[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: mlogit: predicted probabilities with confidence limits

From   "Austin Nichols" <>
Subject   Re: st: mlogit: predicted probabilities with confidence limits
Date   Wed, 26 Sep 2007 10:49:08 -0400

I have a feeling -spostado- can do this (net from  If it does, it might do a lot
of calculations something like this for every observation:

sysuse auto, clear
mlogit rep78 wei, baseout(5)
predict p1-p5
loc s "1"
forv i=1/4 {
 loc s="`s' +exp([`i']_b[wei]*2930+[`i']_b[_cons])"
nlcom 1/(`s')
di "LB: " el(r(b),1,1)-1.96*sqrt(el(r(V),1,1))
di "UB: " el(r(b),1,1)+1.96*sqrt(el(r(V),1,1))
li wei p5 in 1

On 9/26/07, Dedman, Dan <> wrote:
> I run a multinomial logistic regression model, where the outcome variable has three levels. From this I can obtain predicted probabilities for each level of the outcome. I would like to put confidence intervals around these probabilities, but I am not sure how to do it. I can obtain confidence intervals for the linear prediction (using the -stdp- option with -predict-), but I am not sure how to incorporate this information into the equations which relate the linear prediction to the probability:
> P(y=1) =         1/(1+exp(beta2*x)+exp(beta3*x))
> P(y=2) = exp(beta2*x)/(1+exp(beta2*x)+exp(beta3*x))
> P(y=3) = exp(beta3*x)/(1+exp(beta2*x)+exp(beta3*x))
> (where beta2*x is the linear prediction for outcome level 2, etc)
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index