Should three sets of transit paths (based on technology differentiation) be combined into just one?

[bhargavasana]

Currently in ABM3, (apart from access/egress modes) three sets of transit paths are identified - Local (only), Premium (only), and Mix (use both local & premium modes). Premium modes include all transit modes except local bus (Ex. coaster, trolley, express buses, etc.).

Should the three be combined into just one (would result in significant memory and storage space savings)? EMME extended transit assignment does consider multiple paths. Could that be optimized to represent mode competition appropriately in certain markets in San Diego?

[bhargavasana]

@JiaXu1024 performed an analysis of assigning 2023 transit on-board survey trips to ABM3 base year (2022) network. Assignment was performed for with three sets and one set (denoted by _mix suffix) using both EMME 4.3.7 and Open Paths (OP).

Route ridership comparison (model vs. OBS) based on R-squared shows that one set (_mix) assignment is not significantly worse than that based on three sets.

Boardings comparison by access mode and main transit mode.

[bhargavasana]

Observations by @jfdman

• Express bus and rapid bus are significantly over-assigned regardless of method of whether assignments are differentiated by technology.
• Perhaps the headway scaling is causing the service frequencies to look a lot better on these modes or the service frequency is inconsistent through the time period?
• Express bus is also much more significantly overassigned in PNR access\egress. Maybe reviewing PNR connector logic would be helpful.
• Looks like boardings per trip are too high in PNR and KNR assignments, may need to adjust the boarding penalty more across to reduce the over-assignment for these access\egress modes.
• There is a huge improvement in commuter rail boardings when we don’t differentiate by transit technology.
• The other differences are slight, though local bus does get worse when trips are not differentiated. Probably because express bus and rapid get even more overassigned when they are lumped in with all the other local bus only trips in assignment. This will need to be addressed by adjustment of path weights if we decide to collapse the local-only, premium-only, mixed differentiation.
• Don’t see a reason to keep the differentiation, based on these results.

[zouyang401]

Would we ever want to report transit travel time by OD by different modes (premium vs local)? or identify the transit line haul modes used by the path? If so, can the one transit path handle that?

[inrokevin]

There is definitely a trade-off here, but I think it is best to think in terms of the "levers" that can be pulled in the assignment to better calibrate the assignment results. The three transit sets is a pretty powerful lever which can help match the transit results better at an O-D -> mode basis but also adds a lot of complexity and is not the only lever which can be pulled.

It is certainly worth exploring a single assignment specification with different perception factors. To do so efficiently you will want to consider some semi-automated parameter adjustment. In practice the boardings by mode or even by line can be exactly recreated (which would be overfitting) but the focus should be on getting the distribution of O-D -> mode or line/ major corridor within reason such that the assignment is useful for predicting results in alternative scenarios. I'm sure others can add a lot to the detail but the overall idea is to categorize zones or O-D pairs and compare the aggregate behaviour. The O-D -> mode relationship will not be as closely followed as in the three transit sets case.

Perception factors to consider:

• Walk time perception: can differentiate by geography - generally the walk time perception should be a little lower than observed due to the spatial aggregation effect of TAZs. Alternatively (or in addition) the centroid connector walk times / speeds can be adjusted to better reflect real world walk time distributions.
• Stop boarding perception: to account for both perceived reliability and preference of stops, e.g. rail stations may be more attractive for a number of reasons. This can also vary by mode/line as one way of calibrating boardings where the choice does not appear to be dependent on trip length, and can vary by number / sequence of boardings to better match transfer behaviour
• In-vehicle perception: relative attractiveness of different modes where the perception relates to the length of the trip. E.g. for a long trip travellers walk past the bus stop to get the train but not for a short trip
• Wait time perception: can vary by stop / line as well - this is most commonly used to account for schedule perception - use a reduced factor for services which are reliable and travellers generally know the schedules. This can vary by sequence of boardings as well to differentiate transfers.

In addition to these the "flow distribution" options can greatly increase the path variety and better match real world choices.

• Flow distribution at origins: distribute the flow between several connectors using a logit distribution, highly recommend testing this out to better distribute demand where there are multiple connectors close to different stops. Generally pretty easy to understand and apply.
• Flow distribution at regular nodes with auxiliary transit choices: this will greatly increase the path variety and is worth exploring but is also increases the complexity and can make it harder to debug results. It is appropriate where there are multiple nearby stops where travellers may walk between stops, or different travellers will make different choices.
• Flow distribution between lines: more demand will be assigned to faster, less frequent services where they share the same stop as less frequent services. A pretty easy option to turn on and can help reflect preference for express services. I like it as it helps with more proportional sensitivity to small improvements in travel time.

[DavidOry]

Per your email @bhargavasana, I think this change should have a minor effect on the mode choice calibration, given the modest role transit plays for the air passenger market. Regardless of the approach, the key transit movement downtown to the airport on the 992. I suggest we do an initial calibration with the skims you have now and then iterate once more after the new approach is settled. It should only take one or two mode choice runs to tighten things up.