Chapter 17: Public Opinion — What Americans Believe, Why They Believe It, and How We Know

"The people, sir, are a great beast." — attributed to Alexander Hamilton, possibly apocryphal

"Public opinion is what politicians say it is — until an election proves them wrong." — anonymous campaign manager, 2024

Why this chapter exists

Part III of this textbook is about the people: how they form views, how they organize, how they participate, how they are mobilized and demobilized. We start with public opinion because everything downstream — voting, parties, campaigns, polarization, social movements — runs on the raw material of what Americans actually believe. Get that wrong and the rest of Part III becomes guesswork.

The question "what do Americans believe?" sounds simple. It is not. There are 260 million voting-age Americans. They believe many things, often inconsistent things, often things they have not thought carefully about, often things shaped by their family, neighborhood, schooling, religion, race, generation, and the media diet they happened to encounter last Tuesday. There is no single American public mind. There is a distribution of views, and our tools for measuring that distribution — surveys, polls, behavioral data, election outcomes — are imperfect instruments pointed at a moving target.

This chapter does five things. First, it defines what public opinion is and what it is not. Second, it explains how we measure it (briefly — Appendix F goes deeper). Third, it presents the empirical landscape of what Americans believe in 2026 on the questions that matter most. Fourth, it examines how those beliefs are formed — by family, education, geography, race, religion, generation, and media. Fifth, it engages the hardest question in the field: do voters reason about politics, or do they merely identify with groups? The honest answer is "both, in measurable proportions, with consequences."

We will not pretend public opinion is a clean window into a coherent national mind. We will not pretend polling perfectly captures what is there. And we will not pretend ideology means the same thing to a college-educated suburban voter as it does to a non-college rural voter or a third-generation South Asian American immigrant in Houston. Public opinion is messy. The measurement of it is messy. The job of this chapter is to teach you to read that messiness honestly.

17.1 What is public opinion?

Two definitions

A working definition: public opinion is the aggregated views of a population on matters of public concern, as expressed through some measurable channel.

That definition contains a lot of work. Let us unpack it.

"Aggregated" means we are not talking about any single person's view. Public opinion is a distribution — what percentage of Americans hold view X, what percentage hold view Y, what percentage have no view, what percentage hold both views in tension. The aggregate is a statistical object, not a moral one. The aggregate has no will, no rights, no preferences. People have those. The aggregate is just the sum.

"Matters of public concern" narrows the domain. Public opinion in the political sense is about what government should do, what the country's values should be, who should hold power, and how citizens should live together — not about whether pineapple belongs on pizza. (Though pollsters do ask that question. The answer is roughly 50–50, with significant regional variation. Florida favors. New York opposes.)

"Some measurable channel" is the methodological point. Public opinion has to be observed somehow, and how we observe it shapes what we find. The two main channels are:

1. Surveys and polls — direct questioning of a sample of the population. The numbers everyone cites: "62% of Americans favor X."
2. Revealed preferences — what people actually do when given a choice. Votes cast, dollars donated, organizations joined, content shared, products bought, neighborhoods chosen.

These two channels often disagree. Americans tell pollsters they support background checks for gun purchases at rates above 80%. The political coalition that consistently votes for federal background-check legislation is a much smaller share. Americans say they want lower deficits and fewer government services in the abstract. They vote for incumbents who deliver more services and bigger deficits. The gap between what people say and what people do is a permanent feature of public opinion.

The serious analyst uses both kinds of data. Polls measure attitudes; behavior measures revealed priority. Together they produce a richer picture than either alone.

The aggregation problem

Even granting that public opinion is a distribution, how you aggregate matters.

Consider this stylized example. On a hypothetical immigration policy question, suppose Americans split into four roughly equal groups:

• 25% favor expanded legal immigration AND tighter enforcement against unauthorized crossings.
• 25% favor expanded legal immigration AND a path to citizenship for current undocumented residents.
• 25% favor restricted legal immigration AND tighter enforcement.
• 25% favor restricted legal immigration AND a path to citizenship.

If a pollster asks one question — "Do you support expanded immigration?" — they get a 50–50 split. If they ask the other question — "Do you support a path to citizenship?" — they also get a 50–50 split. But the policy combinations are not 50–50; they are four different coalitions of about a quarter each, and no single policy package commands a majority. The "national view on immigration" that many headline numbers suggest does not exist. There are bundles of preferences, and most actual policy proposals match only some of them.

This is not a hypothetical. Real immigration polling shows precisely this pattern (Pew 2024, Gallup 2024). The lesson is general: the single number summary of public opinion can mask complete incoherence underneath. Always read polling at the level of the underlying cross-tabs, not just the headlines.

Public opinion vs. publicized opinion

A separate problem: the views you see in your media feed are not a representative sample of what Americans believe. They are filtered through the platforms' attention economy, through editors' news judgment, through the asymmetric loudness of the politically engaged. About 6% of Americans produce 75% of political content on Twitter/X (Pew 2019, methodology updated 2023). The voices you encounter in political conversation are systematically louder and more partisan than the modal American.

This is why pollsters exist. A poll, done well, samples the quiet alongside the loud. The 70% of Americans who post no political content on social media still answer phone surveys and online panels. They show up in the data. They do not show up in your feed.

Keep this in mind every time you find yourself thinking "everyone I know thinks X." The relevant question is not what your network thinks. It is what a representative sample of 250 million voting-age Americans thinks. Those are different populations.

17.2 How we measure public opinion (lightly)

This section gives the headline. Appendix F gives the full treatment. If you are going to read polls regularly, read Appendix F. Here we cover only what you need to follow this chapter.

Probability sampling

The fundamental theorem behind polling is this: if you select a sample at random from a population, the sample statistics will, with calculable uncertainty, approximate the population statistics. Random selection is the load-bearing assumption. If your sample is not random — if certain types of people are systematically more or less likely to be sampled — the statistics will not approximate the population.

In practice, no real-world sample is perfectly random. Every method has biases. Pollsters correct for these biases through weighting: counting some respondents more than others to make the demographic profile of the (weighted) sample match what the analyst believes the target population looks like. Weighting can fix some sampling problems and introduce others.

Sample frames

A sample frame is the list of people who could be sampled. The four main frames in modern American polling:

• Random Digit Dial (RDD) — random phone numbers, both landline and cell. Traditional gold standard. Response rates have collapsed to under 5%.
• Address-Based Sampling (ABS) — random U.S. addresses, contacted by mail. Used for some long-running academic surveys.
• Voter file samples — registered voters drawn from state voter rolls. Common for campaign and election polling.
• Online panels — pre-recruited groups who agreed to take surveys in exchange for incentives. Now the dominant frame for most public polling.

Each frame misses certain people (RDD misses households without phones; voter files miss the unregistered; online panels miss the offline) and the misses bias the result if the missing people differ systematically from the included ones.

Weighting

After a poll is in the field, the raw data are weighted to match population targets. Standard weighting variables: age, race/ethnicity, gender, education, region. Sometimes income, religion, partisanship. The choice of what to weight on and what target distribution to use is methodological judgment, not science. Different pollsters make different choices, which is a primary source of "house effects" — the consistent tilts that distinguish one pollster's results from another's.

The biggest methodological lesson of the past decade: education weighting matters enormously. Before 2016, most pollsters did not weight by education. The result was systematic under-representation of non-college-educated white voters, who broke heavily for Trump. Polls underestimated Republican support in the Upper Midwest by 3–5 points. After 2016, most pollsters weight by education. Some now also weight by past-vote recall. The corrections have helped but not eliminated the problem.

Modes

How the survey is conducted: phone (live or automated), online, mail, mixed. Each mode produces somewhat different results because the social context of being interviewed shapes the answers. Live phone respondents are more likely to give socially desirable answers (saying they vote when they don't, hiding stigmatized views). Online respondents, taking a survey alone in their browser, are more honest about embarrassing views but more likely to satisfice — picking answers without thinking carefully.

The bottom line for chapter 17 purposes: no single measurement approach is now the gold standard. A serious read of public opinion compares results across modes, frames, and pollsters. A single poll is one observation. The trend across many polls is the signal.

17.3 Major polling organizations

If you are going to read public-opinion data, you should know who is producing it. The major sources, in rough categories:

Academic gold-standard surveys

• American National Election Studies (ANES) — academic pre/post-election survey since 1948. The single most important data source for political-science research on American voters. Conducted in cooperation between the University of Michigan and Stanford. Long, careful interviews. Publicly available data files; a graduate-student rite of passage to learn the ANES.
• General Social Survey (GSS) — broader social attitudes survey since 1972, conducted by NORC at the University of Chicago. Used heavily for tracking long-term shifts in values.
• Cooperative Election Study (CES, formerly CCES) — large-N (60,000+) academic survey run by Harvard with university partners. Useful for state-level estimates and for sub-group analyses where smaller surveys don't have enough cases.

Major non-partisan pollsters

• Pew Research Center — the leading non-partisan source for issue and demographic polling. Methodologically careful, transparent, frequently updated.
• Gallup — the oldest American polling firm, conducting surveys since 1935. Long historical time series, especially for presidential approval and trust in institutions.
• Reuters/Ipsos — frequent issue and political polling.
• Quinnipiac, Marist, Monmouth, Suffolk, Siena, Marquette Law School, Emerson, Selzer & Co. — university-affiliated and independent pollsters with strong reputations. Selzer's Iowa Poll has had unusual accuracy across cycles, though every pollster has good and bad cycles.

Major newspaper and media polls

• New York Times/Siena College — collaboration that produces some of the highest-quality state-level polling in election years.
• Washington Post/ABC News — long-standing partnership.
• NBC News/Wall Street Journal (the partnership ended in 2024; both outlets continue separately).
• Fox News polling — run by a partnership of Beacon Research (Republican) and Shaw & Co. (Democratic). Despite the network's editorial lean, the polls themselves are mainstream and well-regarded.

Partisan and campaign-affiliated pollsters

• Public Policy Polling (PPP) — openly Democratic-affiliated.
• The Trafalgar Group — Republican-leaning, gained credibility for correctly projecting Trump's 2016 wins in PA and MI.
• Rasmussen Reports — historically polls 3–5 points to the right of the average; methodology is opaque.

A poll's sponsor is information. A campaign-internal poll released by the campaign is partisan messaging, not measurement. A Pew survey on healthcare attitudes is a genuine attempt to measure those attitudes.

17.4 What Americans actually believe

This section is the empirical heart of the chapter. We move through several topics where polling data are reliable, frequent, and important.

The role of government

Americans are split on the most basic question of political philosophy: should government do more, or should it do less?

The standard ANES question — "Some people think the government should provide many more services even if it means an increase in spending; other people think the government should provide many fewer services in order to reduce spending. Where would you place yourself?" — produces a distribution that has shifted modestly over decades but always falls within a bounded range. As of 2024:

• Around 45–50% of Americans favor government doing more (varies by question wording).
• Around 35–40% favor government doing less.
• The remainder are in the middle or have no strong view.

The partisan split is large and growing. About 75% of Democrats favor more government services; about 65% of Republicans favor fewer. The split by education and race compounds: college-educated voters and non-white voters are both more likely to favor more government, though for different reasons and with different substantive priorities.

Specific policies: where consensus exists

On certain specific policies, super-majorities cross partisan lines. Some examples drawn from Pew, Gallup, and ANES data over the past decade:

• Social Security and Medicare. Around 80–90% of Americans, including majorities of both parties, oppose cutting either program. Even in the Republican electorate that elects representatives who propose Social Security reforms, the rank-and-file does not want benefit cuts. This is one reason the "third rail" metaphor persists.
• Background checks for firearm purchases. Polling consistently shows 80–90% support for universal background checks, including 70%+ of Republicans and 70%+ of gun owners (Pew, Gallup, multiple cycles). Federal legislation has not passed despite this. The reasons — salience asymmetry, organizational mobilization, the legislative-process biases — appear in Section 17.10.
• Some immigration controls. Around 60–70% of Americans favor securing the border and reducing illegal immigration; this includes majorities of Hispanic Americans, who are often described in headline coverage as if they uniformly oppose enforcement, but who actually split closer to the national average on enforcement questions (Pew 2024).
• Some path to legal status for long-resident undocumented immigrants who came as children. Around 70% favor legal status for so-called "DREAMers" who arrived as minors and have no criminal record (Pew, multiple cycles). This is one of the few immigration questions where there is durable cross-party consensus.
• Marijuana legalization. Around 70% of Americans support legalization for recreational use; majorities of both parties, though Republicans lower (Pew 2023).
• Same-sex marriage. Around 71% of Americans support, including a majority of Republicans for the first time crossing in roughly 2021–2022 (Gallup).

Specific policies: where polarization dominates

On other issues, the partisan gap is enormous and growing.

• Abortion. The aggregate number — around 60% supporting Roe-style legal abortion in most cases — masks deep partisan and religious splits. Republicans oppose; Democrats favor; the intra-party variation is large but the inter-party gap is larger. Post-Dobbs (2022) has shifted attitudes modestly toward more permissive positions, but the partisan structure has held.
• Climate change. Around 75% of Americans believe climate change is occurring; around 60% believe humans are the main cause. Among Republicans, the figures are 60% and 35%. Among Democrats, 95% and 85%. The gap on whether to act and how aggressively is even larger.
• Race and racism. When asked whether systemic racism is a serious problem in America, around 90% of Black Americans agree; around 80% of Democrats; around 30% of Republicans. The descriptive gap is very large and stable.
• Gun policy beyond background checks. Bans on assault weapons, magazine limits, red-flag laws, concealed-carry expansion — partisan splits in the 30–60-point range.
• Immigration enforcement intensity. Mass deportation, Dreamers, refugee admissions, asylum policy — partisan splits typically 30–50 points.

Trust in government

This is the most stable long-term trend in American public-opinion data, and it is bleak.

The ANES "trust in government" question — "How much of the time do you think you can trust the government in Washington to do what is right?" — has been asked since 1958. The trend:

• 1958–1964: around 75% answered "just about always" or "most of the time."
• 1965–1980: decline to around 25–30%, driven by Vietnam, Watergate, stagflation.
• 1980s–early 1990s: modest recovery into the 30–40% range.
• Mid-1990s–2001: another decline and partial recovery.
• September 2001: spike to 60% post-9/11 (the only major positive shock in the 65-year series).
• 2003–2008: decline to around 25%.
• 2008–2024: sustained low, mostly in the 15–25% range, with brief partisan-driven fluctuations depending on which party holds the White House.
• 2024–2026: around 20%.

About one in five Americans trusts the federal government to do what is right most of the time. This is a historic low and a long-term outlier among advanced democracies (cross-national data appear in Case Study 1).

The decline is partisan-asymmetric in a particular way: trust-in-government numbers rise modestly for the in-party when their president holds office and fall for the out-party. Republicans trust government more under Republican presidents; Democrats trust government more under Democratic presidents. But the level among the in-party has also declined over time. Even when "your team" holds the White House, the trust numbers do not return to mid-century levels.

Trust in media

Trust in mass media has tracked a similar long-term decline. Gallup has asked since 1972: "How much trust and confidence do you have in the mass media — such as newspapers, TV and radio — when it comes to reporting the news fully, accurately, and fairly?"

• 1972: 68% had a great deal or fair amount of trust.
• 1976: 72% (post-Watergate, the press was popular).
• 1990s: decline into the 50s.
• 2010s–2024: decline into the 30s.
• 2024: around 32%, the lowest on record.

The partisan asymmetry in media trust is now extreme. Among Democrats, about 70% express trust. Among Republicans, about 12%. Among Independents, about 27%. The gap of nearly 60 points between the parties on whether mass media reports the news accurately is among the largest partisan gaps on any survey question. Both interpretations are partly correct: mainstream legacy media has, in measurable ways (newsroom demographics, story selection, editorial framing), tilted modestly leftward over the past 25 years, while conservative-aligned media (Fox News opinion programming, then Newsmax/OAN, then podcast/YouTube ecosystem) has produced parallel critique that shapes Republican perception.

Trust in courts

Trust in the Supreme Court has historically run higher than trust in Congress or the presidency, but it has declined and polarized since Dobbs (2022).

• 2000s–2015: around 60–65% expressed confidence in the Court (Gallup).
• 2016–2021: decline into the 50s during the Garland fight and the Kavanaugh and Barrett confirmations.
• 2022–2026: further decline. As of 2024, Gallup found Republicans expressing 60%+ confidence in the Court; Democrats expressing under 30%. The partisan gap, which barely existed in 2000, is now among the largest on any institutional-confidence question.

This is itself a case study in how partisan-aligned outcomes shape trust. When the Court delivered Bush v. Gore (2000), trust dropped among Democrats; it recovered. When the Court delivered Obergefell (2015), trust dropped among Republicans; it recovered. After Dobbs, trust dropped sharply among Democrats and has not recovered. The pattern: a Court widely seen as ideological loses trust from the side that loses on important cases.

Presidential approval

Presidential approval polling, conducted weekly or monthly by every major pollster, follows a predictable structure.

• A new president starts with around 50–55% approval (the "honeymoon"), with very high in-party approval and surprisingly high out-party approval that fades within months.
• Approval declines through the first term as the in-party fades and the out-party hardens.
• A "rally-around-the-flag" event (war, terror attack, crisis) can produce a temporary spike of 5–15 points.
• The modern (post-Reagan) approval floor is around 35–40%. Below that level, severe political dysfunction follows — primary challenges, mass resignations, congressional defections. The 25–30% range, common for unpopular presidents in earlier eras, is now rare because partisan loyalty floors approval.
• Approval is highly polarized: in-party approval typically runs 85–95%; out-party disapproval typically runs 85–95%; the meaningful variation is in the 10–25% of voters with weaker partisan attachment.

For political science, the most useful approval statistic is the net approval among Independents, because the overall number is mostly determined by partisan composition that does not change. Movement among Independents is movement that matters.

17.5 The structure of public opinion

We have surveyed what Americans believe on specific issues. Now we ask a deeper question: how is American public opinion organized? Does it have a coherent ideological structure, with consistent left-right positions? Or is it a mosaic of issue positions held without ideological glue?

Ideology, partisanship, and issue position

Three concepts to keep distinct:

• Ideology is a more-or-less coherent worldview about how government should work and what values should be prioritized. Liberalism, conservatism, libertarianism, progressivism, social democracy, traditionalism are ideologies in this sense.
• Partisanship is identification with a political party. Democrat, Republican, Independent, third-party. It is a social and political identity, not a worldview, though it correlates with worldviews.
• Issue position is what a person thinks about a specific policy. Whether to expand Medicare, whether to allow late-term abortion, whether to raise the minimum wage to $15.

These three correlate in modern America, but the correlation has been growing. In 1960, you could find substantial numbers of conservative Democrats (Southern Dixiecrats) and liberal Republicans (Rockefeller Republicans, Northeastern moderates). By 2024, the parties were ideologically much more sorted: the median Democrat and median Republican differ on ideology measures by about a full standard deviation more than they did in the 1970s (Pew, ANES).

The sorting is not the same as polarization. Sorting means people of similar ideology now share a party label. Polarization means the distance between the sides has grown. Both have happened, but sorting started earlier and has been more dramatic.

Converse and the limits of ideology

The most influential single paper in modern public-opinion research is Philip Converse's "The Nature of Belief Systems in Mass Publics" (1964). Converse looked at panel data — surveys of the same respondents over time — and found that on most issues, Americans did not have stable, internally consistent positions. They gave inconsistent answers across surveys. They held views that did not connect ideologically (favoring expanded social spending and lower taxes; favoring strong defense and isolationism). Most Americans, Converse argued, did not really have an ideology in the sense political scientists used the word. They had attitudes — clusters of feelings tied to groups, parties, and issues, but not organized into a coherent system.

Converse identified a small minority — perhaps 10–15% of the public — for whom ideology was a real organizing structure. For the rest, ideological labels meant something looser: "liberal" meant "for the kinds of things people I trust are for"; "conservative" meant the same on the other side.

Subsequent research has nuanced this. Some of Converse's apparent inconsistency was measurement error. Some was real but smaller than he estimated. As polarization has increased over the past 50 years, the ideological coherence of mass publics has also increased — partly because party labels now do more work as ideological markers, and partly because the parties have sorted ideologically. By 2024, more Americans hold internally consistent ideological positions than in 1964. But the share of the public for whom ideology is a sophisticated, abstract organizing principle is still well under half.

The implication: when political commentary describes "the moderate voter" or "the liberal voter" or "the conservative voter" as if those are stable, articulated identities, it is overstating ideological coherence. Most voters' political beliefs are looser, more contradictory, and more anchored in identity than the language of ideology implies.

Moral foundations: Haidt's framework

Where Converse looked at issue positions, Jonathan Haidt and his collaborators (Graham, Haidt, Nosek, and others) have looked at the underlying moral intuitions that shape political views. Their Moral Foundations Theory identifies several distinct moral concerns that humans bring to political questions:

• Care / harm — concern for the suffering and well-being of others.
• Fairness / cheating — concern for proportional outcomes and reciprocity.
• Liberty / oppression — concern for resisting domination.
• Loyalty / betrayal — concern for in-group cohesion and patriotism.
• Authority / subversion — concern for hierarchy, tradition, legitimate institutions.
• Sanctity / degradation — concern for purity, sacredness, what should not be defiled.

The empirical finding from Haidt's research is that liberals and conservatives differ in how heavily they weight these foundations. Liberals tend to weight care and fairness highly and to weight loyalty, authority, and sanctity less. Conservatives tend to weight all six roughly comparably. Libertarians weight liberty most heavily.

Haidt's important normative move is that he treats all of these foundations as legitimate moral concerns, not as one being "real morality" and the others being primitive intuitions to be overcome. A conservative concern about loyalty (to country, to community, to family) is a moral concern in exactly the way a liberal concern about care (for the marginalized, the suffering, the disadvantaged) is. Both are universal human moral concerns. People weight them differently.

This is steel-manning at the moral-psychology level. In a polarized environment, partisans on each side often perceive the other side as morally deficient — as not caring about real moral concerns. Haidt's framework reframes the disagreement: the other side does have moral concerns, but is weighting different concerns more heavily. That does not make disagreements disappear. It does change what the disagreement is about.

The empirical evidence for moral-foundations differences is robust across many studies and many countries. The framework is contested in detail (some psychologists argue the foundations should be reorganized; others question whether liberty is a separate foundation), but the core finding — that left and right differ systematically in moral weighting — replicates.

Personality and politics

A smaller but real correlation: the Big Five personality traits (openness, conscientiousness, extraversion, agreeableness, neuroticism) correlate weakly with political ideology. Liberals score modestly higher on openness; conservatives score modestly higher on conscientiousness. The effects are small (correlations typically 0.1 to 0.2) but appear consistently across studies and across countries. This is not a master key to politics — most variation in political views is not explained by personality — but it suggests that political differences have a partial basis in dispositions rather than purely in reasoning.

17.6 How public opinion is formed

If most Americans do not arrive at their political views through systematic ideological reasoning, then how do they arrive at them? Political socialization research has been answering this question for seven decades.

Family

The single strongest predictor of an American adult's party identification is what their parents' party identification was. Children raised in Democratic households are about 70% likely to identify as Democrats; children raised in Republican households about 70% likely to identify as Republicans. The transmission rate has weakened modestly over time but remains very high.

Family socialization is partly active (political conversation around the dinner table) and partly passive (modeling, the social network of the family, media consumption habits). It happens early. The political identifications of children as young as 10 predict their adult identifications with surprising accuracy.

The transmission is stronger for party identification than for issue positions. Children may grow up to differ from their parents on policy specifics, but the team identification — Democrat or Republican — is sticky.

Education

Higher education correlates with more liberal positions on social issues (race, gender, sexuality, immigration) and with more support for democratic norms (civil liberties, minority rights, support for institutions). It correlates more weakly with positions on economic issues — college-educated voters are not systematically more economically liberal than non-college-educated voters; in some surveys they are more economically conservative.

The causal claim — that education makes people more liberal — is contested. Some of the apparent effect is selection (people who choose college differ from those who do not) and some is exposure (college students encounter more diverse peers, more international students, more critical perspectives on traditional authority). Studies that try to isolate the causal effect — comparing twins, comparing communities before and after college expansion — typically find a real but smaller education effect than the raw correlation suggests.

The most consequential recent shift has been the education realignment: the historical Democratic advantage among non-college voters and Republican advantage among college voters has flipped. By 2024, college-educated white voters lean Democratic; non-college white voters lean Republican strongly. Among Hispanic and Asian voters, the partisan map is in flux but trending toward similar education-based polarization.

This is not just a class story (college and non-college voters do not differ that much in income; the difference between them is more cultural than economic) and not just a race story (the realignment runs across racial lines, with non-college voters of all races moving Republican). It is a structural feature of contemporary American politics that cuts across older categories.

Geography

Where Americans live shapes their politics. Three dimensions:

• Urban-rural. Urban voters are more Democratic; rural voters more Republican; suburban voters split. The gap is large and growing. By 2024, urban-rural is a more powerful predictor of vote choice than income or education.
• State-of-residence. A lifelong Texan and a lifelong Massachusan are politically different even controlling for individual demographics, partly because of state political culture, partly because of the people they grew up around, partly because state government delivers different experiences of government.
• Neighborhood effects. Within metro areas, exposure to neighbors of different views can liberalize or conservatize, depending on which views one is exposed to. Bill Bishop's The Big Sort (2008) documented an increasing geographical clustering of like-minded voters; subsequent research has found the trend continues.

The geography effect is partly self-selection (people choose neighborhoods aligned with their values) and partly socialization (living among certain people shapes values). Both operate.

Race, ethnicity, religion, gender, age, class

Each of these demographic categories predicts political views. Together they predict more.

• Race. Black Americans vote Democratic at very high rates (around 85% in recent presidential elections). Hispanic Americans vote Democratic but with significant variation by national origin, generation, and region; the Republican share has grown to around 40% in some recent elections, especially in Florida, Texas, and parts of the Southwest. Asian Americans vote Democratic, with substantial variation by national origin (Vietnamese-Americans more Republican; Indian-Americans more Democratic). White Americans split 55–60% Republican in recent presidentials, with large internal variation by region, education, and religion.
• Religion. White evangelical Protestants vote Republican at very high rates (around 80%). Non-Christians (Jewish Americans, Muslim Americans, Hindus, secular voters) lean Democratic. Catholics split. Mainline Protestants lean modestly Republican but less so than evangelicals. The frequency-of-attendance variable matters: regular church-attenders of any Christian denomination lean more Republican than infrequent attenders.
• Gender. Since the early 1980s, women have voted slightly more Democratic and men slightly more Republican; the "gender gap" is typically 5–10 points. The gap is larger among younger voters and college-educated voters than among older voters.
• Age. Younger voters are more Democratic; older voters more Republican. But generational cohorts retain their political socialization: today's older voters were socialized in different eras than today's younger voters and the generational effect interacts with the life-cycle effect.
• Class. Income predicts political views less than education or race. Among comparable demographic groups, higher-income voters lean modestly more Republican on economic issues but not on social issues. Class-as-class is a weaker organizing variable in American politics than in most European democracies.

Generation

Each generation's political socialization is shaped by the political moment in which they came of age.

• The Silent Generation (born roughly 1928–1945) came of age in the post-WWII era of high trust in institutions and bipartisan consensus on the Cold War. They retain unusually high trust in institutions and unusually high voter turnout.
• Boomers (1946–1964) came of age during the upheavals of the 1960s and 1970s. They split politically: the early Boomers more shaped by civil rights and Vietnam, the later Boomers more shaped by Reagan and the conservative resurgence.
• Generation X (1965–1980) came of age during the Reagan era and the end of the Cold War. They have been the most politically heterogeneous and least politically engaged of the modern generations.
• Millennials (1981–1996) came of age during 9/11, the Iraq War, the financial crisis, and Obama. They are unusually Democratic-leaning compared to earlier generations at comparable ages.
• Generation Z (1997–2012) came of age during Trump, COVID, and the social-media transformation of public life. The politics of Gen Z is still being established as the cohort enters adulthood, but early data show a continuation of Millennial Democratic lean among college-educated Gen Z and a Republican shift among non-college Gen Z young men, especially.

Generational effects are real but partial. Each generation contains internal variation that exceeds the variation between generations. Saying "Millennials believe X" is always loose talk: about a large majority of Millennials, while substantial minorities believe the opposite.

Media exposure

The final socialization channel — and a topic so important that Chapter 18 is devoted to it — is media. Americans are exposed to political information through whatever channels they choose, and increasingly those channels are partisan-sorted. Conservative voters consume Fox News, conservative talk radio, Newsmax, OAN, and a growing podcast/YouTube ecosystem. Liberal voters consume MSNBC, NPR, the New York Times, the Washington Post, and a parallel set of newer outlets. Independents consume more local news and broadcast network news, the categories whose audiences are shrinking fastest.

The polarized media environment does several things to public opinion. It selects: voters who already lean one direction self-select into outlets confirming that direction. It frames: even on the same factual story, different outlets emphasize different aspects. It mobilizes: emotionally charged coverage drives engagement, which drives turnout. It misinforms: low-quality outlets in both partisan ecosystems publish material that does not survive fact-checking, and consumers often do not check.

We will treat all of this in Chapter 18. For Chapter 17 purposes, the takeaway is that media exposure is a significant socialization channel, especially for the most politically engaged voters who consume the most political content. For the modal American who pays modest attention, family, peers, religion, and neighborhood typically have larger effects than media.

17.7 Heuristics and rational ignorance

Now the harder question: do voters actually reason about politics?

What voters know

Surveys of basic political knowledge produce humbling results. From the Annenberg Public Policy Center's recurring surveys:

• About 70–75% of Americans can name all three branches of government.
• About 50% can name their U.S. House representative.
• About 30% can name both of their U.S. senators.
• About 20% can identify a recent Supreme Court decision.
• About 15% can name the Speaker of the House.
• About 10% can name the Chief Justice.

These numbers vary year to year and by exact question wording, but the pattern is consistent: most Americans do not know the basic structural facts of their political system in the depth that civics-textbook ideals would suggest.

This is not stupidity. It is what economists call rational ignorance. The expected effect of any individual citizen's knowledge on political outcomes is essentially zero. The cost of acquiring political knowledge — in time, attention, and unrewarded study — is non-zero. For most people, the rational allocation of their limited time is to know more about their family, work, and immediate community, and to substitute heuristics for systematic political reasoning.

Heuristics

A heuristic is a cognitive shortcut: a rule that allows you to make a reasonable decision without doing the full analysis. Voters use several:

• Party label. "What does my party say about this candidate or issue?" If you know nothing else about a candidate, knowing their party affiliation gives you most of what you would learn by deeper investigation. The party label aggregates a great deal of information (the candidate's coalition, voting record, ideological commitments) into a single signal.
• Endorsements. "Who is supporting this candidate, and do I trust them?" Endorsements from groups whose values you share — a labor union, a chamber of commerce, an evangelical association, the Sierra Club — communicate alignment.
• Demographic cues. "Is this candidate someone like me?" Voters often vote for candidates from their own demographic group, controlling for ideology. The effect is real but contested in size.
• Performance evaluation. "Is the country / state / district doing well? If yes, reward the incumbent; if no, punish them." This is the basis for retrospective voting, which we will discuss next.

These heuristics are not failures of citizenship. They are reasonable cognitive economies. The question is whether they produce decisions that approximate the decisions a fully informed voter would make. The answer is: sometimes yes, sometimes no.

Bartels and Achen: Democracy for Realists

The most influential recent skeptical account of voter reasoning is Christopher Achen and Larry Bartels's Democracy for Realists (2016). Their argument has several pieces.

First, they show that many voters' issue positions appear to follow their party identification, not the other way around. When parties shift positions on an issue (the Republican Party's shift on free trade, the Democratic Party's shift on immigration enforcement), partisan voters often shift with them — suggesting that the party label is doing the cognitive work, not independent issue reasoning.

Second, they show that voters reward and punish incumbents for events the incumbents did not control: drought, floods, shark attacks, the performance of the local college football team. (The football-team finding is contested but provocative.) If voters were carefully evaluating policy, these events should not affect electoral outcomes. They do.

Third, they argue that the "folk theory" of democracy — that voters carefully consider issues, choose candidates whose positions match their own, and thereby control government — is empirically false. Voters mostly identify with social groups (parties, religious communities, demographic categories) and vote based on group loyalty, not on policy reasoning.

Their conclusion is provocative: democracy, as actually practiced, is mostly a system of group-based identity politics, not policy preference aggregation. They are not anti-democratic; they argue democracy is still valuable for other reasons (allowing peaceful transitions of power, providing some accountability, distributing political voice). But the textbook story of informed-citizen-deliberation does not describe how American democracy works.

The pushback: Healy, Lenz, and others

Other researchers have pushed back on the Bartels-Achen pessimism. Andrew Healy and Gabriel Lenz have shown that voters do respond to economic conditions in ways consistent with retrospective performance evaluation — even if not perfectly, and even if they sometimes weight the most recent year disproportionately. Voters punished Republican incumbents during the Great Recession and Democratic incumbents during the inflation shock of 2022. The signal is noisy but real.

Lenz's work also shows that voters frequently update their issue positions when they learn that their preferred candidate holds an unexpected position. This looks like party-loyalty-driven reasoning rather than independent policy preferences. But in some cases, voters change candidates rather than positions: when a candidate's positions diverge too far from the voter's actual preferences, the voter eventually defects.

The state of the academic debate is roughly: voters are less policy-reasoning than the folk theory suggests and more policy-reasoning than the strongest version of group-identity theory suggests. They use heuristics, including party identification, but the heuristics are not perfectly aligned with party loyalty in all cases. Performance matters. Conditions matter. Identity matters. The correct mental model is not "voters are stupid" or "voters are wise" — it is "voters are economical reasoners doing the best they can with limited time and the tools available."

For an instructor: this is one of the most fertile spots for class discussion. The Bartels-Achen view and the Healy-Lenz view both have strong empirical support; they emphasize different parts of the same complicated reality.

17.8 Cognitive biases in political judgment

A separate strand of research focuses on the systematic ways human cognition goes wrong on political questions. Some highlights.

Confirmation bias and motivated reasoning

The classic finding from cognitive psychology — that people seek out information confirming their existing beliefs and discount disconfirming information — applies to politics with particular force. Partisan voters are more likely to read articles consistent with their views, more likely to remember evidence consistent with their views, and more likely to find arguments consistent with their views persuasive.

Motivated reasoning is a related concept: people reason toward conclusions that align with their identities and emotional commitments. The classic experimental demonstration is the "death-penalty study" (Lord, Ross, and Lepper 1979), which presented partisans on both sides of capital-punishment debates with the same mixed evidence and found that both sides came away more confident in their initial position. Each side accepted the supportive studies as well-conducted and discounted the unsupportive ones.

Cultural cognition: Kahan's research

Dan Kahan's "cultural cognition" research extends this further. Across many studies, Kahan and collaborators have found that people's interpretations of factual claims (about climate change, gun policy, vaccine safety, economic policy) are powerfully shaped by their cultural and political identities. The same facts produce different conclusions in different identity groups.

Kahan's most disturbing finding: higher numeracy and education sometimes makes the bias worse, not better. On politically charged questions, more sophisticated reasoners are better at finding ways to reach the conclusion their identity prefers. They are more capable of motivated reasoning, not less. On non-political questions, more education improves accuracy. On politically polarized questions, it can deepen rationalization.

This is one of the most important findings for anyone teaching American politics. The naïve hope — that if we just teach people to think more rigorously, they will reach better conclusions on politics — does not survive contact with Kahan's data. Education does not by itself overcome motivated reasoning. Something else has to.

What that something else is, the research is less clear about. Exposure to people of different political views, in low-stakes settings where listening is rewarded, seems to help modestly. Forcing people to articulate the strongest version of the other side (the steel-manning exercise) seems to help. Information consumed in mixed-source environments seems to help. None of these are reliably scalable solutions.

The perception gap

A specific application of motivated reasoning: Americans of each party caricature the other party's views. They believe Democrats and Republicans are more extreme and more uniform than the actual median of each party.

The most thorough documentation of this is the "Hidden Tribes" research and subsequent reports by the More in Common organization (2018, 2022, 2024). They have shown:

• Republicans estimated that around 32% of Democrats believe the police should be defunded; the actual figure was around 8%.
• Democrats estimated that around 50% of Republicans believe immigration should be effectively zero; the actual figure was around 18%.
• Both sides systematically overestimate how much the other side dislikes them.
• Both sides systematically overestimate how extreme the other side is on most policy questions.

The perception gap is largest among the most politically engaged voters — those who consume the most political content. The most politically engaged Republicans have the most caricatured view of Democrats, and vice versa. People who consume less political content have more accurate views of the other side.

The mechanisms include partisan-sorted media (which amplifies the most extreme voices on the other side as evidence of their craziness), social-media virality (which selects for outrage), social sorting (Americans increasingly do not have close friends of the other party), and the simple human tendency to weight the loudest voices most heavily.

The good news in the perception-gap research: Americans on both sides are less polarized on policy than they believe they are. The actual median Democrat and the actual median Republican differ less than either thinks. The bad news: the perception gap is itself politically real. Americans treat the other party as if it held more extreme positions, even when it does not, and this affects political behavior.

Case Study 2 develops the perception-gap research and its implications more fully.

17.9 Issue polling vs. issue importance

A puzzle: many policy positions polled at large-majority support do not become law. Universal background checks for gun purchases (above 80% support, no federal law). A path to citizenship for Dreamers (above 70% support, no comprehensive law). Lower prescription drug prices (above 80% support, only partial action through Medicare negotiation in 2022). What gives?

The answer involves the gap between polling support and political force. Three pieces:

Salience asymmetry

A voter who supports background checks but ranks the issue twentieth on their list of priorities is unlikely to change their vote based on it. A voter who opposes additional gun regulation and ranks the issue first or second on their priority list will change their vote on it. The 80% supportive coalition is broad but shallow; the 20% opposing coalition is narrow but deep.

For legislators, the calculus that matters is not "what does the median voter on this issue support?" It is "what will move enough voters to affect my election?" On many issues, the intensely-motivated minority outweighs the broadly-supportive majority.

Organizational mobilization

Issues are not just polled; they are organized. The pro-gun-rights coalition has the National Rifle Association, gun-owners associations, single-issue PACs, and a network of state-level groups with millions of members and substantial campaign-finance resources. The pro-gun-control coalition has organizations too — Everytown, Brady, Giffords — but they are smaller, newer, less embedded in dense local networks, and less well funded historically (though the gap has narrowed since around 2018).

Organizations turn polling support into political pressure: emails to members, voter contact, direct lobbying, primary-election threats, contributions, and the implicit threat of all of these. A policy with broad polling support but weak organizational backing routinely loses to a policy with narrow polling support and strong organizational backing.

The legislative process

Congress is not a direct-democracy aggregator. The filibuster (60 votes for most legislation in the Senate), committee gatekeeping, the conference process, and the executive's veto each provide veto points where minority opposition can block majority preference. On many policies — including ones with strong polling support — the supportive coalition cannot assemble the supermajorities needed to overcome these veto points.

This is not a bug specific to gun policy or immigration. It is a feature of the American constitutional system, which was designed to slow majoritarian action and preserve veto points for various minorities. We will discuss this in detail when we cover Congress (Chapter 8) and the policy process (Chapter 33). For Chapter 17 purposes, the lesson is: polling majority support is a necessary but not sufficient condition for policy change. Many other things have to be true for the legislative machinery to act.

17.10 Public opinion and policy: the responsiveness debate

The largest unresolved question in this field is how responsive policy is to public opinion. The answer matters enormously: if policy responds well, then democracy is doing its central job, even if imperfectly. If policy does not respond — or responds only to certain people's preferences — then "democracy" is doing something other than what we usually claim.

Page and Shapiro: public opinion shapes policy

Benjamin Page and Robert Shapiro's The Rational Public (1992) presented evidence that public opinion does, in aggregate, shape American policy with measurable lags. When the public moves on an issue — toward more support for environmental regulation, or away from support for the Vietnam War — policy moves with it, on a timescale of years. The public is not always right; the public is not always coherent; but the public's direction of motion is generally followed by the political system, eventually.

This is a relatively optimistic finding: democracy works, slowly, on average.

Gilens and Page: whose opinions shape policy?

Martin Gilens and Benjamin Page's later study (2014, building on Gilens's Affluence and Influence 2012) asked a sharper question. Across a large dataset of policy decisions, they compared the policy outcome to the polling preferences of three different groups: median-income voters, top-10%-income voters, and organized interest groups.

Their finding: when median-income voters and top-10% voters disagreed, the policy outcome correlated much more strongly with the top-10% preference than the median preference, controlling for partisan composition of government. Median-income voters had statistically near-zero independent influence on policy outcomes; the wealthy had substantial influence; organized interest groups had substantial influence.

This is a much darker finding. It says American democracy, in practice, is more responsive to the affluent and the organized than to the broad middle of the public. The framing they use — "oligarchic" tendencies in American policymaking — is provocative; the data are real.

Subsequent debate

Their finding has been replicated, qualified, and contested. Several lines of debate:

• What counts as "responsiveness"? Some critics (Bashir 2015, Branham et al. 2017) argue that median-voter and high-income preferences are correlated more strongly than Gilens-Page allowed; once you control for the agreement, the high-income-only effect is smaller.
• The role of political parties. When parties take aligned positions, the analysis collapses; the data only show responsiveness to the affluent on the issues where the parties were divided in particular ways.
• What about descriptive representation? Even if median-income voters' preferences don't predict policy, the parties' coalitions reflect demographic majorities, so the policy may be representing voters by another route.
• The role of organized labor. Historical analyses suggest the wealthy-policy correlation has grown as union density has fallen; when labor unions were stronger, the correlation between mass and elite preferences was stronger.

The state of the field is: yes, the wealthy and organized have outsized influence; the size of that influence and the right way to measure it are contested; and the implications for "American democracy is broken" are contested.

For Chapter 17 purposes, the relevant claim is the empirical one: the median voter is not, in fact, the median of the political system in the way the simplest civics-textbook story implies. Both the affluent (through donations, organization, and personal networks) and the highly motivated minority (through mobilization and primary-election pressure) have leverage that the median does not. A serious account of American politics has to integrate this.

17.11 A note on cross-pressured voters

Before turning to data sources, one more empirical pattern matters: the existence of voters whose demographic and identity profile predicts one party but whose specific issue positions push the other way.

A college-educated white evangelical Hispanic woman in suburban Phoenix may be cross-pressured: education predicts Democratic, white evangelical predicts Republican, Hispanic identity predicts Democratic with significant variation, suburban predicts swingy, woman predicts Democratic by a modest margin. The combination of these signals does not point uniformly. She makes a decision on something like residual issue weighting, candidate appeal, and current conditions.

Cross-pressured voters are politically important because they are the swing electorate. Voters with reinforcing pressures (a non-college rural white evangelical man, a college-educated urban secular Black woman) are highly predictable in their voting and are courted mostly through turnout efforts rather than persuasion. Cross-pressured voters are courted through persuasion and are the small population on which most close election outcomes turn.

The literature on cross-pressured voters goes back to Lazarsfeld, Berelson, and Gaudet's The People's Choice (1944) — the foundational study of how voters decide. They found that cross-pressured voters disproportionately delay deciding, disproportionately decline to vote, and when they do vote, are more responsive to short-term campaign factors than reinforcing-pressure voters. Subsequent research has confirmed the basic structure across many cycles.

In a polarized environment with high partisan sorting, the proportion of cross-pressured voters is smaller than it once was. But the small population that remains is decisive in close races. Political scientists who want to understand how elections turn out spend a disproportionate share of their attention on this population.

17.12 Where to find public-opinion data

If you want to engage with public opinion as a citizen rather than just consume the media's filtered version, here are the primary sources.

• Pew Research Center (pewresearch.org) — extensive issue and demographic polling, generally well-presented.
• Gallup News (news.gallup.com) — historical time series, especially for trust in institutions and presidential approval.
• American National Election Studies (ANES) (electionstudies.org) — academic gold standard, with downloadable data files for serious analysis.
• General Social Survey (GSS) (gss.norc.org) — long-running social-attitudes data.
• Cooperative Election Study (CES) (cces.gov.harvard.edu) — large-N academic survey.
• Roper iPoll Archive — comprehensive archive of polling data going back to the 1930s; subscription, but available through most university libraries.
• More in Common (moreincommon.com) — perception-gap research and "Hidden Tribes" reports.
• AP-NORC Center for Public Affairs Research (apnorc.org) — frequent issue polling.
• Public Religion Research Institute (PRRI) (prri.org) — religion and politics; deep coverage of the religion-politics intersection.

If you are doing the Democracy Audit project, your district-level analysis will rely heavily on Pew and ANES sub-group data combined with district demographics from the Census. The specifics are in exercises.md.

17.13 Looking ahead

We have now mapped what Americans believe and how we know. Several themes recur in the next nine chapters of Part III:

• Chapter 18 (The Media) explores how the media environment shapes public opinion and how media polarization has changed since 2000.
• Chapter 19 (Political Parties) examines how parties aggregate and channel public opinion into electoral coalitions.
• Chapter 20 (Elections and Campaigns) translates opinion into vote choice through the campaign process.
• Chapter 22 (Voting Behavior) drills into the determinants of individual voting decisions.
• Chapter 25 (Polarization) returns to the perception-gap research and asks what is driving the divergence.
• Chapter 26 (Social Movements) examines how organized groups translate public sentiment into political pressure.

A reader who finishes Part III should have a clear picture of how the American political system, at its base layer, processes the views of 260 million people into the small set of decisions that government makes. Public opinion is the raw input. Everything else is the machinery.

A final reminder: the empirical landscape this chapter has described is contested in detail and stable in outline. New polling data will refine the numbers. Major events will shift the trends. But the structural features — the role-of-government split, the institutional-trust decline, the perception gap, the responsiveness puzzle — have been remarkably stable across the last decade and are likely to remain so. They are the terrain on which all American political action takes place. Knowing the terrain is the precondition for understanding the action.

What to do next

• Skim appendix-f-how-to-read-a-poll.md for the methodological depth this chapter touched only lightly.
• Work the exercises in exercises.md — at minimum, the Pew study analysis and the perception-gap measurement.
• Read both case studies — the trust-in-institutions decline and the perception gap. They develop the data this chapter sketched.
• For your Democracy Audit project, this chapter's exercises produce a concrete deliverable: a public-opinion profile of your district, drawn from Pew and ANES sub-group data.

Public opinion is hard. It is messy, contested, partial, and slow to change. So is American democracy. They are connected.