Hacker News Reader

Comment by aldarion | Hacker News Reader

Didn't know that there was anyone arguing it doesn't, but... nice article.

Fisher, one of the great statisticians, argued that insufficient evidence existed to conclude with certainty that smoking causes cancer.

https://pubmed.ncbi.nlm.nih.gov/2000852/

After that, a number of well-designed experiments on rodents demonstrated that smoking causes (strong causation) cancer (in rodents) beyond a reasonable doubt (as in, if you disagree, you'd better bring some impeccable data).

Is this Fisher the one that proposed the "p<0.05" rule?

Yes... are you going to criticize him for that?

It's worth reading the whole backstory (start here: https://en.wikipedia.org/wiki/Ronald_Fisher)

A flawed genius

No, why should I criticize him for that? Btw, I just learned that he made the Iris dataset famous.

Many people criticize p=0.05, https://sites.uw.edu/stlab/2016/03/09/the-arbitrary-magic-of...

> No, why should I criticize him for that?

Because after that academia made it the gold standard of research in most disciplines, leading to what many describe as a "replication crisis" in social sciences.

But surely anyone criticizing him for that would have to prove they've already sufficiently criticized him for the whole eugenics thing.

you seem to adhere to the intellectual norms of the great leap forward rather than those of science or philosophy

in science nobody ever has to 'prove they've already sufficiently criticized' anyone or anything

they just provide evidence

enjoy your auto-da-fe but please do it somewhere else, only involving people who are into that kind of thing

I'm sorry you feel triggered in mentioning that the guy was a leader in the eugenics movement?

Judea Pearl wrote a book, Causality, introducing Bayesian networks to address the inadequacy, even refusal, of Fisher's statistics to model causes and effects. Here's a lecture he gave almost 30 years ago [1] that's entertaining and accessible. He mentions the tobacco issue and Fisher specifically.

[1] http://bayes.cs.ucla.edu/BOOK-2K/causality2-epilogue.pdf

Thanks. I've never been able to understand the math behind Pearl but I appreciated the little diagram in the Fisher smoking section that clearly lays out causality of models diagrammatically.

I think it was Pearl's Book of Why that went into far more detail about how it was decided that tobacco causes cancer from purely observational data. The technique was... relatively convoluted, and supposedly Fischer fought tooth and nail against Bayesian techniques required to adequately answer any such causal questions.

Fisher was a smoker too right? Pretty sure his bosses played a part too

he was a consultant for the tobacco industry. I think he may have raised some legitimate questions of what constitutes "sufficient evidence" for making large-scale public health policy. But yes, I think his position mostly came from his biases.

Reply on news.ycombinator.com