Dear all and @ Jovo Kolev,
I have two explanatory variables (x1 and x2) that I would like to test whether they are substitutes or complements in explaining a dependent variable (y).
I am using weighted least squares regression to estimate a model with panel data (N=454, firms=87). I first estimated a model where with the mentioned two independent variables and an interaction term:
reg y c.x1##c.x2 x3 ... x10 i.year i.industry [aweight=wght2], vce(cluster firm)
The coefficients are: x1 is significantly positive, x2 is negative but not statistically significant, and the interaction term between x1 and x2 is significantly negative. I concluded that there is a substitution effect between x1 and x2. However, I was told this analysis is not correct and that I should use seemingly unrelated regression instead. My data is heterocedastic.
I ran the following model:
sureg (x1 y x3 ... x10 i.year i.industry) (x2 y x3 ... x10 i.year i.industry) [aweight=wght2], notable noheader corr
and I get correlation output positive and significant.
Is this latter approach correct?
Thank you
I have two explanatory variables (x1 and x2) that I would like to test whether they are substitutes or complements in explaining a dependent variable (y).
I am using weighted least squares regression to estimate a model with panel data (N=454, firms=87). I first estimated a model where with the mentioned two independent variables and an interaction term:
reg y c.x1##c.x2 x3 ... x10 i.year i.industry [aweight=wght2], vce(cluster firm)
The coefficients are: x1 is significantly positive, x2 is negative but not statistically significant, and the interaction term between x1 and x2 is significantly negative. I concluded that there is a substitution effect between x1 and x2. However, I was told this analysis is not correct and that I should use seemingly unrelated regression instead. My data is heterocedastic.
I ran the following model:
sureg (x1 y x3 ... x10 i.year i.industry) (x2 y x3 ... x10 i.year i.industry) [aweight=wght2], notable noheader corr
and I get correlation output positive and significant.
Is this latter approach correct?
Thank you
Comment