Chapter 10.3

Down the ballot

In 2024 4.7% of ANES voters reported different parties for president and the House, against 22.3% in 1992, and the presidential and House winners differed in 2.8% of counties in which every district part of the House race was contested, against 31.0% in 1992.

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

In 2024, 4.7% (95% CI 3.7% to 5.7%) of voters in the American National Election Studies (ANES) who reported a major-party vote for president and for the House reported votes for candidates of different parties, against 22.3% (95% CI 19.9% to 24.8%) in 1992 and 29.5% (95% CI 27.1% to 32.0%) in 1972, the highest rate among the 19 studies from 1952 to 2024. In county returns, the presidential and House winners differed in 2.8% of the 2,695 counties in which every district part of the House race was contested in 2024, against 31.0% of 2,735 counties in 1992.

Ticket splitting is a vote for one party's candidate for one office and for another party's candidate for a second office on the same ballot. The ANES asks respondents for both votes, so it measures splitting directly, by self-report.The individual sample is 29,103 respondents in 19 presidential-election studies from 1952 to 2024 who reported a major-party vote for president and the House, and 20,334 who reported one for president and the Senate. Estimates use the survey weight and intervals use the Kish effective sample size [1], which treats the weighted sample as a simple random sample. County returns count each candidate's votes in a place and never match one voter's two choices. They show how often the two winners in a county differed, which splitting can produce, although the two races also draw different numbers of voters and splits in opposite directions cancel in a county total. This chapter calls a county whose presidential and House winners differ a divided county.The county sample is the contiguous states, with 3,107 counties holding presidential votes in 2024, in the 9 presidential elections from 1992 to 2024. Jackson County, Missouri, is left out from 2004 to 2024 because the county file omits the part of the county that lies in Kansas City. The rules for which counties enter each comparison are given below and in the methods note. Chapter 10.1, Nationalization of the county vote measures how uniformly counties moved between presidential elections, and this chapter compares two offices in the same election.

Burden and Kimball [2] studied presidential, House and Senate voting from 1952 to 1996 and linked ticket splitting to the similarity of the candidates' platforms, to candidate characteristics and to changes in congressional campaigns such as the rise of incumbency advantage and of money, and not to voters balancing the parties. Jacobson [3] reports that the electoral advantage of House incumbents fell to levels not seen since the 1950s together with an increase in party loyalty, straight-ticket voting and president-centered nationalization, and Hopkins [4] documents the nationalization of American political behavior. Kuriwaki [5] uses cast vote records, which show one voter's choices across offices, and finds ticket splitting higher in state and local offices than in Congress. This chapter describes how much splitting there is for Congress and does not test any of these explanations.

Self-reported splitting was highest from 1972 to 1988 and lowest in 2024

Figure 1 plots the share of voters who split between president and House, with the Senate beside it. From 1952 to 1968 the president and House rate was between 12.6% and 17.8% in every study. It was 29.5% (95% CI 27.1% to 32.0%) in 1972 and between 24.8% and 29.5% in each study from 1972 to 1988. In 1992 it was 22.3% (95% CI 19.9% to 24.8%), and in each study from 1996 to 2008 it was between 16.7% and 18.3%. The later studies give 9.8% (95% CI 8.5% to 11.1%) in 2012, 15.3% (95% CI 13.4% to 17.1%) in 2016, 11.4% (95% CI 10.1% to 12.6%) in 2020 and 4.7% (95% CI 3.7% to 5.7%) in 2024. The 2024 interval lies below the interval of every earlier study. The rate was 6.7 (95% CI 5.1 to 8.2) points lower in 2024 than in 2020, 17.7 (95% CI 15.0 to 20.3) points lower than in 1992 and 24.9 (95% CI 22.2 to 27.5) points lower than in 1972.

Figure 1

Self-reported ticket splitting between president and House fell from 29.5% of voters in 1972 to 4.7% in 2024

Share of voters who reported different parties' candidates for president and for the House (or Senate), voters with a major-party vote in both races, ANES studies 1952 to 2024; 95% intervals

Show data

Data for Self-reported ticket splitting between president and House fell from 29.5% of voters in 1972 to 4.7% in 2024
Races	Interview mode	Study	Share who split (%)	Lower	Upper	Respondents
President and House	all	1952	12.6	10.5	14.6	1,009
President and House	all	1956	15.6	13.5	17.6	1,151
President and House	all	1960	14.0	11.1	16.9	766
President and House	all	1964	15.1	12.8	17.4	947
President and House	all	1968	17.8	15.1	20.5	776
President and House	all	1972	29.5	27.1	32.0	1,293
President and House	all	1976	24.8	22.1	27.4	1,049
President and House	all	1980	27.7	24.5	30.9	762
President and House	all	1984	25.4	22.9	27.9	1,142
President and House	all	1988	25.3	22.7	28.0	1,030
President and House	all	1992	22.3	19.9	24.8	1,126
President and House	all	1996	17.7	14.9	20.4	926
President and House	all	2000	18.3	15.5	21.2	882
President and House	all	2004	16.8	13.8	19.8	701
President and House	all	2008	16.7	14.1	19.3	1,222
President and House	all	2012	9.8	8.5	11.1	3,501
President and House	all	2016	15.3	13.4	17.1	2,138
President and House	all	2020	11.4	10.1	12.6	5,167
President and House	all	2024	4.7	3.7	5.7	3,515
President and House	ftf	2008	16.8	14.2	19.4	1,222
President and House	ftf	2012	11.0	8.5	13.6	1,065
President and House	ftf	2016	10.2	7.2	13.1	619
President and House	ftf	2024	5.3	2.8	7.8	579
President and House	web	2012	9.2	7.7	10.7	2,436
President and House	web	2016	17.4	15.1	19.6	1,519
President and House	web	2024	5.0	3.4	6.6	1,227
President and Senate	all	1952	12.0	9.8	14.3	797
President and Senate	all	1956	16.1	13.6	18.5	878
President and Senate	all	1960	11.1	7.7	14.5	455
President and Senate	all	1964	17.2	14.5	19.8	787
President and Senate	all	1968	19.2	16.1	22.4	603
President and Senate	all	1972	28.5	25.3	31.7	765
President and Senate	all	1976	23.6	20.5	26.6	768
President and Senate	all	1980	23.7	20.2	27.3	539
President and Senate	all	1984	22.7	19.7	25.8	717
President and Senate	all	1988	23.3	20.4	26.2	819
President and Senate	all	1992	19.2	16.4	21.9	795
President and Senate	all	1996	14.7	11.4	18.0	548
President and Senate	all	2000	14.0	11.2	16.8	726
President and Senate	all	2004	14.5	11.2	17.9	508
President and Senate	all	2008	12.9	9.9	15.9	786
President and Senate	all	2012	11.1	9.5	12.6	2,814
President and Senate	all	2016	13.0	11.0	15.0	1,625
President and Senate	all	2020	7.2	5.9	8.5	2,879
President and Senate	all	2024	6.8	5.4	8.3	2,525
President and Senate	ftf	2008	12.8	9.9	15.8	786
President and Senate	ftf	2012	10.4	7.6	13.1	837
President and Senate	ftf	2016	10.3	6.7	13.9	447
President and Senate	ftf	2024	7.9	4.4	11.4	433
President and Senate	web	2012	11.5	9.6	13.3	1,977
President and Senate	web	2016	14.0	11.6	16.4	1,178
President and Senate	web	2024	6.2	4.1	8.3	923

The rate was highest from 1972 to 1988 and lowest in 2024, and the Senate line is close to the House line in level but not in every change. Each point is the weighted share, among respondents who reported a major-party vote for president and for the House (or Senate), who reported candidates of two different parties; bands are 95% confidence intervals. The buttons show in-person estimates for 2008 (a study with no web sample), 2012, 2016 and 2024 and web estimates for 2012, 2016 and 2024.Note: The 2024 president and House estimate is rebuilt from the presidential and House vote items because the cumulative file's combined item (VCF0709) is missing for 2024. In the 18 earlier studies the rebuilt code equals VCF0709 for all 25,637 respondents with both.Note: The Senate item is asked only where the respondent's state had a Senate race. The 2020 study has no mode weights, so the mode views leave 2020 open. The all-interview estimates use the post-election weight in the studies that have one and VCF0009z in the others, 2012 included; the mode estimates use the in-person and web weights. Weights equal one in 1952, 1956, 1964, 1968, 1972, 1980, 1984 and 1988.Source: ANES Cumulative Data File, 1952-2024; weighted (post-election weight where the file has one, VCF0009z elsewhere; mode weights weight_ftf and weight_web for 2008, 2012, 2016 and 2024)

The cumulative file's combined president and House item (VCF0709) is missing for 2024, so every year here uses a split rebuilt from the presidential vote item (VCF0704) and the House vote item (VCF0707). The rebuilt code covers 3,530 respondents in 2024. In the 18 studies where VCF0709 exists, it equals the rebuilt code for all 25,637 respondents who have both, and 2 respondents in 1984 have VCF0709 but a missing vote item and are left out. The vote items are post-election, so estimates use the post-election weight in the studies that have one and the full-sample weight in the others, 2012 included. 64 respondents with both votes in 2016, 2020 and 2024 lack the post-election weight and are left out.

The president and Senate rate is close to the House rate in level, with a correlation of +0.94 across the 19 studies, but it does not move with it at every step. It was 28.5% (95% CI 25.3% to 31.7%) in 1972, 7.2% (95% CI 5.9% to 8.5%) in 2020 and 6.8% (95% CI 5.4% to 8.3%) in 2024, with overlapping intervals for 2020 and 2024. The Senate item is asked only where the respondent's state held a Senate race. The fall since 2008 is not smooth for the House. The rate dropped by 6.9 (95% CI 4.0 to 9.8) points between 2008 and 2012, rose in 2016 and dropped again by 2024. The 2008 to 2012 fall crosses the addition of the web sample, and among in-person respondents alone the rate went from 16.8% (95% CI 14.2% to 19.4%) to 11.0% (95% CI 8.5% to 13.6%), a fall of 5.7 (95% CI 2.1 to 9.4) points, so the fall is not only a change of mode. The Senate rate did not fall over those years: it was 12.9% (95% CI 9.9% to 15.9%) in 2008 and 11.1% (95% CI 9.5% to 12.6%) in 2012, a change of −1.8 (95% CI −5.2 to 1.6) points with an interval that includes zero.

The 2016 rise appears among web respondents in the House item

The ANES interviewed respondents mostly in person through 2008 and added a web sample alongside the in-person sample in 2012, 2016 and 2024. The 2020 study has no separate weights by interview mode, so it enters only the all-interview series. The all-interview rate for the House rose from 9.8% in 2012 to 15.3% in 2016, a change of +5.5 (95% CI 3.2 to 7.7) points. The rise is larger among web respondents, who moved from 9.2% (95% CI 7.7% to 10.7%) to 17.4% (95% CI 15.1% to 19.6%), a change of +8.2 (95% CI 5.4 to 10.9) points. In-person respondents moved from 11.0% (95% CI 8.5% to 13.6%) to 10.2% (95% CI 7.2% to 13.1%), a change of −0.9 (95% CI −4.8 to 3.1) points whose interval includes zero and also a rise of several points, so the data do not confine the rise to the web sample.

Figure 2 compares the two modes within each study. In 2016 web respondents split more than in-person respondents by +7.2 (95% CI 3.4 to 10.9) points for the House. The Senate item shows less. Web minus in-person was +3.7 (95% CI −0.6 to 8.0) points in 2016, +1.1 (95% CI −2.2 to 4.4) in 2012 and −1.7 (95% CI −5.8 to 2.5) in 2024, and the web change from 2012 to 2016 was +2.5 (95% CI −0.5 to 5.5) points. All four intervals include zero, so the 2016 jump is clear in the House item and not in the Senate item. For the House, the difference was −1.9 (95% CI −4.9 to 1.1) points in 2012 and −0.3 (95% CI −3.2 to 2.7) points in 2024, when in-person respondents split at 5.3% (95% CI 2.8% to 7.8%) and web respondents at 5.0% (95% CI 3.4% to 6.6%). Both of those intervals include zero and exclude a gap the size of the 2016 gap, and both still allow differences of a few points, which is wide against a rate near 4.7%. The fall from 2016 to 2024 appears in both modes, at 4.9 (95% CI 1.0 to 8.8) points among in-person respondents and 12.4 (95% CI 9.6 to 15.2) points among web respondents.

Figure 2

Web respondents reported more ticket splitting than in-person respondents for the House in 2016 by +7.2 points, against −1.9 in 2012 and −0.3 in 2024

Share of voters who reported different parties' candidates for president and the House (or Senate), by interview mode, ANES 2012, 2016 and 2024; 95% intervals

In person
Web

Show data

Data for Web respondents reported more ticket splitting than in-person respondents for the House in 2016 by +7.2 points, against −1.9 in 2012 and −0.3 in 2024
ftf	ftf_hi	ftf_lo	label	n_ftf	n_web	web	web_hi	web_lo
11.04	13.62	8.45	House, 2012	1,065	2,436	9.18	10.69	7.66
10.17	13.15	7.18	House, 2016	619	1,519	17.35	19.62	15.09
5.27	7.76	2.79	House, 2024	579	1,227	5.00	6.63	3.36
10.36	13.09	7.63	Senate, 2012	837	1,977	11.47	13.33	9.60
10.28	13.85	6.71	Senate, 2016	447	1,178	14.01	16.37	11.65
7.88	11.41	4.35	Senate, 2024	433	923	6.23	8.34	4.11

In 2016 the web dot lies to the right of the in-person dot for the House, and in 2012 and 2024 the two dots are close. The Senate rows show a smaller 2016 difference with an interval that includes zero. Each row plots two weighted shares with 95% confidence intervals; the in-person estimate uses weight_ftf and the web estimate weight_web.Note: The 2024 mode weights cover the newly recruited in-person and web samples. The 1,709 voters in the panel and paper samples belong to neither mode row.Note: The 2020 study has no mode weights and is not shown. The file has no post-election mode weights, so these estimates use the full-sample in-person and web weights.Source: ANES Cumulative Data File, 2012, 2016 and 2024; weighted (weight_ftf and weight_web)

The two modes are separately recruited samples, so the comparison cannot separate an effect of the interview mode from a difference between the people who answered. The mode estimates use the in-person and web weights, because the file has no post-election mode weights. The 2024 mode weights cover only the newly recruited samples. The 1,709 panel and paper respondents with both votes are in neither mode, and they split at 3.7% (95% CI 2.4% to 4.9%), which is why the all-interview rate for 2024 sits below the rate in each mode. The fall from 2020 to 2024 is read from the all-interview series because 2020 has no mode weights.

From 1972 to 1988 the usual split was a Republican president and a Democratic House

Figure 3 separates the two kinds of split. Voters who chose a Republican for president and a Democrat for the House were 25.0% (95% CI 22.6% to 27.3%) of voters in 1972 and 18.3% (95% CI 15.9% to 20.6%) in 1988, and in every study from 1972 to 1988 that line is above the other kind, with separate intervals. In 1996 the order reversed, with 13.4% (95% CI 10.9% to 15.8%) choosing a Democrat for president and a Republican for the House against 4.3% (95% CI 2.8% to 5.7%) for the opposite combination. From 1996 on the Democratic-president, Republican-House kind has the higher estimate in 7 of the 8 studies, and the intervals separate only in 1996 and 2020, when the two kinds were 6.8% (95% CI 5.8% to 7.8%) and 4.6% (95% CI 3.8% to 5.4%). In 2024 they are 3.0% (95% CI 2.2% to 3.8%) and 1.7% (95% CI 1.1% to 2.3%).

Splits in opposite directions offset each other in a vote total, so the net effect on the House vote is smaller than the share who split. Among the same voters, the House Democratic share minus the presidential Democratic share was +20.4 (95% CI 17.7 to 23.2) points in 1972 and −9.1 (95% CI −12.0 to −6.1) points in 1996, when the House vote was less Democratic than the presidential vote. In 1992 22.3% of voters split, and the net difference was +2.1 (95% CI −0.6 to 4.9) points, an interval that includes zero. In 2024 it was −1.3 (95% CI −2.3 to −0.2) points. The second view of the figure plots this net effect.

Figure 3

Voters who chose a Republican for president and a Democrat for the House were 25.0% in 1972 and 1.7% in 2024

Share of voters with a major-party vote for president and for the House, by which party won each vote, ANES studies 1952 to 2024; 95% intervals

Show data

Data for Voters who chose a Republican for president and a Democrat for the House were 25.0% in 1972 and 1.7% in 2024
View	Measure	Study	Estimate	Lower	Upper	Respondents
direction	Republican for president, Democratic for House	1952	10.3	8.4	12.2	1,009
direction	Republican for president, Democratic for House	1956	13.5	11.5	15.4	1,151
direction	Republican for president, Democratic for House	1960	9.8	7.3	12.3	766
direction	Republican for president, Democratic for House	1964	6.4	4.9	8.0	947
direction	Republican for president, Democratic for House	1968	10.8	8.6	13.0	776
direction	Republican for president, Democratic for House	1972	25.0	22.6	27.3	1,293
direction	Republican for president, Democratic for House	1976	15.5	13.3	17.7	1,049
direction	Republican for president, Democratic for House	1980	19.8	17.0	22.6	762
direction	Republican for president, Democratic for House	1984	19.8	17.5	22.1	1,142
direction	Republican for president, Democratic for House	1988	18.3	15.9	20.6	1,030
direction	Republican for president, Democratic for House	1992	12.2	10.3	14.2	1,126
direction	Republican for president, Democratic for House	1996	4.3	2.8	5.7	926
direction	Republican for president, Democratic for House	2000	8.5	6.5	10.5	882
direction	Republican for president, Democratic for House	2004	9.6	7.2	12.0	701
direction	Republican for president, Democratic for House	2008	7.9	6.0	9.8	1,222
direction	Republican for president, Democratic for House	2012	4.0	3.2	4.9	3,501
direction	Republican for president, Democratic for House	2016	6.5	5.2	7.8	2,138
direction	Republican for president, Democratic for House	2020	4.6	3.8	5.4	5,167
direction	Republican for president, Democratic for House	2024	1.7	1.1	2.3	3,515
direction	Democratic for president, Republican for House	1952	2.3	1.4	3.2	1,009
direction	Democratic for president, Republican for House	1956	2.1	1.3	2.9	1,151
direction	Democratic for president, Republican for House	1960	4.2	2.5	5.9	766
direction	Democratic for president, Republican for House	1964	8.7	6.9	10.5	947
direction	Democratic for president, Republican for House	1968	7.0	5.2	8.8	776
direction	Democratic for president, Republican for House	1972	4.6	3.4	5.7	1,293
direction	Democratic for president, Republican for House	1976	9.3	7.5	11.0	1,049
direction	Democratic for president, Republican for House	1980	7.9	6.0	9.8	762
direction	Democratic for president, Republican for House	1984	5.6	4.3	6.9	1,142
direction	Democratic for president, Republican for House	1988	7.1	5.5	8.7	1,030
direction	Democratic for president, Republican for House	1992	10.1	8.3	11.9	1,126
direction	Democratic for president, Republican for House	1996	13.4	10.9	15.8	926
direction	Democratic for president, Republican for House	2000	9.8	7.7	12.0	882
direction	Democratic for president, Republican for House	2004	7.2	5.1	9.3	701
direction	Democratic for president, Republican for House	2008	8.9	6.9	10.8	1,222
direction	Democratic for president, Republican for House	2012	5.8	4.7	6.8	3,501
direction	Democratic for president, Republican for House	2016	8.8	7.3	10.2	2,138
direction	Democratic for president, Republican for House	2020	6.8	5.8	7.8	5,167
direction	Democratic for president, Republican for House	2024	3.0	2.2	3.8	3,515
net	House Democratic share minus presidential Democratic share	1952	8.0	5.9	10.2	1,009
net	House Democratic share minus presidential Democratic share	1956	11.4	9.2	13.6	1,151
net	House Democratic share minus presidential Democratic share	1960	5.6	2.5	8.6	766
net	House Democratic share minus presidential Democratic share	1964	−2.2	−4.7	0.3	947
net	House Democratic share minus presidential Democratic share	1968	3.9	0.9	6.8	776
net	House Democratic share minus presidential Democratic share	1972	20.4	17.7	23.2	1,293
net	House Democratic share minus presidential Democratic share	1976	6.3	3.2	9.3	1,049
net	House Democratic share minus presidential Democratic share	1980	11.9	8.3	15.6	762
net	House Democratic share minus presidential Democratic share	1984	14.2	11.4	17.0	1,142
net	House Democratic share minus presidential Democratic share	1988	11.2	8.2	14.2	1,030
net	House Democratic share minus presidential Democratic share	1992	2.1	−0.6	4.9	1,126
net	House Democratic share minus presidential Democratic share	1996	−9.1	−12.0	−6.1	926
net	House Democratic share minus presidential Democratic share	2000	−1.4	−4.5	1.8	882
net	House Democratic share minus presidential Democratic share	2004	2.4	−0.9	5.7	701
net	House Democratic share minus presidential Democratic share	2008	−1.0	−3.8	1.9	1,222
net	House Democratic share minus presidential Democratic share	2012	−1.7	−3.1	−0.3	3,501
net	House Democratic share minus presidential Democratic share	2016	−2.3	−4.3	−0.2	2,138
net	House Democratic share minus presidential Democratic share	2020	−2.2	−3.5	−0.9	5,167
net	House Democratic share minus presidential Democratic share	2024	−1.3	−2.3	−0.2	3,515

The Republican-president, Democratic-House line is above the other line in every study from 1972 to 1988 and below it in 1996, and from 1996 on the other line has the higher estimate in 7 of 8 studies. The Democratic-president, Republican-House line was 13.4% in 1996 and 3.0% in 2024. Bands are 95% confidence intervals. The second view plots the net effect of the two kinds of split on the House vote.Note: Net effect, second view: the weighted mean of the House Democratic vote minus the presidential Democratic vote among the same voters, in points. Zero means the two Democratic shares are equal even if many voters split in each direction.Source: ANES Cumulative Data File, 1952-2024; weighted (post-election weight where the file has one, VCF0009z elsewhere)

A county is divided when the party that won more of the Democratic and Republican presidential votes differs from the party that won more of the House votes. The comparison needs rules for races without a real contest and for defects in the county file. A county is left out for an exact tie, for Jackson County, Missouri, in the 6 elections from 2004 to 2024, because the file leaves out the part of the county in Kansas City, and for Senate races in which the file codes an independent as a major-party candidate (Maine and Vermont in 2012 and Nebraska in 2024, 123 county-years), because it codes the same candidates as independents in other years.

A totals screen leaves out county rows whose House or Senate total votes fall outside 0.7 to 1.3 times the presidential total. It removes 687 of 27,567 House county-years (2.5%) and 398 of 18,645 Senate county-years (2.1%). The named defects are the Iowa 2012 House file, which double counts the votes (99 counties), the Georgia Senate vote in 1992 and 2008, where the file holds a later runoff round (158 and 157 counties), and the Louisiana Senate vote in 1992, which is the October primary (61 counties). The other 588 House and 22 Senate county-years have totals far from the presidential total for reasons the file does not give, and 150 of the House ones would otherwise be in the main sample. Dropping the screen changes the yearly share of divided counties by at most 0.5 points for the House and 1.3 points for the Senate.

A race is a two-party race when the Democratic and Republican candidates together won at least 80% of its votes, which leaves out races with a strong independent and fails in 14 House races and 1 Senate race with candidates of both parties. A race is contested when both candidates received votes. A House race is a district, and a county enters the main sample only when every district part of it was a contested two-party race, because a county that holds parts of several districts sums parts of several races. A Senate race is a state. Among the county-years that pass the first two rules, the race rules remove 3,097 House and 295 Senate county-years. A second sample, all returns, keeps the counties in uncontested two-party races.

Figure 4 plots the share of divided counties in the House sample. It was 31.0% in 1992, among 2,735 counties, and stayed between 25.5% and 31.0% in every election through 2008. It was 9.7% in 2012, which is 16.9 points below 2008, 7.7% in 2016, 3.0% in 2020 and 2.8% in 2024. No election after 2008 exceeded 9.7%. Weighted by votes, the divided counties held 27.9% of the presidential two-party votes cast in the sample in 1992 and 5.5% in 2024, so the divided counties of 2024 were larger than the typical county. Keeping the counties in uncontested races gives 30.6% in 1992 and 2.7% in 2024, and it changes the county share by at most 2.1 points in any election, the largest change being in 2004.

Figure 4

Counties whose presidential and House winners differed fell from 31.0% of counties in 1992 to 2.8% in 2024

Share of counties, and of presidential two-party votes, where the Democratic and Republican presidential winner differs from the House winner, counties with a contested House race, 1992 to 2024

Show data

Data for Counties whose presidential and House winners differed fell from 31.0% of counties in 1992 to 2.8% in 2024
Office	View	Measure	Election	Share (%)	Counties	Counties in sample
House	share	Counties	1992	31.0	n/a	2,735
House	share	Presidential votes	1992	27.9	n/a	2,735
House	share	Counties, uncontested races included	1992	30.6	n/a	2,843
House	direction	Republican for president, Democratic down the ballot	1992	n/a	573	2,735
House	direction	Democratic for president, Republican down the ballot	1992	n/a	276	2,735
House	share	Counties	1996	25.6	n/a	2,703
House	share	Presidential votes	1996	24.8	n/a	2,703
House	share	Counties, uncontested races included	1996	25.4	n/a	2,769
House	direction	Republican for president, Democratic down the ballot	1996	n/a	226	2,703
House	direction	Democratic for president, Republican down the ballot	1996	n/a	466	2,703
House	share	Counties	2000	27.0	n/a	2,555
House	share	Presidential votes	2000	21.8	n/a	2,555
House	share	Counties, uncontested races included	2000	26.2	n/a	2,918
House	direction	Republican for president, Democratic down the ballot	2000	n/a	521	2,555
House	direction	Democratic for president, Republican down the ballot	2000	n/a	170	2,555
House	share	Counties	2004	25.5	n/a	2,479
House	share	Presidential votes	2004	21.5	n/a	2,479
House	share	Counties, uncontested races included	2004	23.3	n/a	2,848
House	direction	Republican for president, Democratic down the ballot	2004	n/a	516	2,479
House	direction	Democratic for president, Republican down the ballot	2004	n/a	115	2,479
House	share	Counties	2008	26.6	n/a	2,661
House	share	Presidential votes	2008	18.3	n/a	2,661
House	share	Counties, uncontested races included	2008	26.8	n/a	2,925
House	direction	Republican for president, Democratic down the ballot	2008	n/a	560	2,661
House	direction	Democratic for president, Republican down the ballot	2008	n/a	148	2,661
House	share	Counties	2012	9.7	n/a	2,609
House	share	Presidential votes	2012	10.9	n/a	2,609
House	share	Counties, uncontested races included	2012	9.2	n/a	2,857
House	direction	Republican for president, Democratic down the ballot	2012	n/a	162	2,609
House	direction	Democratic for president, Republican down the ballot	2012	n/a	90	2,609
House	share	Counties	2016	7.7	n/a	2,427
House	share	Presidential votes	2016	8.7	n/a	2,427
House	share	Counties, uncontested races included	2016	7.1	n/a	2,921
House	direction	Republican for president, Democratic down the ballot	2016	n/a	131	2,427
House	direction	Democratic for president, Republican down the ballot	2016	n/a	57	2,427
House	share	Counties	2020	3.0	n/a	2,884
House	share	Presidential votes	2020	5.1	n/a	2,884
House	share	Counties, uncontested races included	2020	3.4	n/a	3,084
House	direction	Republican for president, Democratic down the ballot	2020	n/a	40	2,884
House	direction	Democratic for president, Republican down the ballot	2020	n/a	47	2,884
House	share	Counties	2024	2.8	n/a	2,695
House	share	Presidential votes	2024	5.5	n/a	2,695
House	share	Counties, uncontested races included	2024	2.7	n/a	3,017
House	direction	Republican for president, Democratic down the ballot	2024	n/a	49	2,695
House	direction	Democratic for president, Republican down the ballot	2024	n/a	27	2,695
Senate	share	Counties	1992	30.9	n/a	1,824
Senate	share	Presidential votes	1992	29.3	n/a	1,824
Senate	share	Counties, uncontested races included	1992	30.9	n/a	1,824
Senate	direction	Republican for president, Democratic down the ballot	1992	n/a	271	1,824
Senate	direction	Democratic for president, Republican down the ballot	1992	n/a	293	1,824
Senate	share	Counties	1996	20.2	n/a	2,288
Senate	share	Presidential votes	1996	18.1	n/a	2,288
Senate	share	Counties, uncontested races included	1996	20.2	n/a	2,288
Senate	direction	Republican for president, Democratic down the ballot	1996	n/a	94	2,288
Senate	direction	Democratic for president, Republican down the ballot	1996	n/a	369	2,288
Senate	share	Counties	2000	21.1	n/a	1,858
Senate	share	Presidential votes	2000	22.9	n/a	1,858
Senate	share	Counties, uncontested races included	2000	21.1	n/a	1,858
Senate	direction	Republican for president, Democratic down the ballot	2000	n/a	256	1,858
Senate	direction	Democratic for president, Republican down the ballot	2000	n/a	136	1,858
Senate	share	Counties	2004	29.6	n/a	2,006
Senate	share	Presidential votes	2004	26.6	n/a	2,006
Senate	share	Counties, uncontested races included	2004	29.0	n/a	2,050
Senate	direction	Republican for president, Democratic down the ballot	2004	n/a	506	2,006
Senate	direction	Democratic for president, Republican down the ballot	2004	n/a	88	2,006
Senate	share	Counties	2008	24.4	n/a	2,055
Senate	share	Presidential votes	2008	15.2	n/a	2,055
Senate	share	Counties, uncontested races included	2008	24.4	n/a	2,055
Senate	direction	Republican for president, Democratic down the ballot	2008	n/a	436	2,055
Senate	direction	Democratic for president, Republican down the ballot	2008	n/a	65	2,055
Senate	share	Counties	2012	18.2	n/a	1,843
Senate	share	Presidential votes	2012	13.4	n/a	1,843
Senate	share	Counties, uncontested races included	2012	18.2	n/a	1,843
Senate	direction	Republican for president, Democratic down the ballot	2012	n/a	315	1,843
Senate	direction	Democratic for president, Republican down the ballot	2012	n/a	20	1,843
Senate	share	Counties	2016	6.2	n/a	1,989
Senate	share	Presidential votes	2016	10.7	n/a	1,989
Senate	share	Counties, uncontested races included	2016	7.2	n/a	2,045
Senate	direction	Republican for president, Democratic down the ballot	2016	n/a	84	1,989
Senate	direction	Democratic for president, Republican down the ballot	2016	n/a	40	1,989
Senate	share	Counties	2020	1.9	n/a	2,212
Senate	share	Presidential votes	2020	3.6	n/a	2,212
Senate	share	Counties, uncontested races included	2020	1.9	n/a	2,212
Senate	direction	Republican for president, Democratic down the ballot	2020	n/a	21	2,212
Senate	direction	Democratic for president, Republican down the ballot	2020	n/a	22	2,212
Senate	share	Counties	2024	3.0	n/a	1,749
Senate	share	Presidential votes	2024	8.7	n/a	1,749
Senate	share	Counties, uncontested races included	2024	3.0	n/a	1,749
Senate	direction	Republican for president, Democratic down the ballot	2024	n/a	45	1,749
Senate	direction	Democratic for president, Republican down the ballot	2024	n/a	8	1,749

The share of counties stays between 25.5% and 31.0% through 2008, drops to 9.7% in 2012 and is 2.8% in 2024. The dashed line adds the counties in uncontested House races. The Senate view covers the states with a contested Senate race in each election, and the direction views count the divided counties of each kind in every election.Note: Full counts of votes, so the county series have no sampling interval.Note: A winner is the party with more Democratic or Republican votes; third-party votes are left out, as in the two-party margin. A county is in the House sample when every district part of it was a contested race (both major parties on the ballot, together at least 80% of the votes). Jackson County, Missouri, is left out from 2004 to 2024 because the file omits the part of the county in Kansas City.Source: County returns for President, U.S. House and U.S. Senate, 1992-2024 (Heideman 2026, with MIT Election Data and Science Lab presidential returns); full counts, unweighted

The Senate fall comes later in the pooled series. The share of counties divided between president and Senate was 30.9% in 1992, 29.6% in 2004 and 24.4% in 2008, then 18.2% in 2012, 6.2% in 2016, 1.9% in 2020 and 3.0% in 2024. It fell by 6.2 points from 2008 to 2012 and by 11.9 points from 2012 to 2016, against 16.9 points for the House from 2008 to 2012. The Senate sample is the 29 states with a contested regular race in 2024 and differs from election to election, so that comparison mixes states. The cleaner comparison is by Senate class, the seats that come up together, so that the same states vote in every third presidential election. In the class that voted in 2000, 2012 and 2024 the share of divided counties was 21.1%, 18.2% and 3.0%. In the class of 1996, 2008 and 2020 it was 20.2%, 24.4% and 1.9%, and in the class of 1992, 2004 and 2016 it was 30.9%, 29.6% and 6.2%. In each class the latest of the three elections has the lowest share, which is consistent with a fall after 2008 or 2012 and does not date it.

Divided counties mostly had a Republican president and a Democratic House, and regions differed by election

The direction views of Figure 4 count the divided counties of each kind. Counties with a Republican for president and a Democrat for the House outnumbered the reverse in 7 of the 9 elections, and the exceptions are 1996 and 2020. The counts were 573 against 276 in 1992, 226 against 466 in 1996, 40 against 47 in 2020 and 49 against 27 in 2024. In the Senate the Republican-president kind led in 6 elections, and the reverse led in 1992, 1996 and 2020: 369 against 94 in 1996 and 22 against 21 in 2020. In 2024 the Senate counts were 45 and 8. The House exceptions, 1996 and 2020, are also the two studies in which the ANES intervals separate with the Democratic-president kind larger. In most other years after 1996 the ANES point estimates favor that kind while the county counts favor the opposite. The two count different things, voters and county winners, and counties weigh a small county as much as a large one.

Figure 5 colors each county in the sample by its two winners. In 1992 the share of counties that were divided was 32.5% in the Census Bureau's South and 29.9% in the rest of the country. In 2004 the South's share was higher, at 30.5% against 22.2%. In 2012 and 2016 it was lower, at 7.6% against 11.2% and 5.1% against 9.6%. In 2024 the two shares were 2.9% and 2.7%. The South stood out in some elections and not in others, in both directions.

Figure 5

Counties with different presidential and House winners were spread across regions in 1992, 32.5% of the South's counties and 29.9% of the other regions' counties

Counties by the party that won the presidential vote and the party that won the House vote, counties in contested House races, each presidential election from 1992 to 2024

Loading county values

Winners of the presidential and House vote

No data (3,143 counties)

Select a class to show only those places.

Find a county

Hover or tap a county for its value, or search by name. Zoom with the buttons on the map, then drag to move it.

Alaska draws as no data. Its election returns are reported by election district, and the district codes cannot be matched to boroughs and census areas.

In 1992 counties of both kinds are spread across the map, in the South and outside it; by 2024 nearly every county in the sample has the same winner for both offices. A county in the sample is colored by its two winners: neutral when the same party won both, blue when a Democrat won the House vote in a county the Republican carried for president, red for the reverse. Move the slider to change the election.Note: Full counts of votes, so the county series have no sampling interval.Note: Only counties in the main sample are colored: contested races in every district part, totals screen passed. Counties in uncontested races, in states without a Senate race (Senate view), counties with no usable returns and Jackson County, Missouri, from 2004 to 2024 (the file omits the part in Kansas City) draw as no data. Alaska and Hawaii are not in the county file. The South is the Census Bureau's South region.Source: County returns for President, U.S. House and U.S. Senate, 1992-2024 (Heideman 2026, with MIT Election Data and Science Lab presidential returns); full counts, unweighted

County margins converged on the presidential margin

A second view of the same counties compares margins and not winners. Figure 6 plots the correlation across counties between the two-party margin for president and the margin for the House, weighted by presidential votes. It was +0.65 (95% CI 0.56 to 0.72) in 1992 and +0.82 (95% CI 0.76 to 0.86) in 2008. It reached +0.95 (95% CI 0.93 to 0.96) in 2012, +0.99 (95% CI 0.98 to 0.99) in 2020 and +0.99 (95% CI 0.99 to 0.99) in 2024. The county-equal correlation, which counts each county once, was +0.58 (95% CI 0.47 to 0.67) in 1992 and +0.98 (95% CI 0.98 to 0.99) in 2024. For the Senate the correlation was +0.63 (95% CI 0.47 to 0.79) in 1992 and +0.98 (95% CI 0.97 to 0.99) in 2024. The intervals resample whole states, so they show how much each value depends on which states are in the sample.

Figure 6

The correlation between county presidential and House margins rose from +0.65 in 1992 to +0.99 in 2024

Vote-weighted correlation across counties between the Democratic minus Republican margin for president and for the House (or Senate), counties in contested races, 1992 to 2024; 95% intervals

Show data

Data for The correlation between county presidential and House margins rose from +0.65 in 1992 to +0.99 in 2024
Office	Measure	Election	Estimate	Lower	Upper	Counties
House	correlation	1992	0.65	0.56	0.72	2,735
House	gap	1992	16.36	14.55	18.69	2,735
House	correlation	1996	0.79	0.73	0.84	2,703
House	gap	1996	16.53	14.96	18.33	2,703
House	correlation	2000	0.74	0.66	0.80	2,555
House	gap	2000	17.98	16.05	19.89	2,555
House	correlation	2004	0.78	0.69	0.84	2,479
House	gap	2004	15.62	13.82	17.68	2,479
House	correlation	2008	0.82	0.76	0.86	2,661
House	gap	2008	13.49	12.08	15.24	2,661
House	correlation	2012	0.95	0.93	0.96	2,609
House	gap	2012	7.63	6.79	8.53	2,609
House	correlation	2016	0.93	0.89	0.96	2,427
House	gap	2016	8.73	7.58	10.01	2,427
House	correlation	2020	0.99	0.98	0.99	2,884
House	gap	2020	4.32	3.77	4.91	2,884
House	correlation	2024	0.99	0.99	0.99	2,695
House	gap	2024	3.70	3.24	4.22	2,695
Senate	correlation	1992	0.63	0.47	0.79	1,824
Senate	gap	1992	15.93	12.12	21.09	1,824
Senate	correlation	1996	0.84	0.76	0.91	2,288
Senate	gap	1996	10.89	7.78	14.96	2,288
Senate	correlation	2000	0.82	0.72	0.88	1,858
Senate	gap	2000	13.89	10.78	18.45	1,858
Senate	correlation	2004	0.78	0.67	0.87	2,006
Senate	gap	2004	18.05	13.51	23.38	2,006
Senate	correlation	2008	0.87	0.80	0.92	2,055
Senate	gap	2008	12.06	8.82	15.77	2,055
Senate	correlation	2012	0.93	0.88	0.97	1,843
Senate	gap	2012	8.71	5.66	12.50	1,843
Senate	correlation	2016	0.93	0.91	0.96	1,989
Senate	gap	2016	10.11	7.72	13.04	1,989
Senate	correlation	2020	0.99	0.98	0.99	2,212
Senate	gap	2020	3.93	3.24	4.90	2,212
Senate	correlation	2024	0.98	0.97	0.99	1,749
Senate	gap	2024	4.57	3.44	6.04	1,749

Both lines rise toward one, though not at every step, with the House line at +0.95 in 2012 and +0.99 in 2024. Each point is a correlation across the counties in the sample, weighted by presidential two-party votes. Bands are 95% intervals from a bootstrap that resamples whole states. The second view plots the average size of the gap between the two margins, in points.Note: County returns are full counts. The intervals show how much the correlation depends on which states are in the sample: whole states are resampled in a bootstrap, and the methods note gives the replicates and seed.Note: The gap view plots the vote-weighted mean absolute difference between the down-ballot margin and the presidential margin, in points of margin.Source: County returns for President, U.S. House and U.S. Senate, 1992-2024 (Heideman 2026, with MIT Election Data and Science Lab presidential returns); full counts, unweighted

A correlation measures whether counties keep their order, and it would be high if every county's House margin were its presidential margin plus a constant. The second view of the figure plots the size of the gap in points. The vote-weighted average absolute gap between the House margin and the presidential margin was 16.4 (95% CI 14.5 to 18.7) points in 1992 and 3.7 (95% CI 3.2 to 4.2) points in 2024. For the Senate it was 15.9 (95% CI 12.1 to 21.1) points in 1992 and 4.6 (95% CI 3.4 to 6.0) points in 2024.

Most of the gap is a shift shared by the counties of one district or state

The gap between the down-ballot margin and the presidential margin can be split into the average gap of the district or state a county sits in, and the variation among counties inside it. Figure 7 plots the root mean square of the gap, which weights large gaps more than the average absolute gap does, and the root mean square of what remains once each district's or state's average is removed. The House gap, for counties wholly inside one district, was 27.0 points in 1992 and 5.5 points in 2024, and the part that varies among counties inside a district was 9.1 and 2.3. The district average is 89% of the mean square of the gap in 1992 and 83% in 2024, and it is between 82% and 94% in every election. For the Senate the gap in 2024 was 6.1 points and 2.5 points of it varied inside states; the state average is between 64% and 91% of the mean square across elections and 83% in 2024.

Figure 7

A shift shared by the counties of one district or state accounts for most of the gap between presidential and down-ballot margins

Vote-weighted root mean square of the down-ballot margin minus the presidential margin, in points of margin, and the part that varies among counties inside one district (House) or state (Senate), 1992 to 2024

Root mean square gap (points of margin)

Show data

Data for A shift shared by the counties of one district or state accounts for most of the gap between presidential and down-ballot margins
Series	Election	Root mean square (points)	Counties
House, whole gap	1992	27.0	2,348
House, variation inside a district	1992	9.1	2,348
House, whole gap	1996	23.4	2,329
House, variation inside a district	1996	7.3	2,329
House, whole gap	2000	28.4	2,190
House, variation inside a district	2000	7.1	2,190
House, whole gap	2004	27.8	2,078
House, variation inside a district	2004	7.4	2,078
House, whole gap	2008	24.8	2,280
House, variation inside a district	2008	7.4	2,280
House, whole gap	2012	12.5	2,229
House, variation inside a district	2012	5.0	2,229
House, whole gap	2016	14.3	2,037
House, variation inside a district	2016	6.0	2,037
House, whole gap	2020	6.3	2,445
House, variation inside a district	2020	2.6	2,445
House, whole gap	2024	5.5	2,277
House, variation inside a district	2024	2.3	2,277
Senate, whole gap	1992	20.5	1,823
Senate, variation inside a state	1992	8.0	1,823
Senate, whole gap	1996	15.2	2,288
Senate, variation inside a state	1996	5.7	2,288
Senate, whole gap	2000	18.5	1,858
Senate, variation inside a state	2000	6.0	1,858
Senate, whole gap	2004	22.7	2,006
Senate, variation inside a state	2004	6.9	2,006
Senate, whole gap	2008	16.0	2,055
Senate, variation inside a state	2008	6.6	2,055
Senate, whole gap	2012	12.6	1,843
Senate, variation inside a state	2012	4.5	1,843
Senate, whole gap	2016	13.4	1,989
Senate, variation inside a state	2016	8.0	1,989
Senate, whole gap	2020	5.5	2,212
Senate, variation inside a state	2020	3.0	2,212
Senate, whole gap	2024	6.1	1,749
Senate, variation inside a state	2024	2.5	1,749

The solid lines are the whole gap and the dashed lines what remains after each district's or state's average gap is removed, so the distance between a solid and a dashed line is the shared shift. In 2024 the root mean square of the House gap is 5.5 points and the part that varies inside districts is 2.3 points; the Senate figures are 6.1 and 2.5. The House lines use counties wholly inside one district.Note: Full counts of votes, so the county series have no sampling interval.Note: A root mean square is the square root of the vote-weighted mean of the squared gap, so large gaps count more than in an average of absolute gaps. Districts and states with one county in the sample are left out, and the part inside each unit is multiplied by n over n minus one for a unit of n counties, because a unit's own average absorbs part of its counties' deviations.Source: County returns for President, U.S. House and U.S. Senate, 1992-2024 (Heideman 2026, with MIT Election Data and Science Lab presidential returns); full counts, unweighted

These shares use districts and states with more than one county in the sample, and they correct the part inside each unit for the number of counties, because a unit's own average absorbs part of its counties' deviations. Without the correction the House share is between 84% and 94%. A county's down-ballot result therefore differs from its presidential result mostly by an amount that its neighbors in the same district or state share. The data describe where the gap lies and not what produces it.

Voters and counties fell together from 2008 to 2012 and not at every step

Figure 8 puts the two measures on one scale for the nine presidential elections. The share of ANES voters who split between president and House fell from 22.3% in 1992 to 4.7% in 2024, a fall of 17.7 points, and the share of divided counties fell by 28.2 points. Both fell between 2008 and 2012, from 16.7% to 9.8% for voters and from 26.6% to 9.7% for counties. They then part. The voter rate rose to 15.3% in 2016, the House-item rise of the mode section, while the county share fell to 7.7%.

Figure 8

Between 1992 and 2024 the share of ANES voters who split their ticket fell by 17.7 points and the share of divided counties by 28.2 points

Share of ANES voters whose presidential and House votes were for different parties (95% intervals), and share of counties in contested House races with different presidential and House winners, 1992 to 2024

Show data

Data for Between 1992 and 2024 the share of ANES voters who split their ticket fell by 17.7 points and the share of divided counties by 28.2 points
Measure	Election	Share (%)	Respondents or counties
Voters who split their ticket (ANES)	1992	22.3	1,126
Counties with different winners	1992	31.0	2,735
Voters who split their ticket (ANES)	1996	17.7	926
Counties with different winners	1996	25.6	2,703
Voters who split their ticket (ANES)	2000	18.3	882
Counties with different winners	2000	27.0	2,555
Voters who split their ticket (ANES)	2004	16.8	701
Counties with different winners	2004	25.5	2,479
Voters who split their ticket (ANES)	2008	16.7	1,222
Counties with different winners	2008	26.6	2,661
Voters who split their ticket (ANES)	2012	9.8	3,501
Counties with different winners	2012	9.7	2,609
Voters who split their ticket (ANES)	2016	15.3	2,138
Counties with different winners	2016	7.7	2,427
Voters who split their ticket (ANES)	2020	11.4	5,167
Counties with different winners	2020	3.0	2,884
Voters who split their ticket (ANES)	2024	4.7	3,515
Counties with different winners	2024	2.8	2,695

Both lines fall between 2008 and 2012 and are lowest in 2024, and they do not move together at every step: the voter line rises in 2016 while the county line keeps falling. The two lines are shares of different things, voters and counties, plotted on one percent scale.Note: The voter line is the ANES rate of self-reported splitting among respondents with a major-party vote for president and the House. The county line is a share of counties and has no sampling interval.Source: ANES Cumulative Data File, 1992-2024, weighted (post-election weight where the file has one, VCF0009z elsewhere); county returns for President and U.S. House, 1992-2024, full counts, unweighted

The two lines are shares of different things and cannot be derived from each other. A county can be divided with few voters splitting, if its two races were close and drew different electorates, and a county can have a high rate of splitting with the same winner in both races, if splits run in both directions. The ANES records a respondent's state and no county, so it cannot say which kinds of place have the most individual splitting.

Which counties are in the comparison

Figure 9 lists the counties and the votes behind each main sample, with the share of divided counties for the House and the Senate. The House sample holds between 71% and 84% of the presidential two-party votes of the contiguous states, and 81% in 2024. The rest are in counties with an uncontested race, no usable returns in the file or a failed totals screen. The Senate sample holds between 49% and 74% because only states with a contested regular Senate race enter. Keeping the counties in uncontested Senate races changes the share of divided counties by at most 1.0 points in any election.

Figure 9

The main House sample holds between 71% and 84% of the presidential votes of the contiguous states and the main Senate sample between 49% and 74%

Counties, share of the 48 contiguous states' presidential two-party votes, and share of divided counties in the main House and Senate samples, by election

Election	Counties with presidential votes	House sample, counties	House sample, share of presidential votes (%)	House sample, divided counties (%)	Senate class	Senate sample, counties	Senate sample, share of presidential votes (%)	Senate sample, divided counties (%)
1992	3,109	2,735	77	31.0	3	1,824	68	30.9
1996	3,108	2,703	83	25.6	2	2,288	53	20.2
2000	3,107	2,555	71	27.0	1	1,858	74	21.1
2004	3,108	2,479	71	25.5	3	2,006	72	29.6
2008	3,108	2,661	77	26.6	2	2,055	49	24.4
2012	3,108	2,609	79	9.7	1	1,843	73	18.2
2016	3,107	2,427	71	7.7	3	1,989	62	6.2
2020	3,107	2,884	84	3.0	2	2,212	51	1.9
2024	3,107	2,695	81	2.8	1	1,749	73	3.0

Show data

Data for The main House sample holds between 71% and 84% of the presidential votes of the contiguous states and the main Senate sample between 49% and 74%
house_counties	house_div	house_votes	pres_counties	senate_class	senate_counties	senate_div	senate_votes	year
2,735	31.04	77.48	3,109	3	1,824	30.92	67.66	1992
2,703	25.60	82.93	3,108	2	2,288	20.24	52.59	1996
2,555	27.05	71.45	3,107	1	1,858	21.10	73.66	2000
2,479	25.45	70.70	3,108	3	2,006	29.61	72.02	2004
2,661	26.61	77.43	3,108	2	2,055	24.38	48.56	2008
2,609	9.66	78.75	3,108	1	1,843	18.18	73.41	2012
2,427	7.75	71.22	3,107	3	1,989	6.23	62.27	2016
2,884	3.02	83.82	3,107	2	2,212	1.94	51.32	2020
2,695	2.82	81.34	3,107	1	1,749	3.03	73.25	2024

Each row gives the counties with presidential votes, the counties in each main sample, the share of the presidential two-party votes cast in them and the share of those counties that are divided. The Senate class groups the elections in which the same states vote, every twelve years, so rows with the same class compare the same states.Note: Full counts of votes, so the county series have no sampling interval.Note: A Senate class is the set of seats that come up together every six years. In presidential years the three classes alternate, so a class comes up in 1992, 2004 and 2016 (class 3), 1996, 2008 and 2020 (class 2) and 2000, 2012 and 2024 (class 1). The Senate sample is the states with a contested regular race.Source: County returns for President, U.S. House and U.S. Senate, 1992-2024 (Heideman 2026, with MIT Election Data and Science Lab presidential returns); full counts, unweighted

What these data cannot show

County returns cannot identify who splits a ticket. A county in which the two winners agree can contain many voters who split in opposite directions, and a divided county can arise from close races and unequal turnout with few splitters. Robinson [6] showed that a correlation across places can differ in size and sign from the relationship among the individuals who live in them. The ANES has state geography only, so it cannot say where individual splitting is highest. Amlani and Algara [7] study nationalization in county presidential, senatorial and gubernatorial returns from 1872 to 2020, a longer period and a different measure than the comparisons here.

The ANES measure is self-reported. Respondents describe their votes after the election, and the file has no validated vote. The ANES voters in the 2024 sample are 55.3% (95% CI 53.0% to 57.7%) Democratic in the major-party presidential vote against 49.2% in the certified national count, so self-reported votes lean Democratic. A lean that differed between voters who split and voters who did not would change the rate of splitting, and the chapter cannot say whether it does. The samples are respondents who reported a major-party vote in both races, so voters who chose a third-party candidate, left a race blank or did not vote are out. The Cooperative Election Study file used in this book has no House, Senate or governor vote choice, so it gives no second individual-level source for 2006 to 2024. The full cumulative file has those items and would add that check, but it is not part of this build.

The chapter describes and does not explain. Incumbency, candidate quality and redistricting are possible reasons for the fall in divided counties and for the shared shifts by district and state, and nothing here tests them. The county fall from 2008 to 2012 comes between two elections separated by the 2010 midterm and by redistricting after the 2010 census, and timing alone does not say which of these matters, if any. Chapter 14.5, Candidate extremity and vote share (planned) asks whether more extreme House candidates lose vote share. The county file holds only president, House and Senate, so it says nothing about governors, state legislators or local offices, where Kuriwaki [5] finds more splitting than in Congress.

The county comparison depends on its rules and on the file. The main sample leaves out uncontested races, and uncontested seats are not a random draw of places. The file has defects that the rules remove or work around, such as the missing Kansas City part of Jackson County, runoffs stored as general elections and an Iowa file that double counts votes, and others may remain. The House samples use different district maps in different decades, the Senate samples use different states in different elections, and presidential margins leave out third-party votes, which were a larger share of the vote in 1992 and 1996 than in later elections. Alaska and Hawaii are not in the county file. County returns are full counts and carry no sampling interval. The intervals on the correlations treat states as independent, and the shared-shift decomposition describes where the gap lies without saying why it lies there.

Data sources

ANES Cumulative Data FileCumulative Data File 1948-2024: president, House and Senate vote items (VCF0704, VCF0707, VCF0708), combined item VCF0709 for 1952 to 2020; post-election weight; mode weights for 2008, 2012, 2016 and 2024Years: 1948-2024 (33 study years; even years only after 2004; no 2006, 2010 or 2014 study) · Weights: weight_postelection (VCF9999) where the file has it and VCF0009z elsewhere (VCF0009z equals 1 for every respondent in 1952, 1956, 1964, 1968, 1972, 1980, 1984 and 1988); weight_ftf and weight_web for the mode split · License: Free download after accepting the ANES terms of use; cite the release.
County election returnsCounty returns for President, U.S. House and U.S. Senate, 1992-2024, 48 contiguous states (Heideman 2026, summary and candidate-level files)Years: President 1868-2024 (40 elections); House and Senate 1990-2024 · Weights: none (full counts); vote-weighted statistics weight each county by its presidential two-party votes · License: The Heideman file is CC BY 4.0 (doi:10.5281/zenodo.22976553), but the state records behind it keep their own terms, and the sites of Florida, Indiana, Colorado and Wisconsin restrict reuse.
State presidential returnsState presidential returns, 2024, certified national Democratic share of the two-party vote (benchmark for the ANES 2024 sample)Years: 1976-2024 (13 elections) · Weights: none (full counts) · License: See the provider's terms of use.

All data sources

Methods

Individuals, sample. ANES Cumulative Data File respondents in the 19 presidential-election studies from 1952 to 2024 who reported a vote for the Democratic or Republican presidential candidate (VCF0704 coded 1 or 2) and for the Democratic or Republican candidate for the House (VCF0707 coded 1 or 2) and who have a weight: 29,103 respondents. The Senate series uses VCF0708 and covers 20,334 respondents in states with a Senate race. Voters for other presidential candidates, respondents who skipped a race and non-voters are left out. Votes are self-reported; the file has no validated vote. Individuals, estimand. A respondent splits when the two reported votes are for different parties. The yearly estimate is the weighted share who split, in percent. The two directions are the weighted shares who voted Democratic for president and Republican for the House and the reverse; the net effect is the weighted mean of the House Democratic vote minus the presidential Democratic vote, in points. The president and House split is rebuilt from VCF0704 and VCF0707 in every study because VCF0709 (the combined item) is missing for 2024. In the 18 studies from 1952 to 2020 the rebuilt code equals VCF0709 for all 25,637 respondents with both, whatever their weight; 2 respondents in 1984 have VCF0709 but a missing vote item and are left out. Weights and intervals. The vote items are post-election, so the all-interview estimates use weight_postelection (VCF9999) in the studies that have it (1996, 2000, 2004, 2008, 2016, 2020 and 2024) and VCF0009z in the others, including 2012, which has no post-election weight in the file. VCF0009z equals one for every respondent in 1952, 1956, 1964, 1968, 1972, 1980, 1984 and 1988, so those studies are unweighted. 64 respondents with both votes in 2016, 2020 and 2024 lack the post-election weight and are left out. Intervals are 95% intervals from the Kish effective sample size with a normal approximation, which treats the weighted sample as a simple random sample and ignores the strata and clusters of the sampling design. Differences between two estimates (two studies, or the two interview modes in one study) combine the two standard errors in quadrature. A yearly cell needs at least 100 respondents and an effective sample of at least 100; no cell was dropped. Mode. The in-person and web estimates use weight_ftf and weight_web (VCF0009x and VCF0009y), the full-sample in-person and web weights, because the cumulative file has no post-election mode weights; the all-interview and mode estimates for one study therefore use different weights. The 2008 study was interviewed in person only, and its in-person estimate is used for the 2008 to 2012 comparison among in-person respondents. The 2020 study has no mode weights, so it enters only the all-interview series. The 2024 mode weights cover the newly recruited in-person and web samples; the 1,709 panel and paper respondents with both votes are in the all-interview estimate and in neither mode. Places, sample. County presidential, House and Senate returns for the nine presidential elections from 1992 to 2024 in the 48 contiguous states (Heideman 2026, version 1.0.0; the presidential rows come from the same file). Oglala Lakota County, South Dakota, appears under two codes (46113 and 46102); the build uses 46102. Rule 1: a county needs presidential two-party votes and returns for the office, and is left out for an exact tie in either race; for Jackson County, Missouri, in 2004, 2008, 2012, 2016, 2020 and 2024, because the file's county rows leave out the part of the county in Kansas City (the file's presidential two-party votes for the county are 55% to 59% of the project's other presidential file in 2004 to 2020, and in 2024 that file reports Kansas City separately); for the Senate races in which the file codes an independent as a Democrat (Maine and Vermont 2012, Nebraska 2024), because the same candidates are coded as independents in other years (123 county-years); and for a House county flagged excludes_unopposed_seat (57 county-years). Rule 2, the totals screen: the office's total votes in the county are between 0.7 and 1.3 times its presidential total votes and the row is not flagged totals_possibly_inflated. It removes 687 of 27,567 House county-years with returns and 398 of 18,645 Senate county-years, among them Iowa 2012 House (99 counties, each vote counted twice), Georgia Senate 1992 and 2008 (158 and 157 counties; the file holds a later runoff round) and Louisiana Senate 1992 (61 counties; the October primary); of the other 588 House and 22 Senate county-years, 150 House county-years would otherwise be in the main sample. Dropping the screen changes the yearly share of divided counties by at most 0.5 points for the House and 1.3 for the Senate. Rule 3, a two-party race: the Democratic and Republican candidates together won at least 80% of the votes in the race (fails in 14 House races and 1 Senate race that have candidates of both parties). A House race is a district (a county can hold parts of several districts, and every district part has to qualify); the district has to be identified in the file and cannot have been decided in a later round. Where the file gives no district (Nevada and Washington in 2000 to 2008, parts of Iowa, Arizona, Delaware and New York), the county counts as one block that qualifies when the two parties hold at least 80% of its votes and as contested when it lists as many Democratic as Republican candidates, at least one of each. A Senate race is the state; special Senate elections are not in the file. Rule 4, a contested race (the main sample): both the Democratic and the Republican candidate received votes, in every district part for the House. Among the county-years that pass rules 1 and 2, rules 3 and 4 remove 3,097 House and 295 Senate county-years. The sensitivity sample, all returns, applies rules 1 to 3 and keeps counties in uncontested two-party races. Places, estimands. The margin is the Democratic minus the Republican votes over their sum. A county is divided when the sign of the presidential margin differs from the sign of the down-ballot margin. The share of counties counts each county once; the share of votes weights each county by its presidential two-party votes. The correlation across counties is weighted by presidential two-party votes (the county-equal version is in the text). The gap is the down-ballot margin minus the presidential margin in points; the average absolute gap and the root mean square are vote-weighted. The shared-shift decomposition subtracts each district's (House, counties wholly inside one district) or state's (Senate) vote-weighted mean gap and compares the mean squares before and after; districts and states with one county are left out, and the part inside each unit is multiplied by n over n minus one for a unit of n counties. Places, intervals. County returns are full counts, so shares of counties and votes carry no sampling interval. The intervals on the correlation and the average gap are percentile intervals from a bootstrap that resamples whole states (2000 replicates, seed 12345), which shows how much each statistic depends on which states are in the sample and treats states as independent. Benchmark and region. The certified 2024 Democratic share of the two-party presidential vote is computed from the state presidential returns for all states and the District of Columbia. The South is the Census Bureau's South region among the contiguous states: AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA and WV.

References

Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Burden, Barry C.; Kimball, David C. (2002). Why Americans Split Their Tickets: Campaigns, Competition, and Divided Government. University of Michigan Press. doi:10.3998/mpub.22995
Jacobson, Gary C. (2015). It's Nothing Personal: The Decline of the Incumbency Advantage in US House Elections. The Journal of Politics 77(3): 861-873. doi:10.1086/681670
Hopkins, Daniel J. (2018). The Increasingly United States: How and Why American Political Behavior Nationalized. University of Chicago Press. doi:10.7208/chicago/9780226530406.001.0001
Kuriwaki, Shiro (2026). Ticket Splitting in a Nationalized Era. The Journal of Politics 88(1): 47-62. doi:10.1086/734263
Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review 15(3): 351-357. doi:10.2307/2087176
Amlani, Sharif; Algara, Carlos (2021). Partisanship & nationalization in American elections: Evidence from presidential, senatorial, & gubernatorial elections in the U.S. counties, 1872-2020. Electoral Studies 73: 102387. doi:10.1016/j.electstud.2021.102387

How to cite

@misc{singh2026downtheballot,
  author = {Singh, Vishal},
  title = {Down the ballot},
  year = {2026},
  url = {https://vishalsingh.org/politics/down-the-ballot},
  note = {In The Public Mind. Published 3 October 2026.}
}

Self-reported splitting was highest from 1972 to 1988 and lowest in 2024

The 2016 rise appears among web respondents in the House item

From 1972 to 1988 the usual split was a Republican president and a Democratic House

The share of divided counties fell from 2008 to 2012 and again by 2020

Divided counties mostly had a Republican president and a Democratic House, and regions differed by election

County margins converged on the presidential margin

Most of the gap is a shift shared by the counties of one district or state

Voters and counties fell together from 2008 to 2012 and not at every step

Which counties are in the comparison

What these data cannot show

Data sources

Methods

References

How to cite