 Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: Referring to coefficients after mlogit

 From Maarten Buis To statalist@hsphsun2.harvard.edu Subject Re: st: Referring to coefficients after mlogit Date Mon, 15 Oct 2012 11:36:56 +0200

```You can refer to the coefficient of the variable x1 in the first
equation as [#1]_b[x1] and for the second equation: [#2]_b[x1].
Alternatively, these equations have names and you can refer to these
by [name1]_b[x1] and [name2]_b[x1]. Useful can also be the
-coeflegend- option present in most estimation commands, including
-mlogit-.

Substantively, I would simply discus the odds ratios. This avoids the
whole problem, see:
M.L. Buis (2010) "Stata tip 87: Interpretation of interactions in
non-linear models", The Stata Journal, 10(2), pp. 305-308.

Hope this helps,
Maarten

On Mon, Oct 15, 2012 at 11:24 AM, Korcan Kavusan
<k.kavusan@tilburguniversity.edu> wrote:
> Hello all,
>
> I am estimating a multinomial logit model with 3 possible outcomes (0, 1 ,
> 2). 0 is the base outcome. Mlogit gives two sets of results, one comparing
> 0 and 1 and the other 0 and 2. So each independent variable has 2
> coefficients, one for each comparison. I have a difficulty in correctly
> referring to these coefficients after the estimation.
>
> Specifically, Bowen (2010) suggests the code below to compute the value and
> significance of a moderating effect for each observation. The code is
> written for logit estimation. I want to adapt it to my mlogit model and do
> these computations for my two sets of results, but cannot figure out how
> tell the code to use the coefficient from the result set that compares 0 and
> 1 (and then separately from 0-2 comparison later). It is probably a small
> technical issue but for me now a real headache. I greatly appreciate any
> clue.
>
> The original code for the logit model:
>
> * Estimate logit model for binary dependent variable `dismissed'
> logit dismissed X1 X2 X12
> predict phat
> * Define values used in computing moderating effects local xb
> _b[X1]*X1+_b[X2]*X2+_b[X12]*X12+_b[_cons]
> local xb0 _b[X1]*X1+_b[X2]*X2+_b[_cons]
> local phat (exp(`xb')/(1+exp(`xb')))
> local phat0 (exp(`xb0')/(1+exp(`xb0')))
> gen phat0 = (exp(`xb0')/(1+exp(`xb0')))
> label var phat0 "Predicted probability (model excludes interaction
> variable)"
> local coef1 (_b[X1]+_b[X12]*X2)
> local coef2 (_b[X2]+_b[X12]*X1)
> * compute value of each moderating effect at each observation predictnl
> total=`phat'*(1-`phat')*(_b[X12]+(1-2*`phat')*`coef1'*`coef2'),
> se(se_total)
> predictnl structural = `phat0'*(1-`phat0')*((1-2*`phat0')*_b[X1]*_b[X2]),
> se(se_structural)
> predictnl secondary =
> `phat'*(1-`phat')*(_b[X12]+(1-2*`phat')*`coef1'*`coef2') ///
> -`phat0'*(1-`phat0')*(1-2*`phat0')*_b[X1]*_b[X2], se(se_secondary) label var
> total "Total Moderating Effect"
> label var secondary "Secondary Moderating Effect"
> label var structural "Structural Moderating Effect"
>
> References:
>
> Bowen, H. P. 2010. Testing Moderating Hypotheses in Limited Dependent
> Variable and Other Nonlinear Models: Secondary Versus Total Interactions.
> Journal of Management.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

--
---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
```