A team of researchers in Europe has retracted a 2016 paper on how people with autism process social cues after finding an error in their analysis.
The article, “Social Bayes: using Bayesian modeling to study autistic trait–related differences in social cognition,” appeared in Biological Psychiatry, an Elsevier journal.
The senior author of the paper is Leonhard Schilbach, of the Max Planck Institute of Psychiatry in Munich and University Hospital Cologne. According to the abstract of the article:
Autism is characterized by impairments of social interaction, but the underlying subpersonal processes are still a matter of controversy. It has been suggested that the autistic spectrum might be characterized by alterations of the brain’s inference on the causes of socially relevant signals. However, it is unclear at what level of processing such trait-related alterations may occur.
…
Our results demonstrate that higher autistic traits in healthy subjects are related to lower scores in a learning task that requires social cue integration. Computational modeling further demonstrates that these trait-related performance differences are not explained by an inability to process the social stimuli and its causes, but rather by the extent to which participants take into account social information during decision making.
The paper has been cited 25 times, according to Clarivate Analytics’ Web of Science. Schilbach tells Retraction Watch:
After publication of this paper, we have used the experimental task to collect new data in different patient and a control group.
When analysing these new data sets, we went back to the original data obtained from a group of healthy controls to check the consistency of our findings. While doing this, we noticed an error in our previous analysis and informed the journal about it. Together with the journal, we then decided to retract the paper to make this transparent.
Following the journal’s advice, we have revised the original paper in light of the different findings and have resubmitted it to the journal for peer review.
The notice reports that
…the authors noticed that an error was made when transforming one of the variables used in computing the outcome prediction in their computational model. More specifically, in order to combine the social and nonsocial (card-related) cues into one prediction about the outcome, the card color probabilities in reference to where the social agent directs the gaze had to be transformed. This transformation was performed incorrectly in the original model, leading to an incorrect estimation of the response model parameters.
The authors reanalyzed the properly transformed data and could reproduce all of the results from the original paper, with the exception of the correlation between AQ scores and the social weighting parameter. Instead, the authors found that the model parameter determining belief precision differed as a function of autistic traits. Upon correction, they found that low AQ scorers took the gaze schedule into account to adjust their learning rates about the card probabilities, and performed better on the task. High AQ scorers, on the other hand, failed to adjust their learning rates according to the gaze schedule and thus relied more often on the gaze in phases when it was more unreliable, in phases of high volatility and low accuracy.
The authors voluntarily informed the Journal of this honest error upon its discovery. Because of the extent and nature of the changes to the paper, the editors and authors concluded that, to ensure maximum clarity and transparency, the only course of action was to retract this version of the paper. The authors are revising the paper, which the Journal will re-review and consider further for publication.
Like Retraction Watch? You can make a tax-deductible contribution to support our work, follow us on Twitter, like us on Facebook, add us to your RSS reader, sign up for an email every time there’s a new post (look for the “follow” button at the lower right part of your screen), or subscribe to our daily digest. If you find a retraction that’s not in our database, you can let us know here. For comments or feedback, email us at [email protected].
It would be interesting to know how many of these kinds of errors are generated by the use of spreadsheets for data analysis, with all their automated and pretty opaque processing of input data. I suspect a lot.