Realtime Information: Facts or Predictions?

Among the cool things that Portland’s transit agency Tri-Met did during our record snowstorm is this:


Instead of telling us a prediction of when the bus will arrive, they told us the fact of how far away it is.  Predictions are not facts.

Free and conscious citizens should always value facts over predictions.  It doesn’t hurt to have both, but predictions without facts can be dangerous.  Humans always want more certainty about the future than the universe provides, so they tend to overvalue predictions, and even treat them as promises.

A rare citywide disruption puts all this in perspective, by highlighting something that is really always true.  The transit agency is in no position to promise when the bus will arrive.  Too many things out of their control might happen.  What they can do is tell you the facts and let you make your own judgment about what to do.

16 Responses to Realtime Information: Facts or Predictions?

  1. Jonathan Hallam January 18, 2017 at 12:01 pm #

    I’m not sure I agree. I mean, what you care about in this context is how long it takes to come. Offering the distance information means the rider is forced to make this estimate – but the transport agency is better equipped to do so. Maybe the time estimate should come with an error bar, e.g. 10minutes +/-3 minutes (70% chance), +/-6minutes (90% chance).

    • Runaurufu January 18, 2017 at 12:34 pm #

      In theory it sounds great to give time estimate…. but in real life it gets you pissed when you see that your tram will arrive in 5 min…. and you have that information for 3 min and then it suddenly changes to 8 minutes. The best choice is to provide both distance and time estimation… but in most cases there is not enough space to do so :/

    • Joseph Brant January 19, 2017 at 11:41 pm #

      Confidence intervals are a good idea, although perhaps not the easiest to understand.

  2. Mitch Walk January 18, 2017 at 1:04 pm #

    Facts are great, with context. Knowing a bus is 1 mile away from my house in London means one thing when it is traveling west, and something entirely different when it is traveling east. Likewise at rush hour or at 08:00 on a Sunday morning. A transit agency, which I hope is using literally hundreds or thousands of data points to run its prediction algorithms is far better placed than me to make the judgment about its time of arrival because the algorithm can contextualize the data. When the algorithm can’t contextualize the data (like during a snowstorm, or major road works scheme), then give me facts.

    • Michael Robinson January 18, 2017 at 1:33 pm #

      I agree totally about the London example. There can be big differences in journey time over the same distance between different routes because of differences such as junctions or roundabouts. At a bus stop with multiple routes, knowing distances is not the same as knowing what bus will get there first.

      I’ve just started to use Amazon Alexa to ask her for the arrival times of buses at my nearest stop in London and if a particular bus is, say, 10 minutes away, I know I can leave within 5 minutes and be at the stop on time allowing for a margin of it being a minute early. If it provided distances, that is all a bit too complicated to work out and really, I don’t need exact times but if the bus should arrive within 10 minutes, 20 minutes or is the service totally messed up.

  3. Arne Nys January 18, 2017 at 1:17 pm #

    I think this approach only works for people who are familiar with the city, its transportation system and who have no issue estimating the remaining time. For a lot of people though, this is not so easy. I think there are some flaws to using miles to show how far away they are:
    – 1 mile does not mean the same wait time for line A as for line B. It depends on the street network and how the traffic flows.
    – It also differs from city to city, so people that are not familiar with the system would have no idea how to value the distance.
    – Let alone that lots of people (non-Americans) do not use miles but kilometres, and vice-versa. Minutes, hoewever, are universal.

    A public transportation system should be accessible and easy to understand in order for people to use it. Inserting an extra step between the question ‘when does my bus come’ and the answer will alienate people from the information.
    Additionally, how would a route planner that takes into account real-time position of vehicles use remaining distance? There will always be a conversion to travel time if you want to give people the fastest route available.

    I agree that estimating remaining wait time is not always very accurate. I see a lot of potential however in new algorithms that look at the historical performance of the system at a particular time of day for individual vehicles, and estimate remaining time based on that. That would make it much more useful and accurate than it is now.

    • Arne Nys January 18, 2017 at 1:21 pm #

      It’s cool however that TriMet does this in situations where no accurate predictions can be made, like a snowstorm. In that case, like you say, it doesn’t make sense to make predictions.

  4. Adam H. January 18, 2017 at 5:23 pm #

    The transit tracker was all kinds of messed up this weekend. Even as I used the mile-based tracker, buses would jump from 2 miles away to 5 miles. Many buses simply fell off the GPS and stopped announcing stops.

    • Michael Miller January 20, 2017 at 10:49 pm #

      @Adam H., maybe you were seeing the bus sliding back down a hill. 😉

      I’m in Portland as well, and I guess I must not have used transit tracker itself directly during those times. I typically use PDXBus except when I am sitting in front of a computer, and I never noticed it display mileage in place of time. However, I rode my bike Thursday and Friday last week, so didn’t use transit at all on the two days combining highest road impacts with relatively heavier traffic.

  5. Kyle B. January 19, 2017 at 6:22 am #

    If this is done properly, you could see the progression of the bus as it inches closer to your stop. On the other hand, with time-based predictions, they generally “freeze” during delays, which says something about the relativity of time, but not much about whether the bus or train is advancing.

    NYCT pairs the countdown clocks with periodic announcements about how far away the next train is, this is a decent stopgap solution.

  6. Voony January 19, 2017 at 9:57 pm #

    I have seen this in Hanoi

    …and has a tourist unfamiliar with both the city and the language, I have greatly appreciated it and didn’t share the concern expressed here…

    waiting for the bus, you get a good idea of the real time traffic condition, and build yourself a sense of how much time you have to wait considering the distance…

    The real issue is not to present fact or prediction, but to present reliable information

    If you can present reliable waiting time please do it! …after all it’s the info I want.
    But if not, please stick to the fact !

    It seems it was the case during the Portland snowstorm, and they did the right thing…

  7. Paul January 21, 2017 at 12:42 am #

    I speak for London and the other UK real time screens, replacing the word “Due” with “Near” would calm some frustration.

    A bus can be at the “due” point at a set of lights, but if the junction is blocked and can’t move it can be “due” for a while.

    • Gag Ha;frunt January 22, 2017 at 1:12 pm #

      I’ve actually seen that happen.

      As far as I know, “Due” appears when the estimated waiting time reaches zero. The positioning data isn’t precise enough to show whether the bus has arrived at the stop or is merely near it, so if “Due” were changed to “Near” a bus that had already arrived would be shown as “Near”, which might cause confusion.

  8. Aaron Priven January 23, 2017 at 11:10 pm #

    I think Arne Nys said it well. Bus location itself is not useful information. Who cares where it is now? It could be right next to you, waiting for its operator to get back from a break. So what? Ultimately, one can’t do anything with that “mile” distance except make one’s own prediction — and is that really the passenger’s job?

    I think the other thing that isn’t mentioned here is that, at least in the systems I’m familiar with, the geolocation isn’t perfect either, both because of the imperfections of the GPS system (with signals bouncing off buildings and so on), and also because of polling times — often the system doesn’t know where the vehicle is *now*, only where it was the last time it was polled, which could have been minutes ago. So the location of the bus really is also a prediction, not a fact.

    For what it’s worth, I do think it would be comprehensible to give ranges, if that’s the margin of error: “Line 42: 3-5, 9-12, 15-20 min”.

  9. bjorn January 25, 2017 at 4:43 pm #

    This may have happened downtown on the screens but the phone app did not display distance. Overall I found the time estimates to be pretty worthless during the snow. In fact several time buses that were showing as a few minutes away simply never came. In this way trimet failed completely, the number one thing that must be accurately shown is has trimet decided to cancel all the buses on the route for the rest of the night, and more than once the what should have been the last bus of the night simply never came. Nothing like waiting 45 minutes to find out you are completely screwed.

  10. Eric Goodman March 18, 2017 at 2:17 pm #

    This appeals to me. Particularly in the context of service compromised by weather effects. Have had discussions on experience with a BRT line. Constant difficulty in giving accurate predictions. Signals complicate travel time estimation, more so over short distance. The research says we experience less wait time with real-time data available. Wonder if anyone has measured people becoming less patient /more irritated when fluctuations or dysfunction in the prediction occur?

    With standard ways to give and get information like gtfs-rt, the transit agency can supply all of it. Developers decide what data to push and users decide what apps to consume for thier own needs and preferences. On an agency’s own assets, in critical times, my preference would be facts. Of course, would still check my phone for prediction fiction.