The problem with comparing faces to other stimuli May 5, 2007Posted by Johan in Face Perception, Neuroscience, Social Neuroscience.
The figure above is an example of the complete opposite of the stimuli you normally see in studies on the neural basis of face perception – the pictures of the faces at the top are quite different from each other in terms of size and angle, while the car pictures at the bottom are taken in virtually identical settings.
In most face processing studies, the faces are presented much like the cars – from the front, straight ahead, at the same size – like passport pictures. In order to get a reasonable comparison, objects from a non-face category are often used as controls, for instance tools, houses, or (as above) cars. But as a rule, pictures of these non-face stimuli tend to display a lot more perceptual variance. So in effect, many studies have compared faces presented at the same angle and size to a control category, presented with widely differing angle and size. If you then find that based on your study, certain characteristics appear for the face group but not the control group, you can no longer be sure that the results are not caused by differing amounts of what has been termed interstimulus perceptual variance, rather than some feature that is intrinsic to faces.
The figure above comes from a paper by Thierry, Martin, Downing and Pegna (2007), who identified this problem and tried to control for it. In studies using Event-Related Potentials (ERP – an EEG technique) and in some cases MEG, a waveform can be produced by averaging all trials in a condition. It has been known for some time that some of the features of this waveform only appear in response to certain situations. A number of studies have shown that a negative wavepeak at 170 ms after stimulus presentation, the N170, appears to be specific to faces. The figure from Thierry et al (2007) below is a good example of what an ERP waveform looks like.
To see whether the N170 is an effect of perceptual variance or faces, Thierry et al (2007) produced high- and low-variance groups of faces and cars, for a total of 4 groups of stimuli. EEG was recorded while the participant were shown the stimuli in random order. To maintain attention, the participants performed a one-back task, where they had to press a button if a stimulus was presented twice.
Thierry et al (2007) found a significant main effect of perceptual variance, meaning that across both faces and cars, the N170 was greater in the low variance condition than in the high-variance condition. There was no significant difference in N170 amplitude between faces and cars – but the contrast most resembling that used in previous studies, comparing high-variance cars to low-variance faces, was significant.
Interestingly, Thierry et al (2007) also found that two microstates in the P1 range (see the figure above), P1f and P1c, did differ significantly between cars and faces, but not between the high- and low- perceptual variance condition. The paper also contains some additional research to rule out other possible explanations for their results, which I won’t go into here.
So on the downside, the commonly accepted truth that the N170 is face-specific may be wrong. On the upside, Thierry et al (2007) have identified two microstates that may prove more resistant to changes in method. While such surprising results need replication, it is quite striking that a face-specific component appears so early. By comparison, a stimulus that is presented for 140 ms (the far end of where P1 appears) would be considered borderline subliminal. Or to take another example, perception experiments sometimes use stimulus presentations of less than 200 ms to ensure that the subject doesn’t have time to shift their gaze, as that’s about the minimum time needed for the eye to make a saccade. So in less time than what is needed to execute a saccade, the brain appears to be carrying out face-specific processing.
While N170 has been reasonably accepted up until now, the analogue finding in neuroimaging studies has been far more controversial. The conventiently-named Fusiform Face Area (FFA), located in the right fusiform gyrus on the underside of the brain, was originally identified and named by Kanwisher, who believes that the area is a modality-specific centre for processing of faces, and nothing else. She has come under fire from other investigators such as Tarr, Gauthier (and sometimes Haxby), who argue that the FFA might just as well deal with expert within-category discrimination, a type of processing that faces obviously depend on. There are a few similarities to the N170 situation here, and it certainly is the case that the same issues of failing to consider differences in perceptual variance apply to many fMRI studies on face processing. Still, it should be emphasised that the problem of differing perceptual variance that Thierry et al (2007) identified with the N170 study simply doesn’t apply to a lot of the fMRI face processing paradigms – for instance, many of Kanwisher’s experiments compare normal faces with scrambled faces. Nevertheless, it will be interesting to see if future investigators manage to pinpoint the origin of the P1 microstates to the FFA.
Update, June 1 2007: Unsurprisingly, there is a bit of controversy brewing regarding this paper. See the comments on the Neurocritic’s excellent post on the same paper to see what all the fuss is about. I will certainly keep an eye for the replies, if they make it into NN.
Thierry, G., Martin, C.D., Downing, P., & Pegna, A.J. (2007). Controlling for Interstimulus Perceptual Variance Abolishes N170 Face Selectivity. Nature Neuroscience, 10, 505-511.