At least that’s the thesis of Adam Millard-Ball in an article posted on Access titled Phantom Trips.  I agree with him that if one was going to use the trip generation rates in ITE’s Trip Generation Manual as the foundation for regional traffic forecasts, you would be significantly over-estimating the traffic in the region.  You could use ITE data, but that’s akin to butchering a cow with a scalpel – the wrong tool for the job.  I think we’re in agreement on that point.
But the large flaw in Millard-Bell’s argument, at least in my experience, is his assertion that traffic engineers/transportation planners are making regional policy decisions based on Trip Generation Manual data.  The four-step regional traffic forecasting model I’m familiar with starts with data from the Census Bureau or a localized data collection effort like the Met Council’s Travel Behavior Inventory in my region, not ITE’s data set (I could be wrong on this point though – it’s possible some parts of the country are basing their four step modeling forecasts on ITE data, I just don’t know of any).
ITE’s data set  has many flaws.  In fact the limitations of the data set would be laughable to most statisticians.  But, it’s the best data set U.S. based traffic engineers have right now to determine what upgrades to roads and intersections are needed adjacent to a proposed development.  This is the forecasting and analysis done at a proposed Walmart, for instance, to figure out if a traffic signal needs to be built at the main access street to get into the Walmart parking lot.
Sure, the traffic coming to the Walmart may be folks that used to go to the Kmart a mile away, but traffic engineers are charged with getting the cars that show up at the new Walmart in and out of that parking lot safely and efficiently.  We need to build new infrastructure at that new Walmart location.  We can’t magically pick up the turn lanes from the declining Kmart driveway and move them over to the new thriving Walmart location.
I personally dealt with the extreme of this situation when I was the traffic engineer in Maple Grove and Krispy Kreme Donuts came to our region.  The Maple Grove store was one of the first in Minnesota and the queue on the opening weekend was crazy.  We did the best we could to safely accommodate that queue within the site (we made the designers significantly change their initial site design) and out on the local street the driveway accessed.  Thirteen years later, the Krispy Kreme is out of business and that building is now a local bank (with excellent circulation for  their 40 cars a day).  I was charged with handling opening year, Krispy Kreme traffic, not hypothetical bank traffic a decade later.
On a macro level, Millard-Bell correctly identifies the difference between marginal and average trips.  The four step method used for regional traffic forecasting correctly models marginal trips (at least in my region).  Where his thesis falls apart is that we need to use average trips (ITE’s, or better yet, recent/local data on similar development’s in the region) when we are designing the infrastructure on a micro-level for a proposed development.
On the better yet front, we occasionally collect our own trip generation data from similar developments in our region (here and here for examples). Â We’ve driven down the expense of doing these studies with our COUNTcam system and plan to do even more trip generation data collection in the near future.
Someone at a TRB Annual Meeting session a few years ago tried to make the inverse of this argument – that travel demand models were woefully underestimating trip generation because ITE Trip Gen was so much higher.
Truthfully, using a household survey to prove/disprove ITE is a bad argument. Household survey data, like the NHTS, measures trips from households via a survey, which has a tendency to underestimate trips (particularly short trips and non-home based trips). It does not measure trips coming into a region (a travel model would traditionally use an external station survey or Airsage data to determine the trips coming into a region). Household surveys frequently miss commercial trips as well, and a travel model will do anything from use a state average, a truck model, or even a full commodity flow model to model commercial trips. Those external trips and commercial trips are included in the ITE Trip Generation rates, but they are not included in the NHTS.
I’m not trying to defend ITE, the lack of transparency regarding data and the potential bias in observations makes it questionable, too. But for a small area, it is the best tool for a job.
Andrew – I don’t think I would agree with the inverse argument, but would need to see the full thought process. Very valid points on the household surveys – they have their own issues. What’s the practitioner to do… realize the practical limitations of the all of the different data sets and apply engineering judgment when making decisions based on forecasts. Mike
This is the author of the article here. Thanks for taking the time to engage. I know this reply is coming several years after your blog post, but I’ve only just come across it, and I wanted to respond to a couple of specific points:
1. ITE trip generation numbers ARE being used to do regional-level analyses, whether you think that is justified or not. Four-step models may be used for planning-level analyses, but ITE rates are often used to assess the regional impacts of individual developments – particularly in EIRs.
The most egregious example is site-level greenhouse gas emission analysis, which is often performed by multiplying an emission factor by the ITE trip “generation” rate, without taking into account how many trips are substituting from other locations. Another example: fees assessed to developers for the impact of their “generated” trips on state highways. A third example: air quality studies. In all cases, the appropriate analysis would consider the marginal number of trips (i.e., taking account of substitution effects). But all too often, ITE’s numbers, which are the average number of trips, are used instead. The implicit assumption: all the trips are new, and none are substituted from other locations.
2. Even at the local scale of an individual intersection, using the average number of trips per ITE overstates traffic impacts. Imagine, for example, a grocery store on a corner where a competing grocery store exists. Standard traffic impact analysis would assume that all the trips are new, and none are drawn away from the existing grocery store. That’s hard to defend in practice.
3. You don’t address the point that ITE’s numbers appear to be 50% too high when compared to travel surveys. This is an issue whether you are doing regional GHG impacts or driveway analysis. Despite the challenges with household surveys, the NHTS follows a rigorous, representative sampling plan, and has additional prompts to aid recall about short trips. The ITE data, in contrast, are a black box – who knows how representative the samples are? In particular, they are likely to overrepresent developments where traffic is perceived to be severe. In my experience, even the collection of local data come with an implicit pressure to sample high-traffic locations, rather than taking a statistically representative sample of local land uses.
Thanks again for the post, and I hope these points provide some helpful clarification.