Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Questions on -mlogit-

From   "Clive Nicholas" <[email protected]>
To   [email protected]
Subject   Re: st: Questions on -mlogit-
Date   Sat, 27 Mar 2004 23:55:15 -0000 (GMT)

Ada Ma wrote:

> If your base category is Labour (i.e. Labour=0) then the Conservatives are
> likely to win if celtic==1, and to calculate the the probability of con==1
> given
> that celtic==1 (and ignoring everything else), the function you should use
> is:
> exp(1.96e-17)/(1+exp(1.96e-17))
> Not 1/1.96e-17 as you have used.  I don't know anything about the rest of
> your post.

That would be the calculation I would use if the coefficient was the log
of the odds. But, as I explained in my post, that coefficient _is_ the
odds-ratio, so there's no need to exponentiate it!

Moreover, the 'Celtic' coefficient in the Con vs. Lab model is as I
asserted it to be. Tory candidates are significantly _less_ likely to win
Scottish and Welsh by-elections than Labour ones: saying anything else
flies in the face of the electoral evidence we have (and not just in
by-elections). Notice that the z score on the odds ratio is _negative_:

> -----------------------------------------------------
>              |               Robust
>        bewin |        RRR   Std. Err.      z    P>|z|
> -------------+---------------------------------------
> con          |
>       celtic |   1.96e-17   2.21e-17   -34.21   0.000

and naturally so, since the model - rightly - estimates that by-elections
held in Scotland and Wales reduces the odds of a Tory victory there (at
least over the 1975-2000 period) to be comprehensively less than 1,
relative to Labour. (Remember: even odds = [P=.5].)

So, in effect, I'm still searching for an answer to my first question: why
is this odds-ratio so bizarrely low?

CLIVE NICHOLAS        |t: 0(44)191 222 5969
Politics              |e: [email protected]
Newcastle University  |
*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index