I recently posted this question, about my dependent variable which looks as in the picture below, leading to the question in the link, about what type of regression to use.
My regression always included certain controls. For example the interaction between, country, company size and level of urbanisation. For computational reasons, I decided to demean my dependent variable with respect to that interaction. When I checked the distribution of my new dependent variable, it was to my surprise much smoother (see second picture).
By demeaning my variables, I am not doing much different than adding the interaction as a control (I am getting a slightly different estimate for my variable of interest though). Nevertheless all of a sudden I do not have to worry about the distribution of my dependent variable anymore (obviously I will have to correct the df)?
EDIT: I read here that demeaning the dependent variable causes attenuation bias. However,if I can use normal OLS, I do not actually have to demean my dependent variable in the first place.
