Guest Blog for January 2014: “On Correlations and Bias” by Dr. Brian Hughes

This month’s guest blog comes from Dr. Brian Hughes of NUI Galway who runs The Science Bit blog which aims “to demystify science and scientists, to show how  scientific ideas can be used in everyday contexts, to highlight abuses of scientific jargon and attempts to exploit public confusion about science, and to record examples of how people in public life (who should know better) can often struggle with scientific concepts.” So, bearing that in mind, Brian tackles the ideas of causality and causation along with the tricky relationships between them.


On Correlations and Bias

I’ve said it before, and I’ll probably end up saying it again: the correlation fallacy (wherein observers attribute causal effect to a coincidence) is not as simple as it is often portrayed. For one thing, the mantra that “correlation does not imply causation” tends to draw attention from the fact that correlation is, after all, a necessary precondition for causation: causality causes, and is therefore correlated with, correlation. The problem is that while a correlation between two variables does not necessarily mean that one caused the other — in other words, while correlation does not ‘cause’ causation — it does elevate the logical probability that causality is in play.

Correlation fallacies of various kinds generate regular confusion among media commentators tasked with interpreting the conclusions of research. One can even recall the time the Daily Mail sent to press an entire article referring to a “casual” link between pornography and crime. For various methodological reasons, correlation-based inference is very common in neuroscience, making the associated problems with media reporting all the more acute: think Susan Greenfield, and other mass panics about internet use and the brain.

One aspect of this has always intrigued me. The inherent ambiguity of a correlation means that interpreting it becomes something like a projective test of personality, wherein the conclusions drawn reflect the scientist’s own dispositions and value-systems rather than the reality under empirical scrutiny. In short, the meaning you extract from a correlation can say as much about you as it does about the data.

Therefore, when neuroscientists study a topic like religion or religiosity, and when journalists report on it in the media, it is tempting to consider whether people’s personal values can be inferred from the way results are explained.

Take this recent study as reported by the UK’s Independent newspaper: Spiritual or religious activity may protect against depression by thickening brain cortex. The original research, which appeared in JAMA Psychiatry, found a correlation between scores on a religiosity questionnaire and thickness of the cortex in the left and right parietal and occipital regions, the mesial frontal lobe of the right hemisphere, and the cuneus and precuneus in the left hemisphere (regions in which thinning has previously been linked with depression risk). My first impression of such a finding is that the biological variable must underpin the psychological one: that cortical thickness reflects a disposition to think in a particular way, rather than a thinking style predisposes a person to develop cortical thickness.

Perhaps this reflects my bias in favour of biological determinism. But the study also found that the correlation was especially pronounced among persons who were genetically predisposed toward depression, defined as having ancestors who themselves were clinically depressed. Further, participants’ current religious behaviours — namely, their church attendance habits — were found to be unrelated to cortical thickness. To my mind, all this reinforces my confidence that the correlation reflects a biological cause to a psychological outcome. After all, one’s genetic predisposition (a) is inherently biological, and (b) is present from birth — it can hardly, therefore, be said to “be caused by” one’s later development of religiosity.

However, the Independent seemed to give precedence to the idea that it was religiosity that caused these people’s cortices to thicken:

Meditation and other such spiritual or religious practices may help prevent depression by prompting a thickening of the brain cortex, researchers have said.

…a recent study by the same team found that spirituality can protect against [depression], particularly in those who are predisposed…Following up on the previous findings, the researchers, led by Lisa Miller, professor and director of Clinical Psychology and director of the Spirituality Mind Body Institute at Teachers College, Columbia University, think they may have discovered why. The new study found that those who placed a high importance in religion or spirituality had thicker cortices than those who did not.

The same slant was offered by the Daily Mail:

[The study] suggests being religious enhances the brain’s resilience against depression in a physical way…in welcome news for the faithful, the researchers think that this thickening could also help to stave off depression…The finding suggests that being religious can change the brain in a physical way and this change could be protective against depression.

And by Fox News:

 For people at high risk of depression because of a family history, spirituality may offer some protection for the brain, a new study hints.

A swift re-cap might be useful. The study found that persons with clinical depression had children or grandchildren in whom an association between religiosity scores and cortical thickness was pronounced. The narrative offered by the media offers the same time sequence essentially in reverse: today’s religiosity causes cortical thickness, and today’s cortical thickness leads to depression (as experienced by your ancestors!).

But the media can hardly be blamed for offering the religiosity-leading-to-brain-thickness-buffering-against-depression interpretation. The authors of the study pushed it hard. Here is one of them, quoted by the Independent:

“The new study links this extremely large protective benefit of spirituality or religion to previous studies which identified large expanses of cortical thinning in specific regions of the brain…”

Note use of the word “benefit.” The author doesn’t emphasise that in this case the measure of “benefit” was derived from events that occurred before the participants were even born (i.e., from their parents’ or grandparents’ mental health histories).

And here is the same author, quoted by the Mail (my emphases added):

“Our beliefs and our moods are reflected in our brain and with new imaging techniques we can begin to see this…The brain is an extraordinary organ. It not only controls, but is controlled by our moods.”

To me, it seems a little strange that a researcher using neuroimaging would take the view that the brain “reflects” and “is controlled by” our emotions, rather than the more mechanically straightforward proposition that serves as the very rationale for neuroimaging itself: namely, that brain activity underlies such psychological experiences. The claim that the brain is controlled by emotions (rather than the other way around) epitomises dualism, and offers something of a Deus Ex Machina worldview.

The above statements appeared in the fluid context of media reporting. What about the peer-reviewed confines of scientific journal publication? Well, in JAMA Psychiatry, the authors again share the interpretation that religiosity is likely to be causal, with cortical thickening the result:

The importance of religion or spirituality appears to confer a neuroanatomical resilience in those who are otherwise predisposed to developing depressive illness.

This is despite the fact that their own data show strong effects for genetic predisposition, which, as I’ve suggested, implies that relevant biological factors temporally precede the psychological and mental health outcomes, thereby meeting the fundamental requirement for such factors to be seen as causal. In fact, an equally valid phrasing would have been:

Neuroanatomical resilience in those who are otherwise predisposed to developing depressive illness appears to confer a tendency toward religion or spirituality.

But that equally belies that fact that the finding is correlational.

It can be noted that all three co-authors have individually received financial support from the Templeton Foundation. This group famously promotes research sympathetic to religion, and is thus boycotted by a number of scientists who wish to separate pro-religious bias from the scientific process. Oddly, for the present study, the Templeton monies were not for data collection (which was funded by the NIMH), but for “analyses and manuscript preparation.”

In the end, the findings are truly ambiguous and so open to both interpretations: of religiosity leading to cortical thickening (the Daily Mail, the Independent, and Fox News) and of cortical thickening leading to religiosity (me). The pre-existence of objectively identifiable genetic predictors lend weight to the latter but, I appreciate, the processes involved could be iterative. Also, it is logically defensible to use grandparental depression status as a marker of future depression risk in a grandchild (although the predictive validity of such indices must be very tenuous, especially in a target group of n = 67).

Simply, my aim here is to show that, given the ambiguous nature of correlation, whatever interpretation arises will owe more to the reader’s own value-judgments than to data.

The obvious recommendation is for some kind of longitudinal study, which the authors duly note, and which many media outlets repeated. This is a good lesson. The other is to not underestimate the subtlety of the correlation fallacy. All that’s needed to cause confusion is some kind of bias.

And in media reporting of science, even (or especially) neuroscience, there are plenty of those.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s