Well-intentioned U.S. Frequency Analysis Causing Needless Angst

The best of intentions lay behind the recent analysis of the availability of high frequency service, described by Jake Anbinder at TransitCenter.  And as one of the original proponents of Frequent Network branding, I’m delighted to see organizations starting to care about whether transit is useful, rather than just whether it exists.

But when we transit professionals design Frequent Networks and their standards, we think really carefully about what counts as frequent service.  I insist that it always means every 15 minutes or better on both peaks and throughout the midday, but we have to have a sensitive local conversation about how late it extends into the evening, and how long on weekends.  It’s easy to say that you want 15 minute service 24 hours a day, but when you look at your fixed resources, you’ll find that if you insist on that, your Frequent Network won’t cover enough of your city to be either useful or politically feasible.

That’s exactly the problem that makes the analysis misleading.  Here’s the chart making the rounds from the their work:

Screen-Shot-2016-06-07-at-10.20.35-AM

The problem is that this analysis defines frequent transit as 672 trips each direction per week on a route segment.  Why 672?  It’s 4 x 24 x 7, that is:

  • 4 trips per hour (so every 15 minutes)
  • 24 hours per day
  • 7 days per week.

For service to count as frequent, by this analysis, it has to be frequent even at 3:00 AM.  

I can’t think of a bus route anywhere in North America that runs every 15 minutes in the middle of the night.  Maybe there are a few in Manhattan, but this calculation is an unrealistic basis for defining Frequent service anywhere in North America, except for the densest, biggest cities..  (Obviously, the other way to meet this target is to be very, very frequent during the day to compensate for not being frequent at night.)

Why quibble about this?  Because this chart caused some needless angst in the media.  Among its worst outcomes would be to signal that frequent service is such a rarefied thing that most cities couldn’t realistically hope for it.  It certainly obscures all the great work that’s actually being done to build realistic but useful frequent networks in many cities.

For example, it triggered this Houston Press article implying that there’s some racial disparity in the distribution of frequent service, based on the bizarre notion that Houston’s frequent service consists only of the areas highlighted in yellow:

screen_shot_2016-06-14_at_5.28.11_pm

The Press’s Megan Flynn writes:

In Houston, according to TransitCenter’s analysis, while nearly 80 percent of people have access to transit at least a half mile away from their doorstep, only 18 percent of them have access to what the map’s algorithm considers the highly reliable, “high-frequency” service.

By the AllTransit definition of high-frequency—service running at least every 15 minutes on average, 672 times a week—only one bus route, the Westheimer route, meets that standard, along with all three of Houston Metro’s light-rail lines.

As TransitCenter noted in its analysis: “Notably, in cities with fewer high-frequency transit lines there tends to be a greater demographic skew among people who live near quality transit.”

And it considered Houston to be one of the two “most notable” examples of that demographic skew.

Even though black and white people in Houston each have equal walking-distance access to public transportation in general—each making up 24 percent of the pie—36 percent of white people can access high-frequency service while only 19 percent of the black population can. Latinos make up 45 percent of people with walking-distance access, yet only 34 percent access high-frequency routes. (See The New York Times‘s map of the racial makeup of Houston’s neighborhoods here to compare with the above map.)

Ouch!  Great basis for rage.  But  it’s not about what matters!  When we designed the Houston network, our standard for frequent service was not 672 buses/week, and the map above illustrates why.  If it had been, we’d have brought frequent service to too small a part of the city, and we could have fairly been accused of racial disparity.  That’s why we didn’t!

As it is, the Houston Frequent network (the red lines on this map) is abundant and citywide.   It may not fit some abstract big-city standard, but it’s the fair and equitable way to cover most of Houston with frequent service given the transit agency’s resources, and it’s sculpted to hit inflection points on the frequency spectrum where ridership begins to take off.  The same is true of many of the cities that “fail” the 672 buses/week test.

TransitCenter is an excellent organization, by the way, and this showed in how open they were to this critique when I shared it in draft.  In our correspondence they asked me what a good benchmark would be.  I suggested “15-15-7”:  15 minute frequency, 15 hours a day, 7 days a week, plus 30 minute service for an additional 5 hours a day.  That gives you a total service day of 20 hours, e.g. 5:00 AM – 1:00 AM.  That’s only 490 bus trips/week, so their estimate of 672 is high by over a third.  The chart would be totally different, and more useful at motivating change in cities that aren’t as dense as New York or San Francisco, if a standard around there had been used.

31 Responses to Well-intentioned U.S. Frequency Analysis Causing Needless Angst

  1. Robert Wightman June 20, 2016 at 9:37 am #

    Check out the schedule fo the Yonge Night bus, route 320, in Toronto.

    http://www.ttc.ca/Schedule/schedule.jsp?Route=320N&Stop=n.b._on_Yonge_St_at_Queen_St_East_%28Queen_Station%29

    None of the other routes operate that frequently but there are a lot of night routes operating at least every half hour.

    • Shaun June 22, 2016 at 7:12 am #

      I was thinking this Robert and saw that you beat me to it! I regularly use “the Blue Night” TTC service between 1am-6am, and can confirm the reasonable frequency of the Yonge (320) and Bloor (300) buses. Both are 15 minutes at their most infrequent time (about 4am), but are more frequent through most of the night. Regular transit riders know this and try to maximize the use of the 300 and 320 for middle of the night trips.

      In the larger public conscience, however, after 1am “the subway is closed”; a slight exaggeration of the actual reduction in transit service. Interestingly, the 300 and 320 face significant crowding issues as the capacities on Yonge & Bloor drop from trains every 4 minutes to buses every 4 minutes at a specific time.

      To my knowledge, every one of the other 28 Blue Night routes generally operates on a half-hour schedule with no guaranteed connection. It’s basically a high frequency grid at low frequency. Within the City of Toronto, this strikes me as the ultimate “coverage over ridership” trade-off. I understand why it happens, but would far rather see us focus energy on dealing with the crowding issues and building more frequent trunk routes. I would gladly walk a bit further at the start or end of a trip than the current scenario of waiting 29 minutes in the middle of the night – and often deciding to just walk/taxi home anyway.

  2. Jacob June 20, 2016 at 10:37 am #

    So basically, you’d rather have a map of frequent transit instead of really frequent transit, which is completely fair. That said, I think a map of really frequent transit is also quite useful, and the disparities between the frequent network and the really frequent network are real, even if they don’t make Houston look great. I agree that a place like Houston may not be justified building a really frequent network now, due to it’s low density. But, by better understanding this, Houston may decide to focus more on increasing density and mixture of uses, in addition to revamping transit, to better compete with cities like San Francisco and Chicago. Generally, I think the goal should be to provide useful information to cities, not necessarily to praise them. As such, I think having a range of information, including both this analysis and the one you propose is helpful to get a more complete understanding of the situation.

  3. Alon Levy June 20, 2016 at 10:40 am #

    In New York, the busiest night buses come every 30 or 40 minutes.

    At any rate, I think a bigger quibble than the metric for how many buses per week should be midday frequency. This is easy to measure (you look at average frequency between 10 or 11 am and 2 or 3 pm). The point of this is that a lot of services can look frequent if they run high frequency at the peak but then drop to nothing off-peak, e.g. some busy commuter lines.

    • Henry June 26, 2016 at 4:35 pm #

      While running a calculation for every hour would be more accurate, it would very quickly become expensive to do an hour-by-hour check for the valid time frame. It could also make comparison harder, since the definition of midday or peak-period can vary from city to city.

      • Jason July 4, 2016 at 11:33 am #

        Unless I’m missing something I don’t see why you couldn’t just find out what each city’s peak hours are defined as and use the total number of peak hours per day as a weighting factor, and do likewise for off-peak service.

  4. calwatch June 20, 2016 at 2:47 pm #

    At this rate many Asian cities, which have continuous 10-12 minute headways during the day, would not qualify because they shut down after midnight. I agree with Alon that midday frequency would be better, and possibly using different standards for routes – one for peaks, one for middays, and one for evenings – since many transit agencies (Los Angeles being the classic example) have 4-6 buses an hour during middays, but drop to hourly service after 8 pm.

  5. Bjorn Swenson June 20, 2016 at 2:55 pm #

    There are two factors here, and both are invariably race-related.

    First, due to the effective lack of land use controls, Houston within METRO’s service area likely has an unusual (for the US) pattern where White and Hispanic neighborhoods are more dense than African-American neighborhoods. African-Americans appear to have a predominant cluster in northeast Houston, in neighborhoods where the housing stock is either 60s and 70s bungalows on quarter to half-acre lots or inward-facing garden apartments. In contrast, unlike basically any other US city White (and middle-income Hispanic) households tend to purchase duplexes and townhouses in formerly single-family traditional neighborhoods. Lower income Whites and Hispanics tend to rent from dense apartments. As an example, the Gulfton complex in southwest Houston, built for single workers in the 70s but occupied by Hispanic families today, has a density of 50,000 ppsm with only two-story apartment buildings. While Houston’s townhouse stock is relatively cheap for wealthier white-collar workers who predominately tend to be white, especially compared to denser neighborhoods on the coasts, for lower income service-sector workers who tend to be Hispanic and African-American, perhaps hopping between two low-paying 30-hour-a-week jobs, those townhouses are well out of reach for their family. (This was a big topic at YIMBY 2016; while upzoning everything in hot neighborhoods like Houston definitely lowers the price of a given unit from perhaps $600,000 to $300,000, even the lower price is out of reach of lower income families, especially people in non-White communities). As a result, poor families that predominately tend to be African American and Hispanic go seek housing that they can afford; $120,000 filtered-down bungalows and $800/month dated apartments tend to be what they can afford to live in.

    The other factor relates to how various ethnic communities use transit, and is also how defining ‘good’ service as a 15-minute mid-day headway is in and of itself a moral judgement. At a very casual level, I’ve noticed that a split between cities that have enough money to run more frequent daytime buses or later evening service, but not both. Cities where the predominant non-white majority is Hispanic (Albuquerque, Tucson, basically the southwestern US) tend to have buses run every 20-30 minutes or so, but have little to no service after 8 or 9. In contrast, cities where the predominant non-White majority is African-American (Dayton, Flint, basically the rust belt) tend to run their buses every 30-60 minutes, but the routes start earlier and run later into the evening. I fully admit this is an area for further research, but the information I do have is two-fold; African American households are less likely to have access to a vehicle (see http://www.fhwa.dot.gov/planning/census_issues/ctpp/articles/vamtw.cfm ) compared to Hispanics. Secondly, Hispanics tend to have larger household sizes (see http://www.pewhispanic.org/2016/04/19/statistical-portrait-of-hispanics-in-the-united-states/ph_2015-03_statistical-portrait-of-hispanics-in-the-united-states-2013_current-40/ ).

    Put the two together and the result is that African-Americans tend to have smaller households that are completely car-less, while Hispanics tend to have larger households that are car-lite. In turn, this helps explain why cities with a large African-American population tend to have later, less-frequent buses (they need bus access for all of their trips, especially involving second or third-shift work commutes at odd hours), while cities with Hispanic communities tend to have relatively frequent daytime service as the one household member who works the graveyard shift can drive themselves to work.

    As a result, demanding 15-minute mid-day service at the expense of night service is, in and of itself, a moral judgement with the potential to have a adverse effect, particularly on African-Americans. Combine this with the reality that Houston’s White and Hispanic populations live at higher densities than its African-American population, while a demand-based restructuring did increase ridership, it also leaves poor African-American neighborhoods with hourly coverage routes that end at dinnertime, presumably well before many workers in those neighborhoods punch out.

    Counting the number of buses per week is a more useful measurement of frequency that reflects the differences in customer travel patterns in communities, instead of a moral judgement that deems midday buses to count but midnight buses to be irrelevant. Like it or not, when a frequency of 672 buses per week is used as the metric the only neighborhoods that qualify are those along the 82-Westheimer that are predominately Hispanic and White, undergoing significant redevelopment, not African-American neighborhoods.

  6. Jay Blazek Crossley June 20, 2016 at 4:38 pm #

    Interestingly, my view of how this article and graph was received in Houston was generally excited that Houston showed a much greater very frequent network than any Southern city – a confirmation of the success of the New Bus Network. The Houston Press has become generally anti-TransitAgency for a while and of course would try as hard as possible to find a way to accuse Metro of doing something wrong, while completely ignoring similar equity questions about TXDOT or ignoring that this map shows that Houston Metro is doing a much better job of providing very good transit access to a lot of people of color than any other southern city.

    Also, while the entire New Bus Network is a triumph and changing Houston, living along the 82 Westheimer is substantially different – better – than anywhere else in Texas in terms of access to transit service that isn’t rail, so it is a meaningful thing for Houston that that vast area has that level of service. So I would love to see the 490 map and chart, but that doesn’t take away from the usefulness of this very frequent map.

  7. Al M June 20, 2016 at 6:20 pm #

    They applied “their” standard equally to each district

    Why is ‘their’ standard less relevant than ‘your’ standard?

    As long as it’s applied equally to each district it’s an accurate measure with informative results

    • Fbfree June 21, 2016 at 12:27 pm #

      “Why is ‘their’ standard less relevant than ‘your’ standard?”

      The information content of the map changes with the cut value. Imagine a map showing only where >1000 vehicles per week passed. The map would be so sparse as to be of little interest. It’s not that there’s no information here (see Jay Crossley’s response above), but the analysis would be better with a lower cut value.

      You also want to remake the map over a range of relevant cut parameters (buses per week, or midday bus frequency, etc…). For hard cuts like this, a small change in the cut parameter could dramatically change the qualitative result of the map. The stability of the analysis should be presented.

  8. Ian Graham June 21, 2016 at 9:33 pm #

    Perhaps the most valuable result of the analysis at this point is the discussion it is provoking. “FTN” has become an unquestioned buzzword, with some justification. After all, frequency is one of the key drivers (if not THE key driver) of ridership, so none of us in the transit world should be fundamentally against it. However, it is still mostly applied in a binary way. If 15 minutes is the threshold (which is still a long time for standing in the rain), then 16 minutes is “infrequent”, and 3 minutes is no better than 15.

    The quality of transit service is so multi-dimensional in space and time that it is impossible to reduce it to a single index, especially when comparing diverse cities, and different types of travel patterns and priorities. But it is an interesting start. Given that all of the transit data used for the analysis presumably comes from an open source API (such as for Google Transit), it would be nice to be able to have an interactive tool that would let users adjust the key criteria (frequency thresholds, walking distances, times of day, connection options, etc.) and regenerate indices, charts and maps, for a set of cities. (I am not sure how much horsepower is would require for data crunching, but some of the elements would seem reasonable to optimize for this type of calculation.) Then one could compare “what percentages of population are, during the middle of a weekday, within 500 metres of service with headways of <=30 minutes (moderate), <=15 minutes (frequent), <=10 minutes (quite frequent), <=5 minutes (very frequent)?”, etc. The challenge is to incorporate calculations that can, to at least some degree, reflect the quality and connectivity of the network (e.g., so that high frequency grid networks get credit for the travel flexibility they provide, even if they have to reflect some inconvenience and delay of transfers). Some cities will score highly on some metrics, while others will shine on different ones. It could help unravel the essential differences between Phoenix and Seattle, for example. (Political benefit: practically everyone will be able to find at least one measure that looks good, while critics and advocates can point to potential improvements.)

    • Mike Robinson June 23, 2016 at 2:06 am #

      I know that in London, TfL consider a 15 minute frequency as the threshold between a “turn up and go” service and a service where people will want to consult a timetable before travelling.

      If you just miss a service, yes, people will need to wait 15 minutes but the average wait time will be less than this.

  9. Chris June 22, 2016 at 9:58 am #

    >>For service to count as frequent, by this analysis, it has to be frequent even at 3:00 AM.

    I can’t think of a bus route anywhere in North America that runs every 15 minutes in the middle of the night.”

    Im surprised at this comment Jarrett. Torontos 320-Yonge bus route runs every 3-4 minutes both directions at 3am:

    http://ttc.ca/Schedule/schedule.jsp?Route=320N&Stop=n.b._on_Yonge_St_at_Bloor_St_East_North_Side_%28Bloor_Station%29

    Torontos 300-Bloor-Danforth runs every 10 minutes at 3am (and every 5 minutes even at 2:30am):

    http://ttc.ca/Schedule/schedule.jsp?Route=300E&Stop=e.b._on_Bloor_St_West_at_Yonge_St

  10. Peter Laws June 22, 2016 at 4:31 pm #

    Let me guess. They looked at the numbers for each route and if it made 672 trips in a week, they decided it was frequent and if it didn’t make that many it wasn’t.

    NO problems with that methodology at all. None.

    • Chris June 23, 2016 at 5:28 am #

      You have to set up the cutoff at some number. I suppose you have a better solution.

      • Sailor Boy June 23, 2016 at 6:30 pm #

        Every 15 minute 6am to 9pm, actually useful.

        • Chris June 26, 2016 at 12:52 pm #

          That’s debatable. What if you’re taking transit to the bar and back at night? It’s not very useful to you if it’s only running every 30 or 40 minutes at that time.

          672 trips per week per direction is far more useful than only every 15 minutes 6am to 9pm.

          • Henry June 26, 2016 at 4:29 pm #

            Not even the greatest transit metropolises run transit that frequently at night, and there are problems with the 672 number (are they counting the fact that trips should run in both directions, not just one way?)

          • Philip July 5, 2016 at 6:42 am #

            @Henry above – London has a number of night bus routes with better than 5 minute frequency on Friday and Saturday nights.

        • R. W. Rynerson July 16, 2016 at 3:38 pm #

          @SailorBoy: In Denver that would be 5 a.m. till 8 p.m. We traced this back to network television, which doesn’t believe in the Mountain Time Zone. With that correction, the times that you suggest fit exactly with RTD’s experience in determining what a generic span for base service is. The next big drop in demand is at 11 p.m., corresponding to midnight at lower elevations.

      • Peter Laws July 3, 2016 at 12:51 pm #

        Of course, but 672 trips over 168 hours is quite different from 672 trips over (say) 6 peak hours over 5 days per week (672/168 vs 672/30). There are surely services that provide 22 trips/hour at peak that don’t even have 24×7 service otherwise (or if they do, it’s not 4 trips/hour at 0300). Pretending that 672 trips are evenly spread over the 168 just isn’t a good metric.

  11. Gavin June 24, 2016 at 4:37 pm #

    How does the algorithm handle combined headway corridors?

    For example in Brisbane, the 174 and 175 are half hourly routes that combine for high frequency along a significant inner stretch of that road, and share a common terminus in the CBD.

    Does their algorithm know when this occurs?

  12. Henry June 26, 2016 at 4:28 pm #

    I actually did something similar using publicly available GTFS data, but it’s a little more flexible in that you can specify the minimum headway and appropriate frequency. It also accounts for multiple routes that combine, by using transparency to show the level of frequency; overlapping lines become darker.

    The problem I have with their methodology is that it doesn’t look like it accounts for overlapping routes, and they made some other erroneous calculations as well assuming they’re using the same data source (you would actually need to double their cutoff because buses run in two directions, not one!)

    • R. W. Rynerson July 16, 2016 at 4:24 pm #

      @Henry: I think you are right about their system not comprehending overlapping routes. On Denver’s South Federal Blvd., Rtes 30, 31, 30L, and 36L provide coordinated service on a predominantly Hispanic and Asian commercial strip, This corridor doesn’t show on their map, which enabled them to lecture us on race relations. Ditto with light rail service on the Central trunk segment (served by C/D/E/F/H-Lines). There are some more segments that should have shown up.

      It was interesting to try neighborhoods along those same high-frequency corridors in their community ratings. Some of the “minority” communities showed high quality scores, and all showed “respectable” scores above 7. This reflects both the weaknesses in the Easterners’ program and the fact that there are multiple types of service needed by these communities.

  13. Baylink June 29, 2016 at 10:17 am #

    The underlying question, as I see it, is:

    Does the budget drive the design? Or does the design drive the budget.

    If you have needs, you *have them*, whether you can afford to fulfill them or not; if you can’t afford to fulfill them, you can’t simply redefine them as “not needs”.

    Or, as Tom Clancy once put it: “either send us the bombs, or rescind the missions”.

    • Baylink June 29, 2016 at 10:18 am #

      In a related topic, this is why, for my particular ‘project’, I’m fond of ATO monorail as a backbone: once the cars are on the rail, the incremental cost of running 10 minute headway is solely in how often you have to do maintenance on them.

  14. Alice Ngigi June 30, 2016 at 10:54 pm #

    Thanks for the post.

  15. Alice Ngigi June 30, 2016 at 10:55 pm #

    Quite useful

  16. Alice Ngigi June 30, 2016 at 10:56 pm #

    Thanks for sharing.

  17. Chris July 18, 2016 at 10:47 am #

    There’s a lot wrong with Atlanta transit, mostly because of land-use but a high fraction of its ridership is on grade-separated, frequent, heavy rail that starts relatively early and ends relatively late and covers a lot of ground. If those lines (at least before the splits) don’t meet the frequent threshold then the bar’s too high.