Step 5 - Mode Choice

This script applies mode choice models to zone data, including accessibility and demographic characteristics. The model series utilizes a consistent structure, analyzing the following modal nests:

  • non-motorized vs motorized

    • (non-motorized) Walk vs. bike

    • (motorized) Auto vs transit

      • (Auto) Driver vs. passenger

      • (Transit) Walk access vs. drive access

These models are applied separately for each travel purpose, and there are separate models for trip productions and trip attractions. This means there are 5 models applied to 4 purposes for both trip ends (5*4*2), making 40 model applications in all.

Additionally, models are applied at the block scale in the “window area” and at the TAZ scale for the rest of the model area.

The application steps rely on two primary types of input:

  • Accessibility scores (generated by mode and purpose) pulled from csv’s.

  • Demographic data (HH types) are pulled from the labeled arrays created during the trip gen process that contain cross-classified households.

A mode choice labeled array is generated with the following dimensions:

  • Zones (TAZs or blocks depending on the scale of anlysis)

  • HH Size (“-“, “HHSize1”, “HHSize2”, “HHSize3”, “HHSize4p”)

  • Vehicle Ownership (“-“, “Veh0”, “Veh1”, “Veh2”, “Veh3p”)

  • Income (“-“, “Inc1”, “Inc2”, “Inc3”, “Inc4”)

  • Workers (“-“, “worker0”, “worker1”, “worker2”, “worker3p”)

  • Purpose (“-“, “HBW”, “HBSch”, “HBO”, “NHB”)

  • End (“P”, “A”)

  • Mode (“Walk”, “Bike”, “Drive”, “Passenger”, “Transit WAT”, “Transit DAT”)

This means mode shares are estimated by zone, household type (“-” for most attractions or NHB productions since these are non-home generators), purpose, and trip end. These shares can then be applied to yet another labeled array: trips-by-mode, which is the product of the mode choice and trip gen arrays.

Model application entails…

  • fetching appropriate sections of the demographic array (HH by type);

  • vectorizing, reindexing, and broadcasting accessibility arrays;

  • transforming values based on model coefficients as needed; and

  • adding model intercept constants.

All of this occurs in an in-memory array dimensioned to match the targeted section of the on-disk mode choice array. The accumulated values effectively apply a linear regression formula by adding coefficient-adjusted factors together. The mode choice models are, however, binomial logistic equations, and the linear results are used to determine the probability of selecting each mode in the nest as follows:

  • choice1 = exp(linear_result)/(1 + exp(linear_result))

  • choice0 = 1 - choice1

The script produces the output files listed below. Note that TAZ- and block-level analyses are run and stored independently. Block-level mode choice estimates are “patched” into the TAZ-level estimates when generating the Trips_by_mode_predist.csv table, but the H5 outputs are unadulterated results from each scale of analysis. In the distribution phase, block-level trip details are again patched into the regional trips by mode estimates for the window area, prior to distributing trips.

  • Mode_choice_taz.h5: LbArray

    TAZ mode choice estimates by the 8 dimensions listed above. These estimates do not account for the number of activities by type in each zone. Rather, they provide estimates of mode shares for a given activity type. These are then applied to trip generation estimates to make trips by mode tables (see below).

  • Mode_choice_block.h5: LbArray

    A block-level version of Mode_choice_taz.h5

  • Trips_by_taz.h5: LbArray

    TAZ trips by mode estimates for the 8 dimensions listed above. These estimates can be summarized for a true mode share estimate based on activities in each zone.

  • Trips_by_block.h5: LbArray

    A block-level version of Trips_by_taz.h5

  • Trips_by_mode_predist.csv csv

    A table of trip productions and attractions by TAZ. Block-level trips-by-mode estimates for the window area are patched into TAZ-level trips-by-mode estimates for the region.

Functions

The following functions are referenced in this script, from the wsa.mode_choice (or mc) submodule: