Change, structure, and rank
Forms for questions about how things sit relative to each other — what moved most, what covaries with what, and which panels share a shape.
Share expressing a great deal of confidence, Democrats minus Republicans, in percentage points. Open dot is the 1973–80 average, filled dot the 2022–24 average; rows are sorted by how far the gap moved.
- Democrats lead now
- Republicans lead now
gap = Dem minus Rep weighted 'great deal' share, percentage points
Pairwise correlations among six policy positions. Mean r rises from 0.33 in the 1970s to 0.44 in the 2020s — positions that were once nearly independent now travel together.
1970s
3 of 15 pairs asked
1980s
mean r = 0.17
1990s
mean r = 0.20
2000s
mean r = 0.23
2010s
mean r = 0.35
2020s
mean r = 0.44
ANES cumulative file, weighted Pearson r. The dashed column is each issue's correlation with 7-point party ID — a different quantity than the issue-by-issue triangle, kept apart so it can't be averaged in by eye. Blank cells were not fielded.
Cohen's d for the South minus non-South gap on each measure, positive where the South sits further in the named direction. Panels share one y scale, so a flat line really is a small gap rather than a rescaled one. Color marks the verdict for each domain.
- realigned
- converged
- deepened
- persisted
ANES cumulative file; gaps expressed as standardized mean differences. Measures: Presidential vote (Democratic); Party identification (Democratic); Ideological self-placement (conservative); Church attendance (frequent); Aid to Black Americans (self-reliance); Traditional gender roles (GSS); Homosexuality 'always wrong' (GSS); Welfare spending 'too little' (GSS).
Grey dots are ZIP code scores; arrows are the variable loadings, co-scaled to the cloud, each named in the margin nearest its tip. Almost every disease and deprivation measure points left and they fan together tightly, so one component absorbs most of the variance. The handful pointing right — finance, education, binge drinking — are what the first component separates them from.
Correlation-based PCA (variables standardized). Scores and loadings co-scaled for a proper biplot.