I am currently running an analysis withh a fixed effects regression on whether name similarities between investor and leadership of invested company has an impact on the weight that investors place on these companies in a given time period. The regression design looks like this:
$$y_{i,j,t} = \alpha + \beta_1 * SameName_{i,j,t} + Controls$$
So far I have not found any results and I would like to test whether less experienced investors (e.g., less than one year) show this bias compared to more experience. I am wondering whether I should run a separate regression with less experienced investors (e.g., subset for investors with less than 1 year experience) or include a dummy for less experienced investors and an interaction term such that the equation would look like this:
$$y_{i,j,t} = \alpha + \beta_1 * SameName_{i,j,t} + \beta_2 * LessExperience_{i,j,t} + \beta_3 * (SameName_{i,j,t} * LessExperienced_{i,j,t}) + Controls$$
From my understanding $\beta_3$ would capture the effect if less experienced investors have name similarities. If this is positive and significant, then this subgroup would fall for this bias. And I would expect SameName to be 0 and non-significant as it was in my first regression. The same for LessExperience because there is no rational explanation why this should have an effect on weights.
What are the differences of running seperate regressions and including the interaction terms? Are there any advantages of either method?
