Logical Fallacies in Neuroscience (Neuroimaging) Research

So last night I spent time reviewing an article. I previously wrote some code for the analysis of these data. The code was designed to compute average source activities in different regions of interest. Source localization of EEG data involves computing the possible cortical sources of electrical activity seen at EEG electrodes. Source localization tries to solve an inverse problem using different mathematical methods. A great description of sources localization procedures could be found here.

EEG Source localization from one of my thesis studies.

As I made my way through the article, I realized something was off. Only after reading about fmri analysis (and the pitfalls) was I able to actually express what was wrong. So here it is. Many of the arguments we make in neuroimaging research are logical fallacies. The flaws in logic stem from two interconnected but separate ideas. The first idea is forward inference- that is, we induce some change and that change results in the activation of some areas of the brain. The second idea is reverse inference, that is, we see some activity in an area of the brain and based on the pattern of activity we say that a certain behaviour is occurring.


Both forward and reverse inferences are possible, and they can be expressed in relation to on another. The logical fallacy occurs when we take a forward inference from an experiment and then argue for using the evidence we obtained using statements that refer to a reverse inference.  An example of this is:

We found increased activity in the PPC when participants moved toward a visual targets. Increased PPC activity is associated visual processing and thus is  related to the encoding of space in a visual extrinsic reference frame. This activity thus could be related to the encoding of space in a visual reference frame when making movements to visual targets. 

This isn't as bad as some others- maybe because it is based on my work 😉. However, there is still something wrong here the increase in PPC activity was found when moving to targets. This last statement links processes associated with the activation to the behaviour. In isolation, this is an error in logic because there is an assumption that the function associated with the activity is directly linked to the behaviour that caused the activity. Although if you look at the results I chose to present, there is nothing indicating that this is the case.

This argument is just as flawed as saying: Dogs likes ice cream, Marlo likes ice cream. Therefore, Marlo is a dog.


This example was taken from a lecture on fmri data analysis- Principles of fmri analysis 1