Is Polarization a Myth?
Is Polarization a Myth? (DOI: https://doi.org/10.1017/s0022381608080493) was written by Alan I. Abramowitz and Kyle L. Saunders in 2008.
Authors attempt to test if voters hold partisan/ideological beliefs.
Fiorina ("not very well-informed about politics, do not hold many of their views very strongly, and are not ideological") and Converse hypothesized that ideology is an elite behavior. Recent body of research contradicts that.
- "studies have found that the political beliefs of Democratic and Republican voters have become much more distinctive... that political divisions within the public increasingly reflect differences in religious beliefs and practices... as well as deep-seated psychological orientations... and that ideological polarization among party elites is explained in part by ideological polarization among party supporters in the electorate..."
Authors attempt to test Fiorina's hypothesis: do voters hold partisan/ideological beliefs. Use ANES data.
- See replication notes for how things are coded.
The authors find that proportion of high polarization grew over time for general population; trend holds across most demographic/vote history splits, too.
Authors also find greater polarization, and greater change in polarization over time, in highly educated, likely voter, politically engaged subpopulations.
Reading notes
I believe that there's a non-negligible recency bias. The authors picked issue spaces that are polarized now, and it's not reasonable to assert that these were the polarized issue spaces in 1982. Polarization becomes a self-fulfilling prophesy.
As noted below, the polarization score is also not relative. Besides artificially deflating the detected polarization, this is an unreasonable constraint to put on a model. Taken to the extreme: in a polity where there is exactly one issue space, polarization of variable degrees can still exist. In this model, high polarization is impossible with any fewer than 3 issue spaces.
Replication notes
I was able to somewhat replicate the ANES analysis. See more below...
The important independent/dependent variables are coded as:
// Definitions of ideological positions recode VCF0803 (1 2 3=1 "Liberal") (4 9=2 "Moderate") (5 6 7=3 "Conservative") (else=.), generate(conserv) recode VCF0830 (1 2 3=1 "Aid") (4 9=2 "Moderate") (5 6 7=3 "Help self") (else=.), generate(blackaid) recode VCF0843 (1 2 3=1 "Decrease") (4 9=2 "Moderate") (5 6 7=3 "Increase") (else=.), generate(defspend) recode VCF0809 (1 2 3=1 "Guarantee") (4 9=2 "Moderate") (5 6 7=3 "Help self") (else=.), generate(entitlement) recode VCF0806 (1 2 3=1 "Public") (4 9=2 "Moderate") (5 6 7=3 "Private") (else=.), generate(healthins) recode VCF0839 (1 2 3=3 "Decrease") (4 9=2 "Moderate") (5 6 7=1 "Increase") (else=.), generate(govtspend) recode VCF0838 (1 2=3 "Illegal") (3 9=2 "Moderate") (4=1 "Legal") (else=.), generate(abortlegal) // Definition of polarization egen libpos = anycount(conserv blackaid defspend entitlement healthins govtspend abortlegal), values(1) egen conspos = anycount(conserv blackaid defspend entitlement healthins govtspend abortlegal), values(3) generate polar = abs(libpos - conspos) // Definition of high polarization recode polar (0 1 2 3=0 "Low to Moderate") (4 5 6 7=1 "High"), generate(r_polar) // Definitions of elites recode VCF0702 (1=0 "Non") (2=1 "Voter") (else=.), generate(voter) recode VCF0140A (1 2 3 4=1 "No college") (5=2 "Some college") (6 7=3 "College") (else=.), generate(educ)
There are significant gaps in the collection of these variables across years. A non-negligible number of cohorts do not have data for all 7 definitions of ideological positions, but the polarization score is not relative to account for this.
- In 1982, 1986, and 1990, only 6 collected.
- In 1998, only 5 collected.
- In 2002, only 1 collected.
The '2000s' data point necessarily is calculated solely from 2004 data.
See also their other paper.