Nuts and longevity: correlation isn't causation

For most people, their death is an unfortunate career move. It's been observed that children are born at most 10 months after their fathers' death and the IRS only collects the taxes for about 12 months after the death. Richard Feynman pointed out that he wouldn't like to die twice because it's so boring.



The eternal president and the dear leader in the representative halls of North Korea.

But there are exceptions. Kim Il Sung (1912-1994) was kind of promised to become an eternal president after he dies; the law was adopted in 1998. However, the Korean leader still liked some old-fashioned benchmarks to measure when the life ends, too. We just learned that he ordered his doctors to prolong his life to 120 years. They failed but 82 years isn't bad.




Maybe the doctors should tell him to eat nuts every day. According to a Harvard study in New England Journal of Medicine,

Association of Nut Consumption with Total and Cause-Specific Mortality (Ying Bao et al.)

(see also a NEJM quick take animation), the probability of a heart disease-related death during a 30-year-long period dropped by 29 percent and the probability of a cancer death dropped by 11 percent among those who were eating nuts more or less every day – leading to something like a 20-percent average from these two causes.




It seems to me that this difference is huge, given the large sample – 119,000 people were tracked. Up to nonlinearities (and the neglected other causes of death than these two), you could say that the "rate of death" was lower by 20 percent as well which may mean that the life of nut eaters should be almost 15 years longer. This is a vastly higher difference than the statistical fluctuation you expect from such a large sample.



Purely numerically, by a mindless calculation, I would bet that the statistical significance of this finding is much greater than 5 standard deviations. Nuts are probably good for you. But I still think that there is a potential catch:

Correlation isn't causation.

Well, more precisely, a statistically significant correlation almost certainly proves "some kind of causation" but it may be a different causation than one naively expects. As far as I understand, the people were not made to change their habits. Why is there this correlation between the reduced death rate and the consumption of nuts?

Nuts are probably healthy.

But I believe that there is an alternative explanation that was arguably not given a sufficient attention:

Healthier people love to eat nuts more than the less healthy people do.

This hypothetical relationship may be interpreted as a categorization of people's DNAs that affects their whole lives; but it may also be interpreted as a time-dependent quantity. In the first case, the "genetically more long-lived" people just have the tastes that attract them to nuts. In the second case, it doesn't even have to be about genetics, but it may be that when the people lose their attraction to nuts, it is a symptom of their getting ill or less immune which makes an early death more likely.

Moreover, I have a feeling that this concern applies to much of the medical research. The actual causal relationship may often be the opposite one to the causal relationship that is being immediately extracted. In some cases, the right explanation of the correlation between A and B may be more complex – there may be another factor, C, that causes both A and B (or makes them more likely) which is also enough for a correlation between A and B.

There are TRF readers who are close to medicine or life sciences. Let me ask you: Do the researchers actually realize this potential loophole? Is this loophole being ruled out in much of the research that uses a correlation to prove a causal relationship?

Thanks.



How Jim Parsons would behave if he were not expected to resemble a string theorist.

Along with biases and a lower required statistical significance, this is one of the major reasons why I intuitively tend to find much of the medical research saying "XY is good or bad for you" somewhat untrustworthy but maybe I should be more welcoming.