Use the data in MURDER for this exercise.

(i) Using the years 1990 and 1993, estimate the equation mrdrteit 5 d0 1 d1d93t 1 b1execit 1 b2unemit 1 ai 1 uit, t 5 1, 2 by pooled OLS and report the results in the usual form. Do not worry that the usual OLS standard errors are inappropriate because of the presence of ai

. Do you estimate a deterrent effect of capital punishment?

(ii) Compute the FD estimates (use only the differences from 1990 to 1993; you should have 51 observations in the FD regression). Now what do you conclude about a deterrent effect?

(iii) In the FD regression from part (ii), obtain the residuals, say, e^i

. Run the Breusch-Pagan regression e^

2 i on Dexeci

, Dunemi and compute the F test for heteroskedasticity. Do the same for the special case of the White test [that is, regress e^

2 i on y^i

, y^

2 i , where the fitted values are from part (ii)]. What do you conclude about heteroskedasticity in the FD equation?

(iv) Run the same regression from part (ii), but obtain the heteroskedasticity-robust t statistics. What happens?

(v) Which t statistic on Dexeci do you feel more comfortable relying on, the usual one or the heteroskedasticity-robust one? Why?