We Don’t Need No Models

The meeting the Hampton Roads/Tidewater members of the General Assembly had with the HRTA folks yesterday in at the state-funded Suffolk modeling and simulation facility held an interesting moment.

I got an email telling how it was revealed that the funds to model Hampton Roads traffic were cut from the budget. So, multi-billion dollar decisions are made on the basis of the bogus analysis I read back in 2002. Even that flawed work projected that building the big projects dreamed up in 1997 – pour concrete, pour concrete, pour concrete – would actually INCREASE congestion when they were completed in 20 years or so.


We don’t need no stinking models to make decisions. Do we?

Speaking of which, guess which Commonwealth doesn’t have a econometric model of its state economy?

I’ve been referred to a spread sheet model which is a modification of one made for Texas about a decade ago. I saw that one. Is there another out there?

I suggested to my Delegate to put the money in the budget for our public universities to put together a consortium to develop a very good model of our economy. It would help to analyze spending and taxing options. Didn’t happen.

Share this article


(comments below)


(comments below)


  1. Jim Bacon Avatar
    Jim Bacon

    Virginia should invest in a world-class transportation simulator for the entire state — a model that can simulate the impact of new development, road improvements, mass transit and other aspects of the transportation system. It would cost a lot to build and maintain, but it could save far more by fine-tuning the efficiency of our transportation investments.

  2. James Atticus Bowden Avatar
    James Atticus Bowden

    Hooah. And we need an econometric model of the Commonwealth.

    I know the visualization piece for the model – a system called the TouchTable (T).

  3. Larry Gross Avatar
    Larry Gross

    There are existing models – in use. They were used to help determine the impacts of the BRAC decisions in the Wash Metro Area.

    The NoVa MPO, the MWCOG/TPB uses models…


    they basically use a concept called a TAZ (transportation analysis zone) which is a boundaried area that is analyzed for trip generation per land-use – existing or comp-plan designated).

    The basic problem is, as we have seen with Loudoun County and other counties, is that many localities actually do NOT want to use such models when it comes to land-use changes or development rezoning proposals AND they don’t want VDOT using them either.

    They actually consider it “interference” for someone like VDOT to come in and tell them what impact a proposed development will have on the roads – because then the public knows who made the decision to increase traffic and congestion – and they don’t get to blame VDOT for the decision they made.

    This is why many in the development and growth community support a statewide gas tax increase…

    and not surprisingly, this is one of the principle reasons that the state is running out of maintenance money – because most of the 400 new lane miles a year that are added (and must be subsequently maintained) are subdivision and arterials affected by new development.

    Modelling..would spell trouble for some BOS that would prefer to go on approving rezones while evading responsbility for the traffic consequences of those rezones.

    They MUCH prefer for the state to be held responsible.

    Modelling would also be a way to prove/demonstrate the difference in traffic impacts between different kinds of development – especially those like mixed-use that claim to generate less traffic.

    It would be easy to use a model to show the difference that a mixed-use would have verses a traditional residential or commercial.

  4. E M Risse Avatar
    E M Risse

    Great post Larry.

    And further if the model was based on the organic components of human settlement patterns instead of arbitry municipal borders or zip codes, etc…

    Even the TAZ based system used for the COG Activity Center work and the VDOT models the late 90s “subregional” plan come close to realistic organic structures.

    They were abandoned for the reasons Larry notes and others just as perverse.

    No one, NO ONE wants to face reality and the cliff is coming closer…


  5. Anonymous Avatar

    The transportaion model MWCOG uses is hopelessly out of date. TAZ zones are smaller thatn the ususal govenment data sets so they depend on GIS data and general plan guidance. Such models are difficult, expensive, and time consuming to operate. You provide inputs, and it provides an equilibrium solution, which is an unrealistic way to view things.

    Land use activities are external to the model so you can only model one scenario at a time. Land use data is basically an input to a traffic demand model. It is blind to socioeconomic factors. It does not integrate land markets or the land model with the traffic model. In short, it is impossible to use the traffic model to predict optimal land use situations.

    An adequate model would be able to predict land use changes based on highway construction and traffic patterns/density. However land use changes very slowly and it is therefore difficult if not impossible to validate such models.

    New models are being developed that track traffic density by the density of cell phone transmissions. (What does that say about talking while driving?)

    “UrbanSim is a simulation model that incorporates the interactions between land use, transportation and public policy in the planning and analysis of urban development. The model integrates market behavior analysis and the analysis of governmental actions with land policy and infrastructure choices to maximize a realistic prediction of how the urban market will behave. The UrbanSim model was developed by Paul Waddell, and is hosted by the University of Washington.

    -The model is free!

    -The University of Washington provides strong support.

    -The model is transparent.

    -The model attempts to replicate the actions of relevant actors such as the impact of government intervention.


    -Calibration/validation, in a traditional sense, is impossible.

    -There is no real federal mandate for integrated transportation and land use modeling.

    -Randomness exists.

    The fact that calibration is impossible and the model is transparent is a good thing in that it means the model cannot be used politically. You buy off on the assumptions (the model is transparent) and then you take the results for what they are worth. If you don’t like the results, you cannot “calibrate” it.

    We should understand that a traffic model is only a tool, and an imperfect one at that. There are many factors in making recommendations on transportation decisions. Many of these factors are either not adequately or not included at all in traffic models.


  6. Larry Gross Avatar
    Larry Gross

    I’d agree.. modeling is very difficult.

    A model that cannot be validated/calibrated is not a model.

    It’s just a bunch of numbers that don’t mean much more than political opinions.

    A model that show the traffic that will be generated from land-use is the first step.

    At least with that model, you can get an idea of how many new trips will result from development.

    AND you CAN calibrate that by using traffic counters for the different land-use types.

    The trip data in use in the models can be proved by actual numbers.

    You CAN alter multiple locations in terms of changes in land use as inputs but you also want to be able to disaggregate the data – i.e. understand what changes produce what effects.

    in terms of what model to use.. i.e. Urban Sim – a good approach to model validation and calibration is to use different models and then to observe where they agree and where they differ and to use that analysis to improve the models.

    But the most basic and doeable model results would show how more intense land use would affect the transportation network.

    At the least, people would understand what roads would need to be upgraded and probably where bottlenecks would occur.

    My basic thesis is that data, information and modeling can help to identify the most optimal road solutions and to marginalize political considerations.

    If UrbanSim or other models alone or in concert with each other can help develop a better understanding of the basic issues and help the public gain some level of trust in decisions, then I view that as better than what we have right now.

    re: TAZs – that’s a whole other issue is terms of what constitutes (or should constitute) a TAZ. It appears to be that the boundaries are somewhat arbitrarily drawn…according to existing land uses…. and yes..if they are not kept up to date.. your model is crap.

    the old adage “garbage in – garbagae out” is never more true than with models…

  7. Anonymous Avatar

    “A model that cannot be validated/calibrated is not a model.”

    Not really. If you have amodel that operates over hundreds or thousands or millions of years, even if you can technically adjust or calibrate the model, paractically you cannot. We have models of global warming, but it will take a very long time to determine if they are correct.

    Before Copernicus we had a model of the sky that was completely wrong, and yet it worked well enough for navigators and travellers to use.

    There may be some truth to the idea that some of the inputs are opinion, but not all of them are. Or, the “opinion” may be factual in basis but unknown in value, so we use probabilities to model them.

    “A model that show the traffic that will be generated from land-use is the first step.”

    Not really. You have to first assume the land use, and that severely limits the conditions you can model. But we know that road construction and even changes in road usage change land use, as much as land use changes road usage. Since the two are co-dependent, the first step you mention is really a baby step. And, assuming some pattern of land use is a major place to insert what you call political opinions.

    By starting off with a model to predict traffic flow, we start off with the wrong assumption: that we can either accomodate or manage traffic flow.

    Instead, you need to start off with the idea of creating a model that predicts the optimum combination of land use and traffic flow, considering economic realities. Such a model would be greatly simplified by ignoring all the present conditions. Start with a blank map, and let the model construct it on the fly. If it turns out that the result is new urbanism, then we have an answer that is usable. If it comes up with something else, we have a known (or presumed) “best pattern”. Whenever we make changes to the existing pattern, we can see how well the proposed changes work toward or awy from the predicted “best”.

    The next step woould be to overlay the “optimal” map with our existing map, and model which changes get the most benefit soonest.

    Instead, we spend enormous resources creating input which matches existing conditions. (Think pre-Copernican astronomers.) These act as constraints on the model, and pretty much guarantee inadequate modeleing (because of the expense), and answers that are wrong – because of the inputs.

    Like you say, garbage in – garbage out.

    My thesis is that instead of trying to optimize the road solutions, you try to optimize the total solution. Then, instead of marginalizing political solutions, you use the results to illuminate and justify political solutions.

    In the end, whatever we do will result from political considerations. All we can do is hope that the models will point out when the political considerations are dead wrong.

    Suppose, for example, we had that optimum map, and overlayed it with our present condition. Then suppose it showed that rail to Dulles was far down the list on thigs to do first, or didn’t even show up. And that decision would be essentially non-political: it would have dollars and time in its favor.

    Then you would have strong incentive to tell the developers to go shove it. Then, you give them the model, a procedure to go through for justifying changes, and tell them to come back when they have a better answer.

    On the other hand, when a developer came with a plan that is high on the list, you would have a very hard time turning him down.

    But, to have the confidence to do that, you need to start off with a world class model – one where the underlying political considerations are so strong that they are generally agreed to.

    If the developers opened that model to make their desired changes, they would find that the winners have to pay off the losers before they can come out ahead. They would find it very difficult to simply modify one “political opinion” in such a way that they come out ahead.

    World class is dificult, but if Virginia can come up with such a model, then the profit from adapting the model to this and other problems, could easily pay for our transportation needs. Here is a case where knowlede and efficiency could pay off in conservation and growth. Jim Bacon is right on this one.


  8. Larry Gross Avatar
    Larry Gross

    A model that does not predict what will happen can still be called a model but it’s usefulness as a tool renders it moot.

    Why do you think the government spends millions of dollars to produce models that accurately predict where a missile will land instead of saving money and using just any model…??>?

    models that “work” have confidence levels that ARE verifiable.

    models that don’t – are not trusted and for good reason.

    Models that are not validated open the door to any old fool claiming that they have a model that proves something – based on nothing more than their assertion that it does.

    re: land-use and modeling

    You can MEASURE the trips that come from different kinds of land-use.

    This is not speculation.. there are actual traffic counters used and when you put those counters on the same kinds of land use and they consistently provide the same data – you CAN MODEL and your MODEL, in turn, CAN PREDICT future conditions and as such it CAN be trusted.

    this has absolutely nothing to do with your views or speculation as to why those traffic counts numbers are what they are and less so that you’d change the models because you don’t understand why the real-life results are what they are.

    Is your premise essentially that if you don’t understand why there are 10 trips a day that you can ignore it even if the traffic counter proves it – because you don’t agree?

    a model that “proves” winners and losers… good luck guy…

    I’d like to see your modeling technique when deciding how big and strong a bridge needed to be…

    I can see you arguing that no bridge could be built until one could prove who the winners and losers would be from building that bridge…

    Why don’t you head down to HR/TW and try this theory out of them?

    I can see it now:

    “PEOPLE, PEOPLE.. we can’t build no stinkin bridges and tunnels until we know who the winners and losers are”.

    I strongly suspect that this is NOT what Jim Bacon (or any other sane person) was advocating.

  9. Anonymous Avatar

    models that “work” have confidence levels that ARE verifiable.

    Not true. As I said navigators traveled by the stars according to models that werent true, and when verified, turned out to be wrong. But, for navigational purposes they worked.

    For example, we have models and rules of thumb that work, that produce verifiable results, and we have no idea why they work. All we know is that A is stongly correlated with B, without necessarily knowing why.

    For example, as a chemist,(this is a true story) I can start out with a known clean environment. Then introduce a known amount of contaminant, that should produce a result of 100 ppm when I sample it.

    Yet, using the best techniques, equipment, and methods available, I run 100 samples and the results are alternating zeros and 50 PPM.

    I don’t know why this happens, but that it the result. And, if I do it with a sample from another known environment that should produce a 200 PPM result, I get alternating results of zero and 100.

    This works all the way up to 1000 PPM, but then the results start changing. Now I get alternating results of 600 and 100, instead of 500 and zero.

    In the end, I get a curve that models actual environmental contamination, and it is based on measured results that look nothing like what I know are the actual levels of contamination.

    In this case, I have a model that “works” because it is calibrated, even though I have no idea why it works that way, and no way to verify it. I have some theories as to why it works that way, but as far as the original problem goes, I don’t care. I have a situation that I can prove is reproducible, which is all I need.

    And that’s just chemistry. It is easy compared to predicting what people will do. You can’t compare physical models like bridge engineering or ballistics with socioeconomic models.

    “You can MEASURE the trips that come from different kinds of land-use.

    This is not speculation.. there are actual traffic counters used and when you put those counters on the same kinds of land use and they consistently provide the same data – you CAN MODEL and your MODEL, in turn, CAN PREDICT future conditions and as such it CAN be trusted.”

    Actually this isn’t true. the handbooks that are used for trip generation are under attack for just this reason.

    Yes, we can measure such things, and there are traffic counters. But, the traffic counters work for the road guys, not the land use guys.

    Very seldom does this information get correlated with land use. Usually we resort to the handbooks and pick a number out. If you ever do go measure traffic, and then try to relate it to land use, you have a very tough problem indeed.

    I can count the traffic on the road in front of my house, but it gives me no idea what the land use at my house is. I don’t know if that traffic is there because it is tourist traffic going to the local wineries, or whether they are on their way to the national forest or Shenandoah river for an outing, marriot ranch for a trail ride, or whether it is a local person on their way to the hair dresser. the traffic count i s meaningless for either local or regional purposes.

    I could identify an area of a certain kind of development, and then count all the vehicles leaving it to generate a traffic count for that “kind” of development. Id have to cordon off the entire area, or make some measurements that would allow simplification.

    When was the last time you saw an entire neighborhood surrounded by traffic counters? One of the real problems in assessing urban development patterns is, in fact, that the predicted traffic patterns don’t match from place to place.

    In fact, on the rare occasions when people do this kind of work, they find answers that don’t always agree with the handbooks. the handbooks give numbers like 10.22 that imply a level of precision that isn’t there.

    What we find is tha “That” kind of neighborhood generates one amount of traffic in one place and a different amount someplace else. or the same neighborhood generates different trip numbers from one decade to the next.

    Anyway, you are missing the point. In order to do what you say, you count data for a neighborhood type, you have to enter the neighborhood data in the model, and you predict traffic, based on the land use given. The result will be just what you put in. It is very difficut to enter all that land use data, so you can’t very well model one scenario of land use against another. Even if you could, you would have no way to either calibrate or verify it: it would take decades.

    What I’m saying is that land use and traffic affect each other, and that is what the model should do. NOT, enter land use and watch the result. When you do that you can only model land use reflexively: that is, external to the model itself and then watch the results. Because it is so expensive, only a handful of models are tried, and thes ae likely to be politically chosen.

    So, rather than do that, you need a world class model that predicts land use and traffic iteratively and serches for the optimum total economic condition. Then you optimize on the best total system rather than just focusing on roads.

    We already agree you can’t build your way out of congestion by building roads, right? So, create a model that tells you what you should build. The current Road and traffic models are the wrong answer to the wrong problem.


    “I can see you arguing that no bridge could be built until one could prove who the winners and losers would be from building that bridge…”

    My argument is that you have to set priorities on spending. The way you set the priorities is by figuring out the best NET benefit is for the money you spend. In order to figure the NET benefit you have to consider both sides of the balance sheet – winners and losers, if you will.

    If you don’t do that, then you are committed to getting political solutions that is the solution s YOU want, instead of optimal solutions.

    With enough participation, the political process should provide what people BELIEVE is the optimal solution. Except that commonly held beliefs are not necessarily true beliefs, as Copernicus showed. And as EMR points out, we have monetized the political process to the point no one has faith in it anymore.

    We are presently arguing about what the right method of funding transportation is. We are arguing about what modes of transportation the money should be spent on, and in which areas it should be spent. And we are arguing about who comes first.

    This will continue forever because of vested interests in money and power.

    But, if we step back from all that and first come to an agreement on how to agree. I we each draw a line in the sand that says “if the evidence crosses this line, then I’m willing to concede part of my position”, then we have some hope of resolution.


    “Is your premise essentially that if you don’t understand why there are 10 trips a day that you can ignore it even if the traffic counter proves it – because you don’t agree?”

    Absolutely not. If that is what you count traffic then that traffic is what you have.

    What has to got to do with modeling? You look up in the book and the book says this kind of neighborhood generates ten trips per day. So you put that in the model.

    Surprise, the model then says ten trips per day.

    That’s not a model. A model whould tell you that if you start with x trips, then y development will take place, and x+x trips will result, which will cause y+y development, wich will result in x+x+x additional trips. Until development moves elsewhere or you spend a few billion on transit.

    What I’m telling you is that when people go out and measure, they get different numbers. What I’m telling you is that the numbers change over time, even if that neighborhood does’t. The present models are not predictive in that manner. You plug in a land use scenarion, and the model comes up with a steady state traffic solution.

    We know this is idiotic and wrong, but we do it any way because it’s the best we’ve got.

    Bacon is right – we need better. To get that we first have to concede the ones we have don’t work, despite what you say.


  10. Larry Gross Avatar
    Larry Gross

    some days.. the rope-a-dope blather factor exceeds my ability to absorb it…politely

    We don’t use models to predict land-use.

    We use models to predict the impacts that will occur from land uses.

    Your logic makes no sense.

    Using your logic.. we’d not use models to predict what water/sewer infrastructure would be needed per type of land-use …instead you’d want that water and sewer model to predict land-use..

    You CAN put traffic counters in one or more neighborhoods and gated communities to prove your point.. which would be.. what?

    that 9 or 11 trips are generated as opposed to 5 or none?

    and then what?

    you will then go tell localities that they are denied from using the data you don’t agree with – to judge, approve or deny or require mitigation ?

    What exactly is your point?

    Is it that if any process is not 100% accurate that it cannot be used and that we must tear everything down and start over?

    Bacon never advocated that we tear everything down and start over.

    You are the guy who advocates that (and perhaps one other).

    sorry to inform you – the world does not work this way.

    Traffic Generation – though not perfectly 100% accurate – IS accurate enough to determine what scope and size the transportation facilities needed to accommodate different kinds of land-use development.

    It’s not just residential.. it’s everything.. from truck depots to stadiums to hospitals and schools..

    The FHWA uses it to determine what infrastructure to build or not on the interstates…

    VDOT uses it to judge what kinds of improvements are needed at subdivision entrances…

    I could go on and on.. but the point is that you’re living on another planet on this and other issues…

    Your basic premise seems to be that if something is not dead on – that it is illegal to use it..

    join the real world guy.

    you’ve got models that while not perfect are good enough to use and always will be improved as more data and experience indicates the need to.

    We do not throw everything away and start over…

  11. Anonymous Avatar

    Look, all I’m trying to say is
    a) a ballistic equation, or a data curve fit are not a model, as I see it.

    b) A model, in my experience is s tool that is capable of comparing or altering many variables while performing goal seeking for a preferred attribute or number of attributes.

    c) the models we presently use for traffic analysis were NOT desigend for land use analysis, and to attempt to use them for that is backwards thinking, and inefficient, even if it works.

    d) Even for traffic analysis they have serious limitations.

    e) We can do much, much better.

    f) We can use the present models and improve on them, but sometimes it is better and cheaper to start over – based onthe experience you gained the last time around.


    Re- goal seeking. In my airline project I wrote a model that calculated the most efficient flight schedules. It took into account the number of potential passengers at each city location, crew replacement schedules, maintenance requirements, number of dead head trips, refueling requirements (considering the load factor, and price of fuel), and cost of overnight housing for the crews stuck away from their home base. In its essence, this was a transportation problem that considered land use (population as it presently existed) But it did not consider future changes in land use (population), because it wasn’t necessary for that application. It took the calculations right down through cash flow, balance sheet, depreciation schedule, all the way to stockholders equity. the most efficient schedule allowed lower costs, which attracted more customers, which changed the flight schedules.

    It was a complex problem, but it had a number of similarities to the auto transportation problem.


    “Your basic premise seems to be that if something is not dead on – that it is illegal to use it..”

    Not at all. But if the limitations of accuracy are plus or minus 30%, then we should say so up front and not do as the handbooks do, which is quote trip generation to two decimals, when the data don’t support it.

    I had to scrap my model several times, when I found it didn’t solve the real problem. It’s hard to throw good work away, but sometimes you just need to drop back and punt.

    The question is, when do we make land use a co-equal participant in the transportation problem, in such a way that we get land use guidance AND transportation guidance.

    If you had a blank sheet of paper, what is the best possible answer? Once you have that, how does it compare to what we have? What is the plan for getting from here to there?

    Or, do we let MWCOG continue to publish half baked analysis based on five pre-determined scenarios? And then accept them as gospel. if youare going to have a model that suggests policy, then input and access to the model should be democratic.


  12. Anonymous Avatar

    “Slow-growth advocates say the downturn will allow localities to “take a breather” and focus on improving existing communities rather than trying to keep up with the impact of booming growth. But others say a stagnant economy prevents governments from playing catch-up because the same economic slump that halted development plans is also ravaging revenue sources. “

    Where is the econometric model that can balance this problem?

    “Throughout the Washington suburbs, the economic downturn has accomplished what the slow-growth movement could not: It has slowed growth significantly.”

    “Once the dominant topic in regional politics, taming residential development has largely been eclipsed by the fiscal woes created by the slowdown.”

    And as I have asked before, how can you expect to do one, without the other?


  13. Larry Gross Avatar
    Larry Gross

    re: models – are tools

    if you don’t like the results then you can offer ways to improve the tool or you can stand off to the side and cry nah nah nah…

    NO MODEL is 100% dead-on – otherwise we’d not call them models. Get it?

    What IS the ALTERNATIVE to the use of models for things like Regional Traffic?

    I don’t even know what you are talking about with respect to land-use modeling…

    we MODEL the impacts of development that, in turn, requires facilities to serve it and we do this so that we can properly plan for the scope and the scale of the facilities that need to be built to accommodate that development.

    How would you approach this from a land-use model ?

    and don’t start into some totally off-the-wall idea that only the fringe groups talk about.

    bring something to the table that has some credence ….and a reasonable rationale…

    You and EMR would do well to use that same approach.

    Basically starting off by saying what we have does not work and that we must tear it all down and start over – is a total and complete non-starter.. non-responsive to the issues and non-productive in terms of dialog.

    Make a case for a different way as the same time you make the claim that the current way does not work and show how to get there – again – a reasonable way to get there.

    As long as cars exit and drive to and from land that has been developed – there WILL be a need to build the transport infrastructure that will be needed.

    Further we need to know whether to build one lane or four, etc.. and the way we do that is with traffic generation models.

    Have you got a better way?

    how would you do this from a land-use perspective if not with a Comp Plan or some other way of designating uses such that infrastructure planning can be based on those prospective uses?

    let’s get off the blather express and deal with things that are worth discussing…

  14. Anonymous Avatar

    “NO MODEL is 100% dead-on – otherwise we’d not call them models. Get it?”

    No, they are not. That’s why you use sensitivitiy analysis and probability. When you look at the handbook data, rather than saying the average home yields ten trips a day it should say ten trips plus or minus two trips with a probability of 90%. It will err on the lows side under these conditions and on the high side onder those conditions.

    What IS the ALTERNATIVE to the use of models for things like Regional Traffic? ….I don’t even know what you are talking about with respect to land-use modeling…

    we MODEL the impacts of development …

    Actually, we don’t. What we do is assume some development and then model the impact of that, and then try to mitigate the results of our assumed inputs. the land use decisions are all made exernal to the model.

    But land use and transportation are co-dependent. Therefore the model ought to have feed back in both directions. land use impacts trasnportation, and transportation impacts land use, otherwise we wouldn’t have all the hollering about developers getting off the land use gravy from the roads that WE supply for them.

    Several Universities are working on such models now, and one is the same as previously mentioned. So the idea is to figure out what is the best mix of land use and roadways, starting from a blank sheet. Then, starting with what we have now and how it is different from the optimum, whats the best way to move towards the optimum.

    Suppose that the right answer in some cases is to tear down some buildings. Our present models have no ability to tell you that.


    I think it is absurd to think that they (developers) get all the gravy, but that seems to be the conventional wisdom.

    The comp plan is nothing but a wish list. And since zoning is required to be based on the comp plan, the comp plan is frequently used as an excuse to not change the zoning. Since comp plans are only updated every 20 years, they just guarantee stasis.

Leave a Reply