The first thing we always explain is that TMODEL2 or Emme/2--or whichever program used--is not a transportation model. It is simply a tool which we can use to develop a traffic model much as a spreadsheet is a tool we can use to set up and calculate a profit and loss statement. Both are deterministic; that is, they will always produce the same answers given the same inputs. The biggest difference, perhaps, is that building a profit and loss statement is a known process based on fixed rules, whereas developing a traffic model is a "squishy" process based largely on some known and some assumed behavioral characteristics.

If you give ten accountants a set of data and the request to produce a profit and loss statement, you will probably get very close to the same result ten times, and the spreadsheet setup and formulas would be extremely similar. If you give ten modelers a set of data and the request to produce a traffic model, you will probably get several significantly different approaches and models which, although they all "calibrated" within acceptable standards, would have significantly different predictive capabilities.

So, what is a model and why do we accept this variation in result? A transportation or traffic model is a mathematical model which takes sets of data and applies formulas and mathematical processes using estimated or determined parameters (coefficients) in specified sequences to ultimately predict what volumes of traffic will use which streets and intersections. These data and parameters are sometimes measured or determined empirically but are also determined through experience and logic in a synthesis we sell as "professional judgement."

The process and collection of data includes descriptions of streets and intersections, especially those related to speed and capacity; the network and its connectedness, e.g., where streets are one-way; traffic controls and signs; turn restrictions and the network's connections to the "outside world." Further, we must decide on functions and parameters which reasonably predict driver perception of the influence of congestion in destination and route choice.

The data includes information on land uses such as residences and employment locations; it includes the associated travel characteristics of the people located at those land uses in terms of trip generation rates and typical trip lengths and trip purposes. It looks at people in groups in order to estimate where they might travel to satisfy their trip purposes. It is not capable of predicting these things for individuals, only probabilistically for groups. It determines routes based on "cost," be it time, or distance, or money, or combinations of them.

Why is this process so "squishy"? Why do we not have a generally applicable set of rules and procedures we can use to best predict traffic flows? The answer comes in two parts:

First, each area to be modeled has its own individual character in terms of unusual network aspects or land uses and in terms of the behavioral characteristics of its people.

"But we should be able to establish some rules which take these differences into account, shouldn't we?"

Perhaps we can, and seeking such methodologies should be one of our goals as we develop more and more models.

Second, though, is the fact that this sort of modeling is a rapidly evolving process. There are established methods for regional modeling which have been used over the past forty years. Unfortunately, these methods have become entrenched and a significant drag on development of new thinking and methods. Microcomputer modeling has changed this, and now we are in a period of transition. During any period of transition, in the trial and error development of ideas, some of the ideas will be good and some will be bad.

Because there are no established "standards" during these transitions, the peoply applying the prototype methods are largely on their own. Some will be good and others will not be.

These "squishy" periods are both unfortunate, because of occasional shoddy work, and yet terrifically hopeful, because of the positive new developments and good people which emerge. These will bring the field forward to new standards where the best of transitional efforts will be synthesized into known procedures, parameters, and tools suitable to the modeling task at hand.

All of us who work at this do our part. Let's do our best. While modeling isn't perfect, it's the best game in town.