Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: How to get Odds Ratios rather than Relative Risk Ratios from -mlogit-


From   sjsamuels@gmail.com
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: How to get Odds Ratios rather than Relative Risk Ratios from -mlogit-
Date   Sat, 19 Sep 2009 21:09:33 -0400

You are welcome, Philip.  However it is dangerous to interpret each OR
or RRR in isolation.  Take for example a 2-level exposure (unexposed,
exposed) )  and a three category outcome, with the following
percentages, in order,
Outcome:            1     2      3
unexposed:       10% 10% 80%
exposed::            2%  8%  90%

If the unexposed group is the baseline predictor and outcome 1 is the
baseline outcome, the OR for outcome 2 vs outcome 1 is: 4.
So: the exposure must increase the probability of outcome 2, right?
No!  Outcome 2  has a lower percentage (8%)
 in the exposed group than in the unexposed group (10%)

The lesson is that you must consider the whole pattern of predicted
probabilities.

-Steve

On Fri, Sep 18, 2009 at 9:48 PM, Philip Burgess
<philip.burgess.uq@gmail.com> wrote:
> Steve;
>
> Excellent - it all reconciles. Very much appreciate your advice.
>
> Would be good if the online documentation for RRR could make it clear
> that RRR = OR in these circumstances.
>
> Again thanks!
>
> Philip
>
> On Sat, Sep 19, 2009 at 1:11 AM,  <sjsamuels@gmail.com> wrote:
>> In your 3x3 table,  compute the ORs for the four  2x2 tables  I
>> outlined (the baseline predictor category  and baseline outcome appear
>> in each) and compare to the corresponding RRRs.
>>
>> -Steve
>>
>> On Fri, Sep 18, 2009 at 10:10 AM, Philip Burgess
>> <philip.burgess.uq@gmail.com> wrote:
>>> Thanks Steve. I'm still a bit confused though.
>>>
>>> The situation I have is typcially a 3 level outcome category & a 3
>>> level predictor category.
>>>
>>> I looked at the
>>> http://www.ats.ucla.edu/stat/stata/output/stata_mlogit_output.htm
>>> notes and they say:
>>>
>>> "However, the exponentiated coefficient are commonly interpreted as
>>> odds ratios".
>>>
>>> But "interpretation" per se is perhaps the easy way out ....
>>>
>>> Thanks for your thoughts.
>>>
>>> Philip
>>>
>>>
>>>
>>> On Fri, Sep 18, 2009 at 10:53 PM,  <sjsamuels@gmail.com> wrote:
>>>> RRRs are ORs for the 2 x 2 table consisting of the baseline outcome
>>>> category, baseline predictor category, and the outcome and predictor
>>>> categories which index the RRR.  See the mlogit manual 3 x 2 example.
>>>>
>>>> -Steve
>>>> On Fri, Sep 18, 2009 at 12:16 AM, Philip Burgess
>>>> <philip.burgess.uq@gmail.com> wrote:
>>>>> I have a complex survey design defined as:
>>>>>
>>>>> svyset _n [pweight=mhsfinwt], jkrweight(wpm0101 - wpm0160)
>>>>> vce(jackknife) singleunit(missing)
>>>>>
>>>>> When I run the following command, I get RRRs:
>>>>>
>>>>> xi: svy jackknife, nodots : mlogit BAgrp_DX i.isevaus, baseoutcome(1) rrr
>>>>>
>>>>> I wanting to report ORs - but using -eform- option doesn't seem to work:
>>>>>
>>>>> xi: svy jackknife, nodots : mlogit BAgrp_DX i.isevaus, baseoutcome(1) eform(or)
>>>>>
>>>>> Any ideas?
>>>>>
>>>>> Thanks;
>>>>>
>>>>> Philip Burgess

Steven Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
845-246-0774

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index