2026-07-21 Executive Team

[joecastiglione]

Agenda

• Telecommute Frequency Model Update and Tests

Notes

Part 1: Telecommute Frequency Model (RSG Update)

• Estimation problem: the mode choice logsum was incorrectly signed (positive) and largely insignificant in the telecommute frequency model.
• Diagnosis steps taken by RSG:
  • Found the logsum was noisy due to the time-sensitivity (value of time) parameter; fixed by setting this parameter to 1 during estimation, which meaningfully reduced noise.
  • Noted the same time-sensitivity parameter is also used in disaggregate accessibility calculations elsewhere in the model and probably should be turned off there too, since accessibility is meant to represent a level playing field across zones.
  • Fixed a data coding issue related to industry variables in the estimation bundles.
  • Added constants for central business district (CBD)/downtown effects to account for unusual live-work patterns.
  • The key fix: adding an intrazonal variable (home and work in the same MGRA) — this had the largest impact and finally produced a correctly signed, significant, though still small, mode choice logsum coefficient. This raises the question of why workers would have an out of home regular work location that is within the same relatively small MGRA as their home. Is this real behavior, or a survey coding / response issue?
• How significant is the intrazonal home/out-of-home work location overlap issue? Sijia noted a potentially related but distinct finding — about 274 workers (~3% of the estimation sample) reported both significant telework and significant out-of-home work time, suggesting possible data/reporting inconsistencies in how telework and work tours are coded.
• Discussion — statistical significance vs. practical impact (Jeff Newman):
  • The 2023 survey data reflects a period where telecommuting was driven more by employer policy than individual choice, which may explain weak/insignitive distance and cost sensitivities.
  • Recommended sensitivity testing to identify which parameters actually matter for model outcomes and for future survey design, distinguishing statistical significance from meaningful impact.
• MWCOG considerations (Ray Ngo): limited capacity for lots of model runs; MWCOG will rerun their prior sensitivity test with the updated specification and report back before scoping further work.
• Agreed next steps:
  • Send the finalized parameters to MWCOG to rerun their original sensitivity test (exact same test and summary statistics as before) using the new logsum.
  • Based on those results, decide whether further estimation work (e.g., testing a direct destination monetary parking cost term) is warranted.
  • After MWCOG findings and discussion, RSG to finalize the specification (eg. decide on retaining vs. dropping insignificant coefficients).
  • Joe to follow up with SANDAG on possible data/coding issues behind the telework vs. out-of-home work discrepancy — flagged as an issue to raise separately, outside current task order scope.

Action Items

• RSG (Joel/David): Finalize telecommute model specification and send to MWCOG for sensitivity retesting.
• MWCOG (Ray): Rerun prior sensitivity test with updated logsum specification and summarize results.
• Joe: Follow up with SANDAG (or relevant party) re: possible data/coding issue behind significance of workers with out-of-home work locations in the same MGRA as home locations.