“For decades, campaigns had been driven by the instincts of consultants rather than the findings of scientists. The new generation of strategists insisted that no one actually knew what worked — and that the only honest answer was to run the experiment and find out.”
— Sasha Issenberg, The Victory Lab: The Secret Science of Winning CampaignsExecutive Summary
For most of the twentieth century, political campaigns ran on instinct. Senior consultants made tactical decisions based on intuition, tradition, and anecdote. Nobody systematically tested whether the mailers were working, whether the canvassers were actually moving voters, or whether the phone calls were changing any minds. Resources were allocated on the basis of what had always been done — not on the basis of what the evidence showed was effective.
Sasha Issenberg’s The Victory Lab documented how that changed. Beginning in the late 1990s and accelerating through the 2000s, a small group of political scientists — led by Yale’s Donald Green and Alan Gerber — began running randomized field experiments on voter contact tactics. They treated campaigns the way medical researchers treat drug trials: assign voters randomly to treatment and control groups, apply a specific tactic to the treatment group, and measure the actual effect on turnout. The results were often surprising, frequently counterintuitive, and always more reliable than consultant intuition.
That scientific revolution happened first in presidential and Senate campaigns with resources to fund it. But the underlying findings apply to every race — and the data infrastructure to act on those findings is now available at price points that work for county commissioners, school board candidates, and township trustees. This paper explains what the science found, why it matters at the local level, and how small R/I campaigns can build a data-informed operation without a seven-figure budget.
The Randomized-Experiment Revolution
The key methodological insight of the Issenberg era was simple but radical: the only way to know whether a campaign tactic works is to run a controlled experiment. Prior to the field experiment movement, campaigns evaluated tactics by asking whether they felt like they were working, whether veteran consultants endorsed them, or whether the candidate won — which tells you almost nothing about the tactic’s contribution to the outcome.
Green and Gerber’s foundational 2000 experiment on door-to-door canvassing in New Haven, Connecticut established the template. They randomly assigned registered voters to receive canvassing, phone calls, or direct mail, and then compared their actual turnout rates against the control group. The findings upended conventional wisdom: personal canvassing generated substantial turnout lift. Generic phone banking showed small effects. Direct mail showed almost none. And the effects were measurable, replicable, and specific enough to drive resource allocation decisions.
What followed was a decade of experimentation across dozens of campaigns, jurisdictions, and demographic groups. By the time The Victory Lab appeared in 2012, there were hundreds of field experiments in the literature. The findings had been replicated enough times across different contexts that they were no longer academic curiosities. They were operational guidance — for the campaigns that knew about them.
Most down-ballot campaigns did not know about them. And many still don’t. That is the gap this paper addresses.
What the Science Actually Says: Four Core Findings
The field experiment literature is extensive, and the details matter. But four core findings are robust enough, and practically important enough, to ground every down-ballot campaign’s tactical planning.
Finding 1: Personal contact is dramatically more effective than impersonal contact
This is the most consistent and most replicated finding in the field experiment literature. A real human being, at someone’s door, asking them to vote in a specific election, generates measurable turnout lift. The same message delivered by robocall, generic mail, or broadcast advertising generates little to none.
The mechanism is social obligation: a personal interaction creates a commitment that a passive exposure to campaign material does not. Voters who are asked face-to-face whether they plan to vote are more likely to actually vote than voters who receive no contact or impersonal contact. The physical presence of the canvasser activates social norms around civic participation that a postcard simply cannot replicate.
For a down-ballot candidate, this finding has a direct corollary: the most valuable campaign resource is not the media buy. It is the door-knock. Every hour the candidate spends in personal contact with identified supporters is worth more, per dollar of equivalent cost, than almost any other tactic.
Finding 2: Social norms messages move behavior in ways generic messages do not
Among the most surprising and practically important findings in the experimental literature is the power of social comparison information. When voters receive a message telling them that a high percentage of their neighbors voted in the last election — rather than generic encouragement to participate — they turn out at significantly higher rates.
Alan Gerber, Donald Green, and Christopher Larimer’s landmark 2008 experiment sent Michigan voters four different mailers: a civic duty message, a message stating that researchers would be watching their turnout, a message showing their own voting history, and a message showing both their voting history and their neighbors’ voting history. The last condition — social comparison with neighbors — generated the largest turnout effect ever measured in a direct mail study.
The implication for local campaigns is significant: the content of your voter contact matters, not just the method. “Your neighbors are voting — will you?” backed by a true local statistic outperforms “your vote matters” by a measurable margin.
Finding 3: Targeting precision determines whether any tactic pays off
Field experiments consistently show that the same tactic produces dramatically different effects depending on which voters it is applied to. A canvass of low-propensity supporters produces large turnout lifts. The same canvass applied to reliable high-turnout voters produces negligible additional lift — they were going to vote anyway.
This is the finding that transformed the analytics revolution from an academic exercise into a practical competitive advantage. If you can identify which voters in your universe are most likely to support you but least likely to show up without contact, and then concentrate your highest-quality tactics on exactly those voters, you generate far more votes per contact dollar than a campaign that applies the same tactics uniformly across the full voter file.
Universe precision is not a luxury for campaigns with data teams. It is the basic mechanism that makes evidence-based tactics efficient enough to work at local scale.
Finding 4: Most campaigns waste significant resources on tactics that do not work
Robocalls, generic direct mail to undifferentiated lists, and broadcast advertising aimed at no specific voter segment all show small-to-negligible effects in controlled experiments. Yet most local campaigns allocate significant portions of their budgets to exactly these tactics — because they have always done so, because consultants endorse them, or because they feel like they should be reaching “everyone.”
The Victory Lab’s most practically valuable contribution was not identifying what works. It was quantifying what does not work — and giving campaigns a principled basis for reallocating those wasted resources toward the tactics that actually move voters.
Testing What Works at the Local Level
The original Victory Lab experiments were run at scale: thousands of voters across large jurisdictions, with statistical power to detect effects that a single local race cannot match. A county commissioner race with 8,000 voters cannot run a randomized controlled trial with textbook rigor and expect to publish the findings in a peer-reviewed journal.
But that does not mean testing is unavailable to local campaigns. It means testing looks different at local scale.
A/B message testing within the universe
Even a small campaign can test two different message framings by randomly dividing its canvass universe or phone banking list and assigning each half to a different script. If one precinct is contacted with a message emphasizing local road conditions and another with a message emphasizing fiscal responsibility, the campaign can observe which version generates more follow-up enthusiasm, more volunteer sign-ups, or more donation responses as a rough signal of message resonance.
This is not a gold-standard experiment. But it is vastly better than no information at all, and it costs nothing beyond the planning it takes to implement it. Campaigns that systematically test and adjust outperform campaigns that run the same message to every voter from announcement day through election eve.
Tracking and feedback from canvass data
The most practical form of real-time testing for a local campaign is systematic tracking of canvass outcomes. When canvassers record voter responses at each door — strong support, soft support, undecided, soft opposition, hard opposition — the campaign accumulates data on which precincts, which voter segments, and which messages are generating the most favorable responses. That data drives the next shift’s walk list prioritization.
Campaigns that debrief canvassers, log responses, and adjust their universe daily are doing an informal version of what the Victory Lab researchers did formally. They are treating their voter contact program as a source of real-time feedback, not just a turnout-generating machine.
Data Finds, Targeting Delivers, GOTV Turns Out
The three functions of a data-informed local campaign are distinct and sequential. Conflating them produces waste. Understanding their relationship produces wins.
Data finds the right voters. The voter file, layered with modeling and public data sources, identifies which registered voters in the jurisdiction are most likely to support the candidate and least likely to show up without contact. This is the universe-building stage. The output is a tiered, scored list of voters prioritized by their value to the campaign — not a flat list of every registered Republican in the county.
Targeting delivers the right message to those voters at the right moment. Digital targeting allows campaigns to serve ads to matched voter file audiences on Facebook and Google — so the persuasion and awareness message reaches exactly the voters the campaign has already identified as high priority, not a general audience that includes unregistered voters, out-of-district residents, and people who will never be in the universe regardless of what message they see.
GOTV turns out the voters the first two functions found and primed. The field experiment evidence on what moves voters to the polls — personal canvassing, genuine phone conversations, social norms messaging — is the execution layer. It only pays off when the universe it is applied to is correctly built. Without the data and targeting functions, GOTV tactics are applied to the wrong voters at high cost and low return.
These three functions correspond to three distinct phases of a local campaign, and to three distinct tools. Treating them as one undifferentiated activity — “voter outreach” — is the mistake that turns a potentially efficient campaign into an expensive one.
Making the Science Affordable for Small Races
The Victory Lab documented the analytics revolution as it unfolded in campaigns with millions of dollars and dedicated data science teams. The Yale researchers, the early political analytics firms, and the campaign operatives who first adopted field experiment methodology were operating at a scale that most down-ballot campaigns cannot approach.
That scale gap has closed substantially over the past decade. The following capabilities, which were either unavailable or prohibitively expensive a decade ago, are now accessible to small local campaigns at costs that fit within a county-level budget.
Voter file access and modeling
State voter files are public records in every state. The voter file for a county commission race — including registration status, party affiliation (where applicable), and voting history going back multiple cycles — typically costs under $50 and is available within days of requesting it. That data, combined with publicly available demographic information, is the raw material for support scoring and universe construction.
Modeling those raw files into scored, tiered universes requires some analytical capacity — but nothing approaching the data science teams that major campaigns employ. The basic logic of universe segmentation (who supports us, who votes consistently, who is a low-propensity supporter worth contacting) can be implemented in a spreadsheet by a campaign manager with basic data literacy.
Digital audience matching
Meta’s Custom Audiences tool allows campaigns to upload a hashed voter file and serve ads to matched users on Facebook and Instagram. Google Customer Match provides the same capability for Google Search, YouTube, and Display. What was previously available only to campaigns with sophisticated data vendors is now accessible through the same ad platforms local campaigns already use for general awareness advertising.
The practical implication: a local campaign can build a voter universe, identify the 2,000 highest-priority persuasion targets, push that list to Facebook, and serve those 2,000 specific registered voters a message tailored to their profile — for a few hundred dollars in ad spend. That is a fundamentally different tactic from buying reach across the full county population.
Canvass management tools
Walk list optimization and canvass management software — once reserved for campaigns with paid field directors and technology budgets — is now available through mobile apps that cost less than a mail piece per month. Canvassers can receive and return data in real time, universe records update as contacts are logged, and the campaign manager can see coverage progress by precinct without maintaining a separate spreadsheet.
The total infrastructure cost: A local campaign can build a data-informed, field-experiment-grounded voter contact program — voter file acquisition, basic universe modeling, digital audience matching, canvass management — for a fraction of what it would spend on a single mail piece to the full county voter file. The question is not whether the science is affordable. The question is whether the campaign has someone who knows it exists.
The Voter File as Testable Hypothesis
The deepest practical lesson from the field experiment revolution is a change in how campaigns should think about the voter file. In the pre-analytics era, the voter list was a contact list: a set of names and addresses to which the campaign would mail, call, or knock doors in sequence until Election Day. It was a static input to the campaign operation.
In the analytics era, the voter file is a testable hypothesis. Every voter record contains a prediction about that voter’s behavior — their likely support for the candidate, their likely turnout propensity, their responsiveness to specific message types. The canvass data, phone banking results, and early vote tracking that accumulate through the campaign are evidence that confirms, revises, or overturns those predictions.
A campaign that treats the voter file as a testable hypothesis adjusts its universe as data comes in. Precincts that are delivering fewer strong-support responses than predicted get deprioritized in favor of precincts that are overperforming. Message framings that are generating confusion or pushback get revised. Volunteers are shifted to the contacts that the evidence shows are worth making.
This is a posture of disciplined empiricism — the same posture that the field experiment researchers brought to campaign tactics. It does not require a PhD in political science. It requires a campaign manager willing to ask “what does the evidence tell us?” every week, and willing to change the plan when the answer demands it.
What This Means for R/I Down-Ballot Candidates
The analytics revolution documented in The Victory Lab is not a set of exotic techniques reserved for presidential campaigns. It is a set of principles — test before assuming, target before broadcasting, measure what matters — that apply at every level of the ballot. Several practical implications follow for Republican and independent candidates running county, township, and school board races.
Build the universe before building the walk list. The first question is not “how many doors do we knock?” It is “which doors are worth knocking?” Start with the voter file, identify the high-probability supporters with inconsistent turnout history, and build the walk list from that universe — not from a map of precincts or a list of every registered voter.
Treat personal contact as the scarcest and most valuable resource. The candidate’s time at the door is the highest-ROI tactic available to the campaign. The field experiment findings on the turnout effect of candidate personal canvassing are among the strongest in the literature. Do not dilute that resource by scheduling events that reach low-priority audiences when the Tier A universe has not been fully contacted.
Use social norms messaging in direct mail. If the budget includes a mail program, at least one piece should include a localized social comparison: the turnout rate among registered voters of the relevant profile in the last comparable election. It is one of the highest-performing messages in the experimental literature, it costs nothing extra to include, and most local campaigns never use it.
Collect canvass data systematically. Every door interaction is a data point. Campaign infrastructure should make it easy for canvassers to log contact quality and voter response at each door, and that data should be reviewed weekly to adjust the universe and walk list priorities. The campaign that learns from its contacts outperforms the one that just counts them.
Do not let consultant intuition override the evidence. The central lesson of The Victory Lab is that experienced political operatives were often confidently wrong about what worked. When a vendor or advisor recommends a tactic that the field experiment literature shows is low-ROI, that recommendation deserves skepticism. The burden of proof belongs to the tactic, not the skeptic.
Conclusion
Sasha Issenberg’s The Victory Lab told the origin story of the analytics revolution in American campaigns. That story began in presidential campaigns and Senate races — the only contexts where millions of dollars and years of research investment could be justified for a single race. But the findings it documented belong to every level of the ballot.
The science of what moves voters — personal contact, social norms messaging, precise universe targeting, real-time feedback from field data — is not proprietary to campaigns with data science teams. It is a set of validated findings available to any campaign willing to act on them. And the technology to implement those findings at local scale has become accessible enough that a county commissioner candidate with a modest budget and a data-informed campaign plan can now run the kind of evidence-based operation that was unimaginable at that level a decade ago.
The campaigns that will win competitive down-ballot races in the next decade are not the ones with the biggest budgets or the most television exposure. They are the ones that find the right voters, deliver the right message, and turn those voters out with personal contact concentrated where the evidence says it pays off. That is the Victory Lab lesson — applied down-ballot, where it matters most.