[R-sig-DCM] segmenting consumers after a dcm

Wed Feb 2 16:36:37 CET 2011

When segmenting based on utility parameters derived from a choice model, you should probably rescale these utilities before clustering based on them, in order to account for differences in error variance. For example, you could use Sawtooth Software's zero-centered diffs transformation.

As to HB-logit in bayesm: The function "rhierMnlRwMixture" is a very flexible function for estimating individual MNL parameters. However, as Chris pointed out, data preparation can be cumbersome and requires solid R programming skills. Also, the transformation and interpretation of the output is not that straightforward.
>From my point of view, you could also modify rhierBinLogit (replace the binary logit likelihood function by an MNL Likelihood function). The code of rhierBinLogit is a bit easier to understand in my eyes, but of course you need both R programming and HB knowledge for using/modifying this function, too.

-----Ursprüngliche Nachricht-----
Von: r-sig-dcm-bounces at r-project.org [mailto:r-sig-dcm-bounces at r-project.org] Im Auftrag von Chris Chapman
Gesendet: Dienstag, 1. Februar 2011 16:31
An: walt at dataanalyticscorp.com; r-sig-dcm at r-project.org
Betreff: Re: [R-sig-DCM] segmenting consumers after a dcm

Yes, segmentation on HB estimated utilities is a common thing to do. There are some theoretical issues (for instance, HB assumes that a distribution applies across the population from which respondents are sampled, and this assumption at least partially conflicts with some ways in which segments may be conceived), and practical issues (they may be difficult to find).  In principle, however, HB utilities are probably as good as many other basis variables.

As for R, yes, the bayesm package can estimate an HB model in R for many kinds of choice models. However, it has very specific requirements for how data must be formatted (as lists of difference-coded vectors). In my experience, translating data into its format and verifying that it is correctly in that format has been a non-trivial exercise for one-off projects. (I've started work on code to help do that automatically but it is incomplete and probably won't be done for quite a while.)

If you want to make bayesm work, best bet is to check out its documentation in the package vignette, and also the appendices to the book by Rossi, Allenby, & McCulloch: Bayesian statistics and marketing (2005). If you're lucky enough to have data in a format close to theirs, it's not difficult to do a first pass.  (And for more learning, Greg Allenby usually teaches a tutorial on Bayesian methods including the bayesm package at the AMA's ART Forum in June.)

Good luck,

-- Chris

-----Original Message-----
From: r-sig-dcm-bounces at r-project.org [mailto:r-sig-dcm-bounces at r-project.org] On Behalf Of Data Analytics Corp.
Sent: Tuesday, February 01, 2011 5:16 AM
To: r-sig-dcm at r-project.org
Subject: Re: [R-sig-DCM] segmenting consumers after a dcm

Good morning,

I've gotten a lot of great responses to my query about the segmentation.  I understand the latent class application; I do that for other problems.  The question now is how do I estimate a choice model in R using HB?  Bayesm was recommended.  I haven't looked at this yet, but will this estimate a logit model?  Any examples or applications someone can share?

Thanks,

Walt

________________________

Walter R. Paczkowski, Ph.D.
Data Analytics Corp.
44 Hamilton Lane
Plainsboro, NJ 08536
________________________
(V) 609-936-8999
(F) 609-936-3733
walt at dataanalyticscorp.com
www.dataanalyticscorp.com
_____________________________________________________
On 2/1/2011 8:09 AM, Shan, Ming (GfK Kynetec US) wrote:
> Dear Walt,
>
>     Nowadays, it is not uncommon at all for people to conduct segmentation analysis after DCM.  But one has to run individual respondent level estimation (e.g. Hierarchical Bayes mdel) or group level estimation (e.g. Latent Class Analysis).  HB is a form of random effect model allowing each individual respondent to have varying coefficients but still "pool" the information from the others.  You would then run cluster analysis on your utilities to group respondents.  In the case of LCA, the segmentation is done while estimation is made.
>
>    Hope this makes sense and helps somewhat.
>
> Kindly,
> Ming Shan
>
> Direct:  +1 (314) 200 4499
> Mobile: +1 (314) 680 2525
>
> www.gfk-kynetec.com
>
>
> -----Original Message-----
> From: r-sig-dcm-bounces at r-project.org
> [mailto:r-sig-dcm-bounces at r-project.org] On Behalf Of
> r-sig-dcm-request at r-project.org
> Sent: Tuesday, February 01, 2011 5:00 AM
> To: r-sig-dcm at r-project.org
> Subject: R-SIG-DCM Digest, Vol 4, Issue 1
>
> Send R-SIG-DCM mailing list submissions to
>          r-sig-dcm at r-project.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
>          https://stat.ethz.ch/mailman/listinfo/r-sig-dcm
> or, via email, send a message with subject or body 'help' to
>          r-sig-dcm-request at r-project.org
>
> You can reach the person managing the list at
>          r-sig-dcm-owner at r-project.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of R-SIG-DCM digest..."
>
>
> Today's Topics:
>
>     1. segmenting consumers after a dcm (Data Analytics Corp.)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Mon, 31 Jan 2011 14:53:26 -0500
> From: "Data Analytics Corp."<walt at dataanalyticscorp.com>
> To: r-sig-dcm at r-project.org
> Subject: [R-sig-DCM] segmenting consumers after a dcm
> Message-ID:<4D471336.2090009 at dataanalyticscorp.com>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> Hi,
>
> I just completed a dcm (not a conjoint) for a client.  Now they would
> like to segment the consumers based on utilities from the model,
> claiming that they had this done for them before although they don't
> know what or how it was done.  My model had coefficients estimated
> across all consumers, not for each one separately.  I don't even think
> that makes sense to do for each individual.  Has anyone ever done
> segmentation after a dcm?
>
> Thanks,
>
> Walt
>
> ________________________
>
> Walter R. Paczkowski, Ph.D.
> Data Analytics Corp.
> 44 Hamilton Lane
> Plainsboro, NJ 08536
> ________________________
> (V) 609-936-8999
> (F) 609-936-3733
> walt at dataanalyticscorp.com
> www.dataanalyticscorp.com
>
>
>
> ------------------------------
>
> _______________________________________________
> R-SIG-DCM mailing list
> R-SIG-DCM at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-dcm
>
>
> End of R-SIG-DCM Digest, Vol 4, Issue 1
> ***************************************
>
>
> GfK SE, Nuremberg, Germany, commercial register Nuremberg HRB 25014;
> Management Board: Professor Dr. Klaus L. W?bbenhorst (CEO), Pamela
> Knapp (CFO), Dr. Gerhard Hausruckinger, Petra Heinlein, Debra A.
> Pruent, Wilhelm R. Wessels; Chairman of the Supervisory Board: Dr.
> Arno Mahlert This email and any attachments may contain confidential
> ...{{dropped:4}}
>
> _______________________________________________
> R-SIG-DCM mailing list
> R-SIG-DCM at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-dcm
>

_______________________________________________
R-SIG-DCM mailing list
R-SIG-DCM at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-dcm

_______________________________________________
R-SIG-DCM mailing list
R-SIG-DCM at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-dcm

GfK SE, Nuremberg, Germany, commercial register Nuremberg HRB 25014; Management Board: Professor Dr. Klaus L. Wübbenhorst (CEO), Pamela Knapp (CFO), Dr. Gerhard Hausruckinger, Petra Heinlein, Debra A. Pruent, Wilhelm R. Wessels; Chairman of the Supervisory Board: Dr. Arno Mahlert
This email and any attachments may contain confidential ...{{dropped:4}}