Re: st: comparing logit (& mlogit?) coefficients with different dependent variables

[Maarten Buis <maartenlbuis@gmail.com>]

Yes you can, as long as you interpret the comparison as describtive
only and do not give it a causal interpretation. In your situation you
would probably not go for a causal interpretation anyhow, since it is
hard to come up with a meaningfull (thought-)experiment that would
correspond with such a comparison. So the "describtive only clause"
above is not much of a limitation in your case.

Ordered model are typically multinomial model with some restrictions
added, so you can use -mlogit- for ordered variables at the price  of
estimating more coefficients/loosing statistical power.

Hope this helps,
Maarten

On Wed, Sep 28, 2011 at 4:44 AM, M Hollis wrote:
> Taking it a step further, could one compare coefficients across different multinomial logit models, again with the same sample and same independent variables. The dependent variables in this case would also have the same structure (same number and types of possible outcomes).
>
> To be more concrete, I'm looking at predictors of whether an individual saw a change in occupational characteristics using three different measures of characteristics (cognitive skill, physical demands, social skill). In the logit case I would just be marking whether the person moved to a different skill level (dependent variable 1), moved to a job with a different physical demand level (dependent 2), or social skill level (dependent 3). I'd want to compare the coefficients of the independent variables in each of these three models.
>
> In the multinomial logit case, for each of the three types of characteristics I would look at three outcomes: an decline in the measure (e.g. cognitive skill requirements), no change (baseline group), or an increase. I realize these three outcomes could be considered ordinal, but I believe that the factors predicting each of the outcomes will be different and therefore the outcomes should have separate sets coefficients. I would, though, like to be able to compare the coefficients in these models across the three different mlogit estimates. Is this a bad idea?

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------

*
*   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/