interactions and interpretations

[owooley]

Hello!
I'm working on a project with three [dummy] variables of interest. The variables can occur singly or together. I know that, technically, one should run all of these variables singly and together with interaction terms. My problem is that my sample size is too small to part everything out.
Is it kosher to run all of the variables singly and then add to get the full effect?
So, for example, if in one observation I have two variables "turned on," the total effect would then = coef(var1) + coef(var2) ??

Thanks in advance, Mike

[startz]

Unfortunately not.

[owooley]

ahh man, Thanks.

[startz]

Out of curiousity, how many observations do you have?

[owooley]

Well, you see, I'm attempting to measure the effect of a political scandal (from corruption) on economic activity.
So I had it set up with a scandal variable, whether the corruption was known or not, and whether the scandal resulted in prosecution.
Of course, neither of the other two could occur without the scandal.
So what I was trying to do is just run each of the variables separately then add/subtract depending on the scenario.
Then today I {think I} started thinking straight and ran it as four separate variables by doing them jointly: unknown corruption scandal, known corruption scandal, unknown with prosecution, known with prosecution.

Long story short, no beans.

There are, in all, 72 different scandals over 296 months but when you start to part it out it gets rather small. But that seems kind of absurd to say as I write it.

[startz]

So you have plenty of observations to identify four effects, it's just that the outcomes don't show much.

Your break down nests scandal/no scandal. You might try running on just the scandal dummy, or just the known scandal dummy to see if anything shows up. Maybe nothing will show. Or maybe you'll find that your data is powerful enough to detect the effect of a scandal, but not powerful enough to detect the differential effects of of different kinds of scandal.