Racial Justice and Transport Modeling

Two important thinkpieces just appeared that seem to be about different topics but should to be read together.

Christof Spieler, the Houston METRO Board member who drove the 2015 network redesign at the political level, has a piece at Kinder Institute called “Racism has shaped public transit, and it’s riddled with inequities.”  Meanwhile, in Vice, Aaron Gordon takes on “The Broken Algorithm that Poisoned American Transportation,” by which me means transportation demand models.

Spieler outlines how transportation, like everything else, has been forged through many decisions that reflected the values of the time, and how this history has created structures that still produce racially unjust outcomes today.  These structures can be literal infrastructure — like bus lanes designed to be useful for suburban commuters but useless for buses linking inner city residents to opportunity — but they can also be a wide range of bureaucratic and analytic procedures that continue those racially unjust practices in more subtle ways that the people executing those procedures don’t have to notice.

One of those procedures is transportation demand modeling, as Gordon describes it.  The best modeling is not nearly as dumb as the examples Gordon highlights.  But the problem of all modeling is that to show the effects of a proposed action, you have to assume that everything else in the background will remain constant, or at least will continue changing only along predictable paths.

When the modeling process considers many possible futures, the one that is most like the past is called the conservative assumption, as if that means “this is the safest thing to assume.”  This assumption seems calm and rational, attracting many people who would never call themselves conservative politically.  But fact, assuming that the future will be like the past can be crazy if the trajectory defined by the past is unsustainable — environmentally, financially, or morally.  “Unsustainable” means that it is going to change, and in that case, the “conservative” assumption is really the “self-delusion” assumption.

Transport modeling can’t be thrown out, but it never tells us what to do.  It is a basic logical fallacy to say that “the modeling shows we must do x.”  All modeling insights are if-then statements.  A full version of this statement, which I would like to see at the beginning of every modeling-drive transportation study, is:  “This report shows that if the future matches our assumptions, then you can expect this outcome.  But the future may not be like that.  In fact, maybe it shouldn’t be like that.  So what really happens is up to you.”

4 Responses to Racial Justice and Transport Modeling

  1. Scott Elaurant August 30, 2020 at 6:20 pm #

    Jarrett your comment that a model “never tells us what to do” hits the nail on the head. The same problem also applies to the economic assessment of transport projects, which has grown more complex over time, but not more accurate. In both cases, the existing modelling and assessment processes cannot cope with change in underlying behaviours and values.

    In my experience too often both processes (transport modelling and economics) post-date the decision on what a transport project will be. They are done to justify funding for a project, not to determine what the best course of action is. This obviously leads to circular reasoning in the assessment process.

    We need to move to a different process for planning transport. We should also change the models, but if we get the decision making process right, we may find that we need the models less than we think.

  2. Ben Ross August 31, 2020 at 10:08 am #

    I heartily agree with what you’re saying. But in my experience, a “conservative assumption” has a different meaning. (Admittedly, my experience is almost entirely in other fields of modeling – risk assessment, hydrogeology, etc.)

    A conservative assumption is one that assumes things go worse than you expect. Such an assumption is usually made when you don’t have a good projection of the future, but you can see several reasonable alternatives. You choose the one that is least favorable to your project. (Almost never the worst possible future, which would give highly misleading results, but the bad end of a reasonable range.) Then your modeling results give you confidence that the course of action you choose will work out well even if other things don’t go as well as you expect.

    An transit modeling example of a conservative assumption in this case would be rail ridership forecasts that undercount trips by university students. There’s no FTA-approved method for calculating student travel, so models typically assume they generate no more trips than anyone else with similar demographics, which is clearly wrong. Some models than add in an extra number, but that number still has to be conservative.

    Examples: Tucson streetcar found their initial computer model gave them sufficient ridership to qualify for FTA funding, so they didn’t bother trying to adjust it upward. Purple Line added the student ridership on existing bus shuttles along the Purple Line route, scaled for projected future enrollment growth, but did not take account of greatly improved frequency, longer span of operation, and added destinations that Purple Line will bring to some of the bus routes. Both these are “conservative” in the sense I’m using here, with the Tucson model more conservative than the Purple Line model.

  3. Roberta Robles September 6, 2020 at 7:24 am #

    Deeply disappointed at your ally-ship with continued use of these demand models. Given the complexity of the data and software involved it is not longer possible for outside groups to audit these inputs, assumptions and outputs. All of these variables get ‘nudged’ into the model with an clear cut eye towards the preferred alternative. Usually the one that skirts any impacts on neighborhoods of high money or influential status.

    Metro data is not free and any scope of transparency and communicating these issues are consistently *lost in translation* and co-opted by commercial and freight activist.

    We need to apply the network based approach that you pioneered for public transit here and apply to the freeway networks. Please acknowledge the wrongs caused in the past by demand based models by taking an activist approach to implementing:

    Freeway Congestion Management Principles:
    1. The most efficient freeway speed is 45 mph in urban areas.
    3. Closing some on and off ramps will improve freeway efficiency and improve local connectivity issues.
    3. Congestion pricing can be an equitable congestion management technique if the funds are used for public transit, walking and cycling facilities.

    Start by acknowledge the wrongs wedged into Albina and start acknowledging the ‘great’ planners who have supported your journey since before your book was published. Come out against Metro 2020 and the poor SW corridor alignment, which puts more Max stations next to freeways.

    Come out against more auxiliary freeway lanes. We don’t want a Phd diplomacy, I want you to fight for what’s right and use your platform to do it.

    Otherwise we will get the same old congestion and poor air quality near Harriet Tubman MS. And yes I do have legal standing here as my kids are tracked into that school next year. So yeah I’m pissed and I don’t mind calling out people by name. Why are we paying these transport economist to pump out bad results?

    You have enough social capitol to whether the good fight. Step up Jarret and fight for what’s right.

  4. Nathanael September 15, 2020 at 4:43 pm #

    These models are frequently trash. They have enough knobs and variables that they can be tweaked to say whatever the model user wants them to. Sometimes, as when the model gives absurdly low results (like the modelling for more train service in Western Massachusetts) they get caught manipulating it and are forced to redo it, but not often.

    I have nothing against some basic gravity modelling or basic principles like “be on the way” but the fancier models are typically used to justify biases.

Leave a Reply