Chapter 29: Voter Targeting and Microtargeting

A week before the general election, a forty-three-year-old suburban homeowner in the Garza-Whitfield race receives something unexpected: a mailer from the Garza campaign that speaks directly to her concerns about property tax assessments in rapidly gentrifying neighborhoods. Her neighbor — a fifty-year-old retired teacher — receives a different Garza mailer, this one emphasizing Garza's record on public pension protection. The retired teacher's spouse, a 48-year-old who runs a small landscaping business, receives nothing from the Garza campaign at all; he's been scored as an unlikely Garza supporter, a lean-Whitfield voter whose outreach would be wasted resources. And across town, a 29-year-old Latino renter who voted in the last presidential election but skipped the last midterm receives a phone call — not a mailer — because research suggests that phone contact is more effective at mobilizing younger, more mobile voters.

None of this happened by accident. Each of these contact decisions reflects a theory about this specific voter: what they care about, what medium will reach them, whether they are persuadable, and whether contact is worth the cost. The political vocabulary for this practice is targeting. When it reaches the level of individual-voter customization, it is called microtargeting. This chapter examines how targeting works, what it is designed to accomplish, what evidence tells us about its effectiveness, and why it raises some of the most important ethical questions in contemporary democratic politics.

29.1 The Logic of Targeting

Targeting is fundamentally about resource allocation under scarcity. No campaign has unlimited money, volunteer hours, or staff time. Every dollar spent calling a committed opponent is a dollar not spent mobilizing a soft supporter. Every volunteer hour spent knocking on a door that will never open for the candidate is an hour not spent on a door that might. Targeting is the systematic attempt to direct scarce campaign resources toward the voters where they will have the greatest impact.

The basic targeting question has three components: Who should the campaign contact? With what message? And through what channel? Getting each of these right matters, and getting any of them badly wrong can waste enormous resources. A campaign that identifies the right voters but delivers the wrong message will fail. A campaign that has the right message for the right voters but delivers it through a channel those voters ignore will fail. Good targeting requires getting all three right simultaneously.

The Electoral Math of Targeting

Before any voter-level targeting is possible, campaigns must think clearly about the numbers. What is the likely vote universe — the total number of votes that will be cast? What share of that universe does the campaign need to win? How many voters are genuinely persuadable versus committed to one side? How many potential supporters are unlikely to vote without mobilization?

These aggregate questions set the parameters within which voter-level targeting operates. If a campaign needs 520,000 votes to win and the current best estimate is that it has about 490,000 votes in its base (voters who will vote for the candidate regardless of campaign contact), then it needs to either persuade roughly 30,000 additional voters or mobilize roughly 30,000 additional base voters who might not otherwise turn out — or some combination of both. The distribution between persuasion and mobilization determines the entire shape of the targeting program.

💡 Intuition: Think of the targeting decision as an investment portfolio. Persuasion targets are high-risk, moderate-reward investments — you're trying to change someone's mind, which is hard and often fails, but each success gains you a net vote (one vote less for the opponent, one more for you). Mobilization targets are lower-risk, lower-reward investments — you're activating someone who already supports you, which is more reliable, but each success gains you only one vote rather than two. The optimal portfolio depends on how many of each type of voter you have and how responsive they are to contact.

29.2 Types of Targeting Universes

Campaign targeting is organized around distinct voter universes, each designed for a different strategic purpose. Understanding these universes — what defines them, who falls into each, and what you do with them — is the operational core of modern field targeting.

The GOTV Universe

The GOTV (get-out-the-vote) universe consists of voters who are expected to support the campaign's candidate but who are at elevated risk of not voting without campaign contact. These are not persuasion targets — they already support the candidate — but they are mobilization targets. The goal is to increase their turnout probability through contact.

A typical GOTV universe is defined by the intersection of two scores: a high support score (the voter is likely to vote for our candidate) and a low-to-moderate turnout propensity score (the voter is not certain to vote). The sweet spot for GOTV investment is a voter who would vote for the candidate if she votes at all, but who has a meaningful probability of not voting. Too high a turnout score and you're spending resources on voters who would have voted anyway. Too low a turnout score and you're reaching voters so unlikely to vote that contact has limited return.

Nadia's GOTV universe for the Garza campaign numbered approximately 380,000 voters at the start of October — a substantial portion of the campaign's overall math. These voters were disproportionately concentrated in the state's urban and suburban counties, particularly among communities of color and younger voters whose historical turnout rates were lower than their registration numbers would suggest.

The Persuasion Universe

The persuasion universe consists of genuinely undecided voters — people whose vote choice is not predetermined by their partisan identification, ideology, or intense issue commitments. These are the voters who are, in principle, accessible to the campaign's message. They're also the rarest and most expensive to move.

Defining the persuasion universe requires both cutting and selecting. The campaign cuts committed supporters (who don't need persuasion) and committed opponents (for whom persuasion is impossible or cost-prohibitive). Within what remains, it selects voters with high enough turnout propensity to be worth persuading — a voter who is genuinely persuadable but extremely unlikely to vote is a low-priority persuasion target regardless of their openness to the campaign's message.

For Garza, the persuasion universe was concentrated in the state's suburban counties — communities that had been reliably Republican in statewide races through most of the prior decade but that had been shifting toward Democrats in recent cycles, particularly among college-educated voters. These voters were not automatic Garza supporters, but they were reachable: skeptical of Whitfield's more explicitly populist messaging, responsive to Garza's emphasis on institutional competence, and genuinely undecided in ways that earlier cycles' electorate had not been.

The Fundraising Universe

A separate universe exists for fundraising outreach: voters who are likely to donate money to the campaign. Fundraising targeting overlaps with support targeting but is not identical to it. High-support voters who are not going to donate are not worth the marginal cost of fundraising contact. High-capacity donors who are low-turnout voters may still be worth aggressive fundraising outreach even if the field team wouldn't prioritize them for canvassing.

Fundraising universes typically incorporate consumer data about income, household wealth indicators, and charitable giving history — data that has no role in GOTV or persuasion targeting. The overlap between campaign analytics and wealth screening is one of the places where the blending of political and commercial data is most visible.

The Volunteer Universe

A fourth targeting universe — less commonly discussed but operationally essential — is the volunteer recruitment universe: voters who are likely to become active campaign volunteers. Volunteer recruitment typically draws heavily on the campaign's existing email list and social network, but some campaigns have built models to identify likely volunteers in the broader voter file — using factors like past civic engagement, organizational membership, and issue intensity.

📊 Real-World Application: The 2008 Obama campaign's "neighbor-to-neighbor" approach required identifying volunteers who lived in or near target precincts, not just volunteers who were willing to drive to wherever they were sent. This required a geographic dimension to volunteer targeting that most campaigns had not previously developed — another example of how experimental insights (neighbors are more persuasive than strangers) drove analytical innovation (build a model of geographically proximate willing volunteers).

29.3 Predictive Scores: The Engine of Targeting

The targeting universe is defined by scores — numerical estimates of each voter's probability of supporting the candidate, turning out to vote, being persuadable, or donating. These scores are the analytical core of modern targeting, and understanding how they are built, what they can and cannot measure, and how they fail is essential for working with them intelligently.

The Support Score

A support score is a 0–100 estimate of the probability that a given voter will vote for the campaign's candidate. A score of 75 means the model estimates the voter has a 75% probability of voting for the candidate; a score of 25 means a 25% probability.

Support scores are typically built using logistic regression or more complex machine learning models applied to features in the voter file and consumer data. The features include:

• Party registration (in states with party registration, the single strongest predictor)
• Past vote history (which elections the voter has participated in, interpreted alongside precinct-level results)
• Demographic characteristics (age, geography, sometimes gender if available)
• Consumer data indicators (income proxies, homeownership, consumer preferences that correlate with political affiliation)
• Past canvass responses (if the voter has been contacted in previous cycles and their response was recorded)
• Issue survey responses (if the voter has responded to campaign or public surveys)

The model is trained on past election cycles where the outcome is known and validated against held-out data. In the current cycle, where no outcome is yet known, the model generates predictions rather than explanations — it tells you what it predicts will happen, not why.

⚠️ Common Pitfall: Support scores are often misread as statements of certainty rather than probability. A score of 72 does not mean this voter is a Garza voter; it means the model estimates a 72% probability that she is. In a large targeting universe, many 72-score voters will vote for Whitfield. This probability-versus-certainty confusion is one of the most common errors in campaign targeting — and one of the most consequential, because it leads to treating model predictions as facts.

The Turnout Propensity Score

The turnout propensity score estimates the probability that a voter will cast a ballot in the current election. The most important predictors of turnout are:

• Past turnout history: The single strongest predictor. A voter who voted in the last five elections will almost certainly vote in this one. A voter who has never voted despite being registered for three years is much less certain.
• Registration date: Newly registered voters have a significantly lower turnout probability in their first election than established registrants.
• Election type: Presidential year turnout is higher than midterm turnout for most demographic groups. Primaries and odd-year elections have dramatically lower turnout.
• Age: Older voters have higher turnout rates; the relationship flattens somewhat at the very oldest ages.
• Geographic density: Urban voters have somewhat lower individual turnout rates than suburban voters, controlling for other factors.

The Persuadability Score

The persuadability score is the trickiest to build. Persuadability — the probability that a voter's choice can be changed by campaign contact — is not directly observable. You observe whom someone voted for, not whether campaign contact changed their vote.

Campaign analytics firms build persuadability models primarily from two sources. First, field experiments (Chapter 30) in which randomly selected voters receive contact and their subsequent behavior is compared to a control group — the voters who respond most to contact in past experiments become the features of the persuadability model. Second, survey-based measures of political ambivalence, weak partisan attachment, and issue cross-pressure — voters who express conflicting views or weak partisan loyalty in surveys are modeled as more persuadable.

Both approaches have limitations. Field experiments measure behavioral responsiveness to contact, which captures GOTV responsiveness more cleanly than persuasion responsiveness. Surveys measure expressed ambivalence, which may or may not translate into actual vote choice flexibility. The persuadability score is the score that campaign practitioners should be most skeptical about, even as it is one of the most consequential for targeting decisions.

✅ Best Practice: Treat persuadability scores as rough categorizations, not precise probability estimates. Use them to distinguish the top third of the distribution from the bottom third, not to make fine distinctions within the middle. The signal-to-noise ratio in persuadability modeling is substantially lower than in turnout or support modeling, and false precision is a genuine operational risk.

29.3a Building a Support Score: A Technical Walk-Through

Understanding support scores at a conceptual level is different from understanding how they are actually built. A brief technical walk-through demystifies the process and clarifies where uncertainty enters.

Data preparation. The starting point is a voter file record for each registered voter in the campaign's target state. Each record includes the voter's registration information, party affiliation (where available), election participation history (which elections they voted in), and whatever consumer data has been matched to their record. The initial data set may have dozens to hundreds of variables per record.

Before modeling, analysts must address data quality problems. Missing values (a voter record with no consumer data match, a voter who registered too recently to have any turnout history) must be handled — either by imputation (filling in estimated values based on similar records), by creating explicit "missing" indicator variables, or by dropping the affected features from the model. Variables with very low variance (e.g., a consumer data variable that is missing for 95% of records) are typically dropped, as they contribute noise rather than signal.

Feature engineering. Raw voter file variables are often transformed into more useful model inputs. Turnout history — a binary variable for each past election (voted/didn't vote) — is typically aggregated into summary features: number of elections voted in out of last five, specific election types participated in (presidential, midterm, primary), trend in participation over time (improving, declining, steady). Canvass response codes from previous cycles, when available, are encoded as categorical variables.

The modeling approach. For support score modeling, logistic regression remains a widely used baseline because its outputs are naturally interpretable as probabilities and it is relatively resistant to overfitting in small to medium datasets. Gradient boosting algorithms (XGBoost, LightGBM) and random forests are widely used alternatives that often produce better predictive accuracy, particularly with large datasets and many features. Ensemble approaches — combining predictions from multiple models — are standard practice in the most sophisticated operations.

The target variable in support score modeling is typically partisan vote choice in a recent comparable election, inferred from precinct-level returns under the assumption of uniform voting within partisan cohorts. This is an imperfect target: the actual vote choice of a specific voter is unobservable, and precinct-level inference introduces error. But it is the best available approximation given the secret ballot.

Validation. A properly built support model is validated on held-out data — records that were not included in training. Typical validation metrics include AUC (area under the receiver operating characteristic curve, which measures the model's ability to discriminate supporters from opponents) and calibration plots (which assess whether voters with a score of, say, 70 actually voted for the supported candidate about 70% of the time in past elections).

Well-calibrated models perform meaningfully better than chance (AUC substantially above 0.5) and show good calibration across the score range. Poorly calibrated models may have reasonable AUC but systematically under- or over-predict at the extremes — overestimating support among high-scoring voters or underestimating it among low-scoring ones. Identifying and correcting calibration problems before the model is used for targeting is essential.

💡 Intuition: The difference between a well-calibrated and poorly calibrated model is not just statistical — it has practical consequences. If the model systematically underestimates support among a specific demographic group (say, newer Latino registrants in suburban precincts), the campaign will under-prioritize GOTV outreach for those voters. The modeling error becomes a mobilization error. This is why model validation is not a bureaucratic formality — it directly affects which voters get contacted and which don't.

29.4 Microtargeting: From Universe to Individual

Standard voter targeting identifies who to contact and roughly what message to emphasize. Microtargeting goes further: it attempts to customize both message and channel to the specific characteristics of the individual voter, using a granular understanding of what each voter cares about and how they prefer to receive information.

The term "microtargeting" was popularized during the 2004 Bush reelection campaign, which used consumer data matching to identify voters who were sympathetic to specific issue clusters — gun rights, traditional marriage, tax cuts — and customized mail and phone contact accordingly. The technique was described at the time as revolutionary; it has since become a standard element of sophisticated campaign practice, enormously refined by the expansion of digital advertising.

Message Microtargeting

Message microtargeting assumes that different voters are most responsive to different appeals. A suburban homeowner worried about property values receives a message about economic stability. A veteran receives a message about VA healthcare access. A parent of school-age children receives a message about education funding. None of these is the campaign's "main message" — they are issue-specific appeals calibrated to what the model predicts the voter cares most about.

Issue affinity models are typically built from survey data — respondents who identify as most concerned about healthcare are used to train a model that predicts healthcare issue affinity from voter file and consumer features. The resulting model assigns issue affinity scores to the full voter file, allowing the campaign to identify the likely healthcare-primary voter who never participated in a survey.

For the Garza campaign, Nadia's team built issue affinity models across six issue clusters: healthcare access, public education, criminal justice reform, economic opportunity, environmental quality, and immigration. These models were derived from a survey of approximately 4,000 registered voters conducted in July, combined with the experimental evidence from the campaign's digital ad testing program, which had been running issue-specific ads to different audiences and measuring engagement differentially.

Channel Microtargeting

Different voters are accessible through different channels. Older, high-turnout voters are more reliably reachable through direct mail and landline phone than through digital advertising. Younger, mobile voters may have no landline and move frequently enough that their registration address is unreliable. High-income suburban voters are more likely to have their mail read and less likely to respond to generic phone banking.

Channel targeting uses voter age, registration vintage, digital engagement history (if available), and consumer data about technology adoption to estimate which contact mode will be most effective for each voter. In practice, channel targeting often interacts with resource constraints — personal canvassing is the most effective contact mode but also the most expensive, so it is typically reserved for the highest-priority persuasion targets.

The Digital Microtargeting Revolution

The transformation that most dramatically changed microtargeting practice was the rise of programmatic digital advertising and the social media advertising platforms. Facebook's Custom Audiences feature, launched in 2012, allowed advertisers to upload lists of email addresses or phone numbers and reach those specific individuals with targeted ads. For campaigns that could match voter file records to Facebook accounts, this created the ability to deliver tailored political messages to specific, identified voters — not to a demographic category, but to a specific person.

This capability represented a fundamental shift from broadcast targeting (reaching everyone who watches a particular TV program, some of whom might be your target voters) to identity-based targeting (reaching specific voters you have already identified by name and voter file record). The efficiency gain is substantial. So are the concerns.

🔴 Critical Thinking: The shift from broadcast to identity-based targeting raises a question that targeting's efficiency benefits can obscure: what is lost when political communication moves from public broadcasting to private, personalized messaging? A TV ad is public — it can be viewed, critiqued, and fact-checked by journalists, opponents, and citizens who weren't the intended audience. A microtargeted Facebook ad visible only to the specific voters the campaign selected can contain claims that no outsider sees. The asymmetry of information between the campaign and the broader public is qualitatively different in microtargeting than in broadcast.

29.5 Nadia's Targeting Strategy: The Suburban Persuasion Play

By the second week of October, Nadia's targeting strategy for the Garza campaign had settled into a relatively stable form, though she was still adjusting it weekly based on incoming canvass data and the campaign's rolling tracker.

The strategy had two main components. The first was a mobilization push targeting the campaign's low-turnout supporters — primarily younger voters and voters in communities of color in the state's urban and suburban counties. For this universe, the campaign was deploying personal canvassing and phone banking, with a message emphasizing the practical stakes of the election: Garza's record on healthcare access and education funding, and the contrast with Whitfield's positions on both.

The second component was a persuasion program targeting suburban college-educated voters — a segment that Nadia's models identified as the most fluid. These voters — concentrated in the three suburban counties surrounding the state's largest city — showed support scores in the 40–60 range, suggesting genuine ambivalence. They had relatively high turnout propensity scores, meaning they would almost certainly vote; the question was who they would vote for.

For the persuasion universe, Nadia had developed two message tracks based on the campaign's issue affinity models. Track A emphasized Garza's background as Attorney General — her record on consumer protection and corporate accountability — and was targeted at voters whose consumer data profile and issue affinity scores suggested concern about economic fairness and institutional integrity. Track B emphasized Garza's positions on education funding and local government, and was targeted at voters whose profiles suggested school-quality concerns.

The channel mix for the persuasion universe was primarily direct mail (for older suburban voters with reliable addresses) and Facebook/Instagram advertising (for younger and middle-aged voters with digital engagement patterns that suggested receptivity to political advertising). Personal canvassing was reserved for the highest-priority persuasion targets — roughly 25,000 voters with support scores between 45 and 58 who were in precincts that Nadia's model identified as potentially decisive.

29.6 Jake's Counter-Strategy: Base Mobilization and Exurban Expansion

Jake Rourke's targeting strategy operated from a different premise. His reading of the state's electoral math said that the path to a Whitfield victory ran primarily through maximizing turnout in the exurban and rural areas that were Whitfield's strongest territory — places where Republican support was high and reliable, but where turnout in non-presidential years sometimes fell short of what the partisan composition of the population would predict.

The Whitfield targeting universe was accordingly mobilization-heavy. The field program prioritized high-support voters in lower-density areas — voters who would vote for Whitfield if they voted, but who had recent histories of inconsistent midterm participation. Jake's logic: a point of turnout increase in Whitfield-dominant territory was worth more than a point of persuasion in contested suburban areas, because mobilized base voters vote reliably for the candidate while persuaded voters are always subject to last-minute reversal.

The secondary component of Jake's targeting strategy was a persuasion play in working-class communities at the outer edge of the metropolitan suburbs — the exurban towns and small cities where the population was older, less college-educated, and more culturally populist than the inner suburbs where Nadia was focused. These voters had Democratic registration histories that predated the partisan realignment of the Obama and Trump years. They had voted Democratic in older elections but split or gone Republican in recent cycles. Jake's bet was that Whitfield's economic populism — his critique of corporate influence on state government, his emphasis on working-family economic concerns — could peel enough of these voters from their residual Democratic affiliation to make a difference.

🔵 Debate: Nadia's focus on college-educated suburban persuasion targets and Jake's focus on non-college working-class mobilization represent genuinely competing theories of how contemporary electoral coalitions are formed and contested. The realignment of American politics along educational lines — college voters moving toward Democrats, non-college voters moving toward Republicans — creates a situation in which both targeting strategies have coherent logic, and the question of which one captures more votes in a particular state in a particular cycle is genuinely uncertain in advance. Neither campaign analytics team has access to the ground truth that would definitively adjudicate between them.

29.6a The Mechanics of Contact: From List to Door

Targeting produces lists. Lists produce contacts. Contacts, ideally, produce votes. The operational chain from analytics to outcome runs through a series of steps that analytics staff often have limited visibility into — and this visibility gap is one of the most important sources of uncertainty in evaluating targeting effectiveness.

List generation. Once the targeting universe is defined — specific voters with specific scores assigned to specific message tracks — the analytics team generates exportable lists for each contact channel. For canvassing, the list is formatted for the VAN walk list generator, which organizes voters by geographic proximity into efficient walking sequences. For mail, the list is formatted for the direct mail vendor's database, with message track assignments attached. For digital, the list is formatted for a custom audience upload to the relevant platforms.

Each of these formatting tasks requires careful attention to data integrity. The same voter should not appear on both the GOTV list and the persuasion list. Voters who have requested no contact should be flagged and excluded. Households where one voter is a target but a coresident is a committed opponent need careful handling (sending a GOTV mail piece to a household where one resident strongly supports the candidate and another strongly opposes means that piece will be seen by both).

Canvassing walk lists. The walk list is the canvasser's primary operational tool — a smartphone app or printed list showing which doors to knock, in what order, with what script. Modern walk lists are dynamic: the VAN system tracks which voters have already been contacted (in previous canvass shifts) and automatically excludes them from subsequent walk lists, avoiding repeated contact with the same voter. Script prompts guide the canvasser through a brief conversation designed to identify the voter's support level and any issue concerns.

The quality of the canvassing interaction — whether the canvasser genuinely engages with the voter versus delivering a scripted pitch and moving on — is enormously variable and largely invisible in the data. A VAN record that shows "contacted, lean support, mentioned healthcare" could represent a five-minute substantive conversation or a fifteen-second doorstep exchange. This measurement problem means that canvass data quality is inherently uncertain in ways that the record count doesn't capture.

The first-contact priority. One operational principle that analytics teams have learned from field experiments is the importance of the first contact. A voter who has been contacted once is different from a voter who has never been contacted — both because the campaign now has better information about their views, and because the social interaction has potentially primed them to pay more attention to subsequent contacts. Targeting programs that are designed around "one and done" contact may be leaving value on the table; research suggests that multiple contacts, particularly when they reinforce a consistent message, are more effective than single contacts.

This has implications for how walk lists are organized. A campaign with limited canvassing capacity might, in a less sophisticated operation, spread its contacts evenly across the persuasion universe, with each voter receiving a single contact. A more sophisticated operation would identify a smaller high-priority persuasion universe and ensure those voters receive multiple contacts across the final weeks — building a cumulative contact program rather than maximizing first-contact breadth.

Channel coordination and suppression. A voter who has already received a canvass visit should probably not also receive three direct mail pieces and six digital ad exposures in the same week. Over-contact creates diminishing returns and can generate the kind of annoyance that reduces rather than increases a voter's willingness to engage. Effective targeting programs include channel suppression logic: once a voter has been contacted through the highest-priority channel (personal canvassing), they are suppressed from the lower-priority channels (mail, phone) or shifted to lower-frequency versions.

In practice, channel coordination is harder than it sounds because different contact programs are often run by different teams — the field program runs the canvassing, the mail vendor runs the direct mail, the digital team runs the advertising — and these teams don't always communicate in real time about who has been contacted through which channel. Building the coordination infrastructure that allows genuine cross-channel suppression is an advanced capability that most campaigns achieve partially at best.

29.7 Third-Party Data: What It Is and Why It Fails

Both campaigns' microtargeting operations rely heavily on third-party consumer data layered onto the voter file. This data — purchased from commercial data brokers like Acxiom, Datalogix, and Experian — purports to capture consumer preferences, household characteristics, lifestyle indicators, and behavioral patterns that are predictive of political preferences. Understanding what this data actually is, and where it fails, is essential for critically evaluating microtargeting claims.

What Consumer Data Covers

Commercial consumer data aggregators collect information from an enormous range of sources: credit card transactions, retail loyalty programs, property records, magazine subscriptions, warranty registrations, internet browsing behavior, mobile app usage, location data, and many other signals. The resulting record for any individual voter may include thousands of data points — purchase categories, estimated income ranges, estimated wealth, property ownership, vehicle ownership, hobbies inferred from purchases, and lifestyle classifications assigned by proprietary algorithms.

This data is then matched to voter file records using email addresses, phone numbers, name-and-address combinations, and probabilistic matching algorithms. The match rates and accuracy of these procedures vary significantly — and this is where the first major problem with third-party consumer data enters.

The Reliability Problem

Campaign analytics practitioners frequently overestimate the accuracy of consumer data, for several interconnected reasons.

Match rate confusion: A 70% match rate between a voter file and a consumer data file sounds reassuring — it means the campaign has consumer data on most of its voter file. But it obscures the accuracy within the matched records. The records that matched perfectly (same email, same address, same name) are more reliable than records that matched probabilistically (similar name, close address, no email match). The aggregate match rate doesn't tell you which individual records are reliable.

Recency decay: Consumer data ages rapidly. A lifestyle classification built on purchase behavior from two years ago may not reflect current preferences, particularly for younger voters who move frequently, change jobs, and whose consumer behavior shifts with life circumstances. A voter who was classified as a suburban homeowner based on a 2021 property record may have sold the house and moved into an apartment.

Prediction versus description: Consumer data can describe past behavior with some accuracy. Its ability to predict future political behavior is substantially weaker, because the link between consumer behavior and political preference, while real, is not deterministic. Aggregate correlations between, say, truck ownership and Republican support are real but noisy at the individual level. Many truck owners vote Democratic. Many Prius owners vote Republican.

⚠️ Common Pitfall: Campaigns sometimes treat consumer data-based predictions as more reliable than their own canvass data from direct voter contact. A canvasser who actually talked to a voter, recorded that the voter expressed concerns about healthcare and was genuinely undecided, is providing a much higher-quality signal than a consumer data algorithm that inferred issue affinity from purchase behavior. Fresh canvass data should be given substantially more weight than stale consumer data in targeting decisions.

The Third-Party Data Ecosystem and Its Problems

The commercial data ecosystem has additional problems beyond accuracy. Data brokers have faced increasing regulatory scrutiny in recent years — California's CCPA and similar state laws have created new restrictions on how consumer data can be used and shared. Campaigns operating in states with strong consumer privacy laws need to track their compliance obligations carefully, as the use of third-party consumer data for political targeting is an active regulatory target.

The accuracy of commercially provided political scores is also contested. Several academic studies have examined the predictive validity of commercial political interest and issue affinity scores — testing whether voters categorized as high-interest in a specific issue actually report that issue as a priority in surveys — and found mixed results. The correlations exist but are often weaker than the scores' precision implies.

29.8 The Ethics of Microtargeting

The efficiency gains from microtargeting are real. Campaigns that target well spend less money reaching voters who would have voted for them anyway, and more time and money reaching voters who might be persuaded. From a campaign resource management perspective, microtargeting is unambiguously good practice.

But microtargeting's ethical dimensions are more complicated, and they deserve careful attention — both because the ethical questions are important in their own right and because campaigns that ignore them run genuine political risks.

Manipulation and Autonomy

The most fundamental concern about microtargeting is that it is designed to be persuasive in ways that bypass voters' rational deliberation. A broadcast political ad makes a public argument that any voter can evaluate — its claims can be fact-checked, its logic can be assessed, its evidence can be challenged. A microtargeted appeal delivered only to a specific voter identified as susceptible to a specific emotional argument operates differently: it is calibrated to exploit a specific psychological or affective vulnerability identified through data analysis, and the voter has no obvious mechanism for recognizing that the appeal is specifically targeted at them.

The political philosophy literature on manipulation versus persuasion is relevant here. Persuasion, in the liberal tradition, works by offering evidence and arguments that a rational agent can evaluate and accept or reject. Manipulation works by bypassing rational evaluation — exploiting emotional vulnerabilities, creating false urgency, or framing choices in ways that distort the decision environment. Microtargeting, at its worst, can cross from persuasion into manipulation when it is calibrated to exploit specific psychological vulnerabilities identified through data analysis rather than to offer genuine arguments.

⚖️ Ethical Analysis: The line between targeting and manipulation is not always clear. A campaign that identifies voters who are primarily motivated by economic anxiety and sends them information about the candidate's economic proposals is targeting. A campaign that identifies voters who are susceptible to fear-based appeals and sends them deliberately misleading economic threat messages is manipulating. The same data infrastructure enables both. The ethical difference lies in the intent and the content of the message — but the campaign doing the targeting is not a neutral evaluator of that distinction.

Voter Suppression and Dark Pattern Messaging

A particularly serious concern is the potential use of microtargeting for voter suppression — using detailed voter profiles to identify and target opposing voters with messaging designed to discourage their participation rather than to change their vote.

Voter suppression through messaging can take several forms: spreading false information about voting procedures (wrong dates, wrong polling places, incorrect ID requirements) targeted at specific voter populations; messaging designed to demobilize opposing voters by convincing them that their preferred candidate will win easily and their vote isn't needed; or fear-based messaging targeted at immigrant-adjacent communities that raises the salience of ICE enforcement to discourage participation.

All of these tactics use the same data infrastructure as legitimate targeting. The distinction is in the intent and content — which, again, the campaign controls and the voter cannot easily evaluate.

📊 Real-World Application: The Facebook advertising ecosystem's role in voter suppression became a significant political controversy during the 2016 and 2020 election cycles, as investigations revealed that both foreign interference operations and domestic campaigns had used Facebook's detailed targeting capabilities to deliver demobilizing messages to specific voter populations. Facebook's subsequent policy changes — restricting political ad targeting options, requiring disclaimers, and publishing a public ad library — were direct responses to these concerns, though critics argue they are insufficient.

Who Gets Targeted, Who Gets Ignored

Microtargeting creates an electorate that is, effectively, divided into two groups: voters who receive extensive campaign communication, and voters who receive essentially nothing. The targeting logic that makes this efficient from a campaign perspective has distributional consequences that the campaigns' resource optimization calculus doesn't account for.

The voters who fall outside the targeting universe — low-support voters, very low-turnout voters, committed partisans on the wrong side — receive less political information, less mobilization pressure, and arguably less democratic attention as a result of their campaigns' data-driven decisions. These neglected voters are not randomly distributed across the population. They are systematically concentrated in communities where past electoral behavior — lower turnout, weaker partisan registration — has made them lower-priority targets.

The communities most systematically excluded from campaign contact tend to be those with already lower political resources: communities of color with historically suppressed turnout, economically marginalized areas with lower civic engagement, and communities where language barriers have historically limited voter registration. The efficiency of microtargeting, in other words, can amplify existing inequalities in political voice.

🔴 Critical Thinking: Campaigns can reasonably respond that their job is to win the election for their candidate, not to maximize democratic participation across the electorate. The resource allocation logic that produces uneven contact is not malicious — it's rational within the campaign's objective function. But the aggregate effect across all campaigns simultaneously is a democratic system in which political communication is heavily concentrated on a relatively small, relatively predictable universe of mobilizable and persuadable voters, while large segments of the citizenry receive negligible electoral attention. Whether this is an acceptable consequence of rational campaign behavior or a structural problem requiring political solutions is a question we return to in Chapters 38 and 39.

Transparency and Disclosure

One important ethical marker is the degree to which campaigns disclose their targeting practices. Broadcast political advertising is publicly regulated — ads must identify their source and, in many cases, comply with disclosure requirements. Digital targeting operates in a substantially less regulated environment. A campaign can deliver thousands of distinct microtargeted messages to thousands of distinct voter segments, and no external observer — journalist, researcher, regulatory body, or opposing campaign — has a comprehensive view of what is being said to whom.

The disclosure gap between broadcast and digital political communication is a genuine democratic accountability problem. When different voters receive different messages, the public conversation about campaigns and their claims is fragmented. Fact-checkers can only fact-check messages they can see. Opposing campaigns can only rebut arguments they're aware of. The public can only evaluate campaigns on what is publicly visible — and microtargeting makes increasingly little of campaign communication publicly visible.

29.8a The Targeting-Suppression Distinction in Legal and Ethical Terms

The previous section introduced voter suppression as an ethical concern. It is worth examining the legal and institutional landscape around this concern more carefully, both because the law shapes what campaigns can and cannot do, and because the ethical analysis should be grounded in an accurate picture of existing legal protections.

The Voting Rights Act and its limitations. Section 2 of the Voting Rights Act prohibits voting practices that discriminate on the basis of race or color. Section 11 prohibits intimidating, threatening, or coercing anyone for voting or attempting to vote. These provisions create legal prohibitions on some of the most obvious voter suppression tactics — false information about polling place locations targeted at specific racial communities, for example. However, the VRA's application to digital microtargeting is largely untested. The statute was not written with data-driven targeting in mind, and courts have not yet extensively addressed whether targeted digital demobilization messaging violates Section 11 or Section 2 when conducted at scale.

FEC jurisdiction and its gaps. The Federal Election Commission has jurisdiction over campaign finance and related disclosure, but its authority does not extend to the content of campaign communications in most cases. A campaign can make false claims about its opponent's positions without FEC liability (state defamation law may provide some remedy, but it is rarely invoked in campaign contexts). The FEC's disclosure rules for digital advertising, extended in recent years, require disclosure of who paid for an advertisement but do not require disclosure of targeting criteria.

Platform policies as governance. In the absence of comprehensive statutory regulation, Facebook's policies, Google's policies, and similar platform rules function as the primary governance layer for digital political targeting. These policies are privately made, inconsistently enforced, and subject to change without democratic input. Their geographic inconsistency — US platforms apply different political advertising rules in different countries, typically stricter ones in Europe — reflects the absence of a coherent regulatory framework rather than a principled policy architecture.

The practical surveillance gap. Campaign compliance infrastructure — lawyers, accountants, FEC filings — is reasonably well developed for campaign finance. It is essentially non-existent for content targeting practices. No mandatory audit process exists to verify that a campaign's digital targeting complied with its stated targeting criteria. No mandatory disclosure requires campaigns to reveal which voter segments received which messages. The gap between what is regulated and what is operationally possible is wide enough to permit a substantial range of practices that would be objectionable if visible but that are currently invisible to any enforcement mechanism.

⚖️ Ethical Analysis: The absence of legal prohibition is not the same as ethical permissibility. Campaigns that operate at the edge of what is legally permitted — running technically legal but substantively manipulative targeting programs — may be complying with the law while violating the norms of fair political competition. The question of who enforces those norms is genuinely difficult in the current regulatory environment. Journalists can identify egregious cases. Researchers can document patterns. But individual voters who are the targets of manipulative messaging have no practical mechanism for recourse.

29.9 The Evidence on Targeting Effectiveness

All of this sophisticated infrastructure raises a basic question: does targeting actually work? The answer from the research literature is "yes, but with important caveats."

GOTV Targeting Evidence

The evidence that GOTV targeting improves campaign efficiency is strong. Field experiments (Chapter 30 covers these in depth) consistently show that personal canvassing, phone banking, and direct mail increase turnout among contacted voters. Campaigns that direct these resources toward voters who would support the candidate if they voted — rather than random registered voters — see better net gains from the same level of outreach.

The key experimental finding is not just that contact helps but that targeting efficiency matters. Studies comparing targeted GOTV programs (directed at likely supporters with lower turnout propensity) to untargeted programs (contacting random registered voters regardless of support or turnout propensity) find meaningfully higher net vote gains from the targeted approach. Targeting isn't magic — the contacts themselves do the work — but targeting ensures that the contacts are delivered where they'll help the right candidate.

Persuasion Targeting Evidence

The evidence on persuasion targeting is more mixed. Several field experiments have found that issue-specific mail tailored to voters' inferred issue priorities performs better than generic political mail — but the effect sizes are small, and many experiments find no significant difference between targeted and untargeted persuasion contact.

Part of the reason may be the reliability problems in persuadability scoring discussed earlier. If the campaign can't accurately identify which voters are genuinely persuadable, the efficiency gains from persuasion targeting shrink substantially. Contact delivered to voters who are already committed — but who scored in the persuasion range due to model noise — produces no persuasion effect and wastes resources.

Digital Microtargeting Evidence

The evidence on digital microtargeting effectiveness is perhaps the most contested area of current campaign research. Several industry studies and platform-published analyses claim large effects for targeted political advertising. Several independent academic studies have found much smaller effects — and some have found effects indistinguishable from zero.

The challenge is that digital platforms' targeting algorithms are proprietary, making rigorous independent study difficult. The studies that find the largest effects are often conducted by the platforms or vendors who benefit from campaigns believing in those effects. The studies that find the smallest effects are typically academic studies that face restrictions on data access.

📊 Real-World Application: A landmark 2020 study by researchers at Stanford and several other universities (published as a large pre-registered experiment on Facebook political advertising) found that exposure to a political campaign's Facebook ads had very small effects on candidate favorability and vote intention — smaller than most campaigns' assumptions about digital advertising effectiveness would predict. The findings provoked substantial controversy and ongoing methodological debate.

The Cumulative Effect Question

Even if individual targeting touchpoints have small effects, campaigns make many contacts. The cumulative effect of receiving issue-targeted mail, two microtargeted Facebook ads, and a personal canvass visit over six weeks may be substantially larger than the effect of any single contact, particularly if the contacts are coordinated to reinforce a consistent message.

The challenge is measuring cumulative effects, because voters receive overlapping contact from multiple sources and it is difficult to isolate the incremental contribution of each. This is a frontier area of campaign research — and it suggests that evaluating targeting effectiveness at the level of individual contacts may systematically understate the aggregate impact of coordinated targeting across channels.

29.9a Cost-per-Vote Calculations: The Operational Logic of Channel Selection

Targeting strategy is not just about identifying the right voters and the right messages — it is also about allocating limited resources across channels in a way that maximizes net votes per dollar spent. The cost-per-vote framework translates experimental effect estimates and contact costs into a common metric that allows direct comparison across strategies.

The basic cost-per-vote calculation. For any contact mode, the cost per vote generated can be estimated as:

Cost per vote = (Cost per contact) / (Effect per contact)

For personal canvassing with a cost of $12 per completed contact and an effect of 2.5 percentage points per contacted voter:

Cost per vote = $12 / 0.025 = $480 per additional vote generated.

For direct mail with a cost of $0.65 per piece delivered and an effect of 0.5 percentage points per piece:

Cost per vote = $0.65 / 0.005 = $130 per additional vote generated.

For phone banking with a cost of $3 per completed call and an effect of 0.8 percentage points per call:

Cost per vote = $3 / 0.008 = $375 per additional vote generated.

On a pure cost-per-vote basis in this stylized example, direct mail appears to dominate — $130 per vote versus $375–$480 for more expensive contact modes. But the calculation is incomplete, because the effect size estimates apply to the marginal voter being contacted — and the marginal voter is different for each channel.

The quality-of-contact adjustment. Personal canvassing is substantially more expensive than mail, but it is also substantially more effective per contact at moving genuinely persuadable voters. A voter who is firmly in the campaign's support universe — who would vote for the candidate if they vote at all, and who has a 72% probability of voting — will respond similarly to a canvass visit and a piece of mail. For that voter, mail is clearly more cost-efficient. But a voter with a support score of 52 and genuine ambivalence about the candidate — a voter who might actually be persuaded in a substantive conversation — is where canvassing's persuasive superiority over mail is most likely to manifest.

The implication is that the cost-per-vote comparison should not be made at the aggregate level but at the margin: for the last dollar spent on this program, what is the best channel for the specific voter being reached? High-propensity supporters who are almost certain to vote for the candidate if they turn out should be reached by the cheapest effective GOTV mode. Genuinely persuadable voters who could go either way might be worth the higher cost of personal contact.

The cost structure of voter contact programs. Building and running a canvassing program has substantial fixed costs — recruiting and training canvassers, setting up the field data infrastructure, maintaining staff to manage the program — in addition to the variable costs of each contact. The cost-per-vote calculation based on variable costs alone understates the true cost of canvassing relative to mail for small programs. As the program scales, the fixed costs are amortized over more contacts, making the true cost per contact converge toward the variable cost.

This is one reason large campaigns have a cost advantage in canvassing that small campaigns lack. A campaign with 200,000 canvass contacts amortizes its fixed canvassing infrastructure costs over a much larger base than a campaign with 10,000 canvass contacts, making its effective cost per contact significantly lower.

The net vote versus gross vote distinction. The cost-per-vote calculation above counts all votes generated by the contact program. But campaigns also care about net votes — votes that improve the margin. A GOTV contact that mobilizes a previously uncertain supporter generates one net vote gain (the candidate gets +1). A persuasion contact that moves a voter from opposing to supporting the candidate generates two net vote gains (the candidate gets +1, the opponent gets -1). This asymmetry means persuasion contacts are worth approximately twice as much as GOTV contacts per voter moved, all else equal — a fact that sometimes gets lost in aggregate cost-per-vote comparisons that treat mobilized and persuaded voters identically.

In practice, persuasion is much harder to achieve than mobilization — the effect sizes per contact are smaller, the target population is harder to identify, and the conversion rate is lower. Most cost-per-vote analyses find that GOTV targeting is more cost-efficient than persuasion targeting in most electoral contexts, because the reliability of mobilizing genuine supporters outweighs the per-unit value of persuading the much smaller universe of genuinely movable voters.

29.10 The Asymmetry of Targeting Across Campaigns

Not all campaigns can target with equal sophistication, and the resource asymmetry in targeting capabilities has political consequences.

Statewide campaigns with substantial budgets — like Garza or Whitfield — can afford Catalist or i360 subscriptions, voter file enrichment, modeled scores, and a professional analytics team. County-level campaigns, state legislative campaigns, and local races typically cannot. These smaller campaigns often operate with a basic voter file and no enrichment, relying on experienced organizers' instincts rather than predictive models.

This creates an asymmetry in targeting precision between resource-rich and resource-constrained campaigns. The wealthy candidate, or the candidate with access to superior party infrastructure, targets more efficiently. The underfunded challenger targets less efficiently — spending more of her scarce contact resources on voters who aren't going to respond, reaching fewer of the voters who would have made the difference.

Whether this asymmetry is a meaningful competitive disadvantage depends partly on context. In many local races, the vote targets are small enough that unsophisticated but enthusiastic ground game operations can be competitive. In highly competitive statewide and national races, the efficiency of targeting may be increasingly decisive at the margins.

🌍 Global Perspective: The ethics and effectiveness of voter microtargeting are contested across democratic systems, not just in the United States. The United Kingdom's experience with Cambridge Analytica — whose claims to have used Facebook data for psychographic microtargeting in the Brexit referendum were widely reported, and then significantly questioned by subsequent research — illustrates both the public concern about microtargeting and the difficulty of evaluating its actual effects. European data protection regulations under GDPR are substantially more restrictive than US regulation about the commercial data that can be used for political targeting, creating a very different microtargeting environment in most EU member states.

29.11 The Future of Targeting

Several technological and regulatory developments are likely to reshape voter targeting in the coming cycles.

Platform policy changes: Facebook's restrictions on political ad targeting, Google's removal of political advertising targeting options, and Twitter's eventual ban on political advertising all represent significant changes to the digital targeting environment. If these restrictions persist and expand, campaigns will lose access to the identity-based digital targeting capabilities that have been central to microtargeting practice since 2012.

Privacy regulation: State-level consumer privacy laws that restrict how commercial data can be collected and used for political purposes will gradually constrain the consumer data ecosystem that third-party enrichment depends on.

Artificial intelligence: AI-generated content and targeting recommendations are beginning to enter campaign operations. The ability to generate large numbers of individually customized messages at low cost — rather than a few standard tracks with targeted distribution — is becoming technically feasible. The regulatory and ethical implications are actively contested.

Research and transparency: Growing academic and journalistic attention to targeting practices — including projects that build archives of political digital advertising — is gradually increasing the transparency of practices that have been largely opaque.

29.12 Conclusion: The Targeting Paradox

Voter targeting is, at its best, a form of political respect — the campaign has done the work to understand what this specific voter cares about, and it is bringing a message that responds to those concerns. At its worst, it is a form of manipulation — the campaign has used data analysis to identify the specific psychological levers that will move this specific voter, and it is pulling those levers in the dark, confident that no one outside the targeted conversation will see the message or hold the campaign accountable for it.

The targeting paradox is that the same infrastructure enables both outcomes. A campaign with a sophisticated voter file model, a well-validated persuasion universe, and a carefully designed message testing program can run a targeting operation that genuinely improves democratic communication — delivering relevant, substantive political information to voters who would otherwise receive none. The same infrastructure can run a targeting operation that fragments the political conversation, delivers different claims to different audiences, and uses personal data to manipulate rather than persuade.

Nadia Osei's operation for the Garza campaign and Jake Rourke's counter-operation for Whitfield are running versions of targeting that are, by the standards of the industry, relatively responsible — focused on genuine mobilization and substantive persuasion, not on voter suppression or dark pattern messaging. But the infrastructure they're using could be turned to darker purposes by campaigns less constrained by the norms of responsible practice. The ethical limits of targeting are not built into the tools; they are enforced, incompletely, by professional norms, legal requirements, and the (imperfect) scrutiny of journalists and researchers.

That is why understanding targeting — technically, analytically, ethically — is essential for any citizen of a democracy in which campaigns are increasingly sophisticated in their use of data. The voters who know how targeting works are at least partially equipped to question why they're receiving the messages they receive, to notice when messages seem designed to exploit rather than inform, and to demand the transparency that democratic accountability requires. That epistemic capacity — understanding the systems that are trying to persuade you — may be one of the most important civic competencies of the data-driven political era.

Key Terms

Targeting — The systematic allocation of campaign contact resources toward specific voters based on their estimated support, turnout propensity, and persuadability.

Microtargeting — The practice of customizing political messages and contact channels to the specific characteristics of individual voters, using voter file data, consumer data, and predictive modeling.

GOTV universe — The group of voters a campaign prioritizes for get-out-the-vote outreach: expected supporters with below-certain turnout propensity.

Persuasion universe — The group of genuinely undecided voters a campaign targets with persuasion contact.

Issue affinity model — A predictive model that estimates which political issues are most salient to a given voter, used to personalize campaign messaging.

Custom Audiences — Facebook's advertising feature that allows campaigns to target specific individuals by matching their email addresses or phone numbers to Facebook accounts.

Identity-based targeting — Digital advertising directed at specific, identified individuals (via voter file matching), as opposed to demographic or interest-based targeting.

Persuadability score — A modeled estimate of the probability that a voter's vote choice can be changed by campaign contact.

Third-party consumer data — Data about individuals' purchasing behavior, lifestyle, and demographic characteristics purchased from commercial data brokers and used to enrich voter file records.

Voter suppression messaging — Campaign or third-party messaging designed to discourage opposing voters from participating in elections.