Integrating stated and revealed choice data in the case of violation behavior

4 posts / 0 new
Last post
sasbhushan
Integrating stated and revealed choice data in the case of violation behavior

Dear all,

I have a question related to integrating stated and revealed choice data together into modelling. I want to get your suggestions.

 

For example we have revealed data on persons violating a traffic rule.

That is, we observed persons following it and persons violating it and surrounding factors.

Now we also performed a survey among people asking whether they will follow it or not in various scenarios.

The first one is revealed choice set and the second one is stated choice set.

We can probably use both sets separately for developing discrete choice models (binary). Is integrating both data help in any way? How do we integrate? Do we have to take some preliminary steps to make this integration possible?

Once we model these, what is the appropriate method to arrive at measures to prevent this violations?

Could you please give some suggestions on this.

Thank you.

Sasanka Bhushan

Associate Professor

VIT Vellore, India.

David A Nelson

Professor Bhushan,

I think you will find that stated vs revealed choice questions about illegal behavior will be trickier than in some other subject areas. That said, it isn't easy to get reliable information from people on many subjects. It has been found that most people are not great at conveying/remembering what they ate over the last few days without keeping a log which is filled out within a few hours of eating. With traffic violations, the complication comes in how honestly a person answers questions about illegal behavior, whether they believe in the legitimacy of the laws in question, whether they fear being shamed or punished (even in a purely internal sense), and so on. In other words, questions about the least controversial elements of a person's life may be answered more accurately than other questions. If you have an excellent way to get ground truth answers to transportation questions, this will likely reveal the disparity between survey questions and reality, but may not reveal socioeconomic factors well (other than make of vehicle).

David Nelson

circella

Dear all,

I will add a little more details - from a modeling perspective - on this
topic: if you conclude that the data can be treated together (but see
David's comments below about the issues that might specifically affect data
on traffic violations, and in general other reservations might apply on
stated-preference data collected on topics that might be ethically charged
and involve considerations of social desirability answers), it is
possible to estimate models with mixed data that include both revealed
preferences and stated preferences. Some modeling limitations apply,
including possible issues with differences in the scale parameters that are
usually confounded with the estimated coefficients in the estimation of a
discrete choice model. You might find the work from Juan De Dios Ortuzar
and other colleagues in this field inspirational to give guidelines on how
to treat this problem. Eventually, a change in model specification might
help - including moving from simpler model structures, e.g. a multinomial
logit model, to an artificial nested logit model, when desirable to solve
the issue.

Best,
Giovanni

*Giovanni Circella, Ph.D. *

*Honda Distinguished Scholar for New Mobility Studies, and*

*Director, 3 Revolutions Future Mobility Program*

*Institute of Transportation Studies*

*University of California, Davis *

*Phone: 1-(530)-554-0838 *

*gcircella@ucdavis.edu *

On Wed, Mar 11, 2020 at 3:07 PM David A Nelson
wrote:

> Professor Bhushan,
>
> I think you will find that stated vs revealed choice questions about
> illegal behavior will be trickier than in some other subject areas. That
> said, it isn't easy to get reliable information from people on many
> subjects. It has been found that most people are not great at
> conveying/remembering what they ate over the last few days without keeping
> a log which is filled out within a few hours of eating. With traffic
> violations, the complication comes in how honestly a person answers
> questions about illegal behavior, whether they believe in the legitimacy of
> the laws in question, whether they fear being shamed or punished (even in a
> purely internal sense), and so on. In other words, questions about the
> least controversial elements of a person's life may be answered more
> accurately than other questions. If you have an excellent way to get ground
> truth answers to transportation questions, this will likely reveal the
> disparity between survey questions and reality, but may not reveal
> socioeconomic factors well (other than make of vehicle).
>
> David Nelson
> --
> Full post:
> https://tmip.org/content/integrating-stated-and-revealed-choice-data-cas...
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13382
>

winufuwub

Hi there and thanks to Giovanni for the reference. The foremost experts in
joint RP/SP models are Jordan Louviere, David Hensher and Jeffrey Swait and
their 2000 book is still one of the best possible sources of info for
formulating and estimating joint RP/SP models, although a lot has been done
since (and quite a bit is summarised in our book with Luis Willumsen,
"Modelling Transport", Wiley 2011, Chapters 3, 7 and 8).
I agree with David Nelson (and Giovanni) that a complex issue here is
whether we can trust people stating choices about "unethical or illegal
behaviour", we have experienced this in the context of fare evasion, for
example, and there are no simple answers.

With best wishes

Juan de Dios
J. de D. Ortúzar
Emeritus Professor
Department of Transport Engineering and Logistics
Pontificia Universidad Catolica de Chile
www.ing.puc.cl/jos

El jue., 12 mar. 2020 a las 9:21, circella ()
escribió:

> Dear all,
>
> I will add a little more details - from a modeling perspective - on this
> topic: if you conclude that the data can be treated together (but see
> David's comments below about the issues that might specifically affect data
> on traffic violations, and in general other reservations might apply on
> stated-preference data collected on topics that might be ethically charged
> and involve considerations of social desirability answers), it is
> possible to estimate models with mixed data that include both revealed
> preferences and stated preferences. Some modeling limitations apply,
> including possible issues with differences in the scale parameters that are
> usually confounded with the estimated coefficients in the estimation of a
> discrete choice model. You might find the work from Juan De Dios Ortuzar
> and other colleagues in this field inspirational to give guidelines on how
> to treat this problem. Eventually, a change in model specification might
> help - including moving from simpler model structures, e.g. a multinomial
> logit model, to an artificial nested logit model, when desirable to solve
> the issue.
>
> Best,
> Giovanni
>
> *Giovanni Circella, Ph.D. *
>
> *Honda Distinguished Scholar for New Mobility Studies, and*
>
> *Director, 3 Revolutions Future Mobility Program*
>
> *Institute of Transportation Studies*
>
> *University of California, Davis *
>
> *Phone: 1-(530)-554-0838 *
>
> *gcircella@ucdavis.edu *
>
> On Wed, Mar 11, 2020 at 3:07 PM David A Nelson
> wrote:
>
> > Professor Bhushan,
> >
> > I think you will find that stated vs revealed choice questions about
> > illegal behavior will be trickier than in some other subject areas. That
> > said, it isn't easy to get reliable information from people on many
> > subjects. It has been found that most people are not great at
> > conveying/remembering what they ate over the last few days without
> keeping
> > a log which is filled out within a few hours of eating. With traffic
> > violations, the complication comes in how honestly a person answers
> > questions about illegal behavior, whether they believe in the legitimacy
> of
> > the laws in question, whether they fear being shamed or punished (even
> in a
> > purely internal sense), and so on. In other words, questions about the
> > least controversial elements of a person's life may be answered more
> > accurately than other questions. If you have an excellent way to get
> ground
> > truth answers to transportation questions, this will likely reveal the
> > disparity between survey questions and reality, but may not reveal
> > socioeconomic factors well (other than make of vehicle).
> >
> > David Nelson
> > --
> > Full post:
> >
> https://tmip.org/content/integrating-stated-and-revealed-choice-data-cas...
>
> > Manage my subscriptions: https://tmip.org/mailinglist
> > Stop emails for this post:
> https://tmip.org/mailinglist/unsubscribe/13382
> >
> --
> Full post:
> https://tmip.org/content/integrating-stated-and-revealed-choice-data-cas...
> Manage my subscriptions: https://tmip.org/mailinglist
> Stop emails for this post: https://tmip.org/mailinglist/unsubscribe/13382
>