A frequent commenter on HT asks this in an email (the links are mine, not his):
On Second Avenue Sagas, one of the discussions went on a tangent that left me wondering about transfer penalties. If you need to walk from one station to another on the street to transfer, do the ridership models assign a higher penalty than if there’s an enclosed corridor between the stations? In addition, for systems that have faregates, is there an extra penalty for transfers that require exiting and
Modeling can be a murky business, and I’m definitely not a modeler. But the bottom line is that the factors used in a ridership model have to be based on some sort of reality. The reality of North America and Australasia is that most connections are not especially pleasant or easy, especially when there are fare penalties or delay-inducing features such as fare gates. Ease of connection varies a lot from one situation to another based on factors that are hard to model — not just walking distances but things like street crossings, vertical circulation systems (stairs or escalator? elevator?), and of course fare coordination. Improvements in other areas can make a big difference. Smartcards, for example, are faster and easier to use than magnetic strip tickets, which must be manually inserted in a slot; these should make connecting easier.
So my guess is that it’s hard to model transfer penalties very exactly. But let me pass on the question to all the anonymous modeling professionals that are reading. What’s your answer to this reader’s question? If you’re working in North America, do your transfer penalties consider very high-quality transfers, such as the cross-platform connections that often occur in European subways? BART in the San Francisco Bay Area may know something about this, since they do have a cross-platform connection in Oakland, though it’s only used for low-demand trip pairs and so may not offer sufficient robust information. Agencies that do a lot of timed-transfer among buses, especially at island stations where no street crossings are required, may also know something about this.
Modelers! If you’re not comfortable posting a comment here, send me an email. I’ll use it anonymously.
I’m not exactly sure what you’re getting at (it seems a bit over my head) but I always enjoyed the way transfers worked in Portland… again, this may be somewhat obvious to you and many readers… but paying a single fare and being able to transfer between modes for up to 2 hours without penalty was eye-opening. I wish more transit services operated this way. It was so convenient and seamless.
I’m no expert, but IMHO the number 1 disincentive to transferring is wait time, which is directly related to frequency (and congestion). Everything else comes a distant second.
Next is probably the NUMBER of transfers required – not only do two transfers double the penalty but they also increase the odds of suffering from the other negative consequences (such as long wait times).
After that, physical comfort seems to me like the next biggest factor – that includes distance/time to walk, crowds, exposure to the elements, etc. Stairs are a minus, but not escalators, IMHO. On the other hand elevators are a minus because they mean more wait time.
Exposure to the elements is a particular disincentive if the city is prone to inclement weather and the waiting itself has to be done in it.
Other penalties are perceived safety of transfer points, the fact that transfers often result in a less direct trip (which means more wasted time for the customer) and the hidden cost of complexity in any trip that involves someone having to figure out schedules and routing for multiple services.
Plea: in discussing this issue, can we please distinguish –
1. actual interchange walk/wait time
2. weighting of interchange walk/wait time to reflect the fact that it is less comfortable than in-vehicle time, so that it can be considered in common units with in-vehicle time;
3. any additional ‘flagfall’ interchange penalty which we may wish to add to reflect the instrinc dislike of interchange.
Transport planning documents often do not make these distinctions clearly.
The MTA’s ridership model is blind to comfort issues, and models the penalty by considering every minute spent transferring or waiting to count 1.75 times as much as a minute spent on a train. However, the MTA has made statements contradicting this simple scheme – namely, that the waiting penalty is reduced when there are countdown clocks.
*puts modelling hat on*.
In my expereicne, walking transfers are rarely modelled differently from sitting-and-waiting transfers, so that walking 10 minutes to make a connection is treated the same as waiting 10 minutes. Now, I’d certainly agree they are not perceived the same, and so shoudl atrtract a differet penalty, but I’ve never found any research on how to penalise walking transfer times. (Or whether you should).
The best option to me woudl be to model the walking as a seperate leg (with zero wait time), then model the wait at the end of the walk for the next leg as a standard sitting-and-waiting wait. That way bus-walk-bus gets modelled as three legs, and bus-sit-bus gets modelled as two legs. (With a penalty per leg).
Toronto’s TTC has published documents showing the various penalties assumed in its modeling. They include a 10-minute penalty for a transfer. It’s not clear from the writeup whether it’s intended to account for waiting, walking, etc.
I agree that it would be more meaningful to measure transfer ease and efficiency, particularly in a network such as Toronto’s where passengers are accustomed to transferring (due to the grid network design, or to transfer from bus to subway) — as opposed to, say, Ottawa, where there is an expectation that all routes will join the Transitway and proceed to downtown without requiring a transfer. I would think nothing of transferring from a north-south 20-minute-headway route to an east-west 4-minute-headway route, but might consider the reverse more carefully.
I would think nothing of transferring from a north-south 20-minute-headway route to an east-west 4-minute-headway route, but might consider the reverse more carefully.
And therein lies the difficulty. Whatever route you take to work in the morning, you’ll take the reverse route in the evening. For that reason, a transfer between a fast service and a slow service isn’t all that much better than a transfer between slow services.
I don’t think it’s generally considered. Models are already complicated enough. I raised this very issue with a project in my city where a rail to rail “transfer” was 650 feet horizontal distance, not including an elevation change requiring stairs and escalators.
I think indirect transfer matter a lot and they are a detriment to rider attraction. They add too much uncertainty to a potential trip.
Prof. Nigel Wilson of MIT and his students have done “revealed preference” studies of transfers on the Boston MTA and other rail systems.
As I recall, in a high quality transfer environment, the transfer time was equivalent to the same number of minutes of walking time in terms of route choice, but in other transfer environments, passengers would rather spend more time walking to or from a subway stop if it would eliminate a transfer, even at the cost of a longer total trip time.
Perhaps someone more familiar with his work could add details, such as whether his research has been able to compare the value of transfer/walking time to on-vehicle time.
I admit to some skepticism regarding how models deal with transfers. Are the penalities derived from “stated preference” surveys? Do the penalties represent proxies for other attributes such as reliability, frequency, or mode bias that perhaps should be in the model, but aren’t?
I accept that models can be calibrated to give reasonable predictions for near-term system changes, but can they deal with predicting ridership in a hypothetical system that is designed to carry two or three times what an existing system is carrying?
I’ve got to disagree about the “low-demand trip pairs” in Oakland — when a Pittsburg train arrives at MacArthur in the evening, nearly half the passengers get off to switch to the Richmond train to go to Berkeley. They can get away with this only because it is a very well-timed transfer.
Eric. I agree re BART transfers, but I was referring to perfect
cross-platform or same-platform transfer situations, and the one you
mention involves a change of level. I was thinking about
Fremont-Pittsburg, Dublin-Pittsburg, and at night only, Richmond-SF.
No, there is no change of level in this transfer. The Richmond and Pittsburg trains pull up side by side on the same platform, and half of the passengers get off the train and walk over to the other side.
(It is if you are coming *from* Berkeley and going to Pittsburg, or vice versa, a much less common pattern, that you have to go downstairs and back up again.)
Eric. I think we are saying the same thing. The trains TO Richmond and Pittsburg are side to side. These are also known as the trains FROM Fremont and TO Pittsburg. The same cross-platform connection is used for SF to Richmond, but only at off-hours when there are no direct trains. Right?
EngineerScotty: Not always. I take one route (5 + 545) to work and a different route (545 + 43/44) home. If I took the 44 in the morning, I would have two transfers, and one of them is at a very ugly and depressing highway stop. Thus, I gladly add 15 minutes to my commute to skip a transfer and wait at a Starbucks.
Given the BART is clearly capable from the discussion here of good timed transfers, I wonder about the possibility of turning BART from a direct service network into a very high frequency transfer network.
Line 1 – Millbrae-SFO-Richmond (alternatively Pittsburg)
Line 2 – Daly City – Fremont (Alternatively Dublin)
Line 3 – Pittsburg – Dublin (Alternatively Richmond-Fremont)
With timed transfers so that everyone can make their journey with <30 seconds transfer time.
OK, maybe we are saying the same thing. I don’t know any more. All I was trying to say was that transfers on BART are not just for low-demand trip pairs because of the large numbers of people who use it between San Francisco and Berkeley even when there is no direct service.
I think the penalties are definitely higher for transfers where you have to cross a street at grade. Why? This adds *unpredictability* to the transfer time, due to the light cycle. Unpredictability is bad for transit.
For me personally. Walking for 10 minutes or waiting for 10 minutes. So both should get an equal weight factor in the model.
As for routing of the transfer. ie whether the transfer takes you outside or takes you underground. Would depend a lot on what the climate is like. In the middle of winter most people would probably prefer staying underground, even if it did had a few minutes. While during the summer most might prefer the outside route.
The only month for good weather in my city is October, so it’s almost always better to be underground.
Clearly, cross-platform transfers should have the LEAST penalty. Not question the evoke the least stress and least uncertainty since one can see the dwell space for the next trip leg from the previous trip leg vehicle.
Zoltan: BART already runs the frequencies it needs to run. South of Daly City there’s almost no demand for service; that’s why two BART services out of four short-turn. Beyond Oakland, once the lines start splitting, BART turns into an expensive commuter rail system, so higher frequencies than 15 minutes are not necessary.
regarding the transfer issue & BART, one of the annoyances for those of us on the Richmond line on the ride into San Francisco is that often we have to stand for the 20-30 minutes from MacArthur into SF since the folks on the Pittsburg line have taken all the seats already.
Rumor is that this is because even though there are more riders on the Richmond line (which includes Berkeley), the Pittsburg line is wealthier and less transit-dependent.