Re: st: marginal effect of the interaction term in the probit model using margins command

[Ridhima Gupta <g.ridhima@gmail.com>]

Maarten to begin with I estimate a bivariate probit model but I arrive
at the conclusion that the model can be estimated as 2 separate
probits so logit is not an option.

The output of inteff gives you 4 things.

1)A stata table that reports the average partial effect of the interaction term

2) A stata dataset that has values for the interaction effect for each
of the observation

3)A graph that plots the predicted probablities along with
coefficintes on interaction terms for each observation

4)A graph that plots the predicted probabilities across Z statistics

If I choose to summarise the average partial effect on the interaction
term by 1 number  then am I right in understanding thatfor consistency
I will have to report the average partial effects for the other
regressors in my equation?

I was wondering what the if and in expressions in the syntax of the
inteff command refer to?

I will place the graph in the appendix.

Thanks,
Ridhima

On 6/30/11, Maarten Buis <maartenlbuis@gmail.com> wrote:
> On Thu, Jun 30, 2011 at 10:54 AM, Ridhima Gupta wrote:
>> I calculated the marginal effect on the interaction term using the
>> inteff command. My understanding is that the output reports the mean
>> interaction effect i.e. it reports the interaction effect when all
>> variables are held at the mean.
>
> No, -inteff- will compute a separate interaction effect for each
> observation. So if you have a 1000 observations, you'll end up with a
> 1000 interaction effects. If you look at the output (a graph) you'll
> see that the interaction effects varies a lot, from significantly
> positive, to non-significant, to significantly negative. As a
> consequence it makes no sense to try to summarize it with one number.
> If you want to report marginal effects and you want to report the
> marginal effect of the interaction term than your only choice is to
> show the graph produced by -inteff-.
>
> An alternative is to go for a logit model and report odds ratios. In
> that case you will have one effect (a ratio of odds ratios) for your
> interaction term, which makes it a lot easier to interpret and to
> report in a table. Downside is that you need to make sure that your
> audience/readers understand what an odds, an odds ratio, and a ratio
> of odds ratios mean. This is however not as hard as many people
> believe/fear. For an example that shows how that can be done 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
>
> --------------------------
> 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/
>
*
*   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/