Dominic Muston <[email protected]> has the following data:
. list
+--------------------------+
| exposed outcome n |
|--------------------------|
1. | 0 0 184 |
2. | 0 1 21 |
3. | 1 0 174 |
4. | 1 1 30 |
+--------------------------+
He reports he types
. logit outcome exposed [weight=n], or
<some output omitted>
-------------------------------------------------------...
outcome | Odds Ratio Std. Err. z P>|z| ...
-------------+-----------------------------------------...
exposed | 1.510673 .4585442 1.36 0.174 ...
-------------------------------------------------------...
Or if he types -cc outcome exposed [fweight=n]" he gets a similar
1.51 odds ratio.
But, he reports, if he types -blogit outcome n exposed, or- he gets a
different answer. Dominic writes,
> It's going to be something really obvious isn't it?
Yes.
-blogit- wants counts for the positive-outcome variable; -blogit- wants the
number of positive outcomes, not a 0/1 variable. Rather than
the second observation of the data reporting 21 trials, outcome=1,
-blogit- wants to hear 21 trials, 21 positive outcomes:
. gen cnt = outcome*n
. blogit cnt n exposed, or
<some output omitted>
-------------------------------------------------------...
_outcome | Odds Ratio Std. Err. z P>|z| ...
-------------+-----------------------------------------...
exposed | 1.510673 .4585442 1.36 0.174 ...
-------------------------------------------------------...
-- Bill
[email protected]
P.S. Actually, Dominic said he tried the -blogit- command with the
-noconstant- option. -noconstant- is not necessary. When you
specify the -or- option, the intercept is not reported, but it
is included in the model, as it should be.
<end>
*
* 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/
