Chapter 5.2

Racial attitudes and ideology

Among white non-Hispanic ANES respondents, the correlation between racial resentment and conservative self-placement rose from +0.27 in 1986 to +0.66 in 2024, and most of that rise is between the parties rather than among people who share a party.

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

Among white non-Hispanic respondents to the American National Election Studies (ANES), the correlation between racial resentment and conservative self-placement rose from +0.27 (95% CI 0.19 to 0.34) in 1986 to +0.66 (95% CI 0.63 to 0.69) in 2024. Most of that rise is between the parties. With party identification held fixed, so that respondents are compared only with others in the same party category, the standardized slope is +0.23 (95% CI 0.16 to 0.31) in 1986 and +0.24 (95% CI 0.19 to 0.29) in 2024. Depending on how party is held fixed, 72% to 97% of the per-decade rise in the correlation disappears.

Racial resentment is measured with four ANES statements about the place of Black Americans in society [1]. Each statement is recoded so that its most resentful answer scores one and its least resentful answer scores zero, and the index averages the statements a respondent answered, at least two of the four, and multiplies by one hundred, so it runs from zero for the least resentful answers to one hundred for the most resentful. Differences in its average are in index points. Self-placement is the seven-point question that runs from extremely liberal to extremely conservative. The chapter reports the weighted correlation between the two by survey year, for the 20,658 white non-Hispanic respondents who have both answers in 13 ANES study years between 1986 and 2024. The series for respondents of every race and ethnicity is a toggle on the first figure.Agreement is the resentful answer for the statements that Black people should work their way up without special favors and that they should try harder. Disagreement is the resentful answer for the statements that slavery and discrimination have made it difficult to rise and that Black people have gotten less than they deserve. The direction comes from the answer labels stored with each item, not from any correlation with ideology. Intervals use the Kish effective sample size [2] and Fisher z [3], treat the weighted sample as a simple random sample and ignore clustering. The same four statements define the resentment series in Chapter 1.3, Sort dates by attitude, which dates when each attitude became a partisan divide.

The correlation rose after 2012

Figure 1 shows the correlation for each survey year. It rose slowly from +0.27 in 1986 to +0.43 in 2008 and was +0.44 in 2012. The change between 2008 and 2012, the first Obama term, was +0.01 (95% CI −0.06 to 0.08), which the data cannot distinguish from zero. The rise came afterwards: +0.23 (95% CI 0.19 to 0.27) from 2012 to 2020, to +0.67 in 2020, and the correlation was +0.66 in 2024. The series does not show a rise at the 2008 election. The increase falls in the years around the 2016 election, and studies of that campaign report that views about race and identity were tied more closely to vote choice than in earlier elections [4]. Chapter 5.1, Fifty years of racial attitudes asks whether racial attitudes became more tied to vote choice during the Obama presidency.

Figure 1

Among white respondents the correlation between resentment and conservative self-placement rose from +0.27 in 1986 to +0.66 in 2024

Weighted correlation of the racial resentment index with seven-point liberal-conservative self-placement, by ANES survey year, white non-Hispanic respondents

Show data

Data for Among white respondents the correlation between resentment and conservative self-placement rose from +0.27 in 1986 to +0.66 in 2024
Year	Sample	Correlation	Interval low	Interval high	Respondents	Effective n
1986	White non-Hispanic respondents	+0.27	+0.19	+0.34	664	664
1988	White non-Hispanic respondents	+0.22	+0.16	+0.27	1,014	1,014
1990	White non-Hispanic respondents	+0.31	+0.23	+0.38	529	529
1992	White non-Hispanic respondents	+0.31	+0.26	+0.36	1,345	1,340
1994	White non-Hispanic respondents	+0.33	+0.27	+0.38	1,099	914
1998	White non-Hispanic respondents	+0.28	+0.21	+0.35	789	670
2000	White non-Hispanic respondents	+0.39	+0.30	+0.47	471	389
2004	White non-Hispanic respondents	+0.40	+0.33	+0.47	619	524
2008	White non-Hispanic respondents	+0.43	+0.37	+0.49	823	660
2012	White non-Hispanic respondents	+0.44	+0.40	+0.48	3,111	1,984
2016	White non-Hispanic respondents	+0.58	+0.55	+0.61	2,191	1,562
2020	White non-Hispanic respondents	+0.67	+0.65	+0.69	4,799	2,785
2024	White non-Hispanic respondents	+0.66	+0.63	+0.69	3,204	1,502
1986	All respondents	+0.22	+0.16	+0.29	812	812
1988	All respondents	+0.26	+0.21	+0.31	1,256	1,256
1990	All respondents	+0.26	+0.19	+0.33	657	657
1992	All respondents	+0.32	+0.28	+0.37	1,652	1,646
1994	All respondents	+0.30	+0.25	+0.36	1,381	1,131
1998	All respondents	+0.27	+0.21	+0.33	999	851
2000	All respondents	+0.34	+0.26	+0.41	577	473
2004	All respondents	+0.39	+0.33	+0.45	824	693
2008	All respondents	+0.39	+0.34	+0.44	1,481	980
2012	All respondents	+0.41	+0.38	+0.44	4,930	2,761
2016	All respondents	+0.55	+0.51	+0.58	2,874	2,017
2020	All respondents	+0.64	+0.62	+0.65	6,333	3,532
2024	All respondents	+0.63	+0.60	+0.65	4,263	2,005

The correlation was level from 2008 to 2012 (+0.43 and +0.44), then rose to +0.58 in 2016 and +0.67 in 2020 and held at +0.66 in 2024. Each point is a weighted correlation and the band is the 95% interval. The toggle shows all respondents.Note: The resentment statements were not asked in 1996 and 2002. The 1998 index has two statements, so that year is marked and is not strictly comparable with the years on either side.Note: ANES added web interviews in 2012, and the 2020 and 2024 studies also mix modes. The mode check later in the chapter compares the two modes.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; interval from Fisher z on the Kish effective sample size

The series for all respondents has the same shape at a lower level, from +0.22 (95% CI 0.16 to 0.29) in 1986 to +0.63 (95% CI 0.60 to 0.65) in 2024. Restricting the headline series to white respondents matches the population that Kinder and Sanders study [1].

The parties moved apart on resentment

Party identification is related to both variables, and both links tightened. Among white respondents the correlation between the resentment index and seven-point party identification rose from +0.06 (95% CI −0.01 to 0.12) in 1986 to +0.62 (95% CI 0.59 to 0.65) in 2024, and the correlation between party identification and self-placement rose from +0.37 to +0.79 (Chapter 1.1, Party and ideology covers the second). The second view of Figure 2 plots both. A correlation across all white respondents therefore mixes two things: how resentment and ideology line up among people who share a party, and how far apart the parties sit on both.

The first view of Figure 2 shows where the first change came from. The gap in the average resentment score between white Republicans and white Democrats, counting leaners with the party they lean toward, went from 4.2 (95% CI 0.9 to 7.6) index points in 1986 to 37.6 (95% CI 35.3 to 39.9) in 2024. It widened in two stages. Through 2012 Republicans' average rose, from 61.2 to 75.3, while Democrats' stayed level, at 57.0 in 1986 and 55.5 in 2012. After 2012 Democrats' average fell by 26.0 (95% CI 23.5 to 28.4) index points and Republicans' by 8.1 (95% CI 6.2 to 10.0).Changes and gaps are computed from unrounded estimates, so a printed change can differ from the difference of the printed endpoints in the last digit. The 1998 study asked only two of the four statements, so the level of its index is not comparable with the years on either side and the figure leaves 1998 out of the averages.

Figure 2

The gap in average resentment between white Republicans and white Democrats widened from 4.2 index points in 1986 to 37.6 in 2024

Weighted mean of the racial resentment index (0 least resentful, 100 most resentful) by party identification, leaners included, and, in the second view, weighted correlations, white non-Hispanic respondents

Show data

Data for The gap in average resentment between white Republicans and white Democrats widened from 4.2 index points in 1986 to 37.6 in 2024
Year	View	Series	Mean or correlation	Interval low	Interval high	Respondents	Effective n
1986	means	Democrats	56.99	54.42	59.56	375	375
1988	means	Democrats	60.70	58.74	62.65	538	538
1990	means	Democrats	55.68	53.02	58.34	338	338
1992	means	Democrats	57.25	55.44	59.06	765	762
1994	means	Democrats	59.93	57.84	62.03	555	454
2000	means	Democrats	58.27	55.96	60.59	516	413
2004	means	Democrats	56.20	53.03	59.37	308	252
2008	means	Democrats	56.20	53.64	58.75	443	353
2012	means	Democrats	55.52	53.81	57.23	1,295	823
2016	means	Democrats	41.37	39.32	43.42	1,037	729
2020	means	Democrats	27.10	25.70	28.50	2,224	1,251
2024	means	Democrats	29.57	27.77	31.37	1,547	727
1986	means	Republicans	61.23	59.10	63.36	356	356
1988	means	Republicans	64.87	63.30	66.44	657	657
1990	means	Republicans	59.40	57.02	61.79	317	317
1992	means	Republicans	65.23	63.77	66.69	747	744
1994	means	Republicans	68.27	66.69	69.85	651	546
2000	means	Republicans	68.99	67.11	70.87	529	445
2004	means	Republicans	69.58	67.43	71.73	377	319
2008	means	Republicans	72.60	70.66	74.53	464	377
2012	means	Republicans	75.27	74.06	76.49	1,524	982
2016	means	Republicans	70.60	69.22	71.98	1,270	904
2020	means	Republicans	67.59	66.47	68.71	2,603	1,530
2024	means	Republicans	67.13	65.68	68.58	1,803	852
1986	means	All white respondents	59.57	58.02	61.11	844	844
1988	means	All white respondents	63.62	62.46	64.79	1,355	1,355
1990	means	All white respondents	58.19	56.50	59.88	742	742
1992	means	All white respondents	61.60	60.50	62.69	1,723	1,716
1994	means	All white respondents	64.82	63.58	66.05	1,354	1,120
2000	means	All white respondents	64.12	62.69	65.55	1,197	977
2004	means	All white respondents	63.76	61.93	65.59	757	628
2008	means	All white respondents	64.77	63.19	66.34	1,040	834
2012	means	All white respondents	66.71	65.70	67.72	3,279	2,102
2016	means	All white respondents	58.27	56.98	59.55	2,626	1,847
2020	means	All white respondents	51.02	49.94	52.10	5,374	3,068
2024	means	All white respondents	51.52	50.11	52.93	3,552	1,668
1986	corr	Resentment and ideology	0.27	0.19	0.34	664	664
1988	corr	Resentment and ideology	0.22	0.16	0.27	1,014	1,014
1990	corr	Resentment and ideology	0.31	0.23	0.38	529	529
1992	corr	Resentment and ideology	0.31	0.26	0.36	1,345	1,340
1994	corr	Resentment and ideology	0.33	0.27	0.38	1,099	914
1998	corr	Resentment and ideology	0.28	0.21	0.35	789	670
2000	corr	Resentment and ideology	0.39	0.30	0.47	471	389
2004	corr	Resentment and ideology	0.40	0.33	0.47	619	524
2008	corr	Resentment and ideology	0.43	0.37	0.49	823	660
2012	corr	Resentment and ideology	0.44	0.40	0.48	3,111	1,984
2016	corr	Resentment and ideology	0.58	0.55	0.61	2,191	1,562
2020	corr	Resentment and ideology	0.67	0.65	0.69	4,799	2,785
2024	corr	Resentment and ideology	0.66	0.63	0.69	3,204	1,502
1986	corr	Resentment and party	0.06	−0.01	0.12	838	838
1988	corr	Resentment and party	0.08	0.03	0.13	1,351	1,351
1990	corr	Resentment and party	0.08	0.01	0.15	740	740
1992	corr	Resentment and party	0.15	0.11	0.20	1,714	1,707
1994	corr	Resentment and party	0.19	0.14	0.25	1,347	1,114
1998	corr	Resentment and party	0.17	0.10	0.23	936	795
2000	corr	Resentment and party	0.21	0.15	0.27	1,189	971
2004	corr	Resentment and party	0.27	0.20	0.34	748	620
2008	corr	Resentment and party	0.33	0.27	0.39	1,029	825
2012	corr	Resentment and party	0.40	0.36	0.43	3,269	2,094
2016	corr	Resentment and party	0.49	0.45	0.52	2,618	1,849
2020	corr	Resentment and party	0.62	0.60	0.64	5,367	3,065
2024	corr	Resentment and party	0.62	0.59	0.65	3,540	1,662
1986	corr	Party and ideology	0.37	0.32	0.42	1,316	1,316
1988	corr	Party and ideology	0.45	0.40	0.50	1,131	1,131
1990	corr	Party and ideology	0.40	0.35	0.45	1,040	1,040
1992	corr	Party and ideology	0.47	0.43	0.51	1,462	1,457
1994	corr	Party and ideology	0.53	0.48	0.58	1,111	924
1998	corr	Party and ideology	0.48	0.42	0.54	793	673
2000	corr	Party and ideology	0.51	0.44	0.58	532	436
2004	corr	Party and ideology	0.63	0.58	0.68	675	571
2008	corr	Party and ideology	0.67	0.63	0.71	884	705
2012	corr	Party and ideology	0.68	0.65	0.70	3,326	2,123
2016	corr	Party and ideology	0.75	0.73	0.77	2,490	1,788
2020	corr	Party and ideology	0.79	0.78	0.80	5,290	3,085
2024	corr	Party and ideology	0.79	0.77	0.81	3,528	1,655

The shaded gap between the Republican and Democratic averages widened in two stages: Republicans' average rose from 61.2 in 1986 to 75.3 in 2012, and Democrats' fell from 55.5 in 2012 to 29.6 in 2024. Points are weighted means and bands are 95% intervals. The toggle shows the correlations that link party identification to resentment and to self-placement.Note: Independents who do not lean to either party are left out of the two party lines and included in the all-respondents line.Note: The 1998 study asked only two of the four statements, so the level of its index is not comparable with the years on either side and 1998 is left out of the means; the lines join 1994 and 2000. The resentment statements were not asked in 1996 and 2002.Note: Web respondents score higher than face-to-face respondents in the same year, and these means are not adjusted for mode; the next figure gives the two modes separately.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; interval from the Kish effective sample size (Fisher z for the correlations)

Web respondents score higher than face-to-face respondents in the same year, by between 4.8 and 9.5 index points in the four party-by-year cells for 2012 and 2024, and the averages in Figure 2 are not adjusted for mode. Figure 3 gives the two modes separately: the fall since 2012 appears in each mode and in both parties.

Figure 3

Average resentment fell from 2012 to 2024 in both interview modes and both parties, and web respondents score 4.8 to 9.5 index points higher

Weighted mean of the racial resentment index (0 least resentful, 100 most resentful) by interview mode, with the mode weights, white non-Hispanic Democrats and Republicans, leaners included, 2012 and 2024

Face-to-face
Web

Democrats

Republicans

Mean resentment index (0 to 100)

Show data

Data for Average resentment fell from 2012 to 2024 in both interview modes and both parties, and web respondents score 4.8 to 9.5 index points higher
Party	Year and mode	Mean index	Interval low	Interval high	Respondents	Effective n
Democrats	2012, face-to-face	52.4	49.3	55.6	363	255
Democrats	2012, web	57.2	55.2	59.2	932	579
Democrats	2024, face-to-face	23.0	19.5	26.5	278	150
Democrats	2024, web	32.5	29.8	35.1	543	327
Republicans	2012, face-to-face	72.0	69.7	74.3	397	295
Republicans	2012, web	77.2	75.9	78.6	1,127	719
Republicans	2024, face-to-face	63.8	60.4	67.2	314	171
Republicans	2024, web	69.9	67.7	72.1	599	358

In each panel the 2024 dots lie left of the 2012 dots within each mode, and the web dots lie right of the face-to-face dots within each year. Each dot is a weighted mean using its own mode weight and each line is a 95% interval.Note: ANES 2020 has no mode weights, and the 2024 panel and paper respondents are in neither mode, so the two modes do not cover every respondent.Source: ANES Cumulative Data File, 2012 and 2024; weighted with the mode weights (weight_ftf for in-person rows, weight_web for web rows); white non-Hispanic respondents; interval from the Kish effective sample size

Holding party fixed removes most of the rise

Figure 4 separates the two parts of that correlation. The net-of-party line is the standardized slope of self-placement on the resentment index in a regression that also contains a dummy variable for each of the seven party-identification categories, fitted separately in each year, so the effect of party can differ by year. The slope was +0.23 (95% CI 0.16 to 0.31) in 1986 and +0.24 (95% CI 0.19 to 0.29) in 2024, while the total correlation went from +0.27 to +0.66.A dummy variable equals one for respondents in a category and zero otherwise. The seven categories are strong, weak and leaning Democrats, independents, and the same three for Republicans. Both variables are standardized with weighted means and standard deviations inside each survey year, so the standardized slope of self-placement on the index equals the weighted correlation and the two lines share an axis. The net-of-party line is a slope in units of the spread of both variables across all white respondents, which is larger than their spread inside party categories, so it differs from a correlation among respondents in the same category.

Figure 4

With party held fixed the slope is +0.23 in 1986 and +0.24 in 2024, against a total correlation of +0.27 and +0.66

Standardized slope of seven-point self-placement on the racial resentment index, without and with the seven party-identification categories held fixed, and, in the second view, the correlation inside each party, white non-Hispanic respondents

Standardized slope of self-placement on the index

Show data

Data for With party held fixed the slope is +0.23 in 1986 and +0.24 in 2024, against a total correlation of +0.27 and +0.66
Year	View	Series	Correlation or slope	Interval low	Interval high	Respondents	Effective n
1986	net	All white respondents	+0.27	+0.19	+0.34	664	664
1988	net	All white respondents	+0.22	+0.16	+0.27	1,014	1,014
1990	net	All white respondents	+0.31	+0.23	+0.38	529	529
1992	net	All white respondents	+0.31	+0.26	+0.36	1,345	1,340
1994	net	All white respondents	+0.33	+0.27	+0.38	1,099	914
1998	net	All white respondents	+0.28	+0.21	+0.35	789	670
2000	net	All white respondents	+0.39	+0.30	+0.47	471	389
2004	net	All white respondents	+0.40	+0.33	+0.47	619	524
2008	net	All white respondents	+0.43	+0.37	+0.49	823	660
2012	net	All white respondents	+0.44	+0.40	+0.48	3,111	1,984
2016	net	All white respondents	+0.58	+0.55	+0.61	2,191	1,562
2020	net	All white respondents	+0.67	+0.65	+0.69	4,799	2,785
2024	net	All white respondents	+0.66	+0.63	+0.69	3,204	1,502
1986	net	Net of party identification	+0.23	+0.16	+0.31	662	662
1988	net	Net of party identification	+0.16	+0.10	+0.22	1,011	1,011
1990	net	Net of party identification	+0.25	+0.17	+0.33	529	529
1992	net	Net of party identification	+0.21	+0.16	+0.26	1,342	1,337
1994	net	Net of party identification	+0.21	+0.15	+0.26	1,097	913
1998	net	Net of party identification	+0.18	+0.11	+0.25	784	666
2000	net	Net of party identification	+0.25	+0.16	+0.34	470	389
2004	net	Net of party identification	+0.21	+0.14	+0.29	614	520
2008	net	Net of party identification	+0.20	+0.14	+0.26	817	655
2012	net	Net of party identification	+0.19	+0.15	+0.22	3,104	1,978
2016	net	Net of party identification	+0.22	+0.18	+0.26	2,189	1,561
2020	net	Net of party identification	+0.24	+0.20	+0.27	4,795	2,782
2024	net	Net of party identification	+0.24	+0.19	+0.29	3,200	1,500
1986	within	All white respondents	+0.27	+0.19	+0.34	664	664
1988	within	All white respondents	+0.22	+0.16	+0.27	1,014	1,014
1990	within	All white respondents	+0.31	+0.23	+0.38	529	529
1992	within	All white respondents	+0.31	+0.26	+0.36	1,345	1,340
1994	within	All white respondents	+0.33	+0.27	+0.38	1,099	914
1998	within	All white respondents	+0.28	+0.21	+0.35	789	670
2000	within	All white respondents	+0.39	+0.30	+0.47	471	389
2004	within	All white respondents	+0.40	+0.33	+0.47	619	524
2008	within	All white respondents	+0.43	+0.37	+0.49	823	660
2012	within	All white respondents	+0.44	+0.40	+0.48	3,111	1,984
2016	within	All white respondents	+0.58	+0.55	+0.61	2,191	1,562
2020	within	All white respondents	+0.67	+0.65	+0.69	4,799	2,785
2024	within	All white respondents	+0.66	+0.63	+0.69	3,204	1,502
1986	within	Democrats	+0.41	+0.31	+0.50	292	292
1988	within	Democrats	+0.37	+0.28	+0.45	385	385
1990	within	Democrats	+0.41	+0.30	+0.51	245	245
1992	within	Democrats	+0.39	+0.32	+0.46	582	580
1994	within	Democrats	+0.40	+0.31	+0.49	429	352
1998	within	Democrats	+0.39	+0.29	+0.48	348	299
2000	within	Democrats	+0.35	+0.21	+0.48	204	166
2004	within	Democrats	+0.51	+0.40	+0.60	250	210
2008	within	Democrats	+0.46	+0.36	+0.55	335	267
2012	within	Democrats	+0.39	+0.33	+0.45	1,229	774
2016	within	Democrats	+0.47	+0.40	+0.53	868	609
2020	within	Democrats	+0.46	+0.41	+0.50	2,036	1,147
2024	within	Democrats	+0.44	+0.37	+0.49	1,447	673
1986	within	Republicans	+0.04	−0.07	+0.16	298	298
1988	within	Republicans	+0.06	−0.02	+0.14	532	532
1990	within	Republicans	+0.16	+0.04	+0.28	246	246
1992	within	Republicans	+0.03	−0.05	+0.11	625	623
1994	within	Republicans	+0.08	−0.01	+0.17	579	487
1998	within	Republicans	+0.05	−0.06	+0.16	363	307
2000	within	Republicans	+0.23	+0.09	+0.36	221	186
2004	within	Republicans	+0.05	−0.07	+0.17	326	279
2008	within	Republicans	+0.04	−0.07	+0.14	404	326
2012	within	Republicans	+0.13	+0.07	+0.19	1,473	946
2016	within	Republicans	+0.16	+0.09	+0.23	1,116	805
2020	within	Republicans	+0.26	+0.21	+0.31	2,346	1,410
2024	within	Republicans	+0.26	+0.20	+0.33	1,631	778

The total correlation climbs while the net-of-party slope stays near +0.23. That slope comes from a regression with the seven party-identification categories held fixed, fitted separately in each year. The toggle shows the correlation inside each party. Bands are 95% intervals.Note: Independents who do not lean to either party are in the total and net-of-party lines and in neither party line.Note: The resentment statements were not asked in 1996 and 2002. The 1998 index has two statements, so that year is marked and is not strictly comparable with the years on either side.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; net-of-party slopes with HC3 intervals

A year-interaction test puts numbers on the difference. It regresses the standardized outcome on the index and on the index times decade, counting each survey year equally, and reads the change per decade from the second coefficient. In the net-of-party version the party categories enter as separate dummy variables in every survey year, as they do in the yearly fits. Figure 5 gives the results. In total the correlation gained 0.112 (95% CI 0.090 to 0.133) per decade. With party identification held fixed the slope gained 0.004 (95% CI −0.011 to 0.018), an interval that includes zero, and the reduction is 97% (95% CI 86% to 109%) of the total gain. The upper end of that interval lies above the whole total gain because some redraws of the bootstrap sample give a net gain below zero.The interval for each share comes from a bootstrap, which redraws the sample of respondents inside each year many times and reads the interval from the spread of the estimates. Each trend interval is the wider of two: a respondent-level HC3 interval, and an interval that treats each survey year as a cluster, with critical values from the t distribution for the number of years. Neither is always the wider, so the cautious choice is the wider of the two.

Figure 5

Depending on the measure, 72% to 97% of the per-decade rise in the correlation disappears when party identification is held fixed

Change per decade in the correlation, or in the standardized slope, of self-placement on the racial resentment index, and the share of the total change that disappears when party is held fixed, white non-Hispanic respondents, survey years 1986 to 2024

Measure	Change per decade	Interval low	Interval high	Share of the total change removed	Interval low	Interval high
All white respondents (total correlation)	+0.112	+0.090	+0.133	n/a	n/a	n/a
Net of party identification (standardized slope)	+0.004	−0.011	+0.018	97%	86%	109%
Partial correlation given party identification	+0.018	+0.004	+0.032	84%	74%	96%
Average of the Democratic and Republican correlations	+0.031	+0.017	+0.045	72%	63%	83%
Inside Democrats	+0.018	−0.002	+0.039	n/a	n/a	n/a
Inside Republicans	+0.044	+0.019	+0.070	n/a	n/a	n/a
Republican minus Democratic change	+0.026	−0.003	+0.055	n/a	n/a	n/a

Show data

Data for Depending on the measure, 72% to 97% of the per-decade rise in the correlation disappears when party identification is held fixed
change	change_high	change_low	measure	share	share_high	share_low
0.112	0.133	0.090	All white respondents (total correlation)	n/a	n/a	n/a
0.004	0.018	−0.011	Net of party identification (standardized slope)	96.71	109.2	86.06
0.018	0.032	0.004	Partial correlation given party identification	84.12	96.24	73.74
0.031	0.045	0.017	Average of the Democratic and Republican correlations	71.90	83.04	62.84
0.018	0.039	−0.002	Inside Democrats	n/a	n/a	n/a
0.044	0.070	0.019	Inside Republicans	n/a	n/a	n/a
0.026	0.055	−0.003	Republican minus Democratic change	n/a	n/a	n/a

The total correlation gained 0.112 per decade. With party held fixed the gain is 0.004 on the slope measure, 0.018 on the partial correlation and 0.031 on the average within-party correlation, so most of the rise is between the parties on all three measures. Each share is one minus the ratio of the held-fixed change to the total change.Note: Each share has a percentile interval from a seeded bootstrap that resamples respondents inside each year. The share passes 100 percent in a replicate whose held-fixed change is negative.Note: Each change is a year-interaction coefficient. Its interval is the wider of a respondent-level HC3 interval and a year-clustered interval; for Democrats the year-clustered interval is narrower and excludes zero.Note: The trend tests pool the survey years and do not model the change in interview mode in 2012.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; year-interaction coefficients, each survey year counting equally; shares with percentile intervals from a seeded bootstrap

The share depends on the measure, and the three measures answer different questions. The slope keeps both variables in whole-sample units. The partial correlation removes party from both variables and then standardizes what is left, so it is expressed in units of the spread inside party categories, and its reduction is 84% (95% CI 74% to 96%). The third measure averages the Democratic and Republican correlations shown in the second view of Figure 4, and its reduction is 72% (95% CI 63% to 83%). The three shares differ because the spread of each variable inside a party changed as the parties sorted. The data do not choose among them, and all three place most of the rise between the parties. That is an accounting statement. It does not say that party affiliation changed anyone's answers, and the section on what these data cannot show returns to the direction of influence.

Resentment still tracks ideology within parties

Inside the parties, the correlation is the same measure applied to Democrats and Republicans separately, shown in the second view of Figure 4. Among Democrats it was +0.41 (95% CI 0.31 to 0.50) in 1986 and +0.44 (95% CI 0.37 to 0.49) in 2024, and it stayed between +0.35 and +0.51 in every survey year. Among Republicans it rose from +0.04 (95% CI −0.07 to 0.16) to +0.26 (95% CI 0.20 to 0.33). The gain per decade is 0.018 (95% CI −0.002 to 0.039) among Democrats, an interval that includes zero, and 0.044 (95% CI 0.019 to 0.070) among Republicans. The difference between the two, 0.026 (95% CI −0.003 to 0.055), also includes zero, so these data do not separate the parties.For Democrats the year-clustered interval for the same gain, 0.018 (95% CI 0.008 to 0.029), is narrower and excludes zero. The text and the figures report the wider interval. The gain inside each party is small next to the total gain of 0.112 per decade.

Figure 6 sorts the 13,305 white respondents in the 2012, 2016, 2020 and 2024 studies into thirds of the resentment index, cutting the thirds inside each survey year. The index takes few distinct values, so the thirds are not equal in size: they hold 37%, 36% and 28% of respondents. The most resentful third averages 5.3 (95% CI 5.2 to 5.3) on the seven-point self-placement scale and the least resentful third 3.2 (95% CI 3.1 to 3.2), a difference of 2.1 (95% CI 2.0 to 2.2) points. The third view shows the share who call themselves conservative, meaning slightly conservative, conservative or extremely conservative.

Figure 6

The ideology gap between resentment thirds is 2.1 points among white respondents, 1.1 among Democrats and 0.6 among Republicans

Gap in mean seven-point self-placement (1 extremely liberal, 7 extremely conservative) between the most and least resentful third of the racial resentment index; the other views show the mean in each third and the share conservative; white non-Hispanic respondents, 2012 to 2024 pooled

All white respondents
Democrats
Republicans

Show data

Data for The ideology gap between resentment thirds is 2.1 points among white respondents, 1.1 among Democrats and 0.6 among Republicans
View	Group	Third of the index	Gap in mean self-placement	Low	High	Mean self-placement	Low	High	Percent conservative	Low	High	Respondents
gap	All white respondents	n/a	2.10	2.02	2.17	n/a	n/a	n/a	n/a	n/a	n/a	8,643
gap	Democrats	n/a	1.08	0.92	1.23	n/a	n/a	n/a	n/a	n/a	n/a	4,249
gap	Republicans	n/a	0.58	0.48	0.68	n/a	n/a	n/a	n/a	n/a	n/a	3,711
terc	All white respondents	Least resentful third	n/a	n/a	n/a	3.19	3.14	3.24	18.7	17.3	20.1	5,160
terc	All white respondents	Middle third	n/a	n/a	n/a	4.72	4.67	4.77	54.1	52.3	56.0	4,662
terc	All white respondents	Most resentful third	n/a	n/a	n/a	5.29	5.23	5.34	69.8	67.8	71.8	3,483
terc	Democrats	Least resentful third	n/a	n/a	n/a	2.59	2.55	2.64	4.2	3.3	5.0	3,838
terc	Democrats	Middle third	n/a	n/a	n/a	3.48	3.40	3.56	10.4	8.2	12.5	1,331
terc	Democrats	Most resentful third	n/a	n/a	n/a	3.67	3.52	3.82	18.1	13.3	23.0	411
terc	Republicans	Least resentful third	n/a	n/a	n/a	5.04	4.95	5.13	69.8	65.9	73.6	926
terc	Republicans	Middle third	n/a	n/a	n/a	5.36	5.31	5.41	78.4	76.5	80.4	2,855
terc	Republicans	Most resentful third	n/a	n/a	n/a	5.62	5.57	5.67	81.7	79.8	83.5	2,785

The gap is largest among all white respondents and smaller inside each party. Each dot is a weighted difference in means with a 95% interval. The other views show the mean self-placement in each third, and the share who call themselves slightly conservative, conservative or extremely conservative.Note: Thirds are cut within each survey year on all white respondents, so they mean the same thing in every group. Group sizes are in the tooltips and the data table.Source: ANES Cumulative Data File, 2012, 2016, 2020 and 2024; weighted (VCF0009z); white non-Hispanic respondents; intervals from the Kish effective sample size

Inside each party the gradient is present and narrower. The difference between the ends of the index is 1.1 (95% CI 0.9 to 1.2) points among Democrats and 0.6 (95% CI 0.5 to 0.7) among Republicans. Democrats in the most resentful third average 3.7 (95% CI 3.5 to 3.8), which is left of the scale midpoint, and Republicans in the least resentful third average 5.0 (95% CI 5.0 to 5.1), which is right of it. Those two groups are small, with 411 Democrats in the most resentful third and 926 Republicans in the least resentful third, so their intervals are wide. Within a party, more resentful respondents are more conservative on average, and the difference across all white respondents is larger because the parties differ on both scores.

Each statement and each mode shows the rise

ANES added web interviews in 2012, and the 2020 and 2024 studies also mix modes, so a change in survey mode could raise or lower the correlation. Figure 7 compares the two modes with mode-specific weights. Among face-to-face respondents the correlation went from +0.43 in 2012 to +0.70 (95% CI 0.63 to 0.75) in 2024, and among web respondents from +0.45 to +0.66 (95% CI 0.62 to 0.70). In each year with both modes the two estimates are within 0.05 of each other, and the face-to-face correlation in 2024 is above its 2008 value of +0.43. The comparison covers 2008, 2012, 2016 and 2024. ANES 2020 has no mode weights, so the 2020 point cannot be checked this way.

Figure 7

The correlation rose in both interview modes from 2012 to 2024, from +0.43 to +0.70 face-to-face and from +0.45 to +0.66 on the web

Weighted correlation of the resentment index with seven-point self-placement by interview mode, white non-Hispanic respondents, years with mode weights

Face-to-face
Web

Show data

Data for The correlation rose in both interview modes from 2012 to 2024, from +0.43 to +0.70 face-to-face and from +0.45 to +0.66 on the web
Year and mode	Correlation	Interval low	Interval high	Respondents	Effective n
2008, face-to-face	+0.43	+0.37	+0.49	823	660
2012, face-to-face	+0.43	+0.36	+0.50	702	501
2012, web	+0.45	+0.41	+0.49	2,409	1,522
2016, face-to-face	+0.55	+0.47	+0.61	581	384
2016, web	+0.59	+0.56	+0.63	1,610	1,183
2024, face-to-face	+0.70	+0.63	+0.75	550	296
2024, web	+0.66	+0.62	+0.70	1,098	657

In both modes the dots move right from 2012 to 2024, and within a year the two modes are within 0.05 of each other. Each dot is a weighted correlation using its own mode weight, and each line is a 95% interval; face-to-face estimates rest on fewer respondents and have wider intervals.Note: ANES 2020 has no mode weights, so it has no row. The 2024 panel and paper respondents are in neither mode.Note: ANES interviews were mainly in person before 2012; the mode weights exist from 2008.Source: ANES Cumulative Data File, 2008-2024; weighted with the mode weights (weight_ftf for in-person rows, weight_web for web rows); white non-Hispanic respondents

The index is an average, and an average can rise because one statement does. Figure 8 shows that all four statements move the same way. The statement about slavery and discrimination has the lowest estimate in 1986, +0.14 (95% CI 0.06 to 0.21), although its interval overlaps those of the other three, and it ends at +0.56 (95% CI 0.52 to 0.59). The statement about special favors goes from +0.22 (95% CI 0.15 to 0.29) to +0.59 (95% CI 0.55 to 0.62), the statement about trying harder from +0.22 to +0.57, and the statement about what Black people deserve from +0.24 to +0.58.

Figure 8

Each of the four resentment statements is more strongly correlated with conservative self-placement in 2024 than in 1986

Weighted correlation of each statement, recoded 0 to 1 so that 1 is the most resentful answer, and of the four-statement index, with seven-point self-placement, white non-Hispanic respondents

1986
2024

Show data

Data for Each of the four resentment statements is more strongly correlated with conservative self-placement in 2024 than in 1986
Row	Statement	Correlation, 1986	Low	High	Respondents, 1986	Correlation, 2024	Low	High	Respondents, 2024
Slavery made it hard	Generations of slavery and discrimination have made it difficult for Black people to work their way up	+0.14	+0.06	+0.21	662	+0.56	+0.52	+0.59	3,205
No special favors	Other minorities overcame prejudice without special favors, and Black people should do the same	+0.22	+0.15	+0.29	660	+0.59	+0.55	+0.62	3,084
Try harder	If Black people would only try harder they could be just as well off as white people	+0.22	+0.15	+0.29	661	+0.57	+0.53	+0.60	3,202
Got less than deserved	Over the past few years Black people have gotten less than they deserve	+0.24	+0.17	+0.31	654	+0.58	+0.54	+0.61	3,200
Four-statement index	Mean of the statements answered (at least two), scaled from 0 to 100	+0.27	+0.19	+0.34	664	+0.66	+0.63	+0.69	3,204

Every statement moves right between the two years, from +0.14 to +0.56 for the statement about slavery and discrimination and from +0.22 to +0.59 for the statement about special favors. The open dot is 1986, the filled dot is 2024, and the lines are 95% intervals. The data table gives the full wording of each statement.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; interval from Fisher z on the Kish effective sample size

Health insurance and services show the same rise

Tesler [5] [6] describes a spillover of racialization: once a policy became attached to President Obama, opinion on it came to line up with racial attitudes, even though the policy had no racial content. The ANES policy items let the same question be asked of the resentment index across the survey years. Health insurance and government services show the strengthening link that a spillover account predicts, and it appears inside both parties. Three features limit what that establishes. The defense item, included as a comparison, rises too. The 2008 to 2012 change is mixed with the move to web interviews. And party composition may have changed in ways that the data do not capture.

Figure 9 shows the health-insurance, government-services and defense-spending items, each oriented so that a high value is the conservative position: a private plan rather than a government plan, fewer government services, and higher defense spending. Defense spending has no claim to be a placebo, because its correlation with the index is already +0.23 (95% CI 0.16 to 0.30) in 1986. The health-insurance item was not asked in 1986, 1990 or 1998, and the defense item was not asked in 1998 or 2012.

Figure 9

By 2020 the resentment index correlated with the health-insurance, government-services and defense-spending items at +0.53, +0.48 and +0.53

Weighted correlation of the racial resentment index with three policy items, each oriented so that a high value is the conservative position, white non-Hispanic respondents

Correlation with the resentment index

Health insurance
Government services
Defense spending

Show data

Data for By 2020 the resentment index correlated with the health-insurance, government-services and defense-spending items at +0.53, +0.48 and +0.53
Year	Sample	Outcome	Correlation	Interval low	Interval high	Respondents	Effective n
1988	All white respondents	Health insurance	+0.03	−0.02	+0.09	1,164	1,164
1992	All white respondents	Health insurance	+0.13	+0.08	+0.18	1,515	1,509
1994	All white respondents	Health insurance	+0.16	+0.10	+0.21	1,266	1,052
2000	All white respondents	Health insurance	+0.10	+0.02	+0.19	618	504
2004	All white respondents	Health insurance	+0.18	+0.10	+0.26	710	589
2008	All white respondents	Health insurance	+0.21	+0.12	+0.31	476	384
2012	All white respondents	Health insurance	+0.37	+0.33	+0.40	3,120	2,001
2016	All white respondents	Health insurance	+0.44	+0.40	+0.48	2,380	1,684
2020	All white respondents	Health insurance	+0.53	+0.51	+0.56	4,912	2,869
2024	All white respondents	Health insurance	+0.48	+0.44	+0.52	3,164	1,480
1988	Net of party identification	Health insurance	0.00	−0.05	+0.06	1,160	1,160
1992	Net of party identification	Health insurance	+0.07	+0.02	+0.12	1,508	1,502
1994	Net of party identification	Health insurance	+0.08	+0.02	+0.14	1,261	1,047
2000	Net of party identification	Health insurance	+0.04	−0.05	+0.13	613	500
2004	Net of party identification	Health insurance	+0.09	+0.01	+0.17	701	581
2008	Net of party identification	Health insurance	+0.11	+0.01	+0.22	471	381
2012	Net of party identification	Health insurance	+0.17	+0.13	+0.22	3,113	1,996
2016	Net of party identification	Health insurance	+0.22	+0.17	+0.26	2,374	1,687
2020	Net of party identification	Health insurance	+0.20	+0.16	+0.24	4,906	2,866
2024	Net of party identification	Health insurance	+0.23	+0.17	+0.30	3,157	1,477
1988	Inside Democrats	Health insurance	+0.04	−0.05	+0.13	457	457
1992	Inside Democrats	Health insurance	+0.13	+0.05	+0.20	679	676
1994	Inside Democrats	Health insurance	+0.07	−0.02	+0.17	510	419
2000	Inside Democrats	Health insurance	+0.07	−0.06	+0.20	280	228
2004	Inside Democrats	Health insurance	+0.13	0.00	+0.26	282	231
2008	Inside Democrats	Health insurance	+0.24	+0.09	+0.38	199	160
2012	Inside Democrats	Health insurance	+0.24	+0.17	+0.30	1,232	782
2016	Inside Democrats	Health insurance	+0.37	+0.30	+0.43	943	661
2020	Inside Democrats	Health insurance	+0.28	+0.23	+0.33	2,094	1,189
2024	Inside Democrats	Health insurance	+0.37	+0.30	+0.44	1,429	671
1988	Inside Republicans	Health insurance	+0.01	−0.07	+0.10	581	581
1992	Inside Republicans	Health insurance	+0.05	−0.03	+0.12	659	657
1994	Inside Republicans	Health insurance	+0.15	+0.06	+0.23	628	529
2000	Inside Republicans	Health insurance	+0.06	−0.07	+0.19	266	226
2004	Inside Republicans	Health insurance	+0.07	−0.04	+0.19	358	302
2008	Inside Republicans	Health insurance	+0.02	−0.13	+0.16	217	177
2012	Inside Republicans	Health insurance	+0.17	+0.11	+0.23	1,473	950
2016	Inside Republicans	Health insurance	+0.15	+0.09	+0.22	1,178	849
2020	Inside Republicans	Health insurance	+0.19	+0.14	+0.24	2,372	1,426
2024	Inside Republicans	Health insurance	+0.16	+0.09	+0.23	1,590	746
1986	All white respondents	Government services	+0.06	−0.01	+0.13	769	769
1988	All white respondents	Government services	+0.07	+0.01	+0.13	1,105	1,105
1990	All white respondents	Government services	+0.12	+0.04	+0.19	633	633
1992	All white respondents	Government services	+0.15	+0.10	+0.20	1,463	1,458
1994	All white respondents	Government services	+0.13	+0.07	+0.19	1,222	1,009
1998	All white respondents	Government services	+0.12	+0.05	+0.19	875	744
2000	All white respondents	Government services	+0.16	+0.07	+0.24	574	470
2004	All white respondents	Government services	+0.17	+0.09	+0.25	674	564
2008	All white respondents	Government services	+0.13	+0.02	+0.23	450	366
2012	All white respondents	Government services	+0.35	+0.31	+0.39	3,027	1,939
2016	All white respondents	Government services	+0.45	+0.41	+0.49	2,300	1,629
2020	All white respondents	Government services	+0.48	+0.45	+0.51	4,693	2,735
2024	All white respondents	Government services	+0.50	+0.46	+0.54	3,131	1,468
1986	Net of party identification	Government services	+0.04	−0.04	+0.11	764	764
1988	Net of party identification	Government services	+0.05	−0.01	+0.11	1,101	1,101
1990	Net of party identification	Government services	+0.09	+0.01	+0.17	632	632
1992	Net of party identification	Government services	+0.09	+0.04	+0.14	1,459	1,454
1994	Net of party identification	Government services	+0.06	0.00	+0.13	1,215	1,003
1998	Net of party identification	Government services	+0.08	+0.01	+0.15	870	740
2000	Net of party identification	Government services	+0.07	−0.03	+0.16	570	467
2004	Net of party identification	Government services	+0.07	−0.02	+0.16	666	556
2008	Net of party identification	Government services	−0.01	−0.10	+0.08	448	364
2012	Net of party identification	Government services	+0.16	+0.12	+0.20	3,020	1,934
2016	Net of party identification	Government services	+0.21	+0.16	+0.26	2,297	1,626
2020	Net of party identification	Government services	+0.18	+0.14	+0.23	4,690	2,734
2024	Net of party identification	Government services	+0.23	+0.18	+0.29	3,125	1,465
1986	Inside Democrats	Government services	+0.15	+0.04	+0.25	335	335
1988	Inside Democrats	Government services	+0.12	+0.03	+0.21	436	436
1990	Inside Democrats	Government services	+0.22	+0.11	+0.33	293	293
1992	Inside Democrats	Government services	+0.14	+0.07	+0.22	656	654
1994	Inside Democrats	Government services	+0.12	+0.02	+0.22	488	398
1998	Inside Democrats	Government services	+0.13	+0.02	+0.23	388	335
2000	Inside Democrats	Government services	+0.04	−0.09	+0.18	261	209
2004	Inside Democrats	Government services	+0.07	−0.06	+0.20	273	227
2008	Inside Democrats	Government services	+0.10	−0.07	+0.25	185	150
2012	Inside Democrats	Government services	+0.24	+0.17	+0.30	1,193	758
2016	Inside Democrats	Government services	+0.34	+0.27	+0.40	930	650
2020	Inside Democrats	Government services	+0.25	+0.20	+0.31	2,042	1,147
2024	Inside Democrats	Government services	+0.32	+0.25	+0.38	1,424	663
1986	Inside Republicans	Government services	−0.08	−0.18	+0.03	330	330
1988	Inside Republicans	Government services	−0.02	−0.10	+0.06	556	556
1990	Inside Republicans	Government services	−0.03	−0.15	+0.09	275	275
1992	Inside Republicans	Government services	+0.10	+0.02	+0.17	656	653
1994	Inside Republicans	Government services	+0.05	−0.04	+0.13	615	516
1998	Inside Republicans	Government services	+0.09	−0.02	+0.20	381	322
2000	Inside Republicans	Government services	+0.07	−0.06	+0.20	251	214
2004	Inside Republicans	Government services	+0.05	−0.07	+0.16	338	287
2008	Inside Republicans	Government services	−0.09	−0.24	+0.06	213	174
2012	Inside Republicans	Government services	+0.15	+0.08	+0.21	1,454	936
2016	Inside Republicans	Government services	+0.16	+0.09	+0.22	1,133	813
2020	Inside Republicans	Government services	+0.14	+0.09	+0.19	2,234	1,356
2024	Inside Republicans	Government services	+0.19	+0.12	+0.26	1,564	741
1986	All white respondents	Defense spending	+0.23	+0.16	+0.30	769	769
1988	All white respondents	Defense spending	+0.23	+0.18	+0.29	1,226	1,226
1990	All white respondents	Defense spending	+0.29	+0.22	+0.36	664	664
1992	All white respondents	Defense spending	+0.25	+0.20	+0.29	1,541	1,535
1994	All white respondents	Defense spending	+0.21	+0.15	+0.27	1,277	1,058
2000	All white respondents	Defense spending	+0.22	+0.13	+0.30	571	464
2004	All white respondents	Defense spending	+0.31	+0.23	+0.38	684	575
2008	All white respondents	Defense spending	+0.30	+0.20	+0.39	459	369
2016	All white respondents	Defense spending	+0.48	+0.44	+0.52	2,346	1,679
2020	All white respondents	Defense spending	+0.53	+0.50	+0.55	4,824	2,781
2024	All white respondents	Defense spending	+0.44	+0.40	+0.48	3,099	1,450
1986	Net of party identification	Defense spending	+0.21	+0.14	+0.28	764	764
1988	Net of party identification	Defense spending	+0.21	+0.16	+0.27	1,224	1,224
1990	Net of party identification	Defense spending	+0.27	+0.20	+0.35	663	663
1992	Net of party identification	Defense spending	+0.21	+0.16	+0.26	1,539	1,533
1994	Net of party identification	Defense spending	+0.16	+0.10	+0.22	1,271	1,053
2000	Net of party identification	Defense spending	+0.16	+0.06	+0.25	566	460
2004	Net of party identification	Defense spending	+0.21	+0.11	+0.30	677	569
2008	Net of party identification	Defense spending	+0.22	+0.11	+0.33	455	366
2016	Net of party identification	Defense spending	+0.35	+0.30	+0.40	2,343	1,676
2020	Net of party identification	Defense spending	+0.34	+0.29	+0.38	4,822	2,779
2024	Net of party identification	Defense spending	+0.31	+0.25	+0.37	3,096	1,449
1986	Inside Democrats	Defense spending	+0.32	+0.22	+0.41	339	339
1988	Inside Democrats	Defense spending	+0.27	+0.18	+0.35	478	478
1990	Inside Democrats	Defense spending	+0.37	+0.27	+0.46	302	302
1992	Inside Democrats	Defense spending	+0.29	+0.22	+0.35	690	687
1994	Inside Democrats	Defense spending	+0.24	+0.15	+0.33	516	422
2000	Inside Democrats	Defense spending	+0.17	+0.03	+0.30	259	207
2004	Inside Democrats	Defense spending	+0.33	+0.21	+0.44	274	229
2008	Inside Democrats	Defense spending	+0.27	+0.12	+0.41	193	155
2016	Inside Democrats	Defense spending	+0.46	+0.40	+0.52	933	664
2020	Inside Democrats	Defense spending	+0.42	+0.37	+0.46	2,029	1,136
2024	Inside Democrats	Defense spending	+0.35	+0.28	+0.42	1,397	656
1986	Inside Republicans	Defense spending	+0.10	−0.01	+0.20	332	332
1988	Inside Republicans	Defense spending	+0.22	+0.14	+0.29	618	618
1990	Inside Republicans	Defense spending	+0.20	+0.09	+0.30	297	297
1992	Inside Republicans	Defense spending	+0.10	+0.03	+0.18	681	679
1994	Inside Republicans	Defense spending	+0.17	+0.09	+0.25	634	533
2000	Inside Republicans	Defense spending	+0.14	0.00	+0.26	255	217
2004	Inside Republicans	Defense spending	+0.13	+0.02	+0.24	349	297
2008	Inside Republicans	Defense spending	+0.23	+0.08	+0.36	215	174
2016	Inside Republicans	Defense spending	+0.23	+0.16	+0.29	1,158	835
2020	Inside Republicans	Defense spending	+0.22	+0.17	+0.27	2,374	1,410
2024	Inside Republicans	Defense spending	+0.23	+0.16	+0.30	1,570	739

The health-insurance and government-services lines rise between 2008 and 2012, from +0.21 to +0.37 and from +0.13 to +0.35, the window in which ANES added web interviews. The defense line starts higher and has no 2012 estimate. The views hold party identification fixed or restrict to one party. Bands are 95% intervals.Note: The left dashed rule marks 1998, when the index had two statements; the right dashed rule marks 2012, when ANES added web interviews.Note: The health-insurance item was not asked in 1986, 1990 or 1998, and the defense item was not asked in 1998 or 2012, so those lines join the years on either side. The resentment statements were not asked in 1996 and 2002.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; net-of-party slopes with HC3 intervals

Among white respondents the correlation of the index with the health-insurance item rose from +0.03 (95% CI −0.02 to 0.09) in 1988 to +0.48 (95% CI 0.44 to 0.52) in 2024, and with the government-services item from +0.06 (95% CI −0.01 to 0.13) in 1986 to +0.50 (95% CI 0.46 to 0.54). The defense item ends at +0.44 (95% CI 0.40 to 0.48). Figure 11 gives the gain per decade for each item.

The 2008 to 2012 window is the comparison that bears on a spillover account, and it is also the window in which ANES added web interviews. Among all respondents the health-insurance correlation changed by +0.15 (95% CI 0.05 to 0.26) and the government-services correlation by +0.22 (95% CI 0.11 to 0.33). Figure 10 separates the modes.

In 2012 web respondents show higher correlations than face-to-face respondents for both items, by +0.14 (95% CI 0.05 to 0.23) for health insurance and +0.14 (95% CI 0.05 to 0.23) for government services. Among face-to-face respondents alone, the only mode in 2008, the change from 2008 to 2012 is +0.06 (95% CI −0.06 to 0.19) for health insurance, an interval that includes zero, and +0.13 (95% CI 0.00 to 0.26) for government services, an interval whose lower end is at zero. The 2008 to 2012 change in the all-interview series therefore cannot be separated from the move to web interviews.

Figure 10

In 2012 web respondents' correlations with the health-insurance and government-services items were +0.42 and +0.40, against +0.28 and +0.26 face-to-face

Weighted correlation of the racial resentment index with each policy item, oriented so that a high value is the conservative position, by interview mode, white non-Hispanic respondents, years with mode weights

Face-to-face
Web

Health insurance

Government services

Correlation with the resentment index

Show data

Data for In 2012 web respondents' correlations with the health-insurance and government-services items were +0.42 and +0.40, against +0.28 and +0.26 face-to-face
Item	Year and mode	Correlation	Interval low	Interval high	Respondents	Effective n
Health insurance	2008, face-to-face	+0.21	+0.12	+0.31	476	384
Health insurance	2012, face-to-face	+0.28	+0.20	+0.35	801	573
Health insurance	2012, web	+0.42	+0.37	+0.46	2,319	1,470
Health insurance	2016, face-to-face	+0.47	+0.40	+0.54	671	437
Health insurance	2016, web	+0.43	+0.38	+0.47	1,709	1,252
Health insurance	2024, face-to-face	+0.51	+0.42	+0.58	591	317
Health insurance	2024, web	+0.47	+0.41	+0.53	1,120	677
Government services	2008, face-to-face	+0.13	+0.02	+0.23	450	366
Government services	2012, face-to-face	+0.26	+0.18	+0.34	755	539
Government services	2012, web	+0.40	+0.35	+0.44	2,272	1,442
Government services	2016, face-to-face	+0.43	+0.35	+0.51	631	410
Government services	2016, web	+0.46	+0.41	+0.50	1,669	1,222
Government services	2024, face-to-face	+0.53	+0.44	+0.61	545	294
Government services	2024, web	+0.47	+0.41	+0.53	1,077	648

In 2012 the web dot lies right of the face-to-face dot in both panels and the intervals do not overlap. ANES interviewed face to face only in 2008, so the 2008 to 2012 change among face-to-face respondents is +0.06 for health insurance and +0.13 for government services, against +0.15 and +0.22 among all respondents. Each dot is a weighted correlation with its own mode weight, and each line is a 95% interval.Note: ANES 2020 has no mode weights, so it has no row. The 2024 panel and paper respondents are in neither mode. The defense item is left out because it was not asked in 2012.Source: ANES Cumulative Data File, 2008-2024; weighted with the mode weights (weight_ftf for in-person rows, weight_web for web rows); white non-Hispanic respondents

Party identification accounts for part of the per-decade gains and not all of them. Holding party fixed removes 56% (95% CI 49% to 64%) to 67% (95% CI 59% to 77%) of the gain across the policy items, against 97% for self-placement, and the gains net of party stay above zero for every item. Inside the Democratic Party the health-insurance correlation rose from +0.04 in 1988 to +0.37 in 2024, and inside the Republican Party from +0.01 to +0.16, while the self-placement correlation inside the Democratic Party gained 0.018 per decade. The health-insurance item therefore moved with resentment inside that party where the ideology label moved little. The same patterns would also arise if the party groups had changed in composition in ways that the seven categories do not capture.

Figure 11

Holding party identification fixed removes 56% to 67% of the per-decade rise in the three policy items, against 97% for self-placement

Change per decade in the standardized slope of each policy item on the racial resentment index, with 95% intervals, white non-Hispanic respondents, survey years 1986 to 2024

All white respondents
Net of party identification
Democrats
Republicans

Health insurance

Government services

Defense spending

Change in the standardized slope per decade

Show data

Data for Holding party identification fixed removes 56% to 67% of the per-decade rise in the three policy items, against 97% for self-placement
Outcome	Sample	Change per decade	Interval low	Interval high	Survey years	Respondents
Health insurance	All white respondents	+0.135	+0.106	+0.165	10	19,264
Health insurance	Net of party identification	+0.060	+0.045	+0.075	10	19,264
Health insurance	Democrats	+0.092	+0.067	+0.117	10	8,105
Health insurance	Republicans	+0.038	+0.016	+0.060	10	9,322
Government services	All white respondents	+0.118	+0.095	+0.141	13	20,857
Government services	Net of party identification	+0.039	+0.017	+0.061	13	20,857
Government services	Democrats	+0.044	+0.016	+0.073	13	8,904
Government services	Republicans	+0.051	+0.028	+0.073	13	10,000
Defense spending	All white respondents	+0.073	+0.045	+0.101	11	17,420
Defense spending	Net of party identification	+0.032	+0.009	+0.054	11	17,420
Defense spending	Democrats	+0.031	0.000	+0.061	11	7,410
Defense spending	Republicans	+0.024	+0.004	+0.044	11	8,483

In every panel the dot net of party identification lies left of the total, and its interval stays above zero. The net-of-party fits let the effect of party differ by year. For health insurance the total change is 0.135 per decade and the change net of party is 0.060.Note: The interval is the wider of a respondent-level HC3 interval and an interval that clusters on survey year with t critical values.Note: The health-insurance item was asked in fewer survey years than the others, and the defense item was not asked in 1998 or 2012.Note: Each test pools the survey years and does not model the change in interview mode in 2012.Source: ANES Cumulative Data File, 1986-2024; weighted (VCF0009z); white non-Hispanic respondents; year-interaction coefficients, each survey year counting equally

The resentment index is contested

The scale has a long history of dispute. Kinder and Sears [7] proposed symbolic racism as a blend of anti-Black affect and traditional moral values, and Sears and Henry [8] report evidence that the measure combines anti-Black affect with conservative values such as individualism. Sniderman and Tetlock [9] argued that inferring prejudice from answers to such items attributes a motive that principled conservatism can also explain. Feldman and Huddy [10] found that resentment means different things to white liberals and conservatives: among liberals it behaved as a measure of prejudice, and among conservatives it was tied to opposition to race-conscious programs regardless of the race of the recipient. Carmines, Sniderman and Easter [11] concluded that the items conflate resentment with racial policy attitudes and do not validly measure racism. The evidence here cannot settle the dispute. Two of the statements, about special favors and about trying harder, invite an answer based on views about government and individual responsibility, and those views are also related to self-placement. A correlation between the index and ideology is therefore partly a correlation between two related political attitudes, and if the scale carries different content on the left and the right, the correlation mixes those contents.

Two checks bear on the measure without resolving it. Among white respondents the index correlates −0.32 (95% CI −0.38 to −0.25) with the feeling thermometer toward Black Americans in 1986 and −0.28 (95% CI −0.32 to −0.23) in 2024. The sign is negative because more resentful respondents rate Black Americans colder. The yearly values run from −0.10 in 1998, when only two statements were asked, to −0.35 in 2016, so the index is related to warmth without being the same thing. And the rise in Figure 8 appears in the statements about slavery and about what Black people deserve as well as in those about favors and effort, so the rise does not come from the items with the most policy content alone.

What these data cannot show

The ANES is a series of cross-sections. A change from one year to the next mixes people who changed their answers, people who changed parties, and the replacement of older cohorts by younger ones, and this chapter cannot separate them. The Republican rise in the within-party correlation, for example, could come from Republicans who changed their views or from a different set of people calling themselves Republican in 2024 than in 1986.

The direction of influence is not identified. Resentment and party identification are measured in the same interview. Holding party fixed compares respondents who share a party label, and it does not show that party identification is an outcome or a cause of resentment. Engelhardt [12] reports, from panel studies in the 1990s and 2000s, that white respondents aligned their racial attitudes with their party loyalties, so a change in resentment scores can follow party as well as precede it. The data also cannot show why the parties moved apart. They hold no measures of the messages voters heard, no panel that follows the same people, and no experiment.

The share of the rise that is between the parties depends on how the association is measured, and the intervals reflect sampling error only. Several other limits are about measurement and coverage. The headline series describes white non-Hispanic respondents, and the toggle for all respondents pools every group as a comparison. The resentment statements were not asked in 1996 and 2002, and only two were asked in 1998. Intervals assume a simple random sample and ignore the ANES sampling design, which can raise or lower the variance. The mode checks cover the years with mode weights and leave out 2020, mean resentment and the policy-item correlations differ by mode within a year, and the trend tests pool the survey years without modeling the change to web interviews in 2012. The spillover comparison rests on policy items that the ANES asked in most years, it cannot separate the 2008 to 2012 change from the move to web interviews, and it cannot tell a racialized policy from one that became partisan for other reasons. The Floyd period in 2020 is in Chapter 5.3, Racial attitudes after George Floyd's death (planned), and other racial attitudes, such as views on interracial marriage and aid to Black Americans, are in Chapter 5.1, Fifty years of racial attitudes.

Data sources

ANES Cumulative Data FileCumulative Data File 1948-2024; racial resentment statements VCF9039 to VCF9042 (asked 1986-2024), self-placement VCF0803, party identification VCF0301, policy items VCF0806, VCF0839 and VCF0843Years: 1948-2024 (33 study years; even years only after 2004; no 2006, 2010 or 2014 study) · Weights: VCF0009z; weight_ftf and weight_web for the mode comparison · License: Free download after accepting the ANES terms of use; cite the release.

All data sources

Methods

Sample: white non-Hispanic respondents in the ANES Cumulative Data File, the 13 study years from 1986 to 2024 with at least two of the four resentment statements (none in 1996 and 2002, two in 1998); 20,658 respondents have an index score and a self-placement. Index: each statement recoded from 0 (least resentful answer) to 1 (most resentful), averaged over the statements answered and multiplied by 100. Weights: VCF0009z; mode estimates use weight_ftf and weight_web (2008, 2012, 2016 and 2024). Estimator: weighted Pearson correlation, equal to the standardized slope when both variables are standardized within the year; net of party means dummy variables for the seven party-identification categories, fitted in each year. Trends: change per decade from a weighted regression on index x decade, with the wider of an HC3 and a year-clustered interval. Intervals: 95%; Fisher z on the Kish effective sample size for correlations; percentile intervals from a seeded bootstrap (500 replicates, seed 12345) for shares; none reflects the survey design. Suppression: at least 150 respondents and an effective sample of 100 per year. Changes and shares use unrounded estimates.

References

Kinder, Donald R.; Sanders, Lynn M. (1996). Divided by Color: Racial Politics and Democratic Ideals. University of Chicago Press.
Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Fisher, Ronald A. (1921). On the Probable Error of a Coefficient of Correlation Deduced from a Small Sample. Metron 1(4): 3-32.
Sides, John; Tesler, Michael; Vavreck, Lynn (2018). Identity Crisis: The 2016 Presidential Campaign and the Battle for the Meaning of America. Princeton University Press. press.princeton.edu/books/hardcover/9780691174198/identity-crisis
Tesler, Michael (2012). The Spillover of Racialization into Health Care: How President Obama Polarized Public Opinion by Racial Attitudes and Race. American Journal of Political Science 56(3): 690-704. doi:10.1111/j.1540-5907.2011.00577.x
Tesler, Michael (2016). Post-Racial or Most-Racial? Race and Politics in the Obama Era. University of Chicago Press. press.uchicago.edu/ucp/books/book/chicago/P/bo22961444.html
Kinder, Donald R.; Sears, David O. (1981). Prejudice and Politics: Symbolic Racism Versus Racial Threats to the Good Life. Journal of Personality and Social Psychology 40(3): 414-431.
Sears, David O.; Henry, P. J. (2003). The Origins of Symbolic Racism. Journal of Personality and Social Psychology 85(2): 259-275.
Sniderman, Paul M.; Tetlock, Philip E. (1986). Symbolic Racism: Problems of Motive Attribution in Political Analysis. Journal of Social Issues 42(2): 129-150. doi:10.1111/j.1540-4560.1986.tb00229.x
Feldman, Stanley; Huddy, Leonie (2005). Racial Resentment and White Opposition to Race-Conscious Programs: Principles or Prejudice?. American Journal of Political Science 49(1): 168-183. doi:10.1111/j.0092-5853.2005.00117.x
Carmines, Edward G.; Sniderman, Paul M.; Easter, Beth C. (2011). On the Meaning, Measurement, and Implications of Racial Resentment. Annals of the American Academy of Political and Social Science 634(1): 98-116. doi:10.1177/0002716210387499
Engelhardt, Andrew M. (2021). Racial Attitudes Through a Partisan Lens. British Journal of Political Science 51(3): 1062-1079. doi:10.1017/S0007123419000437

How to cite

@misc{singh2026racialattitudesandideology,
  author = {Singh, Vishal},
  title = {Racial attitudes and ideology},
  year = {2026},
  url = {https://vishalsingh.org/politics/racial-attitudes-and-ideology},
  note = {In The Public Mind. Published 30 September 2026.}
}

The correlation rose after 2012

The parties moved apart on resentment

Holding party fixed removes most of the rise

Resentment still tracks ideology within parties

Each statement and each mode shows the rise

Health insurance and services show the same rise

The resentment index is contested

What these data cannot show

Data sources

Methods

References

How to cite

Reading paths