(d)_(h).txt

d.

(i)
Since all the estimates of the OLS regression are statistically significant, we are positive that results of the regression lead to the same conclusion as in the previous point; the effect of both school construction and the election of a village head after 1992, bring an increase of health centre presence in a particular village. Moreover, from the OLS regression we can claim that the magnitude of TREATED and POST are rather similar, and that there is a slightly larger increase which is due to the TREATED (0.0499) vs. POST (0.0419), unlike in point (d), where there was a prevalence of health centres in villages where village head was elected post 1992 (0.062042) in comparison to the treated villages (0.053588).

(ii)---This is sort of a wild guess, so feel free to change it since I am not sure whether it is correct---Observing the results of the regression we can see that the effect of interaction variable is now larger, as would be expected since we have omitted the variable TREATED. The effect of POST dummy is negative, as more importance is given to the control years, which come after the year of the dummy variable. An explanation as to why we have not included variable TREATED is the fact that xtreg with option fe i(v_id) uses the within regression estimator and controls for different years, while controlling for TREATED (conducted in the 70s) might distort the regression results given that this is a setting with staggered implementation of treatment.

h.
Most of the unbalanced baseline controls included have statistically insignificant effects, except for the few of them that are significant at 5% and have a positive effect. Concerning the variables previously included in (e), we see that their estimates are all more or less the same, therefore we can claim that our results remain robust despite the inclusion of unbalanced baseline controls. Adding the lagged outcomes in (g), we notice that the variable controlling for year 1993 becomes insignificant, and that the more importance is assigned to the lagged variable, being the most significant one than to the rest of the variables in the model, whose effects have decreased. Our findings however remain the same and we can claim that the results are also robust to the inclusion of these lagged outcomes.