0
$\begingroup$

Is there a way to test the significance of combined (interaction) coefficients in R?

I have a panel regression model of the following (simplified) form:

yi,t = b0 + b1xi,t + b2xi,t*phase_1t+ b3xi,t*phase_2t + li + kt + ei,t

where phase_1t and phase_2t are dummy variables equal to 1 in a non-overlapping subset of T. Hence, b1 captures the effect of x in Phase 0, b1 + b2 the effect in Phase 1 and b1 + b3 the effect in Phase 2.

The results are a weakly significant positive effect of b1 and a significant negative effect of b2. b3 is negative but insignificant. How can I know if the combined effect of b1 + b3 is significantly positive (or negative)? (i.e. does x have a significant positive/ negative effect in Phase 2?) I guess estimating the model without the b3 would be problematic, as b1 would then capture the effect of x in Phase 0 and Phase 2.

I currently estimate the model using the plm function with time (t) and country (i) fixed effects and interpret coefficients clustered at the country level.

$\endgroup$
0
$\begingroup$

I have solved the issue by just changing the coding of the dummy variables. That is, I code a dummy phase_0 instead of phase_2. The baseline coefficient now captures the total effect of x in Phase 2. The same approach would work for Phase 1.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.