Feature: Improve Telecommute Representation

[DavidOry]

A short-coming of the original TM1 that persists, to some degree, in TM2 is the representation of working from home. In TM1, there was no explicit representation of workers working from home. The procedure went as follows:

1. Workers are generated in the synthetic population.
2. Every worker is assigned a "usual work location".
3. Each worker is assigned a daily activity pattern, either "mandatory", "non-mandatory", or "home".
4. Workers assigned a "mandatory" pattern generate at least one work tour to their usual work location.

To simulate workers who worked full time at home (i.e., did not have a usual work place) as well as telecommuters, MTC manipulated the coordinated daily activity pattern model (CDAP) to increase, over time, the number of workers engaged in a "non-mandatory" or "home" pattern, therefore reducing the quantity of work travel. But the model treated those working from home full-time and those telecommuting in an identical manner as workers who stayed home due to an illness.

With TM1.6, MTC modified these procedures as follows:

1. Workers are generated in the synthetic population.
2. Every worker is assigned a "usual work location".
3. A new model predicted whether a worker was likely to telecommute based on the mix of employment by industry at the work location.
4. Each worker is assigned a daily activity pattern, either "mandatory", "non-mandatory", or "home". Workers who are predicted to telecommute in Step 3 are assigned either a "non-mandatory" or "home" pattern.

This approach had the following benefits:

• Telecommuters can be explicitly identified in the simulation results.
• Telecommuting is a function of employment industry.

TM2 is similar to TM1, with the notable addition of an explicit work from home model, which identifies simulated workers that never work outside the home. The steps in TM2 are therefore:

1. Workers are generated in the synthetic population.
2. Some number of workers are assigned to work from home, i.e., never travel to a usual work location outside the home.
3. Workers not assigned to work from home in Step 2 are assigned a "usual work location".
4. Each worker is assigned a daily activity pattern, either "mandatory", "non-mandatory", or "home".
5. Workers assigned a "mandatory" pattern generate at least one work tour to their usual work location.

While this approach improves upon the TM1 formulation by explicitly identifying those who work only from home, it still treats telecommuters in an identical fashion as workers staying home due to illness.

For TM2.2, we will modify the Java code to combine the current TM2 approach with an improved version of the TM1.6 approach. The model steps will be as follows:

1. Workers are generated in the synthetic population.
2. Some number of workers are assigned to work from home, i.e., never travel to a usual work location outside the home.
3. Workers not assigned to work from home in Step 2 are assigned a "usual work location".
4. Each worker is assigned a daily activity pattern, either "mandatory", "non-mandatory", or "home".
5. Each worker assigned a "mandatory" pattern will be assigned a telecommuting status in a new "explicit telecommute" model. This model will identify whether or not the worker is telecommuting on the simulation day and will be taken from the TM1.6 implementation.
6. Explicit telecommuters, as identified in Step 5, will have their daily activity pattern updated to either "non-mandatory" or "home", based on a fixed probability distribution.

The advantages of this approach relative to TM2 are the same as in TM1.6, specifically:

• Telecommuters can be explicitly identified in the simulation results.
• Telecommuting is a function of employment industry.

The minor change in this approach relative to TM1.6 is that the explicit telecommuting model comes after CDAP, not before. We therefore benefit from the coordination across household members that CDAP provides in determining who will work on the simulation day. For workers subsequently identified as telecommuters by the explicit telecommuting model, the mandatory pattern is updated to reflect the lack of a work tour.

cc: @lmz, @gregerhardt, @FlaviaTsang, @AshishKuls, @vivverma9

[DavidOry]

@vivverma9 Here's an employment data crosswalk.

https://github.com/BayAreaMetro/travel-model-one/blob/master/utilities/taz-data-baseyears/naics_recode/2020_naics_recode.csv

Please make a scatter plot showing the employment by superdistrict, comparing your 2015 data in the new categories and the 2015 data in the link below in the old categories.

https://mtcdrive.app.box.com/v/pba50-2015-TM152-IPA-17

[DavidOry]

One minor issue with the implementation. The CDAP model has a joint tour indicator, which helps ensure that the model generates enough households that can generate a joint tour to satisfy the joint tour generation target. For example, a household annotated as NMj is eligible for a joint tour; a household annotated as HM0 is not. With the telecommuting implementation, we are replacing the M for telecommuters with either N or H, via a probability distribution. In order not to disturb the joint tour potential, we will implement the following rule:

1. If the household has a joint eligible pattern, e.g., NMj, telecommuters will be assigned a N pattern.
2. If the household does not have a joint eligible pattern, e.g., HM0, telecommuters will be assigned either N or H, depending on the probability distribution.

@gregerhardt, @lmz, @FlaviaTsang cc: @vivverma9

[vivverma9]

Here is a scatterplot comparing county-wide employment totals for all the six categories. Looks as expected.

[DavidOry]

Placeholder for 2015 is 12 percent work from home (total) with a 6 percent permanently work from home. @gregerhardt, @vivverma9

[gregerhardt]

Here is a link to our TM 1.6 Targeted Calibration & Validation Report, but the relevant bits are below:

Box 3.1: The WFH rate in ACS 2015 is 5.6%, whereas that of the Travel Model 12.4%. Their difference can be attributed to two main factors: 1) Nuances in definitions: In the ACS, the journey to work question (which also captures working from home) is phrased as follows: “How did this person usually get to work LAST WEEK?” Thus, the workers who work from home 1 or 2 (or even 3 in some cases) days a week would not mark the “Worked from home” option even if they frequently work from home. In contrast, the Travel Model represents the Bay Area residents’ travel on a typical weekday. Some portions of the workers who work from home 1 or 2 days a week are included. Because of this definitional difference, the WFH rate in the model is expected to be higher than that reported by the ACS. 2) Potential underestimation by ACS: There is evidence indicating that ACS tends to underestimate actual WFH rates. Staff analyzed the results of the Bay Area Transportation Study -- a comprehensive week-long travel diary survey conducted in the fall of 2018 and spring of 2019. This survey asked respondents whether they traveled to work and/or teleworked on each day of survey participation. This survey question is more aligned with the Travel Model’s definitions than the ACS’s question. Using weighted data representing a “typical” weekday (Monday through Thursday), the survey revealed substantially higher WFH rates (12.4%) in 2018/2019 than in the 6.5% reported by the ACS.

[vivverma9]

After calibration (adding an alternative specific constant of -0.636), the model results in a 6.27% telecommuting rate (from the explicit telecommute model). The population which permanently works from home (works at home) is 2.5% (from the work location choice model).

[vivverma9]

<html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:x="urn:schemas-microsoft-com:office:excel" xmlns="http://www.w3.org/TR/REC-html40">

Funnel - Telecommute and CDAP |   |   -- | -- | -- Total No. of HHs | 276802 |   Total No. of persons | 744456 |   No. of persons who telecommute | 46692 | 6.27% No. of persons who belong to a HH of size >2 | 29211 | 3.92% No. of persons who belong to a HH which has Joint pattern | 4867 | 0.65% No. of persons who have at least two other people in HH apart from telecommuters with M or N pattern | 4867 | 0.65% No. of such households | 4391 | 1.59%

[vivverma9]

Changed calibration constant to 0.355, share of telecommuters is now 12.44% (this excludes the 2.5% who permanently work from home or "WORK_AT_HOME").

[vivverma9]

Updated Funnel after calibration <html xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:x="urn:schemas-microsoft-com:office:excel" xmlns="http://www.w3.org/TR/REC-html40">

Funnel - Telecommute and CDAP |   |   -- | -- | -- Total No. of HHs | 276802 |   Total No. of persons | 744456 |   No. of persons who telecommute | 92623 | 12.44% No. of persons who belong to a HH of size >2 | 57910 | 7.78% No. of persons who belong to a HH which has Joint pattern | 9571 | 1.29% No. of persons who have at least two other people apart from telecommuters with M or N pattern | 8347 | 1.12% No. of such households | 7186 | 2.60%