Bet it didn’t take Musk that long to hack US counting machines - he’d have got it down 2 a click of a mouse! #X_It #HelloQuitteX
Bet it didn’t take Musk that long to hack US counting machines - he’d have got it down 2 a click of a mouse! #X_It #HelloQuitteX
E(y_it|X_i1,...,X_iT,alpha_i)==E(y_it|X_it,alpha_i). Completely different deal. You get somewhat further with the Mundlak device, but in some fields is not discussed, or when the focus is on other differences of the two modelling classes
E(y_it|X_i1,...,X_iT,alpha_i)==E(y_it|X_it,alpha_i). Completely different deal. You get somewhat further with the Mundlak device, but in some fields is not discussed, or when the focus is on other differences of the two modelling classes
you estimate (1) but condition the marginal effects of X_it on D_i (i.e., g1*X_it + g2*X_it*D_i).
if you get different estimates for b1 and b2 in both specifications, what is that telling you?
does interacting controls with a group classifier violate any fundamental assumptions?
you estimate (1) but condition the marginal effects of X_it on D_i (i.e., g1*X_it + g2*X_it*D_i).
if you get different estimates for b1 and b2 in both specifications, what is that telling you?
does interacting controls with a group classifier violate any fundamental assumptions?
D_i is a time-invariant group classification
C_it is an exogenous continuous variable of interest
X_it are controls
you estimate a two-way FE regression:
Y_it = a_i + t_i + b1*C_it + b2*C_it*D_i + g*X_it + e_it (1)
you're interested in whether b1=b2.
D_i is a time-invariant group classification
C_it is an exogenous continuous variable of interest
X_it are controls
you estimate a two-way FE regression:
Y_it = a_i + t_i + b1*C_it + b2*C_it*D_i + g*X_it + e_it (1)
you're interested in whether b1=b2.