In her concurring opinion in Whitford, Justice Kagan suggested that partisan gerrymandering might be unconstitutional not just because it dilutes the votes of a party’s supporters but also because it burdens their associational rights. “Members of the ‘disfavored party’ in the State,” she wrote, “may face difficulties fundraising, registering voters, attracting volunteers, generating support from independents, and recruiting candidates to run for office.” Since Whitford, plaintiffs in several gerrymandering cases have accepted Justice Kagan’s invitation to mount associational challenges. They have introduced testimony and other evidence from voters, candidates, and party officials that their associational activities have indeed been inhibited.

What these litigants haven’t done, though, is test the overall validity of Justice Kagan’s claim. They haven’t analyzed whether, as a general matter (as opposed to in a particular state), parties victimized by gerrymandering incur the injuries alleged by Justice Kagan. In a paper that Chris Warshaw and I just posted, we investigate precisely this issue. Today, I’ll describe our data and methods. Tomorrow, I’ll talk about our findings and their implications.

To capture the extent of gerrymandering, first, we use five separate measures of partisan advantage: the efficiency gap, the mean-median difference, the declination, partisan bias, and an aggregate of all these metrics. Through this inclusive approach, we’re able to bypass the ongoing debate about the measures’ respective strengths and weaknesses. We calculate the five metrics for congressional and state house elections from 1972 to 2016. This large dataset covers the vast majority of redistricting since the one person, one vote revolution of the 1960s.

Second, we use the following four variables to operationalize the associational harms that Justice Kagan cited in Whitford:

1. Contesting seats: The difference between the parties’ respective shares of contested seats. This is a measure of “recruiting candidates to run for office.”
2. Candidate quality: The difference between the aggregate quality of each party’s candidates (where incumbents and challengers who have previously held elected office are considered high-quality). This is also a measure of “recruiting candidates to run for office.”
3. Campaign contributions: The share of total campaign contributions received by each party’s candidates. This is a measure of “fundraising.”
4. Vote share: Each party’s mean vote share across all districts. This is a measure of “generating support from independents.”

Lastly, we run a series of difference-in-difference and dynamic panel models to evaluate the impacts of partisan advantage on these variables. All of these models include fixed effects for states and years; the latter also lag the outcome variables. The models thus compare different years within the same state and estimate how parties’ associational activities are affected when partisan advantage is relatively high or low for that state.