Transcript How Ethnic Diversity - Political methodology

Flooded Communities: Using the Post-Katrina
Migration to Estimate Contextual Effects
Introduction: Overcoming the Selection
Bias that Plagues Contextual Studies
Scholars have long inquired about the impact of
inter-group contact and proximity. Yet observational
studies in this field are commonly plagued by
selection bias and measurement error. This
research uses the post-Katrina migration as an
exogenous demographic shock to measure how
meeting or living near members of a different group
shapes people’s political views.
The Negative Effects of Living in a Host Community
Daniel Hopkins
Ph.D. Candidate
Department of Government
Harvard University
An alternative explanation is that the
observed differences affected everyone in
the community, not just those who made
contact. With U.S. Postal Service data on
evacuee relocation, I estimated the impact
of living in any community that took in
more than 1 evacuee per 200 residents, as
shown in the map on the left. Using OLS
models specified as below with standard
errors clustered by ZIP code and multiple
imputation, we see that those in affected
communities became more supportive of
anti-crime spending and less supportive
of anti-poverty spending. Notice also the
community-specific effects, since Houston
became anti-crime while Baton Rouge
became anti-poor.
Propensity Score Matching: The Small and
Negative Influence of Evacuee Contact
Drawing on a new, clustered phone survey of 3,879
residents in the U.S. South, the analysis first used
propensity score matching to pair respondents who
came into contact with evacuees with highly similar
individuals who happened to live outside the affected
area. This matching procedure significantly
strengthens the claim of ignorability—that the
potential outcomes are independent of the treatment
conditional on the covariates—since people in
affected communities are being compared to the
people who would have made contact had their
community been affected.
The classical approach compares respondents in the first column,
while this research compares respondents in the first row
Each point represents one respondent to the 2006 Social Capital
Community Benchmark Survey
Treatment effects of living in a community that took in
many evacuees for key attitudes
Estimating two sets of OLS regressions—one for the matched sample of
762, the other for the full sample of 1,605—I find that those who had
contact with the evacuees show few attitudinal differences. The
differences we do observe are in a negative direction, as those who
had contact became less supportive of spending on the poor. A
sample set of regression coefficients is below on the left, alongside the
estimated impact of contact on ten dependent variables of interest.
Using Differences-in-Differences to Rule Out
Community Fixed Effects
On account of the exogenous shock that affected
people in some areas but not others, it is
possible through matching to find a control
group of 381 in unaffected communities that was
virtually identical at baseline to the treated
group, as shown below.
Regression coefficients for support for anti-poverty spending
Treatment effects and 95% confidence intervals of intergroup contact for several evacuee-related attitudes
The central assumption underpinning these analyses is that
living in an affected community shapes attitudes only through
exposure to evacuees once we take covariates into account. To
relax that assumption, I used 416 respondents to the 2004
NES to simulate pre-Katrina attitudes for the treatment and
control groups. Subtracting the 2006 estimated treatment
effect from the 2004 difference, we have a differences-indifferences estimate which accounts for pre-existing
community effects. Even doing so, we reaffirm the earlier
findings. Respondents in Houston became more supportive of
anti-crime spending, and respondents in Baton Rouge became
less supportive of spending on the poor.
The central conclusion: those who lived in a community that took in evacuees became more
supportive of spending to fight crime, or else less supportive of anti-poverty spending.