Transit
NextBus accuracy slips, "ghost buses" explained
WMATA released NextBus accuracy statistics last week, and frequent users will not be surprised to find out that accuracy is not where it could be.
NextBus only has predictions for 78% of buses, far below the 92% accuracy target. The buses themselves are also not keeping to schedules particularly well, being "on time" only about 75% of the time even with a generous on time standard of up to 2 minutes early (which buses exceed almost 7% of the time) and 7 minutes late (which they exceed 18% of the time).
Bus manager Jack Requa provided some explanation for the "ghost buses" that riders widely reported: if a bus stands still for 2 minutes, the NextBus system stops showing that bus at all. It also removes the bus if it deviates more than 160 meters from the typical route, meaning that any bus reroutes generally result in disappearing buses.
The Examiner's Kytja Weir noted that this accuracy is lower than the 80% present when Metro "paused" the program for two years. That doesn't mean Metro should take NextBus down again (and they don't plan to), but does suggest some folly in taking it down in the first place, or keeping it under wraps until it was extremely accurate.
Still, when NextBus does work, it is very useful. When predictions do exist, they are usually correct (but not always), and make it possible to leave home or work just in time to catch a bus. Weir quotes Arlington Board member Chris Zimmerman saying, "NextBus has been one of the best things ... It's greatly improved the quality of my life. ... That said, it has to work with a fair amount of accuracy."
Overall, all riders are better off having an imperfect NextBus than no NextBus at all. However, the accuracy needs to improve. It would help for WMATA to detail its specific plans for what can be improved and at what cost.
The presentation misleads about customer satisfaction with its use of statistics. Requa's presentation shows a large pie chart of the total number of uses and a very tiny slice of the number of complaints. That makes it appear that every person not complaining is happy, which we know not to be necessarily the case. Requa said, "With 1.5But many customers don't complain because it's very difficult to complain. The online complaint form is very complex, and there aren't easy links to give feedback from the NextBus interface on the Web or on the phone. If Metro made it easier to complain, they would have more information about problems, but under the "customers are happy because there are few complaints" standard, it would make things seem worse. I hope that's not deterring the IT staff from making it easier to report problems.
It's terrific that Metro is tracking the accuracy for each bus garage and division, giving managers inside the bus system direct feedback about how well their groups are doing. The "dashboard" statistics we see do not reveal the causes of this inaccuracy, however. Are operators not signing on? Are the transponders breaking down? It would be helpful to see what percentage of the inaccuracies creep in at each stage. Zimmerman asked for this as well at the meeting. It looks like the dashboard listed on the Board presentation has the bottom cut off, so perhaps this information is on the full charts.
Accuracy seems to have slipped since January. The previous presentation showed log-on performance of 89.46% for the week ending January 23, 2010. Showing overall predictability instead of log-on performance is a better number, though it would be nice to show both. Bus performance also dropped, from almost 80% in that January week to 75% today.
Metro staff said that they are working on the problems, but gave few details beyond assurances that they were "doing everything that they can" and that "a lot of manpower and resources is needed." It'd be helpful to hear a roadmap for what Metro staff or the NextBus company itself is going to do to improve things. Ultimately, though, the proof will be in the numbers.
Comments
- Bikeshare is a gateway to private biking, not competition
- Judge denies injunction against closing schools
- Long-term closures: A solution to single-tracking?
- Metro policy for refunds after delays falls short, riders say
- M Street cycle track keeps improving, draws church anger
- PG planners propose bold new smart growth future
- Prince George's County struggles to get trails right









by thm on May 17, 2010 11:07 am • link • report
by David Alpert on May 17, 2010 11:09 am • link • report
I have gone back to using the printed schedule and trying to arrive 10-12 minutes before the bus is scheduled to arrive. And then I wait about 10-20 minutes after its stated time before I look for alternative arrangements.
Relying on nextbus has caused me to waste much more time -- whether waiting for buses that never appear or not waiting for a bus because its not listed -- only for it to show up.
Nextbus is not rocket science. The limitations put in place need to be reworked. Stopped for 120 seconds and then the bus disappears? That's outrageous. Whoever approved that needs to be fired.
by StuckinDC on May 17, 2010 11:17 am • link • report
by John on May 17, 2010 11:20 am • link • report
Even with its flaws, though, the amount of new data that WMATA should have on its bus operations is staggering. Are they actually using it to analyze and improve the way things are done? I'd think they could identify trouble spots where delays and bunching occur and then adjust schedules as necessary. They also have verifiable data now to follow up on customer complaints of, say, a bus leaving 5 minutes before the schedule says it is supposed to (which is at times a chronic problem on the A4 and A5 buses that I ride most frequently). They should also be able to accumulate enough late/early/transponder off/deviates from route statistics to flag potential "bad apples" amongst the drivers, and ultimately, quantify the prevalence of bad-appleness amongst the drivers as a whole.
by thm on May 17, 2010 11:22 am • link • report
My biggest complaint about the interface, on a mobile phone "browser" is that you have to type in a full stop number (they are often not posted at terminal bus stops, such as the Pentagon or Silver Spring Metro), or scroll through every single route, alphabetically. It would be great if the data entry accepted typing in "C4" instead of "2000192" and then I could scroll through to find my stop. That should be a very easy change for the developers to make.
I can understand the detour (160meters) causing the bus to fall off the screen, but 2 minutes of idling - really? Have the NextBus people ever sat at a left turn light in Northern Virginia?!?
by Joe in SS on May 17, 2010 11:24 am • link • report
by Adam L on May 17, 2010 11:26 am • link • report
Not sure what phone you have, but the Metro DC app on Android allows you to search by stop, bus name and/or number, and save favorites.
Re: Users v. complaints
I agree completely. If Nextbus was only 80% accurate during that selected two-month period, then Metro should expect to receive 200k complaints. I think WMATA's IT staff should work at making it easier to complain until they reach that goal ratio.
by Adam L on May 17, 2010 11:30 am • link • report
http://www.wmata.com/about_metro/contact_us/ridercomment.cfm
by bryandc on May 17, 2010 11:31 am • link • report
by John on May 17, 2010 11:32 am • link • report
http://www.wmata.com/about_metro/board_of_directors/board_docs/051310_NextBus.pdf
(The linked one is from earlier in the year.)
by thm on May 17, 2010 11:33 am • link • report
Also, I once called to complain about a bus being late, got a follow-up call and spoke with a manager, and I was assured the problem would be resolved. Overall, my only experience with Metro's customer service was positive.
by Teo on May 17, 2010 12:04 pm • link • report
by Jay on May 17, 2010 12:12 pm • link • report
Well, yes. And airplanes, when they don't crash, are also very useful.
John Catoe should have been fired long ago.
by Fritz on May 17, 2010 12:18 pm • link • report
Second, +1 on the comments re: the mobile web page by Joe. Far far too difficult to use with a modern phone browser. Should be much easier to just input your bus number rather than scroll through 100 choices. I don't use the app version, so maybe that fixes it. Seems as if it was designed for the SMS interface rather than mobile web. Needs to be both.
the idea that stopping for two minutes makes a ghost bus is insane.
I'm not sure using this as a metric for driver performance is a good idea. Plenty of things will slow down a bus that don't have to do with the driver. Gives us some baseline data but how many times is there a parked car blocking traffic, unruly passenger who won't pay, etc, that impacts performance?
What would also be interesting to see what are the most checked lines, and whether there in any increase in ridership on them.
by charlie on May 17, 2010 12:25 pm • link • report
First, misleading/wrong information is very harmful. If the system can't be aware of the accuracy of its predictions, it shouldn't give any predictions at all. I think I'll disagree with the statement that it's useful even if it doesn't work all the time. It should be all or nothing. The story about the S2/S9 above is evidence of the harm it can do.
Secondly, I want to note that if you go to nextbus.com on your mobile device, you'll be brought to a screen that asks if you want to add it to your desktop. Do it. It loads a link on your home screen that takes you right to your closest bus stops. I've stopped using the NextBus DC app and the other webpage and have been using this new service on my iPhone.
Third, in response to "charlie", the issue with driver performance isn't actually the performance of their bus operation, but forgetting to log in, or logging in improperly to the onboard bus systems. If the driver doesn't log in, the "system" knows there's a bus out there moving around but doesn't know which route to link it to, therefore, not giving predictions for it. Therefore, a bus shows up even when nextbus says there isn't one coming.
Fourth, I find it surprising that the conclusion slide from the board presentation claims that NextBus was very useful during the snow storms, yet doesn't show its performance during that time, blocking out the data and replacing it with "Snow Day!!!". Its times like these when we need reliable arrival predictions more than ever, and if Metro is going to take credit for being there for us, they should also let us know how well they did the job.
Finally, I agree that the predictions have been really bad lately. Just yesterday waiting for the 42 at Columbia and Ontario, Nextbus predicted a bus in 2 minutes, then switched the prediction to Arriving, but there was no bus to be seen. Except one in the opposite direction that pulled away from the stop at the moment the "arriving" prediction left the screen. Hrm.
All in all, Metro could do a much better job of getting accurate predictions on all their bus services.
by michael on May 17, 2010 12:38 pm • link • report
We often say "the perfect is the enemy of the good" but Nextbus isn't currently anywhere near good enough for that to apply. It shouldn't have been released in this kind of shape.
by jcm on May 17, 2010 1:01 pm • link • report
Have they checked the comment streams here and on unsuckdcmetro and the other blogs to add to the complaint totals? Perhaps they should.
Yes, it is too hard and a pain to make a complaint, so we can assume that those who did were really, really, really, pissed. I've just been somewhat pissed (bad prediction; missed bus by 20 seconds), so I haven't bothered to file a complaint.
I use NextBus as a curiosity tool. I use the schedule for planning, and then I check NextBus just to see how it does. I think the 78% is in the ballpark based on my experience, but I'm not willing to take a 22% chance.
by Steve O on May 17, 2010 1:32 pm • link • report
Ideally, they should have a service where you send a text message with your stop number, and it returns a list of the next buses arriving.
by andrew on May 17, 2010 1:35 pm • link • report
http://www.nextbus.com/wirelessConfig/sms.jsp
by Adam L on May 17, 2010 1:46 pm • link • report
by Joe on May 17, 2010 1:46 pm • link • report
by Michael Perkins on May 17, 2010 1:48 pm • link • report
Couldn't GPS and triangulation be used to figure out where the bus is or isn't.
Lets say the bus leaves the garage it sends a message to whatever saying that it is leaving the garage going to XX destination to start its route
Lets says its 32 when it arrives at Southern Ave couldn't GPS tell that it is at Southern Ave and then continue like it normally does until Friendship Hgts where the machines resets itself.
Why does a human need to be involved in that at all.
by kk on May 17, 2010 2:37 pm • link • report
I stated in a related topic that Nextbus is tied into the radio system that Metro has and the operators have to log onto their radios via their badge number and assignment number. It's probably something with taking the data from the radio system into Nextbus that might be causing some of these ghost bus problems. Even if they were to use GPS triangulation, it still couldn't tell if a bus was doing a 32 36 31 or even heading downtown to start another route. That's all dependent on the AVL system. On a side note, the AVL system that Metro uses is the same as Los Angeles, Chicago, Baltimore, and Montgomery County.
by K Conaway on May 17, 2010 2:54 pm • link • report
by Lauren on May 17, 2010 3:51 pm • link • report
by Xopher on May 17, 2010 8:00 pm • link • report
by Interested on May 18, 2010 8:14 am • link • report
I dunno about everyone else's experience, but for me personally NextBus is a huge leap over looking at paper schedules. This is especially true if you take an infrequently-running bus to work. Before NextBus, you had to try to plan and hope that the bus wasn't 10 minutes early or 10 minutes late or not coming at all. With NextBus, you can see when the bus is coming, or if it's not on there you can make other plans.
I ride multiple buses every day, and I haven't experienced many of the problems that people are always complaining about with NextBus. I've never had a bus that was listed as being 6 or 7 minutes away just not show up (a friend experienced this right after the snowstorm due to a detour). I've never had a bus listed as a couple minutes away suddenly jump to 15 or 20+ minutes away. I have had buses that show up 5 minutes late but I'd attribute that to specific traffic problems the software can't anticipate.
Is NextBus perfect? No it isn't - and the WMATA data backs that up. But is it better than not having it at all? Absolutely.
by MLD on May 18, 2010 9:07 am • link • report
Of course what's the purpose of NextBus if the bus has to follow the schedule to the second in order for NextBus to be right?
by b.o. on May 18, 2010 9:55 am • link • report
For the most part I've had buses show up pretty much as predicted - usually a few minutes later, rarely early or not at all. I've been taking Metrobus since I moved to DC in '93, so I'm already well-versed in changing transportation plans on the fly if there's no bus in sight.
by Serena on May 18, 2010 9:58 am • link • report
by lilkimbo on May 18, 2010 10:54 am • link • report
CAPTCHA- cobbing vehicle
by Mari on May 18, 2010 11:55 am • link • report
by RS on May 18, 2010 4:04 pm • link • report
by danielle on May 18, 2010 6:05 pm • link • report
The fact thatMetro makes is almost impossible to comment, much less complain - about their service is why they believe we are all happy little campers, gleefully riding the bus. The TV investigation of Metro Board members and their parking perks showed many of us riders that they were unfamiliar with Metro Rail... I would love to see them ride the X2 regularly and discover what real bus passengers endure.
by Greggndc on May 19, 2010 12:06 am • link • report
by lou on May 21, 2010 2:20 pm • link • report
I should just learn my lesson and check the schedules.
Trip planner told me 10:23 AM, then 10:53 AM for the next morning (today).
Missed one bus walking to the stop about 9:59. Checked Nextbus at the stop. It said 10:29 then 10:59. It's now 10:04.
Starting walking to local shops to kill time. Bus zooms by at 10:08 or so. Huh!??! F*^&! I could have been on that bus.
Decide to bag the shops and just wait. Bus comes at 10:22--nine minutes earlier than predicted just 20 minutes ago. If I'd gone in a shop, I would have missed this one, too.
So let's see. First NextBus fails to tell me of a bus coming in 5-8 minutes from now--a bus that Trip Planner never noted. Then it misses by almost 50% how long it will be before the following bus. Fail. All three buses were relatively close to the printed schedule, which I got once I was on the bus. Guess I'll stick to that.
So NextBus is worse than worthless. It's just like 20 years ago: go to the bus stop and wait for the next one.
(FYI, this was for the 1A/1B bus EB on Arlington Blvd.)
by Steve O on Jul 27, 2010 4:19 pm • link • report
Add a Comment