3 ms·

> If you think that’s the stupidest thing you’ve ever heard, you can take it up with Sir Ronald Fisher, who actually made this argument with apparent sincerity

> If you think that’s the stupidest thing you’ve ever heard, you can take it up with Sir Ronald Fisher, who actually made this argument with apparent sincerity in a 1957 letter to the British Medical Journal. I mention this in case you didn’t already know what an ass he was. Not only did Fisher state this, and not only was he an ass, but it is almost embarrassing —it turns out he was partly right. You can read all about it in Pearl’s gem The Book of Why. In essence Fisher argued that there could be a confound between the neurochemical and behavioral propensity to smoke, and the risk of developing lung cancer. And strange as it seems to me, the DNA variant in CHRNA5 that is most highly associated with higher risk of nicotine addiction also modulates (independently) lung cancer risk. His principle was kinda-sorta defensible for an avid smoker. Mendelian randomization of people by their CHRNA5 gene status can indeed demonstrate that Fisher was wrong. The majority causal risk is in fact smoking.

npunt 3y ago
Interesting! Within almost every truth there lurks a valid but ultimately very minor counter-point. Sorta a tangent here, but it's annoying that this is the very thing that causes people biased to not believe to latch onto as evidence they're right. Once introduced, further reasoning or curiosity about the subject tends to stop before weighing matters of degree, which leads to missing the big picture. To generalize a lot, this is kinda where we are in the current meta of truth fracturing. Our biases drive us to discover nuances we grasp onto to avoid believing what we don't want to.
gweinberg 3y ago
The idea that there's some confounding factor isn't in itself implausible, but a factor of 16 increase in risk is huge. Of course, to really be sure you should also look at how lung cancer risk and smoking correlates between countries, and within a country over time. And you can segment the population by sex and ethnicity. Oh and also look for a dose-dependent effect, very important. Heavier smokers should be at a higher risk. Of course, the link between cancer is smoking is even stronger if some of the "non-smokers" were exposed to an increased level of cigarette smoke.
- robwwilliams 3y ago
  With Mendelian randomization a genotype can be used as an instrumental variable if it is causally linked to the trait (usually as part of a genome-wide “association” study)—lung cancer in this case. this can provide reasonable assurance of causality. See Book of Why.

npunt 3y ago

Interesting! Within almost every truth there lurks a valid but ultimately very minor counter-point. Sorta a tangent here, but it's annoying that this is the very thing that causes people biased to not believe to latch onto as evidence they're right. Once introduced, further reasoning or curiosity about the subject tends to stop before weighing matters of degree, which leads to missing the big picture. To generalize a lot, this is kinda where we are in the current meta of truth fracturing. Our biases drive us to discover nuances we grasp onto to avoid believing what we don't want to.

gweinberg 3y ago

The idea that there's some confounding factor isn't in itself implausible, but a factor of 16 increase in risk is huge. Of course, to really be sure you should also look at how lung cancer risk and smoking correlates between countries, and within a country over time. And you can segment the population by sex and ethnicity. Oh and also look for a dose-dependent effect, very important. Heavier smokers should be at a higher risk. Of course, the link between cancer is smoking is even stronger if some of the "non-smokers" were exposed to an increased level of cigarette smoke.

robwwilliams 3y ago
With Mendelian randomization a genotype can be used as an instrumental variable if it is causally linked to the trait (usually as part of a genome-wide “association” study)—lung cancer in this case. this can provide reasonable assurance of causality. See Book of Why.