Evidence for campaign tactics
The designs that can carry a claim that a tactic worked, the kinds of account that cannot, and how the published record is tilted.
fter a campaign says a tactic worked, how would you tell whether it actually did anything? What separates a claim that can be checked from a story about a win?
Geddes (1990)
A political-science paper showing that studying only the cases that came out the way you wanted changes the conclusions you reach.
Unsupported block: pub.terrain.quote
Franco, Malhotra & Simonovits (2014)
A study of 221 social-science experiments where the authors could see both the published and the unpublished ones, measuring how much of the record is missing.
Unsupported block: pub.terrain.quote
Unsupported block: pub.terrain.quote
Kalla & Broockman (2018)
A meta-analysis of 40 field experiments, plus nine new canvassing experiments the authors ran themselves, on whether campaign contact changes a general-election vote.
Unsupported block: pub.terrain.quote
Gerber & Green (2000)
The founding get-out-the-vote experiment: about 30,000 registered voters in New Haven in 1998, contacted by canvassers, mail and phone.
Unsupported block: pub.terrain.quote
Mann & Haenschen (2024)
A re-analysis of 240 field experiments on getting people to the polls, sorted by how much attention the election was getting.
Unsupported block: pub.terrain.quote
Unsupported block: pub.terrain.quote
Gelman, Goel, Rivers & Rothschild
A 345,858-person panel followed through the 2012 US campaign, asking whether the poll swings were real changes of mind.
Unsupported block: pub.terrain.quote
Belguise (2025)
French election results from 2001 to 2024, comparing far-right candidates who just barely came first in round one with those who just barely came second.
Unsupported block: pub.terrain.quote
Falcó-Gimeno, Jurado & Wagner (2026)
A study of Spanish towns where the left or the right won the mayoralty by a handful of votes, comparing far-right support afterwards.
Unsupported block: pub.terrain.quote
Omont (2026)
An attempt to model French two-round vote transfers from official results, and to re-run the 2024 seat allocation under different assumptions.
Unsupported block: pub.terrain.quote
Graham-Squire & McCune (2023)
182 American ranked-choice elections, re-counted round by round from the published ballot files.
Unsupported block: pub.terrain.quote
Synthesis
A claim that a tactic worked needs a comparison the campaign did not choose. Without one, the winning campaigns supply the evidence and the losing ones supply nothing.
The number in the paper is often not the number the campaign can use: 8.7 points on the contacted is a different quantity from the effect on a phone list.
What a tactic is credited with moving — the polls, the mood — is usually the sample changing rather than the voters.
Where the count is published round by round, the counterfactual is arithmetic: re-run the rounds and watch the votes flow. Victoria's preferential count is not published that way, so this kind of check is not available here.
What would settle most campaign accounts is the same thing that is missing from them: a comparison group picked without regard to who won.
No field-experiment record of comparable size exists for Australia or Victoria, so the reference scale has to be borrowed from elsewhere.
It is not known how much of a published effect is the tactic and how much is simply that the election was quiet and low-turnout.
None of these studies observes a conversation between people who already know each other, which is a different thing from a knock by a stranger.
Linked notes
/notes/persuasion
/notes/turnout-tactics
/notes/the-front-republicain
Did you enjoy this article?
Recommend it — Standard Reader surfaces well-loved writing to more readers across the network.