Non-Motorized Choice
Intro
TRMG2 uses a disaggregate, binary choice model to split trips into motorized and non-motorized categories. It is common to see the non-motorized mode as a nest within the larger mode choice model, but it was separated for two primary reasons:
- To use a disaggregate model to predict walk trips.
- To simplify the mode choice model and improve prediction power.
In TRMG2, the non-motorized model takes place after trip production but before trips are aggregated for destination and motorized mode choice models. The advantage with a disaggregate model is that person and household level variables were available to improve non-motorized predictions. A potential downside of this approach is that the prediction does not know the mileage or travel time of the trip. This downside is mitigated by using accessibility measures, which in this case are logsums of the gravity model used to distribute walk trips. In this way, aggregate information about trip lengths is included. Further, a person living downtown will make more walk trips than one living in rural areas (all else equal).
The second benefit of a separate non-motorized model is a simplification of the larger mode choice nesting structure. For more detail on that model, and for information on logit choice models in general, see the resident mode choice page.
Estimation Results
All the models that follow are binary choice models with motorized as the reference alternative. Negative coefficients mean that non-motorized travel is less likely. Positive coefficients mean that non-motorized travel is more likely.
Most coefficients are self-explanatory, but two need differentiation:
- Walk Accessibility: this is a measure of how accessible other zones are to the home zone (see the accessibility page).
- Walkability: this is a measure of how walkable the home zone itself is. It encapsulates information on mixed use, the density of the street grid, and other factors. See the accessibility documentation for more detail.
Work tours
The W_HB_W and W_HB_O trip types were estimated in a single model. Behaviorally, the decision to walk to work is made at the tour level and affects all trips made along that tour. A model was not estimated for W_HB_WR trips. Instead, these trips for deliveries, lawn care, and other activities are all assumed to be motorized by the model. Similarly, all K12 escort trips on work tours (W_HB_EK12) are motorized as well.
Children in the household or being a senior make it less likely that trips on a work tour will be non-motorized. Higher car ownership has the same affect. Walk accessibility has a positive relationship, meaning that people who live in urban centers with high walk accessibility are more likely to make walk trips. All coefficients make behavioral sense and the model fit is strong.
| Term | Coefficient | Tstat |
|---|---|---|
| Is Senior | -0.595 | -1.84 |
| Children in HH | -0.641 | -4.59 |
| Vehicles per Adult | -1.440 | -8.71 |
| Walk Accessibility | 1.000 | 11.30 |
| Constant | -2.120 | -11.73 |
Rho-squared: 0.82
Non-work tours
N_HB_K12
The decision to walk to school is driven by vehicle ownership, walk accessibility, and whether or not the child is old enough to drive herself to school. A high rho-squared metric shows a strong model fit.
| Term | Coefficient | Tstat |
|---|---|---|
| Vehicles per Adult | -0.767 | -2.53 |
| Walk Accessibility | 0.585 | 2.90 |
| Age 16 to 18 | -0.359 | -1.10 |
| Constant | -2.710 | -8.08 |
Rho-squared: 0.80
N_HB_OME
The choice for shopping, dining, and maintenance trips is driven by similar variables as previous trips, but also new ones including income and walkability. Income has a positive impact on walking, which may at first appear counter- intuitive. However, many of the factors like vehicle ownership, work status, and accessibility that are often related to income are already controlled for.
| Term | Coefficient | Tstat |
|---|---|---|
| Vehicles per Adult | -1.940000 | -13.60 |
| Walk Accessibility | 0.547000 | 5.12 |
| Is Worker | 0.311000 | 3.03 |
| Income per Person | 0.000008 | 4.48 |
| Walkability | 7.300000 | 3.28 |
| Constant | -2.580000 | -9.32 |
Rho-squared: 0.72
N_HB_OD_Long
These discretionary trips with long activity duration are influenced by many factors. Children are less likely to walk or bike, higher accessibility leads to more walk trips, and the other relationships make sense.
| Term | Coefficient | Tstat |
|---|---|---|
| Vehicles per Adult | -1.1300000 | -8.81 |
| Walk Accessibility | 0.4170000 | 4.54 |
| Is Child | -0.4650000 | -3.90 |
| Is Worker | -0.2830000 | -3.02 |
| Income per Person | 0.0000044 | 2.76 |
| Walkability | 3.7000000 | 2.19 |
| Constant | -2.5800000 | -9.32 |
Rho-squared: 0.59
N_HB_OD_Short
This is the only model that struggles to predict non-motorized behavior as shown by the low rho-squared value. These are “quick stop” discretionary trips, and the poor fit means that choosing to walk for these trips is based on factors not collected in the survey. While the t-stat for walkability was not large, the coefficient makes sense behaviorally and provides sensitivity to land use.
| Term | Coefficient | Tstat |
|---|---|---|
| Is Senior | -0.36100 | -4.30 |
| Is Worker | -0.18800 | -3.13 |
| Walkability | 0.51800 | 0.36 |
| Children in HH | -0.57900 | -8.48 |
| No Autos | 0.89000 | 4.16 |
| Income per Person | 0.00001 | 8.41 |
| Constant | -0.22000 | -1.17 |
Rho-squared: 0.04
N_HB_OMED
The model for medical trips is simple, but does provide important sensitivity for zero-vehicle households and walk accessibility. It also effectively prohibits seniors from walking. The rho-squared is large, but this is inflated given that over 99% of medical trips are motorized. (Randomly assigning 1 out of every 100 medical trips to walk would also have a high rho-squared value.)
| Term | Coefficient | Tstat |
|---|---|---|
| No Autos | 1.73 | 0.90 |
| Is Senior | -157479.00 | -157479.00 |
| Walk Accessibility | 4.64 | 2.19 |
| Constant | -11.20 | -2.42 |
Rho-squared: 0.98
Calibration
Caliper calibrated the model constants to ensure that the final share of non-motorized trips matched the survey. This was done using an auto-calibration utility included in the model’s drop down menu. Caliper then reviewed the results to check that the calibration constants were reasonable and did not dominate model sensitivity.
The parameter file for N_HB_K12_All trips is shown below as an example. Note that the estimated constant is preserved separately from the calibration constant. The additional constant from calibration is small, which means that calibration did not sacrifice model sensitivity to match base year targets.
| Alternative | Expression | Segment | Coefficient | Description |
|---|---|---|---|---|
| motorized | Constant | NA | 0.0000 | NA |
| nonmotorized | person.veh_per_adult | NA | -0.7670 | HH vehicles per adult |
| nonmotorized | person.age_16_18 | NA | -0.3590 | Person is between 16 and 18 (driving age) |
| nonmotorized | se.access_walk.O | NA | 0.5850 | Interzonal walk accessibility |
| nonmotorized | Constant | NA | -2.7090 | Estimated constant |
| nonmotorized | Constant | NA | -0.4293 | Calibration constant |
The non-motorized shares resulting from the model are shown below. Each matches the survey shares.
| Trip Type | NM Share (%) |
|---|---|
| N_HB_K12_All | 2.67 |
| N_HB_OD_Long | 6.71 |
| N_HB_OD_Short | 25.59 |
| N_HB_OMED_All | 0.50 |
| N_HB_OME_All | 4.73 |
| W_HB_EK12_All | 0.00 |
| W_HB_O_All | 2.38 |
| W_HB_W_All | 2.82 |
TransCAD GIS Software, 2022