Project Meeting 2024.12.03

Agenda

• Telecommuting update - sensitivity tests

Technical

• Telecommuting update
  • Sensitivity tests
    • Purpose - test whether existing ActivitySim representation of telecommuting is behaviorally unattractive
    • Survey suggests that telecommuters make different number of non-mandatory tours compared to workers not working; confirmed with statistical tests
      • Size of tickers is observations
    • Ran base SANDAG ABM3 run
      • 65% of workers spent 0 days telecommuting; 10% worked exclusively from home; Rest are evenly split between 1-3 days telecommuting
      • Compared to survey, model has much higher number of non-mandatory tours (note, model may not be calibrated to 2022 survey; SANDAG confirmed this was not a focus of calibration to date)
      • Model has about the same level of non-mandatory tours across number of teleworking days per week
        • Existing ABM - can not calibrate it separately
      • Survey suggests telecommutes engage in shorter non-mandatory tours than those not working on the survey day
        • Model average durations do not change
      • Most likely a high telecommuting scenario will not produce different results
      • Survey suggests telecommuters are more likely to engage in non-mandatory tours outside of common work hours than those not working on the survey day
      • Model does not show differences in rate of escorting activities between rate of teleworking, while survey does show a difference
    • Not clear at this point what exactly needs to change to best enhance the model; Would be ideal to figure out what the simplest or most optimal fix is;
      • Would suggest pausing high teleworking test until this thought through a bit more

Sijia's comment - All these findings are exactly what we expected. The model does not have the same response as survey, which we have confirmed after survey analysis and the initial model run. Next step is exactly what we proposed to do in Task 3 and Task 4, we are going to add the software ability in model to differentiate telecommuters vs not working, give model users the tool to enhance it, figdet the SANDAG model to be closer to the survey. This is going as we expected in the scope and we are on track.