Chapter 9.1

The god gap in America

The Democratic vote gap between seldom and frequent attenders rose from 5.5 points in the ANES studies of 1972 to 1988 to 21.8 points in those of 2008 to 2020 and was 27.7 points in the 2024 CES.

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

Still open: the ANES vote gap is not split by survey mode, the design checks cover the GSS only, and race and tradition are separated only in three views of one figure. Black Protestants, Latinos and non-Christian faiths are left to the chapter on religious tradition and race.

Among voters who reported a choice between the two major-party candidates in the 2024 Cooperative Election Study (CES), those who attend religious services a few times a year or never voted Democratic at a rate 27.7 (95% CI 26.0 to 29.3) points above those who attend weekly or more. This chapter calls that difference between seldom and frequent attenders the god gap. In the American National Election Studies (ANES) of 1972 to 1988 the same gap was 5.5 points, and in those of 2008 to 2020 it was 21.8 points. The chapter follows the gap from the vote to party identification and to ideology, where the measure is a slope of ideology on attendance, and compares the United States with Europe and the rest of the world. In the General Social Survey (GSS), party identification accounts for 46% (95% CI 33% to 67%) of the rise in that slope between the in-person surveys of 1974 to 1978 and those of 2010 to 2018.

Frequent attenders go to services weekly or nearly weekly, and seldom attenders go a few times a year or never. The ANES records attendance in three categories only, so the other surveys are cut to match as closely as their items allow: the CES has no nearly-weekly answer, so its frequent group is weekly or more.Attendance is self-reported, so each group is defined by how respondents answer and not by a count of visits. The cut is the same in the ANES and the GSS and one answer narrower in the CES, and the levels of the groups are compared across surveys only where the text says so. A gap is the Democratic share of the two-party vote among seldom attenders minus the share among frequent attenders, in points, so a positive gap means that seldom attenders voted more Democratic. It describes who votes for whom and does not say why. A pooled estimate combines the respondents of several studies into one estimate.Every interval is a 95% confidence interval. Shares use the Kish effective sample size [1] and treat the weighted sample as a simple random sample, which leaves out strata and clusters and can move an interval either way, and they describe sampling error within one survey and not differences between surveys. The CES is an opt-in online panel, so its respondents chose to join; the same formula is used for it, and its interval does not cover the error that comes from how respondents were recruited. Studies of American party politics through the 1990s described a widening division between religious traditionalists and seculars in party attachments [2], and a later account of American religion documented polarization between religious conservatives and secular liberals [3]. The sections below measure that division in several surveys.

The ANES vote gap has been above zero since 1992

Figure 1 plots the gap in every presidential year. In the ANES it was 10.3 (95% CI 5.3 to 15.3) points in 1972, 0.1 (95% CI −6.5 to 6.6) in 1980, an interval that includes zero, 11.3 (95% CI 6.2 to 16.5) in 1992 and 21.9 (95% CI 18.2 to 25.7) in 2020. The interval lies above zero in 8 of the 8 studies from 1992 on, and it includes zero in 3 studies, those of 1980, 1984 and 1988. Pooled over 1972 to 1988 the ANES gap is 5.5 (95% CI 2.9 to 8.1) points, and pooled over 1992 to 2004 it is 13.0 (95% CI 9.7 to 16.2) points.

Figure 1

In the ANES the Democratic vote gap between seldom and frequent attenders included zero in 1980, 1984 and 1988, and in the 2024 CES it was 27.7 points

Democratic share of the two-party presidential vote among seldom attenders (a few times a year or never) minus frequent attenders (weekly or nearly weekly, and weekly or more in the CES), reported voters, percentage points, 95% intervals; ANES 1972 to 2020 and CES 2008 to 2024

Show data

Data for In the ANES the Democratic vote gap between seldom and frequent attenders included zero in 1980, 1984 and 1988, and in the 2024 CES it was 27.7 points
Sample	Survey	Year	Gap (points)	Lower	Upper	Seldom, Democratic share (%)	Frequent, Democratic share (%)	Seldom voters	Frequent voters
All voters	ANES	1972	10.3	5.3	15.3	40.6	30.3	705	684
All voters	ANES	1976	7.5	1.5	13.5	53.5	46.0	536	578
All voters	ANES	1980	0.1	−6.5	6.6	38.8	38.7	441	408
All voters	ANES	1984	5.2	−0.3	10.8	43.0	37.8	621	580
All voters	ANES	1988	2.1	−4.1	8.2	47.1	45.1	522	497
All voters	ANES	1992	11.3	6.2	16.5	53.4	42.0	745	684
All voters	ANES	1996	14.6	7.7	21.5	61.3	46.7	459	496
All voters	ANES	2000	14.9	7.9	21.9	58.4	43.5	453	481
All voters	ANES	2004	12.6	4.5	20.7	55.9	43.3	371	316
All voters	ANES	2008	21.1	14.4	27.8	62.2	41.2	666	600
All voters	ANES	2012	20.5	16.2	24.8	61.2	40.8	2,198	1,543
All voters	ANES	2016	23.8	18.9	28.7	62.0	38.2	1,432	884
All voters	ANES	2020	21.9	18.2	25.7	62.3	40.4	3,950	1,723
All voters	CES	2008	25.5	23.9	27.1	63.4	37.9	13,147	8,021
All voters	CES	2012	27.3	25.7	28.9	62.3	35.0	22,397	11,984
All voters	CES	2016	18.9	17.4	20.4	56.9	37.9	26,471	10,902
All voters	CES	2020	23.5	22.1	24.9	60.0	36.5	30,080	10,434
All voters	CES	2024	27.7	26.0	29.3	57.6	29.9	29,642	9,415
Protestant and Catholic voters	ANES	1972	6.3	1.1	11.5	36.4	30.1	604	677
Protestant and Catholic voters	ANES	1976	4.5	−1.8	10.8	50.6	46.1	450	570
Protestant and Catholic voters	ANES	1980	0.6	−6.5	7.7	39.4	38.8	340	402
Protestant and Catholic voters	ANES	1984	2.1	−3.8	8.0	39.5	37.3	484	568
Protestant and Catholic voters	ANES	1988	−1.7	−8.2	4.8	43.0	44.7	407	492
Protestant and Catholic voters	ANES	1992	8.3	2.5	14.2	49.8	41.5	488	643
Protestant and Catholic voters	ANES	1996	15.7	7.9	23.5	61.2	45.5	301	469
Protestant and Catholic voters	ANES	2000	9.4	1.3	17.6	53.5	44.0	289	458
Protestant and Catholic voters	ANES	2004	11.0	1.4	20.6	55.1	44.1	211	292
Protestant and Catholic voters	ANES	2008	16.4	8.5	24.3	57.3	40.9	394	568
Protestant and Catholic voters	ANES	2012	14.7	9.5	19.9	55.4	40.7	1,207	1,348
Protestant and Catholic voters	ANES	2016	15.3	9.5	21.2	52.9	37.5	759	797
Protestant and Catholic voters	ANES	2020	11.8	7.2	16.3	52.6	40.9	1,938	1,403
Protestant and Catholic voters	CES	2008	18.7	16.6	20.7	51.9	33.3	6,415	5,786
Protestant and Catholic voters	CES	2012	15.3	13.5	17.2	50.6	35.3	11,833	10,669
Protestant and Catholic voters	CES	2016	9.0	7.3	10.8	45.3	36.3	12,636	9,457
Protestant and Catholic voters	CES	2020	11.0	9.3	12.7	45.7	34.8	12,431	8,674
Protestant and Catholic voters	CES	2024	13.2	11.2	15.2	42.2	29.0	12,221	7,759
White non-Hispanic voters	ANES	1972	9.9	4.8	15.0	35.8	25.9	640	622
White non-Hispanic voters	ANES	1976	8.3	2.1	14.6	49.6	41.2	486	523
White non-Hispanic voters	ANES	1980	3.5	−3.2	10.3	33.6	30.1	384	346
White non-Hispanic voters	ANES	1984	9.0	3.2	14.8	39.4	30.4	545	483
White non-Hispanic voters	ANES	1988	6.5	−0.1	13.1	42.5	36.0	447	400
White non-Hispanic voters	ANES	1992	17.9	12.3	23.4	50.1	32.2	640	542
White non-Hispanic voters	ANES	1996	22.5	15.0	29.9	57.3	34.8	397	386
White non-Hispanic voters	ANES	2000	18.8	11.1	26.4	54.9	36.1	377	383
White non-Hispanic voters	ANES	2004	18.9	9.7	28.0	51.5	32.7	283	230
White non-Hispanic voters	ANES	2008	28.8	20.9	36.7	55.4	26.6	410	284
White non-Hispanic voters	ANES	2012	29.3	24.4	34.1	53.9	24.7	1,460	874
White non-Hispanic voters	ANES	2016	27.8	22.6	33.1	53.5	25.7	1,107	652
White non-Hispanic voters	ANES	2020	29.9	26.0	33.9	55.4	25.4	3,035	1,292
White non-Hispanic voters	CES	2008	32.1	30.5	33.7	58.2	26.2	11,293	6,653
White non-Hispanic voters	CES	2012	33.0	31.3	34.7	57.9	24.9	18,416	9,387
White non-Hispanic voters	CES	2016	25.4	23.9	26.9	51.1	25.7	20,561	7,745
White non-Hispanic voters	CES	2020	31.0	29.6	32.3	54.2	23.2	23,149	7,406
White non-Hispanic voters	CES	2024	33.8	32.1	35.5	53.7	19.9	22,958	6,803

The ANES gap has an interval that includes zero in 1980, 1984 and 1988 and lies above zero from 1992 on, and the CES gap lies above zero in every year from 2008. Each line joins presidential years, each band is a 95% interval, and a value above zero means that seldom attenders voted more Democratic than frequent attenders. The other views restrict the sample by race or by tradition.Note: The ANES cumulative file's 2024 religion and attendance codes cannot be used, so the ANES line ends in 2020.Note: Attendance groups follow the ANES three-category item: frequent is weekly or almost weekly, seldom is a few times a year or never. The CES has no almost-weekly answer, so its frequent group is the top two answers, once a week and more than once a week.Note: A study enters only when each attendance group has at least 100 respondents and an effective sample of the same size. Intervals are 95% intervals on the Kish effective sample size and treat the weighted sample as a simple random sample.Source: ANES Cumulative Data File, 1952-2020, and Cooperative Election Study, 2008-2024; weighted

The ANES and the CES agree in the gap pooled over 2008, 2012, 2016 and 2020, which is 21.8 (95% CI 19.5 to 24.1) points in the ANES and 23.4 (95% CI 22.7 to 24.2) points in the CES.In single years the two surveys differ in both directions: the CES gap was 27.3 (95% CI 25.7 to 28.9) points in 2012 against 20.5 (95% CI 16.2 to 24.8) in the ANES, with intervals that do not overlap, and the ANES gap was 23.8 (95% CI 18.9 to 28.7) points in 2016 against 18.9 (95% CI 17.4 to 20.4) in the CES.

The CES gap did not move in one direction after 2008. It was 25.5 (95% CI 23.9 to 27.1) points in 2008 and 27.3 (95% CI 25.7 to 28.9) in 2012, fell to 18.9 (95% CI 17.4 to 20.4) in 2016, was 23.5 (95% CI 22.1 to 24.9) in 2020 and stood at 27.7 (95% CI 26.0 to 29.3) in 2024. The change from 2020 to 2024 came from the frequent end: the Democratic share among frequent attenders went from 36.5% (95% CI 35.3% to 37.7%) to 29.9% (95% CI 28.5% to 31.3%), while the share among seldom attenders went from 60.0% (95% CI 59.3% to 60.7%) to 57.6% (95% CI 56.7% to 58.5%).

The gap is larger among white voters and smaller among Protestants and Catholics

The gap for all voters mixes several groups, and the other two views of Figure 1 separate some of them. Among white non-Hispanic voters the 2024 CES gap was 33.8 (95% CI 32.1 to 35.5) points, so leaving out Black, Hispanic and other voters raises it. Among Protestant and Catholic voters it was 13.2 (95% CI 11.2 to 15.2) points in 2024, down from 18.7 (95% CI 16.6 to 20.7) in 2008.

The voters left out of the Protestant and Catholic view include the unaffiliated. In the 2024 CES the unaffiliated attended seldom in 96% (95% CI 95% to 96%) of cases and voted 70.9% (95% CI 69.7% to 72.2%) Democratic. They made up 34.2% (95% CI 33.2% to 35.2%) of seldom-attending voters in 2008 and 44.8% (95% CI 43.9% to 45.7%) in 2024, so a growing part of the seldom group is made up of voters who voted Democratic at that rate. See Chapter 9.2, The rise of the nones (planned) for the unaffiliated and Chapter 9.3, Religious tradition and race for white evangelicals, Black Protestants, Latino Catholics and other traditions.

The attendance gap has exceeded the denominational gap since 1992

The ANES records denomination from 1952 and attendance from 1970, so the earliest religious line in the vote that can be measured runs between Catholics and Protestants.In the ANES the Catholic minus Protestant gap in the Democratic vote was 46.2 (95% CI 38.7 to 53.7) points in 1960, the year John F. Kennedy, a Catholic, was elected, and 20.6 (95% CI 17.4 to 23.7) points over the studies of 1952 to 1968 pooled. It was 8.5 (95% CI 5.6 to 11.3) points in 1972 to 1988. The Catholic minus Protestant gap was 5.0 (95% CI 1.3 to 8.6) points in 1992 to 2004 and 8.3 (95% CI 5.6 to 11.1) in 2008 to 2020, and it was 10.4 (95% CI 8.3 to 12.4) in the 2024 CES (Figure 2). The pooled attendance gap was 13.0 (95% CI 9.7 to 16.2) points in 1992 to 2004 and 21.8 (95% CI 19.5 to 24.1) in 2008 to 2020, and its interval lies above the denominational interval in both windows. By study, the attendance gap was larger than the denominational gap in 8 of the 8 ANES studies from 1992 on, against 1 of 5 in 1972 to 1988, and the two intervals are separate in 3 of the 8 studies from 1992 on (those of 2012, 2016 and 2020). In the CES the attendance interval lies wholly above the denominational interval in 5 of 5 presidential years.The ANES cumulative file cannot supply 2024 religion or attendance. It codes 835 respondents as Protestant in 2024 against 3,180 in 2020, and 49% of 2024 respondents have an attendance answer against 99% in 2020. Every ANES series here stops in 2020, and the CES supplies 2024.

Figure 2

Since 1992 the attendance gap has been larger than the Catholic-Protestant gap, and the unaffiliated-Protestant gap is larger still

Democratic share of the two-party presidential vote among reported voters; gaps in percentage points and shares in percent; ANES 1952 to 2020 and CES 2008 to 2024; 95% intervals

Show data

Data for Since 1992 the attendance gap has been larger than the Catholic-Protestant gap, and the unaffiliated-Protestant gap is larger still
View	Tradition or sample	Series	Year	Estimate	Lower	Upper	Voters
gaps	All voters	ANES attendance	1972	10.3	5.3	15.3	1,389
gaps	All voters	ANES attendance	1976	7.5	1.5	13.5	1,114
gaps	All voters	ANES attendance	1980	0.1	−6.5	6.6	849
gaps	All voters	ANES attendance	1984	5.2	−0.3	10.8	1,201
gaps	All voters	ANES attendance	1988	2.1	−4.1	8.2	1,019
gaps	All voters	ANES attendance	1992	11.3	6.2	16.5	1,429
gaps	All voters	ANES attendance	1996	14.6	7.7	21.5	955
gaps	All voters	ANES attendance	2000	14.9	7.9	21.9	934
gaps	All voters	ANES attendance	2004	12.6	4.5	20.7	687
gaps	All voters	ANES attendance	2008	21.1	14.4	27.8	1,266
gaps	All voters	ANES attendance	2012	20.5	16.2	24.8	3,741
gaps	All voters	ANES attendance	2016	23.8	18.9	28.7	2,316
gaps	All voters	ANES attendance	2020	21.9	18.2	25.7	5,673
gaps	All voters	CES attendance	2008	25.5	23.9	27.1	21,168
gaps	All voters	CES attendance	2012	27.3	25.7	28.9	34,381
gaps	All voters	CES attendance	2016	18.9	17.4	20.4	37,373
gaps	All voters	CES attendance	2020	23.5	22.1	24.9	40,514
gaps	All voters	CES attendance	2024	27.7	26.0	29.3	39,057
gaps	Protestant and Catholic voters	ANES denomination	1952	15.3	8.7	22.0	1,104
gaps	Protestant and Catholic voters	ANES denomination	1956	9.9	3.4	16.4	1,181
gaps	Protestant and Catholic voters	ANES denomination	1960	46.2	38.7	53.7	847
gaps	Protestant and Catholic voters	ANES denomination	1964	16.2	10.2	22.1	1,035
gaps	Protestant and Catholic voters	ANES denomination	1968	21.8	14.6	29.0	963
gaps	Protestant and Catholic voters	ANES denomination	1972	8.2	2.7	13.7	1,469
gaps	Protestant and Catholic voters	ANES denomination	1976	11.5	5.1	17.8	1,219
gaps	Protestant and Catholic voters	ANES denomination	1980	1.3	−6.3	8.9	839
gaps	Protestant and Catholic voters	ANES denomination	1984	8.7	2.7	14.7	1,212
gaps	Protestant and Catholic voters	ANES denomination	1988	10.5	3.9	17.1	1,070
gaps	Protestant and Catholic voters	ANES denomination	1992	5.9	0.1	11.7	1,342
gaps	Protestant and Catholic voters	ANES denomination	1996	6.6	−1.0	14.1	912
gaps	Protestant and Catholic voters	ANES denomination	2000	1.4	−6.1	9.0	911
gaps	Protestant and Catholic voters	ANES denomination	2004	5.0	−4.2	14.3	610
gaps	Protestant and Catholic voters	ANES denomination	2008	9.2	1.2	17.1	1,195
gaps	Protestant and Catholic voters	ANES denomination	2012	9.5	4.4	14.7	2,927
gaps	Protestant and Catholic voters	ANES denomination	2016	8.6	2.8	14.4	1,799
gaps	Protestant and Catholic voters	ANES denomination	2020	7.3	2.8	11.8	3,745
gaps	Protestant and Catholic voters	CES denomination	2008	10.1	8.1	12.2	13,915
gaps	Protestant and Catholic voters	CES denomination	2012	7.9	6.1	9.8	25,252
gaps	Protestant and Catholic voters	CES denomination	2016	10.8	9.0	12.5	24,840
gaps	Protestant and Catholic voters	CES denomination	2020	10.7	9.0	12.5	23,634
gaps	Protestant and Catholic voters	CES denomination	2024	10.4	8.3	12.4	22,174
gaps	Unaffiliated and Protestant voters	CES unaffiliated	2008	36.7	34.7	38.6	13,283
gaps	Unaffiliated and Protestant voters	CES unaffiliated	2012	32.3	30.5	34.1	25,423
gaps	Unaffiliated and Protestant voters	CES unaffiliated	2016	30.1	28.5	31.6	27,239
gaps	Unaffiliated and Protestant voters	CES unaffiliated	2020	35.4	34.0	36.8	29,897
gaps	Unaffiliated and Protestant voters	CES unaffiliated	2024	38.0	36.3	39.7	28,512
within	Catholic	Frequent	2008	38.7	36.0	41.3	1,809
within	Catholic	Frequent	2012	37.5	35.0	39.9	3,002
within	Catholic	Frequent	2016	42.9	40.4	45.4	2,901
within	Catholic	Frequent	2020	42.3	39.7	44.9	2,434
within	Catholic	Frequent	2024	36.7	33.5	39.8	2,146
within	Catholic	Seldom	2008	56.8	54.5	59.1	2,705
within	Catholic	Seldom	2012	56.0	53.9	58.1	4,421
within	Catholic	Seldom	2016	51.4	49.5	53.2	5,220
within	Catholic	Seldom	2020	51.6	49.7	53.5	5,232
within	Catholic	Seldom	2024	47.1	44.9	49.2	5,074
within	Protestant	Frequent	2008	30.8	29.1	32.4	3,977
within	Protestant	Frequent	2012	34.5	32.9	36.1	7,667
within	Protestant	Frequent	2016	33.6	32.0	35.1	6,556
within	Protestant	Frequent	2020	32.3	30.8	33.7	6,240
within	Protestant	Frequent	2024	26.5	24.8	28.1	5,613
within	Protestant	Seldom	2008	48.0	46.1	49.9	3,710
within	Protestant	Seldom	2012	47.2	45.6	48.8	7,412
within	Protestant	Seldom	2016	41.4	39.9	43.0	7,416
within	Protestant	Seldom	2020	41.8	40.4	43.3	7,199
within	Protestant	Seldom	2024	38.9	37.2	40.6	7,147
within	Unaffiliated	Frequent	2012	54.5	46.0	62.9	266
within	Unaffiliated	Frequent	2016	53.5	46.8	60.3	386
within	Unaffiliated	Frequent	2020	59.5	52.9	66.2	369
within	Unaffiliated	Seldom	2008	76.6	75.0	78.1	4,351
within	Unaffiliated	Seldom	2012	74.7	73.2	76.2	7,904
within	Unaffiliated	Seldom	2016	69.3	68.1	70.4	10,803
within	Unaffiliated	Seldom	2020	73.8	72.9	74.8	14,096
within	Unaffiliated	Seldom	2024	72.1	70.9	73.4	13,762

The attendance gap (seldom minus frequent attenders) is larger than the denomination gap (Catholics minus Protestants) in 8 of 8 ANES studies from 1992 to 2020, with separate intervals in 3, and in all 5 CES presidential years, and the CES gap between the unaffiliated and Protestants is larger than both. Each point is a study or survey year and each band a 95% interval. Solid lines are the ANES and dashed lines the CES.Note: The ANES cumulative file's 2024 religion codes cannot be used: Protestants number 3,180 in 2020 and 835 in 2024 while Catholics are stable, so the ANES lines end in 2020. The CES gives a Catholic-Protestant gap of 10.4 points in 2024. The ANES religion item cannot separate the unaffiliated from other faiths, so the unaffiliated line is CES only.Note: A group needs at least 100 respondents and an effective sample of the same size in a study. Intervals are 95% intervals on the Kish effective sample size and treat the weighted sample as a simple random sample.Source: ANES Cumulative Data File, 1952-2020, and Cooperative Election Study, 2008-2024; weighted

The two gaps differ in who is compared. The attendance gap counts the unaffiliated in its seldom group, and the Catholic minus Protestant gap leaves them out. In the CES the unaffiliated voted 36.7 (95% CI 34.7 to 38.6) points more Democratic than Protestants in 2008 and 38.0 (95% CI 36.3 to 39.7) points more in 2024, against attendance gaps of 25.5 and 27.7 points, and the unaffiliated-Protestant interval lies above both the attendance interval and the Catholic-Protestant interval in every presidential year. Among Protestant and Catholic voters alone, the two groups that the denominational gap compares, the attendance gap was 13.2 (95% CI 11.2 to 15.2) points in 2024 (the third view of Figure 1) against 10.4 (95% CI 8.3 to 12.4) points for denomination, and the two intervals overlap. Across the 5 CES presidential years the attendance interval among these voters lies wholly above the denominational interval in 2 of them (2008 and 2012), and across the 8 ANES studies from 1992 on in 0. The comparison of the two gaps therefore depends on which voters are included.

The vote differs by attendance inside each tradition, by less than in the pooled comparison. In the 2024 CES seldom Protestants voted 12.5 (95% CI 10.1 to 14.8) points more Democratic than frequent Protestants, and seldom Catholics 10.4 (95% CI 6.6 to 14.2) points more than frequent Catholics, against 27.7 points among all voters. In 2008 the two gaps were 17.2 and 18.1 points. Denomination also separates voters at each level of attendance. In 2024 Catholics who attend often voted 10.2 (95% CI 6.6 to 13.8) points more Democratic than Protestants who attend often, and Catholics who attend seldom voted 8.1 (95% CI 5.4 to 10.9) points more than Protestants who attend seldom. Among the unaffiliated, those who attended weekly or more in 2020 voted 14.3 (95% CI 7.5 to 21.0) points less Democratic than the seldom attenders. The pooled attendance gap therefore combines a gap inside each tradition with the contrast between the unaffiliated and everyone else.

Figure 3 plots the Republican share of Democratic and Republican identifiers by attendance. In the GSS the share among frequent attenders was 37.4% (95% CI 35.5% to 39.2%) in 1972 to 1978 and 53.2% (95% CI 50.9% to 55.5%) in 2010 to 2018, the last in-person surveys, a change of 15.9 (95% CI 12.9 to 18.8) points. Among seldom attenders it went from 32.1% (95% CI 30.5% to 33.7%) to 37.8% (95% CI 36.2% to 39.5%), a change of 5.7 (95% CI 3.4 to 8.0) points. The gap between the two groups went from 5.3 (95% CI 2.9 to 7.7) to 15.4 (95% CI 12.6 to 18.2) points, so the Republican share rose in both groups and rose more among frequent attenders. The GSS has included web interviews since 2021, and the surveys from 2021 on are called the web era below. The surveys of 2018 to 2024, which include web interviews, give a larger gap of 19.5 (95% CI 16.3 to 22.8) points, and in 2022 and 2024 the gap was 18.0 (95% CI 10.3 to 25.8) points among in-person respondents and 23.7 (95% CI 16.8 to 30.5) among web respondents, so part of the larger gap in that window is not separable from the interview mode.In the surveys of 2018 to 2024 the Republican share was 59.7% (95% CI 57.0% to 62.5%) among frequent attenders and 40.2% (95% CI 38.5% to 41.9%) among seldom attenders, changes from 1972 to 1978 of 22.4 (95% CI 19.1 to 25.7) and 8.1 (95% CI 5.8 to 10.4) points.

Figure 3

The Republican share of party identifiers rose more among frequent attenders than among seldom attenders in the GSS

Republican share of Democratic and Republican identifiers (leaners included) among frequent attenders (weekly or nearly weekly, and weekly or more in the CES) and seldom attenders (a few times a year or never), by year, percent, 95% intervals

Show data

Data for The Republican share of party identifiers rose more among frequent attenders than among seldom attenders in the GSS
Survey	Attendance	Year	Republican share (%)	Lower	Upper	Respondents
CES	Frequent	2008	61.5	60.3	62.7	9,264
CES	Frequent	2009	62.6	60.7	64.5	3,447
CES	Frequent	2010	61.7	60.5	63.0	15,341
CES	Frequent	2011	58.2	56.2	60.2	4,711
CES	Frequent	2012	59.6	58.3	60.8	13,849
CES	Frequent	2013	57.2	54.8	59.6	3,336
CES	Frequent	2014	61.4	60.1	62.7	12,466
CES	Frequent	2015	57.0	54.6	59.4	3,019
CES	Frequent	2016	56.5	55.4	57.7	13,736
CES	Frequent	2017	59.3	57.0	61.6	3,608
CES	Frequent	2018	65.1	63.9	66.3	12,055
CES	Frequent	2019	63.9	61.9	65.9	3,690
CES	Frequent	2020	63.8	62.6	65.0	11,385
CES	Frequent	2021	63.5	61.7	65.3	4,609
CES	Frequent	2022	64.5	63.2	65.8	11,125
CES	Frequent	2023	58.2	56.4	60.0	5,168
CES	Frequent	2024	68.1	66.7	69.5	10,949
CES	Seldom	2008	35.0	34.1	35.9	15,164
CES	Seldom	2009	37.6	36.3	39.0	6,317
CES	Seldom	2010	40.8	39.8	41.7	26,251
CES	Seldom	2011	38.8	37.4	40.2	9,845
CES	Seldom	2012	33.8	32.9	34.7	25,858
CES	Seldom	2013	37.2	35.7	38.8	7,801
CES	Seldom	2014	40.1	39.2	40.9	26,501
CES	Seldom	2015	39.9	38.3	41.4	7,057
CES	Seldom	2016	39.1	38.4	39.9	31,636
CES	Seldom	2017	39.2	37.7	40.7	8,894
CES	Seldom	2018	38.7	38.0	39.5	31,182
CES	Seldom	2019	37.6	36.3	38.9	9,363
CES	Seldom	2020	40.3	39.6	41.1	32,768
CES	Seldom	2021	38.4	37.3	39.5	13,718
CES	Seldom	2022	40.0	39.2	40.7	32,774
CES	Seldom	2023	40.1	39.0	41.2	12,096
CES	Seldom	2024	40.7	39.8	41.6	34,618
GSS	Frequent	1972	33.7	29.5	37.9	575
GSS	Frequent	1973	43.0	38.1	47.8	468
GSS	Frequent	1974	36.9	32.1	41.8	473
GSS	Frequent	1975	38.9	34.0	43.9	467
GSS	Frequent	1976	33.3	28.5	38.0	452
GSS	Frequent	1977	35.3	30.5	40.0	482
GSS	Frequent	1978	41.4	35.7	47.1	458
GSS	Frequent	1980	38.6	33.1	44.2	438
GSS	Frequent	1982	39.7	34.3	45.0	579
GSS	Frequent	1983	39.9	34.3	45.4	513
GSS	Frequent	1984	47.8	42.2	53.5	489
GSS	Frequent	1985	45.4	39.8	50.9	507
GSS	Frequent	1986	47.1	42.1	52.1	466
GSS	Frequent	1987	46.3	41.1	51.5	549
GSS	Frequent	1988	51.6	46.5	56.7	444
GSS	Frequent	1989	50.1	45.0	55.2	476
GSS	Frequent	1990	52.8	47.5	58.1	429
GSS	Frequent	1991	50.9	45.7	56.0	458
GSS	Frequent	1993	53.5	48.5	58.6	487
GSS	Frequent	1994	51.1	47.2	55.0	810
GSS	Frequent	1996	50.0	46.0	53.9	748
GSS	Frequent	1998	51.2	47.3	55.2	763
GSS	Frequent	2000	53.9	49.8	58.1	688
GSS	Frequent	2002	54.1	49.9	58.4	705
GSS	Frequent	2004	53.7	49.5	57.8	786
GSS	Frequent	2006	55.3	51.9	58.6	1,165
GSS	Frequent	2008	53.0	47.9	58.1	527
GSS	Frequent	2010	50.7	45.2	56.2	507
GSS	Frequent	2012	50.8	45.2	56.3	496
GSS	Frequent	2014	54.6	50.0	59.3	587
GSS	Frequent	2016	52.2	47.7	56.7	681
GSS	Frequent	2018	57.9	52.1	63.6	489
GSS	Frequent	2021	54.6	50.3	59.0	813
GSS	Frequent	2022	60.7	54.2	67.3	606
GSS	Frequent	2024	66.2	60.9	71.5	623
GSS	Seldom	1972	31.4	27.2	35.6	579
GSS	Seldom	1973	34.3	30.3	38.4	625
GSS	Seldom	1974	29.4	25.3	33.4	571
GSS	Seldom	1975	30.7	26.6	34.7	587
GSS	Seldom	1976	30.8	26.8	34.9	620
GSS	Seldom	1977	32.8	28.7	36.9	631
GSS	Seldom	1978	34.7	30.0	39.4	625
GSS	Seldom	1980	33.2	28.5	37.9	587
GSS	Seldom	1982	38.1	33.7	42.5	753
GSS	Seldom	1983	37.9	33.2	42.7	639
GSS	Seldom	1984	40.5	35.4	45.6	576
GSS	Seldom	1985	44.4	39.7	49.1	650
GSS	Seldom	1986	43.1	38.5	47.6	560
GSS	Seldom	1987	43.5	39.1	47.9	706
GSS	Seldom	1988	40.2	35.8	44.6	612
GSS	Seldom	1989	46.3	42.0	50.7	643
GSS	Seldom	1990	44.7	39.9	49.5	528
GSS	Seldom	1991	49.5	45.1	54.0	596
GSS	Seldom	1993	43.4	39.1	47.8	630
GSS	Seldom	1994	41.8	38.8	44.9	1,250
GSS	Seldom	1996	44.9	41.8	48.1	1,187
GSS	Seldom	1998	41.0	37.8	44.3	1,108
GSS	Seldom	2000	40.4	37.2	43.7	1,110
GSS	Seldom	2002	40.9	37.5	44.4	1,083
GSS	Seldom	2004	42.7	39.3	46.0	1,140
GSS	Seldom	2006	39.3	36.6	42.0	1,733
GSS	Seldom	2008	35.3	31.5	39.1	857
GSS	Seldom	2010	37.1	33.0	41.1	842
GSS	Seldom	2012	36.3	32.2	40.5	788
GSS	Seldom	2014	37.1	33.7	40.5	1,069
GSS	Seldom	2016	36.2	32.9	39.5	1,240
GSS	Seldom	2018	42.3	38.4	46.1	1,055
GSS	Seldom	2021	35.0	32.4	37.6	1,977
GSS	Seldom	2022	43.6	39.9	47.3	1,690
GSS	Seldom	2024	42.0	38.5	45.5	1,452

The two groups start close together in the GSS, and the Republican share rises further among frequent attenders. Each point is one survey year, each band a 95% interval and the dashed rule an even split. The shaded band marks the GSS years from 2021 on, which include web interviews.Note: Pure independents are left out. GSS 2021 to 2024 include web interviews, so the 2018 to 2024 window is not comparable with the in-person window of 2010 to 2018. In 2022 and 2024 pooled the frequent-minus-seldom gap was 18.0 points among in-person respondents and 23.7 points among web respondents.Note: A cell needs at least 100 respondents and an effective sample of the same size. Intervals are 95% intervals on the Kish effective sample size and treat the weighted sample as a simple random sample.Source: General Social Survey, 1972-2024, and Cooperative Election Study, 2008-2024; weighted

The CES begins in 2008 and shows a different pattern. Its gap was 26.5 (95% CI 25.0 to 28.0) points in 2008 and 27.4 (95% CI 25.7 to 29.1) points in 2024. Frequent attenders were 61.5% (95% CI 60.3% to 62.7%) Republican in 2008 and 68.1% (95% CI 66.7% to 69.5%) in 2024, and seldom attenders 35.0% (95% CI 34.1% to 35.9%) and 40.7% (95% CI 39.8% to 41.6%), so the Republican share rose by similar amounts in the two groups. The CES gap of 2008 already exceeds the GSS gap of 15.4 points in 2010 to 2018 and of 19.5 points in 2018 to 2024, and the two surveys differ in sample, mode and party items, so their levels are not comparable. The CES cannot say what happened before 2008.

Party identification accounts for part of the rise in the slope

In the GSS the raw slope of ideology on attendance was 0.15 (95% CI 0.09 to 0.20) in 1974, 0.26 (95% CI 0.21 to 0.31) in 2008, 0.25 (95% CI 0.20 to 0.31) in 2018 and 0.32 (95% CI 0.26 to 0.37) in 2024 (Figure 4). Each slope is the standardized slope of self-placed ideology on attendance in one survey year, with both variables standardized once on all respondents in the sample, so it counts the standard deviations of ideology that go with one standard deviation more attendance.

Net of party identification, with leaners counted with their party and the fit made on the same respondents, the slope was 0.14 (95% CI 0.08 to 0.19) in 1974, 0.19 (95% CI 0.14 to 0.24) in 2018 and 0.20 (95% CI 0.15 to 0.25) in 2024. The 2024 point falls in the web era of the GSS. In 2022 and 2024 pooled, in-person respondents give a slope of 0.23 (95% CI 0.17 to 0.30) and web respondents give 0.37 (95% CI 0.31 to 0.42), a difference of 0.13 (95% CI 0.05 to 0.22), so the 2024 slope differs by mode of interview.

The CES, which is online throughout, does not show a steady rise. Its slope was 0.35 (95% CI 0.34 to 0.36) in 2008 and fell to 0.25 (95% CI 0.24 to 0.26) in 2010, and it was 0.35 (95% CI 0.34 to 0.37) in 2024. Pooled, it was 0.26 (95% CI 0.26 to 0.27) over 2010 to 2017 and 0.35 (95% CI 0.34 to 0.35) over 2018 to 2024. The step between 2017 and 2018 coincides with a rise in the spread of the five-point ideology item, whose standard deviation was 1.00 in 2010 to 2017 and 1.21 in 2018 to 2024. The slope is standardized on all years together, so part of the step is a change in that spread and not in the tie with attendance. The correlation of attendance with ideology, which does not depend on the spread, was +0.29 (95% CI 0.28 to 0.29) over 2010 to 2017 and +0.32 (95% CI 0.31 to 0.32) over 2018 to 2024.

Figure 4

The slope of ideology on attendance rose in the GSS from the 1970s to the 2000s, and party accounts for part of the rise

Standardized slope of ideology on church attendance, by year, raw and net of party identification; 95% intervals

Show data

Data for The slope of ideology on attendance rose in the GSS from the 1970s to the 2000s, and party accounts for part of the rise
Survey	Series	Year	Slope	Lower	Upper	Respondents
CES	Net of party	2008	0.18	0.17	0.20	29,793
CES	Net of party	2009	0.17	0.16	0.19	12,123
CES	Net of party	2010	0.14	0.13	0.15	51,265
CES	Net of party	2011	0.15	0.13	0.16	18,198
CES	Net of party	2012	0.16	0.15	0.17	48,517
CES	Net of party	2013	0.16	0.14	0.17	14,509
CES	Net of party	2014	0.16	0.15	0.17	49,407
CES	Net of party	2015	0.17	0.15	0.19	12,719
CES	Net of party	2016	0.18	0.17	0.19	58,315
CES	Net of party	2017	0.16	0.15	0.18	16,175
CES	Net of party	2018	0.19	0.18	0.20	53,579
CES	Net of party	2019	0.16	0.15	0.18	16,287
CES	Net of party	2020	0.18	0.17	0.19	54,697
CES	Net of party	2021	0.19	0.17	0.20	23,134
CES	Net of party	2022	0.20	0.18	0.21	54,246
CES	Net of party	2023	0.18	0.17	0.20	21,693
CES	Net of party	2024	0.17	0.16	0.18	55,195
CES	Raw slope	2008	0.35	0.34	0.36	29,793
CES	Raw slope	2009	0.31	0.29	0.33	12,123
CES	Raw slope	2010	0.25	0.24	0.26	51,265
CES	Raw slope	2011	0.24	0.23	0.26	18,198
CES	Raw slope	2012	0.28	0.27	0.29	48,517
CES	Raw slope	2013	0.25	0.23	0.27	14,509
CES	Raw slope	2014	0.27	0.26	0.28	49,407
CES	Raw slope	2015	0.26	0.24	0.28	12,719
CES	Raw slope	2016	0.26	0.25	0.27	58,315
CES	Raw slope	2017	0.27	0.25	0.29	16,175
CES	Raw slope	2018	0.36	0.35	0.38	53,579
CES	Raw slope	2019	0.35	0.33	0.37	16,287
CES	Raw slope	2020	0.34	0.33	0.35	54,697
CES	Raw slope	2021	0.35	0.33	0.37	23,134
CES	Raw slope	2022	0.36	0.34	0.37	54,246
CES	Raw slope	2023	0.30	0.28	0.32	21,693
CES	Raw slope	2024	0.35	0.34	0.37	55,195
GSS	Net of party	1974	0.14	0.08	0.19	1,344
GSS	Net of party	1975	0.15	0.09	0.21	1,382
GSS	Net of party	1976	0.12	0.07	0.18	1,387
GSS	Net of party	1977	0.10	0.04	0.16	1,435
GSS	Net of party	1978	0.13	0.08	0.19	1,420
GSS	Net of party	1980	0.11	0.04	0.18	1,414
GSS	Net of party	1982	0.16	0.10	0.21	1,718
GSS	Net of party	1983	−0.02	−0.10	0.06	761
GSS	Net of party	1984	0.18	0.12	0.23	1,380
GSS	Net of party	1985	0.11	0.05	0.17	1,443
GSS	Net of party	1986	0.13	0.07	0.18	1,387
GSS	Net of party	1987	0.12	0.07	0.18	1,658
GSS	Net of party	1988	0.12	0.06	0.18	1,408
GSS	Net of party	1989	0.12	0.06	0.17	1,428
GSS	Net of party	1990	0.15	0.09	0.21	1,267
GSS	Net of party	1991	0.16	0.10	0.21	1,421
GSS	Net of party	1993	0.20	0.15	0.26	1,496
GSS	Net of party	1994	0.18	0.14	0.22	2,765
GSS	Net of party	1996	0.17	0.13	0.21	2,630
GSS	Net of party	1998	0.19	0.15	0.23	2,585
GSS	Net of party	2000	0.18	0.13	0.22	2,527
GSS	Net of party	2002	0.13	0.07	0.19	1,294
GSS	Net of party	2004	0.14	0.08	0.20	1,287
GSS	Net of party	2006	0.16	0.13	0.19	4,244
GSS	Net of party	2008	0.19	0.14	0.24	1,884
GSS	Net of party	2010	0.15	0.10	0.21	1,911
GSS	Net of party	2012	0.19	0.13	0.24	1,809
GSS	Net of party	2014	0.21	0.17	0.25	2,360
GSS	Net of party	2016	0.19	0.15	0.23	2,662
GSS	Net of party	2018	0.19	0.14	0.24	2,139
GSS	Net of party	2021	0.16	0.12	0.19	3,793
GSS	Net of party	2022	0.18	0.14	0.23	3,276
GSS	Net of party	2024	0.20	0.15	0.25	3,039
GSS	Raw slope	1974	0.15	0.09	0.20	1,344
GSS	Raw slope	1975	0.16	0.10	0.21	1,382
GSS	Raw slope	1976	0.14	0.08	0.19	1,387
GSS	Raw slope	1977	0.10	0.04	0.16	1,435
GSS	Raw slope	1978	0.15	0.09	0.21	1,420
GSS	Raw slope	1980	0.12	0.05	0.19	1,414
GSS	Raw slope	1982	0.17	0.11	0.23	1,718
GSS	Raw slope	1983	−0.02	−0.11	0.06	761
GSS	Raw slope	1984	0.19	0.13	0.24	1,380
GSS	Raw slope	1985	0.12	0.05	0.18	1,443
GSS	Raw slope	1986	0.14	0.08	0.19	1,387
GSS	Raw slope	1987	0.13	0.07	0.19	1,658
GSS	Raw slope	1988	0.15	0.09	0.21	1,408
GSS	Raw slope	1989	0.13	0.07	0.19	1,428
GSS	Raw slope	1990	0.17	0.11	0.23	1,267
GSS	Raw slope	1991	0.16	0.10	0.22	1,421
GSS	Raw slope	1993	0.23	0.18	0.29	1,496
GSS	Raw slope	1994	0.20	0.16	0.25	2,765
GSS	Raw slope	1996	0.20	0.15	0.24	2,630
GSS	Raw slope	1998	0.23	0.19	0.27	2,585
GSS	Raw slope	2000	0.22	0.17	0.26	2,527
GSS	Raw slope	2002	0.18	0.11	0.24	1,294
GSS	Raw slope	2004	0.17	0.10	0.24	1,287
GSS	Raw slope	2006	0.22	0.19	0.26	4,244
GSS	Raw slope	2008	0.26	0.21	0.31	1,884
GSS	Raw slope	2010	0.21	0.15	0.27	1,911
GSS	Raw slope	2012	0.24	0.18	0.30	1,809
GSS	Raw slope	2014	0.27	0.23	0.32	2,360
GSS	Raw slope	2016	0.26	0.21	0.30	2,662
GSS	Raw slope	2018	0.25	0.20	0.31	2,139
GSS	Raw slope	2021	0.26	0.22	0.30	3,793
GSS	Raw slope	2022	0.29	0.23	0.35	3,276
GSS	Raw slope	2024	0.32	0.26	0.37	3,039

In the GSS the raw slope rose from 0.15 in 1974 to 0.26 in 2008 and was 0.25 in 2018 and 0.32 in 2024, and the slope net of party stayed lower, at 0.14 in 1974 and 0.19 in 2018. Each point is one survey year. Both fits use the same respondents, and both variables are standardized once on all years together, so the years share one scale. The shaded band marks 2021 to 2024, when web interviews entered the GSS.Note: Ideology is the seven-point liberal-conservative scale in the GSS and a five-point scale in the CES.Note: GSS interviews were in person, with some telephone interviews, through 2018; 2021 was mostly web, and 2022 and 2024 mixed in-person and web interviews. In 2022 and 2024 pooled, in-person respondents give a slope of 0.23 and web respondents 0.37.Note: The GSS 1983 survey asked the ideology item of a subsample only, and that year's slope, marked on the chart, rests on fewer respondents. Years with fewer than 200 respondents are left out. Intervals are 95% intervals on the Kish effective sample size and treat the weighted sample as a simple random sample. Regression intervals use HC3 robust errors.Source: General Social Survey, 1974-2024, and Cooperative Election Study, 2008-2024; weighted

To ask how much of the rise is accounted for by party, the chapter fits each slope on one set of respondents, once raw and once with party identification added, and compares an early and a late window. The main comparison pools the in-person GSS surveys of 1974 to 1978 and of 2010 to 2018. The raw slope rose by 0.11 (95% CI 0.08 to 0.15) between the two windows and the slope net of party by 0.06 (95% CI 0.03 to 0.10), so party identification accounts for 46% (95% CI 33% to 67%) of the rise.The share is one minus the ratio of the two rises. Its interval comes from a bootstrap over respondents, which recomputes both rises on 2,000 resamples of the respondents drawn with replacement and reads the interval from the spread of the results. Four other specifications give shares from 50% to 64% (Figure 5). Every specification puts the rise net of party below the raw rise, and the single-year comparisons have intervals wide enough to include the whole rise.

Figure 5

Party identification accounts for 46% of the rise in the ideology slope (95% interval 33% to 67%) and for 50% to 64% in four other specifications

Early and late GSS periods compared on one common sample of respondents, by specification; 95% bootstrap intervals

The share of the rise that party identification accounts for runs from 46% in the main specification to 64% when 1974 is compared with 2024

Share of the rise in the raw slope that the party control accounts for, in percent

Main specification
Other specifications

The intervals of the three pooled comparisons lie below the whole rise and those of the two single-year comparisons reach past it. A bound above the whole rise comes from resamples in which the rise net of party is negative.

Show data

Data for Party identification accounts for 46% of the rise in the ideology slope (95% interval 33% to 67%) and for 50% to 64% in four other specifications, Share accounted for
Specification	Periods compared	Share accounted for by party (%)	Lower	Upper	Respondents
Pooled in-person windows	1974 to 1978 and 2010 to 2018, pooled	46	33	67	17,849
Seven party categories	1974 to 1978 and 2010 to 2018, seven party categories	52	37	74	17,849
1974 and 2018 alone	1974 and 2018 alone (2018 is the last survey before the web mode)	50	20	163	3,483
1974 and 2024 alone	1974 and 2024 alone (2024 includes web interviews)	64	41	112	4,383
Late window with web	1974 to 1978 and 2018 to 2024, pooled (includes web interviews)	63	49	82	19,215

Each row compares an early and a late period on the same respondents, and the share is one minus the ratio of the rise net of party to the raw rise. With seven party categories the share is 52%, and it is 63% when the late window is 2018 to 2024, which includes web interviews.Note: The main specification pools the in-person GSS surveys of 1974 to 1978 and 2010 to 2018, and the seven-category row repeats that comparison with a finer party control. The single-year rows compare 1974 with 2018, the last survey before the web mode, and with 2024. The late window with web pools 2018 to 2024 against 1974 to 1978.Note: The percentile bootstrap resamples respondents within each period 2,000 times and refits both slopes each time. The party control is two indicators (Republican, independent) unless the row says seven categories.Note: The web mode began in 2021. Rows that end in 2018 or use the survey years 2010 to 2018 contain no web interviews.Source: General Social Survey, 1972-2024; weighted

The share is an accounting of the slope. Party identification and ideology are answered by the same respondents in the same interview, and the tie between them tightened over these years (Chapter 1.1, Party and ideology), so a party control can be expected to remove more of the slope in later years. In the GSS it removes 8% (95% CI 0% to 18%) of the raw slope in 1974 and 38% (95% CI 29% to 48%) in 2024, and in the CES 47% (95% CI 45% to 50%) in 2008 and 53% (95% CI 50% to 55%) in 2024. Panel studies find that religious behavior and identity respond to politics. Partisan identity that forms early can shape how often people attend and whether they identify with a religion [4], and religious and secular orientations respond to political orientations [5]. Cross-sections cannot separate the two directions, so the share is not an estimate of how much of the slope party causes or how much religion does.

Attendance correlates more closely with cultural than with economic attitudes

The GSS asks questions on sexual morality and abortion (homosexuality, premarital sex, extramarital sex, a pornography law and abortion for any reason) and on economic policy (help for the poor, help for the sick and government action on income differences). Averaging each set gives a cultural index and an economic index. The economic items are asked from 1975, so the comparison starts there, and over 1975 to 2024 the correlation of attendance with the cultural index is +0.44 (95% CI 0.43 to 0.45), with the economic index +0.11 (95% CI 0.09 to 0.12) and with self-placed ideology +0.22 (95% CI 0.21 to 0.23). All three use the same survey years. The cultural and economic correlations use the same respondents, and the ideology correlation uses every respondent who answered the ideology item in those years. The cultural correlation was +0.37 (95% CI 0.34 to 0.40) in 1975 to 1984 and +0.46 (95% CI 0.44 to 0.49) in 2021 to 2024, and the economic correlation went from +0.07 (95% CI 0.04 to 0.11) to +0.18 (95% CI 0.15 to 0.22) (Figure 6). The first period rests on the surveys of 1975, 1983 and 1984, the only ones in it that asked the economic items.

Figure 6

Attendance correlates more closely with cultural than with economic conservatism, and the economic correlation has grown

Weighted correlation of church attendance (never to more than weekly) with cultural conservatism, economic conservatism and self-placed ideology, by period, GSS respondents, 95% intervals

Show data

Data for Attendance correlates more closely with cultural than with economic conservatism, and the economic correlation has grown
Outcome	Period	Correlation	Lower	Upper	Respondents
Cultural conservatism	1975 to 1984	0.37	0.34	0.40	4,196
Cultural conservatism	1985 to 1994	0.39	0.37	0.41	9,643
Cultural conservatism	1995 to 2004	0.42	0.40	0.45	6,963
Cultural conservatism	2005 to 2014	0.44	0.42	0.47	7,362
Cultural conservatism	2015 to 2018	0.47	0.44	0.51	3,317
Cultural conservatism	2021 to 2024	0.46	0.44	0.49	6,846
Economic conservatism	1975 to 1984	0.07	0.04	0.11	4,196
Economic conservatism	1985 to 1994	0.06	0.04	0.08	9,643
Economic conservatism	1995 to 2004	0.07	0.05	0.10	6,963
Economic conservatism	2005 to 2014	0.09	0.06	0.11	7,362
Economic conservatism	2015 to 2018	0.12	0.08	0.17	3,317
Economic conservatism	2021 to 2024	0.18	0.15	0.22	6,846
Self-placed ideology	1975 to 1984	0.14	0.10	0.17	3,565
Self-placed ideology	1985 to 1994	0.17	0.15	0.19	12,990
Self-placed ideology	1995 to 2004	0.21	0.19	0.23	10,527
Self-placed ideology	2005 to 2014	0.24	0.22	0.26	12,522
Self-placed ideology	2015 to 2018	0.24	0.21	0.27	4,983
Self-placed ideology	2021 to 2024	0.27	0.24	0.29	10,448

The cultural line lies above the economic line in every period, and the economic line is higher in the last period than in the first. Each point is the correlation in the survey years of one period, placed at the period's midpoint, with a 95% interval band, and a higher value means that more frequent attenders were more conservative. The shaded band marks the period with web interviews.Note: The first period rests on the surveys of 1975, 1983 and 1984, the only ones in it that asked the economic items. All three outcomes use the survey years in which both indices were measured; the ideology points include every respondent who answered the ideology item in those years. The periods are 1975 to 1984, 1985 to 1994, 1995 to 2004, 2005 to 2014, 2015 to 2018 and 2021 to 2024.Note: Cultural conservatism averages five standardized items (homosexuality, premarital sex, extramarital sex, pornography law, abortion for any reason); economic conservatism averages three (help for the poor, help for the sick, government action on income differences). Each index needs at least two answers. The mean correlation among the cultural items is +0.32 and among the economic items +0.42.Note: GSS 2021 to 2024 include web interviews, and 2015 to 2018 is the last period before them. Correlations use the Kish effective sample size for the interval.Source: General Social Survey, 1972-2024; weighted

The two indices have similar internal consistency, meaning that their items agree with one another to a similar degree (standardized alpha of 0.70 for the cultural items and 0.69 for the economic ones), which limits how far unequal measurement error can explain the difference in correlations. The cultural items ask about sexual morality and abortion, topics on which many religious traditions take positions, so part of the tie sits close to the content of religious teaching. For the abortion item by party see Chapter 4.1, The abortion sort, for whether the cultural items form one dimension see Chapter 9.5, Two moralities (planned), and for the economic and cultural dimensions across the whole ideology scale see Chapter 7.2, Economic and cultural dimensions.

The United States sits above the European and world averages

The World Values Survey and the European Values Study (WVS/EVS) ask how important God is in a respondent's life and ask for a left-right self-placement. Inside each country the two answers have a correlation, and in the United States it is +0.28 (95% CI 0.25 to 0.31). The mean across the 104 countries and territories with a usable estimate, counting each once, is +0.10 (95% CI 0.09 to 0.12). It is +0.13 (95% CI 0.10 to 0.16) across 42 European countries and +0.08 (95% CI 0.06 to 0.10) across 61 countries outside Europe and the United States. 4 countries have a higher point estimate than the United States (Croatia, Greece, Slovenia and Spain), and in 2 of them (Croatia and Spain) the interval lies wholly above the United States interval (Figure 7).The cleaned WVS/EVS file holds 118 countries and territories, and 112 were surveyed in 2005 or later. It leaves the country name blank for ten EVS countries: Austria, Belgium, Denmark, Iceland, Ireland, Luxembourg, Malta, Portugal, Kosovo and Northern Cyprus. This chapter restores them from the country code, and 9 have a usable estimate. Without them the mean across 95 countries and territories is +0.10 (95% CI 0.08 to 0.12).

Figure 7

On both measures the United States correlation between religiosity and the right lies above the European and world averages

Weighted correlation between left-right self-placement and the importance of God or attendance, inside each country; the United States on two measures and equal-country means for Europe and the rest of the world; 95% intervals

United States
Europe
Rest of the world

Show data

Data for On both measures the United States correlation between religiosity and the right lies above the European and world averages
View	Row	Drawn in the chart	Correlation	Lower	Upper	Respondents	Countries
summary	US, GSS attendance	yes	0.25	0.23	0.26	27,953	n/a
summary	Europe, ESS attendance (36 countries)	yes	0.11	0.08	0.14	n/a	36
summary	US, WVS/EVS importance of God	yes	0.28	0.25	0.31	5,858	n/a
summary	Europe, WVS/EVS importance of God (42 countries)	yes	0.13	0.10	0.16	n/a	42
summary	Rest of the world, WVS/EVS importance of God (61 countries)	yes	0.08	0.06	0.10	n/a	61
summary	All WVS/EVS countries, importance of God (104 countries)	yes	0.10	0.09	0.12	n/a	104
countries	Albania	no	0.13	0.09	0.18	2,209	n/a
countries	Algeria	no	−0.02	−0.11	0.06	536	n/a
countries	Andorra	yes	0.22	0.17	0.26	1,813	n/a
countries	Argentina	yes	0.22	0.18	0.26	2,326	n/a
countries	Armenia	no	0.04	0.00	0.08	3,758	n/a
countries	Australia	no	0.20	0.16	0.23	4,456	n/a
countries	Austria	no	0.13	0.09	0.17	2,710	n/a
countries	Azerbaijan	no	0.04	0.00	0.09	2,152	n/a
countries	Bangladesh	no	0.01	−0.05	0.06	1,200	n/a
countries	Belarus	no	0.07	0.04	0.11	3,186	n/a
countries	Belgium	no	0.12	0.07	0.17	1,421	n/a
countries	Bolivia	no	0.05	0.01	0.10	1,803	n/a
countries	Bosnia Herzegovina	no	0.19	0.15	0.23	2,269	n/a
countries	Brazil	no	0.09	0.06	0.12	3,673	n/a
countries	Bulgaria	no	0.05	0.01	0.09	2,670	n/a
countries	Burkina Faso	no	−0.04	−0.09	0.02	1,109	n/a
countries	Canada	yes	0.25	0.22	0.28	5,628	n/a
countries	Chile	no	0.09	0.04	0.15	2,085	n/a
countries	Colombia	no	0.12	0.09	0.14	5,005	n/a
countries	Croatia	yes	0.36	0.32	0.40	2,354	n/a
countries	Cyprus	no	0.07	0.04	0.11	3,279	n/a
countries	Czechia	no	0.01	−0.02	0.05	3,856	n/a
countries	Denmark	no	0.06	0.03	0.10	4,678	n/a
countries	Ecuador	no	0.05	0.01	0.09	2,243	n/a
countries	Egypt	no	0.00	−0.04	0.04	2,721	n/a
countries	Estonia	yes	−0.06	−0.10	−0.03	3,216	n/a
countries	Ethiopia	no	0.11	0.07	0.16	1,914	n/a
countries	Finland	no	0.19	0.15	0.23	2,897	n/a
countries	France	no	0.18	0.15	0.22	3,845	n/a
countries	Georgia	no	0.05	0.02	0.08	4,283	n/a
countries	Germany	no	0.14	0.11	0.16	8,654	n/a
countries	Ghana	no	0.03	−0.02	0.07	2,330	n/a
countries	Greece	yes	0.29	0.25	0.33	2,118	n/a
countries	Guatemala	no	0.06	0.00	0.11	1,211	n/a
countries	Haiti	no	0.01	−0.03	0.05	1,929	n/a
countries	Hong Kong SAR	no	0.06	0.02	0.09	3,801	n/a
countries	Hungary	no	0.15	0.12	0.19	3,432	n/a
countries	Iceland	no	0.13	0.08	0.17	2,299	n/a
countries	India	no	0.12	0.10	0.15	5,785	n/a
countries	Indonesia	no	0.07	0.03	0.10	4,082	n/a
countries	Iraq	no	0.07	0.01	0.13	978	n/a
countries	Ireland	yes	0.22	0.14	0.29	682	n/a
countries	Italy	no	0.19	0.16	0.23	3,515	n/a
countries	Japan	no	0.13	0.10	0.16	3,386	n/a
countries	Kazakhstan	no	0.11	0.06	0.16	1,500	n/a
countries	Kenya	no	−0.04	−0.10	0.02	1,160	n/a
countries	Kosovo	yes	−0.06	−0.13	0.02	650	n/a
countries	Kyrgyzstan	no	0.18	0.13	0.23	1,458	n/a
countries	Latvia	no	0.05	0.01	0.10	2,033	n/a
countries	Lebanon	no	0.11	0.04	0.17	834	n/a
countries	Libya	no	0.03	−0.02	0.08	2,594	n/a
countries	Lithuania	no	0.12	0.07	0.16	1,812	n/a
countries	Luxembourg	yes	0.27	0.21	0.33	1,261	n/a
countries	Macau SAR	no	0.06	−0.01	0.12	996	n/a
countries	Malaysia	no	0.15	0.11	0.18	2,613	n/a
countries	Maldives	no	0.09	0.03	0.15	1,023	n/a
countries	Mali	no	0.01	−0.04	0.07	1,140	n/a
countries	Malta	no	0.07	−0.01	0.15	770	n/a
countries	Mexico	no	0.10	0.07	0.13	4,781	n/a
countries	Moldova	no	0.05	0.00	0.10	1,657	n/a
countries	Mongolia	no	0.20	0.15	0.25	1,611	n/a
countries	Montenegro	no	0.15	0.09	0.21	1,218	n/a
countries	Morocco	no	0.04	−0.01	0.08	1,914	n/a
countries	Netherlands	no	0.13	0.10	0.15	7,785	n/a
countries	New Zealand	no	0.13	0.08	0.18	1,361	n/a
countries	Nicaragua	no	0.02	−0.05	0.08	871	n/a
countries	Nigeria	no	0.04	0.00	0.08	2,929	n/a
countries	North Macedonia	no	0.09	0.04	0.14	1,729	n/a
countries	Northern Ireland	no	0.16	0.08	0.24	763	n/a
countries	Norway	no	0.09	0.06	0.13	3,144	n/a
countries	Pakistan	no	0.17	0.11	0.22	1,200	n/a
countries	Palestine	no	0.09	0.02	0.16	728	n/a
countries	Peru	no	0.08	0.04	0.11	3,339	n/a
countries	Philippines	no	0.07	0.03	0.12	2,388	n/a
countries	Poland	yes	0.28	0.25	0.31	3,703	n/a
countries	Portugal	no	0.15	0.09	0.20	1,954	n/a
countries	Puerto Rico	no	0.03	−0.03	0.09	1,042	n/a
countries	Romania	no	0.01	−0.02	0.04	4,728	n/a
countries	Russia	no	0.11	0.07	0.14	4,411	n/a
countries	Rwanda	no	0.00	−0.04	0.04	2,554	n/a
countries	Serbia	no	0.08	0.04	0.12	3,715	n/a
countries	Singapore	no	0.11	0.06	0.15	1,740	n/a
countries	Slovakia	no	−0.01	−0.05	0.03	3,257	n/a
countries	Slovenia	yes	0.33	0.30	0.36	3,098	n/a
countries	South Africa	no	0.07	0.04	0.11	5,568	n/a
countries	South Korea	no	0.06	0.02	0.10	3,631	n/a
countries	Spain	yes	0.36	0.33	0.39	4,244	n/a
countries	Sweden	no	0.03	−0.01	0.06	4,155	n/a
countries	Switzerland	no	0.09	0.07	0.12	5,138	n/a
countries	Taiwan ROC	no	0.02	−0.02	0.05	3,561	n/a
countries	Tajikistan	yes	0.27	0.22	0.32	1,200	n/a
countries	Thailand	no	0.02	−0.02	0.05	3,656	n/a
countries	Trinidad and Tobago	no	0.02	−0.04	0.08	1,210	n/a
countries	Tunisia	no	0.11	0.07	0.16	1,681	n/a
countries	Turkey	no	0.21	0.19	0.24	6,438	n/a
countries	Ukraine	no	0.10	0.06	0.15	4,601	n/a
countries	United Kingdom	no	0.12	0.09	0.15	6,155	n/a
countries	United States	yes	0.28	0.25	0.31	5,858	n/a
countries	Uruguay	no	0.20	0.16	0.24	2,586	n/a
countries	Uzbekistan	no	0.19	0.11	0.27	615	n/a
countries	Venezuela	no	−0.04	−0.09	0.02	1,190	n/a
countries	Vietnam	no	0.09	0.03	0.14	1,373	n/a
countries	Zambia	no	−0.02	−0.08	0.04	946	n/a
countries	Zimbabwe	yes	−0.04	−0.09	0.00	2,627	n/a

The United States dots lie to the right of the European and world averages on both measures, and a higher correlation means that more religious respondents placed themselves further right. Each dot is a correlation inside one country or an average across countries, and the rows share one scale even though the survey items differ. Lines on averages are 95% intervals from a bootstrap over countries.Note: The WVS/EVS file holds ten EVS countries with no country label; they are restored from the country code, and 9 of them enter here. Without them the world mean is +0.10.Note: Europe in the WVS/EVS excludes Turkey and the Caucasus and Central Asian countries. A country needs an effective sample of 300 or more. Northern Ireland is a separate WVS/EVS entry.Note: The rank comparison uses only the WVS/EVS rows, where every country answered the same question.Source: General Social Survey, 2006-2024; European Social Survey, rounds 3-11 (2006-2023); World Values Survey and European Values Study, fieldwork 2005-2023; weighted, with equal weight for each country in averages

The rank comparison uses only the WVS/EVS, where every country answered the same question. Attendance gives the same contrast on other surveys. The correlation of attendance with self-placed ideology in the GSS from 2006 to 2024 is +0.25 (95% CI 0.23 to 0.26) in the United States, and the mean across 36 European countries in the European Social Survey (ESS) is +0.11 (95% CI 0.08 to 0.14). The two projects also order countries alike: across the 36 countries that both surveyed, the attendance correlations in the WVS/EVS and the ESS correlate at +0.81 (95% CI 0.66 to 0.90), with a mean absolute difference of 0.04.

In the GSS the correlation is higher in 2024 than in 2002

A rising correlation would make the United States a place where the tie between religiosity and the right is still growing, and the data do not settle it. In the GSS the correlation of attendance with ideology was +0.18 (95% CI 0.11 to 0.24) in 2002 and +0.30 (95% CI 0.25 to 0.34) in 2024. The 2024 value falls in the web era: in 2022 and 2024 pooled the correlation was +0.22 (95% CI 0.16 to 0.27) among in-person respondents and +0.33 (95% CI 0.28 to 0.37) among web respondents, and the in-person 2018 interval overlaps the 2002 interval, so the separation between 2002 and 2024 rests on the web-era point. In the CES the correlation was +0.34 (95% CI 0.32 to 0.35) in 2008, +0.27 (95% CI 0.26 to 0.28) in 2010 and +0.33 (95% CI 0.32 to 0.34) in 2024. In a fixed panel of 15 European countries present in every ESS round, the mean correlation was +0.17 (95% CI 0.14 to 0.20) in 2002 and +0.15 (95% CI 0.10 to 0.21) in 2023, a paired change of −0.02 (95% CI −0.06 to 0.03) (Figure 8).The panel mean hides movements in both directions: the correlation rose in 6 of the countries and fell in 9, with the largest rise in Poland, from +0.20 (95% CI 0.16 to 0.25) to +0.38 (95% CI 0.33 to 0.43), and the largest fall in Belgium, from +0.14 (95% CI 0.08 to 0.19) to −0.04 (95% CI −0.10 to 0.02). The direction of the United States series therefore depends on the survey and on the years compared. For attendance in Europe see Chapter 9.4, Secularization in Europe (planned).

Figure 8

The United States attendance-ideology correlation depends on the survey and the years compared, and a European panel shows no change

Weighted correlation between church attendance and left-right or ideological self-placement, by survey year; the United States in the GSS and CES and a fixed panel of European countries in the ESS; 95% intervals

Correlation of attendance with ideology

Show data

Data for The United States attendance-ideology correlation depends on the survey and the years compared, and a European panel shows no change
Series	Year	Correlation	Lower	Upper	Respondents
Europe, ESS	2002	0.17	0.14	0.20	26,103
Europe, ESS	2004	0.15	0.12	0.20	25,153
Europe, ESS	2006	0.18	0.14	0.22	25,528
Europe, ESS	2008	0.15	0.12	0.20	25,843
Europe, ESS	2010	0.17	0.13	0.21	25,234
Europe, ESS	2012	0.15	0.11	0.20	26,600
Europe, ESS	2014	0.14	0.09	0.20	25,084
Europe, ESS	2016	0.14	0.09	0.19	24,898
Europe, ESS	2018	0.15	0.10	0.21	23,263
Europe, ESS	2020	0.15	0.09	0.21	29,750
Europe, ESS	2023	0.15	0.10	0.21	22,401
US, CES	2008	0.34	0.32	0.35	29,793
US, CES	2009	0.32	0.30	0.34	12,123
US, CES	2010	0.27	0.26	0.28	51,265
US, CES	2011	0.28	0.26	0.30	18,198
US, CES	2012	0.31	0.30	0.33	48,517
US, CES	2013	0.28	0.25	0.30	14,509
US, CES	2014	0.29	0.28	0.31	49,407
US, CES	2015	0.28	0.25	0.30	12,719
US, CES	2016	0.29	0.28	0.30	58,315
US, CES	2017	0.29	0.27	0.31	16,175
US, CES	2018	0.33	0.32	0.34	53,579
US, CES	2019	0.31	0.29	0.33	16,287
US, CES	2020	0.31	0.30	0.32	54,697
US, CES	2021	0.32	0.30	0.33	23,134
US, CES	2022	0.33	0.32	0.34	54,246
US, CES	2023	0.28	0.27	0.30	21,693
US, CES	2024	0.33	0.32	0.34	55,195
US, GSS	2002	0.18	0.11	0.24	1,325
US, GSS	2004	0.17	0.11	0.23	1,305
US, GSS	2006	0.23	0.19	0.26	4,322
US, GSS	2008	0.26	0.21	0.31	1,926
US, GSS	2010	0.21	0.16	0.26	1,968
US, GSS	2012	0.24	0.18	0.29	1,868
US, GSS	2014	0.28	0.24	0.32	2,438
US, GSS	2016	0.23	0.19	0.28	2,748
US, GSS	2018	0.24	0.19	0.29	2,235
US, GSS	2021	0.25	0.21	0.28	3,916
US, GSS	2022	0.25	0.20	0.30	3,395
US, GSS	2024	0.30	0.25	0.34	3,137

The GSS line is higher in 2024 than in 2002, the CES line falls after 2008 and returns by 2018, and the European panel line stays level. Each line is one survey's correlation by year with a 95% interval band, and the European line is the mean over the countries that took part in every ESS round, placed at the year each round began. The shaded band marks the GSS years with web interviews, and the two rings are the GSS of 2022 and 2024 pooled, split by interview mode.Note: GSS 2021 to 2024 include web interviews: 2021 was mostly web, and 2022 and 2024 mixed in-person and web interviews. The 2024 point falls in that period, and the in-person 2018 interval overlaps the 2002 interval. The three surveys use different attendance and ideology items, so levels are compared less safely than trends.Note: ESS rounds are placed at the year fieldwork began; round 10 ran from 2020 to 2022 and round 11 began in 2023. The panel is the 15 countries present in every round: Belgium, Switzerland, Germany, Spain, Finland, France, the United Kingdom, Hungary, Ireland, the Netherlands, Norway, Poland, Portugal, Sweden and Slovenia.Note: The panel mean hides opposite movements: the correlation rose in 6 of the countries and fell in 9. The largest rise is in Poland, from +0.20 in 2002 to +0.38 in 2023, and the largest fall in Belgium, from +0.14 to −0.04.Source: General Social Survey, 2002-2024; Cooperative Election Study, 2008-2024; European Social Survey, rounds 1-11 (2002-2023); weighted

What these data cannot show

Every estimate here is an association between answers given by the same people in one interview, or between groups defined by such answers. The vote gap does not say that attendance moves the vote, the ideology slope does not say that attendance moves ideology, and the share accounted for by party does not say that party moves either one. The panel studies cited above find religious change that follows political orientation, and the surveys here have no panel, so they cannot say how much of each gap comes from people changing their attendance, their affiliation or their party, and how much from who joins and leaves the electorate.

The GSS moved from in-person interviews to a mix that includes the web in 2021, and the chapter reports the split for 2022 and 2024 for the slope, the Republican-share gap and the attendance-ideology correlation. The ANES has included web interviews since 2012 and has no mode weights for 2020, and the chapter does not split its gap by mode. The CES is an opt-in online panel. The ideology and attendance items differ between the GSS and the CES, so levels are compared within a survey and trends are compared across surveys.

The intervals treat each weighted sample as a simple random sample and ignore the strata and clusters of the sampling designs. The GSS design variables allow two checks. Cluster-robust standard errors for the yearly slopes are 0.92 to 1.15 times the HC3 standard errors, on both sides of one, and a bootstrap over primary sampling units gives the share of the rise that party accounts for as 46% (95% CI 33% to 66%), against 46% (95% CI 33% to 67%) from a bootstrap over respondents. The ANES, CES, WVS, EVS and ESS estimates are not checked this way. Race and tradition enter only as views of one figure, with Black Protestants, Latino Catholics and non-Christian faiths left to the chapter on tradition and race. The cross-national comparison sets the correlation inside each country against an average of other countries, using different survey items, so it does not show why the tie is stronger where it is, and the data hold nothing on belief or on congregations and cannot separate the effect of religion from that of the party system in each country.

Data sources

ANES Cumulative Data FileTime Series Cumulative Data File, studies 1952 to 2020 (2024 religion and attendance codes excluded)Years: 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.
Cooperative Election StudyCumulative Common Content, 2008 to 2024Years: 2006-2024 (every year; self-reported turnout is available in the 10 even years) · Weights: weight_cumulative · License: See the provider's terms of use.
General Social SurveyCumulative file, 1972 to 2024, with interview mode from the raw fileYears: 1972-2024 (35 study years) · Weights: weight_use · License: See the provider's terms of use.
World Values Survey and EVSWorld Values Survey and European Values Study, fieldwork 2005 to 2023; ten EVS country labels restoredYears: 1981-2023 (7 unified waves) · Weights: weight (S017) within each country; every country counted once across countries · License: Free download with registration; users must cite the release.
European Social SurveyEuropean Social Survey rounds 1 to 11; attendance and left-right self-placementYears: 2002-2023 (rounds 1 to 11; round 11 was fielded in 2023 and 2024) · Weights: weight (analysis weight) within each country · License: Open access after free registration in the ESS Data Portal; follow the ESS data citation requirements.

All data sources

Methods

Sample. Reported two-party presidential voters in the ANES Time Series Cumulative Data File (studies from 1952 to 2020; attendance is asked from 1970) and in the Cooperative Election Study (presidential years 2008 to 2024, weight_cumulative); General Social Survey respondents 1972 to 2024 (weight_use, with the interview mode joined from the raw file); World Values Survey and European Values Study respondents with fieldwork in 2005 or later; European Social Survey rounds 1 to 11. The cumulative ANES file's 2024 religion and attendance codes are unusable (the count of Protestants falls from 3,180 in 2020 to 835 in 2024 while Catholics are stable, and 49% of 2024 respondents have a valid attendance answer against 99% in 2020), so every ANES series ends in 2020 and 2024 comes from the CES. Groups. Attendance groups follow the ANES three-category item and are cut as closely as each source allows: frequent is weekly or nearly weekly (ANES Often, GSS attend 6 to 8), which in the CES is weekly or more (5 and 6 on the six-point item) because the CES has no nearly-weekly answer; occasional is monthly; seldom is a few times a year or never. Party identification includes leaners (GSS partyid 0 to 2 Democratic, 3 independent, 4 to 6 Republican; CES pid7 1 to 3, 4 and 5 to 7); pure independents are left out of the Republican share and are a category in the party control. The CES party and ideology series use every CES year from 2008 to 2024, and the vote series the presidential years. Estimands. The attendance gap is the weighted Democratic share of the two-party vote among seldom attenders minus that among frequent attenders; the denominational gap is Catholics minus Protestants, and the CES contrast for the unaffiliated is the unaffiliated minus Protestants. The slope is the weighted least squares coefficient of ideology (GSS polviews 1 to 7; CES five-point ideology with not sure missing) on attendance (GSS 0 to 8; CES 1 to 6), both standardized once on the whole common sample of respondents with attendance, ideology and party (GSS 1974 to 2024; CES 2008 to 2024), fitted year by year raw and with the party control on the same respondents (two indicators for Republican and independent, or six indicators for seven categories). A rise is the late slope minus the early slope; the share accounted for by party is one minus the ratio of the rise net of party to the raw rise. Windows of several survey years rescale each year's weights to average one. The cultural index averages the weighted z-scores of five GSS items (homosexuality, premarital sex, extramarital sex, pornography law, abortion for any reason) and the economic index those of three (help for the poor, help for the sick, government action on income differences), each with at least two answers; the economic items are asked from 1975, so both indices and their correlations start there, the first period holds the surveys of 1975, 1983 and 1984 only, and the correlation with self-placed ideology uses the same survey years. Cross-national correlations are weighted Pearson correlations of left-right self-placement with the importance of God (WVS/EVS), or with attendance (ESS rounds 3 to 11 pooled; the GSS for the United States, 2006 to 2024). Intervals. Weighted shares and means use the Kish effective sample size and a normal interval, which treats the weighted sample as a simple random sample and ignores stratification and clustering, which can raise or lower the variance; the same formula is used for the CES, an opt-in panel, and its interval does not cover the error that comes from how respondents were recruited; gaps combine the two standard errors; correlations use Fisher z on the Kish effective sample size; year-by-year slopes use HC3 robust errors; rises and the shares of a rise or of a slope that party accounts for use a percentile bootstrap over respondents (the estimate is recomputed on 2,000 resamples of the respondents drawn with replacement, each window resampled separately, seed 12345, and the interval is read from the percentiles of the results); means across countries count each country once and use the same bootstrap over countries; the correlation between the WVS/EVS and ESS country values uses Fisher z on the number of countries. The GSS design variables are used for two checks: cluster-robust errors on the primary sampling unit change the standard errors of the yearly slopes by a median ratio of 1.01 against the HC3 errors (range 0.92 to 1.15), and a bootstrap over primary sampling units gives the main share as 46% (95% interval 33% to 66%) against 46% (33% to 67%) from the bootstrap over respondents. The shares and correlations of the ANES, CES, WVS/EVS and ESS are not checked this way. Suppression. A subgroup-year cell needs at least 100 respondents and an effective sample of the same size; a yearly regression needs 200 respondents; a country correlation needs an effective sample of 300; no published cell rests on fewer than 30 respondents. Countries. The cleaned WVS/EVS file has a blank country label for ten EVS countries; they are restored from the country code and 9 of them have a usable correlation. Europe in the WVS/EVS is the region Europe and Central Asia without Turkey and the Caucasus and Central Asian countries. The European time series uses the 15 countries in every ESS round and places each round at the year fieldwork began. Mode. GSS interviews were in person, with some by telephone, through 2018, mostly on the web in 2021, and a mix of in-person and web interviews in 2022 and 2024; the CES is online throughout. Estimates from 2021 on are reported with a mode split or a mode note: the split covers the slope, the Republican-share gap and the attendance-ideology correlation in 2022 and 2024, and the party comparison of the text uses the in-person surveys of 2010 to 2018 with the 2018 to 2024 window as a second comparison. The ANES mixes in-person and web interviews from 2012 and has no mode weights in 2020, and its vote gap is not split by mode.

References

Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Layman, Geoffrey C. (2001). The Great Divide: Religious and Cultural Conflict in American Party Politics. Columbia University Press.
Putnam, Robert D.; Campbell, David E. (2010). American Grace: How Religion Divides and Unites Us. Simon and Schuster.
Margolis, Michele F. (2018). From Politics to the Pews: How Partisanship and the Political Environment Shape Religious Identity. University of Chicago Press.
Campbell, David E.; Layman, Geoffrey C.; Green, John C.; Sumaktoyo, Nathanael G. (2018). Putting Politics First: The Impact of Politics on American Religious and Secular Orientations. American Journal of Political Science 62(3): 551-565. doi:10.1111/ajps.12365

How to cite

@misc{singh2026thegodgapinamerica,
  author = {Singh, Vishal},
  title = {The god gap in America},
  year = {2026},
  url = {https://vishalsingh.org/politics/the-god-gap-in-america},
  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.

Religion
- Chapter 9.4, Secularization in Europe (planned) Countries

The ANES vote gap has been above zero since 1992

The gap is larger among white voters and smaller among Protestants and Catholics

The attendance gap has exceeded the denominational gap since 1992

The Republican share among frequent attenders rose in the GSS

Party identification accounts for part of the rise in the slope

Attendance correlates more closely with cultural than with economic attitudes

The United States sits above the European and world averages

In the GSS the correlation is higher in 2024 than in 2002

What these data cannot show

Data sources

Methods

References

How to cite

Same question at another grain

Reading paths