EViews.com

Hi,

I have a question, regarding testing of coefficients across groups,

I want to see whether the effect of a certain indep. variable differs for two different groups. I've don the following and i was wondering if this is the correct way to do it in eviews:

i've run the following regression: Y = c X1 @expand(group,@dropfirst) @expand(group,@dropfirst)*X1
with X1 the indep. variable and groups is either 1 or 2.

In my opinion it gives me both the intercepts an slopes of the two groups. Anyway is the right way to test whether the slope between the two groups differ is the following:

-Do a wald coefficient test and test whether C(2) = C(4) (since these are the slopes for the 2 different groups on X1
-Low p-value of 0.034 is given which indicates that the slopes are indeed significantly differ.

Furthermore I've read that this test is not very reliable for smaller samples, is this the case?
Is the way i've done it right or conceptually wrong?

thanks a lot

Pitchforkp

This looks fine. It should work a well as any test of coefficients in a linear regression.

Hey startz, or anyone reading this,

I was wondering about the following (sorry about the long story)

I use a model with a K number of independent variables X regressed on Y. Now I want to test whether the effect of one of the predictor variables X differs between two groups in a binary variable.

I've tested for a slope difference of this particular predictor variable using an interaction term. and keeping the other independent variables fixed e.g I assume that the effects of those variables are the same for the two groups but the results are not significant, thus a t-test rejects a difference of the effect of the predictor variable between the two groups.

However I've also done the same regression, but this time i've used interaction variable for the other K predictor variables as well, since I suspect that there are significant differences in the effect of the other predictor variables on the two groups in other words i suspect that by using just one interaction variable the results will be biased.

--> now there is a significant difference in the effect of the predictor variable of interest. Furthermore some of the interaction variables of the other predictor variables also show significant differences. Using predictor variables for all predictor variables furthermore results in a much higher R2 and f-statistic.

Now my question is the following:
-Am i right to conclude that using seperate regressions (e.g interaction variables for all predictor variables) for the two groups does a better job in explaining the difference in the effect of the predictor variable of interest between the two groups?
-Is the right way to test the significance of the difference in the predictor variable of interest between the two groups still done by looking at the t-statistic and associated p-value?

I've read about this in woolridge, and in the part that "testing for differences in regression functions across groups" it's not mentioned how this works when you want to test for a difference in one interaction variable, i know how it works when you want to test whether the dependent variable follows the same model for two groups, but i wonder whether a significant difference in a particular interaction variable is evidence that the effect of this variable differs between the two groups

Anyway if someone knows, please enlighten me!

Interactions are formally just like any other variable in a linear regression. So using a t-test is fine. Since it sounds like a model with many interactions shows the interactions to be significant, that's probably the way to go.