I4R's the CRAVE files: An Investigation into Football Psychology

The CRAVE Files: An Investigation into Football Psychology

Institute for Replication (I4R) | November 2026

In October 2026, the Institute for Replication (I4R) published a five-part series documenting an investigation into research on elite footballers, cognitive performance, and psychological assessments. What began as an attempt to reproduce the findings of a single article published in Proceedings of the National Academy of Sciences (PNAS) developed into a broader investigation involving several journals, overlapping groups of authors, and multiple datasets.

Our investigation has identified duplicated observations, inconsistencies between published statistics and underlying data, discrepancies across supposedly independent studies, and problems with statistical reporting. Two articles have already been formally retracted, while concerns about several others have been communicated to authors and journal editors.

1. From a PNAS article to a broader investigation

The investigation began in May 2025, when I4R replicator Niklas Jakobsson contacted I4R Chair Abel Brodeur about a PNAS article entitled Decoding the elite soccer player's psychological profile, published in January 2025.

The article examined psychological and cognitive differences between elite footballers and control participants, with potential implications for identifying and developing sporting talent.

Jakobsson and fellow I4R replicator Jack Fitzgerald began examining the study. Their initial concerns focused on the selection of control participants and the statistical analyses. I4R submitted a comment to PNAS in July 2025, which was subsequently published in February 2026. Another independent research team also published a comment, and the original authors responded to both.

Initially, we considered the matter largely concluded. However, in January 2026, independent researcher Kristjan Moore contacted I4R with additional findings that substantially changed our understanding of the study.

Moore identified several unusual features of the Brazilian dataset, including unexpected correlations, irregular distributions, exceptionally large differences between groups, and patterns in the control data consistent with randomly generated observations.

Importantly, our initial replication efforts had not identified these problems. Moore's independent examination demonstrated the value of having multiple researchers approach the same study using different methods.

His findings were documented in I4R Discussion Paper 311, and the original authors subsequently published a response in our discussion paper series.

On October 15, 2026, PNAS formally retracted the article. In their retraction statement, the authors acknowledged methodological and reporting inaccuracies, scoring and processing errors, and statistical irregularities affecting portions of the Brazilian dataset.

The retraction represented an important outcome, but our investigation had already expanded to other publications.

2. Duplicated observations in a Brazilian validation study

Following Moore's findings, we examined another article involving one of the researchers responsible for the Brazilian data in the PNAS study.

The article, The CRAVE and ARGE scales for motivation states for physical activity and sedentarism: Brazilian Portuguese translation and single-item versions, had been published in Frontiers in Psychology in 2023.

The CRAVE scale (Cravings for Rest and Volitional Energy Expenditure) measures motivational states associated with physical activity and rest. Its validation is relevant to research examining exercise motivation and sedentary behaviour.

The publicly available Brazilian dataset contained 1,168 observations. Our examination identified an extraordinary pattern: the first 584 observations were almost exact copies of the subsequent 584 observations. Of the 584 matched pairs, 572 were identical across every variable other than participant ID. The remaining 12 pairs differed only slightly.

We also identified problems involving derived variables and participant eligibility.

These findings were documented in I4R Discussion Paper 313.

On July 30, 2026, Frontiers in Psychology retracted the article. The journal's notice acknowledged substantial data-handling errors, including duplicated response patterns, incorrect calculations, and the inclusion of ineligible participants.

3. Recovering deleted data and examining additional football studies

We subsequently examined the public Git repository associated with the original PNAS article. Its version history contained a dataset that had been uploaded and deleted on the same day. We recovered a file containing 53 variables, including information absent from the dataset that remained publicly accessible.

The recovered material allowed us to examine additional measures and challenge the original authors' assertion that the principal irregularities were concentrated in the first testing session of the Brazilian control group.

The recovered dataset also revealed a concerning connection to a different publication.

In BMC Psychology, an article entitled Cognition in soccer and futsal: Evidence of validity of a 4-instrument protocol to assess executive functioning among women athletes reported results involving 51 female footballers from three Brazilian clubs.

However, 14 of the 18 rows in one of its statistical tables contained means and standard deviations identical to those calculated for a group of 30 male footballers in the recovered PNAS dataset. The matching variables included football performance measures and cognitive tests.

The two articles described different samples: female athletes in the BMC study and male athletes in the PNAS study. Both also referenced the same ethics approval.

These similarities raise substantial questions about the independence and provenance of the underlying observations. We communicated our findings to the relevant authors and journal.

We also examined a 2026 study published in the International Journal of Environmental Research and Public Health concerning anxiety, depression, and suicidal ideation among elite male footballers.

Among 110 control observations, 18 were exact duplicates across all 49 variables, including 42 questionnaire responses. The journal informed us that the concerns had been referred to its ethics and integrity team.

These findings illustrate how irregularities identified in one dataset can warrant closer scrutiny of related publications, without implying that problems in one study automatically invalidate others.

4. Concerns involving CRAVE validation studies in other countries

Our investigation subsequently extended to studies validating the CRAVE scale in other languages and populations.

In a Persian-language validation study published in Current Psychology, we initially noticed an unusual distribution of reported statistical significance levels. The nonsignificant p-values discussed in the text were concentrated immediately above the conventional 0.05 threshold, between 0.051 and 0.054. Although such a pattern is not itself evidence of misconduct, it warranted further examination.

After the authors provided their data, we found extensive duplication of questionnaire response profiles. Across three reported samples of 313, 346, and 244 participants, there were only 98, 206, and 83 distinct 26-item CRAVE response profiles, respectively. Moreover, 238 of the 244 observations in Study 3 exactly reproduced a response profile from Study 1.

We also identified inconsistencies between calculated scores and their underlying questionnaire items, as well as discrepancies in reported statistical tests. These findings are discussed in I4R Discussion Paper 336.

In a separate Chinese-language validation published in Frontiers in Psychology, our examination identified repeated survey submissions sharing identifying fields, discrepancies between reported and observed sample sizes, and possible scoring inconsistencies. For example, one control group contained 500 observations despite the article reporting 481. Some recorded scores exceeded the stated scale maximum, and inconsistencies in scoring potentially affected a central comparison in the study.

Our findings and the authors' subsequent response are available through I4R Discussion Paper 324 and Discussion Paper 325.

Other attempts to examine CRAVE validation studies were constrained by data availability. In some cases, authors declined to provide either the data or the analysis code; in others, requests went unanswered.

We also examined an article in PLOS ONE on physical activity and well-being during the COVID-19 pandemic. The publicly deposited dataset contained 497 observations, while the article reported different sample sizes in its Methods and Results sections. We additionally identified observations that did not satisfy the study's stated eligibility criteria.

These matters have been communicated to the relevant authors and editors. Not every investigation has concluded, and the existence of statistical irregularities should not be interpreted as establishing deliberate misconduct.

5. Journal responses and the challenges of post-publication scrutiny

One striking feature of this investigation has been the variation in journal responses.

Some journals have initiated formal investigations, and two have already issued retractions. Other journals have declined to consider our comments because their editorial policies do not provide for the publication of post-publication comments.

In several cases, identifying irregularities required relatively little time. Determining whether the journals would investigate, obtaining responses, and reaching editorial decisions took considerably longer.

This disparity raises important questions about the incentives facing researchers who conduct replication and forensic investigations. These activities can require significant time and expertise, yet opportunities to publish the findings and receive academic recognition remain limited.

At the same time, an important issue must be acknowledged: studies whose authors share data are more susceptible to independent scrutiny than studies whose data are inaccessible