Harvard ethics prof fabricated multiple behavioral science studies

Re:Harvard ethics prof fabricated multiple behavioral scienc... by UniverseHacker on Monday July 10, 2023 @10:01AM (#36666598)

I’ve found the opposite… because of the complexity and noise the social sciences are using more sophisticated statistical methods than the “hard” sciences. In grad school I took Bayesian probability theory classes from social scientists because my “hard” science peers were still using frequentist statistical tests, often applied inappropriately by just copying methods from past papers that also applied the methods inappropriately. The social sciences in general seem to be a decade or two ahead in adopting improved methods, and researchers spend a lot more time understanding these methods, writing custom software and models, etc.

IMO, mathematical and computational “literacy” is particularly low in the medical and life sciences compared to the social sciences, physics and chemistry are not quite as bad.

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by TheMagicHorsey on Monday July 10, 2023 @10:05AM (#36666647)

I think it really varies department to department. Was your program in an American university?

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by UniverseHacker on Monday July 10, 2023 @10:07AM (#36666670)

An R1 research university in the USA

Look at the most modern open source R and Python packages for statistical inference… probably 90 percent are developed and maintained by social science research labs

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by gowld on Monday July 10, 2023 @11:15AM (#36667653)

Indeed, social sciences use the most sophisticated alchemy to product "knowledge" from ignorance. :-)

Bayesian statistics is great, you can get any posterior you want by choosing good priors.

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by wizzwizz4 on Monday July 10, 2023 @11:33AM (#36667921)

But papers don't publish priors. They publish evidence.

"This study provides 5dB of evidence that the pigs do not reactionless flight capabilities, and 300dB of evidence that I don't know how physics works. However, according to my priors, the flying pigs hypothesis still comes out ahead." said no study ever.

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by mike_hearn on Monday July 10, 2023 @03:08PM (#36671409)

In reality a lot of them publish priors dressed up to look like evidence. As an example see Flaxman2020.

It's best to model journals as magazines that publish anything that vaguely resembles science (has equations, charts, citations, claims of data). It doesn't have to actually be science. If you want to get a paper published it's totally normal to write 10,000 lines of C or FORTRAN, claim it models some natural process, publish a paper with the graphs it outputs and call it a day. The code isn't available. The input data isn't available. The assumptions aren't documented properly or at all. And the outputs are unvalidated, therefore not even laying the foundations of being evidence of anything. Doesn't matter, it'll get published, in some fields this is the vast majority of what gets published (e.g. epidemiology).

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by UniverseHacker on Monday July 10, 2023 @03:19PM (#36671550)

An informative prior requires good data… it is a misnomer that Bayesians just “make up” priors, they are generally the result of a past experiment. Indeed a prior can encode just a hunch or intuition if you have nothing else, but in practice that will be nearly identical to a uniform prior.

Personally, I always use a uniform prior unless I am able to cite or present the data that led to something else.

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by chaxor on Monday July 10, 2023 @11:35AM (#36667950)

More sophisticated does not mean more correct

[ Parent | on HN ]

Re:Harvard ethics prof fabricated multiple behavioral scienc... by pclmulqdq on Monday July 10, 2023 @11:41AM (#36668033)

The best statisticians I know are economists, but they don't know how to run a study to avoid the influence of outside variables.

The best experimenters I know in terms of monitoring and controlling external variables are chemists, but sometimes they just use the wrong statistical test.

Biologists seem to generally be bad at both, and physicists seem to generally be good at both. I honestly think this is about the culture of the fields rather than knowledge - they all start as good scientists and just unlearn what they don't "need."

[ Parent | on HN ]