I have never felt any fear of liability from faulty travel forecasts that I have performed or were performed with my software for public agencies. It is not that the methodology is perfect, it’s that expectations are so low. The errors inherent in travel forecasting are well documented, so buyers must beware. While I try to do everything correctly, I know that any mistakes will not be apparent for years, maybe decades, for long-range planning applications. There is little chance of doing more harm than misspending public funds. And we know governments misspend public funds all the time. Lawsuits over traditional travel forecasts are rare, and they most often involve disputes over assumptions rather than the methodology itself.
Lately, travel models have been used as part of the investment process for toll highways. The implications of a forecast failure are much higher, since real people can be hurt financially if things go awry. I have never attempted to sell QRS II to people doing toll-road studies, preferring to stay a very long distance away from irate investors. And inevitably there will be irate investors. I do not trust the power of the end-user license agreement to protect me from misuse of my software.
Applications outside of traditional planning and benefits assessment worry me. I definitely had concerns about the RADIUS model my students, colleagues and I built for WisDOT a few years ago. This model was designed for work zone construction traffic assessment. Briefly, the RADIUS model combined dynamic traffic assignment and synthetic OD table estimation for an area about the same size (but different shape) as SEWRPC’s regional model. Work zone traffic engineers do immediate-term forecasts. If the model fails, we know about it really fast. There is nowhere to hide.
Models of incidents are less susceptible to critique because it is never possible to perfectly replicate the conditions of an incident before it happens. Nonetheless, a methodology employed for good incident management planning should be quite similar to a methodology for work zone traffic planning. And that methodology differs in important ways from traditional travel forecasting.
First, the demand estimation steps of a travel forecasting model are far too crude for work-zone/incident work. While an OD table from a travel forecasting model might be a good starting point, it would need considerable refinement with ground data before it can give reliable traffic estimates. Second, the most obvious effects of work zones and incidents are long queues at freeway bottlenecks and at intersections. Static traffic assignment is inept at dealing with queuing. Volume-delay functions, such as the BPR curve, were never designed for situations where volume-to-capacity ratios exceed 1.0 for long periods of time. Traffic microsimulation or dynamic traffic assignment is required when queue formation is likely. We should be highly suspicious of static traffic assignment when applied to seriously congested conditions caused by work-zone closures.
Beyond the possibility of being identified as incompetent, there are tangible negative effects of a bad work-zone forecast. Detour routes may be chosen improperly. Message signs may be located in the wrong places. Signals both near and well away from the closure may retimed improperly or not retimed at all. Driver safety may be compromised. Multimodal options may be underutilized. And a whole lot of road users may be angry.
So you can imagine my dismay when I hear of MPO models being used for work zone forecasting or when I read academic articles implying that standard, long-range planning methods can conveniently be used for that same purpose. Such behavior and writing are both reckless and dangerous.
Alan Horowitz, Whitefish Bay, January 23, 2019