Bus prediction
Today in the US is Thankgsiving, and it is traditional to be relentlessly positive, each large amounts of turkey, and collapse in front of the football game.
In partial observance of the tradition, I want to praise the new bus prediction system display. The predictions are still too optimistic about remote buses, but at least the new display lets you tell which predictions are based on actual bus locations and which ones are just based on the timetable (eg).
Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »
It seems to me completely reasonable for the prediction function to have some bias: consider the respective utility functions of a bus that is 10 minutes after the predicted time, versus one that is 1 minute before the predicted time…
With the introduction of Hop cards, loading & unloading contribute relatively little “jitter” to the running times of buses; most jitter (and hence “delay”) is due to queuing at intersections, either because of traffic density or traffic light phasing or both, so to a first approximation every intersection can be treated as an independent source of jitter. (Those with traffic lights have a strongly step-wise distribution.)
If I were building such a system, the displayed ETA would be the 99% “early” outlier, after taking into account any time-points where a bus must stop if it is running early. (Some shorter routes only have a starting time-point.)
12 years ago
99th %ile seems a bit much. In that case a “20 minutes” prediction would just mean there is a 1 in 100 chance of a bus within 20 minutes. That would rarely improve on the timetable, removing most of the point of real-time prediction.
12 years ago