Chapter 14: Turnout — Who Votes and Why

The conference room on the fourth floor of Maria Garza's Sun Belt campaign headquarters smelled like cold coffee and dry-erase markers. It was eleven weeks before Election Day, and Nadia Osei had covered the whiteboard with a 3x4 grid of demographic cells, each one holding a percentage: her modeled probability that a registered voter in that cell would cast a ballot in November.

Jake Rourke stood in the doorway, arms crossed. He had managed six Senate campaigns, two of them winning. He had watched campaigns fall in love with data and lose. He had watched campaigns trust their gut and win. He had also, he admitted privately, watched campaigns trust their gut and lose spectacularly. But the memory of the losses was harder to pin on any single decision.

"Your model says we should put our best canvassers on people who voted in 2018 but not 2020," Nadia said. "Low-propensity voters who have shown they can be activated. The persuadability scores cluster there. It's the highest expected-value deployment."

Jake looked at the grid. "Maria's base wants to see us in their neighborhoods. They want to know she hasn't forgotten them."

"I understand that. But from a pure vote-maximization standpoint—"

"Politics is never a pure anything," Jake said. He walked to the whiteboard, uncapped a marker, and circled the cell representing Black voters over 50 in the state's two urban counties. "These people turn out at 72 percent. We don't mobilize them. They mobilize us."

The tension in that room — model versus gut, marginal voter versus base, expected value versus political legitimacy — is among the oldest in applied political analytics. It also happens to be one of the most illuminating entry points into the academic study of voter turnout: who votes, why they vote, and what campaigns can do about it.

This chapter explores those questions systematically. We begin with the puzzle that animates the entire enterprise: why rational people vote at all. We then examine historical and international patterns in turnout, the structural barriers that suppress participation, the science of mobilization, the psychology of voting habit, and the analytics of predicting who will actually show up. By the end, you will understand both why turnout is so hard to predict and why getting it right is often the difference between winning and losing.

14.1 The Paradox of Voting

Before we can analyze turnout, we have to grapple with the fact that, according to one influential line of thinking, virtually no one should vote.

The argument comes from Anthony Downs's 1957 work An Economic Theory of Democracy, which formalized the idea that political actors, including voters, behave rationally in the sense of maximizing expected utility. For a voter, the expected utility of voting can be expressed as:

EU(vote) = P × B − C

Where P is the probability that one's vote is decisive (breaks or creates a tie), B is the benefit from one's preferred candidate winning, and C is the cost of voting — time, information, and so on.

The problem is that P is vanishingly small. In any election with more than a few hundred participants, the probability that a single vote determines the outcome is effectively zero. Even in the famously close 2000 U.S. presidential election in Florida, the margin of 537 votes out of nearly six million cast meant that the ex-ante probability of being decisive was less than one in ten thousand. In a statewide Senate race with three million voters, the probability is even smaller.

If P is approximately zero, then P × B is approximately zero regardless of how large B is, and as long as C is positive — even a trivial cost — the expected utility of voting is negative. Rational agents shouldn't vote.

Yet roughly 130 million Americans voted in 2020. This is the "paradox of participation," and it has generated an enormous literature attempting to resolve it.

14.1.1 Expressive vs. Instrumental Voting

One resolution replaces the purely instrumental model with an expressive model. Voting is not, on this account, primarily about influencing outcomes. It is about expressing identity, values, and group membership. When Nadia's model flags a 58-year-old Black woman in the state's second-largest city as a high-turnout voter, it is in part capturing the legacy of a community that fought for the right to vote and continues to exercise it as an act of collective memory and civic affirmation.

The sociologist Nina Eliasoph, in her ethnographic work on civic disengagement, found that many non-voters were not indifferent to politics but actively disengaged from the formal political sphere as a coping mechanism. They cared; they had opted out. Voting, in this light, is a decision about whether to re-engage with a system that many Americans experience as unresponsive.

The expressive account explains why turnout correlates strongly with measures of civic identity ("duty to vote"), group solidarity, and partisan affect — factors that have little to do with the probability of being decisive.

14.1.2 The D-Term: Adding Civic Duty

A more formal resolution introduces a "D term" to the Downs equation:

EU(vote) = P × B − C + D

Where D represents the intrinsic satisfaction from voting — the "warm glow" of civic participation, the psychological discomfort of free-riding, or what philosophers call expressive utility. If D is large enough to exceed C, rational actors vote even when P ≈ 0.

This formulation is somewhat unsatisfying because it risks becoming unfalsifiable — you can always postulate a large enough D to explain any observed behavior. But as a descriptive matter, it captures something real. Surveys consistently find that large majorities of Americans cite "civic duty" as a primary reason for voting, well ahead of any belief that their vote will be decisive.

💡 Intuition Check: The Calculus of Voting The puzzle of why people vote is not merely academic. It shapes how campaigns think about mobilization. If voting is expressive and identity-driven, then appeals to civic duty and group solidarity should be effective mobilization tools. If it is instrumental, then making people feel their vote is consequential (e.g., "this will be a close race") should work better. Research suggests both levers are real, which is why effective GOTV messaging often combines both: "Your vote matters AND it's your duty to your community."

14.1.3 Social Pressure and Visibility

A third resolution, with enormous practical implications, is social. Voting is at least semi-public. In the United States, voter rolls record whether (though not how) citizens cast ballots. Campaigns can obtain this information; so, in some states, can neighbors.

A landmark 2008 experiment by Alan Gerber, Donald Green, and Christopher Larimer sent different mailers to Michigan households: a control condition, a "civic duty" message, a message showing the voter's own record, and a message showing their household's record alongside those of their neighbors. The social pressure treatment — "we are watching" — increased turnout by 8.1 percentage points, the largest effect ever documented for a low-cost intervention. Even the "civic duty" message increased turnout by 1.8 points.

The social pressure result reveals that voting behavior is partly regulated by norms of social accountability. People vote partly because they will feel judged if they do not. This finding has been extensively replicated, with some controversy about the ethics of such interventions (a theme we will return to in Chapter 38).

14.2 Historical Trends and International Comparisons

14.2.1 The Long Arc of U.S. Turnout

Voter turnout in presidential elections has followed a distinctive pattern over American history. In the nineteenth century, turnout was extraordinarily high by modern standards — regularly exceeding 70 percent of eligible voters and sometimes approaching 80 percent. The election of 1896, which pitted William McKinley against William Jennings Bryan, saw nearly 80 percent participation. These figures are all the more remarkable given that women were largely excluded, making the effective denominator smaller but the percentage of actual eligible participants higher.

The early twentieth century saw a dramatic collapse in turnout, partly due to the disenfranchisement of Black voters in the South following Reconstruction and the related decline in competitive party politics in the one-party Democratic South. The introduction of the Australian ballot (secret voting) and registration requirements also imposed new costs. By 1924, presidential turnout had fallen below 50 percent.

Turnout recovered somewhat in the mid-twentieth century, reaching 63 percent in 1960 — the Kennedy-Nixon election, often cited as the most competitive of the modern era. The 1971 lowering of the voting age to 18, which added millions of low-turnout young voters to the eligible population, produced a short-term decline in measured turnout even as raw vote totals increased.

The 1996 election, with Bill Clinton's comfortable reelection campaign, saw presidential turnout fall to 49 percent — the lowest in more than 70 years. The contested 2000 election began a recovery, and 2020 produced the highest turnout since 1900: roughly 67 percent of eligible voters cast ballots, an astonishing mobilization in the middle of a pandemic, driven in part by expanded mail voting and intense partisan polarization.

📊 Real-World Application: 2020 Turnout Geography The 2020 election showed striking geographic variation in turnout. Minnesota had 80 percent eligible voter turnout; Tennessee had 55 percent. These differences reflect persistent variation in registration rates, electoral competitiveness (many states were non-competitive), and legal barriers. Within states, counties with high Hispanic populations showed some of the largest percentage-point increases in turnout from 2016 to 2020 — a trend with direct implications for states like the one where Maria Garza is running.

14.2.2 The United States as an Outlier

In comparative perspective, U.S. voter turnout is strikingly low. Among OECD countries, the United States typically ranks near the bottom of turnout leagues, often 30–40 percentage points below countries like Belgium, Sweden, or Denmark. Even Australia, whose compulsory voting law sets something of an upper bound, regularly exceeds U.S. figures by 30 points.

Scholars disagree about how much of this gap is attributable to structural features versus cultural factors. The structural case emphasizes:

Registration barriers. Most democracies use automatic or governmental registration; voters are registered by default and must opt out, or are registered by government administrative processes. In the United States, registration is the individual citizen's responsibility, and the process varies enormously by state. Estimates suggest that eliminating registration requirements could add 3–4 percentage points to national turnout.

Weekday voting. The United States holds elections on Tuesday, a legacy of nineteenth-century agricultural scheduling. Most other democracies vote on weekends or declare Election Day a holiday. Research on the cost-reduction effect of moving to weekend voting is mixed — habits and norms matter — but the inconvenience of Tuesday voting almost certainly reduces participation, particularly among hourly workers who cannot easily take time off.

Plurality electoral rules. In many congressional and local races, winner-take-all plurality rules reduce the incentive to vote if you live in a safe district. Proportional representation systems, common in Europe, ensure that every vote "counts" in the sense of affecting seat allocation. This may increase perceived efficacy, though the evidence is contested.

Candidate quality and partisan differentiation. Some scholars argue that lower U.S. turnout partly reflects voter assessments that the parties offer insufficient policy differentiation, reducing B in the Downs equation.

🌍 Global Perspective: Compulsory Voting Several democracies — Australia, Belgium, Brazil, Argentina — legally require citizens to vote. Australia imposes a small fine for non-compliance and has turnout consistently above 90 percent. Critics argue this produces "donkey votes" (alphabetically ordered or random choices) and dilutes the quality of electoral participation. Supporters argue it reduces demographic skews in the electorate, since optional voting systematically underrepresents poor, young, and minority voters. The normative debate is unresolved, but the empirical effect on turnout is unambiguous.

14.3 Registration Barriers and the Architecture of Participation

14.3.1 Automatic vs. Opt-In Registration

The United States is one of a small number of democracies that places the burden of registration on the individual. As of 2024, 21 states and the District of Columbia have implemented some form of Automatic Voter Registration (AVR), which registers eligible citizens automatically when they interact with government agencies (primarily the DMV) unless they opt out.

Early evaluations of AVR suggest meaningful turnout effects, though the estimates vary. A study of Oregon's AVR implementation found that AVR-registered voters turned out at lower rates than voluntarily registered voters in their first election, but the overall effect on turnout was positive because so many more people were registered. The turnout gap between AVR and traditional registrants narrows substantially by the second election, as AVR registrants develop voting habits.

The design of AVR matters greatly. "Back-end" AVR systems, where the transfer of data from DMV to voter rolls happens administratively and citizens can opt out, appear more effective than "front-end" systems that ask citizens to opt in at the point of government contact.

14.3.2 Voter ID Requirements

Voter identification laws have been among the most contentious electoral policy debates of the past two decades. By 2024, most states had some form of ID requirement, ranging from non-strict requests (any ID accepted, including non-photo ID, and voters without ID can sign an affidavit) to strict photo ID requirements.

The empirical debate about ID laws and turnout is genuinely contested. Some studies find significant negative effects on turnout among low-income, minority, elderly, and young voters — the groups least likely to have qualifying ID. Others find minimal effects, particularly after states implement free ID provisions. The disagreement reflects challenges in research design: identifying the causal effect of a law requires careful comparison with similar states or jurisdictions that did not implement the law.

A 2017 study in the Journal of Politics by Hajnal, Lajevardi, and Nielson found that strict voter ID laws reduced minority turnout by a larger margin than white turnout, increasing the white-nonwhite turnout gap. Subsequent critiques challenged the data construction. The unresolved empirical debate does not change the legal stakes: because ID laws differentially affect specific demographic groups, their constitutionality has been repeatedly litigated.

⚖️ Ethical Analysis: ID Requirements and the Right to Vote The policy debate about voter ID pits two legitimate concerns against each other: preventing fraud (instrumental and expressive value of electoral integrity) versus ensuring access (preventing disenfranchisement). The empirical base for substantial fraud rates in U.S. elections is very weak — documented fraud rates are typically measured in the low thousands per state in major elections, while the number of people lacking qualifying ID is measured in the hundreds of thousands. This asymmetry shapes the ethical analysis even if the empirical turnout effects are uncertain.

14.3.3 Roll Purges and Maintenance

Voter rolls require maintenance. Voters who move, die, or become ineligible need to be removed to maintain accuracy. But the mechanics and aggressiveness of roll purging have major consequences for legitimate voters.

The National Voter Registration Act (NVRA) of 1993 established baseline requirements for list maintenance, including protections for voters who have simply become inactive but not moved or died. The Help America Vote Act (HAVA) of 2002 added further requirements for list accuracy. Despite these protections, aggressive purging practices have removed legitimate voters, particularly in states using "use it or lose it" policies that remove voters for failure to vote in recent elections.

The Brennan Center for Justice estimates that approximately 17 million voters were purged from rolls in the two years before the 2018 election. The purge rate in states with histories of voting rights violations was higher than in other states, though causal attribution is complex.

For campaign analysts like Nadia, roll quality is a practical as well as normative concern. A voter file with stale records increases the cost of contact programs — you expend resources contacting people who have moved or been removed — and can distort turnout models trained on vote history.

14.4 The Science of Mobilization

14.4.1 The Green-Gerber Revolution

Before the mid-1990s, claims about what mobilized voters were largely anecdotal. Campaigns did door-knocking because campaigns had always done door-knocking. They made phone calls because it was cheap. Mailers went out because vendors sold them. The evidence base was weak.

Donald Green and Alan Gerber changed this. Their 1999 field experiment in New Haven, Connecticut, randomly assigned registered voters to receive door-to-door canvassing, live phone calls, or direct mail, with a control group receiving nothing. The results were striking: personal canvassing increased turnout by approximately 9 percentage points. Live phone calls produced a modest effect of about 5 points. Direct mail had no detectable effect.

This experiment, and the dozens it spawned, established field experiments as the gold standard for evaluating mobilization interventions and produced a new subfield that transformed campaign practice. The core finding — that personal, high-quality contact is the most effective mobilization tool — has been replicated across hundreds of subsequent experiments.

💡 Intuition Check: Why Does Personal Contact Work? The leading explanation is that personal contact addresses the social dimension of voting. A live conversation with a canvasser signals that your participation is noticed and valued. It activates social norms around civic duty. It also provides a concrete, near-term commitment ("Will you vote on November 5th?") that psychological research on implementation intentions suggests substantially increases follow-through. The more "personal" the contact — genuine conversation versus scripted recitation — the larger the effect.

14.4.2 What Works: A Taxonomy of GOTV Interventions

Three decades of field experiments have produced a reasonably robust taxonomy of what works, what doesn't, and what remains uncertain:

High-evidence, high-effect interventions:

Door-to-door canvassing by trained, enthusiastic volunteers increases turnout by 3–9 percentage points per person contacted. The effect size varies with quality of conversation, neighborhood density, and election salience. Campaigns that conduct genuine conversations — asking about issues, listening — outperform scripted "knock and walk" approaches.

Peer-to-peer (P2P) text messaging has emerged as one of the most cost-effective modern interventions, with well-implemented programs showing effects of 1–3 percentage points among contacted voters. P2P differs from mass texting in that real humans send personalized messages and respond to replies, mimicking some of the relational quality of door-to-door contact.

Social pressure mailers (the Gerber-Green-Larimer design) reliably produce 1–3 percentage point effects, though campaigns face real ethical and backlash risks in how aggressively they deploy them.

Moderate evidence:

Live phone calls by trained callers show effects of approximately 3–5 percentage points in well-implemented programs. The quality of the conversation matters substantially; robocalls and script-reading produce much smaller or null effects.

Early voting and vote-by-mail facilitation — helping voters understand and use expanded voting options — can increase turnout among low-propensity voters who face time or transportation barriers. The evidence is somewhat mixed because elections differ in how easy the mail/early voting process is.

Weak or null evidence:

Robocalls and automated phone messaging typically show null effects in well-powered experiments. Some experiments find negative effects, possibly due to annoyance.

Generic direct mail (issue advocacy without turnout ask) shows null to very small effects in experiments. Targeted mail with a specific turnout ask does slightly better.

Digital advertising (Facebook, display ads) for pure GOTV purposes shows inconsistent effects; some large experiments find null results. The uncertainty here is substantial and the research is evolving rapidly.

14.4.3 Saturation and Diminishing Returns

Effective mobilization programs run into diminishing returns. A voter contacted five times by canvassers, callers, and mailers does not turn out at five times the rate of a once-contacted voter. This has practical implications for resource allocation that Nadia's optimization model explicitly addresses.

The issue of "voter fatigue" — over-contacted voters becoming annoyed and less likely to support the campaign even if they vote — is less well-documented but empirically plausible. Most mobilization research focuses on turnout rather than support scores; the possibility that aggressive GOTV contact moves votes away from your candidate is understudied and potentially important.

📊 Real-World Application: The Garza Campaign's GOTV Stack Nadia's model allocates canvasser-hours based on three inputs: modeled turnout propensity, modeled support score, and contact cost (approximated by density and drive time). The core insight is that the highest-value targets are low-propensity voters with high support scores — people who would vote for Garza if they vote, but are unlikely to vote without encouragement. High-propensity Garza supporters don't need canvassing; they will vote regardless. High-propensity Whitfield supporters are net-negative contacts. The optimization problem is to find the low-propensity pro-Garza voters and get them to the polls.

Jake's objection is not wrong, but it is about a different objective. Deploying canvassers to high-turnout Garza neighborhoods maintains morale, signals respect to base communities, and serves important relational functions. These are real political goods. They just don't appear in a pure vote-maximization model.

14.5 Habit Formation and the Voting Life Cycle

14.5.1 Voting as Habit

One of the most practically important findings in turnout research is that voting is habit-forming. Voters who turn out in one election are substantially more likely to vote in subsequent elections, controlling for their pre-election characteristics. This is not merely because some people are consistently motivated to vote — the effect persists even in regression discontinuity designs that compare people who barely qualified to vote in a given election (and thus had a chance to form the habit) with those who barely did not.

Shankar Vedantam's popularization of the psychology of habit formation describes habits as behaviors triggered by contextual cues that become automatic over time. Voting fits this model: the experience of going to a polling place, receiving a ballot, and completing the act of voting creates a pattern that is more likely to recur in future elections. The habit is stronger when the initial voting experience is positive (low wait times, helpful poll workers, a sense of efficacy).

The habit hypothesis has a practical implication that shapes many modern campaign GOTV strategies: the best investment may be getting low-propensity voters to vote in a lower-stakes election — a midterm, a primary, an off-year municipal election — because doing so increases their probability of voting in high-stakes future elections.

14.5.2 The First-Vote Effect and Youth Mobilization

Young voters illustrate the habit principle most sharply. Voters who cast their first ballot tend to continue voting at higher rates than those who don't vote in their first eligible election. This "first-vote effect" has been documented across multiple countries and election types.

For Maria Garza's campaign, this creates a specific challenge and opportunity. The state's fast-growing population of young Latino voters — many of them first- or second-generation citizens reaching voting age — represents a large pool of potential new voters. If mobilized in this election, they are more likely to continue voting in future elections, building a long-term electoral coalition. But they are low-propensity voters by definition (no voting history), expensive to reach, and require culturally competent outreach that builds trust rather than merely transacting a vote.

Garza's mobilization challenge with this population is not just about November — it is about whether the campaign seizes the opportunity to create habitual voters. Nadia's model assigns them moderate expected-value scores because their short-term persuadability is offset by their low voting propensity. Jake's political instinct is to invest here for reasons that don't show up in expected value calculations: the long-term coalition-building rationale that is genuinely important but hard to model.

🔴 Critical Thinking: What's Missing from Habit Models? Habit models treat voting as a relatively context-free individual behavior. But voting is embedded in social and political context. Voters in communities with a strong organizational infrastructure of churches, unions, ethnic associations, and civic groups show different habit formation patterns than isolated voters. The habit literature largely treats the individual voter, but the institution — the church that offers rides to the polls, the union hall that hosts voter registration drives — is often the causal agent. Models that miss institutional context may systematically underestimate turnout in well-organized communities and overestimate it in atomized ones.

14.5.3 Early Voting, Absentee Voting, and Vote-by-Mail

The expansion of early voting and mail voting over the past two decades has significantly changed the geography and timing of voter mobilization. In states with high mail-ballot usage (Oregon, Washington, Colorado have conducted all-mail elections for years; California significantly expanded mail voting in 2020), campaigns must completely rethink their GOTV timelines. "Get out the vote" becomes "get in the vote" — a weeks-long process of tracking returned ballots, identifying supporters who haven't returned their ballots yet, and nudging them to do so.

The shift to early voting also complicates turnout modeling. Historically, campaigns could model turnout as a binary Election Day outcome. With early voting, they must model not just whether someone will vote but when — and this affects the timing of mobilization contacts. A campaign that spends resources contacting voters who have already voted early is wasting money.

Modern voter file vendors provide daily updates on early ballot return by registered voters, enabling campaigns to build "ballot chase" programs that focus resources on supporters who have received but not returned a ballot. This is one of the cleaner applications of real-time analytics in campaign management.

14.6 Differential Turnout and Electoral Consequences

14.6.1 The Compositional Effect

Turnout is not politically neutral. Because different demographic groups have different voting rates, and because those groups also have distinct political preferences, the composition of the actual electorate diverges from the composition of the eligible electorate. This gap has systematic political consequences.

In the United States, voters are, on average, older, wealthier, whiter, and more educated than non-voters. The partisan implications of these demographic gaps shift over time as party coalitions evolve — in the current era, the college education gap means that the voting electorate is somewhat more Democratic than the eligible electorate among white voters, while the income and age gaps push in the other direction.

The implications for policy representation are debated but potentially significant. Political scientists like Larry Bartels and Martin Gilens have argued that elected officials are more responsive to the preferences of high-turnout, high-income constituents. If this is correct, low turnout among poor, young, and minority voters is not just a descriptive fact but a structural driver of policy outcomes.

14.6.2 The Surge and Decline Pattern

Congressional elections show a systematic "surge and decline" pattern first identified by Angus Campbell: turnout in presidential election years (surge) is substantially higher than in midterm years (decline), and the composition of the midterm electorate differs from the presidential electorate in predictable ways.

The midterm electorate is older, whiter, and more likely to identify as Republican than the presidential electorate — a structural disadvantage that has typically hurt the party holding the White House in midterms. Democrats have faced particular challenges with this pattern because their coalition skews younger and more racially diverse than Republicans', making their voters more subject to the surge-decline effect.

📊 Real-World Application: The 2022 Pattern The 2022 midterms were widely expected to follow the surge-decline pattern, with Democrats suffering major losses after Biden's 2020 surge. In fact, Democratic losses were modest by historical standards, in part due to the Dobbs abortion ruling mobilizing Democratic-leaning voters who might otherwise have stayed home. This illustrates that the structural surge-decline pattern can be disrupted by salient issues — an example of issue salience interacting with the mobilization function.

14.6.3 Turnout Differential in Close Races

In competitive elections, the relative turnout rate of each party's coalition is often more important than persuasion. A 5-point turnout advantage for one side can overwhelm a 3-point persuasion disadvantage. This is why serious campaigns devote enormous resources to turnout modeling and GOTV programs even when they appear to be winning — a collapse in your coalition's turnout can turn a projected 3-point win into a 2-point loss.

Nadia's model calculates what she calls the "turnout frontier": the set of plausible Election Day outcomes given different assumptions about relative turnout rates for Garza and Whitfield supporters. The model shows that Garza wins under most scenarios in which her coalition turns out at or above 2016 rates, but loses if her coalition's turnout falls below 2020 rates by more than 4 points. The key swing is precisely the young Latino voters Jake and Nadia are debating.

14.7 Turnout Modeling: How Campaigns Predict Who Will Vote

14.7.1 The Voter File as Foundation

Modern turnout modeling begins with the voter file, the comprehensive list of registered voters maintained by state election authorities. Commercial vendors — Catalist on the Democratic side, i360 and Data Trust on the Republican side — augment the official voter file with commercial consumer data, census demographic information, and survey-derived attributes to create the "enhanced voter file" that is the basic input to all serious turnout models.

The most powerful predictor of future voting behavior, by a large margin, is past voting behavior. Voters who have voted in three of the last four major elections are almost certainly going to vote in the next one. Voters who have never voted despite being registered for years are unlikely to vote without significant mobilization investment. The vote history columns in the voter file — which elections the voter voted in — are the backbone of any turnout model.

14.7.2 The Propensity Score

A turnout propensity score is a voter-level estimate of the probability that the voter will cast a ballot in a given upcoming election. It is typically expressed as a 0–100 score, where 100 means near-certain to vote and 0 means near-certain not to vote.

A simple but surprisingly effective propensity model might include: - Vote history: How many of the last N elections did the voter participate in? - Registration recency: New registrants have lower propensity, all else equal - Age: Middle-aged and older voters consistently have higher propensity than young voters - Geography: Voters in competitive districts or counties tend to have higher propensity - Party registration: Primary participation history signals higher engagement

More sophisticated models add hundreds of consumer data attributes (magazine subscriptions, purchase behavior, demographic proxies) and use machine learning methods — random forests, gradient boosting, regularized regression — to improve predictions.

The practical value of a well-calibrated turnout model is in resource allocation. If you have 10,000 canvassing hours to deploy, you want to know which of your 500,000 registered supporters are unlikely to vote without outreach (propensity score 30–60) versus which will vote regardless (propensity score 80+) versus which almost certainly won't vote no matter what you do (propensity score 0–20). The middle group is where mobilization resources have the highest expected value.

14.7.3 Calibration vs. Discrimination

A critical distinction in turnout modeling — one that practitioners often confuse — is between calibration and discrimination.

Discrimination refers to the model's ability to rank-order voters correctly: to give higher scores to people who will vote than to those who won't. It is typically measured by the area under the ROC curve (AUC) or Brier skill score.

Calibration refers to whether the scores mean what they say: if the model assigns 60% probability to a set of voters, do approximately 60% of them actually vote? A model can have excellent discrimination (it ranks voters correctly) but poor calibration (its 60% voters turn out at 45%, consistently).

For campaign resource allocation, discrimination is usually more important — you care about ranking. But for strategic planning ("how many total votes will we generate if we contact everyone with a score above 50?"), calibration matters enormously. Nadia's model is well-calibrated against past election outcomes in the state, but she is acutely aware that calibration degrades as you move further from the historical data — a state with changing demographics and new voting patterns is a harder target than a demographically stable one.

⚠️ Common Pitfall: Overconfidence in Propensity Scores Turnout models trained on past elections can fail badly in novel electoral environments. The 2020 COVID election, with its massive expansion of mail voting, produced systematic miscalibration in models that had been trained on in-person Election Day voting patterns. Several campaigns made serious resource allocation errors because their turnout models — built on 2018 and 2016 data — underestimated the share of low-propensity voters who would vote by mail. Model validation against recent analogous elections is essential; treating a model trained five cycles ago as current is a common and costly mistake.

14.7.4 The Nadia-Jake Synthesis

By the time the campaign reaches its final weeks, Nadia and Jake have arrived at a working synthesis. The bulk of the canvassing budget is allocated to Nadia's optimization targets — the low-to-medium propensity pro-Garza voters who need mobilization encouragement. But a meaningful fraction of the field organizer capacity is allocated to base community outreach in Garza's strongest neighborhoods, partly for the morale and organizational reasons Jake emphasizes, and partly because Nadia has updated her model with evidence from a small field experiment the campaign ran in August: relational organizing in tight-knit communities produces larger-than-expected effects, possibly because the social network multiplier (canvassed voters talking to their contacts) amplifies a single contact's impact.

This synthesis is not merely a political compromise. It reflects a genuine epistemic lesson: models built on individual-level data may systematically miss social and organizational dynamics that operate at the community level. Jake's gut about base communities is imprecise, but it is pointing at something real.

14.8 Summary: The Turnout Calculus

Voter turnout is the product of individual psychology, social networks, institutional structures, and campaign strategy. No single factor determines it, and no single model captures all of it. What the research record establishes clearly is:

1. Voting is costly (in the rational choice sense) and participation must be understood in terms of both expressive and instrumental motivations, not just one or the other.

2. Structural barriers matter — registration requirements, ID laws, roll purges, and polling place accessibility all affect participation rates, often in ways that are differentially burdensome by race, income, and age.

3. Personal contact mobilizes — the Green-Gerber research tradition has firmly established that high-quality personal outreach increases turnout, with effects that are meaningful at the margin of competitive elections.

4. Voting is habitual — past behavior is the strongest predictor of future behavior, which means campaigns' mobilization decisions have consequences beyond the current election.

5. Differential turnout shapes outcomes — who votes is at least as important as how they vote in determining electoral outcomes.

6. Models are tools, not oracles — turnout models improve resource allocation but require continuous validation, appropriate uncertainty quantification, and integration with qualitative organizational knowledge.

For Nadia, this means never treating her propensity scores as fixed facts. They are probabilistic estimates subject to revision as new information arrives. For Jake, it means acknowledging that the intuitions shaped by six campaigns are real data, even if they're hard to formalize. The best campaigns integrate both.

Chapter Summary

• The paradox of participation — why rational actors vote given the minuscule probability of being decisive — is resolved through expressive utility models, the D-term (civic duty), and social pressure effects.
• U.S. turnout is historically low by international standards, a consequence of structural barriers (registration requirements, weekday voting, roll purges) as well as cultural factors.
• Automatic voter registration, while imperfect, increases participation. Strict voter ID laws show mixed evidence on turnout effects, with some studies finding differential impacts by race and income.
• The Green-Gerber experiments established door-to-door canvassing as the gold standard GOTV intervention; live phone calls work; robocalls and generic mail do not.
• Voting is habit-forming: first-time voters are substantially more likely to vote in subsequent elections, making youth mobilization a long-term investment.
• Differential turnout creates a compositional electorate that systematically differs from the eligible electorate, with implications for representation and policy.
• Turnout models combine vote history (the dominant predictor) with demographic and consumer data; well-calibrated models support resource allocation but degrade in novel electoral environments.
• The Garza campaign's Nadia-Jake tension illustrates the real challenge of integrating model-based optimization with organizational knowledge about community dynamics.

14.9 The Psychology of Voting Decisions

14.9.1 Behavioral Economics and the Polling Place

The behavioral economics revolution of the late twentieth and early twenty-first centuries has added a rich new layer to turnout research. Traditional political science models treated voting as a deliberate, calculated decision — the product of weighing costs, benefits, and duty. Behavioral economics reveals that many decisions are powerfully shaped by how choices are framed, what cues are present in the environment, and what default options people face.

Several behavioral mechanisms have been documented in the turnout context:

Commitment devices. People who make explicit commitments to vote — even in an informal survey — are significantly more likely to follow through than those who are asked but don't commit. This is consistent with a large literature on implementation intentions: translating a general intention ("I should vote") into a specific plan ("I will vote on November 5 at 7 AM before work at the Riverside Community Center") dramatically increases follow-through. GOTV programs that help voters form concrete voting plans, not just tell them voting is important, show meaningfully larger effects than generic reminders.

Default effects. Automatic voter registration exploits this principle directly. When the default is "registered" (unless you opt out), far more people end up registered than when the default is "not registered" (and you must actively opt in). The behavioral implication extends beyond registration: default options for absentee ballot requests, polling place assignments, and even which candidates appear first on the ballot (the "primacy effect") can meaningfully affect outcomes.

Identity priming. Research by Christopher Bryan and colleagues showed that framing appeals around identity ("Be a voter") rather than action ("Vote tomorrow") produced significantly higher turnout. Asking people "How important is it to you to be someone who votes?" activates civic identity in a way that "Will you vote?" does not. The identity framing approach has been incorporated into modern GOTV messaging that emphasizes not the act but the identity: "Voters like you are making their voices heard."

Anticipated regret. People are averse to actions (or inactions) they expect to regret. Appeals that help voters anticipate how they would feel if they didn't vote and their preferred candidate lost by a small margin can activate this regret-avoidance motivation. This mechanism partly explains why close-race messaging ("This will be decided by a handful of votes") is effective even though the probability of being individually decisive is still near zero — the visceral anticipation of regret is activated independently of the actual probability calculation.

14.9.2 The Social Environment of Voting

Voting is embedded in a social context that individual-level models often miss. Who your neighbors vote for, whether your colleagues discuss politics, whether your religious community emphasizes civic participation — these social factors shape your probability of voting through mechanisms that are distinct from your individual demographic characteristics or partisan identity.

Alan Gerber and colleagues have documented what they call "contagion" effects in voting: turnout in one household is positively correlated with turnout in nearby households beyond what would be predicted by shared demographic characteristics. Part of this correlation is due to shared exposure to GOTV outreach (if a canvasser visits your block, they may visit your neighbors too). But part appears to reflect genuine social contagion — conversations about the election, observed departures for polling places, and the visibility of "I Voted" stickers all contribute to a social norm that activates individual participation.

The "I Voted" sticker is a small but well-studied example. Studies of sticker adoption and posting on social media suggest that visible voting behavior increases social pressure on observers to vote — the sticker functions as a social signal whose transmission is partly spontaneous and partly exploited by campaigns that encourage voters to post their "I Voted" photos.

📊 Real-World Application: Nadia's Behavioral Interventions The Garza campaign's GOTV design incorporates several behavioral insights beyond the basic door-knocking program. Volunteer canvassers are trained to ask voters not just "Will you vote?" but "When are you planning to vote and what's your plan for getting there?" — the implementation intention approach. Text messages to low-propensity voters are sent on a schedule designed to reach them approximately 48 hours before Election Day, when the "commitment period" research suggests maximum impact. Post-contact follow-up texts on Election Day morning include personalized reminders — "Today's the day — your plan was to vote at Lincoln Elementary at 9 AM" — that leverage the specific commitment elicited in the prior contact.

14.9.3 Costs, Inconvenience, and the Elasticity of Participation

A foundational insight from the turnout literature is that voting participation is elastic with respect to cost. Small reductions in the cost of voting — shorter lines, more polling locations, expanded hours, mail ballots — produce meaningful increases in participation, particularly among groups whose baseline participation is low. This elasticity is not uniform: voters who would vote regardless of modest cost variations show little elasticity; voters at the margin of participation show high elasticity.

This elasticity has direct implications for the electoral impact of changes in voting infrastructure. A county that closes half its polling places, doubling average wait times, will deter some voters — but not uniformly. Voters with transportation constraints, unpredictable work schedules, or child care responsibilities face sharply higher costs from longer lines. Voters who are retired, flexible, or highly motivated will absorb the increased cost without significant probability change.

The differential elasticity by income and employment status explains why polling place reductions and consolidations, even when not racially motivated in intent, can produce racially disparate effects: the voters most likely to be deterred by inconvenience are disproportionately represented among lower-income workers who cannot afford the luxury of waiting two hours to vote.

For campaign analysts, the elasticity insight suggests a targeting refinement: among low-propensity voters with high support scores, those who face high structural costs of voting (long commutes to polling places, no car, shift-work schedules) may be more cost-sensitive than the turnout model's propensity score captures. Identifying these "structurally constrained" low-propensity voters and providing transportation assistance, absentee ballot facilitation, or schedule-aware GOTV contacts is a form of targeted intervention that the basic propensity score doesn't optimize.

14.10 Voter Suppression vs. Voter Fraud: Framing the Policy Debate

14.10.1 Two Competing Narratives

The politics of voting access in the United States is organized around two competing narratives that talk past each other because they emphasize different parts of the empirical record.

The voter suppression narrative emphasizes the documented history of systematic disenfranchisement — from literacy tests and poll taxes in the Jim Crow South to the modern patterns of polling place closures, strict ID requirements, and aggressive roll purges that fall disproportionately on minority, poor, and young voters. This narrative focuses on the costs of exclusion: how many legitimate voters are deterred by each policy, and what the cumulative impact of multiple barriers is.

The voter fraud narrative emphasizes the importance of electoral integrity — ensuring that every vote cast represents a legitimate eligible voter who voted once. It focuses on the potential for fraudulent votes to dilute legitimate participation, and argues that measures like ID requirements and roll maintenance are reasonable safeguards that most citizens can easily comply with.

The empirical problem with the voter fraud narrative is that documented fraud rates in U.S. elections are extremely low. Studies by the Brennan Center, the Heritage Foundation (which compiled a list favorable to finding fraud), and academic researchers have consistently found that voter impersonation fraud — the type that ID requirements address — occurs at rates of tens to hundreds of cases per billion votes cast. Mail voting fraud is somewhat more common but still measured in the very low thousands per major election nationally.

The empirical problem with the voter suppression narrative is that demonstrating causal effect requires careful research design, and the effects of any single barrier (ID requirements, reduced hours, roll purges) are individually modest and hard to measure precisely. Critics argue that the literature overstates suppression effects by attributing turnout differences to policies rather than underlying demographic factors.

14.10.2 The Analyst's Role in the Policy Debate

For a political analyst, these competing narratives present a specific professional challenge: the data on voting access is politically contested, methodologically complex, and directly relevant to the competitive electoral environment in which you work. Campaign analytics teams may be tempted to highlight the framing most favorable to their candidate. Journalists and researchers may be drawn to dramatic findings over careful hedging.

The most intellectually honest position is to:

1. Acknowledge that fraud rates are demonstrably very low while acknowledging that proving a negative (no fraud occurred) is inherently difficult.
2. Acknowledge that suppression effects of any single policy are modest while acknowledging that cumulative effects of multiple barriers can be meaningful.
3. Distinguish between the empirical question (what do the data show?) and the normative question (what tradeoff between access and integrity is appropriate?) — and recognize that reasonable people can disagree on the normative question even if they agree on the empirical facts.
4. Apply the same standard of evidence to claims about suppression as to claims about fraud — demanding careful causal identification rather than accepting convenient correlations.

⚖️ Ethical Analysis: Analysts and Electoral Administration Political analysts who work in campaign environments have an interest in electoral outcomes that creates potential conflicts of interest when they analyze voting access policies. An analyst who works for a Democratic campaign has an incentive to find and publicize suppression effects; an analyst who works for a Republican campaign has an incentive to find and publicize fraud. The professional obligation is to follow the evidence regardless of its political implications — and to be transparent about the limitations and uncertainties in any analysis of these politically charged questions.

14.11 Long-Term Trends: Technology, Demography, and the Future of Turnout

14.11.1 Technology and Voting Access

The COVID-19 pandemic accelerated a transformation in voting methods that had been building for decades. By 2020, more than two-thirds of Americans voted by mail or in-person early rather than at their precinct on Election Day. The logistical implications for campaigns are substantial: the "Get Out the Vote" operation is now a weeks-long "Get In the Ballot" operation that requires tracking returned ballots, identifying supporters who haven't yet voted, and managing a complex chase program.

For turnout modelers, the shift to expanded mail and early voting has created both opportunities and challenges. Opportunities: real-time data on returned ballots allows daily updating of who has voted, enabling more precise targeting of outstanding contacts. Challenges: models trained on in-person Election Day voting behavior must be recalibrated for mail voting patterns, which differ systematically by age, income, and partisan preference.

Research on mail voting's effect on overall turnout is mixed. Some studies find meaningful positive effects; others find that convenience voting primarily shifts when existing voters cast their ballots rather than adding new voters. The effect may be larger for specific populations — elderly voters with mobility limitations, rural voters far from polling places, shift workers — than for the overall electorate.

14.11.2 Demographic Change and the Future Electorate

The United States electorate is undergoing substantial demographic transformation that will reshape the compositional dynamics of turnout for decades. The Hispanic/Latino population has grown rapidly and is expected to constitute an increasing share of the voting-age population over the next two decades. Young voters — who currently have among the lowest turnout rates — will become a larger share of the electorate as older high-turnout generations shrink.

These demographic trends do not automatically translate into political outcomes, for at least two reasons. First, demographic change in the eligible electorate only affects electoral outcomes if it translates into change in the voting electorate — and as this chapter has emphasized, there are large gaps between eligibility and participation, particularly for younger and lower-income voters. Second, political preferences are not fixed by demographic category; the partisan preferences of Hispanic voters, Asian American voters, and young voters have shifted and will likely continue to shift as coalitions evolve.

For the Garza campaign, operating in a Sun Belt state with ~32% Hispanic/Latino registered voters, these long-term demographic trends are not merely background context — they are the central strategic reality. The question of whether Garza can turn out the state's younger Latino voters at rates comparable to older non-Hispanic voters is not just a question about this election; it is a question about whether the emerging demographic coalition can be activated before structural shifts in the state's politics crystallize around other partisan alignments.

🔴 Critical Thinking: Demography Is Not Destiny The phrase "demography is destiny," often attributed to Auguste Comte, has become a cliché of American electoral analysis. Its implication — that demographic trends will inevitably favor one political party — is empirically dubious. Parties adapt. Coalitions shift. The political preferences of demographic groups are not fixed. Cuban Americans have historically voted Republican in patterns very different from Mexican Americans and Puerto Ricans, despite all being "Hispanic." Non-college white voters supported Democrats for most of the twentieth century before their dramatic shift toward Republicans in the 2010s. Any analysis that treats demographic categories as politically fixed is not political analysis — it is demography dressed up as politics. The analyst's job is to understand the contingent, contested, and dynamic relationship between demographic characteristics and political behavior.

Chapter Summary

• The paradox of participation — why rational actors vote given the minuscule probability of being decisive — is resolved through expressive utility models, the D-term (civic duty), and social pressure effects.
• U.S. turnout is historically low by international standards, a consequence of structural barriers (registration requirements, weekday voting, roll purges) as well as cultural factors.
• Automatic voter registration, while imperfect, increases participation. Strict voter ID laws show mixed evidence on turnout effects, with some studies finding differential impacts by race and income.
• The Green-Gerber experiments established door-to-door canvassing as the gold standard GOTV intervention; live phone calls work; robocalls and generic mail do not.
• Voting is habit-forming: first-time voters are substantially more likely to vote in subsequent elections, making youth mobilization a long-term investment.
• Behavioral economics adds implementation intentions, identity framing, default effects, and social contagion to the mechanisms through which campaigns can increase turnout.
• Participation is elastic with respect to cost; differential elasticity by income and employment status means that infrastructure changes have disparate effects on different populations.
• Differential turnout creates a compositional electorate that systematically differs from the eligible electorate, with implications for representation and policy.
• Turnout models combine vote history (the dominant predictor) with demographic and consumer data; well-calibrated models support resource allocation but degrade in novel electoral environments.
• The Garza campaign's Nadia-Jake tension illustrates the real challenge of integrating model-based optimization with organizational knowledge about community dynamics.
• Long-term demographic change is not electoral destiny; the relationship between demographic composition and political outcomes is contingent on mobilization, party adaptation, and coalition formation.
• Behavioral economics adds meaningful interventions beyond traditional GOTV: implementation intentions, identity priming, default effects, and anticipated regret all increase follow-through on voting intentions.
• The voting cost elasticity framework predicts that structural barriers — polling place closures, long lines, inconvenient hours — have differentially large effects on lower-income, shift-working, and transportation-constrained voters, producing disparate demographic impacts even without explicit demographic targeting.

A Final Note on What Turnout Research Cannot Tell Us

The extensive research literature on voter turnout tells us a great deal about who votes, under what conditions, in response to what interventions. What it cannot fully tell us is whether any particular individual or community would have voted if circumstances had been different — if the polling place had not been moved, if the canvasser had not knocked, if the community had had more robust organizational infrastructure, if the election had felt more consequential.

These counterfactual questions are ultimately what the normative debate about voting rights and access is about. The quantitative research gives us probability estimates; the normative debate is about what obligations those probabilities create. A turnout suppression effect of 2–3 percentage points in a population of 100,000 eligible voters means 2,000–3,000 people who would otherwise have voted did not. Whether that matters depends on what you believe about the value of political participation — whether it is primarily instrumental (affecting policy outcomes) or primarily expressive (affirming citizenship and belonging).

The political analyst who understands both the quantitative evidence and its normative stakes is better equipped not just to build better models, but to communicate honestly with campaigns, journalists, policymakers, and the public about what the data can and cannot support. That combination of technical rigor and normative awareness is the hallmark of responsible applied work in political analytics.

For Nadia, the weeks of work building propensity scores, optimization models, and persuadability targeting are ultimately in service of a simple goal: making sure that people who want to vote for Maria Garza actually do. The analytical complexity is a means to that end. But the normative stakes that surround voter participation — who is helped to vote and who is hindered, whose voices are amplified and whose are suppressed, what it means that some citizens' participation costs are systematically higher than others' — are the reason the work matters beyond its technical dimensions. Analytics without that awareness is sophisticated but empty. Analytics with that awareness is both technically rigorous and democratically meaningful.

The key analytical lesson of this chapter: turnout is not a fixed attribute of the electorate. It is a distribution shaped by institutions, incentives, social networks, and campaign effort — and that distribution is, at the margin, moveable. The question is always which levers are within your reach, which populations are genuinely persuadable to participate, and what the evidence says about how much movement is plausible. That is the turnout analyst's core task, and it requires holding both the technical precision of the propensity model and the human complexity that no model fully captures.

Next: Chapter 15 asks whether campaigns actually change votes — the question of campaign effects.