One idea that arose from the discussion of the possible Queen Anne-Downtown-First Hill-Madrona restructure that I blogged about a few weeks ago has been mentioned before in other threads, namely the possibility of extending the Queen Anne trolleybus routes up to Fremont. The case for making this change is evident just from looking at a map: the terminus of the 13 is about half a mile from the Fremont Bridge, a gap which is currently filled only by the infrequent daytime-only route 31. This short extension seems to offer the possibility of tying together two city neighborhoods with frequent service.
Unfortunately, it’s not that easy. Such a route entails negotiating the Fremont Bridge — the most frequently-raised drawbridge in Seattle — along with the traffic around the bridge, which can be terrible (especially on the north side). To me, the cost in terms of schedule time and reliability outweigh that advantage, so I’ve always argued against it. Fortunately, you don’t have to take my word for this any more, because I have obtained timepoint data make the situation clear, after the jump.
This chart is a normed histogram of the difference from average travel time for a selection of routes. It tells you what percentage of trips on a given route took between zero and one minute to get between timepoints, one and two minutes, etc. This chart puts in context the unreliability of the Fremont Bridge, using two sets of similar data that I happened to have lying around: the 3/4 and 27 data that was used in a previous reliability comparison. Here’s what I see in the data:
- Route 27 between 3rd Ave and Broadway (traveling via Yesler) is an example of a very reliable route. 80% of trips are within a minute of the average, 96% within two minutes. Only a minuscule number of trips are more than five minutes late. For a well-used bus route using standard buses, this is about as good as you can reasonably expect.
- Route 3/4, traveling a similar distance (but via James) is a well-known example of an unreliable route. Just over half the trips make it within a minute of the average time, about 15% taking more than four minutes longer.
- Routes 26 and 28 show the reliably of the Fremont Bridge approaching or departing via Dexter. These route sections exhibit similar unreliability to the segment of the 3/4 on James — if anything, slightly worse, with more long delays over four minutes.
- Route 30 and 31 approach or depart the Fremont Bridge via Nickerson — the same alignment that would be used by an extension of the Queen Anne buses. It exhibits the worst reliability of all, with just under half the buses making it within a minute of the average.
The bottom line: extending the 13 to Fremont would cause as much additional unreliability as moving the 3/4 wire from James to Yesler would solve, negating the point of that exercise, and perpetuating the current malaise of unreliability that afflicts much of the trolley network in the CBD. I don’t think the benefits are worth the cost in this case.
This analysis has some ramifications elsewhere. Another restructure sometimes floated in comments (and suggested in the 600k restructure scenario) is abolishing the 26 and 28 and putting the 5 on Dexter, passing through Fremont before joining its current route on Phinney Ridge. Like extending the 13, this does a better job of connecting dense neighborhoods, but at a very high price: it inflicts a delay and reliability penalty on many more riders who are merely passing though Fremont than is the case with the 26 and 28, as there are far more riders on the tail of the 5 than on the tails of the 26 and 28.
Ideally, we’d have a bus network that tied together our dense, close-in neighborhoods on a grid comprised of frequent, direct, and reliable buses. Unfortunately, the real world is distinctly non-ideal, and there are no perfect bus routes, only tradeoffs with varying costs and benefits.
Notes on the chart. You can stop reading if you’re not interested in the details of the calculations behind the chart.
The original dataset consisted of recent Metro timepoint data for the nearest timepoints to the Fremont Bridge for routes 26, 28, 30 and 31; and as specified in the previous post for routes 3, 4, and 27. The steps were as follows: for each route, for each direction, for each run, calculate the average time based on all trip; subtract that from all the trip times; bin that data in seven bins from zero to seven minutes, combining the different directions and runs; normalize within that set of bins.
This allows us to ignore dimensions of complexity that we don’t care about (direction, time of day) while preserving the validity of the average with respect to those dimensions. Two minor points: I’ve omitted deltas that were negative and more than seven minutes over, as the former are essentially uniform across all route combinations and the later comprised a very small fraction of any route.