Ideological Realignment and Active Partisans in the American Electorate
Ideological Realignment and Active Partisans in the American Electorate (DOI: https://doi.org/10.1177/1532673X03259195) was written by Alan I. Abramowitz and Kyle L. Saunders in 2004.
Citizens that participate in political activities beyond voting are expected to have an outsized impact. They are recruited by party leaders through several mechanisms, one of which is "attempting to use ideological appeals to build new electoral coalitions".
The use of ideological appeals has an effect on the partisanship of a party. A persuasion effect sees active citizens adopting political opinions in line with the leader. A "selective recruitment/derecruitment" effect sees active citizens sorted across parties.
The authors attempt to test for significant differences between active Republicans and active Democrats in the U.S., with the expectation that these active citizens drive the larger parties in the same directions. Use ANES data.
- See replication notes for how things are coded.
The authors find a correlation between activism and partisanship, even after controlling for demographic/structural factors such as education and being in the South. They also find significant differences in opinions towards the Clinton impeachment--"by far the most salient and controversial issue in American politics during the 1998 election campaign"--when splitting parties by activism.
Replication notes
I was able to mostly replicate the ANES analysis. It's possible that the differences are a result of cases being removed from cumulative data files, and original files as old as 2004 are extremely difficult to source.
The important independent/dependent variables are coded as:
// Definition of party membership (self-reported) recode VCF0302 (5=1 "Dem") (2 3 4 8 9=2 "Indep") (1=3 "Rep") (else=.), generate(party) // Definitions of activities recode VCF0702 (0 1=0 "Non") (2=1 "Voter") (else=.), generate(voter) recode VCF0717 (0 1=0 "Non") (2=1 "Influencer") (else=.), generate(influencer) recode VCF0720 (0 1=0 "Non") (2=1 "Influencer") (else=.), generate(displayer) recode VCF0718 (0 1=0 "Non") (2=1 "Influencer") (else=.), generate(attender) recode VCF0719 (0 1=0 "Non") (2=1 "Influencer") (else=.), generate(worker) recode VCF0721 (0 1=0 "Non") (2=1 "Influencer") (else=.), generate(donor) // Activity score egen activity = rowtotal(voter influencer displayer attender worker donor) recode activity (0=0) (1=1) (2=2) (3 4 5 6=3 "3+"), generate(r_activity) // Definition of partisan generate partisan = voter & (influencer | displayer | attender | worker | donor)
The authors then use a Poisson regression and the clarify statistical software to calculate first differences. Which is still a TODO for me!
See also their other paper.