Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman’s introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called “activity-based models” by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90’s, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, “four-step” approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, “the classic trip-based [four-step] approach remains standard practice in much of the world” – quoting from Eric’s article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of “by how much?” and “is it worth the substantially greater time and effort involved?” would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including “locking into” outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

Chandra et al,

Thank you so much! This is exactly the type of work that I was looking for! I believe that I was vaguely aware of this effort at the time, but if I ever saw the paper, I don’t recall it (perhaps a sign of age?)

In any event, as you and your distinguished colleagues correctly point out, this is a comparison for one, and only one, region, and the results are at least somewhat inconclusive.

Any other such efforts out there?

With gratitude,

Bernie

From: Bhat, Chandra R
Sent: Monday, 8 May 2023 18:27
To: balpern2 ; TMIP
Subject: RE: [TMIP] Current State of Activity based Models and Next Steps

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the “ground truth” to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:

Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version , MS Word version

Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version , MS Word version

Best,

Chandra.

Chandra Bhat

University Distinguished Teaching Professor

Joe J. King Chair in Engineering

Department of Civil, Architectural and Environmental Engineering

Department of Economics (Courtesy Appointment)

The University of Texas at Austin

Austin, Texas 78712

Tel: (512) 771-9166

http://www.ce.utexas.edu/prof/bhat/home.html

Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org > On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP >
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman’s introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called “activity-based models” by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90’s, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, “four-step” approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, “the classic trip-based [four-step] approach remains standard practice in much of the world” – quoting from Eric’s article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of “by how much?” and “is it worth the substantially greater time and effort involved?” would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including “locking into” outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

Agree.

Sent from Yahoo Mail on Android

On Mon, May 8, 2023 at 11:20 AM, kgoulias wrote:
This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner wrote:

> Bernie:
>
> Good to hear from you. And I hear you about comparisons between different
> modeling paradigms. While seemingly straightforward, there are so many
> different considerations that need to be accommodated in these comparisons,
> including at what level (aggregate/disaggregate) to compare outputs and to
> ensure (as far as possible) that we are comparing apples to apples. And,
> then, we also have the constant challenge of what constitutes the "ground
> truth" to compare different model results with.
>
> Please see this peer-reviewed paper of research we undertook at UT Austin
> with other colleagues you will recognize.
>
> Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D.
> Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step
> Versus Tour-Based Models for Prediction of Travel Behavior Before and After
> Transportation System Changes," Transportation Research Record, Vol. 2303,
> pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle
> ownership, work flow distribution, and highway projects). PDF version, MS
> Word version
>
> An executive summary and the full report submitted to ODOT are available
> here:
> Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R.
> Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to
> Transportation System Changes," Executive Summary Report FHWA/OH-2011/4,
> prepared for the Ohio Department of Transportation, February 2011. PDF
> version, MS Word version
> Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R.
> Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting
> Travel Behavior Before and After Transportation System Changes - Results
> Interpretation and Recommendations," Technical Report FHWA/OH-2011/4,
> prepared for the Ohio Department of Transportation, February 2011. PDF
> version, MS Word version
> Best,
> Chandra.
>
> Chandra Bhat
> University Distinguished Teaching Professor
> Joe J. King Chair in Engineering
> Department of Civil, Architectural and Environmental Engineering
> Department of Economics (Courtesy Appointment)
> The University of Texas at Austin
> Austin, Texas 78712
> Tel: (512) 771-9166
> http://www.ce.utexas.edu/prof/bhat/home.html
> Web of Science Researcher:
> https://publons.com/researcher/3178803/chandra-bhat/
>
> I live and work on occupied Indigenous land and acknowledge the Carrizo &
> Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache,
> Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and
> Indigenous Peoples and communities who have been or have become a part of
> the lands and territories known today as Texas.
>
> From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
> Sent: Monday, May 8, 2023 10:07 AM
> To: TMIP
> Subject: Re: [TMIP] Current State of Activity based Models and Next Steps
>
> Krishnan (and Eric), thank you very much for this.
>
> As some of you receiving this email know, I was quite active in the travel
> demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's
> introductory course at MIT in 1975, through my retirement last year (47
> years later). Throughout that period, I never had the opportunity to
> develop, maintain, or operate a tour-based model (which Eric correctly
> points out is the proper name for models currently used in practice that
> are often called "activity-based models" by many people in the field).
>
> I maintained a keen interest in travel demand models, however, and
> rejoiced at the work of John (Bowman) and Moshe when it was presented in
> the mid 90's, as I immediately recognized it as an essential breakthrough
> in developing an appropriate mathematical structure for such models.
>
> One thing that I always found frustrating, however, was that the practice
> adopted as an axiom that because of their theoretically more solid basis in
> traveler behavior, such models HAD to be superior to the older, "four-step"
> approach to travel modeling (even including many, many enhancements over
> the original formulation of such models). Yet to date, I have not come
> across a single straight-up comparison of the performance of tour-based vs.
> four-step models, as measured in appropriate ways. Given that tour-based
> models are, in fact, substantially more complicated, difficult to develop,
> maintain and use, and computationally inefficient relative to four-step
> models (as reflected by the fact that some 30 years after research on
> activity-based modeling began in earnest, "the classic trip-based
> [four-step] approach remains standard practice in much of the world" -
> quoting from Eric's article), you would think that by now several such
> comparisons would have been made and the findings published. Even if the
> tour-based approach were demonstrated to provide superior forecasts, the
> questions of "by how much?" and "is it worth the substantially greater time
> and effort involved?" would still need to be addressed and discussed.
>
> If anyone reading this is aware of any such work that has been published
> (and, preferably, peer-reviewed), I would be extremely happy if I could be
> told how to find it.
>
> Thank you in advance,
>
> Bernard (Bernie) Alpern
>
> Jerusalem, Israel
>
> bernard.alpern@gmail.com
>
> From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
> Sent: Monday, 8 May 2023 14:40
> To: TMIP
> Subject: [TMIP] Current State of Activity based Models and Next Steps
>
> As a practitioner I found Eric Miller's editorial on the current state of
> ABMs in Transport Reviews to be thought provoking and interesting & thought
> it might be of interest to this group. Maybe there will be some discussions
> about the editorial and the proposed solutions at next month's conference
> on Innovations in Travel Analysis & Planning. The proposed solutions seem,
> to me personally, more academic in nature than something the industry can
> adopt. Since I am unsure of Transport Review's restrictions on sharing
> widely, I will share the link to the editorial (
> https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list
> he
> might be willing to share it directly.
>
> The abstract is below.
>
> Despite the clear theoretical advantages of activity-based models of travel
> behaviour relative to trip-based models, adoption of such models in
> planning practice has been slow. This editorial discusses some reasons
> underlying this fact, including "locking into" outmoded model structures
> and software and challenges in translating research advances into practice.
> It argues for more widespread adoption of an activity-scheduling approach
> to the problem and identifies a number of key areas requiring new research
> in order to improve the operational capabilities of these models.
>
> Krishnan
>
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

>--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

By the way, following up on time-use related behaviors in a changing landscape, we had some fun (and also had some interesting insights about human behavior) as we put this paper together –

Dannemiller, K.A., K.E. Asmussen, A. Mondal, and C.R. Bhat (2023), "Autonomous Vehicle Impacts on Travel-Based Activity and Activity-Based Travel," Transportation Research Part C, Vol. 150, 104107 (Keywords: travel-based activity, activity-based travel, multi-tasking, autonomous vehicles, ranked probit, psycho-social constructs).

Thought some of you may also enjoy it. Here is a brief extract from the abstract of the paper –

“Our study indicates that “productive use of time” is not necessarily always tied with activities such as work and study; rather, being able to partake in relatively chill activities (such as sleeping, relaxing, and gazing out the window) is also considered as good use of time. This suggests caution in the interpretation of what are traditionally referred to as “productive” activities and also a need for scholarly restraint in the use of the label “multi-tasking” to exclusively refer to non-passive activities. We suggest that the field move away from subjective/ambiguous terms such as multitasking and “productive” activities, and adopt the more neutral label of “travel-based activity” (TBA). The results also support the notion that the option of opening up travel to pursue work/study activities may itself be engendering stress in individuals; that is, as the option to pursue “non-chill” activities increases in an AV environment, that itself may produce angst in individuals and lead to less enjoyment in travel. This also highlights a need to examine TBAs in the broader context of emotional well-being and quality of life. Indeed, AVs may further erode into our time of tuning-out from the “chatter” of routine life and make it less possible to partake in “calm and mindless” activities. Finally, our study cautions against the use of simple and uniform (across individuals) value of travel time savings (VTTS) factor modifications to study AV impacts on activity-based travel (ABT).”

Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bhat, Chandra R
Sent: Monday, May 8, 2023 2:58 PM
To: Kostas Goulias
Cc: TMIP
Subject: RE: [TMIP] Current State of Activity based Models and Next Steps

Yes, Kostas. The work I referenced was done some time back, but, as Bernie indicated, there has been little empirical comparisons between activity models and trip-based models. Yes, activity models do have much more of a theoretical basis, and should provide important insights at disaggregate levels. Besides, time-use, time-space prisms, and complete activity-travel patterns becomes even more central as we go into an era of increasing virtual participations, and emerging mobility options (including automated vehicles). More the reason to put the emphasis on activities.

My response was to the specific issue of comparing trip-based versus tour/activity-based approaches. Easier said than done, because of so many confounding factors and the basic issue of what constitutes “ground truth”. I do recall we put in considerable thought in the comparisons we made for the ODOT project to compare apples to apples.

If we are talking about ABM implementations in general, in addition to the Activitysim stuff that Kostas references, Kostas led a team (including Ram and myself) for a Doha (Qatar) ABM project before COVID (https://mot.gov.qa/sites/default/files/LT/QTMV/Executive%20Summary%20-%2...), and we also completed a NYMTC ABM project with Tom Rossi, Ram Pendyala, myself, and many other colleagues in 2021 (https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...).

Best to all,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Kostas Goulias >
Sent: Monday, May 8, 2023 12:54 PM
To: Bhat, Chandra R >
Cc: TMIP >
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner > wrote:

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the "ground truth" to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called "activity-based models" by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90's, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, "four-step" approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, "the classic trip-based [four-step] approach remains standard practice in much of the world" - quoting from Eric's article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of "by how much?" and "is it worth the substantially greater time and effort involved?" would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including "locking into" outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

Thanks to Chandra’s post I have updated myself on the New York Best Practice Model (NYBPM) https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...

Their website describes it as follows: The New York Best Practice Model (NYBPM) is extremely complex, incorporating millions of households, a dozen travel modes and millions of journeys with multiple trip purposes and over several time periods.

My first reaction was to ask: Why?

I imagine the trivial answer to that question is: “because the real mobility system in the New York area is very complex, involving the participation of millions of people, dozen ways of moving at different times of the day (and year) and for different purposes”.

Really? There are two main reasons for developing and using transport/mobility models: Research and Delivering intervention advice. The answer above is perfectly valid for research applications; it is not clear to me this answer it is equally valid to support decision making.

Most transport interventions have costs and impacts over many years; therefore, our models should be able to deliver forecasts for 20, 30 or more years. Any investment or intervention decision is taken in the context of future uncertainty (and a bit of present uncertainty as our data is always imperfect). As modellers and planners we cannot provide useful advice to decision-makers pretending than we can forecast the future; that would be unethical and foolish.

Two main sources of uncertainty plague our forecasts: Model uncertainty and External uncertainty. Model uncertainty is a result of our imperfect understanding of human behaviour and the paucity of the data (small sample, noise to signal ratio, etc.) available to feed our models. Eric’s paper list some of the areas where we fall far short of a reliable understanding of human behaviour. Moreover, as most is cross-section data we certainly do not understand how humans change their mind and values over time. More research offers the promise of reducing this type of uncertainty but the ideal of closing the gap is an hallucination based on a mechanical view of what it is to be human.

External uncertainty has many sources: the future evolution of the economy, migration, energy prices, globalisation, inequality. Even our own creativity is a source of uncertainty: the adoption and acceptability of virtual presence, the advent of new mobility services including the long-awaited autonomous vehicle, changes in government priorities, etc. Not to mention geopolitical friction, pandemics and global warming.

Uncertainty raises the question of the level of granularity appropriate to deliver sound advice under conditions of uncertainty. Would a simpler, more agile, model provide better support to explore alternative futures than a more complex but behaviourally richer model? But, how simple is good enough? Each type of model will require a different level of assumptions about future input data and each will pose different demands on limited resources.

It would be interesting to find out if the NYBPM is able to provide a reasonable representation of present (post pandemic) conditions so that it could be used to support decisions today. I do not know the answer to this question as the model is based on 2012 data and apparently produces forecasts only for 2040 and 2055 (I am only using the website as a reference; I have no experience in this model).

There is very little research on what would be the best modelling approach to support decision making under uncertainty. This is an uncomfortable gap in our professional practice as we may be using the wrong tool for this purpose.

Luis Willumsen

……………………………………………

Luis Willumsen

Director

Willumsen Advisory Services

Nommon Solutions and Technologies

London & Madrid

M: +44 7979 53 88 45

www.nommon.es

…………………………………………..

From: on behalf of Chandra Bhat
Date: Monday, 8 May 2023 at 21:04
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Yes, Kostas. The work I referenced was done some time back, but, as Bernie indicated, there has been little empirical comparisons between activity models and trip-based models. Yes, activity models do have much more of a theoretical basis, and should provide important insights at disaggregate levels. Besides, time-use, time-space prisms, and complete activity-travel patterns becomes even more central as we go into an era of increasing virtual participations, and emerging mobility options (including automated vehicles). More the reason to put the emphasis on activities.

My response was to the specific issue of comparing trip-based versus tour/activity-based approaches. Easier said than done, because of so many confounding factors and the basic issue of what constitutes “ground truth”. I do recall we put in considerable thought in the comparisons we made for the ODOT project to compare apples to apples.

If we are talking about ABM implementations in general, in addition to the Activitysim stuff that Kostas references, Kostas led a team (including Ram and myself) for a Doha (Qatar) ABM project before COVID (https://mot.gov.qa/sites/default/files/LT/QTMV/Executive%20Summary%20-%2...), and we also completed a NYMTC ABM project with Tom Rossi, Ram Pendyala, myself, and many other colleagues in 2021 (https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...).

Best to all,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Kostas Goulias
Sent: Monday, May 8, 2023 12:54 PM
To: Bhat, Chandra R
Cc: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner > wrote:

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the "ground truth" to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called "activity-based models" by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90's, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, "four-step" approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, "the classic trip-based [four-step] approach remains standard practice in much of the world" - quoting from Eric's article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of "by how much?" and "is it worth the substantially greater time and effort involved?" would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including "locking into" outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

All:

And thanks to Luis’s post, I felt a need to add a little more to this discussion.

First, the word “complex” is unfortunately used sometimes in our field as a mark of pride and sophistication. The best model is one that is as simple as possible and still “gets the job done”. It is in simple-ness that the power and elegance of an approach lies. I believe Luis and I are in full agreement there.

Having said that, as modelers and planners, it is EXACTLY our job to forecast the future, while also exercising a good dose of humility in the uncertainty surrounding the forecasts we make. We cannot provide useful advice to decision makers if we tell them that we CANNOT forecast the future at all. That would be foolish, because decision-makers have no incentive then to get input from us, and it would be unethical because we would be wasting our collective resources in a profession that does no good for society.

So, now, moving onto forecasting the future. I have written about this in some earlier threads. Our models are not for “predicting” today. We do not need models for that. The idea of a model is to be able to see how patterns may change (1) when technology/infrastructure/travel conditions/demographics in the future are quite different from what exist today, and/or (2) when we bring in some kind of a “shock” in the form of a policy measure that changes the travel environment in the short term. Point (1) above brings in the issue of the stability of behavioral processes (essentially, temporal transferability in the context of stability of coefficients in our models), while point (2) brings in the issue of causality in relationships and how well the behavioral process has been captured in the first place. So, let’s look at these two issues. The first assumption of stability of behavioral processes has been the mainstay of all our models (even if models are based on longitudinal data, somewhere down the line trend changes are assumed to be stable, or assumptions are made about trend extrapolations into the future). Of course, one can (and should) question this temporal transferability assumption due to not completely being able to capture the vicissitudes of human behavior and/or changes in that behavior over time (what Luis terms as model uncertainty). But, intuitively and also based on extensive reviews of studies of temporal transferability, there is little doubt that when we are able to capture causal directions and behavioral processes “better” (the second issue identified above), this leads to improved temporal transferability (the first issue). So, the two are closely interlinked. Of course, given that we are estimating models based on data at different points in space (even if all within the same region), we are also assuming spatial stability in model parameters. Again, the validity of such an assumption improves if we better capture the effects (and causal directions) of built environment and other spatially-varying variables in the behavioral process.

Basically, then, I would not put the kind of schism between “research on behavioral processes” and “decision-making under uncertainty” as some others may make it out to be. These are not two different camps. Besides, just because there may be substantial “external uncertainty” (as Luis puts it) does not justify putting less emphasis on addressing “model uncertainty”. With less model uncertainty, we can at least better capture possible future states through the development of scenarios to introduce “external uncertainty”. So, I hope we do not think of “behaviorally richer models” as not being needed for exploring alternative futures just because of the tremendous uncertainty upon us. In fact, when the noise gets “louder”, that is the time to use all our tools to extract out the signal. That is the mark of a good modeler. Anyone literally can extract out the signal when there is little noise. Of course, we also should not think of “behaviorally rich models” as having to be necessarily monstrous. The trick, I believe, is to strive toward simple-ness while also having our models be behaviorally rich. The two are not necessarily mutually exclusive, though our field may have made it feel that way (besides, conceptual simple-ness is different from mechanical simple-ness; the trip-based approaches may be mechanically simple, but definitively not conceptually simple given the underhanded way of going back and forth between origin/destinations and productions/attractions, but that for another discussion).

On last thing that I alluded to in the first paragraph about exercising a good dose of humility. In a recent article focused more on data collection considerations in a fast evolving transportation landscape (see Bhat, 2022 referenced below), I had these comments:

“The basic point is that when analyzing human behavior in fast-evolving technological, humanitarian, and environmental contexts, we cannot be sure which data collection approach would provide better insights about activity-travel characteristics than other approaches. So, rather than stick to any self-assured hubris regarding what is the right way to collect data about future behavior, a better approach, in this author’s opinion, would be to readily acknowledge what is unknown, and embrace humility as we pursue the scientific path.” , and

“In closing, none of the many data collection approaches singularly are a panacea or necessarily a better representation of future behavior in a fast-evolving landscape. But the combined insights from any such single-data studies, as well as those from multi-data studies that combine data from multiple of these sources, can be beneficial in getting a reasonable sense of what the future holds. Of course, this does not absolve us of our responsibility to pursue the most rigorous methodical designs within any data collection approach (or combination of approaches) employed. But, as scholars, rather than arguing over what kind of singular data collection approach would provide the best insights into future behavior, let’s channel our time more constructively on addressing the following question: “Given we have so many different types of data collection approaches at our disposal, how best might we harness the full potential of the different approaches, either individually or in combination, to gather insights into future activity and travel behaviors?.”

Bhat, C.R. (2022), "Data Collection Methods and Activity-Travel Behavior Analysis in a Fast Evolving Technological Future," ITE Journal, Vol. 92, No. 10, pp. 29-33 (October 2022 Issue) (Keywords: travel-based activity, activity-travel behavior, revealed behavior, stated intentions, autonomous vehicles). PDF version, MS Word version

Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Luis Willumsen
Sent: Tuesday, May 9, 2023 4:00 AM
To: Bhat, Chandra R ; TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Thanks to Chandra’s post I have updated myself on the New York Best Practice Model (NYBPM) https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...

Their website describes it as follows: The New York Best Practice Model (NYBPM) is extremely complex, incorporating millions of households, a dozen travel modes and millions of journeys with multiple trip purposes and over several time periods.

My first reaction was to ask: Why?

I imagine the trivial answer to that question is: “because the real mobility system in the New York area is very complex, involving the participation of millions of people, dozen ways of moving at different times of the day (and year) and for different purposes”.

Really? There are two main reasons for developing and using transport/mobility models: Research and Delivering intervention advice. The answer above is perfectly valid for research applications; it is not clear to me this answer it is equally valid to support decision making.

Most transport interventions have costs and impacts over many years; therefore, our models should be able to deliver forecasts for 20, 30 or more years. Any investment or intervention decision is taken in the context of future uncertainty (and a bit of present uncertainty as our data is always imperfect). As modellers and planners we cannot provide useful advice to decision-makers pretending than we can forecast the future; that would be unethical and foolish.

Two main sources of uncertainty plague our forecasts: Model uncertainty and External uncertainty. Model uncertainty is a result of our imperfect understanding of human behaviour and the paucity of the data (small sample, noise to signal ratio, etc.) available to feed our models. Eric’s paper list some of the areas where we fall far short of a reliable understanding of human behaviour. Moreover, as most is cross-section data we certainly do not understand how humans change their mind and values over time. More research offers the promise of reducing this type of uncertainty but the ideal of closing the gap is an hallucination based on a mechanical view of what it is to be human.

External uncertainty has many sources: the future evolution of the economy, migration, energy prices, globalisation, inequality. Even our own creativity is a source of uncertainty: the adoption and acceptability of virtual presence, the advent of new mobility services including the long-awaited autonomous vehicle, changes in government priorities, etc. Not to mention geopolitical friction, pandemics and global warming.

Uncertainty raises the question of the level of granularity appropriate to deliver sound advice under conditions of uncertainty. Would a simpler, more agile, model provide better support to explore alternative futures than a more complex but behaviourally richer model? But, how simple is good enough? Each type of model will require a different level of assumptions about future input data and each will pose different demands on limited resources.

It would be interesting to find out if the NYBPM is able to provide a reasonable representation of present (post pandemic) conditions so that it could be used to support decisions today. I do not know the answer to this question as the model is based on 2012 data and apparently produces forecasts only for 2040 and 2055 (I am only using the website as a reference; I have no experience in this model).

There is very little research on what would be the best modelling approach to support decision making under uncertainty. This is an uncomfortable gap in our professional practice as we may be using the wrong tool for this purpose.

Luis Willumsen

……………………………………………
Luis Willumsen
Director
Willumsen Advisory Services
Nommon Solutions and Technologies
London & Madrid
M: +44 7979 53 88 45

www.nommon.es
…………………………………………..

From: > on behalf of Chandra Bhat >
Date: Monday, 8 May 2023 at 21:04
To: TMIP >
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Yes, Kostas. The work I referenced was done some time back, but, as Bernie indicated, there has been little empirical comparisons between activity models and trip-based models. Yes, activity models do have much more of a theoretical basis, and should provide important insights at disaggregate levels. Besides, time-use, time-space prisms, and complete activity-travel patterns becomes even more central as we go into an era of increasing virtual participations, and emerging mobility options (including automated vehicles). More the reason to put the emphasis on activities.

My response was to the specific issue of comparing trip-based versus tour/activity-based approaches. Easier said than done, because of so many confounding factors and the basic issue of what constitutes “ground truth”. I do recall we put in considerable thought in the comparisons we made for the ODOT project to compare apples to apples.

If we are talking about ABM implementations in general, in addition to the Activitysim stuff that Kostas references, Kostas led a team (including Ram and myself) for a Doha (Qatar) ABM project before COVID (https://mot.gov.qa/sites/default/files/LT/QTMV/Executive%20Summary%20-%2...), and we also completed a NYMTC ABM project with Tom Rossi, Ram Pendyala, myself, and many other colleagues in 2021 (https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...).

Best to all,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Kostas Goulias
Sent: Monday, May 8, 2023 12:54 PM
To: Bhat, Chandra R
Cc: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner > wrote:

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the "ground truth" to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called "activity-based models" by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90's, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, "four-step" approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, "the classic trip-based [four-step] approach remains standard practice in much of the world" - quoting from Eric's article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of "by how much?" and "is it worth the substantially greater time and effort involved?" would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including "locking into" outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

Okay. Sorry for the multiple e-mails. I think Juan de Dios already posted Luis’s and my messages. But, for easier readability, I am following up with my response on the fresh (re-) post from Luis. Thanks for this discussion, Pilo.

Best to all.

**********************************************************************
All:

And thanks to Luis’s post, I felt a need to add a little more to this discussion.

First, the word “complex” is unfortunately used sometimes in our field as a mark of pride and sophistication. The best model is one that is as simple as possible and still “gets the job done”. It is in simple-ness that the power and elegance of an approach lies. I believe Luis and I are in full agreement there.

Having said that, as modelers and planners, it is EXACTLY our job to forecast the future, while also exercising a good dose of humility in the uncertainty surrounding the forecasts we make. We cannot provide useful advice to decision makers if we tell them that we CANNOT forecast the future at all. That would be foolish, because decision-makers have no incentive then to get input from us, and it would be unethical because we would be wasting our collective resources in a profession that does no good for society.

So, now, moving onto forecasting the future. I have written about this in some earlier threads. Our models are not for “predicting” today. We do not need models for that. The idea of a model is to be able to see how patterns may change (1) when technology/infrastructure/travel conditions/demographics in the future are quite different from what exist today, and/or (2) when we bring in some kind of a “shock” in the form of a policy measure that changes the travel environment in the short term. Point (1) above brings in the issue of the stability of behavioral processes (essentially, temporal transferability in the context of stability of coefficients in our models), while point (2) brings in the issue of causality in relationships and how well the behavioral process has been captured in the first place. So, let’s look at these two issues. The first assumption of stability of behavioral processes has been the mainstay of all our models (even if models are based on longitudinal data, somewhere down the line trend changes are assumed to be stable, or assumptions are made about trend extrapolations into the future). Of course, one can (and should) question this temporal transferability assumption due to not completely being able to capture the vicissitudes of human behavior and/or changes in that behavior over time (what Luis terms as model uncertainty). But, intuitively and also based on extensive reviews of studies of temporal transferability, there is little doubt that when we are able to capture causal directions and behavioral processes “better” (the second issue identified above), this leads to improved temporal transferability (the first issue). So, the two are closely interlinked. Of course, given that we are estimating models based on data at different points in space (even if all within the same region), we are also assuming spatial stability in model parameters. Again, the validity of such an assumption improves if we better capture the effects (and causal directions) of built environment and other spatially-varying variables in the behavioral process.

Basically, then, I would not put the kind of schism between “research on behavioral processes” and “decision-making under uncertainty” as some others may make it out to be. These are not two different camps. Besides, just because there may be substantial “external uncertainty” (as Luis puts it) does not justify putting less emphasis on addressing “model uncertainty”. With less model uncertainty, we can at least better capture possible future states through the development of scenarios to introduce “external uncertainty”. So, I hope we do not think of “behaviorally richer models” as not being needed for exploring alternative futures just because of the tremendous uncertainty upon us. In fact, when the noise gets “louder”, that is the time to use all our tools to extract out the signal. That is the mark of a good modeler. Anyone literally can extract out the signal when there is little noise. Of course, we also should not think of “behaviorally rich models” as having to be necessarily monstrous. The trick, I believe, is to strive toward simple-ness while also having our models be behaviorally rich. The two are not necessarily mutually exclusive, though our field may have made it feel that way (besides, conceptual simple-ness is different from mechanical simple-ness; the trip-based approaches may be mechanically simple, but definitively not conceptually simple given the underhanded way of going back and forth between origin/destinations and productions/attractions, but that for another discussion).

On last thing that I alluded to in the first paragraph about exercising a good dose of humility. In a recent article focused more on data collection considerations in a fast evolving transportation landscape (see Bhat, 2022 referenced below), I had these comments:

“The basic point is that when analyzing human behavior in fast-evolving technological, humanitarian, and environmental contexts, we cannot be sure which data collection approach would provide better insights about activity-travel characteristics than other approaches. So, rather than stick to any self-assured hubris regarding what is the right way to collect data about future behavior, a better approach, in this author’s opinion, would be to readily acknowledge what is unknown, and embrace humility as we pursue the scientific path.” , and

“In closing, none of the many data collection approaches singularly are a panacea or necessarily a better representation of future behavior in a fast-evolving landscape. But the combined insights from any such single-data studies, as well as those from multi-data studies that combine data from multiple of these sources, can be beneficial in getting a reasonable sense of what the future holds. Of course, this does not absolve us of our responsibility to pursue the most rigorous methodical designs within any data collection approach (or combination of approaches) employed. But, as scholars, rather than arguing over what kind of singular data collection approach would provide the best insights into future behavior, let’s channel our time more constructively on addressing the following question: “Given we have so many different types of data collection approaches at our disposal, how best might we harness the full potential of the different approaches, either individually or in combination, to gather insights into future activity and travel behaviors?.”

Bhat, C.R. (2022), "Data Collection Methods and Activity-Travel Behavior Analysis in a Fast Evolving Technological Future," ITE Journal, Vol. 92, No. 10, pp. 29-33 (October 2022 Issue) (Keywords: travel-based activity, activity-travel behavior, revealed behavior, stated intentions, autonomous vehicles). PDF version, MS Word version

Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: luis=luiswillumsen.com@mg.tmip.org On Behalf Of Pilo Willumsen
Sent: Wednesday, May 10, 2023 6:28 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Thanks to Chandra’s post I have updated myself on the New York Best Practice Model (NYBPM) https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...

Their website describes it as follows: The New York Best Practice Model (NYBPM) is extremely complex, incorporating millions of households, a dozen travel modes and millions of journeys with multiple trip purposes and over several time periods.

My first reaction was to ask: Why?

I imagine the trivial answer to that question is: “because the real mobility system in the New York area is very complex, involving the participation of millions of people, dozen ways of moving at different times of the day (and year) and for different purposes”.

Really? There are two main reasons for developing and using transport/mobility models: Research and Delivering intervention advice. The answer above is perfectly valid for research applications; it is not clear to me this answer it is equally valid to support decision making.

Most transport interventions have costs and impacts over many years; therefore, our models should be able to deliver forecasts for 20, 30 or more years. Any investment or intervention decision is taken in the context of future uncertainty (and a bit of present uncertainty as our data is always imperfect). As modellers and planners we cannot provide useful advice to decision-makers pretending than we can forecast the future; that would be unethical and foolish.

Two main sources of uncertainty plague our forecasts: Model uncertainty and External uncertainty. Model uncertainty is a result of our imperfect understanding of human behaviour and the paucity of the data (small sample, noise to signal ratio, etc.) available to feed our models. Eric’s paper list some of the areas where we fall far short of a reliable understanding of human behaviour. Moreover, as most is cross-section data we certainly do not understand how humans change their mind and values over time. More research offers the promise of reducing this type of uncertainty but the ideal of closing the gap is an hallucination based on a mechanical view of what it is to be human.

External uncertainty has many sources: the future evolution of the economy, migration, energy prices, globalisation, inequality. Even our own creativity is a source of uncertainty: the adoption and acceptability of virtual presence, the advent of new mobility services including the long-awaited autonomous vehicle, changes in government priorities, etc. Not to mention geopolitical friction, pandemics and global warming.

Uncertainty raises the question of the level of granularity appropriate to deliver sound advice under conditions of uncertainty. Would a simpler, more agile, model provide better support to explore alternative futures than a more complex but behaviourally richer model? But, how simple is good enough? Each type of model will require a different level of assumptions about future input data and each will pose different demands on limited resources.

It would be interesting to find out if the NYBPM is able to provide a reasonable representation of present (post pandemic) conditions so that it could be used to support decisions today. I do not know the answer to this question as the model is based on 2012 data and apparently produces forecasts only for 2040 and 2055 (I am only using the website as a reference; I have no experience in this model).

There is very little research on what would be the best modelling approach to support decision making under uncertainty. This is an uncomfortable gap in our professional practice as we may be using the wrong tool for this purpose.

Luis Willumsen

……………………………………………

Luis Willumsen

Director

Willumsen Advisory Services

Nommon Solutions and Technologies

London & Madrid

M: +44 7979 53 88 45

www.nommon.es

…………………………………………..

From: on behalf of Chandra Bhat
Date: Monday, 8 May 2023 at 21:04
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Yes, Kostas. The work I referenced was done some time back, but, as Bernie indicated, there has been little empirical comparisons between activity models and trip-based models. Yes, activity models do have much more of a theoretical basis, and should provide important insights at disaggregate levels. Besides, time-use, time-space prisms, and complete activity-travel patterns becomes even more central as we go into an era of increasing virtual participations, and emerging mobility options (including automated vehicles). More the reason to put the emphasis on activities.

My response was to the specific issue of comparing trip-based versus tour/activity-based approaches. Easier said than done, because of so many confounding factors and the basic issue of what constitutes “ground truth”. I do recall we put in considerable thought in the comparisons we made for the ODOT project to compare apples to apples.

If we are talking about ABM implementations in general, in addition to the Activitysim stuff that Kostas references, Kostas led a team (including Ram and myself) for a Doha (Qatar) ABM project before COVID (https://mot.gov.qa/sites/default/files/LT/QTMV/Executive%20Summary%20-%2...), and we also completed a NYMTC ABM project with Tom Rossi, Ram Pendyala, myself, and many other colleagues in 2021 (https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...).

Best to all,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Kostas Goulias
Sent: Monday, May 8, 2023 12:54 PM
To: Bhat, Chandra R
Cc: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner > wrote:

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the "ground truth" to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called "activity-based models" by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90's, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, "four-step" approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, "the classic trip-based [four-step] approach remains standard practice in much of the world" - quoting from Eric's article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of "by how much?" and "is it worth the substantially greater time and effort involved?" would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including "locking into" outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

Me commenting on a post by Willumsen! I am humbled beyond belief! ME2 forever! I apologize that this is an email response, but me posting to TMIP? Really? Really? (Reading emails, yes. Getting some emails from TMIP [ I don’t know how, but I do] And before someone tells me how easy it is to post; I am stuck in my ways. What about old dogs and new tricks don’t you understand)

I could not agree more. Complexity may be beautiful, and it certainly is complex. It thus may be inaccessible except to those who understand that complexity. But that does not mean that it is also correct. Sometimes less complex, and even ugly, is correct. When Johannes Kepler suggested that the planets move on ellipses rather than circles, that struck his contemporaries as too ugly to be true. Kepler himself had developed a beautiful theory of complex circular motion that he abandoned in favor of the ugly theory of ellipses. But the ugly, less complex theory was correct.

ABMs are an effort to develop a trip table. Static Traffic Assignments, STAs, and Dynamic Traffic Assignments, DTAs, are methods to assign that trip table to a network. Together, the development of a trip table of demand and assigning that demand to a network, that is a Travel Demand Model.

There is uncertainty. ABMs explicitly include random variables, typically in the form of a Pseudo Random Number Generator, because there is a desire on the part of the users of the outputs of TDMs for a single, replicable answer. But Pseudo may be a fancy word, just as viaduct is a fancy word, so we say, “Why A Duck”, Pseudo, and act like ABMs are not random.

The more traditional methods of developing trip tables, Frataring, Trip Distribution, Logit Mode Split, and even Maximum Entropy Matrix Estimation, ME2, (which is known in the United States by the propriety eponym ODME) are all random models, they all solve for the most probable outcome among many, not a single outcome, i.e., the mesostate with the greatest number of microstates. They too are thus random, even if it is less obvious.

Life is random, uncertain, stuff happens, so it is not surprising that trip tables of demand are also random. Trying to pretend that it is not random does not change that fact.

Uncertain does NOT mean unknown. Las Vegas casinos are famous for random games, but I am fairly certain that casino operators know what they are doing. I hope that the NYTMC BPM operators know what they are doing as well. (I selfishly agree with their choice of consultants advising them! But the comments above are my own and should not be used against that consulting firm)

.

From: luis=luiswillumsen.com@mg.tmip.org On Behalf Of Pilo Willumsen
Sent: Wednesday, May 10, 2023 7:28 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Thanks to Chandra’s post I have updated myself on the New York Best Practice Model (NYBPM) https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...

Their website describes it as follows: The New York Best Practice Model (NYBPM) is extremely complex, incorporating millions of households, a dozen travel modes and millions of journeys with multiple trip purposes and over several time periods.

My first reaction was to ask: Why?

I imagine the trivial answer to that question is: “because the real mobility system in the New York area is very complex, involving the participation of millions of people, dozen ways of moving at different times of the day (and year) and for different purposes”.

Really? There are two main reasons for developing and using transport/mobility models: Research and Delivering intervention advice. The answer above is perfectly valid for research applications; it is not clear to me this answer it is equally valid to support decision making.

Most transport interventions have costs and impacts over many years; therefore, our models should be able to deliver forecasts for 20, 30 or more years. Any investment or intervention decision is taken in the context of future uncertainty (and a bit of present uncertainty as our data is always imperfect). As modellers and planners we cannot provide useful advice to decision-makers pretending than we can forecast the future; that would be unethical and foolish.

Two main sources of uncertainty plague our forecasts: Model uncertainty and External uncertainty. Model uncertainty is a result of our imperfect understanding of human behaviour and the paucity of the data (small sample, noise to signal ratio, etc.) available to feed our models. Eric’s paper list some of the areas where we fall far short of a reliable understanding of human behaviour. Moreover, as most is cross-section data we certainly do not understand how humans change their mind and values over time. More research offers the promise of reducing this type of uncertainty but the ideal of closing the gap is an hallucination based on a mechanical view of what it is to be human.

External uncertainty has many sources: the future evolution of the economy, migration, energy prices, globalisation, inequality. Even our own creativity is a source of uncertainty: the adoption and acceptability of virtual presence, the advent of new mobility services including the long-awaited autonomous vehicle, changes in government priorities, etc. Not to mention geopolitical friction, pandemics and global warming.

Uncertainty raises the question of the level of granularity appropriate to deliver sound advice under conditions of uncertainty. Would a simpler, more agile, model provide better support to explore alternative futures than a more complex but behaviourally richer model? But, how simple is good enough? Each type of model will require a different level of assumptions about future input data and each will pose different demands on limited resources.

It would be interesting to find out if the NYBPM is able to provide a reasonable representation of present (post pandemic) conditions so that it could be used to support decisions today. I do not know the answer to this question as the model is based on 2012 data and apparently produces forecasts only for 2040 and 2055 (I am only using the website as a reference; I have no experience in this model).

There is very little research on what would be the best modelling approach to support decision making under uncertainty. This is an uncomfortable gap in our professional practice as we may be using the wrong tool for this purpose.

Luis Willumsen

……………………………………………

Luis Willumsen

Director

Willumsen Advisory Services

Nommon Solutions and Technologies

London & Madrid

M: +44 7979 53 88 45

www.nommon.es

…………………………………………..

From: on behalf of Chandra Bhat
Date: Monday, 8 May 2023 at 21:04
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Yes, Kostas. The work I referenced was done some time back, but, as Bernie indicated, there has been little empirical comparisons between activity models and trip-based models. Yes, activity models do have much more of a theoretical basis, and should provide important insights at disaggregate levels. Besides, time-use, time-space prisms, and complete activity-travel patterns becomes even more central as we go into an era of increasing virtual participations, and emerging mobility options (including automated vehicles). More the reason to put the emphasis on activities.

My response was to the specific issue of comparing trip-based versus tour/activity-based approaches. Easier said than done, because of so many confounding factors and the basic issue of what constitutes “ground truth”. I do recall we put in considerable thought in the comparisons we made for the ODOT project to compare apples to apples.

If we are talking about ABM implementations in general, in addition to the Activitysim stuff that Kostas references, Kostas led a team (including Ram and myself) for a Doha (Qatar) ABM project before COVID (https://mot.gov.qa/sites/default/files/LT/QTMV/Executive%20Summary%20-%2...), and we also completed a NYMTC ABM project with Tom Rossi, Ram Pendyala, myself, and many other colleagues in 2021 (https://www.nymtc.org/en-us/Data-and-Modeling/New-York-Best-Practice-Mod...).

Best to all,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Kostas Goulias
Sent: Monday, May 8, 2023 12:54 PM
To: Bhat, Chandra R
Cc: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is old stuff. Check activitysim.org

Konstadinos (Kostas) G. Goulias
Professor of Transportation at UCSB
www.kostasgoulias.com
geotrans.geog.ucsb.edu

On Mon, May 8, 2023 at 10:52 AM wigobiner > wrote:

Bernie:

Good to hear from you. And I hear you about comparisons between different modeling paradigms. While seemingly straightforward, there are so many different considerations that need to be accommodated in these comparisons, including at what level (aggregate/disaggregate) to compare outputs and to ensure (as far as possible) that we are comparing apples to apples. And, then, we also have the constant challenge of what constitutes the "ground truth" to compare different model results with.

Please see this peer-reviewed paper of research we undertook at UT Austin with other colleagues you will recognize.

Ferdous, N., L. Vana, J.L. Bowman, R.M. Pendyala, G. Giaimo, C.R. Bhat, D. Schmitt, M. Bradley, and R. Anderson (2012), "Comparison of Four-Step Versus Tour-Based Models for Prediction of Travel Behavior Before and After Transportation System Changes," Transportation Research Record, Vol. 2303, pp. 46-60 (Keywords: trip-based model, MORPC tour-based model, vehicle ownership, work flow distribution, and highway projects). PDF version, MS Word version

An executive summary and the full report submitted to ODOT are available here:
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Sensitivity of Four-Step Versus Tour-Based Models to Transportation System Changes," Executive Summary Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Ferdous, N., C.R. Bhat, L. Vana, D. Schmitt, J. Bowman, M. Bradley, and R. Pendyala, "Comparison of Four-Step Versus Tour-Based Models in Predicting Travel Behavior Before and After Transportation System Changes - Results Interpretation and Recommendations," Technical Report FHWA/OH-2011/4, prepared for the Ohio Department of Transportation, February 2011. PDF version, MS Word version
Best,
Chandra.

Chandra Bhat
University Distinguished Teaching Professor
Joe J. King Chair in Engineering
Department of Civil, Architectural and Environmental Engineering
Department of Economics (Courtesy Appointment)
The University of Texas at Austin
Austin, Texas 78712
Tel: (512) 771-9166
http://www.ce.utexas.edu/prof/bhat/home.html
Web of Science Researcher: https://publons.com/researcher/3178803/chandra-bhat/

I live and work on occupied Indigenous land and acknowledge the Carrizo & Comecrudo, Coahuiltecan, Caddo, Tonkawa, Comanche, Lipan Apache, Alabama-Coushatta, Kickapoo, Tigua Pueblo, and all the American Indian and Indigenous Peoples and communities who have been or have become a part of the lands and territories known today as Texas.

From: Bernard.Alpern=gmail.com@mg.tmip.org On Behalf Of balpern2
Sent: Monday, May 8, 2023 10:07 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

Krishnan (and Eric), thank you very much for this.

As some of you receiving this email know, I was quite active in the travel demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman's introductory course at MIT in 1975, through my retirement last year (47 years later). Throughout that period, I never had the opportunity to develop, maintain, or operate a tour-based model (which Eric correctly points out is the proper name for models currently used in practice that are often called "activity-based models" by many people in the field).

I maintained a keen interest in travel demand models, however, and rejoiced at the work of John (Bowman) and Moshe when it was presented in the mid 90's, as I immediately recognized it as an essential breakthrough in developing an appropriate mathematical structure for such models.

One thing that I always found frustrating, however, was that the practice adopted as an axiom that because of their theoretically more solid basis in traveler behavior, such models HAD to be superior to the older, "four-step" approach to travel modeling (even including many, many enhancements over the original formulation of such models). Yet to date, I have not come across a single straight-up comparison of the performance of tour-based vs. four-step models, as measured in appropriate ways. Given that tour-based models are, in fact, substantially more complicated, difficult to develop, maintain and use, and computationally inefficient relative to four-step models (as reflected by the fact that some 30 years after research on activity-based modeling began in earnest, "the classic trip-based [four-step] approach remains standard practice in much of the world" - quoting from Eric's article), you would think that by now several such comparisons would have been made and the findings published. Even if the tour-based approach were demonstrated to provide superior forecasts, the questions of "by how much?" and "is it worth the substantially greater time and effort involved?" would still need to be addressed and discussed.

If anyone reading this is aware of any such work that has been published (and, preferably, peer-reviewed), I would be extremely happy if I could be told how to find it.

Thank you in advance,

Bernard (Bernie) Alpern

Jerusalem, Israel

bernard.alpern@gmail.com

From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
Sent: Monday, 8 May 2023 14:40
To: TMIP
Subject: [TMIP] Current State of Activity based Models and Next Steps

As a practitioner I found Eric Miller's editorial on the current state of
ABMs in Transport Reviews to be thought provoking and interesting & thought
it might be of interest to this group. Maybe there will be some discussions
about the editorial and the proposed solutions at next month's conference
on Innovations in Travel Analysis & Planning. The proposed solutions seem,
to me personally, more academic in nature than something the industry can
adopt. Since I am unsure of Transport Review's restrictions on sharing
widely, I will share the link to the editorial (
https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list he
might be willing to share it directly.

The abstract is below.

Despite the clear theoretical advantages of activity-based models of travel
behaviour relative to trip-based models, adoption of such models in
planning practice has been slow. This editorial discusses some reasons
underlying this fact, including "locking into" outmoded model structures
and software and challenges in translating research advances into practice.
It argues for more widespread adoption of an activity-scheduling approach
to the problem and identifies a number of key areas requiring new research
in order to improve the operational capabilities of these models.

Krishnan

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

The original post came from a "practitioner's" perspective. As a
practitioner I'm responsible for carrying out the travel forecasts for
Corridor alternative analyses, subarea (County), Sector, and activity
center plans, and providing inputs facility design and operations
(including ITS) studies. For these we typically:

1. Obtain the adopted regional network Model (usually trip based)
2. Add network, zone, and operational details to capture the subarea and
policy decisions of interest often through pre and post processing to
inject modifications at key points in the overall model stream.
3. Carry out reasonableness checks and validation within the area of
interest.
4. Code and run base and future alternatives
5. Extract and summarize results and measures of effectiveness.

We have worked in several regions that have adopted Activity Based Models
for their regional process and would be very interested in someone looking
at whether they can practically be used to carry out the above
activities for the above bread and butter tasks that we are asked to
perform. This is contrast to their occasional use to carry out the
official regional forecasts associated with an LRP update. We have found:

1. There is often a reticence to share the adopted regional ABM because
it is felt that it is "too" complex for others to implement/operate.
2. When the ABM model is obtained, it can be difficult to implement
locally, often has hard coded folder and drive specifications (just so you
know we have the computer resources and have been able to do this on the
ABM models that have been shared).
3. Even when successfully installed and implemented the models often are
still coded at a "regionally significant" level and do not include the
details needed for subare, corridor, and operational analyses and are
aggregated back up to be assigned in the traditional network assignment
methods and detail.
4. Because of the custom coding and modules it is very very difficult to
modify the ABM to add additional zone and network detail or change other
features (an example is ABMs that use compiled java modules). They are
effectively closed box tools unless you have a full data analysis and
programming staff or are one of the original developers.

Has anyone looked at how these improved models are being used x years after
their adoption? Are they being used for anything other than the regional
LRP and conformity updates? Is their adoption reducing the pool of those
that can/do carry out analyses discussed above? Are they being
distributed/used by the consulting community that supports these studies?
Are they improving the decisions made as part of these studies?

These are just practical issues that impact my day to day work.

JAB

On Mon, May 8, 2023 at 11:07 AM balpern2 wrote:

> Krishnan (and Eric), thank you very much for this.
>
> As some of you receiving this email know, I was quite active in the travel
> demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman’s
> introductory course at MIT in 1975, through my retirement last year (47
> years later). Throughout that period, I never had the opportunity to
> develop, maintain, or operate a tour-based model (which Eric correctly
> points out is the proper name for models currently used in practice that
> are often called “activity-based models” by many people in the field).
>
> I maintained a keen interest in travel demand models, however, and
> rejoiced at the work of John (Bowman) and Moshe when it was presented in
> the mid 90’s, as I immediately recognized it as an essential breakthrough
> in developing an appropriate mathematical structure for such models.
>
> One thing that I always found frustrating, however, was that the practice
> adopted as an axiom that because of their theoretically more solid basis in
> traveler behavior, such models HAD to be superior to the older, “four-step”
> approach to travel modeling (even including many, many enhancements over
> the original formulation of such models). Yet to date, I have not come
> across a single straight-up comparison of the performance of tour-based vs.
> four-step models, as measured in appropriate ways. Given that tour-based
> models are, in fact, substantially more complicated, difficult to develop,
> maintain and use, and computationally inefficient relative to four-step
> models (as reflected by the fact that some 30 years after research on
> activity-based modeling began in earnest, “the classic trip-based
> [four-step] approach remains standard practice in much of the world” –
> quoting from Eric’s article), you would think that by now several such
> comparisons would have been made and the findings published. Even if the
> tour-based approach were demonstrated to provide superior forecasts, the
> questions of “by how much?” and “is it worth the substantially greater time
> and effort involved?” would still need to be addressed and discussed.
>
> If anyone reading this is aware of any such work that has been published
> (and, preferably, peer-reviewed), I would be extremely happy if I could be
> told how to find it.
>
> Thank you in advance,
>
> Bernard (Bernie) Alpern
>
> Jerusalem, Israel
>
> bernard.alpern@gmail.com
>
> From: krisviswanathan=gmail.com@mg.tmip.org On Behalf Of krishnan
> Sent: Monday, 8 May 2023 14:40
> To: TMIP
> Subject: [TMIP] Current State of Activity based Models and Next Steps
>
> As a practitioner I found Eric Miller's editorial on the current state of
> ABMs in Transport Reviews to be thought provoking and interesting & thought
> it might be of interest to this group. Maybe there will be some discussions
> about the editorial and the proposed solutions at next month's conference
> on Innovations in Travel Analysis & Planning. The proposed solutions seem,
> to me personally, more academic in nature than something the industry can
> adopt. Since I am unsure of Transport Review's restrictions on sharing
> widely, I will share the link to the editorial (
> https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list
> he
> might be willing to share it directly.
>
> The abstract is below.
>
> Despite the clear theoretical advantages of activity-based models of travel
> behaviour relative to trip-based models, adoption of such models in
> planning practice has been slow. This editorial discusses some reasons
> underlying this fact, including “locking into” outmoded model structures
> and software and challenges in translating research advances into practice.
> It argues for more widespread adoption of an activity-scheduling approach
> to the problem and identifies a number of key areas requiring new research
> in order to improve the operational capabilities of these models.
>
> Krishnan
>
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
>

--
James (Jim) Allday Bunch, JABunch Transportation Consulting
411 Penwood Road, Silver Spring Maryland, 20901
240-271-3534 jabunch.work@gmail.com

This has been an interesting discussion so far, and although I have never really been an actual practitioner, I thought I would chime in after Jim's comment.

First, I would be willing to leave aside issues with practicalities such as quirks (i.e. poor implementation) from bespoke software and other issues that are not intrinsic from ABMs (or TBMs) for now.

However, I think that there is a link between Jim's comment on a "typical use" and the point Chandra made that models should be as simple as possible to their purpose.

The link is that, in most agencies, (forecasting) transport models are used for a variety of applications, ranging from traffic volume forecasts for new freeway sections, to parking or public transport pricing changes, to congestion charging and beyond.

When you add the need to look into results from an equity or environmental perspectives on top of pure transportation/traffic/patronage outcomes. I would think that most people on this list would be capable of building a simple model capable of any one particular analysis, but I struggle to believe that it would be possible to build a single model that can answer all policy questions modelling teams are asked to answer without having a complex model.

That said, I am not at all saying that we need models are complex as we have today (I am not a practitioner after all), but I find it a bit short-sided to think that ultra-simple models with a focus on uncertainty would fit the bill for any mid-size to large agency.

Finally, I would ask myself why not contemplating the possibility of having multiple purpose-built models if I were a bit more naïve, but then I remember that the mere possibility of having multiple models that could potentially be giving discrepant results for a particular KPI would send shivers down most modelling managers in MPOS and DoTs in the US and abroad.

Cheers,

Pedro

PS - I hope to see most of you in Indianapolis!

---- On Thu, 11 May 2023 06:31:01 +1000 jabunch wrote ---

The original post came from a "practitioner's" perspective. As a
practitioner I'm responsible for carrying out the travel forecasts for
Corridor alternative analyses, subarea (County), Sector, and activity
center plans, and providing inputs facility design and operations
(including ITS) studies. For these we typically:
1. Obtain the adopted regional network Model (usually trip based)
2. Add network, zone, and operational details to capture the subarea and
policy decisions of interest often through pre and post processing to
inject modifications at key points in the overall model stream.
3. Carry out reasonableness checks and validation within the area of
interest.
4. Code and run base and future alternatives
5. Extract and summarize results and measures of effectiveness.
We have worked in several regions that have adopted Activity Based Models
for their regional process and would be very interested in someone looking
at whether they can practically be used to carry out the above
activities for the above bread and butter tasks that we are asked to
perform. This is contrast to their occasional use to carry out the
official regional forecasts associated with an LRP update. We have found:
1. There is often a reticence to share the adopted regional ABM because
it is felt that it is "too" complex for others to implement/operate.
2. When the ABM model is obtained, it can be difficult to implement
locally, often has hard coded folder and drive specifications (just so you
know we have the computer resources and have been able to do this on the
ABM models that have been shared).
3. Even when successfully installed and implemented the models often are
still coded at a "regionally significant" level and do not include the
details needed for subare, corridor, and operational analyses and are
aggregated back up to be assigned in the traditional network assignment
methods and detail.
4. Because of the custom coding and modules it is very very difficult to
modify the ABM to add additional zone and network detail or change other
features (an example is ABMs that use compiled java modules). They are
effectively closed box tools unless you have a full data analysis and
programming staff or are one of the original developers.
Has anyone looked at how these improved models are being used x years after
their adoption? Are they being used for anything other than the regional
LRP and conformity updates? Is their adoption reducing the pool of those
that can/do carry out analyses discussed above? Are they being
distributed/used by the consulting community that supports these studies?
Are they improving the decisions made as part of these studies?
These are just practical issues that impact my day to day work.

JAB

On Mon, May 8, 2023 at 11:07 AM balpern2 wrote:

> Krishnan (and Eric), thank you very much for this.
>
> As some of you receiving this email know, I was quite active in the travel
> demand modeling space from when I took Moshe Ben-Akiva and Steve Lerman’s
> introductory course at MIT in 1975, through my retirement last year (47
> years later). Throughout that period, I never had the opportunity to
> develop, maintain, or operate a tour-based model (which Eric correctly
> points out is the proper name for models currently used in practice that
> are often called “activity-based models” by many people in the field).
>
> I maintained a keen interest in travel demand models, however, and
> rejoiced at the work of John (Bowman) and Moshe when it was presented in
> the mid 90’s, as I immediately recognized it as an essential breakthrough
> in developing an appropriate mathematical structure for such models.
>
> One thing that I always found frustrating, however, was that the practice
> adopted as an axiom that because of their theoretically more solid basis in
> traveler behavior, such models HAD to be superior to the older, “four-step”
> approach to travel modeling (even including many, many enhancements over
> the original formulation of such models). Yet to date, I have not come
> across a single straight-up comparison of the performance of tour-based vs.
> four-step models, as measured in appropriate ways. Given that tour-based
> models are, in fact, substantially more complicated, difficult to develop,
> maintain and use, and computationally inefficient relative to four-step
> models (as reflected by the fact that some 30 years after research on
> activity-based modeling began in earnest, “the classic trip-based
> [four-step] approach remains standard practice in much of the world” –
> quoting from Eric’s article), you would think that by now several such
> comparisons would have been made and the findings published. Even if the
> tour-based approach were demonstrated to provide superior forecasts, the
> questions of “by how much?” and “is it worth the substantially greater time
> and effort involved?” would still need to be addressed and discussed.
>
> If anyone reading this is aware of any such work that has been published
> (and, preferably, peer-reviewed), I would be extremely happy if I could be
> told how to find it.
>
> Thank you in advance,
>
> Bernard (Bernie) Alpern
>
> Jerusalem, Israel
>
> mailto:bernard.alpern@gmail.com
>
> From: krisviswanathan=mailto:gmail.com@mg.tmip.org On Behalf Of krishnan
> Sent: Monday, 8 May 2023 14:40
> To: TMIP
> Subject: [TMIP] Current State of Activity based Models and Next Steps
>
> As a practitioner I found Eric Miller's editorial on the current state of
> ABMs in Transport Reviews to be thought provoking and interesting & thought
> it might be of interest to this group. Maybe there will be some discussions
> about the editorial and the proposed solutions at next month's conference
> on Innovations in Travel Analysis & Planning. The proposed solutions seem,
> to me personally, more academic in nature than something the industry can
> adopt. Since I am unsure of Transport Review's restrictions on sharing
> widely, I will share the link to the editorial (
> https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this list
> he
> might be willing to share it directly.
>
> The abstract is below.
>
> Despite the clear theoretical advantages of activity-based models of travel
> behaviour relative to trip-based models, adoption of such models in
> planning practice has been slow. This editorial discusses some reasons
> underlying this fact, including “locking into” outmoded model structures
> and software and challenges in translating research advances into practice.
> It argues for more widespread adoption of an activity-scheduling approach
> to the problem and identifies a number of key areas requiring new research
> in order to improve the operational capabilities of these models.
>
> Krishnan
>
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
>
--
James (Jim) Allday Bunch, JABunch Transportation Consulting
411 Penwood Road, Silver Spring Maryland, 20901
240-271-3534 mailto:jabunch.work@gmail.com
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960

A worthy line of thought/activity may be how to connect the complex and
simple models in a meaningful way. For a corridor or subarea study I want
to be able to state that my model results are consistent with and an
extension of the adopted long range forecasts and model results. Yet, I
often need more detail. So does anyone have methods that pivot off the
more complex ABM regional models for specific local studies? Or have
methods to reaggregate the point to point tour based travel information to
other than the regional TAZs so more details can be added within a study
area. We have often done this for trip based models where we feel that
there will be insignificant mode shift impacts due to the traffic
operations scenarios we are considering (we further disaggregate the post
mode choice vehicle trip tables). If the connections are thought out there
may be a way to develop a consistent hierarchy of model tools and practices
that capture both levels of modeling and sensitivity.

On Thu, May 11, 2023 at 12:40 PM SuzanneChildress <
childresssuzanne@gmail.com> wrote:

> From a practitioner perspective at large MPO, Pedro’s assessment seems
> correct to me:
>
> *“When you add the need to look into results from an equity or
> environmental perspectives on top of pure transportation/traffic/patronage
> outcomes. I would think that most people on this list would be capable of
> building a simple model capable of any one particular analysis, but I
> struggle to believe that it would be possible to build a single model that
> can answer all policy questions modelling teams are asked to answer without
> having a complex model.”*
>
> The latest question Puget Sound Regional Council data staff were asked with
> our model system was:
>
> What policies and projects can we implement to reach our climate goals
> while maintaining transportation accessibility for people of color, people
> with low incomes, older adults, youth, people with disabilities, and people
> with limited English proficiency? *
>
> My colleague Brice Nichols will discuss this work at the upcoming
> conference
>
> Krishnan and Pedro referenced. See you there!
>
> In the Puget Sound Region, there are separate models to answer specific
> short term project level or land use related questions, maintained to that
> level of accuracy, time frame, and sensitivity. **
>
> Maybe we’re coming to consensus:
>
> *We need different models to answer different needs*! Let’s make more
> models tailored to more specific questions, not less. We need scenario
> tools, ABMs, one-off MNLs, simple linear regressions, and machine-learning.
> Let’s use more data and keep expanding our knowledge, like Chandra noted!
> Let’s keep building our toolsets! And for all these tools and data sources,
> we need to work together and develop standards.
>
> Suzanne Childress
>
> Puget Sound Regional Council
>
> Principal Data Scientist
>
> * This statement is fully my own and not that of PSRC: I bet you know what
> worked, but no policy maker liked it. Come to Brice’s presentation and find
> out!
>
> **My sense with why ABMs are still not in wide use at a project-level is
> that the field of planning is not willing to invest the money in the
> problem to fully solve its usability and functionality. Like if we just had
> $100 million (or something like that, people working on the problem could
> tell me more accurately what that dollar amount is) to spend for 2 years we
> could build out an open source, fast system that could work anywhere in the
> world that could provide useful outputs. ---But maybe that $100 million
> would be better spent elsewhere like figuring out how to merge travel
> survey data with all other available data sources in an *open,
> statistically
> valid, standard way* that is widely accepted by the community (my
> preference, can somebody figure that out finally academics! I don’t have
> time as a practitioner).
>
> ***I don’t do modeling so much anymore-except when asked by my friends-
> because I think my skills are more helpful to planning by creating data
> analytics, data visualizations, statistics, and general programming
> products. I think broadly using my skills developed over the years in
> modeling and surveys is the best way for me to help my region. As we can
> see modeling is so fraught with controversy, and I can’t handle that heat
> anymore! If anyone is interested in transitioning from the usual modeling
> paradigm to general data science in support of planning, I’d love to chat!
> Check out some of the team’s recent work here: https://github.com/psrc
>
> On Thu, May 11, 2023 at 4:07 AM Pedro Camargo wrote:
>
> > This has been an interesting discussion so far, and although I have never
> > really been an actual practitioner, I thought I would chime in after
> Jim's
> > comment.
> >
> > First, I would be willing to leave aside issues with practicalities such
> > as quirks (i.e. poor implementation) from bespoke software and other
> issues
> > that are not intrinsic from ABMs (or TBMs) for now.
> >
> > However, I think that there is a link between Jim's comment on a "typical
> > use" and the point Chandra made that models should be as simple as
> possible
> > to their purpose.
> >
> > The link is that, in most agencies, (forecasting) transport models are
> > used for a variety of applications, ranging from traffic volume forecasts
> > for new freeway sections, to parking or public transport pricing changes,
> > to congestion charging and beyond.
> >
> > When you add the need to look into results from an equity or
> environmental
> > perspectives on top of pure transportation/traffic/patronage outcomes. I
> > would think that most people on this list would be capable of building
> > a simple model capable of any one particular analysis, but I struggle to
> > believe that it would be possible to build a single model that can answer
> > all policy questions modelling teams are asked to answer without having a
> > complex model.
> >
> > That said, I am not at all saying that we need models are complex as we
> > have today (I am not a practitioner after all), but I find it a bit
> > short-sided to think that ultra-simple models with a focus on uncertainty
> > would fit the bill for any mid-size to large agency.
> >
> > Finally, I would ask myself why not contemplating the possibility of
> > having multiple purpose-built models if I were a bit more naïve, but
> then I
> > remember that the mere possibility of having multiple models that could
> > potentially be giving discrepant results for a particular KPI would send
> > shivers down most modelling managers in MPOS and DoTs in the US and
> abroad.
> >
> > Cheers,
> >
> > Pedro
> >
> > PS - I hope to see most of you in Indianapolis!
> >
> > ---- On Thu, 11 May 2023 06:31:01 +1000 jabunch wrote ---
> >
> > The original post came from a "practitioner's" perspective. As a
> > practitioner I'm responsible for carrying out the travel forecasts for
> > Corridor alternative analyses, subarea (County), Sector, and activity
> > center plans, and providing inputs facility design and operations
> > (including ITS) studies. For these we typically:
> > 1. Obtain the adopted regional network Model (usually trip based)
> > 2. Add network, zone, and operational details to capture the subarea and
> > policy decisions of interest often through pre and post processing to
> > inject modifications at key points in the overall model stream.
> > 3. Carry out reasonableness checks and validation within the area of
> > interest.
> > 4. Code and run base and future alternatives
> > 5. Extract and summarize results and measures of effectiveness.
> > We have worked in several regions that have adopted Activity Based Models
> > for their regional process and would be very interested in someone
> looking
> > at whether they can practically be used to carry out the above
> > activities for the above bread and butter tasks that we are asked to
> > perform. This is contrast to their occasional use to carry out the
> > official regional forecasts associated with an LRP update. We have found:
> > 1. There is often a reticence to share the adopted regional ABM because
> > it is felt that it is "too" complex for others to implement/operate.
> > 2. When the ABM model is obtained, it can be difficult to implement
> > locally, often has hard coded folder and drive specifications (just so
> you
> > know we have the computer resources and have been able to do this on the
> > ABM models that have been shared).
> > 3. Even when successfully installed and implemented the models often are
> > still coded at a "regionally significant" level and do not include the
> > details needed for subare, corridor, and operational analyses and are
> > aggregated back up to be assigned in the traditional network assignment
> > methods and detail.
> > 4. Because of the custom coding and modules it is very very difficult to
> > modify the ABM to add additional zone and network detail or change other
> > features (an example is ABMs that use compiled java modules). They are
> > effectively closed box tools unless you have a full data analysis and
> > programming staff or are one of the original developers.
> > Has anyone looked at how these improved models are being used x years
> after
> > their adoption? Are they being used for anything other than the regional
> > LRP and conformity updates? Is their adoption reducing the pool of those
> > that can/do carry out analyses discussed above? Are they being
> > distributed/used by the consulting community that supports these studies?
> > Are they improving the decisions made as part of these studies?
> > These are just practical issues that impact my day to day work.
> >
> > JAB
> >
> > On Mon, May 8, 2023 at 11:07 AM balpern2 wrote:
> >
> > > Krishnan (and Eric), thank you very much for this.
> > >
> > > As some of you receiving this email know, I was quite active in the
> > travel
> > > demand modeling space from when I took Moshe Ben-Akiva and Steve
> Lerman’s
> > > introductory course at MIT in 1975, through my retirement last year (47
> > > years later). Throughout that period, I never had the opportunity to
> > > develop, maintain, or operate a tour-based model (which Eric correctly
> > > points out is the proper name for models currently used in practice
> that
> > > are often called “activity-based models” by many people in the field).
> > >
> > > I maintained a keen interest in travel demand models, however, and
> > > rejoiced at the work of John (Bowman) and Moshe when it was presented
> in
> > > the mid 90’s, as I immediately recognized it as an essential
> breakthrough
> > > in developing an appropriate mathematical structure for such models.
> > >
> > > One thing that I always found frustrating, however, was that the
> practice
> > > adopted as an axiom that because of their theoretically more solid
> basis
> > in
> > > traveler behavior, such models HAD to be superior to the older,
> > “four-step”
> > > approach to travel modeling (even including many, many enhancements
> over
> > > the original formulation of such models). Yet to date, I have not come
> > > across a single straight-up comparison of the performance of tour-based
> > vs.
> > > four-step models, as measured in appropriate ways. Given that
> tour-based
> > > models are, in fact, substantially more complicated, difficult to
> > develop,
> > > maintain and use, and computationally inefficient relative to four-step
> > > models (as reflected by the fact that some 30 years after research on
> > > activity-based modeling began in earnest, “the classic trip-based
> > > [four-step] approach remains standard practice in much of the world” –
> > > quoting from Eric’s article), you would think that by now several such
> > > comparisons would have been made and the findings published. Even if
> the
> > > tour-based approach were demonstrated to provide superior forecasts,
> the
> > > questions of “by how much?” and “is it worth the substantially greater
> > time
> > > and effort involved?” would still need to be addressed and discussed.
> > >
> > > If anyone reading this is aware of any such work that has been
> published
> > > (and, preferably, peer-reviewed), I would be extremely happy if I could
> > be
> > > told how to find it.
> > >
> > > Thank you in advance,
> > >
> > > Bernard (Bernie) Alpern
> > >
> > > Jerusalem, Israel
> > >
> > > mailto:bernard.alpern@gmail.com
> > >
> > > From: krisviswanathan=mailto:gmail.com@mg.tmip.org
>
> > On Behalf Of krishnan
> > > Sent: Monday, 8 May 2023 14:40
> > > To: TMIP
> > > Subject: [TMIP] Current State of Activity based Models and Next Steps
> > >
> > > As a practitioner I found Eric Miller's editorial on the current state
> of
> > > ABMs in Transport Reviews to be thought provoking and interesting &
> > thought
> > > it might be of interest to this group. Maybe there will be some
> > discussions
> > > about the editorial and the proposed solutions at next month's
> conference
> > > on Innovations in Travel Analysis & Planning. The proposed solutions
> > seem,
> > > to me personally, more academic in nature than something the industry
> can
> > > adopt. Since I am unsure of Transport Review's restrictions on sharing
> > > widely, I will share the link to the editorial (
> > > https://doi.org/10.1080/01441647.2023.2198458) & if Eric is on this
> list
> > > he
> > > might be willing to share it directly.
> > >
> > > The abstract is below.
> > >
> > > Despite the clear theoretical advantages of activity-based models of
> > travel
> > > behaviour relative to trip-based models, adoption of such models in
> > > planning practice has been slow. This editorial discusses some reasons
> > > underlying this fact, including “locking into” outmoded model
> structures
> > > and software and challenges in translating research advances into
> > practice.
> > > It argues for more widespread adoption of an activity-scheduling
> approach
> > > to the problem and identifies a number of key areas requiring new
> > research
> > > in order to improve the operational capabilities of these models.
> > >
> > > Krishnan
> > >
> > > --
> > > Full post:
> > >
> >
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> > > Manage my subscriptions: https://tmip.org/mailinglist
> > > Stop emails for this post:
> > https://tmip.org/mailinglist/unsubscribe/13960
> > > --
> > > Full post:
> > >
> >
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> > > Manage my subscriptions: https://tmip.org/mailinglist
> > > Stop emails for this post:
> > https://tmip.org/mailinglist/unsubscribe/13960
> > >
> > --
> > James (Jim) Allday Bunch, JABunch Transportation Consulting
> > 411 Penwood Road, Silver Spring Maryland, 20901
> > 240-271-3534 mailto:jabunch.work@gmail.com
> > --
> > Full post:
> >
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> > Manage my subscriptions: https://tmip.org/mailinglist
> > Stop emails for this post:
> https://tmip.org/mailinglist/unsubscribe/13960
> > --
> > Full post:
> >
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> > Manage my subscriptions: https://tmip.org/mailinglist
> > Stop emails for this post:
> https://tmip.org/mailinglist/unsubscribe/13960
> >
> --
> Full post:
> https://tmip.org/content/current-state-activity-based-models-and-next-steps
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960
>

--
James (Jim) Allday Bunch, JABunch Transportation Consulting
411 Penwood Road, Silver Spring Maryland, 20901
240-271-3534 jabunch.work@gmail.com

This is an extremely good discussion. As folks have already discussed the pros and cons quite in depth both theoretically and practically, I would like to add what I think is a big piece missing in this puzzle: resources. Or more straightforwardly, money and staff. (Well apparently, staffing resource basically can be translated to the money to hire qualified staff.)

I’ll start with the latest post in this thread. Suzanne wrote:

“

Maybe we’re coming to consensus:

*We need different models to answer different needs*! Let’s make more models tailored to more specific questions, not less. We need scenario tools, ABMs, one-off MNLs, simple linear regressions, and machine-learning. Let’s use more data and keep expanding our knowledge, like
Chandra noted! Let’s keep building our toolsets! And for all these tools and data sources, we need to work together and develop standards.

”

This reminds me of the “Let them eat cake” story. Please don’t get me wrong, actually I do think that sounds like a perfect solution, given the fact that an agency has enormous resources, but that’s not always the case, so I am actually jealous of what you got. I just found out that the data and modeling team at your MPO alone is more than twice the size of our whole MPO squad.

In my opinion, whether a modeling approach is good/suitable for an agency, has two dimensions. One is the model itself at the time of delivery and the other is its sustainability. For the first part, agencies may be able to invest a lump sum funding to develop a high-cost model (usually those more complex ones) at some point, however, to maintain and simply use it in the long run, it will require constant both financial input and staffing input. I also learned this lesson after I purchased my first house and I’m sure you don’t need more details. I will just throw one example/topic here. After running a model and getting the output numbers we need, I genuinely believe it is essential for us modelers to understand why we get these outputs as they look like, and, when we run different scenarios, to understand why these outputs are different from each other, “what is the causation or what are the causations of these differences”, and, when evaluating a specific project when a base-year number is off the ground truth more than we expected at that specific location, why is that. As a simple “practitioner” (like Paul referred to himself), with my knowledge set, I am capable of answering all those questions for a four-step model, whether purely based on the knowledge on the model structure or by conducting a further selected links analysis to find out, I can do that, for a four-step model, but I am sorry to admit, for an activity-based model, I am not capable of doing the same thing, at this time, given the fact that there are so many correlations behind the black box. I am sure there are many experts out there who know everything behind these black boxes, but I doubt all agencies can afford consulting them on a routine basis when using the model in the ways Jim has summarized earlier. I think this example also called back what Dan wrote “Complexity may be beautiful, and it certainly is complex. It thus may be inaccessible except to those who understand that complexity. But that does not mean that it is also correct.” and the synthetic Q&A Ken posted earlier:

“

Q2. Your model is predicting that auto VMT (or ridership) in
the corridor of interest increased by XX percent between the calibration year
and 2040, which is noticeably higher (or lower) than the predicted increase in
the corridor’s population and employment. What is the reason for this?

A2. We have no idea. It’s what the model says.

”

I just come up with an example to explain the potential side effect of not knowing what are behind the results. For example, when applying Policy #A to the model and we may get a negative impact toward our Goal #B, but it may actually be caused by the impact of Policy #A having on Factor #C. If we don’t know what’s behind it, we may simply conclude that Policy #A does not comply with our Goal #B. However, if we know the actual causation, we may be able to figure out a Strategy #D to eliminate such side impact and then apply both Policy #A and Strategy #D to the model which could potentially end up with a positive impact toward our Goal #B.

I seem to have gone beyond the topic of resources so I will stop here though I still have much to say. In the end, I do have a question for Peter as you wrote:

“Most of the US MPOs have already switched to ABMs and
abandoned their 4-steps. Most of the cities around the world still use 4-steps
and do not have an ABM yet.”

There are around 408 MPOs in the US and I am sure not all them have travel demand models, but for those who do, I am wondering if most of them have switched to ABMs. I really don’t know. Peter, would you mind providing more source for the information on that? I am just curious. Also, on the second part for the rest of the world, if you can provide some additional information too, I appreciate that. Thank you.

See you all in Indianapolis! 

Best,

Xinbo Mi

Evansville MPO

Hello Xinbo Mi and everyone else,

Xinbo’s post earlier this morning summarizes the overall “situation” very well, so let’s thank this person for writing.
When I sent my earlier post about some made-up (synthesized) Q&A questions, the answers came from multiple practitioners primarily using trip-based rather than activity-based models (but not entirely): so if a typical forecaster is having difficulty “explaining the forecast” from a simpler model, or even performing a “so how well is the model really performing when it comes to predicting actual observed changes in travel between two actual years” exercise, I continue to remain skeptical that the same forecaster is going to somehow “better understand” whatever comes out of a more “sophisticated” modeling environment.

Just for fun, I decided to ask Bing AI for his/her/its/their concerns about activity-based models. After spending five seconds examining billions of electronic documents, it seems to have just “given up” and decided to directly quote from wiki books (first sentence) and RSG’s ActivitySim white paper (second sentence):

One concern with activity-based models is that they require solving many optimization problems simultaneously, which is computationally difficult and behaviorally unrealistic1. Additionally, user concerns about activity-based models’ cost, complexity, and support have limited their wider adoption2.

My personal opinion is that a simpler model construct placed in the hands of a knowledgeable and competent forecaster who actually UNDERSTANDS what’s behind the predictions (with healthy amounts of DATA available to get a grasp on the likely uncertainties), with considerable doses of PUBLIC HUMILITY to tell others that his/her knowledge of the situation is still not anywhere close to perfect, is going to prepare much more useful forecasts to support a meaningful decision-making process than any alternative. It seems the current AI (artificial intelligence) bots have not yet figured out how to summarize the thought-provoking “Hubris or Humility” paper written a number of years ago by David Hartgen.
Ken Cervenka

p.s., if I was at the beginning rather than beyond the ending of a travel forecasting career, I would try to spend more time in not just having multiple analytical tools available, but be sure that one of those tools is grounded in what is commonly called an incremental-based approach that relies heavily on DATA to come up with current-year person travel tables.

From: Xinbo Mi
Sent: Friday, May 12, 2023 7:23 AM
To: TMIP
Subject: Re: [TMIP] Current State of Activity based Models and Next Steps

This is an extremely good discussion. As folks have already discussed the pros and cons quite in depth both theoretically and practically, I would like to add what I think is a big piece missing in this puzzle: resources. Or more straightforwardly, money and staff. (Well apparently, staffing resource basically can be translated to the money to hire qualified staff.)
I’ll start with the latest post in this thread. Suzanne wrote:
“
Maybe we’re coming to consensus:
*We need different models to answer different needs*! Let’s make more models tailored to more specific questions, not less. We need scenario tools, ABMs, one-off MNLs, simple linear regressions, and machine-learning. Let’s use more data and keep expanding our knowledge, like
Chandra noted! Let’s keep building our toolsets! And for all these tools and data sources, we need to work together and develop standards.
”
This reminds me of the “Let them eat cake” story. Please don’t get me wrong, actually I do think that sounds like a perfect solution, given the fact that an agency has enormous resources, but that’s not always the case, so I am actually jealous of what you got. I just found out that the data and modeling team at your MPO alone is more than twice the size of our whole MPO squad.
In my opinion, whether a modeling approach is good/suitable for an agency, has two dimensions. One is the model itself at the time of delivery and the other is its sustainability. For the first part, agencies may be able to invest a lump sum funding to develop a high-cost model (usually those more complex ones) at some point, however, to maintain and simply use it in the long run, it will require constant both financial input and staffing input. I also learned this lesson after I purchased my first house and I’m sure you don’t need more details. I will just throw one example/topic here. After running a model and getting the output numbers we need, I genuinely believe it is essential for us modelers to understand why we get these outputs as they look like, and, when we run different scenarios, to understand why these outputs are different from each other, “what is the causation or what are the causations of these differences”, and, when evaluating a specific project when a base-year number is off the ground truth more than we expected at that specific location, why is that. As a simple “practitioner” (like Paul referred to himself), with my knowledge set, I am capable of answering all those questions for a four-step model, whether purely based on the knowledge on the model structure or by conducting a further selected links analysis to find out, I can do that, for a four-step model, but I am sorry to admit, for an activity-based model, I am not capable of doing the same thing, at this time, given the fact that there are so many correlations behind the black box. I am sure there are many experts out there who know everything behind these black boxes, but I doubt all agencies can afford consulting them on a routine basis when using the model in the ways Jim has summarized earlier. I think this example also called back what Dan wrote “Complexity may be beautiful, and it certainly is complex. It thus may be inaccessible except to those who understand that complexity. But that does not mean that it is also correct.” and the synthetic Q&A Ken posted earlier:
“
Q2. Your model is predicting that auto VMT (or ridership) in
the corridor of interest increased by XX percent between the calibration year
and 2040, which is noticeably higher (or lower) than the predicted increase in
the corridor’s population and employment. What is the reason for this?
A2. We have no idea. It’s what the model says.
”
I just come up with an example to explain the potential side effect of not knowing what are behind the results. For example, when applying Policy #A to the model and we may get a negative impact toward our Goal #B, but it may actually be caused by the impact of Policy #A having on Factor #C. If we don’t know what’s behind it, we may simply conclude that Policy #A does not comply with our Goal #B. However, if we know the actual causation, we may be able to figure out a Strategy #D to eliminate such side impact and then apply both Policy #A and Strategy #D to the model which could potentially end up with a positive impact toward our Goal #B.
I seem to have gone beyond the topic of resources so I will stop here though I still have much to say. In the end, I do have a question for Peter as you wrote:
“Most of the US MPOs have already switched to ABMs and
abandoned their 4-steps. Most of the cities around the world still use 4-steps
and do not have an ABM yet.”
There are around 408 MPOs in the US and I am sure not all them have travel demand models, but for those who do, I am wondering if most of them have switched to ABMs. I really don’t know. Peter, would you mind providing more source for the information on that? I am just curious. Also, on the second part for the rest of the world, if you can provide some additional information too, I appreciate that. Thank you.
See you all in Indianapolis! 
Best,
Xinbo Mi
Evansville MPO
 
 
--
Full post: https://tmip.org/content/current-state-activity-based-models-and-next-steps
Manage my subscriptions: https://tmip.org/mailinglist
Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13960