On December 11, 2014, the journal Science published a paper that read like good news for the apparently broken machinery of democratic persuasion. Its title was “When contact changes minds: An experiment on transmission of support for gay equality.” Its authors were Michael J. LaCour, a UCLA political-science graduate student, and Donald P. Green, a distinguished Columbia professor whose textbooks defined modern field-experiment methodology for political science. Its central claim was that a single, roughly 20-minute, face-to-face conversation with a gay canvasser about same-sex marriage produced large, lasting attitude change in voters who had previously opposed marriage equality — and that the effect was even larger when the canvasser was gay than when they were straight.

The media reception was unusual in its scale even for a high-prestige Science paper. This American Life devoted Episode 555 — “The Incredible Rarity of Changing Your Mind,” first aired on April 24, 2015 — to a long, sympathetic, beautifully reported account of the canvassing technique, built directly on the credibility of the Science paper. The New York Times, The Washington Post, The Atlantic, Vox, and most major political-science outlets covered the finding. A documentary about the canvassers was in production. Civil-rights organizations and political consultants were already redesigning persuasion campaigns around the technique.

Five months later, on May 19, 2015, two graduate students at the University of California, Berkeley — David Broockman and Joshua Kalla, working with Yale political scientist Peter Aronow — released a working paper titled simply “Irregularities in LaCour (2014).” It documented, with statistical precision, that the data on which the Science paper was based had almost certainly been fabricated. Within 24 hours, Donald Green requested a retraction. Within three weeks, Science retracted the paper. Within months, Princeton had withdrawn its assistant-professor offer to LaCour, and UCLA had begun its own misconduct investigation.

The interesting thing about this case is the speed. Diederik Stapel’s fraud went undetected for more than a decade. Brian Wansink’s questionable research practices accumulated public criticism for years before retractions began. The Marc Hauser case at Harvard took roughly three years from internal allegations to federal misconduct finding. LaCour’s took five months from publication to retraction — and the proximate reason is that the graduate-student replicators were attempting methodological extension, not citation. They were trying to run the same kind of experiment themselves, found they could not match LaCour’s reported response rates, and reverse-engineered the dataset to understand why.

This is also the cleanest available case of a research-misconduct story that does not end in pure cynicism. The broader hypothesis LaCour invented evidence for — that brief, empathetic, face-to-face contact can durably change attitudes on contested social issues — turns out to have real empirical support. Broockman and Kalla published a properly executed version of the experiment in Science in April 2016 (“Durably reducing transphobia”) and found a small but genuine effect on anti-transgender attitudes. The specific evidence in LaCour 2014 was fake. The broader question it pretended to answer is, as of the best current evidence, partially affirmative.

This is the story of how it happened, how it was caught, and what the case tells anyone outside academia about how to evaluate striking research-backed claims.

What LaCour & Green 2014 Claimed

The published paper described what was presented as a large, multi-wave field experiment conducted in Los Angeles County in 2013. The reported design was elegant — and, as the field would soon learn, suspiciously so.

LaCour reported that he had partnered with the Los Angeles LGBT Center to deploy canvassers door-to-door in neighborhoods that had voted for California’s anti-same-sex-marriage Proposition 8 in 2008. Some canvassers were openly gay and discussed their personal experiences; others were straight and discussed same-sex marriage in third-person terms. Voters were randomly assigned to receive a canvasser of one type, the other, or no canvasser at all (the control condition). Roughly 9,500 voters were said to have been contacted in the field experiment.

The reported measurement design was a multi-wave online panel survey administered by Qualtrics, with attitude measures collected before the canvassing, immediately after, and at multiple follow-up waves stretching out to nine months. The reported response rates and attrition figures were extraordinarily high for a panel survey of this duration — a detail that would later become important.

The headline findings, as published in Science:

  • A single 20-minute conversation with a gay canvasser produced a large, statistically significant increase in support for same-sex marriage among initially opposed voters.
  • The effect was durable, persisting at least nine months after the single conversation.
  • The effect was transmitted within the household: untreated household members of treated voters also showed attitude change.
  • The effect was substantially smaller when the canvasser was straight rather than gay, suggesting that the personal stakes of the messenger mattered, not just the message.

If real, these findings would have rewritten substantial portions of the political-persuasion literature. The dominant prior had been that durable attitude change on contested social issues is rare, that brief contact effects fade quickly, and that field-experiment persuasion effects are typically small and short-lived. LaCour’s reported effects were large, durable, and theoretically interpretable through Allport’s classic contact hypothesis. The combination made the paper irresistible to journalists and policy advocates, and it landed on the cover-story rotation of every outlet that covered behavioral science.

It also made the paper irresistible to other researchers who wanted to extend the design.

How Broockman And Kalla Caught It

David Broockman and Joshua Kalla, both then graduate students at Berkeley, read the paper and wanted to do something specific with it: they wanted to run a similar experiment on a different attitude — anti-transgender prejudice — using a roughly comparable design. This was not skepticism. This was extension. They believed the finding and wanted to build on it.

What they discovered when they began planning their own study was that the operational logistics LaCour had described were, in their own pilot work, impossible to reproduce. Specifically, LaCour’s paper reported a response rate to the baseline online survey that was dramatically higher than anything they could achieve in their pilot, even with substantially larger financial incentives offered. Broockman and Kalla contacted Qualtrics — the survey firm LaCour had named as the data-collection partner — to ask how the original study had achieved such high response rates.

What Qualtrics told them, the eventual “Irregularities” report documents, was that Qualtrics had no record of the study described in the paper. The firm could find no contract, no project, no employee matching the contact LaCour had named, and no internal record of having conducted the panel survey on the scale or with the methodology described in the published paper.

This was not yet proof of fraud. It was a serious anomaly that warranted further investigation. Broockman, Kalla, and Aronow then began a systematic statistical audit of the dataset LaCour had publicly posted alongside the Science paper. The “Irregularities in LaCour (2014)” report, released on May 19, 2015, documented a series of statistical patterns that, taken individually, were each suggestive; taken together, were essentially impossible without fabrication.

The patterns the report documented include:

  • The baseline data was statistically indistinguishable from a different, existing dataset. Specifically, the baseline survey responses in LaCour’s data closely matched, at distributional levels, the publicly available Cooperative Campaign Analysis Project (CCAP) dataset from the 2012 election cycle. The match was tighter than any genuinely independent data collection should plausibly produce.
  • The over-time changes in attitudes were distributed as nearly perfect normal noise. Real panel data on attitude change typically contains substantial heterogeneity, structural breaks, item-specific patterns, and other non-Gaussian features. LaCour’s reported changes looked, statistically, like someone had taken a baseline dataset and added small Gaussian perturbations across waves to produce the appearance of treatment effects.
  • Canvasser-identifier data was missing in ways that prevented verification of the canvasser-specific effects. The paper’s claim that gay canvassers produced larger effects than straight canvassers depended on linking effects to specific canvassers, but the data needed to verify this was either absent or insufficiently detailed.
  • Response rates and attrition figures were implausible. The panel survey would have had to perform substantially better than any comparable Qualtrics panel survey ever documented in the literature — and would have required participant incentive structures that, as later investigation showed, did not actually exist.

The “Irregularities” report did not, by itself, charge fabrication. It documented patterns that the authors said warranted urgent clarification from LaCour and an opportunity to produce the raw data. But the patterns were specific enough that anyone competent in survey methodology, reading the report, would understand what it was alleging.

The Five-Month Timeline From Publication To Retraction

The case is unusually compressed. The full sequence, drawn from contemporaneous reporting in The New York Times, Retraction Watch, New York magazine, and the principals’ own subsequent accounts:

December 11, 2014. LaCour & Green publish “When contact changes minds” in Science, Vol. 346, Issue 6215, pp. 1366–1369. Media coverage begins immediately.

Late 2014 through April 2015. Sustained popular coverage. This American Life’s Episode 555 airs on April 24, 2015, devoting its prologue and Act One to the canvassing technique. Documentary production begins. Political and advocacy organizations begin redesigning their persuasion campaigns. The result is cited in policy debates, including marriage-equality litigation arguments. The paper is treated as established fact in a wide range of derivative work.

Early 2015. Broockman and Kalla, planning their own field experiment on transphobia, attempt to pilot the recruitment design described in LaCour & Green. They cannot replicate the reported response rates.

Roughly May 5–15, 2015. Broockman and Kalla, joined by Peter Aronow at Yale, contact Qualtrics and begin a systematic statistical audit of LaCour’s publicly posted dataset. The Qualtrics denial and the dataset anomalies converge.

May 16, 2015. Broockman, Kalla, and Aronow contact Donald Green privately to share their preliminary findings.

May 17, 2015. Green, after reviewing the report, tells Broockman and Kalla that he agrees a retraction is warranted unless LaCour can produce evidence to rebut the findings. He immediately begins contacting LaCour to demand the underlying raw data.

May 19, 2015. Green submits a formal retraction request to Science, having determined that LaCour cannot produce the requested raw data files. He posts the retraction notice publicly on his Columbia website the same day. Broockman, Kalla, and Aronow release the “Irregularities in LaCour (2014)” report publicly.

May 20, 2015. Major news coverage begins. Benedict Carey and Pam Belluck publish “Doubts About Study of Gay Canvassers Rattle the Field” in The New York Times. Retraction Watch publishes its first report. The story moves rapidly from political-science circles into mainstream news.

May 28, 2015. Science posts an Editorial Expression of Concern. LaCour posts a 23-page online response attempting to address the irregularities; the response is widely judged by other methodologists as insufficient — it does not produce the raw Qualtrics files, does not produce evidence of the survey having been conducted, and does not address the CCAP-baseline-match finding.

June 5, 2015. Editor-in-chief Marcia McNutt publishes the formal retraction notice in Science, Vol. 348, Issue 6239, p. 1100 (DOI: 10.1126/science.aac6638). The retraction cites three specific grounds: (1) the misrepresentation of the survey incentives — LaCour’s published methodology claimed cash payments to respondents that, per LaCour’s own attorney’s subsequent correspondence with Science, had never actually been paid; (2) the false statement of sponsorship — LaCour had claimed funding from the Williams Institute, the Ford Foundation, and the Evelyn and Walter Haas Jr. Fund that, per the same attorney correspondence, he had not actually received or used; and (3) LaCour’s failure to produce the raw data files that the methodology described.

The total elapsed time from publication to retraction was 176 days — less than six months.

Green’s Response

Donald Green’s behavior in this case is worth examining carefully, because it is structurally important to how research fraud gets caught and to how academic culture handles the senior-author-protection problem.

Green, by 2014, was one of the most prominent field experimentalists in political science. He had moved from Yale, where he had directed the Institution for Social and Policy Studies, to Columbia in 2011. His textbook Field Experiments: Design, Analysis, and Interpretation (with Alan Gerber, 2012) was a standard methodological reference. The implicit credibility of his name on the LaCour paper was, in fairness, a substantial part of why the paper was received as seriously as it was.

Green did not see the raw data before publication. The arrangement, as Green has subsequently described in interviews and in his retraction letter, was that LaCour conducted the field experiment and the survey, and Green served as a senior collaborator on the design and analysis. The data files Green worked with were the cleaned, processed versions LaCour produced — not the raw Qualtrics outputs. This is the same structural failure mode that enabled Stapel’s fraud at Tilburg: a senior researcher trusts a junior collaborator with the data-handling stage, and the junior collaborator’s fabrication is therefore invisible to the senior researcher until external scrutiny forces the question.

What is distinctive about Green’s response, once the question was forced, is the speed of his action.

Within roughly 72 hours of Broockman and Kalla contacting him, Green had reviewed the “Irregularities” report, demanded raw data from LaCour, concluded that LaCour could not or would not produce it, and submitted a formal retraction request to Science. He did not attempt to defend the paper. He did not seek a quieter resolution. He did not delay to protect his own reputation or LaCour’s career prospects. He wrote, in his retraction letter, “Michael LaCour’s failure to produce the raw data coupled with the other concerns noted above undermines the credibility of the findings,” and added that he was “deeply embarrassed by this turn of events.” He apologized to the journal, the reviewers, and the readers.

This is the senior-author-protection problem solved in real time. The default failure mode in research-fraud cases is that senior authors, motivated to protect their reputations and the careers of their collaborators, slow-walk the retraction, demand impossibly high standards of proof before agreeing to it, or attempt to negotiate softer language. Green did not. His behavior is, in retrospect, one of the reasons this case unwound so quickly — and is one of the reasons his own reputation in the field, while damaged, was not destroyed.

This is not a defense of Green’s role in the original mistake. He was a co-author on a paper whose data he had not adequately verified, and he carries the appropriate share of responsibility for that. But his behavior once the question was raised demonstrates what it looks like when a senior co-author refuses to be the chokepoint that protects fabricated data from scrutiny — and the contrast with how other research-fraud cases have unfolded is instructive.

The Aftermath

The institutional consequences for LaCour were swift and substantial.

Princeton withdrew its job offer. LaCour had accepted an assistant-professor position at Princeton’s Department of Politics for the fall of 2015. Princeton rescinded the offer in the weeks following the retraction. The CV LaCour had submitted to Princeton during the hiring process was subsequently found to contain additional fabrications beyond the Science paper — including a teaching award that did not exist and claimed grant receipts totaling, by some reporting, in the range of $793,000 that he had not actually received. The CV falsifications were a substantial additional element of misconduct that came to light only after the data-fabrication investigation began.

UCLA’s misconduct investigation proceeded. UCLA convened a formal investigation, which concluded in 2015 with findings of research misconduct. The question of whether UCLA could retroactively rescind LaCour’s PhD became legally complex, and the public record on the final disposition is not entirely clear. Contemporaneous reporting indicated that UCLA’s procedures for revoking a granted degree posed substantial procedural obstacles. What is documented is that LaCour did not pursue an academic career after the case, and that UCLA’s findings of misconduct stand on the record.

Carnegie withdrew Green’s fellowship. The Carnegie Corporation rescinded the 2015 Andrew Carnegie Fellowship (approximately $200,000) that Green had been awarded for separate work, citing the reputational cloud from the LaCour case. Green retained his Columbia appointment and continued to publish and supervise students; his subsequent work has been received normally in the field.

This American Life retracted its coverage. On May 22, 2015, Ira Glass posted a public retraction of the canvassing-study portions of Episode 555, explaining that the show had only aired the story “because there was solid scientific data published in the journal Science.” The retraction segment was added to the episode. The documentary that had been in production was halted.

LaCour did not face criminal charges. As is typical in academic research-fraud cases in the United States, criminal prosecution requires that prosecutors identify specific elements — fraud on a federal grant, theft of federal funds, false statements in a federal grant application — that fit the conduct. In the LaCour case, the grant-funding claims LaCour had made were, according to his own attorney’s subsequent correspondence with Science, fabrications: he had not actually received or used the grants he claimed. The absence of real federal grant fraud paradoxically made the case harder to prosecute criminally, even as it deepened the misconduct findings.

The Redemption: Broockman & Kalla 2016

The most distinctive feature of this case in the literature on research fraud is what happened next.

Broockman and Kalla did not abandon the broader hypothesis. They were already in the planning stages of their own canvassing field experiment when they uncovered the LaCour fraud. They proceeded with that experiment, properly executed, and published the results in Science on April 8, 2016: “Durably reducing transphobia: A field experiment on door-to-door canvassing” (Broockman & Kalla, 2016, Science, 352(6282), 220–224, DOI: 10.1126/science.aad9713).

The 2016 paper used a roughly 10-minute door-to-door conversation in which canvassers — both transgender and cisgender — engaged voters in South Florida in an active perspective-taking exercise about gender identity. The design was field-experimental, with random assignment of voters to treatment and control conditions. Outcome attitudes were measured by online surveys at multiple follow-ups out to three months.

The results, in summary:

  • The 10-minute conversation produced a measurable, statistically significant reduction in anti-transgender prejudice.
  • The effect persisted at three months — the longest follow-up reported in the published paper.
  • The effect was roughly equivalent whether the canvasser was transgender or cisgender (in contrast to LaCour’s fabricated claim that the canvasser’s identity mattered substantially).
  • The effect size, while real, was small to moderate — comparable in magnitude to a decade of natural attitude change in the population on similar issues.

The 2016 paper has itself been subject to subsequent scrutiny — appropriate for any influential single study — and has been the subject of independent replication efforts and methodological discussion. Broockman and Kalla have continued to publish in this area, including subsequent follow-on field experiments that have refined the findings and clarified the conditions under which “deep canvassing” produces durable effects.

The broader hypothesis is therefore not dead. The specific evidence LaCour fabricated was fake, but the question it pretended to answer — can brief, empathetic, face-to-face contact durably change attitudes on contested social issues? — has been investigated by properly executed experiments and has, with appropriate qualifications about effect size and conditions, an empirical basis.

What’s Honest To Say About Persuasion Research Now

The honest synthesis, given the full record:

Brief in-person contact can change attitudes, in some conditions, in measurable but typically modest ways. Broockman and Kalla’s properly executed 2016 paper, and subsequent work by them and others, provides evidence for this in the specific context of door-to-door canvassing about transgender rights. The effect sizes are real but smaller and more conditional than LaCour’s fabricated claims suggested.

The LaCour-specific findings are gone from the literature. The 2014 Science paper is retracted. Its reported effect sizes, durability, household-transmission claim, and gay-canvasser-versus-straight-canvasser contrast should not be cited as evidence of anything except the case study of how the fabrication was uncovered.

The “deep canvassing” technique — the procedural approach of engaging voters in extended perspective-taking conversations rather than scripted persuasion appeals — has accumulated genuine evidence. It is the methodological core of the work that did replicate, and political-organizing groups using the technique can point to real research support for the practice.

The effects are not magic. Real persuasion effects on contested attitudes are typically modest, conditional on the specific context and the specific outcome measure, and sensitive to the design of the contact. The folk understanding that emerged from the LaCour coverage — that a single 20-minute conversation could durably flip voters on a contested social issue — was always implausibly strong. The retraction confirmed that intuition was correct; the legitimate replication evidence has not restored the original magnitude.

The redemption arc is not a closed argument. The 2016 paper and its successors are themselves single studies that warrant ongoing scrutiny, replication, and methodological discussion. The lesson is not “we have proven LaCour’s hypothesis was right after all” — it is “the hypothesis is alive, the evidence is partial, and the methodological practices that originally caught the fraud are the same practices that should continue to evaluate the legitimate work.”

What This Means For Strategists Evaluating Political/Social Persuasion Research

If you are a CEO, consultant, or strategist commissioning persuasion research, evaluating advocacy spending, designing internal-communications interventions, or otherwise relying on the social-psychology literature on attitude change, the LaCour case offers concrete diagnostic discipline.

Ask who collected the data — by name, by institution, and with verification. The fastest signal that triggered the LaCour unmasking was that the survey firm LaCour had named denied any involvement. If a study cites a specific data-collection partner (Qualtrics, YouGov, Knowledge Networks, an internal university lab), it is legitimate to ask whether that partner has confirmed the work. In a real, honest study, the answer should be trivially yes.

Ask whether the raw data is available for independent scrutiny. LaCour’s data was, in fact, publicly posted — that is what made the audit possible. But the audit depended on Broockman, Kalla, and Aronow having the methodological expertise to spot the patterns. The norm of public data is necessary but not sufficient; the norm of independent audit is what completes the verification chain.

Treat response rates and attrition figures as load-bearing parameters, not boilerplate. The LaCour fraud was first cracked because reported response rates were impossibly high for the described design. In commissioning or evaluating persuasion research, the operational logistics — how many people were contacted, how many responded, how was attrition handled — are not back-matter details. They are the structural skeleton of the claim, and implausible numbers there are a leading indicator of broader problems.

Weight replicated findings, in the same area, far more heavily than single striking studies. The 2016 Broockman & Kalla paper is more credible than the 2014 LaCour & Green paper not just because the 2014 paper was retracted, but because the 2016 paper’s broader claim has been the subject of follow-on work by independent investigators, refining the effect under varying conditions. A finding with three or more independent replications in the same area is qualitatively different evidence than a single dramatic result.

Be calibrated about effect magnitudes. LaCour’s reported effects were extraordinarily large by the standards of comparable field experiments. The properly executed work that followed shows effects that are real but modest — typical for genuine persuasion research. When a study reports effect sizes that are dramatically larger than the surrounding literature, the appropriate response is increased scrutiny, not increased enthusiasm.

Distinguish “the dramatic single study turned out to be fake” from “the underlying question has no answer.” This is the most important distinction the LaCour case teaches. The dramatic single study was fake. The underlying question — whether interpersonal contact durably changes attitudes — has an honest, partial, calibrated answer in the legitimate work. Strategists who concluded after the retraction that persuasion was impossible were as wrong as strategists who had concluded after the original publication that persuasion was easy. The truth was in between, and the legitimate replication infrastructure is the mechanism that revealed it.

The deepest lesson of the LaCour case is not about fraud detection. It is about how the verification chain in social science actually works when it works. Broockman and Kalla did not catch this fraud because they were skeptical. They caught it because they believed the finding strongly enough to try to extend it, and the extension exposed the underlying impossibility. The institutional infrastructure that turns “I cannot reproduce this” into a formal retraction in 176 days — public data, accessible auditing tools, journal responsiveness, senior co-author willingness to act, and a culture of methodological extension as a routine activity — is what good science actually looks like. The LaCour case is, paradoxically, evidence that this infrastructure can work, and work quickly, when researchers behave with the integrity Broockman, Kalla, Aronow, and Green ultimately did.

Sources

  • The Replication Crisis hub — the full set of cases, methods, and decision frameworks for strategists evaluating “research-backed” claims.
  • Diederik Stapel and the 58-Retraction Fraud — the same structural failure mode (senior researcher controls the data, junior collaborators trust without verification) operating over a decade rather than a few months. The contrast in detection speed is itself diagnostic.
  • Brian Wansink and Mindless Eating — a different shape of misconduct (questionable research practices and self-plagiarism rather than outright fabrication) that nevertheless produced a similar retraction cascade in a high-prestige research program.
  • Marc Hauser and the Monkey Cognition Fraud — the parallel case at Harvard, where the misconduct involved fabricated coding of behavioral data rather than fabricated survey panels, and where the senior-author-protection problem played out in the opposite direction from the Green case.
  • Daryl Bem and Precognition — published in JPSP in 2011, the same archive of social-psychology results that produced the broader replication crisis. The LaCour and Bem cases together represent the two failure modes of striking single-study findings in the field.

FAQ

Did deep canvassing actually work?

Yes, in a more limited and conditional form than LaCour’s 2014 paper claimed. Broockman and Kalla’s properly executed 2016 Science paper on transgender-rights canvassing found a real, statistically significant reduction in anti-transgender prejudice from a single roughly 10-minute door-to-door conversation, persisting at three-month follow-up. Subsequent work by the same team and others has refined the conditions under which the effect appears and is robust. The effect sizes are modest — comparable to several years of natural attitude change — not the dramatic, large effects LaCour fabricated. The technique has real research support; it is not a silver bullet.

Is Donald Green credible?

His role in the original paper is a real mark on his record: he co-authored a paper whose data he had not independently verified, and that failure of verification is appropriately his to carry. His behavior once the fraud was raised is the countervailing evidence: within roughly 72 hours of being contacted by Broockman and Kalla, he reviewed the report, demanded raw data from LaCour, and submitted a formal retraction request to Science when LaCour could not produce it. He did not stall, defend, or attempt to negotiate softer language. His subsequent work has been received normally in the field, and his methodological textbooks remain widely used. The honest synthesis is that he made a serious mistake of verification in the original paper and behaved with appropriate institutional integrity in response — both of which inform any reading of his subsequent work.

How was the survey firm’s denial discovered?

When Broockman and Kalla attempted to pilot a similar canvassing experiment, they found they could not match LaCour’s reported response rates even with substantially larger participant incentives. They contacted Qualtrics — the survey firm LaCour had named as the data-collection partner — to ask how the original study had achieved such high response rates. Qualtrics responded that it had no record of the project, no contract matching the described study, and no employee matching the contact LaCour had named. That denial was the first concrete signal that the operational logistics LaCour had described had not occurred. Statistical audit of the publicly posted dataset followed, and the convergence of the Qualtrics denial and the dataset anomalies produced the “Irregularities” report.

Did LaCour go to prison?

No. As is typical in academic research-fraud cases in the United States, criminal prosecution requires prosecutors to identify specific federal statutes the conduct violated — typically fraud on a federal grant, theft of federal funds, or false statements in a federal grant application. In LaCour’s case, the grants he had claimed to receive were, per his own attorney’s subsequent correspondence with Science, fabricated; he had not actually received the federal money he claimed. That absence of real federal grant fraud paradoxically made criminal prosecution harder. The institutional consequences — retraction, withdrawn job offer, UCLA misconduct finding, end of academic career — were the operative sanctions.

Why did this fraud get caught in five months when others took years?

Because the people who caught it were not skeptics looking for fraud. They were extension researchers trying to run a similar experiment, who discovered through their own pilot work that the reported response rates were impossible. The fastest way to expose fabricated data is for someone to try to do the same thing the original paper claimed to have done. Citation does not catch fraud; methodological extension does. The 2016 Science paper on transphobia canvassing was, in effect, the test of LaCour’s methodology — and its failure to reproduce his operational logistics was the leading indicator that turned into the formal audit.

What is “deep canvassing”?

Deep canvassing is the procedural approach of engaging voters in extended, perspective-taking conversations — typically 10 to 20 minutes — rather than brief scripted persuasion appeals or single-message contacts. The technique was developed in practice by political and advocacy organizations before LaCour’s paper, and was the procedural core of the canvassing methodology described in both the retracted 2014 paper and the legitimate 2016 follow-up. The 2016 evidence supports that the technique produces measurable, durable attitude shifts on at least some contested social issues. The technique should not be confused with the specific 2014 LaCour data, which was fabricated.

What did UCLA do about LaCour’s PhD?

UCLA convened a formal research-misconduct investigation that concluded with findings of misconduct in 2015. The legal and procedural question of whether UCLA could retroactively rescind a PhD that had already been granted is genuinely complex, and the public record on the final disposition is incomplete. What is documented is that the misconduct findings stand on the record, that LaCour did not pursue an academic career after the case, and that the institutional question of degree revocation in research-fraud cases remains an unresolved area of university administrative practice.

What is the single most important lesson for someone outside academia?

The verification chain in social science depends, far more than people realize, on someone wanting to do the same thing the original paper claimed to have done. Citation does not test a finding. Coverage does not test a finding. Even peer review does not, structurally, test a finding — peer reviewers do not see the raw data and cannot re-run the study. The thing that tests a finding is somebody else attempting to extend it, and discovering whether the operational logistics, the response patterns, and the effect sizes are reproducible in their own hands. The LaCour case is the cleanest available proof that this verification works — and that it can work quickly, when researchers behave with integrity. When evaluating any single striking research-backed claim, the operative question is not “was this peer-reviewed?” but “has anyone else tried to do this, and what happened when they did?”

replication-crisis lacour-fraud research-misconduct political-science evidence-evaluation
Share this article
LinkedIn (opens in new tab) X / Twitter (opens in new tab)
Atticus Li

Experimentation and growth leader. CXL-certified CRO practitioner, Mindworx-certified behavioral economist (1 of ~1,000 worldwide). 200+ A/B tests across energy, SaaS, fintech, e-commerce, and marketplace verticals.

About LinkedIn Newsletter