st: Re: draw parameter values from distribution

["Chris Chung" <cchungc403@rogers.com>]

Dear Hans,

I had a similar problem. My problem was related to the size of dataset and 
time-consuming computing procedure (i.e. multiple failure-time analysis). 
Although I used a quad xeon 2.2 processor with 8 GB of RAM and 500 GB of 
drive space, I had to wait several days to get the result of multiple 
failure-time analysis. I was really surprised by the slow computing speed as 
I have yet to run more complicated models with more variables. However, the 
2 GB limit that we are experiencing is a Stata/Terminal Services limitation 
(or feature). Any single application running through Stata/Terminal Services 
can only allocate up to 2 GB of memory.  In this case, with a 32 bit 
application it doesn't really matter - the current 32 bit Stata for Windows 
can only allocate a theoretical 2 GB of memory anyway. When MS officially 
releases a 64 bit version of Windows, we will have a 64 bit compile of Stata 
for Windows which will be able to overcome the 2 GB limit. But the time has 
not come yet.

Chris

----- Original Message ----- 
From: "Kelchtermans, Stijn" <Stijn.Kelchtermans@econ.kuleuven.ac.be>
To: <statalist@hsphsun2.harvard.edu>
Sent: Tuesday, December 14, 2004 4:25 AM
Subject: st: draw parameter values from distribution



> All,
>
> I'm estimating a conditional logit model of schooling choice. I want to 
> obtain confidence intervals for the predicted choice probabilities.
>
> The approach I want to use is the following:
> based on the parameter estimates, get 1000 draws from the multivariate 
> normal distribution to generate 1000 alternative parameter values. Using 
> these values, I calculate 1000 choice probabilities. I then drop 2,5% of 
> the values in each tail to get the 95% CI for the choice probabilities.
>
> I can't find Stata commands that allow to automate this procedure. If I 
> use "drawnorm" to get the parameter value, I lose the dataset so I can't 
> calculate the predicted probabilities. The best I can think of is to put 
> the parameter draws in a matrix, read the dataset again and then calculate 
> the probabilities. This seems quite a hassle to program though.
>
> Any ideas to simplify this?
>
> Many thanks,
>
> Stijn Kelchtermans
> Katholieke Universiteit Leuven
> Belgium
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/ 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/