Note to medical journalists: correlation does not prove causation

Some nice person sent me a link to this article earlier this week which appears in the UK’s broadsheet newspaper The Telegraph. I mention the fact that this publication is a ‘broadsheet’ because these larger format newspapers generally have a reputation of higher quality reporting compared to the ‘tabloids’.

Whether that true or not, I don’t know. What I do know is that the ‘story’ here is a familiar one. It actually appears to be about some research which links higher saturated fat with speedier mental decline during ageing, and monounsaturated fat with a protective effect. But the real story here, for me at least, is about how the reporter, Stephen Adams, at no point mentions that fact that this is an epidemiological study, and the limitations of this sort of research.

In short, epidemiological studies look at associations between things. However, just because one thing is associated with another, this doesn’t mean this one thing is causing the other. This is because of things called confounding factors – perhaps true causative factors that are associated with other factors that are ‘innocent bystanders’. For example, let’s saywe have research which links ownership of a television with an increased risk of heart disease. Do televisions cause heart disease? Probably not. It’s more likely that one of the real issues here is watching television (i.e. sedentary behaviour) which, of course, is associated with owning a television. Eating rubbish snack foods in front of the TV might be a contributing factor here too, of course.

So, getting back to the study reported in The Telegraph, what it showed is that saturated fat is associated with enhanced mental decline. However, let’s all remember that for the last 30-odd years saturated fat has been roundly demonised by the nutritional and medical establishment. At the same time, we’ve heard generally positive things about monounsaturated fat. Now imagine we have a generally health-conscious individual hearing this advice loudly and consistently. Chances are they’ll attempt to act on this advice by cutting back on saturated fat and boosting their monounsaturated fat at the same time. Out goes the eggs and beef and in comes the olive oil and avocado. But, remember, this is a health-conscious type of person, so there’s a good chance they’re not going to smoke, or drink too much, and may well be regularly active too.

On the other hand, someone who is, say, less concerned about their health and has antibodies to health advice might choose to ignore advice to cut back on saturated fat and load up on monounsaturated instead, and may also continue to spend more time in front of the TV, drinking, smoking and eating unhealthy snack foods.

So, how do we know if the mental changes were caused by different dietary fat levels or other factor? We don’t. In an attempt to take account of other relevant factors, researchers can ‘control for’ confounding factors in their analyses. The researchers involved in the fat/brain function study did this, but this practice is essentially a very imprecise science: At the end we still end up with data that really doesn’t tell us anything at all. That’s just how epidemiology is, I’m afraid: correlation does not prove causation.

Now, as I say, Stephen Adams does not mention this major deficiency of epidemiological evidence, and his short story there gives, I think, the unwary reader with the impression that the findings are much stronger than they are in reality.

Not so long ago, this would perhaps go relatively unnoticed and unremarked upon. But these days, it appears there are growing numbers of people who are well versed in the scientific method, and who don’t mind pointing out the error of someone else’s ways. Many newspapers now allow readers to make comments. I thought I’d take a few moments to read the comments after Mr Adams’ piece. I was pleased to find that at least some commenters pointed out, fairly and squarely, the limitations of epidemiological evidence. Here’s one of the comments that neatly summed it up, I think.

5 Responses to Note to medical journalists: correlation does not prove causation

  1. chmeee 24 May 2012 at 12:25 am #

    Two things. I sometimes get into discussions at work when articles such as these appear. I tend to say that in reports of studies such as this, words and phrases to look out for include such as ‘Could be implicated in’, ‘May’, Might’, ‘it is possible that…..’, and so on with ‘Could’ and ‘May’ being the key ones to look for. If the reporter has been honest of course….

    Corrlelation is not causation is often too much of a mouthful for many. The example I always give is that I can correlate my hangover with morning and sunrise with around 100% accuracy but that we all know it was the extra wine / beer I drank last night that caused it, not sunrise. Everyone gets that ! 🙂

  2. Neal Matheson 24 May 2012 at 8:33 am #

    Man this stuff makes me cross! When are we going to see some news articles in the amount of sugar in baby foods. Or the sugars in foods deliberately aimed a children you know those obese diabetic porkers that constitute the next generation.
    Of course the truth is probably 100 percent the reverse of what he has written Did he account for the rise in dementia and altzheimers with the decrease in sat fat consumption?

  3. Nancy Bruning 25 May 2012 at 7:50 pm #

    My favorite example of correlation not being causation is umbrellas and rain. You look out the window, see everyone is using umbrellas and voila! Umbrellas cause rain to fall.

  4. Rolf Mueller 26 May 2012 at 1:10 am #

    One of the saddening other ‘oversights’ is the fact that all saturated fats are thrown into the same pot, notwithstanding the many reported benefits of certain saturated fats i.e. Coconut oil etc. This leaves the less discerning reader of this kind of reporting utterly convinced that ALL saturated fats are bad. Far from it!


  1. Note to medical researchers: correlation does not prove causation | Dr Briffa's Blog - A Good Look at Good Health - 7 September 2012

    […] up on processed food or being very sedentary. I actually wrote about these issues most recently here, in a post entitled ‘Note to medical journalists: correlation does not prove causation’. It […]

Leave a Reply