Party, belief and the political map

Among respondents to the American National Election Studies (ANES) who answered both the party-identification and the liberal-conservative questions, the weighted correlationCorrelationA number from minus 1 to plus 1 for how closely two measures move together in a straight-line pattern. Zero means no straight-line pattern; a correlation does not say that one measure causes the other. between the two answers rose from +0.32 (95% CI 0.27 to 0.36)+0.32 correlation coefficient, minus 1 to plus 1Estimandweighted Pearson correlation between party identification and liberal-conservative self-placementInterval95% CI 0.27 to 0.36n1,545Weightsweight (VCF0009z)PopulationANES respondents with a party identification (leaners kept) and a liberal-conservative self-placementPeriod1972SourceANES Cumulative Data File in 1972 to +0.72 (95% CI 0.70 to 0.74)+0.72 correlation coefficient, minus 1 to plus 1Estimandweighted Pearson correlation between party identification and liberal-conservative self-placementInterval95% CI 0.70 to 0.74n4,716Weightsweight (VCF0009z)PopulationANES respondents with a party identification (leaners kept) and a liberal-conservative self-placementPeriod2024SourceANES Cumulative Data File in 2024. Those respondents numbered 1,5451,545 respondentsEstimandcount of respondents who answered both questionsWeightsunweighted count of respondentsPopulationANES respondents with a party identification (leaners kept) and a liberal-conservative self-placementPeriod1972SourceANES Cumulative Data File in 1972 and 4,7164,716 respondentsEstimandcount of respondents who answered both questionsWeightsunweighted count of respondentsPopulationANES respondents with a party identification (leaners kept) and a liberal-conservative self-placementPeriod2024SourceANES Cumulative Data File in 2024. A correlation describes how closely two answers move together; a higher positive value here means that a more Republican identification goes more consistently with a more conservative placement. The intervals describe model-based sampling uncertainty, not every source of survey error.The estimates come from the ANES Cumulative Data File, weighted with the cumulative-file weight, and are built in Chapter 1.1, Party and ideology. Each interval uses the Fisher z transformation [1] with the Kish effective sample size [2] and does not incorporate the full survey design. Table 1 plots the correlation in every survey year, for three surveys.

Table 1. The party-ideology correlation ended higher than it began on all three surveys

The increase is one starting point for this book. It shows that knowing someone's party tells us more about their ideological label than it did in the earlier survey. It does not establish that their views became more extreme, that their positions on different issues became identical to their party's, or that party attachment caused their beliefs to change. Those are separate questions.

The Public Mind studies how political identities, beliefs and electoral choices align, and where they remain distinct. The American evidence is its center. Comparisons across countries test whether relationships familiar in the United States hold elsewhere. Election returns and campaign-finance records extend the inquiry to the places where votes are cast and the people who give money to political campaigns.

Political alignment has several meanings

Party identification is a person's attachment to a political party. Ideological self-placement is the label a person chooses for their views. A policy preference is an answer to a more specific question, such as what government should provide or what the law should allow. These measures often move together, but they describe different things.

Partisan sortingParty sortingThe alignment of party with other traits, such as ideology, religion, race and place, so that people who share a party also tend to share those traits. means that political characteristics become more consistently associated with party. It can occur when people change parties while keeping their opinions. It can also occur when their opinions change, or when the composition of the groups changes. A series of surveys interviewing different people can describe the resulting alignment without identifying which process produced it.

Issue polarization concerns the distance between positions on policy. Affective polarizationAffective polarizationThe gap between how warmly people feel toward their own party and how coldly they feel toward the other party. It measures feeling, not disagreement on issues. concerns the difference between feelings toward one's own party and the other party. A person can dislike the opposing party while holding some views in common with its supporters. An ideological moderate can be a committed partisan. A county that divides its vote closely can contain reporting units with very different margins.

The book keeps these distinctions visible. Its chapter on economic and cultural dimensions finds that the correlation between its economic and cultural indices rose from +0.10+0.10 correlation coefficient, from −1 to +1Estimandweighted Pearson correlation of the economic and cultural conservatism indices (Fisher z interval)Interval95% CI 0.05 to 0.15n1,508Weightsweight (VCF0009z)Populationadults with at least two of three items answered on each indexPeriod1984SourceANES Cumulative Data File in 1984 to +0.51+0.51 correlation coefficient, from −1 to +1Estimandweighted Pearson correlation of the economic and cultural conservatism indices (Fisher z interval)Interval95% CI 0.48 to 0.55n4,387Weightsweight (VCF0009z)Populationadults with at least two of three items answered on each indexPeriod2024SourceANES Cumulative Data File in 2024 and is still below one, at +0.69+0.69 correlation coefficient, from −1 to +1Estimandobserved correlation divided by the square root of the product of the two alphas (bootstrap percentile interval)Interval95% CI 0.65 to 0.74n4,387Weightsweight (VCF0009z)Populationadults with at least two of three items answered on each indexPeriod2024SourceANES Cumulative Data File, after correction for measurement error. Its chapter on partisan feeling finds a long-run fall in ratings of the other party, from 48.048.0 points on the 0 to 100 feeling thermometerEstimandweighted mean rating (95% interval from the Kish effective sample size)Interval95% CI 47.1 to 48.9n1,723Weightsweight (VCF0009z)Populationpartisans, leaners included, with a rating of both partiesPeriod1978SourceANES Cumulative Data File in 1978 to 20.620.6 points on the 0 to 100 feeling thermometerEstimandweighted mean rating (95% interval from the Kish effective sample size)Interval95% CI 19.8 to 21.5n5,080Weightsweight (VCF0009z)Populationpartisans, leaners included, with a rating of both partiesPeriod2024SourceANES Cumulative Data File in 2024 on the feeling thermometer, while the average own-party rating ends near its starting level. The recent movement in those ratings also depends on which interview modes are compared.

The American alignment developed on different schedules

The first volume, Labels, sorting and partisan feeling, examines the changing relationship between party identity and ideology, what the labels contain, and partisan feeling. The second, Coalitions and political differences, follows the evidence through electoral groups, religion, issues, racial attitudes and institutional confidence.

The timing differs by subject. On abortion, the surveys show a widening party gap over their observed periods, from 3.53.5 percentage pointsEstimandweighted Democratic share minus weighted Republican share; the ANES four-category abortion question, answer 'always allowed as a matter of personal choice'Interval95% CI −1.8 to 8.8n1,313Weightsweight (VCF0009z; equals 1 for every respondent from 1980 to 1990, so those years are unweighted)PopulationDemocratic and Republican identifiers (leaners included) who answered the ANES abortion questionPeriod1980SourceANES Cumulative Data File points in 1980 to 52.352.3 percentage pointsEstimandweighted Democratic share minus weighted Republican share; the ANES four-category abortion question, answer 'always allowed as a matter of personal choice'Interval95% CI 48.8 to 55.9n4,666Weightsweight (VCF0009z; equals 1 for every respondent from 1980 to 1990, so those years are unweighted)PopulationDemocratic and Republican identifiers (leaners included) who answered the ANES abortion questionPeriod2024SourceANES Cumulative Data File in 2024 in the ANES. The economic-perception chapter describes a gap that reverses at each change of party in the presidency. Partisans are more likely to say that the national economy got better over the past year when their own party holds the presidency, by 34.034.0 percentage pointsEstimandmean of the yearly premiums, each survey counted once; the interval covers sampling error within each survey onlyInterval95% CI 32.5 to 35.4n20,874Weightsweight (VCF0009z)PopulationANES respondents who identify with or lean toward a party and answered the itemPeriod2012 to 2024 (four surveys)SourceANES Cumulative Data File points on average. A persistent difference on an issue and a difference that changes sign with control of government are different patterns, and the chapters describe each without explaining it.The chapters behind this paragraph are Chapter 4.1, The abortion sort, Chapter 4.7, Sort dates by attitude and Chapter 4.5, Perceptions of the economy.

The composition of political groups matters too. The income chapter finds a change in the comparison between the highest and lowest household-income brackets, from −9.1−9.1 percentage points of Democratic two-party vote shareEstimandDemocratic share in the highest bracket minus that in the lowest; weighted regression on the two brackets, state-clustered standard errorsInterval95% CI −12.9 to −5.4n5,534Weightsweight_cumulativePopulationhighest minus lowest bracket, two-party presidential voters who reported household incomePeriod2008SourceCooperative Election Study points in 2008 to +5.7+5.7 percentage points of Democratic two-party vote shareEstimandDemocratic share in the highest bracket minus that in the lowest; weighted regression on the two brackets, state-clustered standard errorsInterval95% CI 1.1 to 10.3n12,281Weightsweight_cumulativePopulationhighest minus lowest bracket, two-party presidential voters who reported household incomePeriod2024SourceCooperative Election Study in 2024, but the 2024 profile is U-shaped rather than a simple rising ladder. Within education groups the 2024 gap cannot be told apart from zero, although it still rose after 2008. A straight line summarizes a U poorly, so a description of the parties as the parties of the rich or the poor leaves out that structure.See Chapter 3.5, The income ladder.

The first volume also asks how the alignment changed within birth cohorts. A cohortCohortPeople born in the same period, such as the same decade. Comparing cohorts shows whether a difference comes from when people were born, as distinct from their age or the year they were surveyed. is a group of people born in the same years. Its average can change as its members change their views, but also as people enter or leave the observed group. The cohort analysis therefore cannot identify individual switching. Its estimates of the contribution of replacement depend on assumptions about cohort rates.See Chapter 2.1, Cohort change in partisan sorting.

Left and right contain different combinations of beliefs

The first part of the book, Labels and beliefs, examines the content behind political labels. It distinguishes economic from cultural views and considers the relationship between basic values and ideological placement. The second volume studies religious attendance and affiliation.

Cross-national comparisons are useful because the American combination is only one observed combination. In the book's comparative analysis, left-right placement follows economic attitudes most closely in some countries, the importance of God in others and moral attitudes in others, and in the remaining countries no one attitude leads clearly. Even within post-communist Europe, conservation values, meaning the importance people give to conformity, tradition and security, can relate to the left in one country and the right in another. The post-communist average conceals opposite signs, so it does not support one uniform regional account.The chapters behind this paragraph are Chapter 1.3, Left and right across countries, Companion 1.3a, Basic values and left-right and Companion 1.3b, The obedience-autonomy disposition.

These comparisons also test the measures. Questions about the qualities desirable in a child differ from portraits used to measure personal values. Religious attendance differs from the importance of God. When related measures give similar results, that agreement is informative. It does not make the questions interchangeable or prove that they all measure a single disposition.

The comparative chapters written so far therefore qualify the American account. The book asks which relationships recur and which depend on the political label, the survey question or the country. The strength of an association and its direction are empirical questions, even when a theory predicts both.

Election maps describe distributions of votes

The third volume, Votes, giving and places, begins with county election returns and then examines finer reporting units, political donations, demographic characteristics, and local economic and environmental conditions. It ends by relating the geography of voting to individual opinion.

A county's margin describes its result in one election. Its swing describes the change between elections. Counties can swing by similar amounts while remaining far apart in their levels of support. The nationalization chapter measures the spread of those changes. The county-margin chapter measures distance from the national result within an election. The measures answer different questions.The chapters behind this paragraph are Chapter 6.1, Nationalization of the county vote and Chapter 6.2, Distance from the national vote, 1868 to 2024.

Table 2. The share of counties within 5 points of the national swing rose from 21.2% in 1972 to 1976 to 75.6% in 2020 to 2024

Moving below the county line reveals variation that a county average hides. The precinct chapter compares counties that finished near a tie but had different distributions of reporting-unit margins. A close county result does not tell us whether most units were close or whether more one-sided results balanced each other. Nor does a reporting-unit margin reveal the views of every resident; some reporting units are categories of ballots rather than geographic neighborhoods.See Chapter 6.3, Precincts within counties.

Campaign-finance records supply another comparison. Donors are a selected group, and their scores are inferred from giving rather than chosen as survey labels. The donor-pool chapter follows changes in who is active on a scale that assigns each donor a fixed score. The donor-voter chapter compares county pools drawn from a long record of giving with a particular election. Neither comparison directly measures how an individual donor voted.The chapters behind this paragraph are Chapter 7.1, The ideology of the donor pool and Chapter 7.2, Counties where donors and voters diverge.

People and places answer different questions

A relationship across places can differ from a relationship among the people who live in them. Across ZIP codes, for example, more capital-gains filers went with a swing toward the Democrats, while within education groups investors were more Republican than non-investors; with education held fixed, the two results agree in sign. The book places those comparisons alongside one another and identifies the different populations and periods they describe.See Companion 8.1b, Wages, capital and the ZIP code vote.

The final part ends by returning directly to survey respondents. Chapter 8.3, Party and place compares Democrats and Republicans living in the same kind of county, then compares members of the same party across strongly Democratic and strongly Republican counties. On every issue it examines (Table 3), the party gap is larger than the within-party place gap, averaging 43.143.1 points (difference in the share taking the Democratic side, on a 0 to 100 scale)Estimandaverage over the issues of the party gap: Democratic minus Republican share taking the Democratic side, averaged over deep-red and deep-blue countiesInterval95% CI 41.8 to 44.5n118,638Weightsweight_cumulative (CES cumulative weight; each survey year carries the same total weight)PopulationDemocrats and Republicans (three-point party identification, leaners excluded) in counties with a two-party presidential margin of 20 points or morePeriodCES survey years 2013 to 2021, each issue in the years it was askedSourceCooperative Election Study and 5.65.6 points (difference in the share taking the Democratic side, on a 0 to 100 scale)Estimandaverage over the issues of the place gap: deep-blue minus deep-red county share taking the Democratic side, averaged over Democrats and RepublicansInterval95% CI 4.7 to 6.6n118,638Weightsweight_cumulative (CES cumulative weight; each survey year carries the same total weight)PopulationDemocrats and Republicans (three-point party identification, leaners excluded) in counties with a two-party presidential margin of 20 points or morePeriodCES survey years 2013 to 2021, each issue in the years it was askedSourceCooperative Election Study points. The place differences remain observable, and their size depends on the issue and adjustment.

Table 3. On every issue the gap between parties is larger than the gap between deep-red and deep-blue counties

That comparison does not establish what moving to another county would do to someone's views. People select where they live, party attachments relate to other characteristics, and the measured county margin captures only one aspect of local context. Separating influence from selection requires different evidence.

Reading the evidence

Each chapter defines the comparison its figures make. Survey weightsSurvey weightsNumbers that scale each respondent so the sample matches the population on known traits such as age, race, education and region. A weighted average counts a respondent with a large weight for more than one with a small weight. adjust how much each respondent contributes so that the sample matches specified population characteristics, not every unobserved one. Confidence intervalsConfidence intervalA range around an estimate that would contain the true value in 95% of repeated samples if the sampling model held. The book reports 95% intervals unless it says otherwise. describe uncertainty under an estimator's assumptions and do not by themselves include errors from wording, nonresponse or selection into an online panel.

Election returns count recorded votes. Estimates allocated from precincts to other geographic areas add modeling assumptions. Historical returns also describe the people who could and did vote under the rules of their period. Campaign-finance files describe the contributions covered by their reporting and scoring procedures. These sources cannot be treated as equivalent samples of the public. The methods reference explains the recurring choices.

The chapters can be read in sequence, through the map-first path or by subject. Each provides its data sources, methods and limitations so that it can be read on its own. Planned chapters are identified as questions still to be investigated; their presence in the contents is not evidence of a finding.

What these data cannot show

The analyses in the current chapters describe relationships more often than they identify causes. They cannot by themselves determine whether party leaders changed voters' views, whether voters selected leaders who matched them, or whether both responded to other changes. Those limits apply to the interpretation of the patterns, even when the patterns themselves are measured precisely.