0
$\begingroup$

I have data with 3 columns: y represents dependent continuous variable, xcont represents independent continuous variable, xcat represents independent categorical variable with 3 levels: xcat1, xcat2, and xcat3. I am doing simple model with interaction:

y = xcont + xcat + xcon*xcat

The statistical software I am using (in my case, R) shows the standard table with coefficients, standard errors and p-values, respectively, like this (using made up numbers for simplicity):

xcont 5 (2) 0.05

xcat1 10 (2) 0.02

xcat2 18 (1.5) 0.01

...

Now, I see that xcont is significant. But it is significant for the full model (with xcat3 used as reference level for categorical variable xcat).

Is there a way I could explore significance of xcont for specific levels of xcat? I.e. estimate the coefficient and standard error? For example, I would like to conclude something like "conditioning on xcat1, xcont is significant at this level".

It seems like the simplest solution is to do the regression of:

y = xcont + xcat1 + xcont*xcat1

Is there a "smarter" way though?

Improve this question
$\endgroup$
2
  • $\begingroup$ You should have 6 estimated regression coefficients. Apart from 3 you listed, other 3 are: intercept, xcontxcat1, and xcontxcat2. $\endgroup$
    – user158565
    Commented May 30, 2017 at 2:22
  • $\begingroup$ You are looking for "simple main effects," and yes, the "smarter" way includes variance from the entire sample—this is precisely what simple main effects analyses are. There are guides on how to do it in SPSS, R, SAS, etc. if you Google for "simple main effects." $\endgroup$
    – Mark White
    Commented May 31, 2017 at 2:29

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.