Questions and Answers
The presentation seemed to mainly focus on researchers' various attempts to correlate increases in VMT with increases in lane miles. Are there any studies that instead focus directly on what motivates people to change their travel behavior, and then use that information to build assumptions into the transportation model? It would seem to be the logical approach to simply ask people what they would do, given hypothetical choices. Would they shop at a different place if they could get there sooner? Would they alter the hours they travel if they could save time by doing so? What actually enters into a person's decision when they are choosing a route to get to work or home from work? Are there any instances that would cause them to consider making an additional trip that they don't make now, depending on the cost of time and money to do so? (For example, coming home from work for lunch, or stopping by a certain store after work to run an errand?)
I think this is an excellent idea in theory. One interpretation of this suggestion is that you'd like to find a way to forecast the "natural" demand for travel, starting from first causes. That is, first ascertain what is the travel that people want to do of their own accord (i.e. based on their sociodemographic traits like age and presence of children and employment status and income, but also based on personalities, lifestyle aspirations, etc.), and then figure out to what extent the transportation system is inhibiting (suppressing), and/or possibly amplifying, that demand.
There are many studies of the type you mention, but there are a couple of serious challenges to incorporating those results into regional travel forecasting models. The first is the difficulty and unreliability of getting people to give hypothetical responses. Just to give a simple example: I have a theory that people *subconsciously* have a desired amount of time that they want to spend traveling for daily life activities (i.e. not counting long vacation trips). If that's true, it would really be valuable to model that, since people traveling less than their target amount will have a propensity to increase their travel if some constraint is lifted (contrary to our ubiquitous assumption that people seek to minimize their travel time). But I wouldn't have much faith in directly asking people how much time they'd like to travel per day, on average, because I don't think people articulate such a mental target to themselves, nor would they be able to articulate it to us, very clearly.
Second though, we do even have a number of studies not only of what people say they *would* do, but what they actually *do* do in response to some change in the system. And suppose, for the sake of argument, that those studies are reliable. The problem is how to incorporate the variables that actually affect people's choices, into the region-wide models. For example, my propensity to stop by the mall on the way home from work (or go there on my lunch hour) is probably related to whether shopping is recreational for me, or just a chore that I do as seldom as possible. That propensity can in fact be quantified, but it is never done so for a representative sample, incorporated as an explanatory variable into a regional planning model, forecast into the future, etc.
So I think the short answer to your question might be that our ability to identify variables relevant to travel behavior, and reactions to changes in the system, exceeds our ability to operationalize those relationships in a practical way for region-wide modeling.
The second comment I'd like to offer relates to the correlation between VMT and lane miles. Historically, the rise in VMT has been significantly greater than the rise in population and the rise in lane miles. Therefore, a 1:1 correlation between VMT and lane miles would actually reflect a decrease from the historical pattern.
Right -- we were talking further about your situation there in San Diego after the videoconference concluded. I think what you are seeing there is a lot of "background growth that Would Have Occurred Anyway". I.e., San Diego is a desirable region that is attracting considerable population and employment increases, almost independently of the transportation system. So in your particular context, it seems reasonable to me that if there IS an induced demand effect of adding new capacity, it is so far outweighed by background growth that it is even harder than usual to find. I personally expect that there still is an induced demand effect, but that it would be a smaller proportion of total growth for your district than for a more stable area. And as I tried to indicate, assigning a number to it (xx% of the growth in VMT is due to induced demand, versus yy% due to natural growth) is an uncertain proposition. However, you might be interested in following up on Kevin Heanue's attempt at doing just that for the Milwaukee area, that I referred to in my talk (Transportation Research Circular Number 481, February 1998). It would be very interesting to replicate his approach, in what he characterizes as a "relatively slow growing" region, for your far more rapidly growing region.
I understood from your presentation that the model calculates VMT as a function of lane-miles. Can the model be used or changed to calculate effects from factors (like those shown graphically in slide 19) such as technology (e.g. Intelligent Transportation Systems), policy (e.g. value/congestion pricing), and others where the number of lanes-miles remains constant?
I guess the short answer is, "not too well!"
The longer answer is, the models that my student Sangho Choo is estimating (using the graphical model shown there) are at such a high level of aggregation (US-wide) that they would be probably be too crude to capture an effect such as value pricing, that would be only selectively implemented (they could capture, more generally, an effect of increasing the price of travel overall). And the data on all the variables in the model would not be available (some would be, but not all) at a fine-level grain such as metro region (and even then, if value pricing were implemented only for selected corridors or subareas of the region, and/or only for certain times of day, it would be even harder to capture its effect). As for the effect of ITS, we just don't have enough (any) years of data with which to calibrate such an effect, so we can't really do a true forecast. But I think the basic problem in both cases is the scale mismatch -- the effects of ITS and value pricing would be expected to occur at a much lower geographic level than our model is designed for.
As for the other models I talked about, most of them would have had transportation price variables, so you could incorporate some rough scenario of, e.g. just increasing the price of travel by a certain amount that you'd attribute to value pricing. None of them have technology variables per se (again, just not enough experience to give us real data yet). But you might be able to simulate an effect of ITS by assuming it increased *effective* capacity and therefore you'd increase lane-miles accordingly, representing the "virtual" effect? But that defeats your suggestion of keeping lane-miles constant.
Probably the best way right now to approach your question is not with the kinds of models that I was talking about yesterday, but with the regional travel demand forecasting models that are currently in use, that are set up to treat system changes at the corridor or subarea level. In fact I know there has been some study of that -- using those models to simulate potential effects of ITS and value pricing (not necessarily in the same study).
Have you evaluated the model/data in terms of effect size as well as statistical significance?
Subtle point! By "stronger" I did mean in terms of size rather than statistical significance. And I agree, you *would* want to know size of effect, not just significance. I tell my students you can estimate a negligible effect (small size) very precisely (high significance) if your sample size is large enough, and conversely if you have a small sample and/or some explanatory variables are highly correlated (an effect may be large in terms of practical implications if true, but imprecisely estimated, i.e. with low statistical significance).