Re: st: xtlogit: panel data transformation's recast to double makes model incomputable

[Tom <tommedema@gmail.com>]

Hi Jay,

Just to be clear -- I am also "on Nick's side". Anyway I don't want to
be talking about sides, somehow that's somewhat associated with a
battle and that is absolutely unrelated to this fruitful discussion.

I want to stress that I did start with theory, and that it made me
choose a fixed effects logistic regression. I am not somehow "choosing
the model that gives me an answer". For example, I used - clogit,
group()- but that is identical to -xtlogit, fe- so there is no change
there. Moreover, I used -logit- but only to verify if this problem was
related to panel data or not. For your information, I already started
over.

Anyway, I think I may have a very important pointer that could help in
finding the cause of this issue:
when I run -ladder close- it returns absolutely nothing, i.e.:

 ladder close

Transformation         formula               chi2(2)       P(chi2)
------------------------------------------------------------------
cubic                  close^3                    .            .
square                 close^2                    .            .
identity               close                      .            .
square root            sqrt(close)                .            .
log                    log(close)                 .            .
1/(square root)        1/sqrt(close)              .            .
inverse                1/close                    .            .
1/square               1/(close^2)                .            .
1/cubic                1/(close^3)                .            .

I believe this is very much related to the problems I am having with
this variable, close. If I can determine why ladder refuses to give me
any transformations, then I may be able to figure out what is the
source of this issue.

I am not just trying to get an answer, I am interested in why ladder
is not returning any transformations because this may give me
information on the root cause of the problem.

Thanks again.

Tom

On Wed, Apr 3, 2013 at 2:13 PM, JVerkuilen (Gmail)
<jvverkuilen@gmail.com> wrote:
> On Wed, Apr 3, 2013 at 7:54 AM, Tom <tommedema@gmail.com> wrote:
>> I understand what you mean. Note that I did not use any of these other
>> models until unexpected computational problems appeared.
>
> This is crazily simple but try rescaling the relevant price data. I
> forget what the scale of them is, but try dividing them by, say, 1000.
> The QR decomposition which is used computationally effectively does
> this, but your horribly skewed and massively differently scaled data
> may well be overwhelming even it.
>
> That said, I'm with Nick. This problem needs a major rethink. There
> are simply too many levers being pulled right now to get a believable
> answer. Even if a model converges it is still got the attached funk of
> all the troubleshooting and tricks you needed to do to make things
> work.
>
> In order:
>
> -Regenerate the data file "clean". Make sure all variables are scaled
> appropriately and candidate transformations are chosen. (e.g., with
> price data, log transform). This rules out mistakes or problems with
> the datafile.
> -Then lay out the model assumptions and framework.
> -Then fit candidate models in a predictable sequence.
>
> You may well have done a lot of this, but going back to square one and
> checking in sequence will help nail down what the problem is. You may
> well end up back where you started, but the effort will not be wasted.
> *
> *   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/