TAZ | TOTHH | HHPOP | RETEMP | INDEMP | OTHEMP | TOTEMP |
|---|---|---|---|---|---|---|
2,138 | 0 | 0 | 0 | 0.0000 | 0 | 0.0000 |
2,140 | 0 | 0 | 0 | 0.0000 | 0 | 0.0000 |
2,141 | 0 | 0 | 0 | 0.0000 | 277 | 277.0000 |
2,149 | 0 | 0 | 0 | 0.0000 | 796 | 796.0000 |
2,170 | 0 | 0 | 3 | 359.1667 | 71 | 433.1667 |
TAZ | TOTHH | HHPOP | RETEMP | INDEMP | OTHEMP | TOTEMP |
|---|---|---|---|---|---|---|
2,138 | 0 | 0 | 0 | 0 | 0 | 0 |
2,140 | 0 | 0 | 0 | 0 | 0 | 0 |
2,141 | 0 | 0 | 0 | 0 | 277 | 277 |
2,149 | 0 | 0 | 0 | 0 | 796 | 796 |
2,170 | 0 | 0 | 3 | 359 | 71 | 433 |
TAZ | TOTHH | HHPOP | RETEMP | INDEMP | OTHEMP | TOTEMP |
|---|---|---|---|---|---|---|
2,138 | 7,430.833 | 17,810.84 | 4.333333 | 0.0000 | 76.16667 | 80.5000 |
2,140 | 0.000 | 0.00 | 610.333333 | 4.0000 | 7,389.83333 | 8,004.1667 |
2,141 | 0.000 | 0.00 | 1,449.333333 | 0.0000 | 5,363.16667 | 6,812.5000 |
2,149 | 0.000 | 0.00 | 962.000000 | 1.5000 | 7,372.50000 | 8,336.0000 |
2,170 | 0.000 | 0.00 | 7.000000 | 357.3333 | 106.33333 | 470.6667 |
TAZ | TOTHH | HHPOP | RETEMP | INDEMP | OTHEMP | TOTEMP |
|---|---|---|---|---|---|---|
2,138 | 7,431 | 17,811 | 4 | 0 | 76 | 80 |
2,140 | 0 | 0 | 610 | 4 | 7,390 | 8,004 |
2,141 | 0 | 0 | 1,449 | 0 | 5,363 | 6,812 |
2,149 | 0 | 0 | 962 | 2 | 7,372 | 8,336 |
2,170 | 0 | 0 | 7 | 357 | 106 | 471 |
In the WFRC model, this change is trivial to implement. The model uses the land use/socioeconomic data directly, so the only adjustment needed is replacing the data for the specific TAZs with the 2050 data. As noted previously, all other TAZs have the same land use data as in the 2019 baseline scenario.
-ActivitySim requires two changes for this scenario. The first is an update to the TAZ-level land use and socioeconomic data, which is identical to the process for the WFRC model. The second is an updated synthetic population. In order to keep consistency between model scenarios, a new population was created only for the 5 affected TAZs and joined to the existing synthetic population. There were no individuals or households in the affected zones in the existing synthetic population, so no individuals or households needed to be removed before joining the two populations.
-Creating the new synthetic population followed a similar process as in the baseline scenario (Section 3.2.1), but used the new land use data as new TAZ-level controls. However, many of the controls for PopulationSim use tract-level data from the Census, and existing Census data for these controls is unrepresentative of the new development. Because of this, the Census tract covering the Gateway area in Salt Lake City (see ?fig-gateway-tract) is used to represent the new development patterns at The Point. The income distribution, etc. of this area will therefore match that of the Gateway area, though the TAZ-level controls and land use/socioeconomic data will match the WFRC projections for 2050.
-? Note in realistic scenario you could use just the land use forecast as controls directly, rahter than the whole census shenanigans but we didn’t do that since we wanted the models to be independent.
+ActivitySim requires two changes for this scenario. The first is an update to the TAZ-level land use and socioeconomic data, which is identical to the process for the WFRC model. The second is an updated synthetic population.
+In order to keep consistency between model scenarios, a new population was created only for the 5 affected TAZs and joined to the existing synthetic population. There were no individuals or households in the affected zones in the existing synthetic population, so no individuals or households needed to be removed before joining the two populations.
+Creating the new synthetic population followed a similar process as in the baseline scenario (?sec-populationsim), but used the new land use data as new TAZ-level controls. However, many of the controls for PopulationSim use tract-level data from the Census, and existing Census data for these controls is unrepresentative of the new development, due to the lack of households and economic activity at the site of the development. Because of this, the Census tract covering the Gateway area in Salt Lake City (see Figure 3.3) is used to represent the new development patterns at The Point. The income distribution, etc. of this area will therefore match that of the Gateway area, though the TAZ-level controls and land use/socioeconomic data in the area will match the WFRC projections for 2050.
+
+In a more realistic case, a transportation agency would have forecasted land use and socioeconomic data that could be used as controls to PopulationSim, rather than using a (different) representative Census tract for new development. However, our ActivitySim implementation only needs to be within a rough approximation of the WFRC model for the purposes of this project, and the method used here results in reasonable accuracy between the models.
+Additionally, our ActivitySim implementation is designed to be independent from the WFRC model where feasible.
In a trip-based model, it is relatively easy to calculate person-miles traveled from trips produced in the new development zones (see Figure 4.3), and where they are being attracted (see Figure 4.5 (a)). However, those living in the new development make many more trips than only those produced in their home zone. In a trip-based model, this is modeled with “non–home-based” trips. However, it is difficult to know how best to distribute non–home-based trips, since by definition these trips do not have an origin or destination in the zone that generated them. cite something about how it’s a problem
+There are several kinds of analyses an agency likely would want to do in assessing the effects of a change in land use. Chief among them would be an analysis of the new trips resulting from the development. This could include analysis in the number of trips, the distance traveled, and where the trips are being made.
+Both model types allow for very easy analysis of trip numbers and lengths, as the WFRC model outputs origin-destination trip tables directly by mode and purpose, and ActivitySim outputs a list of trips containing information on origin, destination, and mode. Figure 3.4, ?fig-lu-personmiles-abm, for example, show the new trip-miles produced in the updated zones for the WFRC mode and ActivitySim, respectively. However, there is a crucial difference between the model types, and that is the treatment of trips that do not begin or end at the home.
By contrast, since an ABM models individuals explicitly, it is easy to follow the daily trips of any individual. ?fig-lu-personmiles-abm shows a similar plot of person-miles traveled, but of all trips made by individuals living in the new development zones. We can additionally make a distinction between trips produced in the individuals’ home zones and those produced elsewhere. Note that in a trip-based model, a round-trip from home to work and back is regarded as two trips produced in the home zone. ActivitySim does not deal with productions and attractions in the same way, so for comparison between models we are counting a trip with an origin or destination in the home zone as produced by that zone (e.g. in Figures 4.4 and (asim_lu_new_desire_map?)).
Figure 4.5 shows desire line plots of trips generated by the new development zones. For home-based trips, it is easy to see where trips are produced and attracted to, and this is shown in Figure 4.5 (a). However, the non–home-based trips are more difficult to analyze. Because non–home-based trips are relocated to zones other than where they were generated, it is impossible to filter to only those non–home-based trips generated by the new zones.
-The best approach is to take the difference in non–home-based trips between the scenarios (Figure 4.5 (b)), though this presents two problems. The first is that for real-world analyses, where land use changes are not isolated, it becomes impossible to tell which non–home-based trips are generated by which developments, as the trips are seen only in aggregate. The second problem is with the distribution of the non–home-based trips. Though the exact distribution process depends on the specific model used, Figure 4.5 (b) shows inconsistencies with the approach in the WFRC model. Between the baseline scenario and the updated land use scenario, many pairs of zones saw an increase in the number of non–home-based trips between them, but other pairs saw a decrease. Additionally, all pairs of zones that saw an increase in non–home-based trips include a production or attraction to the new development zones. The WFRC model redistributes non–home-based trips as part of its network assignment step, which occurs after all trip matrices are created. it’s either this or the only increase in nhb trips is from non-residents, which doesn’t make much sense. The only measure of the increase of non–home-based trips after redistribution is therefore in the highway network, reported as roadway volumes. This compounds with the first problem and makes it even more difficult to tell where the non–home-based trips are coming from.
+In the WFRC model (and in many trip-based models), homes produce trips with different trip purposes, including home-based work, home-based other, and non–home-based trips. “Home-based” trips have an origin or destination at the home, and are fairly straightforward to model, as the destination choice step can take for granted that these trips have one trip end in the zone that produced them. In addition to home-based trips, though, individuals make many “non–home-based” trips, which do not have an origin or destination at the home (e.g. traveling from work to a grocery store). Non–home-based trips can be a significant portion of total travel, as Figure 3.4 shows, but are not as straightforward to model as home-based trips.
+By contrast, an ABM models individuals and their travel explicitly, and this makes the treatment of non–home-based trips much more straightforward. Each trip is tied to a specific individual with a defined home location, and so no extra “redistribution” step is needed to analyze non–home-based trips: these are “built-in” to each individual’s tour pattern. In fact, as Figure 3.5 shows, non–home-based trips can occur as part of any tour type/purpose; there is no separate “non–home-based” purpose in ActivitySim. Note that Figure 3.5 counts person-miles by tour purpose, using the purposes as defined in ActivitySim, rather than converting the ActivitySim trips to the “common” trip purposes as discussed in ?sec-baseline-calibration.
+In addition to looking at total person-miles traveled, it is also useful to analyze the origins and destinations of the new trips. One common way to visualize trip origins and destinations is with desire lines, which show lines for each trip origin/destination pair. The thickness of the line represents the number of trips between the pair of zones.
+Figure 3.6 shows a desire line plot by mode for all home-based trips produced in the new development zones. This figure is in line with what is expected: non-motorized trips are quite short, transit trips are exclusively to downtown areas, and many auto trips are made with varying lengths. Figure 3.6 also shows a similar mode split to Figure 3.4. Although the former depicts the number of trips and the latter depicts trip distance, there is a rough correlation between trip count and miles traveled, so it is not surprising that the mode split is similar between the figures.
+
There is difficulty in analyzing the non–home-based trips, however. Because non–home-based trips by definition have neither an origin or destination at the home (where trips are produced in the trip generation step), these trips happen exclusively between zones that did not produce them. It is difficult therefore to know how best to redistribute non–home-based trips, as they could in reality have any number of origins and/or destinations. Though modeling the destination choice for non–home-based trips could be done via a similar process to that of home-based trips, the origins of these trips need to be modeled as well.
+There are several methods to redistribute non–home-based trips in trip-based models. One approach is to assign non–home-based trip origins in a similar manner to trip destinations as part of the trip distribution step, either with a gravity model or some other distance-decay function. The destinations of these non–home-based trips can then be assigned as if they were any other trip. This results in non–home-based trips that are more likely to have both an origin and destination relatively near to the home.
+The WFRC model takes a different approach. Here there are two sources of information for non–home-based trip ends: a production model and an attraction model. In the trip generation step, households produce non–home-based trips similarly to any other trip purpose. However, the trips produced in this step determine only the quantity of non–home-based trips, not the trip ends. The distribution of non–home-based trips is determined by a trip attraction model (largely based on TAZ employment). This distribution is then globally scaled to match the total quantity of non–home-based trips produced in the trip generation step.
+Typically in a trip-based model, once non–home-based trips are assigned trip ends, they have no connection to the homes/zones that produced them, and are treated as “belonging” to either the origin or destination zone. Because of this, it is not possible to simply filter trips by origin or destination as can be done with the home-based trips. Instead, we took the difference between the entire non–home-based trip matrices in both this scenario and the baseline scenario.
+Figure 3.7 shows the desire line plot for the difference in non–home-based trips between this scenario and the baseline scenario. Two things are immediately noticeable from this plot.
+The first is that many pairs of zones saw a decrease in non–home-based trips between them compared to the baseline scenario (i.e. there were more non–home-based trips in the baseline scenario between these zones). Certainly it makes little sense to predict fewer trips as the result of added population and employment. However, this is in fact not an overall decrease in non–home-based trips; these trips are simply being assigned trip ends in different locations due to the nearby change in land use.
+The second is that all pairs of zones that saw an increase in non–home-based trips include an origin or destination in the new development, i.e. the home zones of the new population. Because the change in employment was much more significant than the change in population (see Table 3.1), many more non–home-based trip ends were attracted to the development zones compared to the relatively little global increase in non–home-based trips due to the population. Both effects (the global increase in and the changed distribution of non–home-based trips) are present in the model, but the effects are impossible to separate.
+
+
As mentioned, an ABM allows for tracking of individuals explicitly, and so analyzing non–home-based trips is much more straightforward. ?fig-lu-desire-abm shows desire lines of all trips made by individuals living in the new development zones, colored by place of production. It is also easy to see how trips are related to each other, as each individual has a specific sequence of trips. The individual nature of an ABM avoids entirely the problems trip-based models have with non–home-based trips. In a complicated land use forecast, each development’s full contribution to network congestion can be analyzed individually.
+As mentioned, an ABM allows for tracking of individuals explicitly, and so analyzing non–home-based trips is much more straightforward. ?fig-lu-desire-abm shows desire lines of all trips made by individuals living in the new development zones. Non–home-based trips are colored differently from home-based trips.
+It is easy to connect non–home-based trips to their place of production, as each trip is linked to a specific individual (who has a defined home location). It is also easy to see how trips are related to each other, as each individual has a specific sequence of trips. The individual nature of an ABM avoids entirely the problems trip-based models have with non–home-based trips. In a complicated land use forecast, each development’s full contribution to network congestion can be analyzed individually.
purpose | mode | cube_tr | cube_by | cube_diff_pct | asim_tr | asim_by | asim_diff_pct |
|---|---|---|---|---|---|---|---|
hbo | carpool | 2,701,032.32 | 2,702,272.42 | -0.00045891062 | 2,145,135 | 2,148,429 | -0.0015332133 |
hbo | drive_alone | 1,394,094.88 | 1,394,415.10 | -0.00022964582 | 698,809 | 700,133 | -0.0018910693 |
hbo | nonmotor | 510,103.39 | 510,143.12 | -0.00007788579 | 611,996 | 613,134 | -0.0018560380 |
hbo | transit | 38,911.62 | 37,346.03 | 0.04192117151 | 395,524 | 389,780 | 0.0147365180 |
hbw | carpool | 256,673.69 | 257,805.01 | -0.00438827233 | 256,550 | 258,459 | -0.0073860844 |
hbw | drive_alone | 1,326,190.82 | 1,328,609.31 | -0.00182031601 | 1,010,565 | 1,012,180 | -0.0015955660 |
hbw | nonmotor | 76,396.23 | 76,506.40 | -0.00144004292 | 145,845 | 145,957 | -0.0007673493 |
hbw | transit | 52,379.76 | 48,751.83 | 0.07441618268 | 256,790 | 253,176 | 0.0142746548 |
nhb | carpool | 1,273,013.66 | 1,273,317.20 | -0.00023838255 | 936,408 | 938,056 | -0.0017568248 |
nhb | drive_alone | 951,406.76 | 951,560.65 | -0.00016172558 | 714,854 | 716,143 | -0.0017999198 |
nhb | nonmotor | 146,409.27 | 146,403.78 | 0.00003748332 | 156,587 | 156,819 | -0.0014794126 |
nhb | transit | 13,869.89 | 13,453.28 | 0.03096704282 | 161,800 | 159,935 | 0.0116609873 |
Purpose | WFRC Model | ActivitySim | |||||
|---|---|---|---|---|---|---|---|
Baseline Trips | Improved Transit Trips | Change | Baseline Trips | Improved Transit Trips | Change | ||
Home-based Work | Drive Alone | 1,328,609 | 1,326,191 | -0.2% | 1,012,180 | 1,010,565 | -0.2% |
Carpool | 257,783 | 256,654 | -0.4% | 258,459 | 256,550 | -0.7% | |
Local Transit | 37,935 | 36,494 | -3.8% | 232,222 | 233,426 | 0.5% | |
Commuter Rail | 10,821 | 15,891 | 46.9% | 19,846 | 22,265 | 12.2% | |
Ridehail | 1,108 | 1,099 | -0.8% | ||||
Non-motorized | 76,506 | 76,396 | -0.1% | 145,957 | 145,845 | -0.1% | |
Home-based Other | Drive Alone | 1,394,415 | 1,394,095 | 0.0% | 700,133 | 698,809 | -0.2% |
Carpool | 2,702,277 | 2,701,039 | 0.0% | 2,148,429 | 2,145,135 | -0.2% | |
Local Transit | 33,168 | 32,583 | -1.8% | 195,062 | 194,649 | -0.2% | |
Commuter Rail | 4,180 | 6,332 | 51.5% | 81,094 | 87,337 | 7.7% | |
Ridehail | 113,624 | 113,538 | -0.1% | ||||
Non-motorized | 510,143 | 510,103 | 0.0% | 613,134 | 611,996 | -0.2% | |
Non–home-based | Drive Alone | 951,561 | 951,407 | 0.0% | 716,143 | 714,854 | -0.2% |
Carpool | 1,273,279 | 1,272,977 | 0.0% | 938,056 | 936,408 | -0.2% | |
Local Transit | 12,213 | 12,068 | -1.2% | 107,526 | 108,395 | 0.8% | |
Commuter Rail | 1,243 | 1,806 | 45.3% | 12,317 | 13,344 | 8.3% | |
Ridehail | 40,092 | 40,061 | -0.1% | ||||
Non-motorized | 146,404 | 146,409 | 0.0% | 156,819 | 156,587 | -0.1% | |
+
With greater access the commuter rail by decreasing the headways, we wanted to see how the ridership changed in this scenario. We also wanted to see where the commuter rail riders were coming from.
-Analyzing this with the trip-based model was straightforward since one of the outputs is a file listing the amount of trips made by each form of transit. There are also more detailed matrices that shows commuter rail transit (CRT) trip productions and attractions. There is a matrix for driving to the CRT and a matrix for walking to the CRT. With these matrices and the taz shape file, we were able to to visualize the catchment area in a plot like (another figure that we can show)
-Implications: We saw that there was more than a 30% increase in commuter rail transit trips in this scenario but not much change in the other transit trips. We could see the areas where peoples’ CRT trips began on a zonal and a district level and noticed a big increase in the people coming from zones that were closer to the new stations that were part of the extended rail line. (I think more can be added here as well)
-Limitations: Apart from the increase in CRT trips and the catchment areas, there wasn’t much more we could see with the trip-based model. By connecting the catchment areas with what we know of the zonal SE data, we could make some assumptions about the income of the new people taking the commuter rail, but we couldn’t find anything more about the demographics of the riders.
+One particularly interesting analysis that can be done with an ABM is to see who changed modes with the improved transit. Because trips are modeled individually rather than in aggregate, it is possible to identify trips that switch modes between the scenarios. Figure 4.2 shows the distribution of these “switched” trips, grouped by tour purpose. These are trips that are “the same” between scenarios and differ only by mode.
+For the purposes of this analysis, trips are considered “the same” between scenarios if they share the following: person id, origin and destination zones, time of day1, and tour and trip purpose. Most of these trips also share the same mode, which is to be expected, but many do not. Figure 4.2 is filtered to show only trips that do not share the same mode between scenarios.
+There is some amount of randomness in the way ActivitySim determines trip modes, though. This randomness is seen partly in trips that switch away from commuter rail despite the improved commuter rail service, as well as some trips that switch to modes other than commuter rail, especially to “Drive Alone”. Although, part of the switch from “Carpool” to “Drive Alone” can be explained as previously-carpool trips where all but one vehicle occupant switched to another mode, leaving one person in the vehicle for the trip. Overall, though, the randomness is not a significant percentage of the overall mode switching seen in Figure 4.2.
purpose | transit_trips | TOTHH | ALLEMP | med_income |
|---|---|---|---|---|
hbo | 38,911.62 | 455.8946 | 425.9752 | 51,230.00 |
hbw | 52,379.76 | 467.8071 | 356.6345 | 60,187.16 |
nhb | 13,869.89 | 103.1667 | 1,364.6551 | 50,921.00 |
purpose | transit_trips | income | age |
|---|---|---|---|
hbw | 256,790 | 79,647.82 | 37 |
hbo | 395,524 | 70,783.23 | 33 |
nhb | 161,800 | 69,411.20 | 36 |
Mode choice is not the only step of ActivitySim affected by the improved transit service, however. In fact, there are many trips that do not have a match between scenarios, where origin, destination, time of day and/or purpose differ. The number of trips an individual makes may also differ between scenarios, as each person’s DAP is partially dependent on accessibility measures. Notably, Figure 4.2 also does not include any of these trips; the figure only shows trips which do have a match between scenarios.
+ABMs also allow for even more granular analysis than shown in Figure 4.2. For example, Figure 4.3 shows the trip modes of at-work subtours made by individuals who switched their work tour mode away from “Drive Alone”. The figure shows the at-work subtour trip modes for all these individuals, not just those who also switched their at-work subtour trip modes. These results are essentially as expected. All trips that were “Drive Alone” in the baseline scenario switched to carpool, and there was virtually no mode switching otherwise, except a few trips that switched from Carpool to Non-motorized. This switching can again be largely explained by the randomness in ActivitySim’s mode choice models, and again is relatively insignificant.
+
Figure 8 shows the increased productions and attractions of the “drive to CRT” mode by district. These could as an example be further analyzed by TAZ/district median income or similar variables. However, there is no indication of which types of individuals are switching their mode. In an ABM, this can be analyzed. Figure 9 shows the trips that switched modes from the base scenario, as well as which mode they switched to. Though some of this switching is due to the internal randomness in ActivitySim, the majority of the mode-switching was from auto to transit, and this shows a clear increase in transit usage over the base scenario.
-ActivitySim also has an “at-work” tour purpose, indicating a subtour from the workplace. The team analyzed the trip modes of these subtours for the individuals who switched to transit for their work tour. Figure 10 shows that most people did not switch subtour modes from the base scenario. The similar number of switches between auto and non-motorized modes indicates that the switching here may be mostly due to ActivitySim’s randomness and not any changes in the network.
+
+
+
Analyzing this with the trip-based model was straightforward since one of the outputs is a file listing the amount of trips made by each form of transit. There are also more detailed matrices that shows commuter rail transit (CRT) trip productions and attractions. There is a matrix for driving to the CRT and a matrix for walking to the CRT. With these matrices and the taz shape file, we were able to to visualize the catchment area in a plot like (another figure that we can show)
+Implications: We saw that there was more than a 30% increase in commuter rail transit trips in this scenario but not much change in the other transit trips. We could see the areas where peoples’ CRT trips began on a zonal and a district level and noticed a big increase in the people coming from zones that were closer to the new stations that were part of the extended rail line. (I think more can be added here as well)
+Limitations: Apart from the increase in CRT trips and the catchment areas, there wasn’t much more we could see with the trip-based model. By connecting the catchment areas with what we know of the zonal SE data, we could make some assumptions about the income of the new people taking the commuter rail, but we couldn’t find anything more about the demographics of the riders.
purpose | transit_trips | TOTHH | ALLEMP | med_income |
|---|
purpose | transit_trips | income | age |
|---|---|---|---|
hbw | 256,790 | 79,647.82 | 37 |
hbo | 395,524 | 70,783.23 | 33 |
nhb | 161,800 | 69,411.20 | 36 |
+
ActivitySim also has an “at-work” tour purpose, indicating a subtour from the workplace. The team analyzed the trip modes of these subtours for the individuals who switched to transit for their work tour. Figure 10 shows that most people did not switch subtour modes from the base scenario. The similar number of switches between auto and non-motorized modes indicates that the switching here may be mostly due to ActivitySim’s randomness and not any changes in the network.
+ +ActivitySim models time of day as the “departure hour” for each trip. If two trips share the same departure hour, they are considered here to have happened at the same time.↩︎
name | 2019_tc | 2050_tc | 2019_wfh | 2050_wfh |
|---|---|---|---|---|
Retail | 0.0270 | 0.0725 | 0.0212 | 0.0250 |
Food | 0.0187 | 0.0503 | 0.0146 | 0.0173 |
Manufacturing | 0.0202 | 0.0545 | 0.0159 | 0.0188 |
Office | 0.0666 | 0.1801 | 0.0522 | 0.0623 |
Gov't/Education | 0.0167 | 0.0456 | 0.0131 | 0.0157 |
Health | 0.0286 | 0.0721 | 0.0211 | 0.0249 |
Agriculture | 0.0693 | 0.1683 | 0.0544 | 0.0582 |
Mining | 0.0053 | 0.0143 | 0.0042 | 0.0050 |
Construction | 0.0328 | 0.0882 | 0.0257 | 0.0304 |
Other | 0.0537 | 0.1458 | 0.0421 | 0.0504 |
We adjusted the remote work models in ActivitySim using the same process as in Section 3.3.2, but with the 2050 targets from the WFRC model. The “target work-from-home percent” value in ActivitySim’s work-from-home submodel was changed to 3.5% based on a weighted average from the 2050 WFRC data, and the job type coefficients in the telecommute frequency submodel were calibrated to match the WFRC target telecommute shares by job type. Table 6.2 shows the WFRC 2050 telecommute percentages with the ActivitySim telecommute utility coefficients. As in the baseline scenario, this calibration allowed ActivitySim to match the WFRC telecommute percentages exactly.
+We adjusted the remote work models in ActivitySim using the same process as in ?sec-baseline-calibration, but with the 2050 targets from the WFRC model. The “target work-from-home percent” value in ActivitySim’s work-from-home submodel was changed to 3.5% based on a weighted average from the 2050 WFRC data, and the job type coefficients in the telecommute frequency submodel were calibrated to match the WFRC target telecommute shares by job type.
+Table 5.2 shows the WFRC 2050 telecommute percentages with the ActivitySim telecommute utility coefficients. As in the baseline scenario, this calibration allowed ActivitySim to match the WFRC telecommute percentages exactly.
name | pct | 1 day | 2–3 days | 4 days |
|---|---|---|---|---|
Retail | 0.0725 | 2.021 | 0.809 | 0.505 |
Food | 0.0503 | 1.376 | 0.551 | 0.344 |
Manufacturing | 0.0545 | 1.636 | 0.655 | 0.408 |
Office | 0.1801 | 4.792 | 1.916 | 1.197 |
Gov't/Education | 0.0456 | 1.199 | 0.480 | 0.301 |
Health | 0.0721 | 1.929 | 0.771 | 0.482 |
Agriculture | 0.1683 | 4.764 | 1.906 | 1.191 |
Mining | 0.0143 | -0.694 | -0.277 | -0.174 |
Construction | 0.0882 | 2.544 | 1.018 | 0.637 |
Other | 0.1458 | 3.804 | 1.521 | 0.951 |
name | pct | 1 day | 2–3 days | 4 days |
|---|---|---|---|---|
Retail | 0.0725 | 2.021 | 0.809 | 0.505 |
Food | 0.0503 | 1.376 | 0.551 | 0.344 |
Manufacturing | 0.0545 | 1.636 | 0.655 | 0.408 |
Office | 0.1801 | 4.792 | 1.916 | 1.197 |
Gov't/Education | 0.0456 | 1.199 | 0.480 | 0.301 |
Health | 0.0721 | 1.929 | 0.771 | 0.482 |
Agriculture | 0.1683 | 4.764 | 1.906 | 1.191 |
Mining | 0.0143 | -0.694 | -0.277 | -0.174 |
Construction | 0.0882 | 2.544 | 1.018 | 0.637 |
Other | 0.1458 | 3.804 | 1.521 | 0.951 |
Both models decrease the number of work trips made as remote work rates increase. However, the WFRC model does not account for a potential “rebound effect” where more discretionary trips are made by those who do not travel to their workplace on a given day. This is seen in Table 6.3, where the WFRC model shows a decrease in home-based work and non–home-based trips (many of which begin or end at work), but virtually no change in home-based other trips. ActivitySim on the other hand does account for this, in that individuals working remotely on any given day may be more likely to make discretionary tours. Table 6.3 shows this as well, where ActivitySim predicts a noticeable increase in home-based other trips as well as a decrease in work trips.
+Both models decrease the number of work trips made as remote work rates increase. However, the WFRC model does not account for a potential “rebound effect” where more discretionary trips are made by those who do not travel to their workplace on a given day. This is seen in Table 5.3, where the WFRC model shows a decrease in home-based work and non–home-based trips (many of which begin or end at work), but virtually no change in home-based other trips. ActivitySim on the other hand does account for this, in that individuals working remotely on any given day may be more likely to make discretionary tours. Table 5.3 shows this as well, where ActivitySim predicts a noticeable increase in home-based other trips as well as a decrease in work trips.
purpose | mode | cube_wfh | cube_by | cube_diff_pct | asim_wfh | asim_by | asim_diff_pct |
|---|---|---|---|---|---|---|---|
hbo | carpool | 2,702,624.95 | 2,702,272.42 | 0.0001304563 | 2,171,566 | 2,148,429 | 0.010769264 |
hbo | drive_alone | 1,395,195.81 | 1,394,415.10 | 0.0005598856 | 709,957 | 700,133 | 0.014031620 |
hbo | nonmotor | 508,869.39 | 510,143.12 | -0.0024968107 | 617,480 | 613,134 | 0.007088173 |
hbo | transit | 37,358.88 | 37,346.03 | 0.0003439830 | 396,815 | 389,780 | 0.018048643 |
hbw | carpool | 238,668.79 | 257,805.01 | -0.0742275139 | 242,497 | 258,459 | -0.061758345 |
hbw | drive_alone | 1,244,450.85 | 1,328,609.31 | -0.0633432745 | 950,306 | 1,012,180 | -0.061129443 |
hbw | nonmotor | 71,062.67 | 76,506.40 | -0.0711539505 | 137,684 | 145,957 | -0.056681077 |
hbw | transit | 44,976.92 | 48,751.83 | -0.0774313205 | 237,881 | 253,176 | -0.060412519 |
nhb | carpool | 1,254,548.32 | 1,273,317.20 | -0.0147401446 | 922,662 | 938,056 | -0.016410534 |
nhb | drive_alone | 938,652.85 | 951,560.65 | -0.0135648736 | 687,935 | 716,143 | -0.039388781 |
nhb | nonmotor | 144,126.37 | 146,403.78 | -0.0155556895 | 152,688 | 156,819 | -0.026342471 |
nhb | transit | 13,198.72 | 13,453.28 | -0.0189220575 | 158,366 | 159,935 | -0.009810235 |
purpose | mode | cube_wfh | cube_by | cube_diff_pct | asim_wfh | asim_by | asim_diff_pct |
|---|---|---|---|---|---|---|---|
hbo | carpool | 2,702,624.95 | 2,702,272.42 | 0.0001304563 | 2,171,566 | 2,148,429 | 0.010769264 |
hbo | drive_alone | 1,395,195.81 | 1,394,415.10 | 0.0005598856 | 709,957 | 700,133 | 0.014031620 |
hbo | nonmotor | 508,869.39 | 510,143.12 | -0.0024968107 | 617,480 | 613,134 | 0.007088173 |
hbo | transit | 37,358.88 | 37,346.03 | 0.0003439830 | 396,815 | 389,780 | 0.018048643 |
hbw | carpool | 238,668.79 | 257,805.01 | -0.0742275139 | 242,497 | 258,459 | -0.061758345 |
hbw | drive_alone | 1,244,450.85 | 1,328,609.31 | -0.0633432745 | 950,306 | 1,012,180 | -0.061129443 |
hbw | nonmotor | 71,062.67 | 76,506.40 | -0.0711539505 | 137,684 | 145,957 | -0.056681077 |
hbw | transit | 44,976.92 | 48,751.83 | -0.0774313205 | 237,881 | 253,176 | -0.060412519 |
nhb | carpool | 1,254,548.32 | 1,273,317.20 | -0.0147401446 | 922,662 | 938,056 | -0.016410534 |
nhb | drive_alone | 938,652.85 | 951,560.65 | -0.0135648736 | 687,935 | 716,143 | -0.039388781 |
nhb | nonmotor | 144,126.37 | 146,403.78 | -0.0155556895 | 152,688 | 156,819 | -0.026342471 |
nhb | transit | 13,198.72 | 13,453.28 | -0.0189220575 | 158,366 | 159,935 | -0.009810235 |
In addition to the number of trips, increasing remote work rates can also have an effect on the length of trips that are made. The “travel time budget” proposed by Moreno and Moeckel (2017) suggests that longer trips would be made less frequently, and Moeckel (2017) additionally found that those who travel to their job site less frequently are more likely to live further away from their job site, and so their longer but infrequent commute is dropped on remote work days, perhaps in favor of shorter, discretionary trips.
-The WFRC model does not consider trip length when adjusting trip rates due to remote work. There is perhaps an implicit consideration in that remote work rates differ by job type and some job types are concentrated in certain areas, but there is no reference to trip length explicitly. Table 6.4 illustrates this, where for example home-based work driving trips decreased by 6.5% relative to the baseline scenario, but person-miles traveled decreased only by 5.8%. This shows that in fact the shorter work trips are being made less frequently with increased remote work rates, though notably this is only a side-effect of the WFRC model and the two specific model scenarios.
+The WFRC model does not consider trip length when adjusting trip rates due to remote work. There is perhaps an implicit consideration in that remote work rates differ by job type and some job types are concentrated in certain areas, but there is no reference to trip length explicitly. Table 5.4 illustrates this, where for example home-based work driving trips decreased by 6.5% relative to the baseline scenario, but person-miles traveled decreased only by 5.8%. This shows that in fact the shorter work trips are being made less frequently with increased remote work rates, though notably this is only a side-effect of the WFRC model and the two specific model scenarios.
purpose | mode | wfh_trips | by_trips | trips_pct | wfh_pmt | by_pmt | pmt_pct |
|---|---|---|---|---|---|---|---|
hbo | carpool | 2,702,624.95 | 2,702,272.42 | 0.0001304563 | 13,448,783.75 | 13,420,596.31 | 0.0021003115 |
hbo | drive_alone | 1,395,195.81 | 1,394,415.10 | 0.0005598856 | 6,122,516.59 | 6,088,804.31 | 0.0055367659 |
hbo | nonmotor | 508,869.39 | 510,143.12 | -0.0024968107 | 590,348.63 | 591,297.33 | -0.0016044324 |
hbo | transit | 37,358.88 | 37,346.03 | 0.0003439830 | 264,432.21 | 264,203.19 | 0.0008668473 |
hbw | carpool | 238,668.79 | 257,805.01 | -0.0742275139 | 2,945,150.50 | 3,204,552.45 | -0.0809479501 |
hbw | drive_alone | 1,244,450.85 | 1,328,609.31 | -0.0633432745 | 12,070,213.27 | 12,736,969.62 | -0.0523481151 |
hbw | nonmotor | 71,062.67 | 76,506.40 | -0.0711539505 | 122,930.22 | 132,215.78 | -0.0702303750 |
hbw | transit | 44,976.92 | 48,751.83 | -0.0774313205 | 500,952.87 | 547,803.97 | -0.0855253032 |
nhb | carpool | 1,254,548.32 | 1,273,317.20 | -0.0147401446 | 7,538,595.77 | 7,650,624.71 | -0.0146431105 |
nhb | drive_alone | 938,652.85 | 951,560.65 | -0.0135648736 | 4,736,978.68 | 4,777,297.29 | -0.0084396287 |
nhb | nonmotor | 144,126.37 | 146,403.78 | -0.0155556895 | 134,783.78 | 136,913.82 | -0.0155575308 |
nhb | transit | 13,198.72 | 13,453.28 | -0.0189220575 | 72,017.93 | 73,563.35 | -0.0210079893 |
purpose | mode | wfh_trips | by_trips | trips_pct | wfh_pmt | by_pmt | pmt_pct |
|---|---|---|---|---|---|---|---|
hbo | carpool | 2,702,624.95 | 2,702,272.42 | 0.0001304563 | 13,448,783.75 | 13,420,596.31 | 0.0021003115 |
hbo | drive_alone | 1,395,195.81 | 1,394,415.10 | 0.0005598856 | 6,122,516.59 | 6,088,804.31 | 0.0055367659 |
hbo | nonmotor | 508,869.39 | 510,143.12 | -0.0024968107 | 590,348.63 | 591,297.33 | -0.0016044324 |
hbo | transit | 37,358.88 | 37,346.03 | 0.0003439830 | 264,432.21 | 264,203.19 | 0.0008668473 |
hbw | carpool | 238,668.79 | 257,805.01 | -0.0742275139 | 2,945,150.50 | 3,204,552.45 | -0.0809479501 |
hbw | drive_alone | 1,244,450.85 | 1,328,609.31 | -0.0633432745 | 12,070,213.27 | 12,736,969.62 | -0.0523481151 |
hbw | nonmotor | 71,062.67 | 76,506.40 | -0.0711539505 | 122,930.22 | 132,215.78 | -0.0702303750 |
hbw | transit | 44,976.92 | 48,751.83 | -0.0774313205 | 500,952.87 | 547,803.97 | -0.0855253032 |
nhb | carpool | 1,254,548.32 | 1,273,317.20 | -0.0147401446 | 7,538,595.77 | 7,650,624.71 | -0.0146431105 |
nhb | drive_alone | 938,652.85 | 951,560.65 | -0.0135648736 | 4,736,978.68 | 4,777,297.29 | -0.0084396287 |
nhb | nonmotor | 144,126.37 | 146,403.78 | -0.0155556895 | 134,783.78 | 136,913.82 | -0.0155575308 |
nhb | transit | 13,198.72 | 13,453.28 | -0.0189220575 | 72,017.93 | 73,563.35 | -0.0210079893 |
ActivitySim does model distance to work directly when predicting remote work status (see Section 3.3.2 and Table 3.5), so those who live further away from their job site are more likely to work remotely. ActivitySim therefore predicts a greater decrease in miles traveled than in trips for home-based work, as seen in Table 6.5.
+ActivitySim does model distance to work directly when predicting remote work status (see ?sec-baseline-calibration and ?tbl-asim-tc-model-coeffs), so those who live further away from their job site are more likely to work remotely. ActivitySim therefore predicts a greater decrease in miles traveled than in trips for home-based work, as seen in Table 5.5.
purpose | mode | wfh_trips | by_trips | trips_pct | wfh_pmt | by_pmt | pmt_pct |
|---|---|---|---|---|---|---|---|
hbw | drive_alone | 950,306 | 1,012,180 | -0.061129443 | 9,021,681.0 | 9,632,251.2 | -0.063388118 |
hbw | carpool | 242,497 | 258,459 | -0.061758345 | 2,463,551.7 | 2,631,886.1 | -0.063959629 |
hbw | transit | 237,881 | 253,176 | -0.060412519 | 2,728,897.4 | 2,911,615.6 | -0.062754932 |
hbw | nonmotor | 137,684 | 145,957 | -0.056681077 | 332,978.2 | 353,246.2 | -0.057376492 |
hbo | drive_alone | 709,957 | 700,133 | 0.014031620 | 4,332,319.3 | 4,280,005.8 | 0.012222761 |
hbo | carpool | 2,171,566 | 2,148,429 | 0.010769264 | 11,624,928.0 | 11,498,993.6 | 0.010951780 |
hbo | transit | 396,815 | 389,780 | 0.018048643 | 3,583,629.9 | 3,547,052.4 | 0.010312089 |
hbo | nonmotor | 617,480 | 613,134 | 0.007088173 | 1,098,043.0 | 1,090,175.6 | 0.007216663 |
nhb | drive_alone | 687,935 | 716,143 | -0.039388781 | 3,804,674.1 | 3,984,191.3 | -0.045057358 |
nhb | carpool | 922,662 | 938,056 | -0.016410534 | 3,898,220.0 | 3,962,840.2 | -0.016306519 |
nhb | transit | 158,366 | 159,935 | -0.009810235 | 852,242.6 | 867,866.9 | -0.018003094 |
nhb | nonmotor | 152,688 | 156,819 | -0.026342471 | 189,483.4 | 194,493.3 | -0.025758980 |
purpose | mode | wfh_trips | by_trips | trips_pct | wfh_pmt | by_pmt | pmt_pct |
|---|---|---|---|---|---|---|---|
hbw | drive_alone | 950,306 | 1,012,180 | -0.061129443 | 9,021,681.0 | 9,632,251.2 | -0.063388118 |
hbw | carpool | 242,497 | 258,459 | -0.061758345 | 2,463,551.7 | 2,631,886.1 | -0.063959629 |
hbw | transit | 237,881 | 253,176 | -0.060412519 | 2,728,897.4 | 2,911,615.6 | -0.062754932 |
hbw | nonmotor | 137,684 | 145,957 | -0.056681077 | 332,978.2 | 353,246.2 | -0.057376492 |
hbo | drive_alone | 709,957 | 700,133 | 0.014031620 | 4,332,319.3 | 4,280,005.8 | 0.012222761 |
hbo | carpool | 2,171,566 | 2,148,429 | 0.010769264 | 11,624,928.0 | 11,498,993.6 | 0.010951780 |
hbo | transit | 396,815 | 389,780 | 0.018048643 | 3,583,629.9 | 3,547,052.4 | 0.010312089 |
hbo | nonmotor | 617,480 | 613,134 | 0.007088173 | 1,098,043.0 | 1,090,175.6 | 0.007216663 |
nhb | drive_alone | 687,935 | 716,143 | -0.039388781 | 3,804,674.1 | 3,984,191.3 | -0.045057358 |
nhb | carpool | 922,662 | 938,056 | -0.016410534 | 3,898,220.0 | 3,962,840.2 | -0.016306519 |
nhb | transit | 158,366 | 159,935 | -0.009810235 | 852,242.6 | 867,866.9 | -0.018003094 |
nhb | nonmotor | 152,688 | 156,819 | -0.026342471 | 189,483.4 | 194,493.3 | -0.025758980 |
Figures 6.2 and 6.3 show the trip length frequency distribution of “unmade” trips in the increased remote work scenario (i.e. the trip length frequency distribution of the difference in trips) compared to that of the baseline scenario. Similar to Tables 6.4 and 6.5, this shows that ActivitySim “removes” longer trips more frequently than shorter trips, and the WFRC model makes no distinction.
+Figures 5.2 and 5.3 show the trip length frequency distribution of “unmade” trips in the increased remote work scenario (i.e. the trip length frequency distribution of the difference in trips) compared to that of the baseline scenario. Similar to Tables 5.4 and 5.5, this shows that ActivitySim “removes” longer trips more frequently than shorter trips, and the WFRC model makes no distinction.
mpg | cyl | disp | hp | drat | wt | qsec | vs | am | gear | carb |
|---|---|---|---|---|---|---|---|---|---|---|
21.0 | 6 | 160 | 110 | 3.90 | 2.620 | 16.46 | 0 | 1 | 4 | 4 |
21.0 | 6 | 160 | 110 | 3.90 | 2.875 | 17.02 | 0 | 1 | 4 | 4 |
22.8 | 4 | 108 | 93 | 3.85 | 2.320 | 18.61 | 1 | 1 | 4 | 1 |
21.4 | 6 | 258 | 110 | 3.08 | 3.215 | 19.44 | 1 | 0 | 3 | 1 |
18.7 | 8 | 360 | 175 | 3.15 | 3.440 | 17.02 | 0 | 0 | 3 | 2 |
18.1 | 6 | 225 | 105 | 2.76 | 3.460 | 20.22 | 1 | 0 | 3 | 1 |
mpg | cyl | disp | hp | drat | wt | qsec | vs | am | gear | carb |
|---|---|---|---|---|---|---|---|---|---|---|
21.0 | 6 | 160 | 110 | 3.90 | 2.620 | 16.46 | 0 | 1 | 4 | 4 |
21.0 | 6 | 160 | 110 | 3.90 | 2.875 | 17.02 | 0 | 1 | 4 | 4 |
22.8 | 4 | 108 | 93 | 3.85 | 2.320 | 18.61 | 1 | 1 | 4 | 1 |
21.4 | 6 | 258 | 110 | 3.08 | 3.215 | 19.44 | 1 | 0 | 3 | 1 |
18.7 | 8 | 360 | 175 | 3.15 | 3.440 | 17.02 | 0 | 0 | 3 | 2 |
18.1 | 6 | 225 | 105 | 2.76 | 3.460 | 20.22 | 1 | 0 | 3 | 1 |
task | hours |
|---|---|
Synthetic population creation | 50 |
Baseline mode choice calibration | 20 |
Add remote work models to ActivitySim | 20 |
Baseline remote work calibration | 10 |
Scenario creation: Land Use | 15 |
Scenario creation: Transit | 2 |
Scenario creation: Remote Work | 5 |
task | hours |
|---|---|
Synthetic population creation | 50 |
Baseline mode choice calibration | 20 |
Add remote work models to ActivitySim | 20 |
Baseline remote work calibration | 10 |
Scenario creation: Land Use | 15 |
Scenario creation: Transit | 2 |
Scenario creation: Remote Work | 5 |
Each model works differently. - There will be a learning curve when changing from TBM to ABM - Planning agencies should take this into account when switching
ABM is more malleable. It was easy to throw on the WFH model without having to completely change the rest of the model, and it seemed more realistic than the TBM. - For planning agencies wanting a model that more easily adapts to unforeseen travel behavior change, an ABM would be preferable
More analyses can be done with ABMs - We were able to replicate each TBM analysis with the ABM and more - We could make more demographic-type analyses with ABM - We were able to compare changes in individual behavior with ABM - Simpler to go about the analysis when thinking on the individual level
@@ -662,7 +656,7 @@Activity-based models (ABMs) have been championed by researchers and many practitioners as being theoretically superior to the trip-based models historically used in transportation planning efforts since the 1950s.
Despite the theoretical benefits, many agencies have delayed or declined to transition to an ABM citing additional data requirements, staff training, computational resources, and related concerns. There is also not a consensus that ABMs result in better infrastructure volume forecasts, the primary—or even sole—purpose of many regions’ travel demand analysis efforts.
In this research, we investigate the quality and characteristics of travel analyses enabled by an ABM. We do this by applying an array of infrastructure and behavior scenarios to both an ABM and a trip-based model representing the Wasatch Front (Salt Lake City) region of Utah, USA. The results will compare the implications of each model and help inform agencies that are considering adopting an ABM.
-The document proceeds in a typical fashion: Chapter 2 provides a discussion of the differences between trip-based models and ABMs, alongside a discussion of previous studies examining the theoretical and analytical benefits of each framework. Chapter 3 first describes the model frameworks used in this research, namely the regional trip-based model and an activity-based model constructed to support research activities in the region; this section also describes three scenarios designed to test the usefulness and applicability of the different model frameworks. Chapters 4–6 describe the findings from each scenario, alongside a discussion of their limitations and implications. Chapter 7 provides a series of recommendations and opportunities for future research.
+The document proceeds in a typical fashion: Chapter 2 provides a discussion of the differences between trip-based models and ABMs, alongside a discussion of previous studies examining the theoretical and analytical benefits of each framework. ?sec-methods first describes the model frameworks used in this research, namely the regional trip-based model and an activity-based model constructed to support research activities in the region; this section also describes three scenarios designed to test the usefulness and applicability of the different model frameworks. Chapters 3–5 describe the findings from each scenario, alongside a discussion of their limitations and implications. Chapter 6 provides a series of recommendations and opportunities for future research.
diff --git a/qmd/lit-review.html b/qmd/lit-review.html index a455757..408fd40 100644 --- a/qmd/lit-review.html +++ b/qmd/lit-review.html @@ -50,7 +50,7 @@ - + @@ -143,35 +143,29 @@ -This approach is relatively straightforward: the required input data is usually easy to obtain, the trip generation models are often simple, and it is computationally inexpensive (National Academies 2012). However, the types of analyses possible are limited by the initial segmentation of the aggregate population data. An analysis based on parents’/adults’ highest received education, for example, would require determining the number of households in each TAZ with each possible combination of education level. This can theoretically be done, but more detailed and varied analyses would require more levels of segmentation, greatly increasing the number of classifications needed. Aggregation at any point precludes that segmentation from use in subsequent model steps as well as in any post-hoc analysis. Since these segmentations need to be carried through each model step, trip rates, mode choice equations, etc. need to be estimated for every classification, and while relevant real-world data may exist, sample sizes approach zero very quickly, and so the estimates have little statistical value (Moeckel et al. 2020; National Academies 2012).
This becomes a particular issue in equity analysis because it is perhaps impossible to determine equitable distribution of “winners” and “losers” of a potential policy without using demographic variables in the trip generation and destination and mode choice steps (Bills and Walker 2017). Though many studies have shown that trip production and mode choice behavior differ by ethnic group even after controlling for income Bhat and Naumann (2013), including such variables in trip-based models is problematic. Does coding such a variable in a mode choice model represent discrimination? Or does doing so assert that present differences resulting from unequal opportunity will persist into the future planning years? Regardless the reasons for their exclusion, these variables consequently cannot be used in a post-hoc analysis of a transportation policy because the trip matrices do not contain the adequate segmentation.
An alternative approach to population data is to use a full synthetic population. A synthetic population takes demographic and socioeconomic data at various levels of detail to create a “population” with generally the same distribution as the study area (National Academies 2012, 93). The goal is to have a population that is functionally similar to the actual population, but without the privacy concerns of using real individual household data. Castiglione et al. (2006) argue that the major advantage with this approach is that the demographic and socioeconomic data is known at the person and household level, rather than the zone level, and this data remains available throughout the modeling process. This allows, for example, an equity analysis to determine the “winners” and “losers” of a proposed development without needing to encode demographic variables into each step of the model.
-Bills and Walker (2017) used the 2000 Bay Area Travel Survey to create a synthetic population and compare the effects that certain scenarios had on high income and low income populations. With a 20% reduction in travel cost, they found that high income workers benefited more than low income workers. They did similar comparisons for scenarios involving reduced travel times for different mode choices and saw the effects each scenario had on the high and low income workers. These types of analysis, which are not possible with aggregate population data (this is actually possible; a lot of trip-based models segment high/low income), can be very valuable in transportation planning and policy making.
+Bills and Walker (2017) used the 2000 Bay Area Travel Survey to create a synthetic population and compare the effects that certain scenarios had on high income and low income populations. With a 20% reduction in travel cost, they found that high income workers benefited more than low income workers. They did similar comparisons for scenarios involving reduced travel times for different mode choices and saw the effects each scenario had on the high and low income workers. These types of analysis, which are difficult with aggregate population data, can be very valuable in transportation planning and policy making.
It is important to note that while many connect them only with ABMs, synthetic populations can be used in running trip-based models as well. Trip-based models using a synthetic population—often called trip-based microsimulation models—do exist (see Walker (2005) and Moeckel et al. (2020)), but these are relatively rare.
-Figure 2.1 gives a visualization of an example “information pipeline” for a trip-based model using aggregate data and an ABM using a synthetic population. In the aggregate data model, it is impossible to know which trips are made by e.g. 2-worker, 1-vehicle, low-income households; it only describes which trips are made by households with fewer vehicles than workers. With a synthetic population, however, individuals are being modeled, and so each trip can be traced to a specific person. All information is known at each point in the model regardless of which data is used in previous steps.
+Figure 2.1 gives a visualization of an example “information pipeline” for a model using aggregate data and a model using a synthetic population. In the aggregate data model, it is impossible to know which trips are made by e.g. 2-worker, 1-vehicle, low-income households; it only describes which trips are made by households with fewer vehicles than workers. With a synthetic population, however, individuals are being modeled, and so each trip can be traced to a specific person. All information is known at each point in the model regardless of which data is used in previous steps.
