Chapter 22: Voting Behavior — Who Votes, Who Doesn't, and Why It Matters

"The franchise is the central instrument of democracy. The franchise practiced is the central reality of it."

Learning objectives

By the end of this chapter you will be able to:

1. Distinguish among the three denominators used to calculate voter turnout (VAP, VEP, registered voters) and explain why VEP is the cleanest.
2. Describe long-term trends in U.S. presidential and midterm turnout and place American turnout in international context.
3. Identify the major demographic predictors of turnout — age, education, income, race/ethnicity, gender, marital status, religious attendance — and rank them by predictive strength.
4. Describe the major partisan demographic correlates as of 2024, including the post-2016 education realignment and the 2024 shifts among Hispanic and working-class voters.
5. Explain four theories of why people vote despite the rational-choice paradox: civic-duty, expressive, resource, and mobilization.
6. Describe four theories of how people choose: party identification, issue voting, candidate evaluation, and retrospective/economic voting.
7. Map the legal architecture of voter registration in the United States: NVRA, HAVA, online registration, same-day registration, automatic voter registration, voter ID.
8. Steel-man both the progressive critique of restrictive voting laws and the conservative defense of election-integrity measures, and describe what the empirical literature does and does not establish.
9. Apply the chapter's frameworks to your own congressional district.

1. Why this chapter exists

Two facts. American politics is decided by voters. American voters are not a representative cross-section of Americans.

Both facts are obvious. Together, they shape almost everything in this book — who wins elections, which policies get enacted, which constituencies politicians court and which they ignore. Chapter 17 told you what Americans think. Chapter 19 told you about the parties they belong to. Chapter 20 told you about the elections that translate opinion into office. Chapter 21 told you how campaigns operate. This chapter tells you about the people who actually show up.

It is a behavioral chapter. Empirical, data-heavy, comparative. We will look at what predicts whether a person casts a ballot, what predicts how they cast it once they do, and how the legal architecture of voting — registration, voter ID, early voting, mail voting — shapes the electorate that exists.

We will also handle a contested question carefully. The legal architecture of voting is the subject of intense partisan disagreement. Progressives describe many recent state-level measures as "voter suppression" — laws that disproportionately reduce turnout among Democratic-leaning groups. Conservatives describe the same measures as "election integrity" — reasonable safeguards against fraud that apply equally to all voters. Both sides have empirical claims. Both sides have legitimate concerns. The honest answer about most of these laws is that the effects are smaller than activists on either side claim. We will work through the evidence carefully.

What this chapter is not: a moral verdict on who votes or who doesn't. The patterns are what they are. We will describe them.

Why behavioral data matters more than ideology

There is a temptation in American political analysis to substitute strong moral claims for empirical observation. The temptation is bipartisan. The progressive version says: turnout is low because elites have rigged the system to keep working people out. The conservative version says: turnout is low because half the country is too disengaged or too busy free-riding to participate in self-government. Neither claim, on its own, holds up against the data we will look at in this chapter.

What does hold up: a structural account in which different demographic groups have different costs of voting, different psychological and social rewards from voting, and different mobilization environments around them. The differences are not random and they are not destiny. They are patterns — measurable, comparable across time and place, and sometimes responsive to policy change.

A political scientist looks at the patterns and resists the temptation to turn them into a verdict. A citizen who wants to act in the political world has to do something more than that — they have to choose what to do. But they choose better when they start from the data than when they start from the verdict.

The structure of this chapter

We move from descriptive to explanatory to legal-institutional, and end with the policy and reform debate.

• Sections 2-3 describe who turns out and how this varies across demographic groups.
• Section 4 describes the partisan demographic correlates in 2024 — the ways the existing electorate sorts into the two parties.
• Sections 5-6 ask why people vote and how they choose, drawing on the political-behavior research literature.
• Sections 7-8 describe the legal architecture of voter registration, voter ID, mail voting, and early voting.
• Section 9 addresses the contested suppression debate and the empirical evidence on the effects of restrictive laws.
• Section 10 closes with the registration cliff — the 25% of eligible Americans who are not on any voter roll.

By the end, you should be able to read post-election commentary critically. You should know which numbers are reliable and which are spin. You should know how to make the case from each side without misrepresenting the other. And you should know what your own political community looks like when measured against the national patterns.

2. Turnout: who votes

The denominator problem

Before we can talk about how many Americans vote, we have to decide what number we are dividing by. (Appendix G covers this in more detail; we summarize here.)

There are three candidate denominators:

• Voting-Age Population (VAP). All persons over 18 living in the United States. Includes about 22 million non-citizens. Includes (in some states) people with felony convictions who cannot vote. The largest, dirtiest denominator. Useful for demographic description, not for measuring civic participation.
• Voting-Eligible Population (VEP). VAP minus people legally ineligible to vote — non-citizens, and (in states that disenfranchise them) people with felony convictions. This is the cleanest measure of "Americans who could legally vote." Maintained authoritatively by Michael McDonald at the University of Florida Election Lab (electlab.org).
• Registered voters. People on the voter rolls. About 70–75% of VEP is registered. Turnout as a percentage of registered voters is always higher than as a percentage of VEP — but it ignores the unregistered, who are the very people most political reform efforts try to reach.

In this chapter, every "turnout" number is a percentage of VEP unless otherwise noted.

Long-term presidential turnout

Modern American presidential turnout has hovered between 55% and 65% of VEP since the early 1970s, with two recent surges:

The 2020 surge had multiple drivers — pandemic-driven expansion of mail voting, intense partisan engagement, both campaigns mobilizing aggressively, and a sense among voters of both parties that the election had unusually high stakes. What makes the 2020 number historically remarkable is not just its size but its bipartisan character: both Democratic and Republican turnout were at modern record levels. Neither side dragged the other up by alone. Both showed up.

2024 stayed elevated. Whether the 2020-2024 plateau is a durable engagement shift or a function of the Trump-era political environment is a question political scientists are still working on.

Midterm turnout

Midterm turnout is structurally lower than presidential turnout because the presidency itself is the main mobilizing force for low-engagement voters. Even so, midterm turnout has been climbing.

The 2018 and 2022 cycles broke a long pattern of declining midterm engagement and reset the baseline upward. As with 2020, whether the new baseline is durable is unsettled.

Off-year, special, and primary turnout

Below the midterm, turnout falls precipitously.

• Off-year state and local elections. In states that hold gubernatorial elections in odd years (Virginia, New Jersey, Kentucky, Mississippi, Louisiana), turnout is typically 40–50%. Mayoral and city-council elections frequently see 15–25%, sometimes lower.
• Special elections. Highly variable. A high-profile special election with national attention (Pennsylvania-12 in 2018, Wisconsin-3 in 2023) may approach 30–40%; a low-profile one may hit 5–10%.
• Primaries. Primary turnout in presidential cycles is roughly 25–35% of registered voters. Down-ballot primary turnout is often under 15%.

The point: when you read about a "voter mandate" from any election, the first question is what fraction of the eligible electorate actually weighed in. A representative who wins a low-turnout primary may be the choice of 5% of the voters they will go on to represent.

International comparison

American turnout, in presidential years, runs near the bottom of advanced democracies — but the comparison is complicated.

Two observations. First, when Americans turn out at 2020 levels, U.S. participation is competitive with Canada, the U.K., and Germany — voluntary-voting peer countries. The United States is not uniquely apathetic; it just has a registration-burden architecture that depresses turnout relative to peer countries with automatic registration. Second, the truly high-turnout countries (Australia, Belgium) achieve their numbers through compulsory voting laws — fines for non-voters. Americans considering whether the United States should adopt compulsory voting confront a genuine trade-off between participation and the negative liberty to abstain.

A third observation worth registering: the international comparison gets noisier when it is broken down by election type. American turnout in midterm elections (40-50%) compares unfavorably with peer countries' comparable cycles (which are usually general elections, treated as equivalent to U.S. presidential cycles). And American turnout in primaries — where many congressional races are effectively decided in safe districts — falls below 30% in most cases, with no real peer-country equivalent because most other democracies use party-internal candidate-selection rather than open primaries. The comparison most damaging to U.S. turnout is not at the presidential level but at the level of primaries and downballot races, where the American system asks voters to participate in many more elections than the typical peer democracy does, and to do so under a registration architecture that imposes a meaningful per-cycle cost.

Cross-national comparison also surfaces structural reasons U.S. turnout is lower beyond compulsory voting. Election day is a workday in the United States. Most peer democracies hold elections on weekends or treat election day as a national holiday. Voter registration is the citizen's responsibility in the United States. In most peer democracies, governments register eligible voters automatically using existing administrative records (tax rolls, civil registries). Polling places vary in accessibility. In peer democracies with smaller populations and more centralized administration, polling-place lines are rare; in the United States, lines of multiple hours have been documented in particular states and particular precincts.

These structural differences mean that the question "why do Americans turn out at lower rates than (say) Germans?" has answers that go beyond civic culture. Some of the gap is administrative architecture; some is registration burden; some is workday-vs-weekend timing; some is the sheer number of elections Americans are asked to participate in across federal, state, and local levels. To shift U.S. turnout meaningfully would require addressing several of these structural factors simultaneously.

3. Demographics of turnout

Six demographic variables predict whether a given person votes, with overlapping but distinct effects.

Age

The single strongest pattern is that older Americans vote at much higher rates than younger Americans.

Two things drive the age gradient. First, life-stage effects. Older voters are more likely to own homes, pay property taxes, have school-age children or grandchildren, and have stable addresses — all of which connect them to local political institutions. Younger voters move more, change jobs more, and have less stake in local outcomes. Second, aging-into-voting. Voting is habit-forming (more on this below). A 65-year-old has had 47 chances to vote in presidential elections; a 22-year-old has had at most one. Repetition matters.

The 18–29 turnout number has climbed since 2016. Pre-2016, young-voter turnout was typically 40–45%. In 2018, 2020, and 2022, it touched 50%. The 2024 number was about 47%, lower than 2020 but well above pre-2016 norms.

Education

The strongest demographic predictor of turnout is education — even stronger than age, when controlled for other variables.

The 25-point gap between college and non-college turnout is the single most important demographic fact in American politics. It shapes who shows up and, increasingly, which party they vote for.

Why does education predict turnout so strongly? Several mechanisms. Education provides civic skills — reading bureaucratic forms, navigating registration requirements, finding polling places, parsing ballot initiatives. Education builds social networks that include other voters, which generates social pressure to participate. Education raises civic knowledge — what the offices on the ballot do, what is at stake in this election. And education correlates with other turnout-predicting variables (income, residential stability) without being reducible to them.

Income

Higher-income voters turn out at higher rates than lower-income voters.

A ~33-point gap between the top and bottom income quartiles. Some of this is mechanical — higher-income people are also better-educated, on average — but income has independent predictive power even after controlling for education.

Race and ethnicity

The white-Black turnout gap has narrowed substantially over the last 20 years, especially in presidential years and especially in states with strong Black mobilization infrastructure. In 2008 and 2012 — the Obama elections — Black turnout actually equaled or exceeded white turnout. It has slipped slightly since but remains close.

The white-Hispanic and white-AAPI gaps are larger and have been more stable. Several factors. Citizenship structure. Many Hispanic and Asian American adults are non-citizens; this lowers VAP-based turnout dramatically but only modestly affects VEP-based turnout. Age structure. Hispanic and Asian American populations are younger on average, and younger voters always vote less. Mobilization investment. Both parties invest substantially less in Hispanic and Asian American outreach than in Black or non-college white outreach. The gap is partly because the parties expect lower turnout — and partly self-fulfilling.

Gender

Since 1980, women have voted at slightly higher rates than men.

The gender turnout gap is small but real: women are now consistently 3–5 points more likely to vote than men. Combined with the gender gap in vote choice (described below), this means more votes are cast by women than by men in every recent election cycle.

Marital status, religious attendance, civic engagement

A handful of other variables predict turnout, mostly because they index integration into community life.

• Married > unmarried. Married voters turn out at ~10 points higher than never-married voters of similar age and education. Marriage correlates with home ownership, stability, and social networks of other married (voting) people.
• Frequent religious-service attenders > infrequent. Regular churchgoers (or synagogue/mosque attenders) turn out at ~8 points higher than non-attenders, even controlling for education and age. Religious institutions function as civic-skill incubators; they are also the largest voluntary mobilization networks in American life.
• Civic engagement (volunteering, organization membership) > none. Members of voluntary associations turn out at substantially higher rates.

These variables don't determine turnout. They index a cluster of habits — community embeddedness, social trust, regular participation in collective action — that all push toward voting.

Putting the demographic predictors together

A useful exercise: imagine two Americans who differ on all the variables we have just discussed.

Voter A is a 67-year-old white college graduate with a household income in the top quartile, who is married, attends religious services weekly, owns her home, and has lived in the same congressional district for 30 years. Predicted probability she votes in 2024: above 90%. The model says nearly every one of her demographic markers points the same way.

Voter B is a 22-year-old Hispanic high school graduate with a household income in the bottom quartile, who is unmarried, does not attend religious services, rents an apartment, and has moved twice in the past three years. Predicted probability he votes in 2024: below 35%. The model says nearly every one of his demographic markers points the same way.

The 55-percentage-point gap between Voter A and Voter B is not principally about ideology or even about partisan preference. It is about resources, habits, and integration into community life. The two voters experience the act of voting differently. For Voter A, voting is a low-cost, habitual, socially reinforced activity. For Voter B, voting may require checking whether his registration follows him to his new apartment, finding out where his polling place is, navigating new ID rules in his state, and either leaving work early or voting before his shift starts. The two voters are not deciding whether or not to participate against the same backdrop. They are facing different worlds.

Most political-science models of turnout, when they try to predict who votes, reach about 75-85% predictive accuracy using demographic and habit variables alone. The remaining 15-25% is the consequential terrain — the voters whose turnout decision is genuinely contingent on the specific election, the specific candidate, and the specific mobilization effort. Campaigns spend most of their resources on those marginal voters, because the rest are either certain to vote or certain not to.

4. Partisan demographic correlates (post-2016)

The previous section asked who votes. This one asks how they vote. The American electoral coalitions have been reshuffling for the past decade, and the post-2016 patterns differ meaningfully from the pre-2012 patterns.

Education realignment — the dominant fact

The single most important shift in American voting behavior since 2012 is the education realignment. College-educated voters, who used to lean Republican (Reagan and Bush coalitions), now lean Democratic. Non-college voters, who used to lean Democratic (the New Deal coalition), now lean Republican.

The realignment hit hardest among white voters, where the diploma divide is now ~25 points (white college voters lean Democratic; white non-college voters lean Republican by large margins). It is more muted, but visible, among non-white voters as well.

Why does education predict vote choice now when it didn't 20 years ago? Several theories.

• Cultural sorting. Issues that previously cross-cut both coalitions — immigration, race, religion, gender — have polarized along educational lines. College-educated voters have become more cosmopolitan; non-college voters more populist.
• Geographic sorting. Educated voters concentrate in metro areas with high density of other educated voters, which become culturally and politically distinct from less-dense areas.
• Information environments. Educated voters consume different media, including elite-prestige media that has trended Democratic.
• Class-based realignment in the Republican Party. Republican messaging has shifted toward economic populism on trade and immigration, attracting non-college voters who once voted Democratic on bread-and-butter economics.

What you make of these explanations is partly contested. The fact of the realignment is not.

Race and ethnicity

The race/ethnicity gap in vote choice remains the largest demographic divide in American politics, but it is narrowing in important ways.

Black voters remain the most reliably Democratic constituency, with 85–90% of the vote in every recent election. The 2024 number (~85%) was slightly below the 2012 peak (~93% for Obama) but remained the largest racial-ethnic Democratic bloc. Black men shifted modestly toward Republicans in 2024 (~25% Trump vote, up from ~12% in 2012), but Black women remain ~92% Democratic.

Hispanic voters show the largest movement. The 2024 election delivered ~45% of the Hispanic vote to Trump — up from ~28% in 2016 and ~32% in 2020. Hispanic men in 2024 split nearly evenly between the parties; Hispanic women still leaned Democratic but by single digits. Cuban-Americans have been Republican-leaning for decades; Mexican-Americans were the swing portion in 2024; Puerto Rican and Dominican voters mostly stayed Democratic. (Case Study 2 unpacks this shift in detail.)

Asian American voters also moved modestly toward Republicans, especially in 2024. Indian-American voters remain the most Democratic Asian American subgroup; Vietnamese-American voters the most Republican.

Religion

Religious affiliation correlates with vote choice in patterns that have been stable for a generation.

White evangelicals remain the most reliably Republican religious group, voting 78–82% Republican in every recent presidential election. Catholics are the most internally divided major group, with white Catholics leaning Republican and Hispanic Catholics leaning Democratic — meaning the "Catholic vote" splits roughly evenly. Jewish voters remain ~70% Democratic, though the margin has narrowed slightly. Muslim voters are demographically diverse — Indian-American Muslims, Arab-American Muslims, Black Muslims, and Southeast Asian Muslims have different voting patterns; the 2024 election saw notable Muslim defections from Democrats over Israel-Gaza policy in some swing states (Michigan in particular).

The "Nones" — Americans with no religious affiliation, now about 28% of the adult population — are the fastest-growing demographic and lean ~60-40 Democratic.

Geography

The rural-urban divide is now one of the strongest demographic predictors of vote choice. The relationship is roughly linear in population density.

The urban-rural divide cuts across race. Rural Black voters lean Democratic but less than urban Black voters; rural Hispanic voters lean Republican more than urban Hispanic voters; rural white voters lean Republican more than urban white voters. Population density has become a more powerful predictor of vote choice than income.

The contested terrain in modern elections is the suburb — and specifically the inner suburb, which has been moving Democratic, and the outer suburb / exurb, which has been moving Republican.

Class and income

The income-vote relationship is now complicated by the education realignment. The traditional pattern (higher income votes Republican) still holds among non-college voters but is reversed among college voters.

• High-income, college-educated voters now lean Democratic — a reversal from 1980-2000 norms.
• High-income, non-college voters lean strongly Republican.
• Low-income, college-educated voters lean strongly Democratic.
• Low-income, non-college whites lean strongly Republican.
• Low-income, non-college non-whites still lean Democratic, but with significant 2020-2024 movement.

The cleanest summary: education and race-ethnicity now explain more of vote variance than income alone. Income still matters, but it interacts with the other variables in complex ways.

Gender

Women have voted more Democratic than men in every presidential election since 1980. The "gender gap" — the difference between male and female Democratic vote share — has been remarkably stable at 7–12 points.

The 2024 gap was modestly above the long-run average. The gender gap interacts with marital status — single women lean substantially more Democratic than married women, and single men have been moving toward Republicans faster than married men. Some analysts argue the modern primary axis is not a "gender gap" but a "marriage gap" or a "family-formation gap."

Generation

Younger generations vote more Democratic than older generations on aggregate — but Gen Z's pattern is more complicated than Millennials' was. Young men in Gen Z have moved toward Republicans faster than any other generation; young women in Gen Z lean substantially Democratic. The gender gap among 18–29 voters in 2024 was the largest ever measured.

Whether young voters' Democratic lean is a generational effect (their cohort will stay Democratic as they age) or a life-stage effect (they will move rightward as they marry, buy homes, and pay property taxes) is one of the most consequential open questions in American electoral analysis.

A note on what the demographic data does not tell you

Before we move to the explanation section, an honest caveat. The demographic correlates above are descriptive averages. They are not iron laws. Within every demographic group, there is wide variance in voting behavior. Not every white evangelical votes Republican. Not every college-educated suburbanite votes Democratic. Not every young voter is engaged. Not every retiree turns out.

The averages are useful for understanding aggregate electoral coalitions. They are not useful for understanding any particular voter. A voter is not a statistical artifact of their demographic categories; they are a person with their own history, family, religious life, occupation, neighbors, and personal experience of politics. Demographic predictors give you the prior probability of how a voter behaves; they do not give you the voter.

This matters for two reasons. First, respecting voters' agency is a democratic commitment, not just a methodological one. The framing of "Hispanic voters" or "non-college whites" as monolithic blocs erases the reality that millions of individuals are making individual decisions, often in tension with their demographic group's modal pattern. Second, the demographic frame is also a strategic frame, used by parties and campaigns to allocate mobilization resources. When the frame becomes too coarse, it produces strategic errors — campaigns that assume "Hispanic voters" want a certain message based on the average without recognizing the within-group diversity, and consequently fail to reach the voters whose support they assumed.

The chapter uses demographic data because it is the cleanest way to describe broad electoral patterns. But you should always read demographic claims with a small mental caveat — on average, with substantial variance, with individual voters retaining the agency to defy the average. The data is a starting point, not a verdict.

5. Why people vote (or don't)

If turnout patterns describe what happens, this section asks why. Why do some people vote and others not? Why do people vote at all, given that any individual vote has essentially zero probability of deciding the election?

The rational-choice paradox

The rational-choice model of voting — first formalized by Anthony Downs in 1957 — points out a problem. The expected payoff of voting equals the probability your vote is decisive, multiplied by the policy benefit you gain from your candidate winning, minus the cost of voting (time, transportation, information). For any plausible election, the probability that your single vote is decisive is in the range of one in a million to one in a hundred million. So the expected payoff of voting, narrowly understood, is essentially zero, and any positive cost — even five minutes of your time — should make voting irrational.

But people vote. So either Downs's model is wrong, or there must be benefits to voting that are not captured by the "probability of decisiveness times policy payoff" framework. Theorists have proposed several.

• Civic-duty value. People vote because they believe they ought to. Citizenship implies a duty to participate, and discharging that duty has intrinsic value to the voter. The size of this duty value varies by individual; people for whom it is large vote, people for whom it is small don't.
• Expressive value. Voting is a form of self-expression. A vote for a candidate is a statement of identity and values, with symbolic payoff to the voter independent of whether the vote affects the outcome. (Compare: cheering at a sports event. Your cheer doesn't determine the score, but the cheering is itself the point.)
• Group-conformity value. People vote because their family, friends, or community vote, and they want to be the kind of person their family/friends/community is. This explains why turnout is contagious within networks.
• Investment value. A vote is part of a long-term investment in democratic institutions. No single vote is decisive, but the system depends on enough votes being cast, and you contribute your share.

None of these is fully satisfying as a complete explanation. Together, they suggest that voting is a complex social and psychological behavior, not a narrowly instrumental one.

The resource model

Sidney Verba, Kay Schlozman, and Henry Brady's landmark 1995 book Voice and Equality argues that political participation requires three resources: time, money, and civic skills. All three are unequally distributed.

• Time. Voting takes time — registration, learning what's on the ballot, getting to the polling place, waiting in line. Lower-income workers with multiple jobs and inflexible schedules have less of this resource.
• Money. Voting itself is free, but related activities (researching candidates, taking time off work, paying for childcare or transportation) cost money.
• Civic skills. The ability to read forms, navigate bureaucracies, understand ballot language, and parse political information. These skills are taught in school, in workplaces, and in voluntary associations — but unevenly.

The resource model predicts (correctly) that turnout will be highest among voters with the most time, money, and civic skills, and lowest among those with the least. It also explains why the participation gap maps closely onto educational, racial, and class lines.

Mobilization

Voters are more likely to vote when contacted by a campaign or community organization. Field experiments by political scientists Donald Green and Alan Gerber have measured this effect rigorously since the late 1990s.

Door-knocking is by far the most effective mobilization technique, but also the most expensive per contact. Robocalls are nearly free per contact and nearly worthless. Most modern campaign mobilization is a calculation of cost-per-marginal-vote.

Mobilization effects are most powerful for infrequent voters — people who would not have voted absent contact. Campaigns target their resources accordingly: the goal is not to persuade reliable voters who are going to vote anyway, but to mobilize unreliable voters who match the campaign's coalition.

Habit

Voting is habit-forming. People who vote in their first eligible election are substantially more likely to vote in subsequent elections than people who don't.

A key finding from Gerber, Green, and Shachar (2003): a person mobilized to vote in one election is roughly 7 percentage points more likely to vote in the next election, even without being mobilized again. The first vote creates the habit; the habit persists.

This has two implications. First, first-vote mobilization compounds. A campaign that mobilizes a 19-year-old to vote in 2024 is also indirectly affecting that voter's behavior in 2028, 2032, and onward. Second, non-voting is also habit-forming. Someone who is eligible to vote but doesn't in their first three elections will, statistically, struggle to start voting later in life. The under-30 turnout problem is partly a habit problem — voters who skip their first elections may never become regular voters.

Pre-registration and youth-voting policy

The habit-formation finding has motivated a particular policy reform. About 18 states now permit pre-registration of 16- and 17-year-olds, who are added to the rolls automatically upon turning 18. The theory is that capturing voters before they leave high school — when they have a stable address, a known school context, and a likely civics class teaching them about the process — converts more would-be eligible voters into registered ones, who are then more likely to vote in their first eligible election, which sets the habit.

The empirical evidence on pre-registration is moderately positive: states with pre-registration see modest increases in 18-19-year-old turnout (~2-4 percentage points) without obvious offsetting effects elsewhere. The intervention is small but consistent with what habit theory predicts. The fact that it is the kind of policy that has bipartisan-pragmatic appeal — both parties want their voters to develop voting habits — has helped it spread to states with diverse partisan control.

The mobilization-habit-resource framework, integrated

Putting sections 5's pieces together: a person votes when (a) they have the time, money, and civic skills to do so cheaply; (b) they were mobilized by a campaign or community to do so this cycle; (c) they have voted before, generating a habit; or (d) they hold values (civic duty, expressive identification, group conformity) that make voting intrinsically rewarding. These four pathways are mutually reinforcing rather than competitive. A high-resource voter who was mobilized last cycle and developed the habit, and who values civic participation, will vote at near-100% probability. A low-resource voter who was not mobilized, has no habit, and does not place high intrinsic value on voting will vote at near-0% probability.

Most Americans are between these poles. Most voters have moderate resources, are sometimes-but-not-always mobilized, have variable habits, and assign moderate intrinsic value to voting. Their turnout decisions are the consequential ones. Whether they vote in any given cycle depends on whether the cycle's stakes feel high enough to overcome the costs, whether anyone reaches them, whether they happen to be in a habit groove, and whether their life circumstances that month (a new job, a child's illness, a recent move) leave them the bandwidth to participate.

This framework is humbling for anyone who wants to "fix" American turnout with a single intervention. There is no single fix. There are multiple small interventions — automatic registration, pre-registration, mobilization investment, mail voting, vote-by-mail infrastructure, civic education, weekend or holiday voting — each of which moves turnout by 1-3 points. Adding them together can move turnout substantially. Doing only one of them moves it modestly.

6. Why people choose how they choose

Once a person has decided to vote, what determines their choice?

Party identification

The single strongest predictor of vote choice is party identification — whether a voter thinks of themselves as a Democrat, Republican, or independent.

• About 90% of self-identified Democrats vote for Democratic candidates in any given election.
• About 90% of self-identified Republicans vote for Republican candidates.
• The roughly 30% of voters who call themselves "independent" lean toward one party in nearly all cases. "Pure independents" — voters who don't lean either way — are roughly 7–10% of the electorate.

Party identification is acquired in childhood and early adulthood, often inherited from parents and peer groups. Once formed, it is remarkably stable: the modal voter dies with the same party identification they acquired by age 25.

This stability is why most "persuasion" in elections is actually mobilization. The number of voters who switch parties between elections is small. The number of voters who choose whether to show up at all is much larger, and that's where most campaigns focus.

Issue positions

Issues matter, but at the margin. Most voters' issue positions are correlated with their party identification — Democrats favor Democratic positions on issues, Republicans favor Republican positions. The minority of voters whose issue positions are cross-pressured (Democrats opposed to abortion, Republicans favoring climate action) are the genuine swing vote on those issues.

For the 10–15% of voters who are genuinely persuadable, issue positions can move votes. The 2024 election saw immigration salience drive significant shifts: voters who ranked immigration as their top issue went heavily for Trump regardless of their other commitments.

Candidate evaluation

Voters evaluate candidates as people — perceived competence, honesty, empathy, strength, "looks presidential." These evaluations are partly endogenous to partisan filtering (voters perceive their own party's candidates as more competent than the opposing party's), but they have real independent effects.

In primary elections, where party identification is held constant, candidate evaluation is decisive.

Retrospective voting

Voters reward incumbents in good times and punish them in bad. When the economy is growing, the incumbent party tends to gain. When the economy is contracting, the incumbent party tends to lose. This is retrospective voting.

The classic case: 1980. Inflation was 13%. Unemployment was 7%. Hostages were held in Iran. Carter lost. The classic counter-case: 1996. Economy was strong. Clinton won easily.

How much retrospective voting reflects rational performance evaluation — vs. arbitrary blame-assignment for events outside the incumbent's control — is contested.

• Christopher Achen and Larry Bartels (Democracy for Realists, 2016): Retrospective voting is closer to arbitrary blame-assignment than rational evaluation. Voters punish incumbents for shark attacks (Achen and Bartels' famous example) and for droughts and floods. Voters' performance evaluations are more partisan-rationalized than evidence-based.
• Andrew Healy and Gabriel Lenz, in subsequent work, push back: voters are clumsy retrospective evaluators, but they do respond to actual performance, especially on the economy.

The most defensible synthesis: voters do reward and punish incumbents based on perceived performance, but the perceived-performance signal is noisy, partisan-filtered, and sometimes anchored on irrelevant cues. It is a form of accountability, but a flawed one.

Economic voting

A subset of retrospective voting: voters reward or punish based on the economy. James Carville's 1992 phrase — "It's the economy, stupid" — captures the conventional wisdom.

The reality is more nuanced. Subjective economic perceptions correlate strongly with vote choice. Objective economic indicators — GDP growth, unemployment, real wages — correlate less strongly, especially in the polarized era. In 2020 and 2024, voters' perceptions of economic conditions varied dramatically by party: Republicans who saw the economy as terrible under Biden in 2024 mostly believed it was terrific under Trump in 2020, and vice versa, despite indicators that didn't change much across the boundary.

This suggests economic voting is now heavily partisan-filtered: voters perceive the economy through their partisan lens, and the resulting perceptions drive their votes. The exception is when objective indicators are extreme enough to override partisan filtering — sustained 7%+ inflation, double-digit unemployment, or comparable shocks. The 2022-2024 inflation episode broke through partisan filtering for many voters and drove a real, broad-based shift away from the incumbent Democratic Party.

Information environments and vote choice

A separate strand of research, increasingly important, looks at how the information environment shapes vote choice. Most American voters get most of their political information from a small number of sources: cable news (different channels for different audiences), social media platforms (different algorithms for different audiences), local newspapers (where they still exist), and trusted personal networks (family, friends, coworkers, religious community).

The post-2010 fragmentation of these information environments means that voters in different partisan coalitions are often getting different facts about the same events. They are not just disagreeing about how to interpret shared facts; they are disagreeing about what the facts are.

This has consequences for vote choice. A Republican-leaning voter watching Fox News, listening to conservative talk radio, and following conservative figures on X is being told a story about the country in which (say) inflation is severe, the border is in crisis, and crime is rising. A Democratic-leaning voter watching MSNBC, reading the New York Times, and following progressive figures is being told a story in which inflation is moderating, the border is being managed, and crime is declining. Both stories have some empirical basis; both also reflect editorial choices about emphasis and framing. The resulting disagreement is not entirely about values; it is partly about which set of facts each voter trusts.

How much information environments drive vote choice — vs. information environments being effects of partisan identification rather than causes — is contested. The most defensible synthesis: information environments and partisan identification reinforce each other in a feedback loop. Voters self-select into information environments congenial to their partisan priors; the environments then reinforce and intensify those priors. Breaking the loop is hard.

The persuasion-mobilization tradeoff

A practical implication for campaigns. In the highly polarized post-2016 environment, the population of voters who are persuadable — who could plausibly switch from one party to the other — is much smaller than it was in earlier decades. Estimates from validated voter analyses suggest that perhaps 7-12% of voters in any given cycle are genuinely cross-pressured or undecided.

This means that persuasion (changing voters' minds about which candidate to support) yields lower returns per dollar than mobilization (turning out voters who already prefer your candidate but might not show up). The strategic implication is that modern campaigns spend more on mobilization and less on persuasion than they did 30 years ago. The 2024 cycle saw both campaigns invest heavily in mobilization on the assumption that the persuadable middle was small.

Whether this is good for democracy is contested. Mobilization-focused campaigns reach more voters but reinforce existing partisan divides. Persuasion-focused campaigns try to win cross-pressured voters but reach fewer people. The trade-off is real and the equilibrium depends partly on how polarized the electorate is. As polarization has grown, mobilization has dominated. If polarization eases — a question we return to in Chapter 25 — persuasion may again become more central.

7. Voter registration and the franchise

How a person becomes eligible to vote is a state-by-state legal architecture.

National Voter Registration Act (1993, "Motor Voter")

Federal law signed by President Clinton, requiring states to:

• Allow voter registration when applying for or renewing a driver's license.
• Provide registration through public-assistance offices.
• Accept mail-in registration applications.
• Maintain accurate voter rolls (with restrictions on how purges can be conducted).

Motor Voter is the foundation of modern American voter registration. It dramatically increased registration rates without altering the basic state-level architecture.

Help America Vote Act (2002)

Federal law passed in response to the 2000 Florida controversy:

• Required states to replace punch-card and lever voting machines with newer technology.
• Established the Election Assistance Commission (EAC).
• Required states to maintain centralized voter-registration databases.
• Required first-time registrants who registered by mail to provide ID at the polling place.

HAVA modernized voting infrastructure but did not create a uniform national system; states retained control of most administrative details.

Online registration

About 42 states now permit online voter registration, typically through the state's secretary-of-state portal. Online registration has been associated with modest increases in registration rates (~3 percentage points), particularly among younger and lower-income voters.

Same-day registration / Election-day registration

About 22 states permit voters to register and cast a ballot on the same day, including election day itself. These states typically have higher turnout — by ~5–10 points, after controlling for demographics — than states without same-day registration. The mechanism is partly mechanical (it captures voters who realize on election day that they're not registered) and partly that the option signals a more accessible voting environment.

Examples of same-day-registration states: Minnesota, Wisconsin, Iowa, Maine, North Carolina, Hawaii, the District of Columbia.

Automatic voter registration

About 22 states have adopted automatic voter registration (AVR), under which citizens are automatically registered when they interact with the state DMV (or other specified state agencies), unless they affirmatively opt out. Oregon was the first state to adopt AVR in 2015. AVR has been associated with modest registration gains (~2-5 points) and somewhat smaller turnout gains.

Voter ID

About 36 states require some form of voter identification at the polls. The strictness varies considerably.

• Strict photo ID (about 9 states): voters must present a current government-issued photo ID; provisional ballots without ID are not counted unless ID is provided after the fact.
• Non-strict photo ID (about 7 states): photo ID requested, but voters without one can sign an affidavit or vote provisionally.
• Strict non-photo ID (about 4 states): government-issued ID required, but it doesn't have to be photo.
• Non-strict non-photo ID (about 16 states): ID requested, alternatives accepted.
• No ID required at the polls (about 14 states + DC): voter signature on the rolls suffices.

Voter ID is one of the most contested election-administration questions in American politics, and we treat the contested parts in section 9.

State-by-state variability

The most important fact about American voting administration is that it is not national. The Constitution, in Article I, Section 4, gives states primary authority over the "Times, Places and Manner" of elections, with Congress retaining the power to override state rules. Federal law (NVRA, HAVA, the Voting Rights Act) sets some floors, but the substantive shape of voting administration is determined state by state.

This variability cuts in multiple directions.

• Two voters in different states are voting under genuinely different rules. A 22-year-old in Oregon, where every registered voter receives a ballot in the mail and registration is automatic at the DMV, faces almost no friction in voting. A 22-year-old in Tennessee, where there is no online registration and no early voting except in person at a courthouse, faces substantial friction. The 22-year-old's "decision to vote" is conditioned on the architecture in which they live.
• State variability allows policy experimentation. Different rules in different states make possible the comparative empirical analysis that anchors much of the political-science literature on voting laws. Researchers can study what happened to turnout when (say) Wisconsin imposed a strict voter-ID law while Minnesota next door did not, holding most state-level confounders roughly constant.
• State variability also produces inequities of access. Voters in some states have substantially easier paths to participation than voters in other states, for reasons that have little to do with voter behavior or interest and everything to do with state legislative choices.

The federal-vs-state question is one of the deeper structural debates in American electoral policy. Reformers who want a national voting-administration baseline (a federal "right to vote" statute, automatic federal voter registration, uniform mail-voting access) argue that the state-level patchwork produces unjustifiable inequality of access. Reformers who want to preserve state authority argue that decentralization protects against federal capture and allows states to tailor administration to local conditions. Both sides have constitutional and policy arguments. We return to this in Chapter 36 (Voting Rights).

8. Mail and early voting

How votes are cast — and counted — has been transformed in the past decade.

The 2020 transformation

In 2020, partly because of COVID-19 and partly as the culmination of pre-existing trends, mail and early voting exploded.

The 2020 numbers show the pandemic effect: about 65% of votes were cast before election day. By 2024, in-person election-day voting had partly rebounded but the share of votes cast before election day (~53%) remained higher than the pre-2020 baseline.

Partisan asymmetry

In 2020, Democratic voters disproportionately used mail and early voting; Republican voters disproportionately used election-day in-person voting. This asymmetry was partly endogenous to messaging — Trump campaigned hard against mail voting, his voters listened, and they voted in person on election day instead.

In 2022, the asymmetry remained but began to close. By 2024, the Republican Party leadership had reversed strategy and begun actively promoting early and mail voting. The 2024 cycle saw substantial Republican mail and early-voting investment, and the partisan gap in vote modes narrowed substantially. Trump in 2024 still outperformed Harris on election day and Harris outperformed Trump in early/mail voting, but the gap was much smaller than in 2020.

Permanent mail-ballot states

Eight states now mail every registered voter a ballot for every election, with no application required: California, Colorado, Hawaii, Nevada, Oregon, Utah, Vermont, and Washington. The systems vary in detail (some require returning by mail; some allow drop boxes; some allow in-person delivery), but all share the principle that the ballot comes to the voter rather than the voter going to a polling place.

These states have higher turnout, on average, than otherwise-similar states with traditional polling-place voting. Whether the turnout gain is purely a function of the mail-ballot system or partly reflects unobserved state-level factors is contested. Multiple peer-reviewed studies have found modest but real turnout gains (~2–4 percentage points) attributable to all-mail systems.

Drop boxes and ballot collection

Many mail-voting states permit voters to return their ballots to drop boxes or to have a third party (a family member, a community organization) return the ballot. Drop boxes were a focal point of the 2020 election controversies, with some states subsequently restricting their use (Wisconsin restricted drop boxes after 2020 but reversed after a 2024 court ruling). "Ballot collection" rules — who can return another person's ballot — also vary state by state and have become a partisan flashpoint.

9. Structural barriers and the "voter suppression" debate

We come now to the genuinely contested terrain. The legal architecture of voting in the United States is the subject of intense partisan disagreement. The disagreement has empirical components (do these laws actually reduce turnout?) and normative components (what should the rules be?). We have to handle both carefully.

The progressive critique

The progressive position is that many post-2010 state-level voting laws — voter ID, polling-place closures, registration purges, signature-match requirements, restrictions on early voting and mail voting, restrictions on ballot collection — have been targeted, designed, and implemented in ways that disproportionately reduce turnout among Democratic-leaning voters: Black voters, Hispanic voters, young voters, low-income voters, and voters in densely populated urban areas. The progressive critique has several components.

• Disparate impact. Even race-neutral laws (voter ID, polling-place consolidation) have demonstrable disparate impacts on minority and lower-income voters, who are less likely to possess the relevant ID, more likely to lack transportation, more likely to work shifts that conflict with restricted polling hours.
• Targeted design. Some laws appear designed with disparate impact in mind. The 2013 North Carolina voting law, struck down in 2016 by the Fourth Circuit (NC NAACP v. McCrory), was found to have targeted Black voters "with almost surgical precision."
• Cumulative effect. Individual laws may have small effects, but the cumulative architecture of multiple restrictive laws stacked together has substantial effects.
• Asymmetric advocacy. The push for these laws has come overwhelmingly from Republican-controlled legislatures, in states where Republican electoral fortunes have been threatened by changing demographics.

The conservative defense

The conservative position is that election-administration measures — voter ID, signature matching, citizenship verification, voter-roll maintenance, restrictions on universal mail balloting and ballot collection — are reasonable safeguards against fraud and administrative error, that they apply equally to all voters, and that the empirical claim of disparate impact has been overstated. The conservative case has several components.

• Election integrity has independent value. Voters need to trust that elections are run accurately. Procedures that protect against fraud and error are valuable even if their effect on turnout is small.
• Voter ID is normal. The vast majority of advanced democracies require ID to vote. The United States is unusual in lacking a national ID system, which makes voter-ID implementation messier than elsewhere — but the underlying principle (verifying identity) is not extreme.
• The empirical evidence on suppression effects is contested. Multiple peer-reviewed studies have found small, mixed, or zero aggregate effects of voter ID laws on turnout. The studies that find large effects often have methodological problems.
• Universal mail balloting carries risks. Permanent mail-ballot lists can include people who have moved, died, or never wanted a ballot. The risks are administrative (ballots going to wrong addresses, dead voters, non-citizens registered through DMV records) more than fraudulent, but they are real.
• Ballot collection is a particular concern. Permitting third parties to collect and return others' ballots creates a documented (though small-scale) risk of coercion and pressure within households or communities.

What the empirical evidence does and doesn't establish

Several rigorous studies have attempted to measure the actual turnout effects of these laws.

• Hajnal, Lajevardi, and Nielson (2017), Journal of Politics: found that strict voter ID laws produced substantial declines in minority turnout. (Subsequent reanalysis disputed the magnitude.)
• Hopkins, Rocco, and Hill (2017–18): reanalyzed the same data with different methodological choices and found smaller, less consistent effects.
• Cantoni and Pons (2021), Quarterly Journal of Economics: using a different research design (administrative voter-file data), found that strict voter ID laws had no statistically significant aggregate effect on turnout, including no significant differential effect on minority turnout. This study attracted considerable attention as a serious challenge to the larger-effect findings.
• Anoll, Engelhardt, and Pearson-Merkowitz (2022): found that restrictive laws do reduce turnout, but the effects are small in aggregate and concentrated in specific subpopulations.
• Multiple studies on polling-place closures: consistent finding of small but measurable turnout effects, especially for voters whose polling place is moved farther away.
• Multiple studies on early-voting restrictions: generally find that reducing early-voting days has modest aggregate turnout effects.

The most defensible synthesis: most voter-administration measures have smaller effects on aggregate turnout than activists on either side claim, but they can be consequential at the margin (sufficient to change a close election outcome) and they can have disparate effects on specific demographic groups. The disparate-effects question is harder to litigate than the aggregate-effects question — a law that reduces overall turnout by 0.5 points and minority turnout by 1.5 points has both a small aggregate effect and a real disparate impact, and how to weigh those facts depends on values, not just data.

Steel-manning both sides

The genuine progressive concern is real: many post-2010 voting laws were enacted by Republican-controlled legislatures in states where electoral demographics were shifting, and the laws were designed in ways that, on average, disadvantaged Democratic-leaning constituencies. The fact that effects are smaller than activists claim does not mean effects are zero, and the fact that effects are disparate matters even when aggregate effects are small.

The genuine conservative concern is also real: election integrity is a legitimate value; voter ID is not extreme by international standards; universal mail balloting and unrestricted ballot collection carry administrative and integrity risks; and accusations of "voter suppression" applied to laws with measurable but small effects can feel like demagoguery to voters who support those laws on substantive grounds.

The honest reading: the United States has a state-by-state administrative patchwork that is messier than most peer democracies, and reform proposals from both directions are responding to genuine problems even when their advocates overstate their case. Chapter 36 (Voting Rights) revisits these issues in a more historical and constitutional frame.

10. The "registration cliff"

We close with a number. About 25% of voting-eligible Americans — around 60 million people — are not registered to vote. They are disproportionately younger, more mobile (recent movers), lower-income, and non-white. They are the voters who would most affect election outcomes if they were to register and vote, and they are the hardest to reach.

The reasons for non-registration are partly logistical (people move, registration doesn't follow them) and partly motivational (people don't believe their vote will matter, don't see candidates speaking to them, or don't trust the system). Reaching them requires sustained, expensive outreach — exactly the kind that under-resourced campaigns and weakly mobilized constituencies struggle to do.

Who the unregistered are

Demographic profiles of the unregistered population (drawn from Census data, the Pew Research Center, and the Current Population Survey) consistently show:

• Younger. About 35% of eligible 18-29-year-olds are unregistered, compared to 15% of eligible 65+ Americans.
• More mobile. People who have moved in the past two years are about three times as likely to be unregistered as people who have lived at the same address for more than five years.
• Lower-income. Bottom-quartile income households are about twice as likely to have unregistered adults as top-quartile households.
• Less educated. Adults without high school diplomas have unregistration rates of ~40%, compared to ~12% for college graduates.
• More likely to be non-white. Hispanic and Asian American eligible adults have unregistration rates well above the white average; Black unregistration rates are roughly comparable to white.
• Concentrated in certain states. States with restrictive registration requirements have higher unregistered populations; states with automatic registration have lower.

Note that these are correlations among the same variables we have seen throughout the chapter. The unregistered population is, in demographic terms, a more extreme version of the population that votes at lower rates among the registered. The factors that depress turnout among registered voters also depress registration to begin with, in a stacked architecture.

What automatic registration could change

A natural question: what if registration were automatic? In states that have adopted some form of automatic voter registration (AVR), unregistered populations have shrunk modestly. Oregon, the first AVR state (2015), saw its registration rate climb from about 73% to about 87% within several years, with corresponding turnout gains in the lower single digits.

Universalizing AVR — through a federal law mandating automatic registration via DMV interactions, tax filings, or other administrative encounters — would, by most analysts' estimates, add 30-50 million Americans to the voter rolls. Whether they would then vote is a separate question; some research suggests AVR-added voters turn out at substantially lower rates than self-registered voters, partly because the population added is the population most likely to be infrequent voters even when registered. But registration is the necessary first step. You cannot vote in most American jurisdictions if you are not registered.

The federal-AVR debate is a useful illustration of the tradeoffs we have seen throughout this chapter. Supporters argue that the registration burden is the largest fixable barrier to American voting, that automatic registration is normal in peer democracies, and that adding 30-50 million eligible-but-unregistered voters to the rolls would correct an obvious participation deficit. Opponents argue that registration as a citizen-initiated act has democratic value, that AVR risks placing non-citizens or otherwise ineligible people on the rolls (a contested empirical claim), and that state-level decisions on registration architecture should be preserved against federal preemption. Both sides have arguments. The 2021 federal voting-rights legislation that would have expanded AVR nationally did not pass; whether it (or something like it) eventually does is one of the open questions in voting-administration policy.

The closing thought

This is the deeper version of the turnout question. American elections are not decided by the median voter alone. They are decided by the median voter who shows up. The composition of that group is shaped by every law, habit, mobilization effort, and demographic trend we've covered in this chapter.

The chapter has presented the data without normative verdicts on most of the contested questions. That is by design. The empirical patterns are what they are — turnout varies by demographic group, the post-2016 education realignment is real, the 2024 Hispanic and working-class shift is real, voter ID's effects are smaller than activists claim but real and disparate. What you make of these facts depends on values that go beyond the data.

What the data does establish: power flows to those who show up. American politics rewards the constituencies that mobilize. The Republican coalition's growing working-class multiethnic base, the Democratic coalition's growing college-educated suburban base, the older voters who decide so many elections, the young voters who could decide more if they showed up — these are the people whose decisions to vote or not to vote will shape the next several decades of American policy. The chapter does not tell you whose side to take. It tells you who is in the room.

That is who votes. That is who doesn't. Now you know why it matters.

Continue to: Exercises · Quiz · Case Study 1: The 2020 Turnout Surge · Case Study 2: The 2024 Hispanic and Working-Class Shift · Key Takeaways · Further Reading