9. Gerber and Green conducted a mobilization experiment in which calls from a large com-mercial phone bank urged voters in Iowa and Michigan to vote in the November 2002 election.23 The randomization was conducted within four blocks: uncompetitive con-gressional districts in Iowa, competitive congressional districts in Iowa, uncompetitive congressional districts in Michigan, and competitive congressional districts in Michigan. Table 4.3 presents results only for one-voter households in order to sidestep the complica-tions of cluster assignment.
(a) Within each of the four blocks, what was the apparent effect of being called by a phone bank on voter turnout?
(b) When all of the subjects in this experiment are combined (see the rightmost column of the table), turnout seems substantially higher in the treatment group than the control group. Explain why this comparison gives a biased estimate of the ATE.
(c) Using the weighted estimator described in Chapter 3, show the calculations used to provide an unbiased estimate of the overall ATE.
(d) When analyzing block randomized experiments, researchers frequently use regres-sion to estimate the ATE by regressing the outcome on the treatment and indicator variables for each of the blocks (omitting one indicator variable if the regression includes an intercept.) This regression estimator places extra weight on blocks that allocate approximately half of the subjects to the treatment condition (i.e., P. = 0.5)
