Re: st: wtpcikr for models not estimated in STATA

["P. Wilner Jeanty" <pwjeanty@gmail.com>]

On Wed, Nov 14, 2012 at 2:31 PM, sanja_lutzeyer@ncsu.edu
<slutzey@ncsu.edu> wrote:
> Hi Wilner.
> Thank you very much for the response.
> I am actually running a latent class model and so the logit is indeed
> the correct distributional assumption to assume in this case.
> There are two further queries I have based on your question.
> My variables (miles5, miles8, asc etc.) are all effects coded
> variables. miles5, for example, is only one of the 4 levels of the
> distance attribute and so I have the variables coded as follows:
>
>                     miles5 miles8 miles12
> miles5            1            0           0
> miles8             0           1            0
> miles12           0           0            1
> miles18          -1           -1          -1
>
> The asc variable is simply coded as 1 for the status quo and zero
> otherwise. And price is the only continuous variable in the model.
> Besides this, I have interactions between the different effects coded variables.

-wtpcikr- does not handle estimation results with interaction terms at
this point. Information contained in the interaction terms will be
lost.

> Do I enter the average based on the effects codes in the mymean()
> vector? Or is there a different approach to take?

-mymean()- will accept the mean values as long as they are placed into a matrix.

>
> Also, given I currently only have the varcov matrix and the
> coefficient estimates vector in stata - would it be enough to "trick"
> -wtpcikr- by trying to run a logit model which results in an error?

That is a legitimate concern. But remember, because -wtpcikr- feeds on
estimation results, the first thing it does is to verify whether a
model has been estimated.It will abort with an error when a model has
not been estimated prior to its call. Even if you estimate a logit
model, results from that estimation will not be part of the
calculations. Only the information provided using the bmat(), vmat(),
and mymean() options will be used. That is why it  is immaterial as to
what data go into the estimation.

> If not, I guess I'll just have to append the current dataset with a
> few data entries that would allow for running a logit.

As mentioned above, for that logit model, any 0/1 dependent variable
and any independent variables would do. Those can be generated
variables for example. However, if you are calculating the mean values
in Stata, you will need to bring in the correct dataset. If that is
the case, why don't you use Stata to carry out the analysis?

You can check with me off-list for further discussions.

>
> Thanks a ton for the help.
> Kind regards
> Sanja
> *
> *   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/

Best,

Wilner

___________________
The Kinder Institute for Urban Research & The Hobby Center for the
Study of Texas
Rice University
6100 Main Street
Houston, Texas, 77005
Email: pwjeanty@rice.edu
*
*   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/