Chapter 1.4

Cohort change in partisan sorting

Conservative Democrats and liberal Republicans fell from 19.6% to 6.3% of partisans who place themselves on the ideology scale from 1972 to 2024; the fall appears inside cohorts born 1927 to 1974, and how much of it is cohort replacement depends on assumptions the data cannot test.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 30 September 2026

DraftFigures and numbers may change as the analysis is rebuilt.what is still being checked

This draft describes birth cohorts and cannot follow individuals. It does not yet use a panel study that re-interviews the same people.

In 1972, 19.6% of the Democrats and Republicans in the American National Election Studies (ANES) who placed themselves on the seven-point liberal-conservative scale were conservative Democrats or liberal Republicans, whom this chapter calls cross-pressured partisans. In 2024 the share was 6.3%.The word has another meaning in Chapter 15.4, Partisans in the other party's territory (planned), which is about partisans who live where their party loses. The correlation between party identification and ideology is the subject of Chapter 1.1, Party and ideology, and Chapter 1.3, Sort dates by attitude dates the sort attitude by attitude. How tightly issue positions bundle inside each party is the subject of Chapter 1.5, Constraint among issue positions (planned). This chapter describes the fall by birth cohort. Two accounts differ in where it occurs: the older birth cohorts that held those mixed positions were replaced by younger cohorts that did not, or the share fell inside the cohorts that held them. The fall appears inside the cohorts born 1927 to 1974, and the share of it that replacement accounts for depends on assumptions the data cannot test.

Under replacement, each cohort keeps roughly the rate it had when it entered the electorate, and the national share falls because high-rate cohorts leave the sample while low-rate cohorts arrive. Under within-cohort change, the rate falls inside each cohort as the years pass. The ANES draws a new sample at each election, so it cannot follow the same people and cannot show that any individual switched. It can show whether the fall occurs within groups of people born in the same years, so every statement in this chapter is about cohorts.

A placed partisan is a Democrat or Republican, with leaners counted with the party they lean toward, who gives a position on the scale. Positions on the liberal side are liberal, the midpoint is moderate and positions on the conservative side are conservative. A fully sorted partisan is a liberal Democrat or a conservative Republican, the pattern that party sorting produces. The chapter uses the 22 ANES study years since 1972 that carry the item, which gives 39,662 placed partisans, and every estimate uses the survey weights.Each interval is a confidence interval computed with the Wilson method [1] on the Kish effective sample size [2]. The intervals treat the weighted sample as a simple random sample and ignore clustering and stratification, which can raise or lower the variance. The last section says which verdicts a larger variance would change.

Figure 1 plots the yearly share. From 1972 to 1990 it stayed between 17.0% (95% CI 15.1% to 19.0%) in 1984 and 20.3% (95% CI 17.8% to 23.1%) in 1980. It was 19.7% (95% CI 17.5% to 22.0%) in 1990, 11.5% (95% CI 9.4% to 14.0%) in 2004 and 5.2% (95% CI 4.5% to 5.9%) in 2020, and the fall from 1972 to 2024 is 13.3 (95% CI 10.9 to 15.6) points. The share in 2024, 6.3% (95% CI 5.4% to 7.4%), differs from the 2020 share by +1.2 (95% CI −0.1 to 2.4) points, an interval that includes zero and runs up to 2.4 points, so the data cannot separate a plateau from a rise of up to that size. The 2024 sample mixes interview modes, and the notes to the figure give the estimate for each mode. The 2000 point comes from a study that put the item to a subsample of respondents, which the section on unplaced partisans describes.

Figure 1

Cross-pressured partisans fell from 19.6% of placed partisans in 1972 to 6.3% in 2024

Percent of placed partisans who are conservative Democrats or liberal Republicans, ANES 1972-2024, with 95% intervals

Show data

Data for Cross-pressured partisans fell from 19.6% of placed partisans in 1972 to 6.3% in 2024
Measure	Year	Share (%)	Interval low	Interval high	Respondents in the base
Cross-pressured share of placed partisans	1972	19.6	17.6	21.8	1,371
Cross-pressured share of placed partisans	1974	18.4	15.7	21.3	977
Cross-pressured share of placed partisans	1976	17.8	15.8	20.0	1,306
Cross-pressured share of placed partisans	1978	18.0	16.1	20.0	1,430
Cross-pressured share of placed partisans	1980	20.3	17.8	23.1	892
Cross-pressured share of placed partisans	1982	18.1	15.6	20.9	806
Cross-pressured share of placed partisans	1984	17.0	15.1	19.0	1,401
Cross-pressured share of placed partisans	1986	18.5	16.5	20.5	1,441
Cross-pressured share of placed partisans	1988	18.6	16.6	20.8	1,286
Cross-pressured share of placed partisans	1990	19.7	17.5	22.0	1,220
Cross-pressured share of placed partisans	1992	16.9	15.2	18.8	1,630
Cross-pressured share of placed partisans	1994	13.6	11.7	15.8	1,279
Cross-pressured share of placed partisans	1996	14.5	12.5	16.9	1,233
Cross-pressured share of placed partisans	1998	15.6	13.2	18.3	922
Cross-pressured share of placed partisans	2000	16.0	13.0	19.5	603
Cross-pressured share of placed partisans	2002	14.9	12.6	17.5	1,171
Cross-pressured share of placed partisans	2004	11.5	9.4	14.0	849
Cross-pressured share of placed partisans	2008	11.3	9.4	13.4	1,448
Cross-pressured share of placed partisans	2012	9.8	8.7	11.0	4,613
Cross-pressured share of placed partisans	2016	7.2	6.2	8.4	2,949
Cross-pressured share of placed partisans	2020	5.2	4.5	5.9	6,353
Cross-pressured share of placed partisans	2024	6.3	5.4	7.4	4,482
Conservative share of placed Democrats	1972	24.6	21.7	27.8	776
Conservative share of placed Democrats	1974	23.0	19.2	27.1	573
Conservative share of placed Democrats	1976	24.1	21.0	27.4	725
Conservative share of placed Democrats	1978	23.1	20.4	26.0	867
Conservative share of placed Democrats	1980	28.0	24.3	32.0	515
Conservative share of placed Democrats	1982	25.5	21.8	29.6	479
Conservative share of placed Democrats	1984	22.3	19.4	25.4	719
Conservative share of placed Democrats	1986	24.6	21.7	27.7	793
Conservative share of placed Democrats	1988	27.4	24.1	31.0	638
Conservative share of placed Democrats	1990	25.9	22.7	29.3	673
Conservative share of placed Democrats	1992	23.4	20.7	26.3	871
Conservative share of placed Democrats	1994	22.6	19.2	26.5	622
Conservative share of placed Democrats	1996	21.8	18.5	25.4	667
Conservative share of placed Democrats	1998	21.5	17.9	25.6	515
Conservative share of placed Democrats	2000	22.3	17.6	27.7	319
Conservative share of placed Democrats	2002	21.1	17.3	25.6	568
Conservative share of placed Democrats	2004	16.1	12.7	20.2	438
Conservative share of placed Democrats	2008	16.6	13.7	19.9	894
Conservative share of placed Democrats	2012	14.5	12.7	16.4	2,709
Conservative share of placed Democrats	2016	10.3	8.6	12.3	1,467
Conservative share of placed Democrats	2020	6.6	5.5	7.8	3,277
Conservative share of placed Democrats	2024	8.2	6.7	10.0	2,310
Liberal share of placed Republicans	1972	13.1	10.6	16.1	595
Liberal share of placed Republicans	1974	11.4	8.3	15.4	404
Liberal share of placed Republicans	1976	9.8	7.6	12.5	581
Liberal share of placed Republicans	1978	10.1	7.9	12.9	563
Liberal share of placed Republicans	1980	9.8	7.2	13.2	377
Liberal share of placed Republicans	1982	7.3	5.0	10.7	327
Liberal share of placed Republicans	1984	11.4	9.3	14.0	682
Liberal share of placed Republicans	1986	11.0	8.8	13.6	648
Liberal share of placed Republicans	1988	9.9	7.8	12.4	648
Liberal share of placed Republicans	1990	12.1	9.6	15.1	547
Liberal share of placed Republicans	1992	9.5	7.6	11.8	759
Liberal share of placed Republicans	1994	5.1	3.5	7.3	657
Liberal share of placed Republicans	1996	6.1	4.3	8.7	566
Liberal share of placed Republicans	1998	8.3	5.9	11.7	407
Liberal share of placed Republicans	2000	8.7	5.8	13.0	284
Liberal share of placed Republicans	2002	8.4	6.1	11.4	603
Liberal share of placed Republicans	2004	6.6	4.4	9.7	411
Liberal share of placed Republicans	2008	5.2	3.5	7.8	554
Liberal share of placed Republicans	2012	4.6	3.5	5.9	1,904
Liberal share of placed Republicans	2016	4.0	3.0	5.4	1,482
Liberal share of placed Republicans	2020	3.7	3.0	4.7	3,076
Liberal share of placed Republicans	2024	4.4	3.3	5.8	2,172
Fully sorted share of placed partisans	1972	43.6	41.0	46.3	1,371
Fully sorted share of placed partisans	1974	46.0	42.4	49.6	977
Fully sorted share of placed partisans	1976	46.9	44.2	49.7	1,306
Fully sorted share of placed partisans	1978	46.7	44.1	49.3	1,430
Fully sorted share of placed partisans	1980	50.2	46.9	53.5	892
Fully sorted share of placed partisans	1982	47.5	44.1	51.0	806
Fully sorted share of placed partisans	1984	51.6	49.0	54.2	1,401
Fully sorted share of placed partisans	1986	46.1	43.5	48.7	1,441
Fully sorted share of placed partisans	1988	51.8	49.1	54.5	1,286
Fully sorted share of placed partisans	1990	44.8	42.1	47.6	1,220
Fully sorted share of placed partisans	1992	53.2	50.7	55.6	1,630
Fully sorted share of placed partisans	1994	53.2	50.2	56.2	1,279
Fully sorted share of placed partisans	1996	55.0	51.9	58.0	1,233
Fully sorted share of placed partisans	1998	50.2	46.7	53.7	922
Fully sorted share of placed partisans	2000	54.2	49.7	58.5	603
Fully sorted share of placed partisans	2002	57.7	54.3	61.1	1,171
Fully sorted share of placed partisans	2004	57.0	53.3	60.6	849
Fully sorted share of placed partisans	2008	63.1	60.0	66.1	1,448
Fully sorted share of placed partisans	2012	60.6	58.7	62.5	4,613
Fully sorted share of placed partisans	2016	68.8	66.8	70.7	2,949
Fully sorted share of placed partisans	2020	71.7	70.2	73.1	6,353
Fully sorted share of placed partisans	2024	69.3	67.3	71.2	4,482
Moderate share of placed partisans	1972	36.8	34.2	39.3	1,371
Moderate share of placed partisans	1974	35.6	32.2	39.2	977
Moderate share of placed partisans	1976	35.3	32.7	38.0	1,306
Moderate share of placed partisans	1978	35.3	32.9	37.8	1,430
Moderate share of placed partisans	1980	29.5	26.6	32.6	892
Moderate share of placed partisans	1982	34.4	31.2	37.7	806
Moderate share of placed partisans	1984	31.4	29.0	33.9	1,401
Moderate share of placed partisans	1986	35.5	33.0	38.0	1,441
Moderate share of placed partisans	1988	29.6	27.2	32.2	1,286
Moderate share of placed partisans	1990	35.5	32.9	38.2	1,220
Moderate share of placed partisans	1992	29.9	27.8	32.2	1,630
Moderate share of placed partisans	1994	33.1	30.4	36.0	1,279
Moderate share of placed partisans	1996	30.5	27.7	33.4	1,233
Moderate share of placed partisans	1998	34.2	31.0	37.6	922
Moderate share of placed partisans	2000	29.9	26.0	34.1	603
Moderate share of placed partisans	2002	27.4	24.4	30.6	1,171
Moderate share of placed partisans	2004	31.5	28.2	35.0	849
Moderate share of placed partisans	2008	25.6	22.9	28.5	1,448
Moderate share of placed partisans	2012	29.6	27.9	31.4	4,613
Moderate share of placed partisans	2016	24.0	22.2	25.9	2,949
Moderate share of placed partisans	2020	23.2	21.8	24.6	6,353
Moderate share of placed partisans	2024	24.4	22.6	26.2	4,482
Cross-pressured share of placed partisans	1972	19.6	17.6	21.8	1,371
Cross-pressured share of placed partisans	1974	18.4	15.7	21.3	977
Cross-pressured share of placed partisans	1976	17.8	15.8	20.0	1,306
Cross-pressured share of placed partisans	1978	18.0	16.1	20.0	1,430
Cross-pressured share of placed partisans	1980	20.3	17.8	23.1	892
Cross-pressured share of placed partisans	1982	18.1	15.6	20.9	806
Cross-pressured share of placed partisans	1984	17.0	15.1	19.0	1,401
Cross-pressured share of placed partisans	1986	18.5	16.5	20.5	1,441
Cross-pressured share of placed partisans	1988	18.6	16.6	20.8	1,286
Cross-pressured share of placed partisans	1990	19.7	17.5	22.0	1,220
Cross-pressured share of placed partisans	1992	16.9	15.2	18.8	1,630
Cross-pressured share of placed partisans	1994	13.6	11.7	15.8	1,279
Cross-pressured share of placed partisans	1996	14.5	12.5	16.9	1,233
Cross-pressured share of placed partisans	1998	15.6	13.2	18.3	922
Cross-pressured share of placed partisans	2000	16.0	13.0	19.5	603
Cross-pressured share of placed partisans	2002	14.9	12.6	17.5	1,171
Cross-pressured share of placed partisans	2004	11.5	9.4	14.0	849
Cross-pressured share of placed partisans	2008	11.3	9.4	13.4	1,448
Cross-pressured share of placed partisans	2012	9.8	8.7	11.0	4,613
Cross-pressured share of placed partisans	2016	7.2	6.2	8.4	2,949
Cross-pressured share of placed partisans	2020	5.2	4.5	5.9	6,353
Cross-pressured share of placed partisans	2024	6.3	5.4	7.4	4,482

The share moved between 17.0% and 20.3% in every study year to 1990, then fell to 11.5% in 2004 and 5.2% in 2020. Each point is a weighted share of the placed partisans interviewed that year, and the bands are confidence intervals. The By party tab splits the share by party and the Sorted and moderate tab sets it beside the fully sorted and moderate shares.Note: There was no ANES study in 2006, 2010 or 2014, so the line joins the neighboring study years.Note: The 2000 point rests on 603 placed partisans, the fewest in any year. The seven-point item went to a subsample of respondents that year: 52.1% of respondents have no recorded answer, and 35.6% of partisans placed themselves.Note: The ANES adds an online sample in 2012, and 2020 has no separate weights by interview mode. In 2024 the share is 7.2% in the face-to-face sample, 4.6% in the web sample and 7.1% among the 2,049 panel and paper cases (45.7% of the placed partisans in 2024), which have no separate weights by interview mode.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included

The fall appears in both parties (the By party tab of Figure 1). Conservatives were 24.6% (95% CI 21.7% to 27.8%) of placed Democrats in 1972 and 8.2% (95% CI 6.7% to 10.0%) in 2024. Liberals were 13.1% (95% CI 10.6% to 16.1%) of placed Republicans in 1972 and 4.4% (95% CI 3.3% to 5.8%) in 2024. The 2024 share is 0.33 (95% CI 0.26 to 0.42) of the 1972 share among Democrats and 0.33 (95% CI 0.23 to 0.47) among Republicans, although conservative Democrats started higher and fell further in points. Over the same years the fully sorted share rose from 43.6% (95% CI 41.0% to 46.3%) to 69.3% (95% CI 67.3% to 71.2%) and the moderate share fell from 36.8% (95% CI 34.2% to 39.3%) to 24.4% (95% CI 22.6% to 26.2%) (the Sorted and moderate tab). This is the pattern in Levendusky's account of the partisan sort, in which liberals became Democrats and conservatives became Republicans [3].

The fall appears in cohorts born 1927 to 1974

Figure 2 traces one birth cohort per panel, and Figure 3 gives each cohort's change from its first to its last plotted decade. The 1943 to 1958 cohort went from 17.6% (95% CI 15.7% to 19.6%) in the 1970s to 8.9% (95% CI 7.5% to 10.5%) in the 2010s and 5.0% (95% CI 4.0% to 6.1%) in the 2020s, a change of −12.6 (95% CI −14.9 to −10.4) points. The cohorts born 1927 to 1942 and 1959 to 1974 fell as well, with changes of −8.7 (95% CI −12.6 to −4.7) and −14.9 (95% CI −17.9 to −11.9) points, and the cohort born 1991 or later changed by −4.9 (95% CI −8.6 to −1.1) points between the 2010s and the 2020s. The two oldest cohorts have enough respondents only through the 1980s (born 1895 to 1910) and the 2000s (born 1911 to 1926), and their changes are −3.3 (95% CI −8.5 to 1.9) and −1.5 (95% CI −7.9 to 4.9) points; the change for the 1975 to 1990 cohort, whose first cell has 153 respondents, is −4.9 (95% CI −10.5 to 0.6) points. All three intervals include zero.

Figure 2

Every cohort born 1927 to 1974 fell, mostly after 1990

Cross-pressured share of placed partisans by birth cohort (solid) and for all placed partisans (dashed), ANES survey years pooled by decade, 1972-2024, with 95% intervals

Share of placed partisans (%)

This cohort
All placed partisans

Survey years pooled by decade (plotted at the midpoint)

Show data

Data for Every cohort born 1927 to 1974 fell, mostly after 1990
Panel	Series	Survey years pooled	Share (%)	Interval low	Interval high	Placed partisans
Born 1895-1910	This cohort	1972-1978	19.7	16.6	23.3	644
Born 1895-1910	This cohort	1980-1988	16.5	12.8	20.9	328
Born 1895-1910	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1895-1910	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1895-1910	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1895-1910	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1895-1910	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1895-1910	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1911-1926	This cohort	1972-1978	18.1	15.9	20.5	1,262
Born 1911-1926	This cohort	1980-1988	18.5	16.3	21.0	1,053
Born 1911-1926	This cohort	1990-1998	19.8	17.1	22.7	835
Born 1911-1926	This cohort	2000-2008	16.6	11.5	23.4	201
Born 1911-1926	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1911-1926	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1911-1926	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1911-1926	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1911-1926	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1911-1926	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1927-1942	This cohort	1972-1978	18.7	16.5	21.0	1,324
Born 1927-1942	This cohort	1980-1988	18.3	16.2	20.5	1,276
Born 1927-1942	This cohort	1990-1998	16.6	14.5	19.0	1,174
Born 1927-1942	This cohort	2000-2008	12.2	9.6	15.5	678
Born 1927-1942	This cohort	2012-2016	7.9	5.8	10.7	797
Born 1927-1942	This cohort	2020-2024	10.0	7.3	13.7	625
Born 1927-1942	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1927-1942	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1927-1942	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1927-1942	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1927-1942	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1927-1942	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1943-1958	This cohort	1972-1978	17.6	15.7	19.6	1,702
Born 1943-1958	This cohort	1980-1988	17.5	16.0	19.1	2,279
Born 1943-1958	This cohort	1990-1998	15.1	13.5	16.7	2,120
Born 1943-1958	This cohort	2000-2008	13.5	11.4	15.8	1,263
Born 1943-1958	This cohort	2012-2016	8.9	7.5	10.5	2,392
Born 1943-1958	This cohort	2020-2024	5.0	4.0	6.1	2,910
Born 1943-1958	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1943-1958	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1943-1958	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1943-1958	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1943-1958	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1943-1958	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1959-1974	This cohort	1980-1988	21.4	18.8	24.3	863
Born 1959-1974	This cohort	1990-1998	16.5	14.8	18.3	1,881
Born 1959-1974	This cohort	2000-2008	12.7	10.7	15.0	1,248
Born 1959-1974	This cohort	2012-2016	9.4	7.9	11.1	2,037
Born 1959-1974	This cohort	2020-2024	6.5	5.4	7.9	2,785
Born 1959-1974	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1959-1974	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1959-1974	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1959-1974	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1959-1974	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1959-1974	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1975-1990	This cohort	1990-1998	10.4	6.2	17.1	153
Born 1975-1990	This cohort	2000-2008	12.3	9.6	15.8	641
Born 1975-1990	This cohort	2012-2016	8.4	6.9	10.3	1,758
Born 1975-1990	This cohort	2020-2024	5.5	4.5	6.8	2,789
Born 1975-1990	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1975-1990	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1975-1990	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1975-1990	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1975-1990	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1975-1990	All placed partisans	2020-2024	5.7	5.1	6.3	10,835
Born 1991 or later	This cohort	2012-2016	8.9	6.0	13.0	407
Born 1991 or later	This cohort	2020-2024	4.0	2.8	5.7	1,363
Born 1991 or later	All placed partisans	1972-1978	18.4	17.3	19.6	5,084
Born 1991 or later	All placed partisans	1980-1988	18.4	17.4	19.4	5,826
Born 1991 or later	All placed partisans	1990-1998	16.2	15.2	17.1	6,284
Born 1991 or later	All placed partisans	2000-2008	12.9	11.8	14.2	4,071
Born 1991 or later	All placed partisans	2012-2016	8.8	8.0	9.7	7,562
Born 1991 or later	All placed partisans	2020-2024	5.7	5.1	6.3	10,835

The 1943-1958 cohort falls from 17.6% in the 1970s to 5.0% in the 2020s, and the 1927-1942 and 1959-1974 cohorts also fall, mostly after 1990. In each panel the solid line is the cohort and the dashed line is all placed partisans, and the bands are confidence intervals.Note: A cell is plotted only when it has 100 or more placed partisans and an effective sample (the sample size after weighting) of 100 or more, the floor of this chapter, so a line starts and ends where the cohort is well observed. The two oldest cohorts are plotted for only two and four decades.Note: The pooled 2020s combine 2020 and 2024, and 553 placed partisans with no birth cohort in the file are excluded from the cohort lines. The dashed line uses all placed partisans, including those with no birth cohort.Note: Two cells below the floor are not plotted: the 1895-1910 cohort in the 1990s (13.1%, 105 respondents) and the 1911-1926 cohort in the 2010s (12.8%, 60 respondents).Note: The 1927-1942 cohort is at 7.9% in the 2010s and 10.0% in the 2020s, a difference of +2.1 points that is within sampling error.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included

Figure 3

The changes for the two oldest cohorts and the 1975 to 1990 cohort include zero, and the other cohorts fell

Change in the cross-pressured share of placed partisans from a cohort's first to its last plotted decade, ANES 1972-2024, with 95% intervals

Show data

Data for The changes for the two oldest cohorts and the 1975 to 1990 cohort include zero, and the other cohorts fell
Cohort	First plotted decade	Last plotted decade	Change (points)	Interval low	Interval high	Placed partisans in the two cells
Born 1895-1910	1970s	1980s	−3.3	−8.5	+1.9	972
Born 1911-1926	1970s	2000s	−1.5	−7.9	+4.9	1,463
Born 1927-1942	1970s	2020s	−8.7	−12.6	−4.7	1,949
Born 1943-1958	1970s	2020s	−12.6	−14.9	−10.4	4,612
Born 1959-1974	1980s	2020s	−14.9	−17.9	−11.9	3,648
Born 1975-1990	1990s	2020s	−4.9	−10.5	+0.6	2,942
Born 1991 or later	2010s	2020s	−4.9	−8.6	−1.1	1,770

The intervals for the cohorts born 1927 to 1974 and for the cohort born 1991 or later lie below zero, and those for the two oldest cohorts and the cohort born 1975 to 1990 include it. Each dot is the change in a cohort's share from its first to its last plotted decade, and the line is a 95% interval.Note: Plotted decades by cohort: Born 1895-1910, 1970s to 1980s; Born 1911-1926, 1970s to 2000s; Born 1927-1942, 1970s to 2020s; Born 1943-1958, 1970s to 2020s; Born 1959-1974, 1980s to 2020s; Born 1975-1990, 1990s to 2020s; Born 1991 or later, 2010s to 2020s.Note: The 1975-1990 cohort's first plotted cell has 153 placed partisans, close to the floor of 100.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included

What a cohort line contains

A cohort line follows a group of people born in the same years, and the group changes as the years pass. It mixes four things: individuals who changed party or ideology; loss, because some members of a cohort die or stop answering surveys; entry, because a cohort keeps arriving in the sample as its youngest members reach adulthood; and changes in who is a partisan and who places themselves on the scale.The 1943 to 1958 cohort was not fully adult in the 1970s. People born from 1955 on reached voting age during that decade, so its 1970s cell includes newcomers to the sample.

A simple bound shows what loss alone would require. In the 1943 to 1958 cohort the odds of being cross-pressured in the 2020s are 0.24 (95% CI 0.19 to 0.32) times the odds in the 1970s. If loss alone produced that fall, the cross-pressured would have had to survive, and stay in the survey samples, at only that fraction of the rate of other partisans over the whole period. The bound covers one channel, loss, and gives no estimate of anyone's mortality or of the other three channels.

Studies that re-interview the same people find movement in both directions. A panel of the 1992 to 1996 elections found that people who saw party differences on an issue they cared about changed party to fit their views, and people who saw differences on an issue they did not care about changed their views to fit their party [4]. A panel that followed one cohort from 1973 to 1997 found that partisanship shaped issue orientations early and that issue orientations shaped partisanship later [5]. Work on the 1992 to 1994 ANES panel describes citizens choosing a party identification to match their policy preferences as the parties' positions became clearer [6]. A cohort line also mixes age and period, because a cohort ages a decade in each step and age, period and cohort effects cannot be separated without further assumptions [7]. Chapter 3.6, Age and generation takes up what repeated surveys can and cannot separate.

Figure 4 puts a number on replacement. The exercise takes the observed mix of birth cohorts among placed partisans in each decade and applies fixed cohort rates to it. The mix changed: 38.5% of placed partisans in the 1970s were born before 1927, and 73.9% of those in the 2020s were born after 1958.

The first version freezes every cohort at its 1970s rate. Cohorts without enough respondents in the 1970s, which are the cohorts born after 1958 and the oldest, take the mean 1970s cohort rate of 18.5%. This version predicts a 2020s share of 18.3% (95% CI 17.1% to 19.5%), against 5.6% (95% CI 5.0% to 6.3%) observed, and the change in the mix alone moves the share by +0.1 (95% CI −0.2 to 0.3) points, which is −0.4% (95% CI −2.2% to 1.4%) of the fall. That result is close to built in, because the assumption gives later cohorts the rate of the older ones.

The second version freezes every cohort at its entry rate, the rate it had in the first decade in which it has enough respondents. The older cohorts keep their 1970s rates, and the later ones take 21.4% for people born 1959 to 1974, 10.4% for 1975 to 1990 and 8.9% for 1991 or later. This version predicts 15.1% (95% CI 13.5% to 16.7%) for the 2020s. The mix alone then lowers the share by 3.2 (95% CI 1.3 to 5.0) points, which is 25.2% (95% CI 10.8% to 38.0%) of the fall in the observed share from the 1970s to the 2020s, 12.8 (95% CI 11.5 to 14.2) points. This version credits replacement with everything that distinguishes later cohorts when they arrive, including the part that comes from the years in which they arrived. It is the more generous of the two versions and is not an upper bound.

Figure 4

Freezing cohort rates predicts a 2020s cross-pressured share of 15.1% to 18.3%, against 5.6% observed

Cross-pressured share of placed partisans, observed and under two frozen-rate counterfactuals, ANES survey years pooled by decade, 1972-2024, with 95% bootstrap intervals

Show data

Data for Freezing cohort rates predicts a 2020s cross-pressured share of 15.1% to 18.3%, against 5.6% observed
Series	Survey years pooled	Share (%)	Interval low	Interval high	Placed partisans
Observed	1972-1978	18.5	17.3	19.6	5,059
Observed	1980-1988	18.4	17.4	19.3	5,808
Observed	1990-1998	16.2	15.2	17.1	6,268
Observed	2000-2008	12.9	11.8	14.2	4,034
Observed	2012-2016	8.8	8.0	9.7	7,451
Observed	2020-2024	5.6	5.0	6.3	10,489
1970s rates	1972-1978	18.3	17.1	19.5	5,059
1970s rates	1980-1988	18.2	16.9	19.4	5,808
1970s rates	1990-1998	18.2	17.0	19.4	6,268
1970s rates	2000-2008	18.3	17.0	19.5	4,034
1970s rates	2012-2016	18.3	17.0	19.4	7,451
1970s rates	2020-2024	18.3	17.1	19.5	10,489
Entry rates	1972-1978	18.3	17.1	19.5	5,059
Entry rates	1980-1988	18.6	17.5	19.7	5,808
Entry rates	1990-1998	18.8	17.6	20.0	6,268
Entry rates	2000-2008	17.5	16.1	19.1	4,034
Entry rates	2012-2016	16.2	14.6	17.9	7,451
Entry rates	2020-2024	15.1	13.5	16.7	10,489

The observed line falls from 18.5% to 5.6%, the line with cohorts frozen at their entry rates (a cohort's rate in its first well-observed decade) ends at 15.1%, and the line frozen at 1970s rates stays near its start. The mix of cohorts accounts for 25.2% of the fall under the entry-rate version and −0.4% of it under the 1970s-rate version, which is within sampling error of zero. Both versions apply the observed cohort mix of each decade to fixed cohort rates and are accounting exercises, not fitted models.Note: Placed partisans with no birth cohort in the file are left out of all three series so that the accounting is exact.Note: Cohorts not observed with 100 or more placed partisans in the 1970s take the mean 1970s cohort rate of 18.5% in the first counterfactual.Note: The within-cohort part of the fall is the observed fall minus the mix part, so it includes a baseline residual: the frozen rates applied to the 1970s mix predict a 1970s share 0.2 points (1970s rates) and 0.2 points (entry rates) below the observed one.Note: The bootstrap re-draws respondents with replacement within each pooled decade many times and re-estimates every quantity each time; the intervals are the 2.5th and 97.5th percentiles.Note: Bands are drawn for the observed and entry-rate lines; the interval of the 1970s-rate line is in the data table. Differences are computed from unrounded values, so a printed difference can differ in the last digit from the difference of the printed shares.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included; intervals from a bootstrap over respondents within pooled decades (1,000 replicates, seed 12345)

Figure 5 gives the mix effect under those two versions and three variations of the second: a stricter and a looser rule for a well-observed cell, and a limiting case that holds the cohorts born after 1958 at their own 2020s rates instead of their entry rates. The limiting case counts the fall inside those cohorts as a fixed cohort trait, so it is not a replacement counterfactual, and it shows that the answer depends on the rate given to a cohort before it is observed.

Figure 5

The share of the fall due to the mix of cohorts ranges from −0.4% to 74.0% depending on the rates given to later cohorts

Change in the cross-pressured share of placed partisans from the 1970s to the 2020s that the mix of birth cohorts produces under five assumptions about cohort rates, with 95% bootstrap intervals

Assumption about cohort rates	Change from the mix (points)	95% interval low (points)	95% interval high (points)	Share of the fall (%)	95% interval low (%)	95% interval high (%)
Every cohort at its 1970s rate	+0.1	−0.2	+0.3	−0.4	−2.2	1.4
Every cohort at its entry rate	−3.2	−5.0	−1.3	25.2	10.8	38.0
Entry rates, stricter rule for a well-observed cell	−2.8	−4.4	−1.1	21.4	9.0	32.6
Entry rates, looser rule for a well-observed cell	−3.3	−5.1	−1.4	26.0	11.6	38.9
Cohorts born after 1958 at their own 2020s rates (limiting case)	−9.5	−10.7	−8.4	74.0	70.6	77.8

Show data

Data for The share of the fall due to the mix of cohorts ranges from −0.4% to 74.0% depending on the rates given to later cohorts
change	change_high	change_low	share	share_high	share_low	version
0.053	0.290	−0.182	−0.410	1.44	−2.24	Every cohort at its 1970s rate
−3.24	−1.33	−4.99	25.23	37.96	10.78	Every cohort at its entry rate
−2.75	−1.08	−4.40	21.44	32.64	9.00	Entry rates, stricter rule for a well-observed cell
−3.34	−1.40	−5.14	26.05	38.87	11.61	Entry rates, looser rule for a well-observed cell
−9.50	−8.43	−10.65	74.02	77.76	70.57	Cohorts born after 1958 at their own 2020s rates (limiting case)

The share is −0.4% with every cohort at its 1970s rate, 25.2% at entry rates (21.4% with a stricter rule for a well-observed cell and 26.0% with a looser one) and 74.0% when the cohorts born after 1958 keep their own 2020s rates, so the data do not choose among the assumptions. The last row is a limiting case: an extreme assumption that counts the fall inside those cohorts as a fixed cohort trait.Note: The change from the mix is the frozen-rate share in the 2020s minus the frozen-rate share in the 1970s, in percentage points; the share of the fall is that change divided by the observed change. Differences are computed from unrounded values, so a printed difference can differ in the last digit from the difference of the printed shares.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included; intervals from a bootstrap over respondents within pooled decades

The parts sum to the fall. Under the second version the share falls by 9.6 (95% CI 7.9 to 11.4) points inside cohorts and by 3.2 points through the mix, together 12.8 points. Under the first version the inside-cohort part is 12.9 (95% CI 11.5 to 14.3) points and the mix moves the share by +0.1 points. The inside-cohort part is the observed fall minus the mix part, so it also contains a small residual that the notes to Figure 4 give. Both versions are accounting statements. They hold cohort rates fixed by assumption and do not say why the rates fell.

The education gap widened, then returned to its 1970s size

The change in the education mix accounts for 1.3 (95% CI 0.9 to 1.8) points, or 10.3% (95% CI 6.6% to 15.1%), of the fall of 12.7 (95% CI 11.4 to 13.9) points in this sample. College graduates went from 20.6% (95% CI 19.4% to 21.8%) of placed partisans in the 1970s to 41.3% (95% CI 40.0% to 42.6%) in the 2020s, and applying the 1970s education mix to the 2020s group rates gives a cross-pressured share of 7.0% (95% CI 6.1% to 8.1%) in the 2020s, against 5.7% (95% CI 5.1% to 6.4%) observed. The remainder of the fall occurs inside education groups.

Figure 6 plots the cross-pressured share for each education group. The gap between partisans with a high school education or less and college graduates was 5.4 (95% CI 2.6 to 8.3) points in the 1970s, 11.3 (95% CI 9.2 to 13.5) in the 1990s, 11.0 (95% CI 8.1 to 14.0) in the 2000s and 5.1 (95% CI 3.4 to 6.9) in the 2020s. The 2020s interval excludes zero and overlaps the 1970s interval, so the gap returned to about its starting size and did not close.

Figure 6

The education gap in the cross-pressured share was 5.4 points in the 1970s, 11.3 in the 1990s and 5.1 in the 2020s

Cross-pressured share of placed partisans by highest education and for all placed partisans (dashed), ANES survey years pooled by decade, 1972-2024, with 95% intervals

Show data

Data for The education gap in the cross-pressured share was 5.4 points in the 1970s, 11.3 in the 1990s and 5.1 in the 2020s
Series	Survey years pooled	Share (%)	Interval low	Interval high	Placed partisans
College degree	1972-1978	15.1	12.9	17.6	1,039
College degree	1980-1988	12.6	11.0	14.4	1,448
College degree	1990-1998	9.8	8.5	11.2	1,960
College degree	2000-2008	6.9	5.5	8.7	1,438
College degree	2012-2016	5.2	4.3	6.3	2,953
College degree	2020-2024	3.6	3.0	4.3	5,419
Some college	1972-1978	15.9	13.7	18.3	1,124
Some college	1980-1988	17.8	16.0	19.8	1,616
Some college	1990-1998	14.8	13.1	16.7	1,706
Some college	2000-2008	13.6	11.6	15.9	1,320
Some college	2012-2016	8.0	6.8	9.5	2,518
Some college	2020-2024	5.6	4.6	6.7	3,431
High school or less	1972-1978	20.5	19.0	22.2	2,906
High school or less	1980-1988	21.7	20.2	23.3	2,712
High school or less	1990-1998	21.1	19.5	22.8	2,513
High school or less	2000-2008	18.0	15.6	20.6	1,294
High school or less	2012-2016	13.3	11.5	15.3	2,011
High school or less	2020-2024	8.7	7.2	10.5	1,874
All placed partisans	1972-1978	18.4	17.3	19.6	5,084
All placed partisans	1980-1988	18.4	17.4	19.4	5,826
All placed partisans	1990-1998	16.2	15.2	17.1	6,284
All placed partisans	2000-2008	12.9	11.8	14.2	4,071
All placed partisans	2012-2016	8.8	8.0	9.7	7,562
All placed partisans	2020-2024	5.7	5.1	6.3	10,835

The gap is the vertical distance between the high school or less line and the college degree line: it widens from the 1970s to the 1990s and returns to about its 1970s size by the 2020s. College graduates fell first, from 15.1% in the 1970s to 6.9% in the 2000s, when partisans with a high school education or less were at 18.0%, and by the 2020s the two groups were at 3.6% and 8.7%. The bands are confidence intervals.Note: Grade school and high school are combined because the grade-school cells fall below the floor of 100 respondents in the 2000s and 2010s.Note: The pooled 2020s combine 2020 and 2024. The gap is the high school or less share minus the college share. Differences are computed from unrounded values, so a printed difference can differ in the last digit from the difference of the printed shares.Note: Bands are drawn for the three education groups; the interval of the dashed line is in the data table.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included

College graduates fell first. Their share lies below its 1970s interval from the 1990s on, and the shares of partisans with some college and with a high school education or less lie below theirs from the 2010s on. The timing is descriptive, since the data hold no measure of exposure to elite messages.

Figure 7 compares the South with the rest of the country. In the 1970s 48.6% (95% CI 44.5% to 52.8%) of placed Southern conservatives were Democrats, against 30.6% (95% CI 27.9% to 33.3%) outside the South, a gap of 18.1 (95% CI 13.1 to 23.0) points. In the 2020s the shares were 9.3% (95% CI 7.7% to 11.2%) and 8.3% (95% CI 7.0% to 9.9%), and the gap of 1.0 (95% CI −1.3 to 3.3) points is within sampling error of zero. The cross-pressured share of all placed partisans fell in both regions, from 22.9% (95% CI 20.7% to 25.2%) to 6.9% (95% CI 5.9% to 8.1%) in the South and from 16.4% (95% CI 15.1% to 17.8%) to 4.9% (95% CI 4.2% to 5.6%) elsewhere, and a gap of 2.0 (95% CI 0.7 to 3.3) points remains, down from 6.5 (95% CI 3.9 to 9.1) points in the 1970s.

Figure 7

Southern and other partisans converged as the Democratic share of conservatives fell from 48.6% to 9.3% in the South and 30.6% to 8.3% elsewhere

Two shares of placed partisans by region (Census South and the rest of the country), ANES survey years pooled by decade, 1972-2024, with 95% intervals

South
Non-South

Show data

Data for Southern and other partisans converged as the Democratic share of conservatives fell from 48.6% to 9.3% in the South and 30.6% to 8.3% elsewhere
Measure	Region	Survey years pooled	Share (%)	Interval low	Interval high	Placed partisans
Cross-pressured, % of placed partisans	South	1972-1978	22.9	20.7	25.2	1,591
Cross-pressured, % of placed partisans	South	1980-1988	22.6	20.7	24.6	1,813
Cross-pressured, % of placed partisans	South	1990-1998	18.6	16.8	20.4	2,024
Cross-pressured, % of placed partisans	South	2000-2008	14.5	12.5	16.7	1,548
Cross-pressured, % of placed partisans	South	2012-2016	10.3	8.9	11.8	2,847
Cross-pressured, % of placed partisans	South	2020-2024	6.9	5.9	8.1	3,993
Cross-pressured, % of placed partisans	Non-South	1972-1978	16.4	15.1	17.8	3,493
Cross-pressured, % of placed partisans	Non-South	1980-1988	16.4	15.3	17.6	4,013
Cross-pressured, % of placed partisans	Non-South	1990-1998	15.0	13.9	16.2	4,260
Cross-pressured, % of placed partisans	Non-South	2000-2008	12.0	10.6	13.6	2,521
Cross-pressured, % of placed partisans	Non-South	2012-2016	8.0	7.0	9.0	4,715
Cross-pressured, % of placed partisans	Non-South	2020-2024	4.9	4.2	5.6	6,842
Democrats, % of placed conservatives	South	1972-1978	48.6	44.5	52.8	653
Democrats, % of placed conservatives	South	1980-1988	40.9	37.6	44.4	799
Democrats, % of placed conservatives	South	1990-1998	32.4	29.4	35.7	943
Democrats, % of placed conservatives	South	2000-2008	22.7	19.3	26.5	735
Democrats, % of placed conservatives	South	2012-2016	17.2	14.8	20.0	1,291
Democrats, % of placed conservatives	South	2020-2024	9.3	7.7	11.2	1,910
Democrats, % of placed conservatives	Non-South	1972-1978	30.6	27.9	33.3	1,299
Democrats, % of placed conservatives	Non-South	1980-1988	27.5	25.4	29.7	1,705
Democrats, % of placed conservatives	Non-South	1990-1998	26.8	24.7	29.1	1,744
Democrats, % of placed conservatives	Non-South	2000-2008	21.5	18.7	24.7	999
Democrats, % of placed conservatives	Non-South	2012-2016	14.8	12.9	17.0	1,825
Democrats, % of placed conservatives	Non-South	2020-2024	8.3	7.0	9.9	2,583

In the right panel the South started 18.1 points above the rest of the country in the Democratic share of placed conservatives and ended within sampling error of it. In the left panel the cross-pressured share fell in both regions, and a smaller gap of 2.0 points remains in the 2020s. Solid lines are the South and dashed lines the rest of the country, and the bands are confidence intervals.Note: The cumulative file has no region for the 2,049 placed partisans among the 2024 panel and paper cases; their region is the Census region of the sample location in the ANES 2024 Time Series, which reproduces the cumulative-file flag for every 2024 respondent who has both.Note: Conservatives are placed partisans at points 5 to 7 of the seven-point scale. Differences are computed from unrounded values, so a printed difference can differ in the last digit from the difference of the printed shares.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included; region is the Census South as coded by the ANES, and for the 2024 panel and paper cases the Census region of the sample location in the ANES 2024 Time Series

The rise of Southern Republicans has been traced through offices and elections [8], and the data here describe how respondents label themselves. The cumulative file has no region for the 2,049 placed partisans among the 2024 panel and paper cases, the same cases that have no mode weight. Their region comes from the sample location recorded in the 2024 Time Series file, which reproduces the cumulative-file flag for every 2024 respondent who has both. Chapter 1.7, The South (planned) compares the region with the rest of the country on other gaps.

The fall is smaller with unplaced partisans in the base

The base of every share so far is the placed partisans, and that base changed. The share of partisans who placed themselves was 59.7% (95% CI 57.7% to 61.7%) in 1972, between 65.2% and 81.2% in the study years from 1974 to 2008 other than 2000, between 77.1% and 92.5% from 2012 to 2024, and 85.8% (95% CI 84.3% to 87.1%) in 2024. The 2000 value, 35.6% (95% CI 33.1% to 38.3%), reflects the design of that study. 52.1% of the 2000 respondents have no recorded answer to the item, against at most 0.9% in 1996, 1998, 2002 and 2004, and among the respondents who have an answer the share choosing "don't know" is 22.3%, inside the 16.8% to 24.0% range of those years. That pattern fits an item put to a subsample of respondents. The 2000 point therefore rests on 603 placed partisans, and its low placement rate is not evidence that more partisans declined the scale. The design is inferred from the response pattern and has not been checked against the codebook.

Partisans who decline the scale are outside every share above, and many Americans do not describe their views with ideological labels [9]. Three checks ask whether the fall depends on who is in the base. The first reweights placed partisans, within each year, to the party, education, age group and race mix of all partisans in that year. The second divides cross-pressured partisans by everyone with a party identification and a placement, so pure independents stay in the denominator. The third divides by all partisans and counts those who did not place themselves as not cross-pressured; Figure 8 leaves that series blank for 2000, when the item went to a subsample. The reweighted share goes from 19.9% (95% CI 17.8% to 22.2%) to 6.9% (95% CI 5.8% to 8.1%), a fall of 13.0 (95% CI 10.5 to 15.5) points, and the share with independents included from 17.4% (95% CI 15.6% to 19.4%) to 6.0% (95% CI 5.1% to 7.1%), a fall of 11.4 (95% CI 9.2 to 13.5) points, against 19.6% and 6.3% for the placed partisans. The share of all partisans goes from 11.7% (95% CI 10.5% to 13.1%) to 5.4% (95% CI 4.6% to 6.4%), a fall of 6.3 (95% CI 4.7 to 7.9) points against 13.3 points for the placed partisans, because the share of partisans who placed themselves rose from 59.7% to 85.8%.

Figure 8

The fall is 11.4 to 13.3 points under three bases and 6.3 points when unplaced partisans count as not cross-pressured

Cross-pressured share under four bases, ANES study years 1972-2024, with 95% intervals for the Placed line

Show data

Data for The fall is 11.4 to 13.3 points under three bases and 6.3 points when unplaced partisans count as not cross-pressured
Series	Year	Share (%)	Interval low	Interval high	Respondents in the base
Placed	1972	19.6	17.6	21.8	1,371
Placed	1974	18.4	15.7	21.3	977
Placed	1976	17.8	15.8	20.0	1,306
Placed	1978	18.0	16.1	20.0	1,430
Placed	1980	20.3	17.8	23.1	892
Placed	1982	18.1	15.6	20.9	806
Placed	1984	17.0	15.1	19.0	1,401
Placed	1986	18.5	16.5	20.5	1,441
Placed	1988	18.6	16.6	20.8	1,286
Placed	1990	19.7	17.5	22.0	1,220
Placed	1992	16.9	15.2	18.8	1,630
Placed	1994	13.6	11.7	15.8	1,279
Placed	1996	14.5	12.5	16.9	1,233
Placed	1998	15.6	13.2	18.3	922
Placed	2000	16.0	13.0	19.5	603
Placed	2002	14.9	12.6	17.5	1,171
Placed	2004	11.5	9.4	14.0	849
Placed	2008	11.3	9.4	13.4	1,448
Placed	2012	9.8	8.7	11.0	4,613
Placed	2016	7.2	6.2	8.4	2,949
Placed	2020	5.2	4.5	5.9	6,353
Placed	2024	6.3	5.4	7.4	4,482
Reweighted	1972	19.9	17.8	22.2	1,365
Reweighted	1974	19.5	16.7	22.6	958
Reweighted	1976	19.0	16.8	21.4	1,290
Reweighted	1978	18.7	16.7	20.9	1,420
Reweighted	1980	21.4	18.7	24.4	888
Reweighted	1982	20.6	17.7	23.8	803
Reweighted	1984	18.4	16.4	20.6	1,381
Reweighted	1986	18.8	16.8	21.0	1,426
Reweighted	1988	19.9	17.7	22.3	1,254
Reweighted	1990	20.7	18.4	23.2	1,197
Reweighted	1992	18.2	16.3	20.2	1,582
Reweighted	1994	15.3	13.1	17.8	1,217
Reweighted	1996	15.9	13.6	18.4	1,224
Reweighted	1998	17.4	14.7	20.4	901
Reweighted	2000	18.8	15.3	23.0	596
Reweighted	2002	16.0	13.4	18.9	1,152
Reweighted	2004	13.1	10.7	15.9	843
Reweighted	2008	13.0	11.0	15.3	1,402
Reweighted	2012	10.1	9.0	11.3	4,507
Reweighted	2016	8.4	7.2	9.8	2,846
Reweighted	2020	5.7	4.9	6.6	6,036
Reweighted	2024	6.9	5.8	8.1	4,207
With independents	1972	17.4	15.6	19.4	1,545
With independents	1974	15.5	13.2	18.1	1,140
With independents	1976	15.4	13.7	17.4	1,497
With independents	1978	15.5	13.8	17.3	1,658
With independents	1980	18.0	15.8	20.5	1,004
With independents	1982	16.4	14.1	19.0	891
With independents	1984	15.4	13.7	17.3	1,546
With independents	1986	16.4	14.7	18.3	1,622
With independents	1988	16.8	15.0	18.9	1,419
With independents	1990	18.3	16.3	20.4	1,315
With independents	1992	15.2	13.6	16.9	1,819
With independents	1994	12.4	10.6	14.5	1,398
With independents	1996	13.5	11.6	15.7	1,326
With independents	1998	14.2	12.0	16.7	1,012
With independents	2000	14.4	11.7	17.6	669
With independents	2002	14.2	12.0	16.8	1,223
With independents	2004	10.8	8.8	13.1	913
With independents	2008	10.2	8.5	12.2	1,607
With independents	2012	8.5	7.6	9.6	5,288
With independents	2016	6.4	5.5	7.5	3,298
With independents	2020	4.7	4.1	5.4	7,041
With independents	2024	6.0	5.1	7.1	4,716
All partisans	1972	11.7	10.5	13.1	2,295
All partisans	1974	13.6	11.6	15.9	1,305
All partisans	1976	12.3	10.8	13.9	1,879
All partisans	1978	13.5	12.1	15.2	1,898
All partisans	1980	13.2	11.5	15.1	1,369
All partisans	1982	12.0	10.3	13.9	1,220
All partisans	1984	12.2	10.8	13.7	1,948
All partisans	1986	14.3	12.8	15.9	1,863
All partisans	1988	13.4	11.9	15.1	1,783
All partisans	1990	13.9	12.3	15.6	1,729
All partisans	1992	12.8	11.4	14.2	2,155
All partisans	1994	10.7	9.1	12.5	1,589
All partisans	1996	11.2	9.6	13.1	1,549
All partisans	1998	12.5	10.5	14.7	1,124
All partisans	2000	n/a	n/a	n/a	1,568
All partisans	2002	12.1	10.1	14.4	1,372
All partisans	2004	9.0	7.3	11.1	1,074
All partisans	2008	8.7	7.3	10.3	2,018
All partisans	2012	9.1	8.1	10.2	5,098
All partisans	2016	5.5	4.7	6.5	3,668
All partisans	2020	4.4	3.8	5.1	7,277
All partisans	2024	5.4	4.6	6.4	5,103

The Placed, Reweighted and With independents lines fall together, and the All partisans line falls by less because the share of partisans who place themselves rose from 59.7% in 1972 to 85.8% in 2024. The All partisans line has a gap at 2000, when the item went to a subsample of respondents.Note: The share of partisans who place themselves was 59.7% in 1972, 35.6% in 2000, 84.9% in 2020 and 85.8% in 2024.Note: The All partisans line has no value for 2000: 52.1% of the 2000 respondents have no recorded answer to the item, so the share of partisans who place themselves reflects the design of that study and not the partisans.Note: With no assumption about the unplaced, the 1972 share lies between 11.7% and 52.0% and the 2024 share between 5.4% and 19.6%.Note: Intervals for the reweighted series use the effective sample size of the adjusted weights and ignore the estimation of the adjustment factors.Note: Bands are drawn for the Placed line only; the intervals of the other lines are in the data table.Source: ANES Cumulative Data File, 1972-2024; weighted (VCF0009z); placed partisans, leaners included; the reweighted series adjusts the weights within each year

The reweighting corrects for the traits it uses. The 13.3-point fall for placed partisans carries over to all partisans only if the partisans who place themselves resemble those who do not on those traits, an assumption the ANES cannot test. With no assumption about the unplaced, the 1972 share lies between 11.7% and 52.0% and the 2024 share between 5.4% and 19.6%. The two ranges overlap, so the data alone fix neither the direction nor the size of the change.

A second survey follows a different path, 2008 to 2016

The Cooperative Election Study (CES) asks party identification and a five-point ideology item every year from 2006. It is an opt-in online sample, and its intervals, like the ANES intervals, describe sampling variation and leave out selection into the sample. This chapter uses the even-year waves, which are the main national waves, and leaves out the smaller odd-year waves. The series starts after most of the ANES fall, so it cannot show whether the fall happened. Within a survey the coding is held fixed, so a path needs no mapping between the scales. From 2008 to 2016 the ANES share fell by 4.1 (95% CI 1.8 to 6.4) points and the CES share rose by 3.0 (95% CI 2.5 to 3.4) points (Figure 9). From 2008 to 2024 both fell, the ANES by 4.9 (95% CI 2.7 to 7.2) points and the CES by 1.8 (95% CI 1.4 to 2.2). The chapter takes from the CES only that both surveys read lower in 2024 than in 2008 and that the CES does not follow the ANES path from 2008 to 2016. It does not use the CES to corroborate the size of the ANES fall.

Figure 9

From 2008 to 2016 the ANES cross-pressured share fell while the CES share rose

Cross-pressured share of placed partisans in the ANES (seven-point ideology scale) and the CES (five-point scale, even-year waves), 2004-2024, with 95% intervals

Show data

Data for From 2008 to 2016 the ANES cross-pressured share fell while the CES share rose
Series	Year	Share (%)	Interval low	Interval high	Placed partisans
ANES	2004	11.5	9.4	14.0	849
ANES	2008	11.3	9.4	13.4	1,448
ANES	2012	9.8	8.7	11.0	4,613
ANES	2016	7.2	6.2	8.4	2,949
ANES	2020	5.2	4.5	5.9	6,353
ANES	2024	6.3	5.4	7.4	4,482
CES	2006	4.3	4.0	4.5	30,222
CES	2008	5.5	5.2	5.8	27,354
CES	2010	5.9	5.6	6.3	46,430
CES	2012	7.0	6.7	7.4	43,847
CES	2014	7.5	7.1	7.9	42,901
CES	2016	8.4	8.1	8.8	50,082
CES	2018	5.9	5.6	6.3	47,245
CES	2020	5.3	5.1	5.6	48,071
CES	2022	4.5	4.2	4.7	47,741
CES	2024	3.7	3.4	4.0	49,320

From 2008 to 2016 the ANES share fell by 4.1 points while the CES share rose by 3.0 points. Both are lower in 2024 than in 2008, the ANES by 4.9 points and the CES by 1.8. The CES begins in 2006, after most of the ANES fall, so it cannot show whether the fall happened, and the scales differ, so the figure compares paths and not levels.Note: The CES is an opt-in online sample. Its intervals are narrow because it samples tens of thousands of respondents each wave; like the ANES intervals they are model based and do not reflect selection into the sample.Note: Only the even-year CES waves are plotted. The CES also fields smaller odd-year waves, which are left out. The ANES line before 2004 is in the first figure of the chapter.Note: CES partisans are pid7 points 1 to 3 and 5 to 7; conservatives are the two conservative points of the five-point scale and 'not sure' counts as not placed.Source: ANES Cumulative Data File, 1972-2024, and Cooperative Election Study cumulative file, 2006-2024; weighted (VCF0009z in the ANES, the survey weight in each CES wave); placed partisans, leaners included

Levels are a different matter, because the two surveys use different scales and the level of the cross-pressured share depends on how the seven-point scale is mapped to the five-point one. Figure 10 gives the CES-to-ANES ratio of the share under two mappings. When every point on each side of the midpoint counts as liberal or conservative, the ratio is 0.49 (95% CI 0.40 to 0.59) in 2008 and 0.58 (95% CI 0.48 to 0.70) in 2024, so the CES share is the lower one in both years. When only the two outer points on each side count, the ratio is 1.22 (95% CI 0.90 to 1.64) and 1.41 (95% CI 1.07 to 1.85), and its interval lies above one in every shared year except 2008. The sign of the level difference follows the mapping, and no mapping is correct by construction, so the chapter draws no conclusion about which survey reads higher.

Figure 10

The CES share is above or below the ANES share depending on how the seven-point scale is mapped to the five-point one

Ratio of the CES cross-pressured share to the ANES share in the five shared study years under two mappings of the ANES scale, with 95% intervals; a ratio of one means equal shares

Year	ANES points counted	ANES share (%)	CES share (%)	CES over ANES	95% interval low	95% interval high	ANES respondents	CES respondents
2008	Three points each side	11.3	5.5	0.49	0.40	0.59	1,448	27,354
2008	Two outer points each side	4.5	5.5	1.22	0.90	1.64	1,448	27,354
2012	Three points each side	9.8	7.0	0.72	0.63	0.81	4,613	43,847
2012	Two outer points each side	3.9	7.0	1.80	1.47	2.19	4,613	43,847
2016	Three points each side	7.2	8.4	1.17	1.00	1.38	2,949	50,082
2016	Two outer points each side	2.6	8.4	3.23	2.48	4.22	2,949	50,082
2020	Three points each side	5.2	5.3	1.03	0.89	1.20	6,353	48,071
2020	Two outer points each side	1.9	5.3	2.83	2.22	3.61	6,353	48,071
2024	Three points each side	6.3	3.7	0.58	0.48	0.70	4,482	49,320
2024	Two outer points each side	2.6	3.7	1.41	1.07	1.85	4,482	49,320

Show data

Data for The CES share is above or below the ANES share depending on how the seven-point scale is mapped to the five-point one
anes	ces	hi	lo	mapping	n_anes	n_ces	ratio	year
11.27	5.48	0.587	0.403	Three points each side	1,448	27,354	0.486	2008
4.50	5.48	1.64	0.904	Two outer points each side	1,448	27,354	1.22	2008
9.81	7.02	0.813	0.629	Three points each side	4,613	43,847	0.716	2012
3.91	7.02	2.19	1.47	Two outer points each side	4,613	43,847	1.80	2012
7.19	8.43	1.38	1.00	Three points each side	2,949	50,082	1.17	2016
2.61	8.43	4.22	2.48	Two outer points each side	2,949	50,082	3.23	2016
5.17	5.34	1.20	0.889	Three points each side	6,353	48,071	1.03	2020
1.89	5.34	3.61	2.22	Two outer points each side	6,353	48,071	2.83	2020
6.33	3.67	0.695	0.484	Three points each side	4,482	49,320	0.580	2024
2.61	3.67	1.85	1.07	Two outer points each side	4,482	49,320	1.41	2024

When every point on each side of the midpoint counts as liberal or conservative, the ratio is 0.49 in 2008 and 0.58 in 2024. When only the two outer points on each side count, it is 1.22 and 1.41. The CES share is the same in both mappings; only the ANES share changes.Note: The CES asks a five-point item, so no mapping of the seven-point ANES item to it is correct by construction. The two shown are two natural readings of liberal and conservative on the seven-point scale.Source: ANES Cumulative Data File, 1972-2024, and Cooperative Election Study cumulative file, 2006-2024; weighted (VCF0009z in the ANES, the survey weight in each CES wave); placed partisans, leaners included; intervals by the delta method

What these data cannot show

Repeated cross-sections cannot show that any person switched, and no panel is in this chapter's data.

The chapter cannot say which identity moved. A conservative Democrat who leaves the cross-pressured group may have changed party to match ideology or changed ideology to match party, and the share counts both the same way. Age, period and cohort are not separated. The accounting exercise rests on assumptions about the rates of cohorts that were not observed in the 1970s. The versions above put the share of the fall due to the mix of cohorts at −0.4% in the first version, at 25.2% in the second and at 74.0% in the limiting case, and the data do not choose among them.

The measure covers only people who place themselves on the scale. The reweighting handles differences in observed traits, and the ANES cannot show what people who decline the scale would have answered; with them in the base the fall is smaller. The scale may mean different things across decades and education groups, and the CES comparison shows that a comparison of levels across scales depends on the mapping. The ANES added an online sample in 2012, has no weights by interview mode for 2020, and gives a 2024 estimate that differs by mode within sampling error.

The intervals are model based and ignore clustering and stratification, which can raise or lower the variance, and the chapter runs no design-based check. As an illustration, if clustering made the sampling variance larger by a design effect (the ratio of the true sampling variance to the model-based variance) of 2, every interval would extend 41% further from its estimate on each side. Among the cohort changes, the cohort-mix effect, the regional and education gaps and the fall from 1972 to 2024, one verdict would change: the interval for the change in the cohort born 1991 or later would become −4.9 (95% CI −10.1 to 0.4) and include zero.

The data hold no measure of what parties and elites did. One account in the literature holds that party leaders clarified what the labels meant [3], and this chapter tests none of it. Chapter 1.3, Sort dates by attitude dates the sort issue by issue.

Data sources

ANES Cumulative Data FileCumulative Data File, study years 1972-2024 that carry the seven-point ideology itemYears: 1948-2024 (33 study years; even years only after 2004; no 2006, 2010 or 2014 study) · Weights: weight (VCF0009z) · License: Free download after accepting the ANES terms of use; cite the release.
ANES 2024 Time Series2024 Time Series, Census region of the sample location (V243007) for the 2024 panel and paper casesYears: 2024 (pre-election and post-election interviews) · Weights: none (a respondent attribute) · License: Free download after accepting the ANES terms of use; cite the release.
Cooperative Election StudyCumulative file, even-year waves 2006-2024, party identification and five-point ideology, for the second-survey comparisonYears: 2006-2024 (every year; self-reported turnout is available in the 10 even years) · Weights: weight of each wave (the cumulative weight gives the same within-year shares) · License: See the provider's terms of use.

All data sources

Methods

Sample: 50,884 Democrats and Republicans (leaners counted with the party they lean toward: pid_7pt 1 to 3 and 5 to 7) interviewed in the 22 ANES study years from 1972 to 2024 that carry the seven-point liberal-conservative item; 39,662 of them place themselves on the scale and are the placed partisans on whom every share is based. Coding fixed in advance from the item wording (1 = liberal, 7 = conservative): liberal 1 to 3, moderate 4, conservative 5 to 7. Cross-pressured means a conservative Democrat or a liberal Republican; fully sorted means a liberal Democrat or a conservative Republican. Decades pool study years (1972-1978, 1980-1988, 1990-1998, 2000-2008, 2012-2016, 2020-2024) and are plotted at the midpoint of the years pooled. Birth cohort is the ANES cohort variable; it is missing for 553 placed partisans, who are left out of the cohort lines and of the accounting exercise. Education has three levels (high school or less, some college, college degree). Region is the Census South as coded by the ANES. The cumulative file has no region for the 2,049 placed partisans among the 2024 panel and paper cases (the same cases that have no mode weight); their region is the Census region of the sample location in the ANES 2024 Time Series file (V243007, matched on the respondent identifier), which reproduces the cumulative-file South flag for every 2024 respondent who has both. Estimand: the weighted share within each cell. Weights: the ANES full-sample weight (VCF0009z); mode-specific weights (weight_ftf, weight_web) are used only for the 2024 face-to-face and web comparison, and the 2024 panel and paper cases, which have no mode weight, take the full-sample weight. Intervals: 95% Wilson intervals on the Kish effective sample size, which are model based and ignore clustering and stratification, which can raise or lower the variance, and no design-based interval was computed; differences between two independent estimates combine their standard errors in quadrature. A design effect of 2 is used only to show which verdicts are close to zero: every half-width is multiplied by the square root of 2. Suppression: a yearly cell needs at least 150 placed partisans and an effective sample of 100 or more, and a subgroup-by-decade cell needs at least 100 and an effective sample of 100 or more. Two cohort-by-decade cells that fail the floor are reported in a note to the cohort figure with their sample sizes and are not plotted. Placement in 2000: 52% of the 2000 respondents have no recorded answer to the seven-point item, against at most 1% in 1996, 1998, 2002 and 2004, and among those with an answer the share who chose don't know is in the range of those years, which fits an item put to a subsample of respondents; the 2000 placement rate is not read as partisans declining the scale. The design was inferred from the response pattern and not checked against the codebook, and the all-partisans series is left blank for 2000. Accounting exercise: the observed mix of cohorts in each decade is combined with fixed cohort rates (version A, the 1970s rate; version B, the rate in the cohort's first well-observed decade; a stricter rule (200 placed partisans and an effective sample of 150) and a looser rule (50 and 50) for a well-observed cell; and a limiting case that holds the cohorts born after 1958 at their 2020s rates). The within-cohort part is the observed fall minus the mix part. Intervals are percentile intervals from 1,000 bootstrap replicates that resample respondents within decades (seed 12345). Education mix: the 2020s rates of the three education groups are combined with the 1970s education mix, with bootstrap intervals of the same kind. Bases: within each year the placed partisans are weighted to the party, education, age-group and race mix of all partisans; pure independents are put back in the denominator; and all partisans are the base with the unplaced counted as not cross-pressured (lower bound) or as cross-pressured (upper bound). Second survey: the Cooperative Election Study cumulative file, even-year waves 2006 to 2024, with party identification (pid7 1 to 3 and 5 to 7) and the five-point ideology item (liberal 1 to 2, moderate 3, conservative 4 to 5; 'not sure' is not placed) and the survey weight of each wave; the ratio of the CES share to the ANES share is shown under two mappings of the seven-point scale (points 1-3 and 5-7, and points 1-2 and 6-7) with 95% delta-method intervals, and no verdict on the level is drawn. Mode: the ANES adds an online sample in 2012, has mode weights for 2012, 2016 and 2024 only, and none for 2020. Displayed differences are computed from unrounded values, so a printed difference can differ in the last digit from the difference of the printed shares.

References

Wilson, Edwin B. (1927). Probable Inference, the Law of Succession, and Statistical Inference. Journal of the American Statistical Association 22(158): 209-212. doi:10.1080/01621459.1927.10502953
Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Levendusky, Matthew (2009). The Partisan Sort: How Liberals Became Democrats and Conservatives Became Republicans. University of Chicago Press.
Carsey, Thomas M.; Layman, Geoffrey C. (2006). Changing Sides or Changing Minds? Party Identification and Policy Preferences in the American Electorate. American Journal of Political Science 50(2): 464-477. doi:10.1111/j.1540-5907.2006.00196.x
Highton, Benjamin; Kam, Cindy D. (2011). The Long-Term Dynamics of Partisanship and Issue Orientations. Journal of Politics 73(1): 202-215.
Abramowitz, Alan I.; Saunders, Kyle L. (1998). Ideological Realignment in the U.S. Electorate. Journal of Politics 60(3): 634-652. doi:10.2307/2647642
Glenn, Norval D. (1976). Cohort Analysts' Futile Quest: Statistical Attempts to Separate Age, Period and Cohort Effects. American Sociological Review 41(5): 900-904. doi:10.2307/2094738
Black, Earl; Black, Merle (2002). The Rise of Southern Republicans. Belknap Press of Harvard University Press. www.hup.harvard.edu/books/9780674012486
Kinder, Donald R.; Kalmoe, Nathan P. (2017). Neither Liberal nor Conservative: Ideological Innocence in the American Public. University of Chicago Press. press.uchicago.edu/ucp/books/book/chicago/N/bo25841664.html

How to cite

@misc{singh2026howthesorthappened,
  author = {Singh, Vishal},
  title = {Cohort change in partisan sorting},
  year = {2026},
  url = {https://vishalsingh.org/politics/how-the-sort-happened},
  note = {In The Public Mind. Published 30 September 2026.}
}

Same question at another grain

These chapters take up the same axis with a different unit of analysis: individuals, places or countries.

Party and place
- Chapter 15.1, Party and place Individuals
- Chapter 15.5, Composition and behavior in county change (planned) Places

Reading paths

Does place matter

The cross-pressured share was level until 1990, then fell

The fall appears in cohorts born 1927 to 1974

What a cohort line contains

The cohort-mix share of the fall depends on assumptions

The education gap widened, then returned to its 1970s size

The South's gap in the Democratic share of conservatives closed

The fall is smaller with unplaced partisans in the base

A second survey follows a different path, 2008 to 2016

What these data cannot show

Data sources

Methods

References

How to cite

Same question at another grain

Reading paths