Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Re: st: Clogit does not converge (Stata returns “not concave” for every iteration) – possible dummy variable problem in a DCE


From   James Buchanan <james.buchanan@dph.ox.ac.uk>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   RE: Re: st: Clogit does not converge (Stata returns “not concave” for every iteration) – possible dummy variable problem in a DCE
Date   Fri, 22 Mar 2013 15:51:10 +0000

Many thanks to both yourself and JVerkuilen for your comments. I've now increased the sample size to 504 patients (8064 choices) and the problem persists so I am ruling out sample size being an issue.

In response to your specific comments:

- I'm unclear as to what it means to have an unidentified model?
- Are the variables you added constant within the options? Each patient makes 16 choices and each choice is between two options. Each patient gets exactly the same 16 choices. I think this means that the answer to your question is "Yes".
- My choice sets have been derived from an efficient design. The actual choices are simulated - each patient makes a totally random choice between each choice pair. I don't think it's a particularly complex simulation? Regardless, I've tried many different combinations of random choices yet the problem persists.
- I've tried using factor variable notation but this makes no difference
- I've tried using the -difficult- option but this makes no difference
- I've downloaded the mixed logit function but I am unclear as to which variables should have fixed coefficients and which should be random - is there a process by which I should be making this decision?

Thank you again.

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Richard Williams
Sent: 22 March 2013 01:34
To: statalist@hsphsun2.harvard.edu
Subject: st: Re: st: Clogit does not converge (Stata returns “not concave” for every iteration) – possible dummy variable problem in a DCE

This sounds like an incredibly complicated simulation. Are you sure that you did it right? When I can't get models to converge it is usually because of problems with the data or with the model. I would run various descriptive statistics on the data. If you are convinced you are doing things right, try a different random seed for the simulation.

Sometimes adding the -difficult- option works miracles.

Also, I would use factor variable notation rather than create my own dummies. Partly because it is easier, but also because it gives you an additional error check, e.g. you'll know to be suspicious if you get
573 dummy variables when you were expecting 3.

Sent from my iPad

On Mar 21, 2013, at 7:51 PM, James Buchanan <james.buchanan@dph.ox.ac.uk> wrote:

> Totally skimpy. But it's only a simulation, and it probably helps to think of the simulation as 160 choices, rather than 10 patients. Regardless, I've tried running the same model on 200 patients (i.e. 3200 choices) and I get the same problem. I should have made the latter clear in my initial post, apologies.
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Richard 
> Williams
> Sent: 21 March 2013 23:39
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Re: st: Clogit does not converge (Stata returns “not 
> concave” for every iteration) – possible dummy variable problem in a 
> DCE
>
> I've never run a model like this, but isn't ten subjects a little on the skimpy side.?
>
> Sent from my iPad
>
> On Mar 21, 2013, at 5:17 PM, James Buchanan <james.buchanan@dph.ox.ac.uk> wrote:
>
>> I am analysing the results of a simulated discrete choice experiment. I want to make sure my data matrix is set up correctly prior to launching the DCE. I have simulated 10 patients, each of whom makes 16 choices (2 options in each choice). Each choice has 6 attributes, and each attribute has four levels. The first four attributes are continuous variables. If I regress these four attributes against my dependent variable (pref) using clogit everything works fine. The final two attributes are categorical variables. As soon as I add these variables as dummies (either standard (1,0) dummy variables or using effects coding (1,0,-1)) clogit refuses to converge. Each iteration returns a not concave message, and I get no results.
>>
>> The syntax I am using is: clogit Pref ASCa Time Cost Eff Rel Info1
>> Info2 Info3 Who1 Who2 Who3, group(Unique_Q)
>>
>> Pref = choice variable (1 or 0)
>> ASCa = alternative specific constant for option A Time, Cost, Eff, 
>> Rel = continuous attributes Info1, Info2, Info3 = categorical 
>> variable 1 converted to 3 dummy variables (there is an Info0 but this 
>> has been dropped to avoid the usual dummy variable problem) Who1, 
>> Who2, Who3 = categorical variable 2 converted to 3 dummy variables 
>> (Who0 dropped) Unique_Q = a unique id value for each choice
>>
>> If I add iter(20) to the end of the syntax, I see that the standard errors for Info2 and Info3 are undefined (“.”), and Stata warns me that convergence has not been achieved.
>>
>> My design is efficient (d-error of 0.051899), there is perfect level balance, minimal overlap, and there are no dominated alternatives. I am using the current version of Stata (12.1) on Windows 7.
>>
>> Does anybody have any idea why I'm getting this error?
>>
>>
>>
>> James Buchanan
>> NIHR Doctoral Research Fellow
>> Health Economics Research Centre
>>
>> t: 01865 289262 | e: james.buchanan@dph.ox.ac.uk | w:
>> http://www.herc.ox.ac.uk/people/james
>> Department of Public Health | University of Oxford | Old Road Campus 
>> | Headington | Oxford | OX3 7LF
>>
>> T: http://twitter.com/HERC_Oxford | F:
>> http://www.facebook.com/healtheconomicsresearchcentre
>>
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/statalist-faq/
>> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index