different from 1, so the nested structure is justified. -
- ASCs were estimated for all but one of the clusters. Several of them had values close to zero with poor significance. These were dropped and only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intrazonal effects are strong. -
- Travel time is more important up to the first 30 minutes. Excess time beyond this threshold is perceived as being less onerous. -
- Auto and transit logsums (by the relevant auto sufficiency category) play a role in the perception of zones’ attractiveness. -
- Home cluster effects are strong, with residents often choosing to shop in the same cluster as their residence. -
- Attractions are driven by office, retail and service employment, which is expected for a trip purpose that involves shopping. -
- ASCs were estimated for all but one of the clusters. Several of them had values close to zero with poor significance. These were dropped and only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intrazonal effects are strong. -
- Travel time is more important up to the first 30 minutes. Excess time beyond this threshold is perceived as being less onerous. -
- Auto logsums (by the relevant auto sufficiency category) play a role in the perception of zones’ attractiveness. -
- Accessibility to hospitals plays a positive role in boosting a zone’s attractiveness, which is apt for this trip purpose. -
- Home cluster effects are also strong, particularly with residents of Garner, Raleigh, Southern Durham and the north-eastern. -
- Attractions are driven by retail and service employment, with an added boost from service employment related to the presence of hospitals. -
- ASCs were estimated for all but one of the clusters. Several of them had values close to zero with poor significance. These were dropped and only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior. -
- Intra-zonal and intra-cluster effects are strong. -
- Attractions are driven by school enrollment. Some of the cluster nest coefficients are significantly different from 1, so the nested structure is justified. Others default to an MNL-like behavior. -
- Longer drive times penalize a destination’s appeal, though times over 30 minutes are not perceived to be as onerous. -
- Hospital accessibility impacts destination attractiveness in a positive manner, likely accounting for longer medical appointments. -
- Walk access is viewed positively, especially in h -
- Most home cluster effects are also strong. -
- Attractions are driven by employment and school enrollment totals as well as the residential population, the latter due to longer social visits. -
- Drive time attenuates the attractiveness of farther destinations, though times beyond 20 minutes are perceived as less onerous than the first 20. -
- Home cluster effects are also strong across all clusters, indicating a strong preference for proximate destinations for relatively short non-mandatory activities. -
- Additionally, intra-zonal destinations are accorded higher preference. -
- Mode choice logsums related to public transit and non-household auto modes (such as ride-hailing) play a positive role in elevating the desirability of destinations. -
- Attractions are driven by employment and school enrollment totals as well as each zone’s residential population, an indication of the propensity to participate in social visits. -
- Work tours
@@ -692,7 +627,7 @@
Non-Home-Based trips
- N_NH_O
- NHB Work Tours by Auto
- NHB Non-work Tours by Auto @@ -701,72 +636,43 @@
- Drive time has the expected negative sensitivity, with times above 30 minutes penalized to a lesser extent than the first 30 minutes. -
- The generic intra-cluster effect is strong. -
- Attractions are driven by a variety of employment totals, as well as school enrollment. -
- Drive time has the expected negative sensitivity, with times above 30 minutes penalized to a lesser extent than the first 30 minutes. -
- Intra-cluster and intra-zonal effects, both generic, are strong. -
- Attractions are driven by school enrollment (perhaps related to after-school activities) and various employment totals. -
- Higher values of the parking logsum lead to more desirable destination owing to enhanced ease of parking. -
- Both in-vehicle time and the number of transfers are significant predictors in the attractiveness of destinations accessible by public transit modes. -
- Transit accessibility positively impacts destination choice. -
- Attractions are driven by office, retail and service employment totals. -
- As expected, walk distance poses a highly significant deterrent to destinations that are farther away from the origin. A high penalty is imposed on destinations that are more than a mile away. -
- Intra-zonal and intra-cluster effects are also strong, favoring short-range trips by non-motorized modes. -
- Walk accessibility is key to choosing destinations. -
- Attractions are driven by office, retail and service employment totals. -
- Cluster ASC values: For a given purpose, the percent of trips attracted to each cluster can be determined (both from the model and the survey). The cluster ASC values can be adjusted to match the aggregated survey percentages. This additional adjustment for each cluster is the natural logarithm of the ratio of the target percentage to the model percentage. -
- Intra-Cluster Constants: The model has provisions for intra cluster dummies. These dummies can likewise be adjusted to match the normalized intra-cluster totals from the survey. -
- Intra-Zonal Dummy: Finally, the specific intra zonal dummy was adjusted for a few purposes to match the intra-zonal percentages. +
- Cluster ASC values: For a given purpose, the percent of trips attracted to each cluster can be determined (both from the model and the survey). The cluster ASC values can be adjusted to match the aggregated survey percentages. This additional adjustment for each cluster is the natural logarithm of the ratio of the target percentage to the model percentage. +
- Intra-Cluster Constants: The model has provisions for intra cluster dummies. These dummies can likewise be adjusted to match the normalized intra-cluster totals from the survey. +
- Intra-Zonal Dummy: Finally, the specific intra zonal dummy was adjusted for a few purposes to match the intra-zonal percentages.
- The mode choice model is run for the given purpose, segment and time period, taking care to use the appropriate time period skim matrices. The output logsum matrix (by mode, segment and period) and the output mode choice probability matrix are stored.
- The DC nested choice is run, which produces a zone-zone probability matrix, such that for each row (zone), the destination probabilities sum to 1. The determination of this probability matrix itself is the result of the nested DC, implemented as a two-stage destination choice model with a zone level model and cluster level model informed by zone-zone DC logsums. diff --git a/docs/resident_mode_choice.Rmd b/docs/resident_mode_choice.Rmd index 318653b7..d519bccd 100644 --- a/docs/resident_mode_choice.Rmd +++ b/docs/resident_mode_choice.Rmd @@ -17,46 +17,612 @@ library(knitr) library(kableExtra) ``` -```{r, include=FALSE} -hh_df <- read_csv("data/output/_PRIVATE/survey_processing/hh_processed.csv") -trips_df <- read_csv("data/output/_PRIVATE/survey_processing/trips_processed.csv") +This document summarizes the mode choice (MC) models estimated using travel +survey and other data from the Triangle region. These are nested logit models +that predict which modes of travel will be used based on things like travel +time, transit fare, and transit headway. Mode choice models play a crucial role +in model estimation, not only directly but also through logsums that are then +used to inform destination choices. For more information on mode choice models, +click [here](https://tfresource.org/topics/Mode_choice.html). + +## Key considerations + +In addition to the usual key drivers of mode choices such as level-of-service +variables, there were key modeling requirements for the Triangle region. + +One of the key modeling requirements for the Triangle region is the focus on +transit usage and ability to analyze the congestion mitigation potential of +future transit projects. To that end, a detailed transit mode choice component +is warranted. In addition to the household travel survey, Caliper had access to +a transit onboard survey. While the proportion of transit trips in the household +survey is in accordance with regional shares, this data alone was not sufficient +to estimate parameters for transit sub-modes. Rather than combine the onboard +survey with the household survey (for various reasons including inconsistency +between trip type), Caliper estimated detailed transit sub-mode models +conditional on the main transit choice using the onboard survey. This is +illustrated in later sections. + +A second consideration was the need to measure future transit modes such as bus +rapid transit, light rail, and commuter rail. These modes do not exist today, so +Caliper leaned on FTA/STOPS guidance. + +A key mode identified during exploratory analysis was the Non-HH Auto mode, that +is comprised of ride sharing and rental cars (classified together as AutoPay) +and borrowed cars (e.g. a friend's car and classified as OtherAuto). + +Parking was a major consideration for the auto modes, especially in designated +parking districts in the CBD/downtown/university areas of cities in the Triangle +region, where parking space is limited. + +For Non-Home Based (NHB) trips, the trip mode is conditional based on the mode +of the home-based leg of the tour. Hence mode choice models were not estimated +for these trips. + +## Approach + +The initial determination was whether to combine the household travel survey +with the on-board survey and subsequently estimate a model using the entire set +of modes (including the transit sub-modes). This would entail potential +re-calculation of the survey weights. + +The alternate approach was to estimate conditional transit sub-mode models using +the onboard survey and then estimate a nested mode choice model using all the +modes but fixing the transit sub-mode utility coefficients as appropriate. +Essentially, the logsum computed from the transit nest would be the key utility +component for the main transit alternative. The transit sub-mode model can also +be viewed as a model that predicts transit sub-mode shares conditional on the +main mode being chosen as transit. + +A closer look at the onboard survey revealed that while the data regarding the +transit sub-mode was rich, the information regarding the tour type was not +available. Whether trips are work, other or shop related could be ascertained +from the onboard survey but there was no information regarding tour types or the +overall travel pattern of the individual respondent. Given that the trip +purposes in TRMG2 are designed based on the tour type and trip type, records +from the onboard survey could not be matched directly to the TRM trip purpose +definitions. Therefore, the alternate approach of estimating transit sub-models +from the onboard survey was chosen. + +Given the above, the mode choice estimation approach for the TRM followed a +two-stage process: + +### Stage 1 + +Estimate Transit sub-mode models (nest) for three main types (Work, Shop and +Other) from the onboard survey. The transit nest is organized by access mode and +transit mode, which was revealed by the estimation process. The three access +modes are Walk, Park & Ride (PNR) and Kiss & Ride (KNR). The modes in the base +year include local bus (LB) and express bus (EB), and these are the only modes +with estimated coefficients. + +### Stage 2 + +Estimate a nested model for each trip purpose in TRMG2. The top-level choices +are the main modes namely Auto, Non-HH Auto and Transit. The Auto nest includes +the sub-modes SOV, HOV2 and HOV3. The Non-HH Auto nest includes the modes +AutoPay and OtherAuto. The appropriate Transit nest is stitched by keeping the +utilities and the coefficients from the transit nest models fixed (essentially +fixing the logsums). For some trip purposes, certain transit access modes were +dropped if there were no observations in the survey. The parameters of the +utility function of the various auto modes, ASC’s and nest coefficients for the +top-level alternatives are estimated during the process. Certain key utility +terms common to a particular nest (e.g., Transit) were added to the appropriate +nest alternative (e.g., to the Transit alternative) to capture the fundamental +choice of the main mode. A weighted estimation was performed using the trip +final weights. + +The conceptual figure of a typical nested model is presented in the figure. Note +that the utility coefficients and structure for the Transit come from the +estimations in Stage 1 described above. For instance, for the w_hb_w purpose, +where the tour mode is work and the trip mode is work, the transit nest for the +work transit purpose is used in the final mode choice specification. + +```{r, fig.align='center', out.width="90%"} +knitr::include_graphics("img/mc/full_tree.png") +``` + +In the following sections, the three Transit nest models are first described, +followed by the completed models for each of the TRM trip purposes. + +# Transit Submodels + +## Work + +This model is estimated from the onboard survey for work trips. All six +combinations of transit and access mode are considered, and the estimated model +has the nested structure below, nested by access mode and then by transit mode. +The estimation process indicated that this nesting structure (rather than main +mode first followed by access mode), was more appropriate for the Triangle. This +reflects the large differences in behavior between households with and without +autos. For most people, they must first determine if driving is an option. If it +isn't, they can only take transit if they can walk to it. + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/transit_sub_work.png") +``` + +### Utility specification + +The utility specification mainly comprises of the level of service transit +variables and auto sufficiency market segmentation terms. The utility equations +are shown in the first table below where the letter ‘X’ indicates that the +variable is included in the utility equation for the appropriate alternative +(column). The alternative-specific constant (ASC) row shows the estimated +constant values with the corresponding tStat within the brackets. The three nest +coefficients are also shown with their tStats. + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/transit_sub_work.png") +``` + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/transit_sub_work2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/transit_sub_work3.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/transit_sub_work4.png") +``` + +As expected, transit LOS variables are significant. A fare coefficient could not +be directly estimated. Instead, a VOT of 11.9 $/hr was used to convert the fare +to equivalent time and a time coefficient was estimated. This ensures that the +model is sensitive to changes in fare policy. + +Nested Logit estimation is known to be non-unique and is influenced by the +starting values of the nest coefficients. This implies that potentially several +estimations must be run with a different set of starting nest coefficients. +After a wide range of experiments on the TRM data, it was necessary to fix the +PNR nest coefficient to 1 (implying that PNR_LB and PNR_EB are treated as +top-level modes). + +As expected, households with zero vehicles tend to prefer walk access to transit +options. Similarly, vehicle insufficient households prefer not to use PNR since +that would presumably make the vehicle unusable while in the parking lot. + +## Other + +The transit sub-model estimated for the other purpose has a similar structure to +the work purpose. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/transit_sub_other1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/transit_sub_other2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/transit_sub_other3.png") ``` -```{r} -samples_df <- trips_df %>% - filter(tour_type != "H") %>% - group_by(trip_type, choice_segment) %>% - summarize(samples = n()) %>% - mutate(choice_segment = factor(choice_segment, levels = c("v0", "ilvi", "ilvs", "ihvi", "ihvs"), ordered = TRUE)) %>% - arrange(choice_segment) %>% - pivot_wider(names_from = choice_segment, values_from = samples, values_fill = 0) +Again, a VOT of 11.9 $/hr was used to obtain the fare coefficient. -# write_csv(samples_df, "mc_samples.csv") +## Shop + +The onboard survey revealed that it is very unlikely that shop trips use PNR and +KNR, which is intuitive. If a household has access to a car with which to make a +PNR/KNR trip, they would prefer to simply drive directly to the store and have +an easy way to transport groceries or other purchases. As a result, the transit +alternatives consist only of the walk access modes. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/transit_sub_shop1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/transit_sub_shop2.png") +``` + +### Notes + +As with the other submodels, a VOT of 11.9 $/hr was used to obtain the Fare +coefficient. + +# Homebased work trips + +## W_HB_W + +These are trips on a work tour with one end being home and the other being work. +During the estimation process, the full nesting structure including all modes in +the transit sub-nest were specified. The utility equations from the work transit +nest model along with their coefficients were retained (and fixed), except the +transit sub-mode ASCs. These were allowed to be determined by the estimation. +This was done to enable the estimation to adjust the constants of the transit +mode to fit the top-level shares from the household travel survey. A weighted +estimation using the trip weights was performed. + +The utility spec for the combined data is shown below. Note that the transit +sub-mode utilities are not shown in the table (see above). + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/w_hb_w1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_w2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_w4.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_w3.png") +``` + +As expected, the time coefficients are highly significant. The TNC time +coefficient is deemed less onerous than other auto modes, which is an intuitive +result given that you don’t have to drive and can (e.g.) read a book. + +Toll coefficients for the auto modes were computed using a VOT of 16.4 $/hr. + +The nest structure for the NonHHAuto is justified, given the tStat of the nest +coefficient of -11.8. The Auto nest is also retained despite the nest +coefficient having a low tStat. + +Parking logsums have a significant effect on the SOV and HOV modes as well as +the OtherAuto mode. As parking conditions deteriorate, trips shift into modes +like Transit, AutoPay, and OtherAuto. + +High Income has a positive effect on the SOV and HOV modes. Similarly, vehicle +insufficient households have a higher tendency to prefer AutoPay and OtherAuto +modes as opposed to SOV and HOV modes. + +During application, the model is applied to five segments and four time periods +(20 combinations). The segments are v0 (Zero Autos); Low-Income and High-Income +vehicle insufficient HHs (ilvi and ihvi); and Low-Income and High-Income vehicle +sufficient HHs (ilvs and ihvs). The coefficients of the high income and vehicle +insufficient utility terms are enforced depending on the segment. + +TNC wait times and fares are used as utility variables for the AutoPay mode and +the computation of these variables is described in the Accessibility page. + +## W_HB_O + +The survey revealed that nearly all the transit trips for this type are walk +access. Therefore, the PNR and KNR branches were removed from the transit nest +in the final specification. The Transit node only has the walk-local and +walk-express alternatives. The appropriate utility equations and coefficients +(except the ASCs) for the walk-local and walk-express modes from the work +transit sub-nest model are fixed based on the transit submodel estimation. The +final nesting structure is shown below. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/w_hb_o1.png") ``` -```{r} -hh_by_segment <- hh_df %>% - group_by(choice_segment) %>% - summarize(weight = sum(hh_weight_combined, na.rm = TRUE)) %>% - mutate(pct = weight / sum(weight)) - +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/w_hb_o2.png") +``` -percents <- trips_df %>% - filter(!(mode_final %in% c("other", "school_bus", "XXX"))) %>% - group_by(choice_segment, mode_final) %>% - summarize(trips = sum(trip_weight_combined)) %>% - mutate(pct = trips / sum(trips)) %>% - select(-trips) %>% - pivot_wider(names_from = mode_final, values_from = pct) +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_o3.png") +``` -percents <- trips_df %>% - filter(!(mode_final %in% c("other", "school_bus", "XXX"))) %>% - group_by(mode_final, choice_segment) %>% - summarize(trips = sum(trip_weight_combined)) %>% - mutate(pct = trips / sum(trips)) %>% - select(-trips) %>% - pivot_wider(names_from = choice_segment, values_from = pct) +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_o4.png") +``` -# write_csv(percents, "test.csv") +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/w_hb_o5.png") ``` +As with W_HB_W, the time coefficients are highly significant in this model. The +VOT for the Auto modes computed from the time and cost estimates is close to the +desired value of time of $16.4/hr. + +Although a nest coefficient for the Auto mode could not be estimated, this +coefficient was set to 0.75 to match the estimated value from the w_hb_w model. +Since this purpose consists of short errands on the work tour, parking logsums +for work were not found to be significant. + +Other similarities to W_HB_W included the positive effect of high income on auto +modes persisted and the preference of vehicle insufficient households to use +OtherAuto and Transit modes + +## W_HB_EK12 + +These are school pick up and drop off trips and the survey indicates that the +two predominant modes were HOV2 and HOV3. Rather than estimate a mode choice +model, a simple probability split of 50.4% for HOV2 and 49.6% for HOV3 was used. + +# Homebased non-work trips + +The purposes N_HB_OME, N_HB_ODShort, N_HB_ODLong, N_HB_OMED and N_HB_K12 fall +under this category. Typically, Caliper observed that transit PNR and KNR modes +were seldom used for these purposes. The tree structure for these purposes is +shown below. + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/nhb_full_nest.png") +``` + +N_HB_K12 is the one exception to the above graphic, as it includes a SchoolBus +mode rather than walk to express bus. + +## N_HB_OME + +These are home based trips on non-work tours that involve shopping, eating out +and other maintenance cash spending activities. Therefore, the transit nest +model for the shop purpose (with the walk-local and walk-express alternatives) +is used. This is the only model where the ‘Shop’ transit sub-nest coefficients +are used. As expected, for trips such as shopping, PNR and KNR modes were not +chosen. If a person has access to a car, they will use it buy groceries rather +than try and carry them on the bus. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/n_hb_ome1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_ome2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_ome3.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_ome4.png") +``` + + +Parking logsums have an important effect on choice of auto modes, which makes +sense. For work trips, travelers have to work near their job and parking is less +of a consideration. For non-mandatory trips like shopping, people are more +sensitive to parking cost and availability. + +High income has a positive effect on HOV and an even larger positive effect on +SOV trips. At the same time, the HH Size 1 coefficient is highly negative for +HOV trips (which is intuitive). Vehicle sufficiency plays a significant role in +choice of SOV and other modes (with varying effects). + +Because estimation is done on dissaggregate survey records, the high-income +housing variable was either 1 or 0. During model application, which is +aggregate, the percent of the zone that is high income is used. This can take +values between 0 and 1. + +## N_HB_OD_Short + +These are home based other trips with a short duration that are part of a +non-work tour. The walk access branch (with the walk-local and walk-express +alternatives) from the ‘Other’ transit nest model is used. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/n_hb_od_short1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_short2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_short3.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_short4.png") +``` + +The coefficients for the Non-HHAuto modes during the estimation had incorrect +signs and were generally not significant. Therefore, a binary choice model with +the alternatives AutoPay and OtherAuto was estimated from the survey using all +the home-based trips on non-work tours. Only trips that chose these two +alternatives were used in the estimation. The binary choice model coefficients +and utility equations were fixed in the final mode choice model. This increased +sample size allowed Caliper to achieve appropriate coefficients. + +Coefficients match similar patterns shown by other purposes and are logical. + + +## N_HB_OD_Long + +These are home based other trips with a long duration that are part of a +non-work tour. The walk access branch (with the walk-local and walk-express +alternatives) from the ‘Other’ transit nest model is used. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/n_hb_od_long1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_long2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_long3.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_od_long4.png") +``` + +All coefficients are intuitive as seen in other models. For example, time +coefficients are significant and parking logsums indidicate correct model +response to parking variables. + +## N_HB_OMED + +These are home based medical related trips (including pharmacies) that are part +of a non-work tour. The walk access branch (with the walk-local and walk-express +alternatives) from the ‘Other’ transit nest model is used. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/n_hb_omed1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_omed2.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_omed3.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_omed4.png") +``` + +During the initial estimation, the time coefficient (along with an asserted toll +coefficient derived from the time coefficient) for the auto modes had a positive +sign. A model without the toll coefficient however yielded a feasible value for +the time coefficient. This value was fixed in the final estimation along with +the asserted toll coefficient using the value of time of $11.9/hr. + +Other coefficients follow the same logical patterns as other purposes. + +## N_HB_K12 + +These are home based school related trips that are part of a non-work tour and +include school trips made by kids and drop off trips made by adults. The survey +indicated that walk-to-local bus was the only transit mode used. AutoPay was +also not used. Therefore, a simple MNL model with the modes SOV, HOV2, HOV3, +OtherAuto, Walk-Local transit, and SchoolBus was estimated. The utility equation +for the walk local transit was borrowed from the ‘Other’ transit sub-mode model. + +### Utility specification + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/n_hb_k12_1.png") +``` + +```{r, fig.align='center', out.width="50%"} +knitr::include_graphics("img/mc/n_hb_k12_2.png") +``` + +### Notes + +The time coefficients are significant as expected, and the school bus time is +actually more onerous compared to driving kids to school. This is an amusing +result that is nevertheless intuitive for anyone who rode the bus as a child. + +Large HHs prefer school pickup/drop-offs compared to riding the bus. + +Choice of mode for this purpose is probably determined by several other factors +specific to individual households and are thus not adequately captured by a +choice model. Nevertheless, a choice model helps fix the shares of the primary +modes for this purpose, namely SchoolBus and HOV + +# Calibration + +Mode choice ASCs were calibrated to match the adjusted targets shown below. The +mode choice targets are generated from the survey for each combination of trip +purpose and HH market segment. Note that target shares are not split by transit +sub-mode due to the lack of transit trips in the survey. The shares vary +significantly by market segment, especially for the zero-auto segment as +expected. Further, during the transit ridership validation, the survey transit +targets were scaled down to match transit boardings. + +The additional ASC adjustments were not large, which increased confidence in the +estimated coefficients. + +```{r, fig.align='center', out.width="85%"} +knitr::include_graphics("img/mc/asc_calibration.png") +``` + +# Application + +As mentioned, the mode choice output shares can only be applied after the +destination choice model has been run. However, the destination choice model +requires mode choice logsums by mode, period, purpose and segment. To facilitate +this, the mode choice models are run first and the probability matrices and +logsums are calculated and stored. After the destination choice model is +complete, the stored probability matrices are used to split the appropriate PA +matrices by mode. The following rules are used during the mode choice +application for a specific purpose, segment and time period. + +- For the zero-vehicle segment, SOV, HOV2 and HOV3 are made unavailable. +- The appropriate skims are used depending on the time period. +- Transit modes are automatically dropped for matrix cells with missing transit +skim values. + +# Rail and BRT + +Rail and Bus Rapid Transit (BRT) do not exist in the Triangle in 2020; however, +the region is actively studying these modes for the future. As a result, Caliper +asserted mode choice nests for Light Rail (LR), Commuter Rail (LR), and BRT. +This assertion combined revealed modal data from the on-board survey (just as +value of time coefficients) and guidelines provided by FTA and used in the STOPS +model +([link1](https://www.transit.dot.gov/funding/grant-programs/capital-investments/stops-workshop-atlantic-city-may-17-2015) +[link2](https://www.fsutmsonline.net/images/uploads/Task_1_Guidebook_for_Florida_STOPS_Application.pdf)). + +The first FTA factor to consider is what they call "Visibility". In short, as +more of the service operates on fixed guideway, the higher this factor goes +(bounded between 0 and 1). + +```{r, fig.align='center', out.width="90%"} +knitr::include_graphics("img/mc/fta_visibility.png") +``` + +For TRMG2, the following visibility factors were assumed: + +- BRT: 0.25 +- LRT: 0.50 +- CRT: 1.00 + +Additionally, the graphic below shows the FTA guidance on path-based constants. +These are presented as penalties added to bus, but Caliper applied them as +bonuses for rail and BRT. + +```{r, fig.align='center', out.width="90%"} +knitr::include_graphics("img/mc/fta_path_constants.png") +``` + +TRMG2 market segmentation does not match the above chart exactly. Caliper made +the following assumptions about equivalency: + +- zero-vehicle segment (v0): 0 vehicle households +- vehicle insufficient segment (vi): 1 vehicle households +- vehicle sufficient segment (vs): 2+ vehicle households + +Additionally, the cost coefficients from estimation (generally around .022) were +within the FTA guidance of .02 - .03. Caliper applied the following discounts +to in-vehicle time (adjusted based on our assumed visibility scores): + +- BRT: 5% +- LRT: 10% +- CRT: 20% + +The table below is a concrete example of how LRT was modified for the +N_HB_OD_Short purpose. In the gray cells, the numbers in blue show the +modifications. The IVTT time is discounted and the mode is given bonus path type +constants based on visibility, market segment, and scaled by G2s cost +coefficient. + +```{r, fig.align='center', out.width="100%"} +knitr::include_graphics("img/mc/lr_adjustment.png") +``` + +Additionally, Caliper compared these adjustments to the existing TRMV6.2 model. +In that model, a single premium mode exists with the following adjustments: + +- IVTT discount of 15% +- 30 bonus minutes added to the constant + +All else equal, TRMG2s smaller discounts and bonuses should lead to more +conservative ridership estimates. However, during planning analysis, the region +can adjust the visibility factors to reflect the actual service being planned. +For example, if planned BRT service operates fully inside reserved right of +way, the visibility factor should be increased. \ No newline at end of file diff --git a/docs/resident_mode_choice.html b/docs/resident_mode_choice.html index 15887371..4d8558fa 100644 --- a/docs/resident_mode_choice.html +++ b/docs/resident_mode_choice.html @@ -188,6 +188,9 @@
- Destination Choice +
- + Mode Choice +
- Non-homebased @@ -203,6 +206,9 @@
- Accessibility +
- + University +
- Airport @@ -233,7 +239,55 @@
- Key considerations +
- Approach +
- Transit Submodels
-
+
- Work +
- Other +
- Shop
-
+
- Utility specification +
- Notes +
+
+ - Homebased work trips +
- Homebased non-work trips
-
+
- N_HB_OME +
- N_HB_OD_Short +
- N_HB_OD_Long +
- N_HB_OMED +
- N_HB_K12
-
+
- Utility specification +
- Notes +
+
+ - Calibration +
- Application +
- Rail and BRT +
- For the zero-vehicle segment, SOV, HOV2 and HOV3 are made unavailable. +
- The appropriate skims are used depending on the time period. +
- Transit modes are automatically dropped for matrix cells with missing transit skim values. +
- BRT: 0.25 +
- LRT: 0.50 +
- CRT: 1.00 +
- zero-vehicle segment (v0): 0 vehicle households +
- vehicle insufficient segment (vi): 1 vehicle households +
- vehicle sufficient segment (vs): 2+ vehicle households +
- BRT: 5% +
- LRT: 10% +
- CRT: 20% +
- IVTT discount of 15% +
- 30 bonus minutes added to the constant +
- Destination Choice +
- + Mode Choice +
- Non-homebased @@ -205,6 +208,9 @@
- Accessibility +
- + University +
- Airport @@ -403,13 +409,13 @@
Home-Based Trips
Utility Specification
The utility specification for the DC models consisted of the following set of variables. Not all these variables are present in every specification since variables are retained depending on the significance of their estimated coefficients or owing to strong apriori assumptions.
+Size term
+This is the natural logarithm of the attractions at a destination zone, which itself is a linear combination of employment variables and their coefficients. These coefficients are simultaneously estimated with other parameters. Since it is required that these coefficients be positive, the utility formulation specifies coefficients that are then exponentiated. Successful estimation requires that one of these attraction variable coefficients is fixed. The estimation also provides a coefficient for the size variable, which theoretically should be between 0 and 1.
+Mode Choice Logsums
+Using mode choice root logsums typically presents a problem in destination choice. The root logsum value is dominated by the auto nest which washes out effects from non-dominant modes. For example, drastically improving the transit accessibility to a particular zone will not affect the probability of choosing that zone if the transit logsums are dwarfed by the auto logsums. To circumvent this issue, separate auto, transit, and non-household auto logsums from the respective nests are used in the TRM specifications. The logsums are segmented by vehicle sufficiency market segments for added explanatory power. Work trip logsums are further distinguished by high- and low-income segments. Finally, note that there is no auto logsum component for the zero-vehicle households market segment.
+Others
-
-
The rich utility specification allows the model to capture the decision-making process of choosing destination zones and capturing cluster-to-cluster flows.
General Estimation Observations
--
-
ASCs were estimated for all but one of the clusters, but only the significant ASCs were kept for each model purpose. The ASCs with poor significance generally had values close to zero. Further, these ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables with the ability to capture location choice behavior.
+Intrazonal effects were strong for a few of the purposes, but intra-cluster effects were very strong implying that there is a tendency to choose locations that belong in the same cluster as the home zone.
+Mode choice logsums for the auto model (and for vehicle sufficient segments) caused unrealistic signs as well as much poorer model fit. This can be explained by the fact that these logsums are primarily driven by auto skims, which are already a part of the utility equation. Hence most specifications do not have an auto logsum.
W_HB_W
@@ -387,23 +388,16 @@W_HB_W
Estimated coefficients and t statistics
-Notes
--
-
ASCs were estimated for all but one of the clusters. Several had values close to zero with poor significance and were dropped. The ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables and is sensitive to changes in model inputs.
+Intrazonal and home cluster effects are strong. Residents of Cary, Durham, Chapel Hill, Garner often live and work in the same cluster. Similar effects are also true for the peripheral clusters toward the northeast and south.
+Root mode choice logsums are significant for workers from zero-auto households, capturing a coupling with access to public transit modes. Using root mode choice logsums for the other auto sufficiency markets caused unrealistic signs as well as much poorer model fit. This can be explained by the fact that these logsums are primarily driven by auto skims, which are already a part of the utility equation (the “Time” term).
+Hospitals are major sources of employment in the region, and accessibility to hospital jobs helps drive work destination attractiveness. Transit and walk accessibilities are also drivers of a zone’s attractiveness for work tours.
+Most of the cluster nest coefficients are significantly different from 1, which means the nested structure is justified. Research Triangle Park is a special cluster (very low residential land use, very high technology employment, etc.). It’s coefficient is effectively 1, which implies an MNL-type sub-nest.
+The size variables are stratified by income using the percentage of low and high paying jobs in each zone. This is how the model pairs high-income workers (e.g. in the Regency area of Cary) with high-paying jobs in RTP. This stratification is critical for accurate work flows in region.
Double constraint and attraction model
-All trip types other than work are singly-constrained. This means that the row sums of the resulting trip table will match starting productions, but column sums will not necessarily be proportional to the amount of employment in each zone. As an example, for two zones with the same employment, the more accessible zone will attract more trips.
+All trip types (other than work) are singly-constrained. This means that the row sums of the resulting trip table will match starting productions, but column sums will not necessarily be proportional to the amount of employment in each zone. As an example, for two zones with the same employment, the more accessible zone will attract more trips.
For work trips, a traditional assumption to make is that each job of the same type must attract the same work trips - even if the zone is in a remote location. The justification for this assumption has weakened in recent years with the rise in telecommuting, flex schedules, and other changes, but the TRMG2 model implements double constraint for work trips. In the context of destination choice models, this is achieved by assigning each zone a “shadow price”. This extra term in the utility equation is adjusted in an iterative fashion to match predicted attractions.
These predicted attractions come from a regression model estimated from the survey data.
During model application, predicted attractions are always scaled to match predicted productions to match double constraint. As a consequence, the attraction model should predict total attractions that is close to predicted productions. This is referred to as PA balance (“production/attraction balance”).
-The coefficients in the table above were multiplied by 2.45 during model calibration to achieve the appropriate PA balance of 1.0. If the ratio is too high or too low, adding additional employment to a zone will have unexpectedly high or low impact on trips attracted (given the scaling that must be done to attractions).
+During model application, predicted attractions are always scaled to match predicted productions before applying double constraint. As a consequence, the attraction model should predict total attractions that is close to predicted productions. This is referred to as PA balance (“production/attraction balance”). If it does not, adding employment to a zone will have unexpectedly high (or low) impacts on total zonal attractions.
+The coefficients in the table above were multiplied by 2.45 during model calibration to achieve the appropriate PA balance of 1.0.
W_HB_O
-W_HB_O corresponds to trips that are part of a work tour, have one end at home, and the other end at a non-work location. This segment captures trips that directly connect the home to an intermediate stop on the way to or from work and translates to the HBO trip purpose in traditional models. Destination choice in this context relates to a short-term decision of intermediate stop location.
+W_HB_O trips are part of a work tour, have one end at home, and the other end at a non-work location. This segment captures trips that directly connect the home to an intermediate stop on the way to or from work and is part of the HBO trip purpose in traditional models. Destination choice in this context relates to a short-term decision of intermediate stop location.
Estimated coefficients and t statistics
-Notes
--
-
ASCs were estimated for all but one of the clusters, but only the significant ones were retained. These ASCs are very small in magnitude, confirming that the model’s explanatory power is derived from other variables and is sensitive to changes in model inputs.
+The model has a sophisticated approach to travel time. Travel time is more important up to the first 30 minutes. Excess time beyond this threshold is gets a slight discount. This reflects a difference in the behavior for these longer trips, which are dominated more by what is at the destination rather than the time to get there.
+Home cluster effects are strong for residents of Cary, Garner, Raleigh, Southern Durham and the southeastern periphery. People in these clusters often stop in the same cluster for their intermediate activities.
+The public transit mode choice logsum is significant for certain segments like zero-vehicle households. Non-household auto logsums (primarily taxis and TNCs) are also significant.
+The number of households is a major factor in the calculation of the size variable, reflecting the inclusion of personal visits in this trip purpose. Logically, the retail and service employment also help drive trips to specific destinations. Income-based stratification of employment did not make a difference in model performance.
+Most cluster nest coefficients are significantly different from 1, so the nested structure is justified. As with W_HB_W, RTP is an exception to this.
W_HB_EK12
-W_HB_EK12 corresponds to trips that are part of a work tour, have one end at home, and the other end at a school. This segment captures trips that directly connect the home to school on the way to or from work and translates to the HB School trip purpose in traditional models. It includes worker dropping off/picking up children on the way to/from work, and workers who attend school.
+W_HB_EK12 corresponds to trips that are part of a work tour, have one end at home, and the other end at a school. This segment captures trips that directly connect the home to school on the way to or from work and translates to the HB School trip purpose in traditional models. It includes workers dropping off/picking up children on the way to/from work, and young workers who attend school.
Estimated coefficients and t statistics
-Notes
--
-
School location choice is driven by travel time and school enrollment. Some of the cluster nest coefficients are significantly different from 1, so the nested structure is justified for those clusters. The remaining zones are treated as top-level choices (MNL).
N_HB_OME
Estimated coefficients and t statistics
-Notes
--
-
The cluster nest coefficients are mostly significantly different from 1, so the nested structure is justified.
+As with the previous models, cluster-based ASCs are small and only applied to a few clusters, which implies a model that is appropriately sensitive. Travel time is again more important up to the first 30 minutes. Excess time beyond this threshold is perceived as being less onerous.
+Attractions are driven by retail and service employment, which is appropriate for shopping/maintenance trips.
N_HB_OMED
Estimated coefficients and t statistics
-Notes
--
-
Most of the cluster nest coefficients are significantly different from 1, so the nested structure is justified.
+A zone’s accessibility to a hospital is a strong predictor in this model. This captures the important clustering of doctor’s offices, clinics, and pharmacies around major hospitals.
+Travel time above 30 minutes has almost no impact on desintation choice. This is an encouraging result. People traveling long distances to see a specialist are not impacted by travel time in a major way.
+Attractions are driven by retail and service employment, with an added boost from service employment related to the presence of hospitals.
N_HB_K12
Estimated coefficients and t statistics
-Notes
--
-
This model has a strong fit and is primarily based on K12 enrollment and travel time (as expected). Additionally, the negative coefficient on the Durham and Northeast clusters means that (all else equal) zones in that cluster are less likely to be chosen. The positive intrazonal coefficient is expected given the tendency for people to be assigned to schools near their home.
N_HB_OD_Long
-N_HB_ODLong corresponds to trips that are not part of a work tour, have one end at home, and the other end at an “other” activity that is at least 30 minutes long. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+N_HB_ODLong corresponds to trips that are not part of a work tour, have one end at home, and the other end at an “other” activity that is at least 30 minutes long. Visiting a friend is one example of this trip type. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
Estimated coefficients and t statistics
-Notes
--
-
The cluster nest coefficients are all significantly different from 1, so the nested structure is justified.
+The penalty on travel time above 30 minutes is reduced as seen in multiple trip trypes. The positive coefficient on walk accessibility means that these trips are more attracted to zones in area that are more dense. Home cluster effects are strong, meaning that most of these trips stay within the home cluster. Employment, school enrollment, and residential population all are significant components of the size term. This reflects the catch-all nature of this trip type.
N_HB_OD_Short
-N_HB_ODShort corresponds to trips that are not part of a work tour, have one end at home, and the other end at an “other” activity shorter than 30 minutes in duration. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+N_HB_OD_Short corresponds to trips that are not part of a work tour, have one end at home, and the other end at an “other” activity shorter than 30 minutes in duration. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
Estimated coefficients and t statistics
-Notes
--
-
All cluster nest coefficients are significantly different from 1, so the nested structure is justified.
+The attenuation on for drive time above 20 minutes is not as strong as other trip types. This indicates that these trips are likely just trying to reach the nearest attraction that satisfies the trip purpose. Similarly, home cluster effects and the intrazonal term are strong meaning that people stay nearby.
+Logsums for household auto, non-household auto, and transit all play an important role in determining where someone travels. Similar to the N_HB_OD_Long trip, the size term is made up of employment, enrollment, and household population.
Non-Home-Based trips
-The non-home-based trip purposes are multinomial logit (MNL) specifications. Though the cluster-based nested structure of the home-based purposes was not adopted, cluster-level alternative-specific constants (ASCs) are included to help capture cluster-level preferences.
+The non-home-based trip purposes are multinomial logit (MNL) specifications. The cluster-based nesting structure did not improve model results and was not adopted for these purposes. Even so, the cluster-level alternative-specific constants (ASCs) are included to help capture cluster-level preferences.
The non-home-based trip purposes consist of:
Given the strong dependence of destination choice on the prior home-based trip’s travel mode for non-home-based trips, separate destination choice models have been estimated by mode. For simplicity and statistical efficiency of model estimation, the above trip purposes were combined into the following four categories that combine trip purpose with travel mode:
+Recall that non-home-based trips are generated by mode based on the results of the home-based models. Given this, separate destination choice models have been estimated by mode. For simplicity and statistical efficiency of model estimation, the above trip purposes were combined into the following four categories that combine trip purpose with travel mode:
Non-Home-Based trips
NHB Work Auto
-This segment corresponds to trips that are part of a work tour, do not have either end at home, have at least one end at school, work, work-related or “other” activities, and are associated with a drive mode. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+This segment corresponds to trips that are part of a work tour, do not have either end at home, and are associated with the auto mode. This segment would have been rolled into the NHBW (Non-Home-Based Work) trip purpose in traditional models.
Estimated coefficients and t statistics
-Notes
--
-
Drive time has the expected negative sensitivity, with times above 30 minutes penalized to a lesser extent than the first 30 minutes. The generic intra-cluster effect is strong, which is expected given shorter trip lengths of NHB trips. Attractions are driven by a variety of employment as well as school enrollment.
NHB NonWork Auto
-This segment corresponds to trips that are part of a non-work tour, do not have either end at home, and have at least one end at school or “other” activities. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+This segment corresponds to trips that are part of a non-work tour, do not have either end at home, and use the auto mode. This segment would have been rolled into the NHBO (Non-Home-Based Other) trip purpose in traditional models.
Estimated coefficients and t statistics
-Notes
--
-
The non-work auto trip model looks similar to work, but the inclusion of parking logsums captures a major difference between the two trip types: work trips are not impacted by parking variables while non-work trips are.
NHB Transit
-This segment corresponds to trips that are part of a non-work tour, and do not have either end at home. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+All NHB transit trips are handled with a single destination choice model.
Estimated coefficients and t statistics
-Notes
--
-
Both the total time of the trip and the number of transfers are significant predictors in the attractiveness of destinations accessible by public transit modes. Transit accessibility of the destination also positively impacts destination choice, which matches expectations.
NHB NonMotorized
-This segment corresponds to trips that are part of a non-work tour, do not have either end at home, and are associated with a non-motorized (e.g. bike, walk) mode. This segment would have been rolled into the HBO (Home-Based Other) trip purpose in traditional models.
+All NHB non-motorized trips are handled with a single destination choice model.
Estimated coefficients and t statistics
-Notes
--
-
As expected, walk distance poses a highly significant deterrent to destinations that are farther away from the origin. An extra penalty is imposed on destinations that are more than a mile away. This captures the sharp drop off in trip lengths beyond one mile in the survey.
+Intra-zonal and intra-cluster effects are strong, which also reflects the short-range nature of non-motorized trips. Finally, the walk accessibility of the destination is important for zero vehicle households. This captures an important reality: these households cannot use autos to make home-based trips.
DC Model Adjustments
During the model calibration stage, further updates to the DC models were performed to match the model cluster to cluster patterns to the weighted patterns from the survey. These adjustments can be thought of as the adjustment to the ASCs of a mode choice model to match aggregated shares.
Given the nested destination choice approach and the assumption behind the structure, the model lends itself very well to such post process adjustments. The following parameters were adjusted to try and match the survey patterns. It is worth mentioning that these post process adjustments turned out to be rather small, thereby inspiring additional confidence in the estimation results:
-
-
While this model form has additional levers to control behavior, the adjustments made were small and do not meaningfully reduce model sensitivity.
DC Application
-This section has a short note on the application of the destination choice models in the TRM model framework. In the TRM model, the destination choice is the higher-level choice which is then followed by the mode choice for a given purpose, market segment and period. However, the utility equation for the DC models contains mode choice logsums. Therefore, a three-step process is followed for each combination of purpose, segment and time period:
+This section is a short note on the application of the destination choice models in the TRM model framework. In the TRM model, the destination choice is the higher-level choice which is then followed by the mode choice for a given purpose, market segment and period. However, the utility equation for the DC models contains mode choice logsums. Therefore, a three-step process is followed for each combination of purpose, segment and time period:
Resident Mode Choice
Resident Mode Choice
-
+
Caliper Corporation
@@ -246,7 +300,240 @@Resident Mode Choice
This document summarizes the mode choice (MC) models estimated using travel survey and other data from the Triangle region. These are nested logit models that predict which modes of travel will be used based on things like travel time, transit fare, and transit headway. Mode choice models play a crucial role in model estimation, not only directly but also through logsums that are then used to inform destination choices. For more information on mode choice models, click here.
+Key considerations
+In addition to the usual key drivers of mode choices such as level-of-service variables, there were key modeling requirements for the Triangle region.
+One of the key modeling requirements for the Triangle region is the focus on transit usage and ability to analyze the congestion mitigation potential of future transit projects. To that end, a detailed transit mode choice component is warranted. In addition to the household travel survey, Caliper had access to a transit onboard survey. While the proportion of transit trips in the household survey is in accordance with regional shares, this data alone was not sufficient to estimate parameters for transit sub-modes. Rather than combine the onboard survey with the household survey (for various reasons including inconsistency between trip type), Caliper estimated detailed transit sub-mode models conditional on the main transit choice using the onboard survey. This is illustrated in later sections.
+A second consideration was the need to measure future transit modes such as bus rapid transit, light rail, and commuter rail. These modes do not exist today, so Caliper leaned on FTA/STOPS guidance.
+A key mode identified during exploratory analysis was the Non-HH Auto mode, that is comprised of ride sharing and rental cars (classified together as AutoPay) and borrowed cars (e.g. a friend’s car and classified as OtherAuto).
+Parking was a major consideration for the auto modes, especially in designated parking districts in the CBD/downtown/university areas of cities in the Triangle region, where parking space is limited.
+For Non-Home Based (NHB) trips, the trip mode is conditional based on the mode of the home-based leg of the tour. Hence mode choice models were not estimated for these trips.
+Approach
+The initial determination was whether to combine the household travel survey with the on-board survey and subsequently estimate a model using the entire set of modes (including the transit sub-modes). This would entail potential re-calculation of the survey weights.
+The alternate approach was to estimate conditional transit sub-mode models using the onboard survey and then estimate a nested mode choice model using all the modes but fixing the transit sub-mode utility coefficients as appropriate. Essentially, the logsum computed from the transit nest would be the key utility component for the main transit alternative. The transit sub-mode model can also be viewed as a model that predicts transit sub-mode shares conditional on the main mode being chosen as transit.
+A closer look at the onboard survey revealed that while the data regarding the transit sub-mode was rich, the information regarding the tour type was not available. Whether trips are work, other or shop related could be ascertained from the onboard survey but there was no information regarding tour types or the overall travel pattern of the individual respondent. Given that the trip purposes in TRMG2 are designed based on the tour type and trip type, records from the onboard survey could not be matched directly to the TRM trip purpose definitions. Therefore, the alternate approach of estimating transit sub-models from the onboard survey was chosen.
+Given the above, the mode choice estimation approach for the TRM followed a two-stage process:
+Stage 1
+Estimate Transit sub-mode models (nest) for three main types (Work, Shop and Other) from the onboard survey. The transit nest is organized by access mode and transit mode, which was revealed by the estimation process. The three access modes are Walk, Park & Ride (PNR) and Kiss & Ride (KNR). The modes in the base year include local bus (LB) and express bus (EB), and these are the only modes with estimated coefficients.
+Stage 2
+Estimate a nested model for each trip purpose in TRMG2. The top-level choices are the main modes namely Auto, Non-HH Auto and Transit. The Auto nest includes the sub-modes SOV, HOV2 and HOV3. The Non-HH Auto nest includes the modes AutoPay and OtherAuto. The appropriate Transit nest is stitched by keeping the utilities and the coefficients from the transit nest models fixed (essentially fixing the logsums). For some trip purposes, certain transit access modes were dropped if there were no observations in the survey. The parameters of the utility function of the various auto modes, ASC’s and nest coefficients for the top-level alternatives are estimated during the process. Certain key utility terms common to a particular nest (e.g., Transit) were added to the appropriate nest alternative (e.g., to the Transit alternative) to capture the fundamental choice of the main mode. A weighted estimation was performed using the trip final weights.
+The conceptual figure of a typical nested model is presented in the figure. Note that the utility coefficients and structure for the Transit come from the estimations in Stage 1 described above. For instance, for the w_hb_w purpose, where the tour mode is work and the trip mode is work, the transit nest for the work transit purpose is used in the final mode choice specification.
+ +In the following sections, the three Transit nest models are first described, followed by the completed models for each of the TRM trip purposes.
+Transit Submodels
+Work
+This model is estimated from the onboard survey for work trips. All six combinations of transit and access mode are considered, and the estimated model has the nested structure below, nested by access mode and then by transit mode. The estimation process indicated that this nesting structure (rather than main mode first followed by access mode), was more appropriate for the Triangle. This reflects the large differences in behavior between households with and without autos. For most people, they must first determine if driving is an option. If it isn’t, they can only take transit if they can walk to it.
+ +Utility specification
+The utility specification mainly comprises of the level of service transit variables and auto sufficiency market segmentation terms. The utility equations are shown in the first table below where the letter ‘X’ indicates that the variable is included in the utility equation for the appropriate alternative (column). The alternative-specific constant (ASC) row shows the estimated constant values with the corresponding tStat within the brackets. The three nest coefficients are also shown with their tStats.
+ + + + +As expected, transit LOS variables are significant. A fare coefficient could not be directly estimated. Instead, a VOT of 11.9 $/hr was used to convert the fare to equivalent time and a time coefficient was estimated. This ensures that the model is sensitive to changes in fare policy.
+Nested Logit estimation is known to be non-unique and is influenced by the starting values of the nest coefficients. This implies that potentially several estimations must be run with a different set of starting nest coefficients. After a wide range of experiments on the TRM data, it was necessary to fix the PNR nest coefficient to 1 (implying that PNR_LB and PNR_EB are treated as top-level modes).
+As expected, households with zero vehicles tend to prefer walk access to transit options. Similarly, vehicle insufficient households prefer not to use PNR since that would presumably make the vehicle unusable while in the parking lot.
+Other
+The transit sub-model estimated for the other purpose has a similar structure to the work purpose.
+Utility specification
+ + + +Again, a VOT of 11.9 $/hr was used to obtain the fare coefficient.
+Shop
+The onboard survey revealed that it is very unlikely that shop trips use PNR and KNR, which is intuitive. If a household has access to a car with which to make a PNR/KNR trip, they would prefer to simply drive directly to the store and have an easy way to transport groceries or other purchases. As a result, the transit alternatives consist only of the walk access modes.
+Utility specification
+ + +Notes
+As with the other submodels, a VOT of 11.9 $/hr was used to obtain the Fare coefficient.
+Homebased work trips
+W_HB_W
+These are trips on a work tour with one end being home and the other being work. During the estimation process, the full nesting structure including all modes in the transit sub-nest were specified. The utility equations from the work transit nest model along with their coefficients were retained (and fixed), except the transit sub-mode ASCs. These were allowed to be determined by the estimation. This was done to enable the estimation to adjust the constants of the transit mode to fit the top-level shares from the household travel survey. A weighted estimation using the trip weights was performed.
+The utility spec for the combined data is shown below. Note that the transit sub-mode utilities are not shown in the table (see above).
+Utility specification
+ + + + +As expected, the time coefficients are highly significant. The TNC time coefficient is deemed less onerous than other auto modes, which is an intuitive result given that you don’t have to drive and can (e.g.) read a book.
+Toll coefficients for the auto modes were computed using a VOT of 16.4 $/hr.
+The nest structure for the NonHHAuto is justified, given the tStat of the nest coefficient of -11.8. The Auto nest is also retained despite the nest coefficient having a low tStat.
+Parking logsums have a significant effect on the SOV and HOV modes as well as the OtherAuto mode. As parking conditions deteriorate, trips shift into modes like Transit, AutoPay, and OtherAuto.
+High Income has a positive effect on the SOV and HOV modes. Similarly, vehicle insufficient households have a higher tendency to prefer AutoPay and OtherAuto modes as opposed to SOV and HOV modes.
+During application, the model is applied to five segments and four time periods (20 combinations). The segments are v0 (Zero Autos); Low-Income and High-Income vehicle insufficient HHs (ilvi and ihvi); and Low-Income and High-Income vehicle sufficient HHs (ilvs and ihvs). The coefficients of the high income and vehicle insufficient utility terms are enforced depending on the segment.
+TNC wait times and fares are used as utility variables for the AutoPay mode and the computation of these variables is described in the Accessibility page.
+W_HB_O
+The survey revealed that nearly all the transit trips for this type are walk access. Therefore, the PNR and KNR branches were removed from the transit nest in the final specification. The Transit node only has the walk-local and walk-express alternatives. The appropriate utility equations and coefficients (except the ASCs) for the walk-local and walk-express modes from the work transit sub-nest model are fixed based on the transit submodel estimation. The final nesting structure is shown below.
+Utility specification
+ + + + + +As with W_HB_W, the time coefficients are highly significant in this model. The VOT for the Auto modes computed from the time and cost estimates is close to the desired value of time of $16.4/hr.
+Although a nest coefficient for the Auto mode could not be estimated, this coefficient was set to 0.75 to match the estimated value from the w_hb_w model. Since this purpose consists of short errands on the work tour, parking logsums for work were not found to be significant.
+Other similarities to W_HB_W included the positive effect of high income on auto modes persisted and the preference of vehicle insufficient households to use OtherAuto and Transit modes
+W_HB_EK12
+These are school pick up and drop off trips and the survey indicates that the two predominant modes were HOV2 and HOV3. Rather than estimate a mode choice model, a simple probability split of 50.4% for HOV2 and 49.6% for HOV3 was used.
+Homebased non-work trips
+The purposes N_HB_OME, N_HB_ODShort, N_HB_ODLong, N_HB_OMED and N_HB_K12 fall under this category. Typically, Caliper observed that transit PNR and KNR modes were seldom used for these purposes. The tree structure for these purposes is shown below.
+ +N_HB_K12 is the one exception to the above graphic, as it includes a SchoolBus mode rather than walk to express bus.
+N_HB_OME
+These are home based trips on non-work tours that involve shopping, eating out and other maintenance cash spending activities. Therefore, the transit nest model for the shop purpose (with the walk-local and walk-express alternatives) is used. This is the only model where the ‘Shop’ transit sub-nest coefficients are used. As expected, for trips such as shopping, PNR and KNR modes were not chosen. If a person has access to a car, they will use it buy groceries rather than try and carry them on the bus.
+Utility specification
+ + + + +Parking logsums have an important effect on choice of auto modes, which makes sense. For work trips, travelers have to work near their job and parking is less of a consideration. For non-mandatory trips like shopping, people are more sensitive to parking cost and availability.
+High income has a positive effect on HOV and an even larger positive effect on SOV trips. At the same time, the HH Size 1 coefficient is highly negative for HOV trips (which is intuitive). Vehicle sufficiency plays a significant role in choice of SOV and other modes (with varying effects).
+Because estimation is done on dissaggregate survey records, the high-income housing variable was either 1 or 0. During model application, which is aggregate, the percent of the zone that is high income is used. This can take values between 0 and 1.
+N_HB_OD_Short
+These are home based other trips with a short duration that are part of a non-work tour. The walk access branch (with the walk-local and walk-express alternatives) from the ‘Other’ transit nest model is used.
+Utility specification
+ + + + +The coefficients for the Non-HHAuto modes during the estimation had incorrect signs and were generally not significant. Therefore, a binary choice model with the alternatives AutoPay and OtherAuto was estimated from the survey using all the home-based trips on non-work tours. Only trips that chose these two alternatives were used in the estimation. The binary choice model coefficients and utility equations were fixed in the final mode choice model. This increased sample size allowed Caliper to achieve appropriate coefficients.
+Coefficients match similar patterns shown by other purposes and are logical.
+N_HB_OD_Long
+These are home based other trips with a long duration that are part of a non-work tour. The walk access branch (with the walk-local and walk-express alternatives) from the ‘Other’ transit nest model is used.
+Utility specification
+ + + + +All coefficients are intuitive as seen in other models. For example, time coefficients are significant and parking logsums indidicate correct model response to parking variables.
+N_HB_OMED
+These are home based medical related trips (including pharmacies) that are part of a non-work tour. The walk access branch (with the walk-local and walk-express alternatives) from the ‘Other’ transit nest model is used.
+Utility specification
+ + + + +During the initial estimation, the time coefficient (along with an asserted toll coefficient derived from the time coefficient) for the auto modes had a positive sign. A model without the toll coefficient however yielded a feasible value for the time coefficient. This value was fixed in the final estimation along with the asserted toll coefficient using the value of time of $11.9/hr.
+Other coefficients follow the same logical patterns as other purposes.
+N_HB_K12
+These are home based school related trips that are part of a non-work tour and include school trips made by kids and drop off trips made by adults. The survey indicated that walk-to-local bus was the only transit mode used. AutoPay was also not used. Therefore, a simple MNL model with the modes SOV, HOV2, HOV3, OtherAuto, Walk-Local transit, and SchoolBus was estimated. The utility equation for the walk local transit was borrowed from the ‘Other’ transit sub-mode model.
+Utility specification
+ + +Notes
+The time coefficients are significant as expected, and the school bus time is actually more onerous compared to driving kids to school. This is an amusing result that is nevertheless intuitive for anyone who rode the bus as a child.
+Large HHs prefer school pickup/drop-offs compared to riding the bus.
+Choice of mode for this purpose is probably determined by several other factors specific to individual households and are thus not adequately captured by a choice model. Nevertheless, a choice model helps fix the shares of the primary modes for this purpose, namely SchoolBus and HOV
+Calibration
+Mode choice ASCs were calibrated to match the adjusted targets shown below. The mode choice targets are generated from the survey for each combination of trip purpose and HH market segment. Note that target shares are not split by transit sub-mode due to the lack of transit trips in the survey. The shares vary significantly by market segment, especially for the zero-auto segment as expected. Further, during the transit ridership validation, the survey transit targets were scaled down to match transit boardings.
+The additional ASC adjustments were not large, which increased confidence in the estimated coefficients.
+ +Application
+As mentioned, the mode choice output shares can only be applied after the destination choice model has been run. However, the destination choice model requires mode choice logsums by mode, period, purpose and segment. To facilitate this, the mode choice models are run first and the probability matrices and logsums are calculated and stored. After the destination choice model is complete, the stored probability matrices are used to split the appropriate PA matrices by mode. The following rules are used during the mode choice application for a specific purpose, segment and time period.
+-
+
Rail and BRT
+Rail and Bus Rapid Transit (BRT) do not exist in the Triangle in 2020; however, the region is actively studying these modes for the future. As a result, Caliper asserted mode choice nests for Light Rail (LR), Commuter Rail (LR), and BRT. This assertion combined revealed modal data from the on-board survey (just as value of time coefficients) and guidelines provided by FTA and used in the STOPS model (link1 link2).
+The first FTA factor to consider is what they call “Visibility”. In short, as more of the service operates on fixed guideway, the higher this factor goes (bounded between 0 and 1).
+ +For TRMG2, the following visibility factors were assumed:
+-
+
Additionally, the graphic below shows the FTA guidance on path-based constants. These are presented as penalties added to bus, but Caliper applied them as bonuses for rail and BRT.
+ +TRMG2 market segmentation does not match the above chart exactly. Caliper made the following assumptions about equivalency:
+-
+
Additionally, the cost coefficients from estimation (generally around .022) were within the FTA guidance of .02 - .03. Caliper applied the following discounts to in-vehicle time (adjusted based on our assumed visibility scores):
+-
+
The table below is a concrete example of how LRT was modified for the N_HB_OD_Short purpose. In the gray cells, the numbers in blue show the modifications. The IVTT time is discounted and the mode is given bonus path type constants based on visibility, market segment, and scaled by G2s cost coefficient.
+ +Additionally, Caliper compared these adjustments to the existing TRMV6.2 model. In that model, a single premium mode exists with the following adjustments:
+-
+
All else equal, TRMG2s smaller discounts and bonuses should lead to more conservative ridership estimates. However, during planning analysis, the region can adjust the visibility factors to reflect the actual service being planned. For example, if planned BRT service operates fully inside reserved right of way, the visibility factor should be increased.
+