To clarify what Maarten says, it does not only "make sense to have all
"lesser" interaction in the equation, rather it is a requirement to include
all additive terms and lower order interaction terms in order to estimate a
correct model. Excluding any of them is the same as pegging those
coefficients to zero, which may have severe impact on the remaining
coefficients.
Best,
Johan Hellstr=F6m
--- Ismail Ait Saadi <a.saadi@curtin.edu.my> wrote:
> I am using logistic model which consisits of 6 variables with 4
> intercation terms:
>
> the fist with 3 factors ( x1*x2*x3)
> the second with 2 factors (x1*x2)
> the third with 2 factors (x4*x5)
> the fourth with 3 factors (x1*x2*x6)
>
> So the model would be like this:
>
> logit (p)=3D b1x1+b2x2+b3x3+b4x4+b5x5+b6x6 +b7(
> x1*x2*x3)+b8(x1*x2)+b9(x4*x5)+b10(x1*x2*x6)
>
> I know that inteff command can be used only when there are two
> variables interacting with each other, but in my case I need to find
> interaction term and effect of three. furthermore there is more than
> one interaction in the model (4 interactions).
-inteff- will compute marginal effects, but you can just estimate the
model with -logit- by creating the interaction using -generate-. e.g.
the first interaction term could be generated with:
gen x1Xx2Xx3 =3D x1*x2*x3
Alternatively have a look at -findit xi3-.
The big problem with three way interactions is how to interpret the
results. One thing that always helps is to make sure that the value
zero for each x is meaningfull, e.g. the mean for a continuous
variable, or one category for a dichotomous variable (i.e. don't code
gender as 1 male, 2 female, but as 0 male, 1 female) Furthermore it
often makes sense to have all ``lesser'' interactions in there as well
(so also x1*x3, x2*x3, x1*x6, and x2*x6)
Hope this helps,
Maarten
-----------------------------------------
*
* 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/