Update SDIA Nonmandatory Tour Destination Model Configurations Using Survey Data

[mwe-sandag]

The current airport.SAN destination choice (non_mandatory_tour_destination) model uses a mixture of geographic characteristics and distance from the airport to distribute trips across the SD region for each market segment. We would like the WSP team to explore and potentially implement removing the distance considerations from the destination choice model and replacing them with a PMSA-based approach derived from the survey data. We feel that this simplifies model assumptions, simplifies model performance analysis, and mitigates assumptions that decisions to use SDIA are greatly affected by proximity to SDIA (particularly for residents).

We would like to have a conversation about the impact of these changes before estimating new NMTD coefficients, as well as a short literature review of other MPO's airport and appropriate special market models to understand other's approaches.

The current NMTD configs are here: nmtd specs. Note that the NMTD configs on the airport_update branch are currently different from the configs on the main branch because of the exclusion of income data in trip mode choice modeling.

Additional config files:

• size terms that determine MAZ's generation based on land_use data
• nmtd yaml
• nmtd coef values

[mwe-sandag]

@rcu @DavidOry would like to have a conversation about WSP working on an updated destination choice model using the survey data at the meeting Oct. 14th!

[DavidOry]

@rcu @DavidOry would like to have a conversation about WSP working on an updated destination choice model using the survey data at the meeting Oct. 14th!

Sounds good @mwe-sandag. CC'ing @AshishKuls and @rickcurry (Rick's new GitHub handle). One question @mwe-sandag: by "PMSA-based approach", do you mean removing all of the distance coefficients and replacing them with PMSA-specific dummy coefficients? Can you please post here a link to the PMSA boundaries? thanks.

[rickcurry]

We have a copy of PMSA that was provided during the Airport Survey work that should conform to SR15 geographies.

[rickcurry]

Comments from Dave:

• I am able to re-run the scripts Michael created.
• For the size terms, because the scale is arbitrary, it is useful to rescale the coefficients so the largest value is 1.0.
• For the polynomial to spine, I had an assumption in the script that forced the spline to be monotonically decreasing. This assumption fails in cases like the one Michael identified when the slope is positive. I’ve updated the script (https://github.com/SANDAG/Survey_Airport/blob/model_estimation/scripts/polynomial-to-splines.Rmd) to add a parameter, so the user can decide whether they want the slopes to be monotonically flattening. The polynomials and splines can now be as similar to each other as we’d like. Note that, generally speaking, positive impedances are not desirable behaviorally and a sign that the size terms are missing something. Meaning, the assumption of a location choice model is that the size term sufficiently describes the attractiveness of a place. If that’s true, the impedance should be strictly negative (i.e., why would I go farther to reach a place that is identical to another place?). Our size terms, of course, never adequately describe places, and very desirable places, such as La Jolla, that are far from the airport, can lead to positive impedance expressions.