4 ms·

Failures in statistical methods are not covered by 'everything has flaws'. Those are fatal, existential flaws. Something that is not likely true (though it is e

Failures in statistical methods are not covered by 'everything has flaws'. Those are fatal, existential flaws. Something that is not likely true (though it is erroneously presented as likely of being true) are likely false.

mmooss 2y ago
Working backwards ... > Something that is not likely true (though it is erroneously presented as likely of being true) are likely false. That doesn't make sense to me. Because something isn't proven here, that doesn't make it more likely to be false; it's just uncertain. Poor evidence is not evidence either way. To say it's false, it would need to be proven false, with good evidence. If I assert, 'the universe is expanding because Pluto is further away today than yesterday', my argument wouldn't support the claim but that doesn't logically imply that the universe is not expanding. > Failures in statistical methods are not covered by 'everything has flaws'. Those are fatal, existential flaws. Why are errors in statistical methods -- if they exist here: we have a hot take by a random, anonymous Internet commenter (using a new account) against scientists who spent a long time on this work, and put their names and reputations on it -- somehow more fatal than other errors? For example, some statistical errors lead to weaker results, but results nonetheless. Some lead to results with a somewhat different meaning. (Some lead to stronger results.) We need to deal with imperfect information all the time and find value in it, or we would have almost no information. I spent last week solving a problem with several routers interacting; I had some clear data, some unreliable information, and some black-hole uncertainty; I had to work with what I had and solve the problem. The idea that science is exempt from that is a fantasy of non-scientists, of the religion of science.

mmooss 2y ago

Working backwards ... > Something that is not likely true (though it is erroneously presented as likely of being true) are likely false. That doesn't make sense to me. Because something isn't proven here, that doesn't make it more likely to be false; it's just uncertain. Poor evidence is not evidence either way. To say it's false, it would need to be proven false, with good evidence. If I assert, 'the universe is expanding because Pluto is further away today than yesterday', my argument wouldn't support the claim but that doesn't logically imply that the universe is not expanding. > Failures in statistical methods are not covered by 'everything has flaws'. Those are fatal, existential flaws. Why are errors in statistical methods -- if they exist here: we have a hot take by a random, anonymous Internet commenter (using a new account) against scientists who spent a long time on this work, and put their names and reputations on it -- somehow more fatal than other errors? For example, some statistical errors lead to weaker results, but results nonetheless. Some lead to results with a somewhat different meaning. (Some lead to stronger results.) We need to deal with imperfect information all the time and find value in it, or we would have almost no information. I spent last week solving a problem with several routers interacting; I had some clear data, some unreliable information, and some black-hole uncertainty; I had to work with what I had and solve the problem. The idea that science is exempt from that is a fantasy of non-scientists, of the religion of science.