Looking at the results of their almost 3 million comparisons, the researchers expected to find a false positive rate of about 5 percent. Instead, they found that the top three fMRI analysis programs—Statistical Parametric Mapping, the fMRIB Software Library, and Analysis of Functional Neuroimages—yielded false positive rates up to 70 percent. That’s right: seven out of 10 scans analyzed using these methods came to inaccurate conclusions. And the authors say that may be a conservative estimate.
If these findings are correct, it has a huge impact on neuroscience research. In a literature search, the authors found more than 40,000 neuroscience studies that relied on fMRI results. Still, there may be no point in going back and re-doing these studies. It’s better to look to the future, the authors say: "Considering it is now possible to evaluate common statistical methods using real fMRI data, the fMRI community should, in our opinion, focus on validation of existing methods.”
No comments:
Post a Comment