1

How to obtain SE for covariates that enter a regression directly and through a interaction term?

Any suggestions of how to interpret the statistical significance of covariate that enters a regression directly and through a interaction term?

The problem, to put is simple, is that there are more than one standard error. The following very simple piece of code illustrates the problem.

data(trees) 
trees$isHigh <- trees$Height > 76 
trees$isThick <- trees$Girth > 13 
m <- lm(trees$Volume ~ trees$isHigh+trees$isThick+trees$isHigh:trees$isThick) 
summary(m)

I’d like to ask the question, for high trees, whether it makes a difference (for volume) whether a tree is thick.

(If my interpretation is correct, for low trees, i.e. for which trees$isHigh == FALSE, the answer is yes. For high trees, I have no clue.)

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
rm.
  • 111
  • 1
  • A related post [link](http://stats.stackexchange.com/questions/3653/adding-coefficients-to-obtain-interaction-effects-what-to-do-with-ses) has some discussion on this, but I would very much appreciate more detailed advice, e.g. in the form of code. – rm. Nov 09 '13 at 20:29

0 Answers0