Why is Bus Ridership Falling? — Notes on the Famous Mineta Paper

When respected authorities speculate about why transit ridership is falling in the US, they usually cite a 2015 paper by the (respected) Mineta Transportation Institute, authored by Bhuiyan Alam, Hilary Nixon, and Qiong Zhang.

The paper has one blindingly obvious conclusion that we shouldn’t need statistics to prove: If you want ridership, you have to run service.  The quantity of service, measured several ways, overwhelmingly determines ridership outcomes.  My comments about the paper in no way question this conclusion.

Still, the paper has problems that are common in papers in the statistical social sciences.  To some extent, I’m not even critiquing the paper so much as the discourse from which it arises.

(And if this is tl;dr, by the way, there’s a “Conclusion” section you can scroll down to.)

Useful Findings, Misleading Interpretation

The study seeks to explain the variation in passenger boardings per capita, an imperfect but easily calculated measure of ridership.  It looks at a large set of things that could explain this variation and concludes that:

The results indicate that gas price, transit fare, transit supply, revenue hours, average headway, safety, transit coverage, and service intensity show statistically significant impacts on transit demand by bus.

All of these except gas price are internal.  In simple language, an internal variable is a thing that someone could change, while an external one (like the weather) is one that they can’t. [1]

But their interpretation of this is deeply misleading:

The results show that the internal variables, the factors that transit managers and operators control, are predominantly the significant predictors of transit travel demand by bus mode. Seven out of eight internal variables in the OLS regression model proved to be significant factors in determining travel demand by bus. [emphasis added]

It is utterly false to say that the internal variables are under the control of “transit managers and operators,” unless you think they can print money. Again, these variables are “transit fare, transit supply, revenue hours, average headway, safety, transit coverage, and service intensity.”

Transit managements can turn fares up or down, and they can be more or less careful about safety, and they can hire firms like mine to help them redesign their networks.  But the larger reality of service quantity — the most important point of the entire paper — is mostly the result of investment decisions made above their level.  Transit is mostly subsidized due to its public benefits, and the level of that subsidy is controlled by some mix of elected officials and voters.  Management has a marginal role in how efficiently that subsidy is translated into service quantity. [2]  Mostly, you get in service what you’ve paid for in subsidy.

This mistake goes to a critical problem in the way the bus ridership decline is being discussed, and why so much activism around the issue is misfiring.  More people are yelling at their transit agencies than are yelling at the elected officials who actually control service quantity   The Mineta paper’s careless language encourages that confusion.  There is no point in telling transit managers that they should run more service.  They agree with you.  You need to tell elected officials this.

What Doesn’t Matter?

The findings about what matter to transit ridership are interesting, but so are claims about what factors don’t matter. Here the authors make sweeping claims about how clueless we transit planners are:

The study found that certain variables that many transit planners view as important determinants of transit demand did not have significant impacts on transit demand. … Variables such as transit orientation pattern, median household income, percentage of college population, percentage of immigrant population, vehicles per household, and MSAs in the South … do not impart significant effects on transit demand by bus. … Population density and the percentage of households without cars show insignificant impacts on transit demand …

To which I can only say, it depends on how you measure these things.

The finding that population density doesn’t matter is based on a common mistake.  The authors measure the total density of Metropolitan Statistical Areas, which are aggregations of US counties that may contain vast expanses of rural area and even wilderness. (More on this, with photos, here.) But even average density of the truly urban area is not what matters.  What matters is density adjacent to transit service.  (The late Paul Mees made the same mistake in his book Transport for Suburbia, as I discuss here.)

This is an example of a case where some geometric thinking would have helped, which is what I tried to do in my transit ridership explainer.  The residential density that matters to transit is the number of people within a fixed radius of a transit stop.  If ridership wasn’t twice as high where density is twice as high, this would mean that individuals living at low density are more likely to use transit than those living at high density.  Such a claim would not only be wildly counterintuitive, it’s also disproven by virtually every transit agency’s stop-by-stop ridership data, as long as you focus narrowly on density near transit.

dot map dublin

A segment of Dublin’s bus route 13, with passenger boardings as dots and residential density in the background. More density right around the route means more ridership. Makes sense, doesn’t it?

Problems with Rates

The confusion about how to measure density also affects a confusion about terms expressed in percentages, medians or averages, as most of the demographic variables are.

The authors run correlations with the percentage of people who are immigrants or college students. Vast parts of an urban area where transit isn’t abundant enough to be useful, if it exists at all, are counted in these rates.   The median income of an entire city matters less to ridership than the median income of the parts of the city served by useful transit.  These are always very different things.

That’s why our firm almost never studies or maps percentages; instead, we study densities.  We draw maps of the density of poor people, or students, or seniors, or whatever.  Because only that shows you how many people we’re really talking about, and where they are.

Conclusions

Too often, social science papers rely on correlations without thinking about what transit is spatially.  Everything I’ve said here will be blindingly obvious to any practicing transit planner —  not just because they show up in properly granular analyses but because the mathematical consequences of them not being true are so nonsensical.  That’s why I’m confident in relying on geometric claims in my own work.  They tend to win arguments in the political world because they don’t require anyone trust a black box of analysis that’s studded with assumptions, or to assume that experts always know best.

The biggest single mistake in this and most similar studies is the false confidence in aggregating data across a metro region.  It is geometrically inevitable that any remotely viable transit agency will distribute its benefits very unequally across the land area of its region — especially in response to density — and the way that transit’s benefits are distributed over a city means that total inputs and outputs at the citywide scale don’t matter very much.

But these papers are at their most exasperating when their resistance to geometric thought is coupled with unfounded claims about how clueless we practitioners were before this paper enlightened us. As a PhD myself, I have the highest respect for the work of scholarship.  But a regression analysis is only as good as the assumptions that went into it, and these need much firmer grounding in geometric reality, as well as in the reality of how transit decisions are actually made.

Nevertheless, the main point of the study, and the one for which it’s most often cited, is indisputable:  Network design projects can help improve ridership for a given amount of money, but for step-changes in ridership, you have to fund more service.

 

[1] The internal/external distinction, routine in the social sciences, is entirely relative in ways that should be more clearly marked in papers.  Whether a factor is internal or external depends on the selected point of view.  My reaction to seeing transit quantity described as internal is that it is not in control of the stated point of view, namely “transit managers and operators.”  Weather used to be the paradigmatic example of an external variable, but now that we know it’s partly the result of human actions, it could be internal if you take the long view of human agency.

[2] Not a zero role, but very small compared to the magnitude of cost involved.

 

10 Responses to Why is Bus Ridership Falling? — Notes on the Famous Mineta Paper

  1. MLD August 29, 2017 at 12:40 pm #

    People try to propose to me that we should do these types of analyses – tell people what kinds of social factors influence ridership! What’s the real ridership recipe?

    Except those analyses are useless without granular data like stop-level ridership data and tract- or block-level social factors data. And that level of ridership data just isn’t available in very many places, if at all!

  2. Standard Error August 29, 2017 at 12:59 pm #

    Unlike a private practice, that can go broke if it issues bad advice, quite a lot of academic research has major reproducibility problems.

    Part of this could be the pressure to publish, which leads to a large volume of “noise” papers being pushed out.

  3. Owen Evans August 30, 2017 at 5:59 am #

    To me the statement that “All else being equal, there is no significant correlation between density and ridership rates” could make sense.

    In the abstract, it’s not the density itself that makes it more likely for a given individual to use transit; it’s the better service that typically comes along with density. In other words: all else equal, if you have half as many people close to a given bus stop, you can expect half the ridership. An analogy: If you take the NYC MTA and all its subways, buses, commuter rail, etc, and transplant it into my home city of Raleigh, NC, while keeping other factors like fares, frequency, and total revenue service hours, you could expect the same percentage of people in the service area to use it here in Raleigh, as currently do in New York. The ridership NUMBERS wouldn’t match New York, of course, but the ridership RATES would remain relatively unchanged.

    On another note, some other factors that I’d particularly like to see studied for correlation with transit use: Roadway congestion, Roadway capacity (eg. freeway lane-miles per capita), and Congestion.

  4. Brian O'Malley August 30, 2017 at 11:43 am #

    Thank you for this. At the transportation advocacy nonprofit where I work (www.cmtalliance.org) we found this study very instructive and useful in the big picture, but we had difficulty with some of its details. In particular we struggled with the definition of and discussion of the service intensity variable. It seems the relationship the authors found between service intensity and ridership is the inverse of what we would expect. We reached out to a couple of experts we know to see if they could read the article and explain it to us, but they ended up having similar questions. So a colleague of ours reached out to the authors. The principal investigator responded to ask for time to get back to us due to other factors. That’s still pending and we’re looking forward to learning more.

    The authors derive the service intensity variable from two data points in the National Transit Database (NTD). They find that service intensity is significantly correlated to service demand. On p. 36/69 and p. 44/69 there is discussion of the service intensity variable. They say that a high ratio of vehicle miles to route miles shows that too much mileage is being driven “outside the map” (i.e. non-revenue service). But that seems to ignore or overlook that the ratio of vehicle miles to route miles would increase with frequency of service.

    Here’s what we suggest: use two variables from the NTD, Actual Vehicle Revenue Miles (VRM) and Directional Route Miles (DRM). Construct a ‘service intensity’ variable by dividing VRM by DRM. That should represent the frequency of service. We would expect it to have a positive correlation with outcomes like ridership; not a negative one.

  5. Alan August 30, 2017 at 1:58 pm #

    Jarrett,

    There may not be a general answer to this question – but on the whole, how much control do transit managers and operators have over what proportion of transit resources to devote to different modes within the network (such as rail versus bus lines)?

    Overall system budgets may not be under the control of transit officials, but allocation of resources within those overall constraints might be. If transit managers have the ability to make trade-offs between bus and rail systems, then budget would certainly be an internal factor when looking at what determines transit use for just the bus mode.

    Alan

    • Adam Tauno Williams August 31, 2017 at 3:09 am #

      > what proportion of transit resources to devote to different modes within the network
      > (such as rail versus bus lines)?

      The **VAST** majority of transit systems in the United States are bus only; so for the vast majority of transit planners this is not a question. Maybe if they are lucky they have a BRT mixed in there, still, mostly not.

  6. Adam Tauno Williams August 31, 2017 at 3:12 am #

    “Transit managements can turn fares up or down”

    They may not even be able to do that.

    • el_slapper August 31, 2017 at 6:50 am #

      @Adam : example in the Parisian Area, where the regional elected assembly decides the fare level(and it’s a huge point in each election, given the importance of the Transilien Network in most people’s daily life, over there).

      The future subway lines have even been decided at the national level, by former president Sarkozy. I’m not complaining, the design is not that bad. Still, when I’m looking at the parameters, I feel poor for the transit agency :

      gas price : of course no lever on that
      transit far : no lever on that, decided on the regional level
      transit supply : mostly dependant on budget, decided on the regional level
      revenue hours : some lever on that, but trade unions mostly forbid any change at this level.
      average headway : (my english is not good enough to get what it means, sorry…)
      safety : OK, they have leverage there. They have their own police – with debatable results, to stay polite
      transit coverage : no lever, it’s decided at the national level.
      service intensity : directly dependant on funding level.

      In other words, besides safety, Parisian transit authority has no lever on the strategic decisions that can make or unmake a network. They are lucky to be backed up by a rather motivated regional electorate board(that knows that weakening the Transilien would be political suicide), and unlucky to have been deprived of national resources diverted for building the nationwide high-speed train system(other french regions have been similarly mistreated in the same context).

      • Max Wyss September 2, 2017 at 1:22 am #

        “average headway” can be translated as “fréquence” or “interval”, and means the number of minutes between scheduled departures, or its inverse, number of services per hour.

        • el_slapper September 4, 2017 at 3:30 am #

          Ah, thanks. So it is also a byproduct of the level of funding the transit authority receives…

Leave a Reply