Johns Hopkins University Econometrics Worksheet

1) In 1980, Cuba “exported” a large number of former prisoners to Miami, in what was called theMariel Boatlift. This increased the supply of labor in Miami by 7%. David Card proposed to usethis “natural experiment” to create a Differences in Differences estimate of the effect ofimmigration on the local unemployment rate for whites. To do so he compared Miami in 1979 and1981 versus a set of other similar cities that were not affected by the Boatlift. His resultsfor the unemployment rate for whites were as follows:1979 1981Miami 5.1% 3.9%Other cities 4.4% 4.3%

a) Calculate the Differences in Differences estimator of the effect of the Boatlift on theunemployment rate

b) Explain the assumptions that are needed if the Differences in Differences estimator is toprovide an unbiased estimate of the effect of the Boatlift

c) Do your results seem plausible? If so: explain why you think the assumptions in b) aresatisfied. If not: explain what could be wrong with the assumptions.

d) Suppose we implement this differences-in-differences estimator using a regression equation asfollows, where “Unemp” is the white unemployment rate as a percentage, “Miami” is a dummy =1 forMiami and =0 for comparison cities, and “Y81” is a dummy=1 for 1981 :Unemp = a + b*Miami + c*Y81 + d*(Miami*Y81) + errorUsing the results in the table, state the values of the regression estimates for parameters a,b, c, and d.

2) This exercise uses the data from EZUNEM.DTA, which are described in Example 13.8. The idea isto see if designating a city as an enterprise zone (which usually means granting big tax breaksto employers) will reduce the unemployment rate.

a) Run a regression of the log of the number of unemployment claims (luclms) against a set ofyear dummies and the enterprise zone variable (ez), using heteroskedasticity-robuststandard errors. What do you conclude about the effect of enterprise zones? What do thetime dummies tell you about trends in unemployment? What problems might this model have?

b) Run the same regression using first differences (FD) for luclms and for ez; still useheteroskedasticity-robust standard errors. Now what do you conclude about enterprisezones? Explain how this model differs from the model in a) and why that might matter.What problems might this model have?

c) Test the model in b) for the presence of serial correlation in the differenced errorterms, using the method Wooldridge introduces in Section 13.5. What do you conclude?

d) Rerun the model in b) with standard errors that are robust to heteroskedasticity AND toserial correlation of observations from the same city. What changes?

e) Rerun the regression using city fixed effects (FE), with heteroskedasticity and serialcorrelation robust standard errors. Now what is your conclusion?

f) Rerun the equation using city random effects (RE), with heteroskedasticity and serialcorrelation robust standard errors. Now what is your conclusion?g) After re-reading the section in Wooldridge called Fixed Effects or First Differencing andalso the section called Random Effects or Fixed Effects, explain which estimates (OLS, FD,FE, or RE) you think give the best estimate of the effect of enterprise zones on countyunemployment. Would you argue that enterprise zones have any effect, based on thisevidence?