Chapter 3.6

Age and generation

Two-party voters under thirty were more Democratic than those sixty-five and older by +22.2 points in the 2020 CES and +15.0 in 2024, a fall that is detectable only among all CES voters, and repeated surveys cannot separate age, election and birth cohort without an assumption that moves the size of the age effect.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 3 October 2026

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

This draft uses repeated cross-sections only. The ANES 2016-2020-2024 panel and the CES panel files, which re-interview the same people, are not in the data store.

In the Cooperative Election Study (CES), 67.2% (95% CI 65.2% to 69.2%) of two-party voters under thirty and 42.7% (95% CI 41.0% to 44.5%) of two-party voters sixty-five and older voted for the Democratic presidential candidate in 2008, a gap of +24.5 (95% CI 21.8 to 27.1) points. The gap was +22.2 (95% CI 19.9 to 24.4) in 2020 and +15.0 (95% CI 11.8 to 18.3) in 2024. The American National Election Studies (ANES) give a gap of +17.0 (95% CI 11.0 to 23.1) in 2020 and −1.1 (95% CI −9.4 to 7.1) in 2024, an interval that includes zero, from a 2024 sample whose Democratic share, 54.5% (95% CI 52.3% to 56.7%), is above the certified count, 49.2%, while the CES share, 49.0% (95% CI 48.2% to 49.7%), is close to it. Among CES voters whose registration was matched to voter-file records the 2024 gap is +23.0 (95% CI 19.1 to 26.9), so the size of the recent fall depends on which voters are counted.

A difference between age groups mixes three things. Age is what happens to a person as they get older. Period is what happens to every voter at one election, such as a landslide or a recession. Birth cohort is what people born in the same years share, such as the events of their teenage years, and a cohort is the group born in those years. The three are tied by arithmetic, because birth year equals election year minus age. A survey that records age and the election year already holds birth year, and no repeated survey can separate all three effects without an assumption [1] [2]. This chapter reports the age gap in the vote and in party identification from two surveys, follows birth cohorts from election to election, and then shows in the same data how far an estimated age effect moves with the assumption chosen.

The ANES Cumulative Data File supplies the vote of 33,768 respondents in 19 presidential elections from 1952 to 2024 who reported voting for the Democratic or the Republican candidate and gave an age of eighteen or older. The CES, an online panel, supplies 188,145 such respondents in the elections from 2008 to 2024. Such a respondent is a two-party voter, and a Democratic share is the share of two-party voters who reported a Democratic vote. Party identification counts Democrats and Republicans, with leaners included. The age groups are eighteen to twenty-nine, thirty to forty-four, forty-five to sixty-four and sixty-five and older, the bands that Chapter 3.1, The two coalitions uses for the age make-up of each party's voters.Every interval is a 95% confidence interval from the Kish effective sample size [3] on the survey weights, and it treats the weighted sample as a simple random sample. The CES is an opt-in panel, so its intervals leave out how respondents joined the panel. A cell needs at least 100 respondents and an effective sample of 100 to be shown, and smaller cells are left blank. ANES vote shares use the post-election weight in the elections where the file has one (1996, 2000, 2004, 2008, 2016, 2020 and 2024) and the full-sample weight otherwise, which includes 2012. CES shares use the cumulative-file weight, which is proportional within each election to the pre-election weight and is the one weight defined for every election. The vote is a post-election item, and the 2024 section checks the choice against the post-election weight. The methods note in the appendix names the weight of each estimate.

Three comparisons, and what each can say

A repeated survey offers three ways to compare voters, and each holds a different quantity fixed. Comparing age groups inside one election holds the period fixed, and it mixes age with birth cohort, because the older voters in an election were also born earlier. Comparing one age group across elections holds age fixed, and it mixes period with birth cohort, because the voters under thirty in two elections were born in different years. Following a birth cohort from one election to the next holds the cohort fixed, and it mixes age with period, because the cohort grows older and the election changes. Each comparison is the sum of two of the three effects, and none isolates one of them. The estimates of all three come near the end of the chapter (Figure 8 and Figure 9), after the descriptive evidence that uses them.

The youngest voters were more Democratic than the oldest in every election from 2008 to 2020 in both surveys

Figure 1 plots the Democratic share of the two-party vote in each ANES election by age group. Before 2004 the sign of the gap between the youngest and oldest groups (Figure 3) varies from one election to the next. The gap was +2.9 (95% CI −5.9 to 11.8) points in 1992. Its interval is above zero in 1964, in 1972 and in every election from 2004 to 2020, which is 7 of the 16 elections with enough voters under thirty to publish, and it includes zero in the rest. In 3 elections (1960, 1996 and 2000) the youngest group has too few respondents for a published cell. The gap was +22.6 (95% CI 12.2 to 33.0) in 2008 and +17.0 (95% CI 11.0 to 23.1) in 2020. In 2024 the youngest group, at 56.3% (95% CI 48.9% to 63.6%), is within about a point of the oldest, at 57.4% (95% CI 53.7% to 61.1%), and the interval for the gap, −1.1 (95% CI −9.4 to 7.1), includes zero.

Figure 1

The ANES age gap in the Democratic vote was above zero in every election from 2004 to 2020 and included zero in 2024

Democratic share of the two-party presidential vote by age group (18 to 29, 30 to 44, 45 to 64 and 65 and over), ANES 1952 to 2024; 95% intervals

Democratic share of two-party vote (%)

Show data

Data for The ANES age gap in the Democratic vote was above zero in every election from 2004 to 2020 and included zero in 2024
Interviews	Election	Age group	Democratic share (%)	95% interval low	95% interval high	Voters
all	1952	18 to 29	42.3	35.1	49.5	182
all	1956	18 to 29	42.1	34.9	49.4	178
all	1960	18 to 29	n/a	n/a	n/a	n/a
all	1964	18 to 29	72.3	65.5	79.1	166
all	1968	18 to 29	44.9	36.6	53.3	138
all	1972	18 to 29	47.3	42.4	52.2	402
all	1976	18 to 29	52.5	46.5	58.5	272
all	1980	18 to 29	41.2	33.8	48.6	170
all	1984	18 to 29	43.2	37.1	49.3	257
all	1988	18 to 29	47.8	40.5	55.1	182
all	1992	18 to 29	63.4	56.5	70.2	192
all	1996	18 to 29	n/a	n/a	n/a	n/a
all	2000	18 to 29	n/a	n/a	n/a	n/a
all	2004	18 to 29	66.0	57.3	74.7	137
all	2008	18 to 29	67.5	59.8	75.2	222
all	2012	18 to 29	62.6	56.8	68.4	501
all	2016	18 to 29	61.9	54.9	68.8	299
all	2020	18 to 29	69.7	64.5	74.9	588
all	2024	18 to 29	56.3	48.9	63.6	285
all	1952	30 to 44	46.9	42.1	51.6	429
all	1956	30 to 44	41.8	37.4	46.2	486
all	1960	30 to 44	51.2	44.8	57.6	327
all	1964	30 to 44	68.7	64.0	73.4	377
all	1968	30 to 44	52.1	46.3	58.0	282
all	1972	30 to 44	34.2	29.6	38.8	415
all	1976	30 to 44	48.5	43.2	53.9	352
all	1980	30 to 44	41.3	35.3	47.3	259
all	1984	30 to 44	40.7	36.3	45.1	482
all	1988	30 to 44	46.4	41.7	51.1	433
all	1992	30 to 44	56.1	51.6	60.6	474
all	1996	30 to 44	54.0	48.3	59.6	352
all	2000	30 to 44	46.2	40.5	51.9	354
all	2004	30 to 44	44.5	37.1	52.0	196
all	2008	30 to 44	58.4	52.4	64.5	411
all	2012	30 to 44	55.7	51.4	59.9	904
all	2016	30 to 44	61.9	57.3	66.6	568
all	2020	30 to 44	58.8	55.4	62.3	1,373
all	2024	30 to 44	55.0	50.3	59.6	841
all	1952	45 to 64	38.8	34.0	43.5	405
all	1956	45 to 64	36.6	32.1	41.2	434
all	1960	45 to 64	50.2	44.1	56.2	356
all	1964	45 to 64	69.1	64.6	73.6	408
all	1968	45 to 64	42.4	37.1	47.7	335
all	1972	45 to 64	29.8	25.9	33.7	527
all	1976	45 to 64	53.3	48.5	58.0	445
all	1980	45 to 64	46.1	40.2	52.1	271
all	1984	45 to 64	41.7	36.8	46.6	391
all	1988	45 to 64	47.2	42.0	52.5	345
all	1992	45 to 64	57.5	52.6	62.3	395
all	1996	45 to 64	59.2	53.4	65.0	331
all	2000	45 to 64	53.1	47.8	58.4	416
all	2004	45 to 64	45.0	39.1	50.9	332
all	2008	45 to 64	52.3	47.3	57.2	591
all	2012	45 to 64	52.3	49.2	55.4	1,760
all	2016	45 to 64	48.4	44.7	52.2	976
all	2020	45 to 64	50.6	47.6	53.6	2,014
all	2024	45 to 64	51.8	47.9	55.7	1,236
all	1952	65 and over	36.7	28.9	44.4	150
all	1956	65 and over	43.9	35.9	51.9	148
all	1960	65 and over	39.2	29.9	48.4	148
all	1964	65 and over	55.3	47.6	63.1	159
all	1968	65 and over	45.1	37.2	53.0	153
all	1972	65 and over	31.9	25.9	37.9	235
all	1976	65 and over	49.4	43.0	55.8	245
all	1980	65 and over	46.3	38.9	53.7	175
all	1984	65 and over	42.9	36.5	49.3	233
all	1988	65 and over	47.9	41.4	54.3	234
all	1992	65 and over	60.4	54.8	66.0	296
all	1996	65 and over	60.6	54.0	67.3	248
all	2000	65 and over	54.2	47.0	61.4	228
all	2004	65 and over	52.1	43.2	61.0	146
all	2008	65 and over	44.9	37.9	52.0	294
all	2012	65 and over	46.9	42.7	51.0	983
all	2016	65 and over	46.4	41.9	50.8	679
all	2020	65 and over	52.7	49.6	55.8	1,883
all	2024	65 and over	57.4	53.7	61.1	1,390
ftf	2012	18 to 29	63.2	53.8	72.6	225
ftf	2012	30 to 44	58.0	51.0	65.0	376
ftf	2016	30 to 44	68.4	59.7	77.2	144
ftf	2012	45 to 64	51.7	45.6	57.9	474
ftf	2016	45 to 64	51.0	43.8	58.3	281
ftf	2024	45 to 64	45.8	37.0	54.7	224
ftf	2012	65 and over	41.5	33.1	49.9	221
ftf	2016	65 and over	45.2	36.9	53.5	216
ftf	2024	65 and over	54.9	46.6	63.2	255
web	2012	18 to 29	62.3	55.0	69.6	276
web	2016	18 to 29	66.7	58.9	74.5	210
web	2012	30 to 44	54.4	49.1	59.7	528
web	2016	30 to 44	59.7	54.3	65.1	426
web	2024	30 to 44	55.4	48.4	62.4	314
web	2012	45 to 64	52.6	49.0	56.1	1,286
web	2016	45 to 64	47.6	43.3	51.9	703
web	2024	45 to 64	57.6	51.3	63.9	410
web	2012	65 and over	49.5	44.8	54.1	762
web	2016	65 and over	47.4	42.3	52.5	473
web	2024	65 and over	59.8	53.4	66.2	407

The line for the youngest group runs above the line for the oldest from 2004 to 2020, by +22.6 points in 2008, and the two estimates are within about a point in 2024, at 56.3% and 57.4%, with a gap whose interval includes zero. Before 2004 the sign of the gap varies and its interval is above zero only in 1964 and 1972. Each line is the weighted Democratic share among two-party voters in one age group, and the bands are 95% intervals. The buttons restrict the sample to face-to-face or web interviews, which the file separates for 2012, 2016 and 2024.Note: A cell is plotted only when it has 100 or more respondents and a Kish effective sample size (the sample size after weighting) of 100 or more, the floor of this chapter. A blank cell leaves a gap in the line. The youngest group falls below the floor in 3 elections (1960, 1996 and 2000).Note: The mode views use the face-to-face weight on face-to-face respondents and the web weight on web respondents. The 2020 study has no interview-mode weights and no in-person interviews, so the mode views leave 2020 open, and many youngest-group cells fall below the floor within a mode.Note: Age is top-coded at 80 in the 2020 and 2024 files, 649 two-party voters with no recorded age are outside every cell, and respondents under 18 (a handful in the file) are left out. Intervals are model based: they treat the weighted sample as a simple random sample and ignore clustering and stratification.Source: ANES Cumulative Data File, presidential elections 1952-2024; weighted (post-election weight where the file has one, otherwise VCF0009z); two-party presidential voters

Figure 2 gives the same comparison in the CES, whose larger samples of young voters allow five age bands. In 2008, 2012, 2016 and 2020 the band aged thirty to forty-four is more Democratic than the band forty-five to sixty-four, which is more Democratic than the band sixty-five and older, and every interval for those two differences excludes zero. The two youngest bands are not always separated from their neighbors: in 2012 the band eighteen to twenty-four differs from the band twenty-five to twenty-nine by +0.5 (95% CI −4.3 to 5.2) points, and the band twenty-five to twenty-nine differs from the band thirty to forty-four by +2.9 (95% CI −0.9 to 6.7). The gap between the two end groups was between +18.7 and +24.5 points in those four elections, each with an interval above zero, and it was +15.0 (95% CI 11.8 to 18.3) in 2024. In 2024 the order of the two oldest bands is not settled: the band forty-five to sixty-four differs from the band sixty-five and older by −1.6 (95% CI −3.2 to 0.0) points. The two youngest bands differ by +0.5 to +5.0 points across the five elections, so the gap between the end groups is not mainly a difference between the very youngest voters and the rest.

Figure 2

In every CES election from 2008 to 2020 the Democratic share was lower among voters forty-five to sixty-four than among voters thirty to forty-four, and lower again among voters sixty-five and older

Democratic share of the two-party presidential vote by age band (18 to 24, 25 to 29, 30 to 44, 45 to 64 and 65 and over), CES 2008 to 2024; 95% intervals

Democratic share of two-party vote (%)

Show data

Data for In every CES election from 2008 to 2020 the Democratic share was lower among voters forty-five to sixty-four than among voters thirty to forty-four, and lower again among voters sixty-five and older
Weight	Election	Age band	Democratic share (%)	95% interval low	95% interval high	Voters
standard	2008	18 to 24	69.3	66.5	72.1	1,641
standard	2012	18 to 24	62.2	58.8	65.6	1,633
standard	2016	18 to 24	67.7	64.8	70.7	1,569
standard	2020	18 to 24	70.5	67.6	73.4	1,795
standard	2024	18 to 24	63.4	58.7	68.2	1,478
standard	2008	25 to 29	64.3	61.7	67.0	1,686
standard	2012	25 to 29	61.7	58.4	65.0	1,919
standard	2016	25 to 29	64.0	61.5	66.6	2,449
standard	2020	25 to 29	65.9	63.4	68.5	2,565
standard	2024	25 to 29	58.5	54.5	62.4	1,604
standard	2008	30 to 44	56.4	54.9	57.9	5,702
standard	2012	30 to 44	58.8	56.9	60.7	4,370
standard	2016	30 to 44	57.5	56.2	58.9	8,868
standard	2020	30 to 44	61.5	60.2	62.7	9,870
standard	2024	30 to 44	55.1	53.5	56.8	7,073
standard	2008	45 to 64	50.6	49.4	51.7	10,079
standard	2012	45 to 64	51.3	50.3	52.2	18,715
standard	2016	45 to 64	47.4	46.5	48.3	18,133
standard	2020	45 to 64	49.2	48.3	50.2	17,747
standard	2024	45 to 64	43.9	42.8	45.1	16,848
standard	2008	65 and over	42.7	41.0	44.5	4,477
standard	2012	65 and over	43.2	41.9	44.6	11,152
standard	2016	65 and over	42.8	41.3	44.3	9,872
standard	2020	65 and over	45.6	44.4	46.8	11,895
standard	2024	65 and over	45.5	44.4	46.7	15,005
validated	2020	18 to 24	70.4	67.0	73.9	1,180
validated	2024	18 to 24	68.7	62.9	74.6	530
validated	2020	25 to 29	67.9	64.8	70.9	1,736
validated	2024	25 to 29	62.3	57.5	67.0	873
validated	2020	30 to 44	62.8	61.4	64.1	7,532
validated	2024	30 to 44	57.0	55.2	58.7	5,208
validated	2020	45 to 64	48.7	47.7	49.7	14,897
validated	2024	45 to 64	45.9	44.8	47.0	13,637
validated	2020	65 and over	46.1	44.8	47.4	10,401
validated	2024	65 and over	41.6	40.4	42.9	12,700

The three oldest lines keep their order in 2008, 2012, 2016 and 2020, and the intervals for the two differences between neighboring bands exclude zero. The two youngest bands are not always separated from their neighbors: in 2012 the difference between the two youngest is +0.5 points and between the second and third bands +2.9. In 2024 the band aged forty-five to sixty-four differs from the band sixty-five and older by −1.6 points, an interval that reaches zero, so the order of those two bands is not settled. The two youngest bands differ by +0.5 to +5.0 points across the five elections. Each line is the weighted Democratic share among two-party voters in one age band, and the bands are 95% intervals.Note: The CES holds far more young voters than the ANES, so no cell falls below the floor. The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel.Note: Intervals are model based: they treat the weighted sample as a simple random sample and ignore clustering and stratification.Source: CES Cumulative File, presidential elections 2008-2024; weighted (weight_cumulative); two-party presidential voters

Both figures compare age groups inside each election and follow one age group across elections, so both mix age with birth year. The voters under thirty in 2008 and in 2024 were born in different years, and so were the voters sixty-five and older.

The 2024 gap fell among all CES voters, with no detectable fall among voters matched to registration records

Between 2020 and 2024 the CES gap changed by −7.1 (95% CI −11.1 to −3.2) points and the ANES gap by −18.2 (95% CI −28.4 to −7.9) points. The CES file also holds a weight, called the validated-voter weight, for voters whose registration could be matched to a voter-file record, which this chapter calls matched voters. Among them the gap was +22.8 (95% CI 20.1 to 25.4) in 2020 and +23.0 (95% CI 19.1 to 26.9) in 2024, a change of +0.2 (95% CI −4.5 to 4.9) points. With the standard weight on the same matched voters the 2024 gap is +22.2 (95% CI 17.9 to 26.5) and its change from 2020 is −0.6 (95% CI −5.6 to 4.5) points (Figure 3). Both intervals extend below zero by several points, so the data show no detectable fall among matched voters and do not show that none occurred. The change among all voters minus the change among matched voters is −6.6 (95% CI −10.2 to −2.8) points with the standard weight and −7.3 (95% CI −11.8 to −3.0) with the validated-voter weight (bootstrap intervals), so the two changes differ by more than sampling error.

For the age gap, the difference between the estimates is associated with the sample and not with the weight. On the same matched voters the validated-voter weight moves the 2024 gap by +0.8 (95% CI −3.0 to 4.8) points, an interval that includes zero, while the gap among matched voters is above the gap among all voters by +7.2 (95% CI 3.8 to 10.6) points (bootstrap intervals). Chapter 3.5, The gender gap in attitudes and votes finds that the weight matters on the same respondents for the gender gap. The share of voters under thirty who could be matched fell from 66.9% in 2020 to 45.5% in 2024, and the share among voters sixty-five and older was 87.4% and 84.6%. In 2024 matched voters under thirty were 68.2% (95% CI 64.1% to 72.4%) Democratic and unmatched voters under thirty 54.5% (95% CI 50.2% to 58.8%), a difference of +13.8 (95% CI 7.8 to 19.7) points. In 2020 the difference was +3.0 (95% CI −1.1 to 7.1) points, an interval that includes zero. The data cannot say why the unmatched young voters of 2024 were less Democratic. Some may be voters whose records the match missed, and some may be respondents who said they voted and did not.

The standard CES weight is proportional to the pre-election weight, and the vote is a post-election item. The post-election weight exists for 2012, 2016 and 2020 and gives age gaps of +19.9, +22.4 and +20.4 (95% CI 17.9 to 22.9), against +18.7, +22.9 and +22.2 with the standard weight. It does not exist for 2024, so the change from 2020 with the post-election weight to 2024 with the standard weight mixes two weights. That change is −5.3 (95% CI −9.4 to −1.2) points, smaller in size than the −7.1 with the standard weight, and its interval excludes zero. The fall among all voters therefore does not depend on this choice, and its size does.

The ANES gives no cleaner reading, because its 2024 sample leans Democratic. Its two-party Democratic share was 54.5% (95% CI 52.3% to 56.7%) against 49.2% in the certified count and 49.0% (95% CI 48.2% to 49.7%) in the CES. The ANES exceeds the count by +5.3 (95% CI 3.0 to 7.5) points, and the CES differs from it by −0.3 (95% CI −1.0 to 0.5). The ANES also exceeded the count in 2020, by +3.5 (95% CI 1.7 to 5.2) points. The excess is largest among voters sixty-five and older, where the ANES share, 57.4% (95% CI 53.7% to 61.1%), is above the CES share by +11.9 (95% CI 8.0 to 15.7) points and above the CES share among matched voters of that age, 46.0% (95% CI 44.8% to 47.2%). It is also positive among voters forty-five to sixty-four, at +7.8 (95% CI 3.8 to 11.9) points. Among voters under thirty the ANES is below the CES by −4.3 (95% CI −12.3 to 3.6) points, an interval that includes zero. The near-zero ANES gap in 2024 therefore comes mostly from its oldest group. No official count gives the vote by age, so which survey is closer within an age group cannot be tested here.

Split by interview mode, the 2024 ANES cells for voters under thirty have effective sample sizes of 86 in the web sample and 31 in the face-to-face sample, below the floor, so the ANES cannot show whether the fall depends on mode. In the web sample the gap was +12.8 (95% CI 4.2 to 21.5) in 2012 and +19.3 (95% CI 10.0 to 28.7) in 2016, and in the face-to-face sample it was +21.7 (95% CI 9.2 to 34.3) in 2012 (the second button of Figure 3). In 2024 1,884 of the 3,933 two-party voters (48%) are panel or paper cases with no mode weight and fall outside every mode cell, and the 2020 study has no mode weights.

Figure 3

Between 2020 and 2024 the CES age gap in the vote changed by −7.1 points among all voters and by −0.6, an interval that includes zero, among matched voters

Democratic share of two-party voters aged 18 to 29 minus the share among voters aged 65 and over, ANES 1952 to 2024 and CES 2008 to 2024; 95% intervals

Youngest minus oldest group (percentage points)

Show data

Data for Between 2020 and 2024 the CES age gap in the vote changed by −7.1 points among all voters and by −0.6, an interval that includes zero, among matched voters
Measure	Survey and weight	Year	Gap (points)	95% interval low	95% interval high	Respondents in the two groups
vote	ANES	1952	+5.6	−4.9	+16.2	332
vote	ANES	1956	−1.8	−12.6	+9.0	326
vote	ANES	1960	n/a	n/a	n/a	n/a
vote	ANES	1964	+16.9	+6.6	+27.3	325
vote	ANES	1968	−0.2	−11.7	+11.3	291
vote	ANES	1972	+15.3	+7.6	+23.1	637
vote	ANES	1976	+3.1	−5.7	+11.9	517
vote	ANES	1980	−5.1	−15.6	+5.4	345
vote	ANES	1984	+0.3	−8.5	+9.1	490
vote	ANES	1988	−0.1	−9.8	+9.6	416
vote	ANES	1992	+2.9	−5.9	+11.8	488
vote	ANES	1996	n/a	n/a	n/a	n/a
vote	ANES	2000	n/a	n/a	n/a	n/a
vote	ANES	2004	+13.9	+1.4	+26.3	283
vote	ANES	2008	+22.6	+12.2	+33.0	516
vote	ANES	2012	+15.7	+8.6	+22.8	1,484
vote	ANES	2016	+15.5	+7.2	+23.7	978
vote	ANES	2020	+17.0	+11.0	+23.1	2,471
vote	ANES	2024	−1.1	−9.4	+7.1	1,675
vote	CES	2008	+24.5	+21.8	+27.1	7,804
vote	CES	2012	+18.7	+16.0	+21.4	14,704
vote	CES	2016	+22.9	+20.4	+25.3	13,890
vote	CES	2020	+22.2	+19.9	+24.4	16,255
vote	CES	2024	+15.0	+11.8	+18.3	18,087
vote	CES, matched voters, validated-voter weight	2020	+22.8	+20.1	+25.4	13,317
vote	CES, matched voters, validated-voter weight	2024	+23.0	+19.1	+26.9	14,103
mode	ANES, all interviews	2008	+22.6	+12.2	+33.0	516
mode	ANES, all interviews	2012	+15.7	+8.6	+22.8	1,484
mode	ANES, all interviews	2016	+15.5	+7.2	+23.7	978
mode	ANES, all interviews	2020	+17.0	+11.0	+23.1	2,471
mode	ANES, all interviews	2024	−1.1	−9.4	+7.1	1,675
mode	ANES, face to face	2012	+21.7	+9.2	+34.3	446
mode	ANES, web	2012	+12.8	+4.2	+21.5	1,038
mode	ANES, web	2016	+19.3	+10.0	+28.7	683
weight	CES, standard weight	2008	+24.5	+21.8	+27.1	7,804
weight	CES, standard weight	2012	+18.7	+16.0	+21.4	14,704
weight	CES, standard weight	2016	+22.9	+20.4	+25.3	13,890
weight	CES, standard weight	2020	+22.2	+19.9	+24.4	16,255
weight	CES, standard weight	2024	+15.0	+11.8	+18.3	18,087
weight	CES, post-election weight	2012	+19.9	+17.2	+22.6	14,704
weight	CES, post-election weight	2016	+22.4	+19.7	+25.0	13,890
weight	CES, post-election weight	2020	+20.4	+17.9	+22.9	16,254
weight	CES, matched voters, validated-voter weight	2020	+22.8	+20.1	+25.4	13,317
weight	CES, matched voters, validated-voter weight	2024	+23.0	+19.1	+26.9	14,103
pid	ANES	1952	+10.2	+1.4	+19.0	486
pid	ANES	1956	+8.5	−0.9	+18.0	438
pid	ANES	1960	n/a	n/a	n/a	n/a
pid	ANES	1964	+17.7	+9.0	+26.3	472
pid	ANES	1968	+0.2	−8.6	+9.1	469
pid	ANES	1972	+7.5	+1.2	+13.8	973
pid	ANES	1976	+7.1	+0.3	+13.9	825
pid	ANES	1980	−2.6	−10.6	+5.4	588
pid	ANES	1984	+4.8	−2.1	+11.7	815
pid	ANES	1988	−5.3	−12.6	+2.1	704
pid	ANES	1992	−6.1	−12.8	+0.7	832
pid	ANES	1996	+0.6	−8.6	+9.8	548
pid	ANES	2000	+1.7	−7.8	+11.1	505
pid	ANES	2004	+8.9	−1.8	+19.6	393
pid	ANES	2008	+13.1	+4.1	+22.0	711
pid	ANES	2012	+7.0	+0.9	+13.2	1,853
pid	ANES	2016	+10.2	+3.8	+16.6	1,395
pid	ANES	2020	+14.5	+9.2	+19.8	2,883
pid	ANES	2024	+1.9	−4.7	+8.6	2,110
pid	CES	2006	+16.3	+14.2	+18.4	9,392
pid	CES	2008	+19.9	+17.6	+22.3	9,848
pid	CES	2010	+21.3	+18.7	+23.9	15,016
pid	CES	2012	+17.5	+15.2	+19.7	17,855
pid	CES	2014	+14.5	+12.1	+16.9	17,157
pid	CES	2016	+14.3	+12.3	+16.4	18,625
pid	CES	2018	+17.2	+15.1	+19.3	20,577
pid	CES	2020	+14.9	+12.9	+17.0	20,364
pid	CES	2022	+21.1	+19.0	+23.3	20,148
pid	CES	2024	+13.9	+11.2	+16.5	22,842

The CES gap is above zero in every election and lower in 2024 than in 2020 among all voters. Among voters whose registration was matched to voter-file records the 2020 gap, +22.8, and the 2024 gap with the validated-voter weight, +23.0, are within sampling error of each other, and the change among all voters differs from the change among matched voters by −6.6 points. The ANES gap is +17.0 in 2020 and −1.1 in 2024, from a 2024 sample that exceeds the certified count by +5.3 points. A gap above zero means the youngest group is more Democratic. Each point is the difference of two weighted shares, and the bands are 95% intervals. The other buttons show the CES by weight, split the ANES by interview mode and show party identification.Note: A cell is plotted only when it has 100 or more respondents and a Kish effective sample size (the sample size after weighting) of 100 or more, the floor of this chapter. A blank cell leaves a gap in the line. Three ANES elections (1960, 1996 and 2000) have too few voters under thirty for a published gap and show a break in the line.Note: The matched-voter series uses voters with a validated-voter weight, the CES weight for voters whose registration was matched to voter-file records (validated registered voters), which exists for 2018 to 2024 in the file. In 2020 it is a constant multiple of the standard weight. The post-election weight exists for 2012, 2016 and 2020.Note: In the mode view the face-to-face gap for 2016 and the face-to-face and web gaps for 2024 fall below the floor and are not drawn. The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel.Source: ANES Cumulative Data File, presidential elections 1952-2024; weighted (post-election weight where the file has one, otherwise VCF0009z); two-party presidential voters; CES weighted (weight_cumulative; vvweight for the matched-voter series)

The CES gap in party identification was lowest in 2024, and whether it fell depends on the baseline wave

Party identification is asked of every respondent, voters or not, and it does not depend on the two candidates of one election, so it offers a check on the vote comparison. Figure 4 plots the Democratic share of party identifiers by age group. In the CES the gap between the youngest and oldest groups was between +13.9 and +21.3 points in each of the 10 even-year waves from 2006 to 2024, with every interval above zero, and 2024 is the lowest of the ten. It was +14.9 (95% CI 12.9 to 17.0) in 2020 and +13.9 (95% CI 11.2 to 16.5) in 2024, a change of −1.1 (95% CI −4.4 to 2.3) points that includes zero. The baseline matters. The gap was +17.2 (95% CI 15.1 to 19.3) in 2018, a change to 2024 of −3.3 (95% CI −6.7 to 0.1) points that includes zero, and +21.1 (95% CI 19.0 to 23.3) in 2022, a change to 2024 of −7.3 (95% CI −10.6 to −3.9) points. The ANES gap fell from +14.5 (95% CI 9.2 to 19.8) to +1.9 (95% CI −4.7 to 8.6), a change of −12.6 (95% CI −21.1 to −4.0) points, and the web sample alone gives +1.1 (95% CI −8.9 to 11.2) in 2024, while the face-to-face sample has too few respondents under thirty to publish.

Figure 4

Among party identifiers in the CES the youngest group was more Democratic than the oldest in every wave from 2006 to 2024

Democratic share of party identifiers (leaners included) by age group, CES even-year waves 2006 to 2024; the ANES view covers presidential-election years 1952 to 2024; 95% intervals

Democratic share of party identifiers (%)

Show data

Data for Among party identifiers in the CES the youngest group was more Democratic than the oldest in every wave from 2006 to 2024
Survey	Year	Age group	Democratic share (%)	95% interval low	95% interval high	Identifiers
ces	2006	18 to 29	63.9	62.3	65.5	3,805
ces	2008	18 to 29	67.1	65.4	68.8	4,827
ces	2010	18 to 29	63.9	61.8	65.9	4,218
ces	2012	18 to 29	65.8	64.0	67.7	6,229
ces	2014	18 to 29	59.1	57.4	60.8	7,389
ces	2016	18 to 29	62.8	61.5	64.2	8,667
ces	2018	18 to 29	61.9	60.3	63.5	9,327
ces	2020	18 to 29	60.3	58.7	61.9	8,824
ces	2022	18 to 29	63.8	62.1	65.5	7,083
ces	2024	18 to 29	60.3	57.9	62.7	7,167
ces	2006	30 to 44	52.7	51.6	53.9	8,096
ces	2008	30 to 44	58.2	56.8	59.6	7,118
ces	2010	30 to 44	56.5	55.0	58.0	8,104
ces	2012	30 to 44	63.6	62.0	65.2	6,253
ces	2014	30 to 44	57.3	55.8	58.8	7,695
ces	2016	30 to 44	60.7	59.6	61.8	13,395
ces	2018	30 to 44	60.2	59.0	61.5	11,724
ces	2020	30 to 44	59.2	58.0	60.4	11,969
ces	2022	30 to 44	59.6	58.3	60.9	11,526
ces	2024	30 to 44	57.8	56.2	59.4	9,128
ces	2006	45 to 64	50.1	49.2	51.0	14,304
ces	2008	45 to 64	53.8	52.7	54.9	11,634
ces	2010	45 to 64	52.5	51.6	53.4	24,741
ces	2012	45 to 64	56.0	55.1	56.9	21,739
ces	2014	45 to 64	51.5	50.5	52.5	19,924
ces	2016	45 to 64	52.6	51.7	53.5	19,966
ces	2018	45 to 64	50.2	49.1	51.2	17,026
ces	2020	45 to 64	49.6	48.6	50.6	17,436
ces	2022	45 to 64	48.8	47.8	49.8	17,846
ces	2024	45 to 64	45.4	44.2	46.5	18,877
ces	2006	65 and over	47.6	46.2	49.0	5,587
ces	2008	65 and over	47.2	45.5	48.9	5,021
ces	2010	65 and over	42.6	40.9	44.2	10,798
ces	2012	65 and over	48.4	47.0	49.7	11,626
ces	2014	65 and over	44.6	42.9	46.3	9,768
ces	2016	65 and over	48.5	47.0	50.0	9,958
ces	2018	65 and over	44.7	43.4	46.1	11,250
ces	2020	65 and over	45.3	44.0	46.6	11,540
ces	2022	65 and over	42.6	41.4	43.9	13,065
ces	2024	65 and over	46.4	45.2	47.6	15,675
anes	1952	18 to 29	65.1	59.6	70.7	284
anes	1956	18 to 29	60.8	54.8	66.7	260
anes	1960	18 to 29	n/a	n/a	n/a	n/a
anes	1964	18 to 29	70.9	65.3	76.5	254
anes	1968	18 to 29	61.7	55.4	68.0	230
anes	1972	18 to 29	63.0	59.1	66.9	589
anes	1976	18 to 29	65.1	60.7	69.5	461
anes	1980	18 to 29	59.8	54.5	65.0	333
anes	1984	18 to 29	56.3	51.8	60.9	458
anes	1988	18 to 29	50.7	45.6	55.7	375
anes	1992	18 to 29	54.4	49.5	59.3	399
anes	1996	18 to 29	60.1	53.1	67.2	225
anes	2000	18 to 29	61.7	54.6	68.8	225
anes	2004	18 to 29	66.4	59.3	73.5	212
anes	2008	18 to 29	66.1	59.7	72.4	342
anes	2012	18 to 29	57.5	52.7	62.2	774
anes	2016	18 to 29	59.9	54.9	64.9	540
anes	2020	18 to 29	63.9	59.4	68.4	841
anes	2024	18 to 29	55.9	50.3	61.6	493
anes	1952	30 to 44	67.4	63.6	71.2	589
anes	1956	30 to 44	61.4	57.4	65.3	590
anes	1960	30 to 44	62.4	56.6	68.3	363
anes	1964	30 to 44	68.2	63.9	72.6	444
anes	1968	30 to 44	65.5	60.8	70.1	411
anes	1972	30 to 44	58.9	54.9	63.0	565
anes	1976	30 to 44	58.9	54.2	63.5	453
anes	1980	30 to 44	61.8	57.0	66.7	380
anes	1984	30 to 44	50.7	46.7	54.6	606
anes	1988	30 to 44	52.5	48.5	56.4	612
anes	1992	30 to 44	56.0	52.4	59.6	748
anes	1996	30 to 44	55.9	51.3	60.4	548
anes	2000	30 to 44	53.0	48.3	57.7	519
anes	2004	30 to 44	49.8	43.4	56.2	273
anes	2008	30 to 44	58.4	53.1	63.8	540
anes	2012	30 to 44	55.0	51.3	58.8	1,142
anes	2016	30 to 44	60.5	56.7	64.4	876
anes	2020	30 to 44	55.3	52.3	58.3	1,739
anes	2024	30 to 44	51.7	47.6	55.7	1,190
anes	1952	45 to 64	58.4	54.1	62.7	510
anes	1956	45 to 64	52.0	47.6	56.5	488
anes	1960	45 to 64	57.4	51.6	63.1	388
anes	1964	45 to 64	69.5	65.4	73.6	495
anes	1968	45 to 64	61.6	57.2	65.9	484
anes	1972	45 to 64	60.9	57.4	64.4	739
anes	1976	45 to 64	60.9	56.9	65.0	588
anes	1980	45 to 64	61.9	57.1	66.7	399
anes	1984	45 to 64	59.8	55.5	64.0	507
anes	1988	45 to 64	55.5	50.9	60.0	456
anes	1992	45 to 64	57.6	53.5	61.6	574
anes	1996	45 to 64	56.4	51.4	61.4	451
anes	2000	45 to 64	58.0	53.4	62.6	536
anes	2004	45 to 64	50.9	45.6	56.2	408
anes	2008	45 to 64	55.6	51.2	60.1	730
anes	2012	45 to 64	53.6	50.7	56.5	2,052
anes	2016	45 to 64	49.5	46.3	52.7	1,306
anes	2020	45 to 64	47.5	44.9	50.1	2,399
anes	2024	45 to 64	48.9	45.3	52.4	1,559
anes	1952	65 and over	55.0	48.1	61.8	202
anes	1956	65 and over	52.2	44.9	59.6	178
anes	1960	65 and over	49.6	41.2	58.1	184
anes	1964	65 and over	53.2	46.6	59.8	218
anes	1968	65 and over	61.5	55.3	67.7	239
anes	1972	65 and over	55.5	50.5	60.4	384
anes	1976	65 and over	57.9	52.8	63.1	364
anes	1980	65 and over	62.4	56.4	68.3	255
anes	1984	65 and over	51.5	46.3	56.7	357
anes	1988	65 and over	55.9	50.6	61.3	329
anes	1992	65 and over	60.5	55.9	65.1	433
anes	1996	65 and over	59.5	53.7	65.3	323
anes	2000	65 and over	60.0	53.7	66.3	280
anes	2004	65 and over	57.5	49.5	65.5	181
anes	2008	65 and over	53.0	46.7	59.3	369
anes	2012	65 and over	50.4	46.5	54.4	1,079
anes	2016	65 and over	49.7	45.8	53.7	855
anes	2020	65 and over	49.4	46.5	52.3	2,042
anes	2024	65 and over	54.0	50.5	57.5	1,617

The line for adults aged eighteen to twenty-nine stays above the line for adults sixty-five and older in all ten waves, by +13.9 to +21.3 points, and the gap is lowest in 2024. Each line is the weighted Democratic share among respondents who identify with a party or lean toward one, and the bands are 95% intervals. The ANES button shows the same measure from the smaller survey.Note: A cell is plotted only when it has 100 or more respondents and a Kish effective sample size (the sample size after weighting) of 100 or more, the floor of this chapter. A blank cell leaves a gap in the line. Party identification counts Democrats and Republicans with leaners included and leaves out pure independents.Note: The ANES estimates for 2012, 2016 and 2024 mix face-to-face and web interviews and the 2020 study has no interview-mode weights. In 2024 the web sample alone gives a gap of +1.1.Note: The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel. Intervals are model based: they treat the weighted sample as a simple random sample and ignore clustering and stratification.Source: CES Cumulative File, even-year waves 2006-2024; weighted (weight_cumulative); party identifiers, leaners included; the ANES view: ANES Cumulative Data File, presidential-election years 1952-2024; weighted (VCF0009z); party identifiers, leaners included

The CES therefore shows no detectable fall in party identification against 2020 or 2018, and a fall against 2022. The CES gap moved by several points between adjacent waves, so a fall measured against one wave is not a trend. The ANES shows a fall in both the vote and party identification against 2020, from a sample whose oldest group leans Democratic relative to the CES, as the previous section reports. An account in which party attachment strengthens with the years a person has held it gives age a role of its own in party identification [4], and this chapter does not test it.

Two cohorts that began with a large Democratic lead held a smaller one by 2024

A birth cohort follows the same group of people as far as a survey can. Figure 5 follows each cohort of the ANES cumulative file, in the sixteen-year birth groups that Chapter 1.4, Cohort change in partisan sorting uses, through the elections in which the cohort has enough voters. It plots the cohort's Democratic share minus the share among voters born in other years in the same election. The subtraction removes anything that moves all voters together in one election, such as a landslide, and it leaves age and the changing make-up of the other voters in place. Cohort cells include every voter with a known cohort, whether or not the age is known.

The cohort born 1975 to 1990 voted 66.0% (95% CI 57.3% to 74.7%) Democratic in 2004, +19.7 (95% CI 10.1 to 29.3) points above other voters, and 55.6% (95% CI 51.3% to 60.0%) in 2024, +1.5 (95% CI −3.7 to 6.6) points above them, a change in its lead of −18.2 (95% CI −29.1 to −7.3) points. The cohort born 1943 to 1958 led by +15.6 (95% CI 10.1 to 21.2) points in 1972 and by +2.7 (95% CI −2.1 to 7.5) in 2024, a change of −12.9 (95% CI −20.3 to −5.5) points (Figure 6). The CES, which starts in 2008, shows the same direction for the 1975 to 1990 cohort: its share fell from 66.3% (95% CI 64.6% to 68.0%) in 2008 to 54.6% (95% CI 53.0% to 56.2%) in 2024, and its lead was +16.6 (95% CI 14.7 to 18.5) points in 2008 and +7.5 (95% CI 5.7 to 9.3) in 2024, a change of −9.1 (95% CI −11.7 to −6.5) points.

Figure 5

The cohorts born 1943 to 1958 and 1975 to 1990 began with a Democratic lead over other voters in the ANES and held a smaller one in 2024

Democratic share of the two-party presidential vote among voters in each birth cohort minus the share among voters born in other years, points, ANES 1952 to 2024; 95% intervals; the buttons switch survey and measure

Show data

Data for The cohorts born 1943 to 1958 and 1975 to 1990 began with a Democratic lead over other voters in the ANES and held a smaller one in 2024
Survey	Birth cohort	Election	Democratic share (%)	95% interval low	95% interval high	Lead over other voters (points)	95% interval low (lead)	95% interval high (lead)	Voters
anes	Born 1895 to 1910	1952	38.9	33.9	43.8	−4.6	−10.7	+1.4	373
anes	Born 1895 to 1910	1956	39.3	34.2	44.4	−1.4	−7.5	+4.6	354
anes	Born 1895 to 1910	1960	49.5	42.4	56.6	+0.4	−8.1	+8.9	261
anes	Born 1895 to 1910	1964	65.9	60.1	71.7	−2.1	−8.6	+4.5	261
anes	Born 1895 to 1910	1968	46.1	39.0	53.2	−0.2	−8.2	+7.7	191
anes	Born 1895 to 1910	1972	32.0	26.3	37.8	−4.5	−10.8	+1.9	253
anes	Born 1895 to 1910	1976	52.3	45.2	59.3	+1.4	−6.3	+9.0	202
anes	Born 1895 to 1910	1980	50.0	40.2	59.8	+7.1	−3.3	+17.4	102
anes	Born 1911 to 1926	1952	46.8	42.2	51.4	+7.8	+2.0	+13.6	453
anes	Born 1911 to 1926	1956	41.1	36.8	45.3	+1.4	−4.2	+6.9	516
anes	Born 1911 to 1926	1960	51.2	45.2	57.1	+3.3	−4.6	+11.1	372
anes	Born 1911 to 1926	1964	68.6	64.0	73.1	+1.7	−4.0	+7.4	401
anes	Born 1911 to 1926	1968	43.1	37.6	48.6	−4.8	−11.6	+2.0	313
anes	Born 1911 to 1926	1972	29.6	25.3	33.9	−8.4	−13.6	−3.3	432
anes	Born 1911 to 1926	1976	53.1	47.9	58.4	+2.7	−3.4	+8.9	363
anes	Born 1911 to 1926	1980	46.9	40.3	53.4	+4.2	−3.4	+11.7	224
anes	Born 1911 to 1926	1984	44.4	38.5	50.2	+3.2	−3.4	+9.8	275
anes	Born 1911 to 1926	1988	44.3	37.8	50.8	−3.5	−10.7	+3.7	228
anes	Born 1911 to 1926	1992	61.9	55.7	68.1	+4.1	−2.8	+11.0	237
anes	Born 1911 to 1926	1996	64.0	56.0	72.0	+6.5	−2.4	+15.3	163
anes	Born 1927 to 1942	1956	42.1	34.9	49.4	+2.2	−5.7	+10.0	178
anes	Born 1927 to 1942	1960	n/a	n/a	n/a	n/a	n/a	n/a	n/a
anes	Born 1927 to 1942	1964	70.5	65.6	75.3	+4.3	−1.6	+10.2	342
anes	Born 1927 to 1942	1968	50.4	44.5	56.2	+6.0	−1.0	+13.0	284
anes	Born 1927 to 1942	1972	34.8	30.3	39.2	−1.4	−6.6	+3.8	446
anes	Born 1927 to 1942	1976	47.2	41.9	52.6	−5.3	−11.5	+1.0	351
anes	Born 1927 to 1942	1980	38.4	32.0	44.8	−7.2	−14.7	+0.2	224
anes	Born 1927 to 1942	1984	40.4	35.0	45.8	−1.8	−8.0	+4.4	319
anes	Born 1927 to 1942	1988	48.9	43.0	54.8	+2.3	−4.4	+9.1	276
anes	Born 1927 to 1942	1992	56.0	50.3	61.8	−3.2	−9.6	+3.3	289
anes	Born 1927 to 1942	1996	55.9	49.0	62.9	−3.1	−11.0	+4.8	230
anes	Born 1927 to 1942	2000	48.4	41.2	55.6	−4.7	−12.8	+3.4	226
anes	Born 1927 to 1942	2004	50.5	42.3	58.8	+0.9	−8.4	+10.1	170
anes	Born 1927 to 1942	2008	41.2	33.1	49.2	−16.4	−25.1	−7.7	224
anes	Born 1927 to 1942	2012	43.8	38.1	49.4	−11.2	−17.3	−5.2	520
anes	Born 1927 to 1942	2016	40.9	33.4	48.3	−13.3	−21.1	−5.4	256
anes	Born 1927 to 1942	2020	47.5	40.9	54.2	−9.2	−16.1	−2.3	415
anes	Born 1927 to 1942	2024	59.0	49.3	68.6	+4.6	−5.3	+14.6	204
anes	Born 1943 to 1958	1972	47.4	42.5	52.3	+15.6	+10.1	+21.2	403
anes	Born 1943 to 1958	1976	53.0	47.8	58.1	+2.7	−3.4	+8.8	378
anes	Born 1943 to 1958	1980	41.3	35.5	47.1	−3.6	−10.7	+3.5	276
anes	Born 1943 to 1958	1984	41.0	36.8	45.2	−1.3	−6.7	+4.1	529
anes	Born 1943 to 1958	1988	47.0	42.4	51.6	−0.2	−6.0	+5.7	449
anes	Born 1943 to 1958	1992	58.5	53.9	63.0	−0.1	−5.6	+5.5	453
anes	Born 1943 to 1958	1996	58.0	52.2	63.9	−0.5	−7.7	+6.6	336
anes	Born 1943 to 1958	2000	53.3	47.8	58.8	+1.6	−5.3	+8.4	380
anes	Born 1943 to 1958	2004	44.0	37.3	50.7	−8.3	−16.4	−0.2	259
anes	Born 1943 to 1958	2008	53.5	47.8	59.3	−2.2	−9.0	+4.7	451
anes	Born 1943 to 1958	2012	50.6	47.2	54.1	−4.3	−8.5	0.0	1,481
anes	Born 1943 to 1958	2016	48.1	44.0	52.2	−6.8	−11.8	−1.8	802
anes	Born 1943 to 1958	2020	53.9	50.7	57.1	−3.1	−6.9	+0.8	1,835
anes	Born 1943 to 1958	2024	56.7	52.7	60.7	+2.7	−2.1	+7.5	1,155
anes	Born 1959 to 1974	1984	42.4	34.5	50.3	+0.6	−7.7	+9.0	151
anes	Born 1959 to 1974	1988	48.3	40.9	55.7	+1.4	−6.6	+9.4	176
anes	Born 1959 to 1974	1992	58.5	53.1	63.8	−0.1	−6.2	+6.1	330
anes	Born 1959 to 1974	1996	57.6	51.0	64.2	−1.1	−8.8	+6.6	263
anes	Born 1959 to 1974	2000	46.5	40.6	52.4	−8.2	−15.2	−1.1	328
anes	Born 1959 to 1974	2004	45.2	38.0	52.4	−6.5	−14.9	+2.0	211
anes	Born 1959 to 1974	2008	55.7	50.0	61.3	+0.8	−6.0	+7.6	462
anes	Born 1959 to 1974	2012	53.9	50.1	57.7	+0.4	−4.1	+4.9	1,171
anes	Born 1959 to 1974	2016	49.8	45.3	54.2	−4.4	−9.7	+0.8	664
anes	Born 1959 to 1974	2020	50.2	46.9	53.6	−8.5	−12.4	−4.5	1,561
anes	Born 1959 to 1974	2024	50.6	46.4	54.8	−5.5	−10.5	−0.5	1,061
anes	Born 1975 to 1990	2004	66.0	57.3	74.7	+19.7	+10.1	+29.3	137
anes	Born 1975 to 1990	2008	64.9	58.5	71.4	+13.1	+5.7	+20.4	337
anes	Born 1975 to 1990	2012	60.0	55.5	64.6	+8.5	+3.4	+13.5	827
anes	Born 1975 to 1990	2016	62.8	58.2	67.4	+12.7	+7.4	+18.1	604
anes	Born 1975 to 1990	2020	58.4	55.1	61.8	+2.9	−1.0	+6.9	1,454
anes	Born 1975 to 1990	2024	55.6	51.3	60.0	+1.5	−3.7	+6.6	964
anes	Born 1991 or later	2016	62.9	54.2	71.6	+11.2	+2.1	+20.2	177
anes	Born 1991 or later	2020	69.6	64.3	74.9	+15.9	+10.4	+21.5	573
anes	Born 1991 or later	2024	55.1	49.0	61.1	+0.7	−5.9	+7.2	452
ces	Born 1927 to 1942	2008	42.2	40.3	44.0	−13.8	−15.8	−11.7	3,830
ces	Born 1927 to 1942	2012	40.7	38.9	42.5	−13.8	−15.8	−11.8	5,912
ces	Born 1927 to 1942	2016	38.0	35.6	40.5	−14.1	−16.7	−11.6	3,561
ces	Born 1927 to 1942	2020	36.8	34.2	39.4	−16.6	−19.3	−14.0	2,320
ces	Born 1927 to 1942	2024	35.4	31.8	39.0	−13.9	−17.5	−10.2	1,150
ces	Born 1943 to 1958	2008	50.8	49.5	52.1	−4.3	−5.9	−2.7	7,691
ces	Born 1943 to 1958	2012	49.2	48.2	50.2	−5.9	−7.3	−4.4	17,779
ces	Born 1943 to 1958	2016	46.4	45.2	47.5	−6.9	−8.3	−5.5	13,874
ces	Born 1943 to 1958	2020	48.1	46.9	49.3	−6.4	−7.8	−5.0	12,855
ces	Born 1943 to 1958	2024	46.3	45.1	47.6	−3.5	−5.1	−2.0	12,790
ces	Born 1959 to 1974	2008	52.7	51.4	54.1	−1.7	−3.3	0.0	7,281
ces	Born 1959 to 1974	2012	53.4	51.9	54.8	+0.4	−1.3	+2.1	8,300
ces	Born 1959 to 1974	2016	48.2	47.0	49.3	−3.9	−5.3	−2.5	12,150
ces	Born 1959 to 1974	2020	48.9	47.8	49.9	−5.3	−6.6	−4.0	13,938
ces	Born 1959 to 1974	2024	42.9	41.7	44.1	−9.0	−10.5	−7.5	15,336
ces	Born 1975 to 1990	2008	66.3	64.6	68.0	+16.6	+14.7	+18.5	4,555
ces	Born 1975 to 1990	2012	62.5	60.6	64.4	+12.4	+10.3	+14.4	4,875
ces	Born 1975 to 1990	2016	60.3	59.0	61.6	+12.2	+10.7	+13.7	9,329
ces	Born 1975 to 1990	2020	61.3	60.1	62.5	+11.3	+9.9	+12.7	10,393
ces	Born 1975 to 1990	2024	54.6	53.0	56.2	+7.5	+5.7	+9.3	8,460
ces	Born 1991 or later	2012	58.8	53.8	63.9	+6.0	+0.9	+11.2	758
ces	Born 1991 or later	2016	67.7	65.0	70.4	+18.0	+15.2	+20.8	1,925
ces	Born 1991 or later	2020	67.8	65.9	69.7	+16.9	+14.8	+18.9	4,360
ces	Born 1991 or later	2024	59.0	56.5	61.6	+11.9	+9.3	+14.6	4,272

The 1943 to 1958 cohort led other voters by +15.6 points in 1972 and by +2.7 in 2024, and the 1975 to 1990 cohort by +19.7 points in 2004 and +1.5 in 2024. The cohort born 1911 to 1926 has a lead of similar size in its first cell, in 1952, and its last, in 1996. A cohort's first points include only its members old enough to vote. A line is drawn from the first election in which the cohort has enough voters, and bands are 95% intervals.Note: Cohorts are the 16-year birth groups of the ANES cumulative file, and the CES uses the same groups from reported birth year. Cohort cells include every voter with a known cohort, whether or not the age is known. A cell is plotted only when it has 100 or more respondents and a Kish effective sample size (the sample size after weighting) of 100 or more, the floor of this chapter. A blank cell leaves a gap in the line.Note: The comparison group is every other voter in the same election with a known cohort, so the cohort and the comparison group are different respondents. The measure removes anything that moves all voters together in an election, and it does not remove the effect of age or the changing make-up of the comparison group. The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel.Source: ANES Cumulative Data File, presidential elections 1952-2024; weighted (post-election weight where the file has one, otherwise VCF0009z); two-party presidential voters; CES view: CES Cumulative File, presidential elections 2008-2024; weighted (weight_cumulative); two-party presidential voters

The pattern is not general. The cohort born 1927 to 1942 was below other voters in the ANES in every election from 2008 to 2020, and its 2024 lead, +4.6 (95% CI −5.3 to 14.6) points, is above them, so its change between 1956 and 2024 is +2.5 (95% CI −10.2 to 15.1) points, an interval that includes zero. In the CES the same cohort's 2024 lead is −13.9 (95% CI −17.5 to −10.2) points, so the two surveys disagree about the oldest voters, as the 2024 section reports. The cohort born 1911 to 1926 led other voters by +7.8 (95% CI 2.0 to 13.6) points in 1952 and by +6.5 (95% CI −2.4 to 15.3) in 1996, a change of −1.4 (95% CI −11.9 to 9.2) points. That interval does not show a loss of the cohort's lead between 1952 and 1996, and its lower end is close to the change for the 1943 to 1958 cohort. The cohort born 1959 to 1974 began near the average, at +0.6 (95% CI −7.7 to 9.0) points in 1984, and ended below it, at −5.5 (95% CI −10.5 to −0.5) in 2024. In the CES the lead of that cohort changed by −7.3 (95% CI −9.6 to −5.1) points between 2008 and 2024 and the lead of the cohort born 1943 to 1958 by +0.8 (95% CI −1.5 to 3.0). The lead of the cohort born 1991 or later changed by −10.5 (95% CI −21.7 to 0.6) points in the ANES between 2016 and 2024, an interval that includes zero, and by +5.9 (95% CI 0.1 to 11.7) points in the CES between 2012 and 2024, from a first cell that holds only voters aged eighteen to twenty-one.

Party identification moves in the CES, where the cohort samples are large. The party-identification leads of 4 of the 5 cohorts changed by more than sampling error between their first and last waves: by −9.6 (95% CI −13.6 to −5.6) points for the cohort born 1927 to 1942, −7.2 (95% CI −9.2 to −5.2) for 1959 to 1974, −4.8 (95% CI −7.2 to −2.4) for 1975 to 1990 and −7.3 (95% CI −13.9 to −0.7) for 1991 or later. The change for the cohort born 1943 to 1958, −1.7 (95% CI −3.8 to 0.3), includes zero. The cohort born 1991 or later therefore lost party-identification lead while its CES vote lead rose, and the two measures do not agree. No ANES cohort's lead in party identification changed by more than sampling error, and those intervals are wide.

Figure 6

The vote lead of the 1975 to 1990 cohort over other voters changed by −18.2 points in the ANES between 2004 and 2024 and by −9.1 in the CES between 2008 and 2024

Democratic share of a cohort's two-party voters minus the share among voters born in other years, first and last cell with enough voters; ANES presidential elections 1952 to 2024; 95% intervals; the buttons switch survey and measure

First cell
Last cell

Show data

Data for The vote lead of the 1975 to 1990 cohort over other voters changed by −18.2 points in the ANES between 2004 and 2024 and by −9.1 in the CES between 2008 and 2024
Measure	Birth cohort and cells	Lead in first cell (points)	95% interval low (first)	95% interval high (first)	Lead in last cell (points)	95% interval low (last)	95% interval high (last)	Change (points)	Respondents, first cell	Respondents, last cell
anes_vote	Born 1895 to 1910 (1952 to 1980)	−4.6	−10.7	+1.4	+7.1	−3.3	+17.4	+11.7	373	102
anes_vote	Born 1911 to 1926 (1952 to 1996)	+7.8	+2.0	+13.6	+6.5	−2.4	+15.3	−1.4	453	163
anes_vote	Born 1927 to 1942 (1956 to 2024)	+2.2	−5.7	+10.0	+4.6	−5.3	+14.6	+2.5	178	204
anes_vote	Born 1943 to 1958 (1972 to 2024)	+15.6	+10.1	+21.2	+2.7	−2.1	+7.5	−12.9	403	1,155
anes_vote	Born 1959 to 1974 (1984 to 2024)	+0.6	−7.7	+9.0	−5.5	−10.5	−0.5	−6.1	151	1,061
anes_vote	Born 1975 to 1990 (2004 to 2024)	+19.7	+10.1	+29.3	+1.5	−3.7	+6.6	−18.2	137	964
anes_vote	Born 1991 or later (2016 to 2024)	+11.2	+2.1	+20.2	+0.7	−5.9	+7.2	−10.5	177	452
ces_vote	Born 1927 to 1942 (2008 to 2024)	−13.8	−15.8	−11.7	−13.9	−17.5	−10.2	−0.1	3,830	1,150
ces_vote	Born 1943 to 1958 (2008 to 2024)	−4.3	−5.9	−2.7	−3.5	−5.1	−2.0	+0.8	7,691	12,790
ces_vote	Born 1959 to 1974 (2008 to 2024)	−1.7	−3.3	0.0	−9.0	−10.5	−7.5	−7.3	7,281	15,336
ces_vote	Born 1975 to 1990 (2008 to 2024)	+16.6	+14.7	+18.5	+7.5	+5.7	+9.3	−9.1	4,555	8,460
ces_vote	Born 1991 or later (2012 to 2024)	+6.0	+0.9	+11.2	+11.9	+9.3	+14.6	+5.9	758	4,272
anes_pid	Born 1895 to 1910 (1952 to 1988)	−1.7	−6.9	+3.5	−0.5	−10.6	+9.6	+1.2	484	100
anes_pid	Born 1911 to 1926 (1952 to 2000)	+6.7	+1.8	+11.5	+11.2	+2.2	+20.3	+4.6	625	141
anes_pid	Born 1927 to 1942 (1952 to 2024)	+5.9	−2.4	+14.2	+4.6	−4.7	+13.9	−1.3	134	240
anes_pid	Born 1943 to 1958 (1968 to 2024)	+4.6	−4.2	+13.4	+1.2	−3.3	+5.7	−3.4	121	1,339
anes_pid	Born 1959 to 1974 (1984 to 2024)	−0.2	−6.5	+6.1	−5.9	−10.4	−1.4	−5.7	285	1,324
anes_pid	Born 1975 to 1990 (2000 to 2024)	+8.6	−0.8	+18.0	−0.3	−4.8	+4.3	−8.8	135	1,301
anes_pid	Born 1991 or later (2012 to 2024)	+10.2	+0.7	+19.6	+4.2	−1.0	+9.4	−6.0	189	765
ces_pid	Born 1927 to 1942 (2006 to 2024)	−6.2	−7.7	−4.7	−15.8	−19.5	−12.0	−9.6	5,626	1,178
ces_pid	Born 1943 to 1958 (2006 to 2024)	−3.4	−4.7	−2.1	−5.2	−6.7	−3.6	−1.7	11,395	13,339
ces_pid	Born 1959 to 1974 (2006 to 2024)	−2.6	−3.9	−1.3	−9.8	−11.3	−8.2	−7.2	9,611	16,895
ces_pid	Born 1975 to 1990 (2006 to 2024)	+12.1	+10.5	+13.7	+7.3	+5.5	+9.0	−4.8	4,631	10,549
ces_pid	Born 1991 or later (2010 to 2024)	+18.7	+12.4	+25.0	+11.4	+9.2	+13.6	−7.3	302	8,886

The ring is a cohort's lead in the first election with enough voters and the dot is its lead in the last. The lead of the 1943 to 1958 cohort changed by −12.9 points and that of the cohort born 1911 to 1926 by −1.4, an interval that includes zero. Each end is the difference of two weighted shares, and whiskers are 95% intervals.Note: A cohort's first cell holds only the members old enough to vote at that election, and the first cell of an older cohort is often long after it entered the electorate. The row label gives the years of the two cells.Note: The two ends of a row are different respondents, and the interval of the change combines their standard errors. Intervals are model based: they treat the weighted sample as a simple random sample and ignore clustering and stratification.Source: ANES Cumulative Data File, presidential elections 1952-2024; weighted (post-election weight where the file has one, otherwise VCF0009z); two-party presidential voters; other views: CES and party identification as named on the buttons

These comparisons follow a cohort, so they mix age with period. A cohort that is more Democratic than other voters when young and closer to them when older fits an age effect, in which people become more like other voters as they age. It fits equally well a cohort effect that fades, a gap between the first cell, which holds only the cohort's members old enough to vote, and the last cell, which holds all of them, and an election in which the young moved more than the old. Ghitza, Gelman and Auerbach model presidential voting as the cumulative impression of political events, which count most in a voter's teenage and early adult years, and divide voters into five generations [5]. The model needs a measured series of events, the Gallup presidential approval series, and an assumed profile of impressionability by age, and this chapter supplies neither.

The age gap by gender and race

Figure 7 splits the CES age gap by gender and race, pooling 2008 with 2012 and 2020 with 2024 so that every cell clears the floor. The gap among men fell from +27.0 (95% CI 24.1 to 29.8) to +19.9 (95% CI 16.6 to 23.1) points, a change of −7.1 (95% CI −11.4 to −2.8), and the gap among women went from +17.3 (95% CI 14.9 to 19.7) to +16.1 (95% CI 13.4 to 18.7), a change of −1.3 (95% CI −4.9 to 2.3) that includes zero. The gap among women minus the gap among men was −9.6 (95% CI −13.4 to −5.9) points in the earlier period and −3.8 (95% CI −8.0 to 0.4) in the later one, so men's age gap was larger in 2008 and 2012 and is no longer clearly larger. Chapter 3.5, The gender gap in attitudes and votes compares women and men within the youngest group of voters, and this chapter compares young and old within each gender.

Race separates the age gaps more than gender does. Among white voters the gap was +20.2 (95% CI 17.7 to 22.7) points in 2020 and 2024 pooled. Among Black voters it was −12.3 (95% CI −17.7 to −6.8) points, so older Black voters were more Democratic than younger Black voters, and the white gap minus the Black gap is +32.5 (95% CI 26.5 to 38.5) points. Among Hispanic voters the gap fell from +22.1 (95% CI 12.9 to 31.2) to +5.3 (95% CI −1.7 to 12.3) points, a change of −16.8 (95% CI −28.3 to −5.2), and the later interval includes zero. Asian voters under thirty are too few to publish a gap for 2008 and 2012, and for 2020 and 2024 pooled the gap is +12.9 (95% CI 2.5 to 23.4) points, from 142 effective respondents under thirty. In 2024 alone the effective sample of Black voters under thirty is 95, below the floor, so the pooled estimates are the ones reported. Chapter 3.4, Hispanic, Black and Asian voters (planned) follows how each of these groups voted, and this chapter compares the young and old within each.

Figure 7

In 2020 and 2024 the age gap in the vote was +20.2 points among white voters and −12.3 among Black voters

Democratic share of two-party voters aged 18 to 29 minus the share among voters aged 65 and over, by gender and race, CES 2008 and 2012 pooled and 2020 and 2024 pooled; 95% intervals

2008 and 2012
2020 and 2024

Show data

Data for In 2020 and 2024 the age gap in the vote was +20.2 points among white voters and −12.3 among Black voters
Voters	Gap 2008 and 2012 (points)	95% interval low	95% interval high	Respondents, early	Gap 2020 and 2024 (points)	95% interval low (late)	95% interval high (late)	Respondents, late	Change (points)
All voters	+21.6	+19.7	+23.5	22,508	+18.1	+16.0	+20.2	34,342	−3.5
Women	+17.3	+14.9	+19.7	11,936	+16.1	+13.4	+18.7	18,877	−1.3
Men	+27.0	+24.1	+29.8	10,572	+19.9	+16.6	+23.1	15,342	−7.1
White	+18.0	+15.9	+20.2	19,158	+20.2	+17.7	+22.7	26,464	+2.1
Black	+0.6	−2.8	+4.0	1,153	−12.3	−17.7	−6.8	3,055	−12.9
Hispanic	+22.1	+12.9	+31.2	1,208	+5.3	−1.7	+12.3	2,555	−16.8

Black voters aged sixty-five and older were more Democratic than Black voters under thirty in 2020 and 2024, the reverse of the pattern among white voters. The gap among men changed by −7.1 points and the change among women, −1.3, includes zero. The ring is 2008 and 2012 pooled and the dot is 2020 and 2024 pooled. Each end is the difference of two weighted shares, and whiskers are 95% intervals.Note: Pooling two elections gives every cell the floor of 100 respondents and an effective sample of 100. Asian voters fall below it in the earlier period and are left out. In 2020 and 2024 pooled their gap is +12.9.Note: Race is the CES single-race category with Hispanic voters of any race as a separate group, and the groups overlap with the gender rows. The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel. Intervals are model based: they treat the weighted sample as a simple random sample and ignore clustering and stratification.Source: CES Cumulative File, presidential elections 2008-2024; weighted (weight_cumulative); two-party presidential voters; elections pooled in two groups

The estimated age effect depends on which of three assumptions holds

Figure 8 estimates the three comparisons as straight-line slopes, in points of Democratic vote share per decade, on voters aged eighteen to seventy-three. In the CES, one decade of age within an election goes with −4.3 (95% CI −4.6 to −4.1) points, one decade of time at a fixed age with −1.2 (95% CI −1.8 to −0.5), and one decade along a cohort with −5.2 (95% CI −5.9 to −4.6). In the ANES the slopes are −2.2 (95% CI −2.6 to −1.7), +1.6 (95% CI 1.3 to 1.9) and +0.3 (95% CI −0.3 to 0.8), and only the first two intervals exclude zero. When the effects are straight lines the third slope equals the sum of the first two. The sum of the first two minus the third is −0.8 (95% CI −1.0 to −0.6) in the ANES and −0.3 (95% CI −0.5 to −0.1) in the CES (bootstrap intervals), and both intervals exclude zero, so the data do not support effects that are exactly straight lines.

Figure 8

In the CES the vote moved by −4.3 points per decade of age within an election, −1.2 per decade at a fixed age and −5.2 along a cohort

Change in the Democratic share of the two-party vote per decade, three straight-line comparisons, voters aged 18 to 73, ANES 1952 to 2024 and CES 2008 to 2024; 95% intervals (HC3)

ANES 1952 to 2024
CES 2008 to 2024

Show data

Data for In the CES the vote moved by −4.3 points per decade of age within an election, −1.2 per decade at a fixed age and −5.2 along a cohort
Comparison	Survey	Change (points per decade)	95% interval low	95% interval high	Voters aged 18 to 73
One decade older, same election (age and birth year)	ANES 1952 to 2024	−2.2	−2.6	−1.7	30,571
One decade older, same election (age and birth year)	CES 2008 to 2024	−4.3	−4.6	−4.1	169,858
One decade later, same age (election and birth year)	ANES 1952 to 2024	+1.6	+1.3	+1.9	30,571
One decade later, same age (election and birth year)	CES 2008 to 2024	−1.2	−1.8	−0.5	169,858
One decade later, same birth cohort (age and election)	ANES 1952 to 2024	+0.3	−0.3	+0.8	30,571
One decade later, same birth cohort (age and election)	CES 2008 to 2024	−5.2	−5.9	−4.6	169,858

The three comparisons give different answers from the same voters: in the ANES the slopes are −2.2, +1.6 and +0.3, and the third interval includes zero. The sum of the first two slopes minus the third is −0.8 in the ANES and −0.3 in the CES, with bootstrap intervals that exclude zero, and it would be zero if the three effects were straight lines. Each dot is a weighted regression coefficient and whiskers are 95% intervals.Note: Cross-section: weighted regression of the 0/1 Democratic vote on age in decades with a fixed effect for each election. Same age: regression on election year in decades with a fixed effect for each 4-year age band. Same cohort: regression on election year in decades with a fixed effect for each 4-year birth band. Voters aged 18 to 73 only, so the age and birth bands share one grid.Note: The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel.Source: ANES Cumulative Data File, presidential elections 1952-2024; weighted (post-election weight where the file has one, otherwise VCF0009z); two-party presidential voters; CES Cumulative File, presidential elections 2008-2024; weighted (weight_cumulative); two-party presidential voters; weighted regression coefficients

If the effects are straight lines, each assumption turns these slopes into an age slope. If birth cohorts have no trend, the age slope is the first slope, −4.3 points per decade in the CES. If elections have no trend, the age slope is the third slope, −5.2 in the CES and +0.3 in the ANES. If age has no trend, the age slope is zero, and the whole difference between older and younger voters inside an election is a difference between birth cohorts. The age slope in the ANES therefore runs between −2.2 and +0.3 points per decade depending on the assumption, and in the CES between −5.2 and zero.

A model that gives every age band, election and birth cohort its own effect shows the same dependence in full (Figure 9). It is a linear probability model that gives each of 14 four-year age bands, each presidential election and each four-year birth cohort an effect on the Democratic share, fitted to the share in every age-by-election cell. It cannot be estimated without a constraint, because adding a straight line to the age effects and to the cohort effects, and subtracting a straight line from the election effects, leaves every fitted share unchanged. Three constraints, each setting the straight-line trend of one set of effects to zero, give profiles that fit the data identically and differ only by a tilt. The span of a profile is the effect of the oldest band minus the effect of the youngest. It combines the slope with the bends in the profile, so a profile whose age slope is zero still has a span that is not exactly zero. In the CES the span is −22.2 (95% CI −25.0 to −19.2) points if birth cohorts have no trend, −27.3 (95% CI −31.5 to −23.3) if elections have none and +1.0 (95% CI −0.8 to 3.1) if ages have none. In the ANES the three spans are −13.3 (95% CI −18.5 to −8.7), −4.8 (95% CI −10.3 to 0.0) and −3.8 (95% CI −7.8 to −0.2). Where each profile bends is the same under all three constraints and is identified from the data, and the overall slope is not.

Figure 9

In the CES the age effect from the youngest to the oldest band is −22.2 points if birth years carry no trend, −27.3 if elections carry none and +1.0 if ages carry none

Age effect on the Democratic share of the two-party vote, points, 4-year age bands from 18 to 73, from a model with an effect for every age band, election and birth cohort, under three constraints; 95% bootstrap intervals

Age effect on the Democratic share (points)

Show data

Data for In the CES the age effect from the youngest to the oldest band is −22.2 points if birth years carry no trend, −27.3 if elections carry none and +1.0 if ages carry none
Survey	Constraint	Age band midpoint (years)	Age effect (points)	95% bootstrap interval low	95% bootstrap interval high
ces	No trend across birth years	19.5	+11.9	+9.5	+14.2
ces	No trend across birth years	23.5	+11.5	+9.5	+13.4
ces	No trend across birth years	27.5	+8.1	+6.3	+9.8
ces	No trend across birth years	31.5	+6.8	+5.1	+8.5
ces	No trend across birth years	35.5	+3.9	+2.3	+5.6
ces	No trend across birth years	39.5	+2.1	+0.6	+3.5
ces	No trend across birth years	43.5	+0.8	−0.6	+2.2
ces	No trend across birth years	47.5	−1.5	−2.8	−0.2
ces	No trend across birth years	51.5	−4.6	−5.8	−3.3
ces	No trend across birth years	55.5	−6.0	−7.2	−4.7
ces	No trend across birth years	59.5	−7.0	−8.3	−5.8
ces	No trend across birth years	63.5	−7.7	−9.0	−6.4
ces	No trend across birth years	67.5	−8.0	−9.7	−6.6
ces	No trend across birth years	71.5	−10.3	−11.8	−8.6
ces	No trend across elections	19.5	+14.5	+11.6	+17.3
ces	No trend across elections	23.5	+13.6	+11.5	+15.8
ces	No trend across elections	27.5	+9.8	+7.9	+11.8
ces	No trend across elections	31.5	+8.1	+6.4	+10.0
ces	No trend across elections	35.5	+4.9	+3.3	+6.6
ces	No trend across elections	39.5	+2.7	+1.1	+4.1
ces	No trend across elections	43.5	+1.0	−0.4	+2.3
ces	No trend across elections	47.5	−1.7	−3.0	−0.4
ces	No trend across elections	51.5	−5.2	−6.6	−3.8
ces	No trend across elections	55.5	−7.0	−8.3	−5.6
ces	No trend across elections	59.5	−8.4	−9.7	−7.0
ces	No trend across elections	63.5	−9.4	−10.9	−8.0
ces	No trend across elections	67.5	−10.2	−12.1	−8.6
ces	No trend across elections	71.5	−12.8	−14.7	−10.8
ces	No trend across ages	19.5	+0.3	−1.6	+2.2
ces	No trend across ages	23.5	+1.6	0.0	+3.2
ces	No trend across ages	27.5	0.0	−1.4	+1.4
ces	No trend across ages	31.5	+0.5	−0.9	+2.0
ces	No trend across ages	35.5	−0.6	−2.0	+0.9
ces	No trend across ages	39.5	−0.6	−2.0	+0.8
ces	No trend across ages	43.5	−0.1	−1.5	+1.3
ces	No trend across ages	47.5	−0.6	−1.9	+0.7
ces	No trend across ages	51.5	−1.9	−3.1	−0.7
ces	No trend across ages	55.5	−1.5	−2.5	−0.4
ces	No trend across ages	59.5	−0.8	−1.8	+0.2
ces	No trend across ages	63.5	+0.4	−0.5	+1.3
ces	No trend across ages	67.5	+1.8	+0.6	+2.8
ces	No trend across ages	71.5	+1.3	+0.1	+2.7
anes	No trend across birth years	19.5	+10.4	+6.8	+14.2
anes	No trend across birth years	23.5	+4.2	+1.4	+7.2
anes	No trend across birth years	27.5	+3.1	+0.5	+5.9
anes	No trend across birth years	31.5	+0.8	−1.6	+3.2
anes	No trend across birth years	35.5	+0.1	−2.1	+2.4
anes	No trend across birth years	39.5	−2.8	−5.0	−0.7
anes	No trend across birth years	43.5	−1.1	−3.4	+1.2
anes	No trend across birth years	47.5	−0.2	−2.6	+2.2
anes	No trend across birth years	51.5	−0.7	−3.2	+1.5
anes	No trend across birth years	55.5	−3.3	−5.7	−1.1
anes	No trend across birth years	59.5	−2.5	−5.0	−0.2
anes	No trend across birth years	63.5	−3.3	−5.7	−0.9
anes	No trend across birth years	67.5	−1.9	−4.6	+0.6
anes	No trend across birth years	71.5	−2.9	−5.6	−0.1
anes	No trend across elections	19.5	+6.2	+2.4	+10.3
anes	No trend across elections	23.5	+0.6	−2.2	+3.6
anes	No trend across elections	27.5	+0.2	−2.3	+2.9
anes	No trend across elections	31.5	−1.5	−3.9	+1.0
anes	No trend across elections	35.5	−1.5	−3.7	+0.7
anes	No trend across elections	39.5	−3.8	−5.9	−1.6
anes	No trend across elections	43.5	−1.4	−3.7	+0.9
anes	No trend across elections	47.5	+0.2	−2.3	+2.6
anes	No trend across elections	51.5	+0.3	−2.2	+2.5
anes	No trend across elections	55.5	−1.7	−4.1	+0.6
anes	No trend across elections	59.5	−0.2	−2.5	+2.0
anes	No trend across elections	63.5	−0.4	−2.7	+2.1
anes	No trend across elections	67.5	+1.6	−1.1	+4.2
anes	No trend across elections	71.5	+1.3	−1.6	+4.1
anes	No trend across ages	19.5	+5.7	+2.5	+9.0
anes	No trend across ages	23.5	+0.2	−2.3	+2.8
anes	No trend across ages	27.5	−0.2	−2.5	+2.5
anes	No trend across ages	31.5	−1.8	−4.2	+0.6
anes	No trend across ages	35.5	−1.7	−3.8	+0.5
anes	No trend across ages	39.5	−3.9	−6.1	−1.7
anes	No trend across ages	43.5	−1.4	−3.7	+0.8
anes	No trend across ages	47.5	+0.2	−2.2	+2.6
anes	No trend across ages	51.5	+0.4	−2.0	+2.5
anes	No trend across ages	55.5	−1.5	−3.7	+0.7
anes	No trend across ages	59.5	0.0	−2.2	+2.3
anes	No trend across ages	63.5	0.0	−2.4	+2.3
anes	No trend across ages	67.5	+2.1	−0.2	+4.4
anes	No trend across ages	71.5	+1.8	−0.5	+4.1

The three profiles fit the vote shares of every age band in every election identically, and they differ by a straight tilt: each constraint sets the straight-line trend of one of the three effects to zero. The curvature is the same in all three, so where a profile bends is identified, and its overall slope is not. Each line is centred on its mean, and bands are 95% intervals from 1,000 bootstrap replicates.Note: The model gives each 4-year age band from 18 to 73, each presidential election and each 4-year birth cohort an effect on the Democratic share, and is fitted by weighted least squares to the cell means (the linear probability model). Because birth year equals election year minus age, the three sets of effects are not identified: adding a straight line to the age effects and the cohort effects and subtracting one from the election effects leaves every fitted cell unchanged. The profile shown is the minimum-norm fit shifted by the straight line that gives the stated constraint.Note: The fitted shares in every cell are identical under the three constraints, to within floating-point rounding. Voters aged 74 and over fall outside the 4-year grid. The CES is an opt-in online panel, so its intervals describe sampling variation among panel members and leave out how respondents joined the panel.Source: CES Cumulative File, presidential elections 2008-2024; weighted (weight_cumulative); two-party presidential voters; weighted least squares on the age-by-election grid, seeded respondent bootstrap

Which constraint to use is a question for theory, and the data do not settle it. Fosse and Winship review the methods that give a point estimate by adding a restriction, among them equality constraints, Moore-Penrose estimators (the minimum-norm fit used here as the base solution) and multilevel models, and an approach that reports bounds and sensitivity to the restriction instead [6]. This chapter takes the second course. It reports the range of the age effect across three constraints and does not choose among them.

These shares describe voters

Every vote share here is a share of respondents who reported voting for a major-party candidate, so an age gap in the vote is a gap among voters. It says nothing about the adults who did not vote. Turnout differs by age group, and Chapter 3.7, Turnout reports it by age and by other characteristics in the Current Population Survey. The gap among voters can change because preferences change or because different people vote, and these surveys cannot separate the two. Chapter 3.1, The two coalitions reports the age make-up of each party's voters. The comparison of matched and unmatched voters above is a reminder that who is counted affects the estimate even among voters.

What these data cannot show

The surveys interview different people in each election, so a cohort line does not follow anyone. It mixes people who changed, people who joined or left the sample (a cohort is smaller at older ages and its youngest members enter as they turn eighteen), and the effects of age and election. A panel study that re-interviews the same voters separates a person's change from the replacement of cohorts, and it still cannot separate the effect of age from the effect of the election in which the person is re-interviewed. The ANES 2024 Time Series holds a sample interviewed in 2016, 2020 and 2024 and the CES has panel files, but the earlier waves of both are not in the data store, so no panel evidence appears here.

The chapter chooses no constraint for the age-period-cohort model and reports the range. A model with more structure can give a narrower answer only by assuming that structure, and the surveys do not test it. The generation labels in the literature summarize cohort differences, and they do not explain why those differences arise. The data hold no measure of the events, campaigns or candidates that might have shaped any cohort.

The vote is the respondent's report, restricted to two-party voters. The ANES Democratic share exceeds the certified count in 2020 and 2024 by more than its interval, and the excess is largest among voters sixty-five and older, so its age gaps rest on a sample that departs from the count most in the oldest group. The CES share is close to the count in 2024. Neither survey can be checked against an official count by age group. The two surveys disagree about 2024, and the explanations offered here, who is matched to voter records and interview mode, cannot be settled with these files. The ANES youngest-group cells are small, 3 elections have none to publish, 649 two-party voters have no recorded age and fall outside every age cell, and age is top-coded at eighty in 2020 and 2024. The 2012, 2016 and 2024 ANES samples mix modes and 2020 has no mode weights, so a change between those elections can include a change of mode. The CES is an opt-in panel and its intervals leave out how respondents joined it. The ANES intervals ignore clustering and stratification, so they can be too narrow or too wide. The chapter reports many comparisons, and no interval is adjusted for that. Cohorts are sixteen-year groups, and a finer grouping could show movements that these groups average away.

Data sources

ANES Cumulative Data FileCumulative Data File: presidential vote, party identification, age and birth cohort, presidential elections 1952-2024Years: 1948-2024 (33 study years; even years only after 2004; no 2006, 2010 or 2014 study) · Weights: post-election weight where available, otherwise VCF0009z; weight_ftf and weight_web for interview-mode estimates · License: Free download after accepting the ANES terms of use; cite the release.
Cooperative Election StudyCumulative Common Content: presidential vote 2008-2024 and party identification 2006-2024Years: 2006-2024 (every year; self-reported turnout is available in the 10 even years) · Weights: weight_cumulative; weight_post (2012, 2016, 2020) and vvweight for sensitivity rows · License: See the provider's terms of use.
State presidential returnsState presidential returns 1976-2024, used only for the certified national two-party shareYears: 1976-2024 (13 elections) · Weights: none (full counts) · License: See the provider's terms of use.

All data sources

Methods

Samples: ANES Cumulative Data File respondents in the 19 presidential elections from 1952 to 2024 who reported voting for the Democratic or the Republican candidate (two-party voters) and have a recorded age of 18 or older (33,768 respondents; 649 other two-party voters have no age and are outside every age cell but inside the cohort cells when their cohort is known; a handful of 17-year-olds are left out), and party identifiers in the same presidential-election years (44,932 respondents with a known age); CES cumulative file respondents who reported such a vote in 2008, 2012, 2016, 2020 and 2024 (188,145) and party identifiers in the even-year waves 2006 to 2024 (450,325). voted_dem_pres equal to 0 in the ANES also holds third-party voters, so the vote is built from voted_dem_pres equal to 1 and voted_rep_pres equal to 1. Weights: ANES vote shares use weight_postelection in the years where the file has one (1996, 2000, 2004, 2008, 2016, 2020, 2024), otherwise weight (VCF0009z), the rule used by the two-coalitions and gender-gap chapters; ANES party identification uses weight (VCF0009z); the ANES interview-mode estimates use weight_ftf on face-to-face respondents and weight_web on web respondents (2012, 2016 and 2024; the 2020 study has no mode weights and no in-person interviews); CES estimates use weight_cumulative, the cumulative-file weight defined for all five elections, which is proportional within each election to the single-wave pre-election weight; the vote is a post-election item, and the post-election weight (weight_post), which exists for 2012, 2016 and 2020 only, is used for a sensitivity check; the matched-voter rows use vvweight, the weight for validated registered voters (respondents whose registration was matched to a voter-file record, not whose vote was validated), which the file holds for 2018, 2020, 2022 and 2024 and which equals a constant multiple of weight_cumulative in 2020. ANES 2012 uses the full-sample weight because the file has no post-election weight for that study; in 2024, a share of respondents (panel and paper cases) have no interview-mode weight and are outside the mode cells. Groups: age 18 to 29, 30 to 44, 45 to 64 and 65 and over from age in years (the CES vote figure splits the youngest group at 24 and 25); birth cohort is the ANES cohort variable (16-year groups; cohort cells include every voter with a known cohort) and, in the CES, the same groups from reported birth year; party identification is Democratic (pid7 or pid_7pt 1 to 3) against Republican (5 to 7), pure independents excluded. Estimands: a share is a weighted mean of a 0/1 outcome; the age gap is the share among the youngest group minus the share among the oldest; a cohort's lead is its share minus the share among all other voters with a known cohort in the same election (a disjoint group, so the two shares are independent); a change is a difference of two independent estimates from different elections. Intervals: 95%. Shares and differences within one election use the Kish effective sample size and a normal approximation, and combine the standard errors of independent groups in quadrature. The 2024 CES difference between the gap among matched respondents and the gap among all voters, the difference between the two weights on the same matched respondents, and the difference between the change in the gap from 2020 to 2024 among all voters and the same change among matched respondents are seeded percentile bootstraps over respondents, resampling within each year (1,000 replicates, seed 12345), because those estimates share respondents. The certified benchmark is the two-party share of the state presidential returns (full counts). Three comparisons: weighted least squares with HC3 errors on voters aged 18 to 73 (ANES 30,571, CES 169,858 respondents): the vote on age in decades with election fixed effects; on election year in decades with fixed effects for 4-year age bands; and on election year in decades with fixed effects for 4-year birth bands. The sum of the first two slopes minus the third has a seeded respondent bootstrap interval that recomputes the three slopes on each resample. Age-period-cohort model: voters aged 18 to 73 are grouped into 14 age bands of 4 years and the presidential elections (4 years apart, so birth cohort is the election index minus the age-band index); the 0/1 vote is regressed on an effect for every age band, election and cohort by weighted least squares on the cell means (the linear probability model), taking the minimum-norm solution. Adding a straight line to the age and cohort effects and subtracting one from the election effects leaves every fitted cell unchanged, and each constraint chooses the shift that sets the straight-line component of one effect to zero (no trend across birth years, across elections or across ages). The largest difference in fitted cell means across the three constraints is 2.2e-14 percentage points in the ANES and 1.1e-14 in the CES. Intervals for the model are percentile intervals from 1,000 respondent-level bootstrap replicates (seed 12345). Suppression: a published cell needs at least 100 respondents and a Kish effective sample of at least 100; a suppressed cell is a row with no value, so no line is drawn across it. All intervals are model based. The CES is an opt-in panel, so its intervals leave out how respondents joined the panel, and the ANES intervals ignore clustering and stratification. Survey mode: the ANES mixes face-to-face and web interviews in 2012, 2016 and 2024 and used no in-person interviews in 2020; the CES is an online panel in every year.

References

Glenn, Norval D. (1976). Cohort Analysts' Futile Quest: Statistical Attempts to Separate Age, Period and Cohort Effects. American Sociological Review 41(5): 900-904. doi:10.2307/2094738
Mason, Karen Oppenheim; Mason, William M.; Winsborough, H. H.; Poole, W. Kenneth (1973). Some Methodological Issues in Cohort Analysis of Archival Data. American Sociological Review 38(2): 242-258. doi:10.2307/2094398
Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Converse, Philip E. (1969). Of Time and Partisan Stability. Comparative Political Studies 2(2): 139-171. doi:10.1177/001041406900200201
Ghitza, Yair; Gelman, Andrew; Auerbach, Jonathan (2023). The Great Society, Reagan's Revolution, and Generations of Presidential Voting. American Journal of Political Science 67(3): 520-537. doi:10.1111/ajps.12713
Fosse, Ethan; Winship, Christopher (2019). Analyzing Age-Period-Cohort Data: A Review and Critique. Annual Review of Sociology 45: 467-492. doi:10.1146/annurev-soc-073018-022616

How to cite

@misc{singh2026ageandgeneration,
  author = {Singh, Vishal},
  title = {Age and generation},
  year = {2026},
  url = {https://vishalsingh.org/politics/age-and-generation},
  note = {In The Public Mind. Published 3 October 2026.}
}