[jdietrich]

>Also, philosophy underpins science. Whenever a hypothesis is tested, there is are philosophically-grounded assumptions being made. The epistemological implications for any given scientific finding depend on the underlying philosophical framework being assumed.

P(A|B) = P(B|A)P(A)/P(B). To the extent that philosophy underpins science, it does so because scientists are bad at math.

[kuhewa]

Your comment appears to imply Bayes theorem is all you need for a scientific framework and outside of that is just inadequate mathematical know-how to deploy it. I find it amazing really, a quintessential straight-from-the-summit [1] HN comment for a couple reasons:

First, it appears to imply that such a probabilistic inductive approach to science would be free of any philosophical baggage or assumptions, when deploying Bayesianism requires an interpretation about what a probability itself is. Don't take it from me though, perhaps from Andrew Gelman, the guy who wrote the book on Bayesian data analysis [2,3].

Then, wrt the charge that those who do not use such an inductive approach (or outright rejected it, e.g. in favour of falsificationism) are bad at math. Which would include the statisticians who developed the null hypothesis significance testing framework that is still pretty dominant in science today: Jerzy Neyman, Egon Pearson, Ronald Fisher (who literally coined the term 'Bayesian') etc. There's a lot of criticism worth making about Fisher, but I'm not sure if anyone has called the guy that developed linear discriminant analysis bad at math before.

[1] https://www.smbc-comics.com/comic/2011-12-28

[2] Bayesian Data Analysis http://www.stat.columbia.edu/~gelman/book/

[3] Philosophy and the practice of Bayesian statistics http://www.stat.columbia.edu/~gelman/research/published/phil...

[jdietrich]

>First, it appears to imply that such a probabilistic inductive approach to science would be free of any philosophical baggage or assumptions

I'm arguing that philosophical baggage is irrelevant to the current practice of science, because the overwhelming majority of published papers have serious and obvious methodological deficiencies that we have collectively agreed to ignore. Science as practised today is a desperate struggle to demonstrate something (p ≤ 0.05) by any means necessary. Established statistical methods have become a means to conceal rather than illuminate. This isn't the fault of individual working scientists, but the fault of the basic information architecture of science and a ritualistic, cargo-cult approach to understanding data. Bayes theorem probably isn't all we need, but it is all we need to spark a scientific renaissance if only we would use the damned thing.

[RoyalHenOil]

Something to keep in mind is that science is not only practiced in cutthroat academic institutions. Arguably most of the science being practiced today is happening in R&D departments across the world, where there is a strong financial incentive to beware of false positives due to the expense of policy changes.

I worked in private agricultural research for many years, and the issues you describe here did not apply to us. A single instance of statistical significance was insufficient to make any assumptions or adjust any policies. Of course we would present those results if we got them, but nobody was getting excited about it until we got the same results consistently several more times, in several different locations, with several different research teams. Being the first to get a positive result was no more meaningful that being the 10th to get a positive result. Promotions were based on your ability to design and conduct solid research, regardless of the outcome.