The TTC is an interesting transit system in this regard: as it is laid out in a grid, most trips involve some amount of travel diagonal to the grid and thus present the traveler with a choice of which route to take. In a Manhattan-distance metric space, there is often not one shortest path but many.
Sometimes the decision is easy – perhaps it’s a lovely day and one prefers the views from the streetcar to the darkness of the subway. Perhaps one prefers variety and takes the path not recently taken, or familiarity with the opposite consequence.
Generally, a transit user will prefer to minimize the number of transfers needed in order to save time and effort, but this constraint still leaves two distinct options in Manhattan space. Should one go over and then up? Or up and then over? I’ve struggled with this question myself since moving to my new apartment, at a diagonal from my office.
When taking transit, I have three distinct ways of getting to work:
I mostly use the third option when I have some shopping to do in Chinatown or the fabric district, both of which it passes through. When I’m just trying to get to the office though, the first two options feel like a genuine toss-up.
I just want to get there quickly and I haven’t yet figured out which is the better strategy. It seems like at any given moment, the choice totally depends on where the next north-bound #47 bus is and what traffic is like for the #506.
I gave up my cellular data plan last year when it got a bit more expensive, so any knowledge I might have – about the positions of vehicles or traffic conditions – knowledge that might come through the Internet, can’t help me once I walk out the door and away from my WiFi. Real-time predictions are of little value here.
What I do have is a lot of archived AVL data. Using some code I wrote last year, I can convert this AVL data, which is essentially a bunch of timestamped vehicle GPS positions, into a retrospective GTFS file where trips and stop-arrival times correspond to actual observed vehicle trajectories. This lets me answer the question in an interesting way – by using a routing engine like OpenTripPlanner to plan a trip (in the past) on this retrospective (literally backward looking) schedule, I can find what would have actually been the fastest way to get to the office at any given time.
So what’s the answer?
According to the standard GTFS schedule data: the 47-2 option is fastest 49% of the time (during daytime service), followed by the 506-only with 19%, and 16% of the time the best option is a quick three stop trip on the 47 followed by a transfer to the 506.
According too the retrospective data however, the 47;506 combo is actually the best, but only 37% of the time, followed by 47;2 at 33% and 506 at 17%.
What’s a boy to do? A 37%-33% split isn’t great odds in favour of either alternative. The retrospective data seems to be telling me I ought to wait for the north-bound #47 on most days, but where should I get off?
Trivial as this may all seem, I think this is an interesting problem, because making the wrong choice, or rather, perceiving that you’ve done so can be quite a frustrating experience. Sometimes I take the 506, only to spend an hour standing in a crammed streetcar and simply hating that I didn’t rather walk up to the subway. Sometimes I take the subway and spend twenty minutes in the cold wondering if I should wait for the next #47 bus or spend twenty minutes walking home from the station. There’s a real cognitive stress that comes with too much choice, and it would be interesting, I think, to see who else in Toronto may be suffering daily under this terrible, terrible burden.
Jarrett Walker, of Human Transit fame, has often advocated that transit agencies restructure their systems into high-frequency grids and seems to have actually led many transit agencies in doing just that, or at least moving in that direction. Living with just such a system in Toronto however, I think it’s interesting to examine some of the side effects of such a network structure that most people have not yet had the pleasure of experiencing directly.
Well anyway, this is the topic that I plan to explore for my next dissertation paper, so keep an eye out for more on this. By the way, the previous paper, a look at the accuracy of schedule-based GTFS data for computing time-based accessibility metrics, is currently under review, but I jumped the gun a bit and posted a pre-print version over at SocArXiv. I’ll also be presenting that research at a poster-session in the upcoming 2019 TRB annual meeting.