Snow Leopard said:
What are the common biases?
Last installment, but some of the real biggies here....
Multiple Comparisons and ROI’s
Another problem that arises in functional neuroimaging research is that fMRI collects so much data. No-one can interpret it all. Researchers do a few things to correct for this. First, they use
statistical corrections (a bit like Bonferroni if you know that), so that only activation in areas of the brain that are marked and highly statistically reliable get reported.
But another thing happens sometimes too. The researchers focus on a “
region of interest” (ROI), and look only at that. This is helpful statistically, as you don’t produce as many statistical results and you don’t have to statistically correct to such a degree, so you’re more like to find significant stuff, at least in your region of interest (ROI). But there’s a cost: you simply ignore any differences in other areas – handy for results that might be troublesome for you to explain or might go against your hypothesis.
Most studies do both a whole brain and a ROI analysis, so you can see the pro's and cons’ of both – but watch for those that do just the ROI one.
The tricky business of hypothesis testing.
A bigger problem: you can use functional neuroimaging to test a hypothesis about
brain activity But you can’t use it to test a
psychological hypothesis.
An example of this conundrum was published a few years ago. It was a case study of someone with “hysterical” leg weakness (their word not mine!). Some really clever researchers (who should have known better) found heightened activation in the anterior cingulate when this person was asked to try and move her bad leg. Since the anterior cingulate has been associated with effortful control, they argued this activation demonstrated an active effort to inhibit leg movement. Hey presto – hysteria confirmed!
But this type of activation is also observed whenever a person is faced with
an especially challenging task – like, er, perhaps moving a leg that is not functioning properly?
The problem: you’re
jumping levels of description. They don’t map one-to-one onto each other. There could be multiple psychological explanations for such activation differences. Its the problem of reverse inference all over again.
The confusion of cause and effect.
A huge problem with interpretation of functional neuroimaging studies of special populations – especially in the media – is that anything observed is often assumed to be the root cause of whatever ails that population.
Take the example of the amygdala in psychopaths. There have been a few studies showing that people who meet the criteria for psychopathy respond differently to pictures of people in pain or being hurt than non-psychopathic prisoners. The amygdala, a region known to be involved in emotional processing, is underactivated in these individuals during this task (usually in these studies, it’s a subtraction design so they compare activation in this task with a “control” or “baseline” task). It has been claimed that such studies demonstrate that psychopaths have a brain aberration which renders them unable to empathise.
But this makes no sense. Psychopaths, by definition, do not feel distress when other people are hurt. This is how the condition is defined. So its not at all surprising that brain regions such as the amygdala, which are known to be activated when people feel distress, are not as highly activated in these people when viewing others' pain. It’s a no-brainer (sorry, couldn’t resist
). In this study
, the neural correlates are just that, neuronal correlates, not causes.
The best way to think of functional neuroimaging measures – especially ones that compare activation in at least two active tasks – is as
another type of response, in the same general group as reaction time measures, self-report scales, etc.. Okay, they don’t involve action in the real world, and they’re not controllable by the person in quite the same way, but when it comes to interpretation, they're responses. They can be really useful measures in this way. They are a lot richer than, say response times, because they don’t just have one dimension – either fast or slow – they also have a second dimension: location. Plus it’s a
new measure, and we could do with some better ones (self-report, well we know that one sucks). But its still pretty much a response.
Psychopaths don’t have brains that are “hard wired” to render them unempathetic (at least as far as we know). They are simply unempathetic, and their brain responses, like their verbal responses and their behaviour, are consistent with this. The neuroimaging data have no privileged status (see below). If you wanna test the causal role of the amygdala in psychopathy, you need a totally different method. For example, take a group of patients with amygdala dysfunction/damage due to a known condition (or damage to other regions implicated, like the orbitofrontal cortex). Are these people psychopathic? If so, it might suggest a causal role. But it turns out they're not.
So if someone’s proposing to use functional neuroimaging to study a special population (emphasis on functional here, structural imaging is another thing),
don’t think biomedical study, think psychological study. Because most of the time, that’s exactly what it is. The brain activation is used as a measure of response to some set of stimuli or scenarios and the question is how this response differs from controls. The whole kit and caboodle is interpreted in psychological terms.
The “privileged status” fallacy
This one is kind of related to the last – it’s the idea that if we can describe behaviour in neuronal terminology, the implications somehow change. This is the concern I was talking about on the main thread when people confuse levels. Its illustrated nicely in the psychopath example above. Somehow the psychopaths don’t seem as responsible for their behaviour any more, now we know it goes on in their brain. But think about it: Of course if goes on in their brain – where else would it be?
I worry about this with ME because people think
fMRI findings will demonstrate they have a real physical illness. This is totally incorrect. As I mentioned above, fMRI has been used at length to study “hysteria”, and “conversion disorder” and the purpose has always been to show how “psychological” the condition really is. In my view, none of these studies is persuasive (lots of bad reverse inferencing, poor control groups and a selective discussion of “compatible” findings).
But the point is, be very, very wary.