The LaCour canvassing study
A celebrated study found that a short conversation with a gay canvasser durably changed minds on same-sex marriage. The data was invented. The finding was then partially vindicated by researchers who did the work properly — which is the part worth knowing.
Why this verdict
Debunked and retracted: the survey data underlying the 2014 finding was fabricated, established by graduate students attempting a replication, and the paper was withdrawn by its co-author within days.
- Period
- Region
- Reading time
- Sources
01The claim
That a single brief, personal conversation with a gay canvasser produced large and lasting changes in voters' attitudes toward same-sex marriage — a result published in Science in December 2014 by Michael LaCour and Donald Green, and widely reported as evidence that prejudice could be reduced cheaply and at scale.
02Where it came from
The paper was unusually influential for a single study. It was covered by major outlets, discussed on This American Life, and taken up by advocacy organisations designing campaigns around it. LaCour was a UCLA doctoral candidate; Green, a senior political scientist at Columbia, was co-author and had not collected or held the raw data.
03What the record shows
✕ DEBUNKED — Investigated and refuted. Here is what actually happened.
The unravelling is a model of how this is supposed to work. David Broockman and Joshua Kalla, then graduate students at Berkeley, wanted to run an extension and requested the materials to design it. The cost estimates would not reconcile: replicating the described survey would have cost far more than any plausible grant. They obtained the dataset and, with Peter Aronow, found that its distribution matched an existing public dataset, the Cooperative Campaign Analysis Project, with statistical noise added — the signature of simulation rather than collection.
They contacted the survey firm named in the paper, which said it had no record of the work and lacked the capability described. LaCour was unable to produce raw data, saying it had been deleted. Green, on seeing the analysis, requested retraction immediately; Science retracted the paper in May 2015. UCLA investigated, and a job offer LaCour had accepted from Princeton was withdrawn.
The entry is here rather than in an academic-misconduct file because of what the fabrication was attached to: the claim was seized on and repeated far beyond its evidence, and the incentive that produced it — a result too good and too useful to interrogate — is the same incentive that carries much of the material in this archive.
The coda matters and is usually left off. Broockman and Kalla went on to run the study properly, on transgender prejudice in Miami, and published in Science in 2016: they found a real, measurable and durable reduction in prejudice from deep canvassing, smaller than LaCour's invented effect but genuine and since replicated. The idea survived its fraudulent evidence, because someone did the work.
04What it cost
Advocacy organisations had built training and spending on a fabricated result, and had to rebuild. Green's reputation absorbed damage for a data-handling arrangement — co-authoring without ever holding the data — that was then common and is now much less so. The episode is a standard teaching case for data-sharing requirements and for the specific vulnerability of a discipline where a single author can hold the only copy of the evidence.
05Connections
- A Rape on CampusDEBUNKED
- The Amarna PrincessCONFIRMED
- A Gay Girl in DamascusDEBUNKED
- The grievance studies affairDEBUNKED
- The Himalayan fossil hoaxDEBUNKED
- The Seigenthaler Wikipedia hoaxDEBUNKED
- The Sokal affairDEBUNKED
- Study 329CONFIRMED
- The Sugar Research Foundation's funded scienceCONFIRMED
- The tobacco industry's manufactured doubtCONFIRMED
- Vaccines and autismDEBUNKED