Telecommute Model
The Work Location model that predicts each worker’s usual work parcel already predicts workers who always work from home (the usual work parcel is the home parcel). Prior to this project, however, DaySim did not have a way to predict workers who have an out-of-work place but still choose to do paid work at home during the simulated weekday. The model could predict that a workers would not make a work tour to the usual workplace, but working at home was treated no differently than other reasons for not going to work, (a sick day, on vacation, scheduled day off, etc.).
An important feature of the work-at-home model (or telecommute model) is that it only predicts whether workers with out-of-home usual workplaces choose to work at home for 3+ hours on the simulated day. It does not directly predict whether that person (also) makes a commute tour or other type of work tour on that day. That is done indirectly in DAYSIM by using the prediction from the Work at Home model as an input to the Individual Person Day Pattern Model that predicts whether a person makes tours and/or extra stops for various travel purposes. Note that this is the same way that “work from home” (usual work parcel = home parcel) is handled in the Individual Person Day Pattern Model but using separate variables for “work at home” versus “work from home”. This two-stage method of predicting teleworking provides greater flexibility in applying the models and is consistent with how the survey data was collected.
The 2018 SACOG survey asked each worker at the end of each day how many hours of paid work the person had done at home during that day, regardless of whether that person had (also) commuted to work on that day or not. Figure 1 shows the unweighted distribution of responses to that question. About 80% of full-time and part-time workers with out-of-home workplaces did no paid work at home, while only 35% of those whose workplace is the home parcel did no paid work there on that day.
Figure 1: Hours doing paid work at home on survey day by worker type
Another 7% or so of full-time and part-time workers worked at home only or 1 or 2 hours. For those who worked at home more than two hours, the most common answers were 8 or 9 hours for full-time workers and those whose usual workplace is at home, but a much flatter distribution for part-time workers.
Many of the workers with out-of-home workplaces who reported doing paid work at home also commuted to the usual workplace on the same day. Table 1 shows that for those who worked at home for 0-2 hours, over 50% made commute tours on that day—similar to the overall percentage of 56.7%. For those who worked at home for 3 or more hours, only about one-third made commute tours to the usual workplace, with the values all quite similar ranging from 26.8% to 37.7%. Based on this result, a threshold of 2.5 hours was used for the binary “Work at Home” choice model, with those working 3 or more hours at home choosing the “work at home” alternative.
Table 1: Fraction of workers commuting to usual workplace by hours worked at home (unweighted)
The Work at Home model was estimated as a binary logit model using 9,221 worker-weekday observations (excluding workers whose usual work parcel is the home parcel). In 1,180 (12.8%) of the cases, the person chose to do paid work at home for more than 2.5 hours. The estimated coefficients are shown in Table 2. Part-time workers were somewhat less likely to work at home relative to full-time workers, while those in households with income below $50,000 (roughly $35,000 in year 2000 dollars), with income above $150,000, with no vehicles, and with at least one non-working adult and at least one child under age 18 were somewhat more likely to work at home.
Table 2: Estimation results for the Work at Home model
The rest of the variables in the model are based on the distribution of employment at the worker’s usual work parcel interacted with the worker’s household income (above or below $50,000). With the fraction of service jobs used as the base category, those whose work parcels have high fractions of medical, government, or retail/food service jobs were somewhat less likely to work at home. For most employment types, those with incomes above $50,000 were more likely to work from home. For industrial and office jobs, the coefficients have different signs depending on income level.
These results are generally consistent with findings elsewhere about what types of workers were most likely to work from home prior to the COVID-19 pandemic. These patterns may change post-pandemic, and the model provides the flexibility to adjust the coefficients to represent different future telework scenarios.
Note that it would probably give a better model if we could use the worker’s actual specific industry type/occupation and the worker’s own personal income, but neither of those were asked in the survey and neither are available in the synthetic population used for model application. Those questions are asked in the American Community Survey (ACS), so it may be worthwhile to ask them in future travel surveys and possibly include them in the population synthesis process.
The IndividualPersonDayPattern model has several new variables using the outcome from the new Work at Home model. These new variables are presented in Table 3. The most important variable is 134-W-WATHOME which has a coefficient of -1.21 and a t-statistic of -16.6. This indicates that those who work at home for more than 2.5 hours in the day are much less likely to make any work tours. This can be compared to variable 124-W-WAHOME which applies to people whose usual work parcel is the home parcel. That coefficient is more negative (-2.66) but with a similar t-statistic (-16.1). Other key new variables are 834-T-WATHOME (0.272 with t-statistic of 2.3) and 934-I-WATHOME (-0.191 with t-statistic of -1.6). These two variables indicate that those who work at home may make more home-based tours (somewhat offsetting the negative coefficients on most of the purpose-specific variables) but with fewer intermediate stops per tour.
Table 3: New Variables in Individual Person Day Pattern Model
| Variable | Coeff. | T-Stat. | Description/Alternative |
|---|---|---|---|
| W-WATHOME | -1.21 | -16.6 | workPurposeConstant * WorkAtHomeDuringTheDay |
| S-WATHOME | 0.533 | 1.3 | schoolPurposeConstant* WorkAtHomeDuringTheDay |
| E-WATHOME | -0.127 | -1.4 | escortPurposeConstant* WorkAtHomeDuringTheDay |
| P-WATHOME | -0.128 | -1.7 | personBusinessPurposeConstant* WorkAtHomeDuringTheDay |
| H-WATHOME | -0.154 | -2 | shopPurposeConstant* WorkAtHomeDuringTheDay |
| M-WATHOME | -0.0524 | -0.7 | mealPurposeConstant* WorkAtHomeDuringTheDay |
| D-WATHOME | -0.0958 | -1.2 | socialOrRecreationPurposeConstant* WorkAtHomeDuringTheDay |
| T-WATHOME | 0.272 | 2.3 | LN(max(1, number of tour purposes) * WorkAtHomeDuringTheDay |
| I-WATHOME | -0.191 | -1.6 | LN(max(1, number of stop purposes) * WorkAtHomeDuringTheDay |
The code for the new Work at Home choice model was programmed as a “pre-model” to the Individual Person Day Pattern Model, in the same was as the Automated Vehicle (AV) choice model had been programmed as a “pre-model” to the Auto Ownership model. This means that the coefficients to the model are input via new entries in the properties configuration file rather than as a separate coefficient (.F12) file.
In model application the outcome from the Work at Home model is written to the Person-Day file in the “workathome” field. A value of 0 indicates that the person does not work at home, while any value greater than 0 indicates that the person does work at home for more than the threshold number of hours.