Fraud topples second neuroscience word processing paper

We have a second retraction from a group of neuroscience researchers in Belgium who discovered fatal errors in their work on how the brain sets about the task of reading written language. Spoiler alert: Turns out those errors weren’t errors after all.

As we reported back in May, the group, from the University of Leuven, was unable to replicate certain fMRI findings in a November 2012 article in Neuroscience. At the time, Hans P. Op de Beeck, who led the group, told us:

As PI I have made the decision to retract once an internal lab investigation provided me with a full understanding of what had gone wrong and why. All authors support this decision, the author who made the errors fully collaborated with the investigation, and thus there was no reason to wait for lawyers or administrative reports (which can give very long delays and related frustration, as you know very well). The decision to retract is also a conservative decision, in the sense that one could wonder whether it would have been possible to save parts of the paper through a correction, but we preferred to follow a high standard in this matter.

That’s changed a bit, as we’ll see in a moment.

The second paper, titled “The visual word form area is organized according to orthography,” appeared in NeuroImage, and has been cited 13 times, according to Thomson Scientific’s Web of Knowledge. According to the abstract:

Efficient word reading depends on a left fusiform brain region, the Visual Word Form Area (VWFA). The internal organization of this brain area is currently unknown, as are the potential factors that might underlie this organization. Here we combine functional magnetic resonance imaging (fMRI) with multi-voxel pattern analyses (MVPA) to probe the internal organization of the VWFA. Our findings reveal distinct activation patterns in the VWFA associated with individual letter strings, as well as for pictures. This result demonstrates that the VWFA is organized in such a way that similar words (as well as pictures) activate neurons consistently, irrespective of their exact visual appearance or location. The activation patterns for letter strings were driven primarily by orthographic similarity, and we observed no effects of semantics or lexical status (words versus pseudowords). This effect of orthographic similarity was significant only in the VWFA and not in retinotopic areas and object-selective control regions. Given the relationships between the internal organization and how visual input is processed in a brain region, our findings provide important constraints for computational models of how visual word forms are represented in the brain.

Here’s the retraction notice:

This article has been retracted at the request of the Authors.

Because of errors by the first/corresponding author the fMRI data reported in this article were incorrectly analyzed. The original data analyses by the first/corresponding author included several operations that the authors have been unable to replicate. Additionally, the methods as agreed upon with the coauthors and described in the paper were not accurately followed. For this reason the results are untrustworthy.

The first author regrets the errors and takes responsibility for them, and has cooperated fully with the investigation. In case of (suspected) plagiarism, it is mandatory to refer to the plagiarized work here by < YNIMG, 59/3 (2012) 2751–2759 >, http://dx.doi.org10.1016/j.neuroimage.2011.10.032.

We’re not quite sure what that last sentence means, and have asked the journal’s publisher, Elsevier, to clarify.

But more to the point: Op de Beeck tells us that although the retraction notice doesn’t mention fraud, the investigation concluded that fraud had indeed occurred. Here’s the official notification from Leuven, which has yet to make it into the retraction notices:

This publication has been the object of an investigation by the KU Leuven Commission of Scientific Integrity (CSI). The conclusion of the investigation is that the analysis of the data represented in this paper was manipulated intentionally by the first author: Wouter Braet.

Wouter Braet has fully cooperated with the investigation of the CSI and has admitted the fraud. He has also confirmed that the coauthors were not informed about his manipulation of the analysis of the data. The CSI wishes to stress that the coauthors, Wouter Braet’s colleagues and supervisor have consistently observed the relevant methodological and scientific standards.

The CSI has acted upon the conclusion that the conduct of Wouter Braet was highly questionable in terms of scientific integrity.

This publication has been retracted, and the results described in it cannot be seen as valid or truthful.

Op de Beeck says the NeuroImage notice will eventually catch up.

What you see on their website, is the authors’ retraction notice as it was submitted by us in April and published by the journal late in the summer. It is almost the same notice as the old Cerebral Cortex which you referred to in your blog post 9 months ago (except the odd addition about what to do in the case of plagiarism).

What I refer to as the most latest version which defines the misconduct very clearly, is not visible yet on the NeuroImage website (but it is visible for the Cerebral Cortex paper). Despite the fact that this latest version was already submitted by our university many months ago (I think in October).

Braet’s name appears on at least eight publications that we can see. But Op de Beeck tells us there will be only two retractions.

All relevant retractions were made public at one point in time, April/May 2013. This is it. Nor we as coauthors, nor the formal institutional investigation, found any indication of the opposite.

And I truly hope I will never have to experience this again. Dealing with (and minimizing) uncertainty is part of being an empirical scientist, but the uncertainty should come from the data, not from having to mistrust one’s colleagues.