HW2: Long-Term Effects of 9/11 on the Political Behavior. of Victims’ Families
In this exercise, we examine a hypothesis that individuals who lost someone in the terrorist
attacks of 9/11, whether a family relative or a neighbor, will have become more politically
engaged.
This exercise is based on: Hersh, E. D. 2013. “Long-Term Effect of September 11 on the Political
Behavior. of Victims’ Families and Neighbors.” Proceedings of the National Academy of Sciences
110(52): 20959-63.
We will examine this hypothesis using several different estimation techniques, focusing
throughout on the effect of the attacks on the victims’ families rather than their neighbors. The
CSV data file, victims9-11.csv, contains the following variables:
Name Description
voter.id Unique identifiers of relatives and neighbors of the victims
treatment Families and neighbors of actual victims (1) vs control group (0)
victim.status Families (2) vs neighbors (3) of victims and controls
ge20xx Voting in the 20xx general election (Y=at the polls, A=absentee, E=early,
M=by mail)
fam.members Number of family members living with voter at their address
age Voter’s age
party Voter’s party affiliation (D=Democrat, R=Republicans, N=no affiliation)
sex Voter’s sex
pct.white Proportion of non-Hispanic white voters living on the same block
median.income Median income of voters living on the same block
Voters were included in the data on the basis of their relationship to actual victims - these
constitute the two treatment groups - or if no such relationship existed but they were, otherwise,
sufficiently similar to voters in the treatment groups - this constitutes the control group.
Question 1
We begin by reformatting the data to facilitate our analysis. The three variables that contain non-
numerical values are sex, party, and voting records. Rewrite these variables using the following
coding rules: for the party variable, assign the value of to all Democrats, to all Republicans,
and for all other parties and non-affiliated citizens; for the sex variable, assign the value of to
all female citizens and to all male citizens; finally, for each of the seven voting record variables,
assign the value of if the individual voted in a given election, and otherwise.
Question 2
We will now test the hypothesis. Calculate the sample Average Treatment Effect of the terrorist
attacks on voter turnout separately for each of the seven general elections between 2000 and 2012.
Focus on families of victims as the treatment, and disregard the neighbor category. For each of
the seven point-estimates, compute the corresponding standard error. Assume that both the
treatment and control groups form. two random samples, and that the two samples are statistically
independent. Then, using the quantiles of the standard normal distribution, calculate the 95%
confidence intervals for each of the seven point-estimates. Plot the results where the horizontal
axis represents the elections. Provide a brief interpretation, with a particular focus on the meaning
of the confidence intervals you computed. The plot could look similar to the following.
Question 3
To examine the validity of the cross-sectional comparisons conducted in Question 2, check
whether possible confounders are balanced between the treatment and control groups. Compare
the means of the last six variables in the table above across the two groups along with their 95%
confidence intervals. Provide a brief interpretation of the results. What can you conclude about
the validity of the cross-sectional comparisons?
Question 4
Now, focus on the treatment group only. Compute the before-and-after estimate of the effect of
the terrorist attacks on voter turnout for families of the victims by using the 2000 general election
as a baseline for each of the subsequent six general elections. Next, calculate the standard errors
corresponding to the six point-estimates. Using the standard errors, compute the 95% confidence
intervals for each of the six point-estimates. Plot the results. Provide a brief interpretation of your
findings. Hint: Since you are now tracing the same sample evolve through time, you can no
longer assume the independence of sample means in calculating the standard errors; take care to
incorporate the covariance between voter turnout in 2000 and a subsequent election into your
calculations. The plot could look similar to the following.
Question 5
We will now repeat the same analysis as in Question 4, but now using voters in the control group
only. As before, be sure to restrict your analysis to family relatives of control victims rather than
their neighbors. What does this analysis imply about the validity of the analysis in the previous
question? The plot could look similar to the following.
Question 6
Calculate the difference-in-differences estimate for the general elections immediately surrounding
the attacks, 2000 and 2002. Calculate the standard error for your estimate, and provide the 95%
confidence interval. Once again, you are allowed to use the quantiles of the standard normal
distribution in calculating the confidence interval. Provide a brief interpretation of the result.
Question 7
Repeat the previous difference-in-differences analysis for all six elections following the attacks.
Throughout this analysis, use the 2000 election as the baseline as done in the previous question.
Plot the results with the horizontal axis representing different elections. Provide a substantive
interpretation of the results. The plot could look similar to the following.