Activity Based Models (ABM) New Features?

2 posts / 0 new
Last post
Claire Bozic
Activity Based Models (ABM) New Features?

The Chicago Metropolitan Agency for Planning is considering a project to improve/update our Activity Based Model, ABM.  It would be useful to us to hear about any new features or useful changes other organizations have implemented in their own ABMs over the last few years.   If you have updated or changed you ABM, can you please send me a note or post here what changes you made or considered making?

One thing we are thinking about is simplifying and removing some detail that we think didn't provide much benefit but did make it run slower and be more complicated.  I think we've heard of some other agencies doing this.  

KenCervenka

Hi Claire,

The following is not specially about anything CMAP is doing, so please don't think I am picking on you in particular, instead I am picking on "modeling tradition" that seems to be heavily focused since the 1980s on "models and model outputs," and declaring to the decision-makers "well this is what the model says," rather than trying to (first) understand the plausibility and uncertainty in both near-year and far-year forecasts, and then figuring out how to present this to decision-makers in a way that helps them make, well, decisions among different investment choices.

Many MPOs are in a position similar to CMAP: whether it is trip-based, tour-based, or some in-between "hybrid, they have "a regional multimodal travel model" in operation (i.e., used for Air Quality Conformity modeling purposes), and are wondering what changes to make. If I was a travel forecasting manager for an MPO again, and regardless of whether the model takes two hours or three days to run, I would first try to learn as much as I can about how the model's predictions of travel in a pre-COVID period (let's say 2019) compare to observed travel in the same period, and if there are any big differences, to try to understand why. And then do this again when there is some level of post-COVID "new normal" that has been reached (let's say 2022). I would also want to examine the model's predictions of post-COVID changes in person and truck travel, in response to changes in the model's inputs of road/transit networks and/or zonal demographics, even though this does get somewhat speculative if it is only about confirming general plausibility in the model-based prediction of change, rather than comparing to a real-world observation of change in two different post-COVID years. Many transit agencies in the country are using FTA's STOPS model for some of their transit ridership forecasting needs (including an Incremental STOPS that has zone-to-zone transit rider survey data as an input), which opens up an opportunity to compare near-year and far-year model-based ridership predictions: not to declare that one is right so therefore the other must be wrong, but to first understand the reasons for any big differences. The same seems to be happening for lots of traffic-focused studies, where "something else" is used in a corridor of immediate "analysis of alternatives" interest.

If the investigations noted above were conducted in many different regions, in a very robust (and honest) manner, with significant amounts of real-world data available (e.g., rather than just average weekday counts, to have access to lots of directional time-of-day auto and truck counts for each link on many well-placed screen lines that are purposely designed to detect possible vehicle trip table problems; or not just average weekday transit passenger boardings by mode, but rather rider-survey derived area-to-area flows of transit passenger trips), it is not so clear to me that one big/complex multimodal travel model for an entire region will ever manage to consistently pass such tests better (or even "as well") than much simpler models that are: more heavily-grounded in real-world data; highly-focused on a particular decision-making need (e.g., perhaps a lot smaller zones in a corridor of immediate "analysis of alternatives" interest than can be "afforded" in a full-region model, plus make direct use out of OD travel information collected in a technically-defensible way); and used by ethically-influenced forecasters who understand the limitations. Plus, ALL models, regardless of their structure, will probably be way-off the actual Year 2030 observed when it comes to 30-year forecasts, which brings up some legitimate consideration about whether the forecasters and decision-makers might focus most of their attention on what is revealed from near-year forecasts. Or if long-range forecasts are still considered necessary, to at least try to "better understand" the uncertainties in the long-range forecasts by comparing to the near-year (or even a "current year") forecast.

For what it's worth, this idea of active usage of a "suite of models" can be seen by listening this week, here in the USA, to the hurricane forecasters who are predicting the storm's path, although this is just a few days in advance, rather than 30 years: "the models are showing the path as falling within this cone, but the European model has been consistently out-performing the other models in tracking previous hurricanes in this area, and shows the following path, so it would be a good idea to get out of the low-lying areas shown on this map." And then a day later: "a plane just flew over the hurricane and the models have been updated with the latest available data, and the cone has now shifted to the following." Wow, talk about a continuous planning effort, that is updated based on real-world data!

Ken C.

From: cbozic=cmap.illinois.gov@mg.tmip.org [mailto:cbozic=cmap.illinois.gov@mg.tmip.org] On Behalf Of Claire Bozic
Sent: Monday, August 24, 2020 4:17 PM
To: TMIP
Subject: [TMIP] Activity Based Models (ABM) New Features?

The Chicago Metropolitan Agency for Planning is considering a project to improve/update our Activity Based Model, ABM. It would be useful to us to hear about any new features or useful changes other organizations have implemented in their own ABMs over the last few years. If you have updated or changed you ABM, can you please send me a note or post here what changes you made or considered making?

One thing we are thinking about is simplifying and removing some detail that we think didn't provide much benefit but did make it run slower and be more complicated. I think we've heard of some other agencies doing this.
--
Full post: https://tmip.org/content/activity-based-models-abm-new-features
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13476