Hacker News new | ask | show | jobs
by alexose 1871 days ago
Great analysis. I wonder, is there a ratings agency for scientific studies? It would be neat to see a score associated with experimental design.
3 comments

You would find that most studies are complete garbage. Especially social science ones.
And the world needs this exposed... in aggregate!
And unfortunately, many nutrition studies are garbage as well.

From the link, first sentence:

> Next time you make a salad, you might want to consider adding mushrooms to it. That's because higher mushroom consumption is associated with a lower risk of cancer

Who writes garbage like this? They say it’s "linked" but draw causal conclusions anyways.

What do you think would be a correct statement?

"If you're the kind of person who adds mushrooms to their food, you might or might not have lower probability of getting cancer"?

It always drives me crazy, because I'm not sure if I'm that kind of person. Am I a person who doesn't smoke or who only doesn't because it's unhealthy? Am I a person who doesn't eat butter or just avoid it because it might be bad for me? I genuinely think I can't know.

> What do you think would be a correct statement?

"because this is not a properly designed study and there is so much chance that this is pure coincidence or related to a non-captured factor, we don't recommend you do anything about your diet. And you can skip reading this paper as well."

That's kind of the whole point - the study should explicitly state that the new findings (as of now) do not provide any diagnostic about your personal chances of getting cancer, and do not provide any dietary recommendation whatsoever.

It's an interesting finding that may motivate further, more specific studies that can result in something actionable, which has value for researchers in the field but (as of now) it should not be used by general public for analysing their diet.

This, is one brilliant idea folks. If it doesn't exist, someone should make it.
Very much want to see that. It's a project idea of mine.
In my opinion publication bias is a bigger problem than poor experimental design. Do you have opinions on how to fix that?
Register the experimental design before hand and a commitment to publish non-results. That’s the logistical piece. The harder one is the incentive structure in academia that focuses on publishing new/novel research. There should be similar rewards for doing good replication work on significantly important results (eg if everyone based their research on a study that hasn’t been replicated which happens with alarming regularity in the social sciences and psychology). High energy particle physics has this problem now too to some extent and they’re tackling it in other ways (you only have one LHC for example)
This should be possible. I think it is important for scientists to do both novel and replication studies.

One way to improve on the status quo would be to make students do both types of work (the point of a PhD is to demonstrate that you can do research, by doing it. So why not do both kinds). This actually could fit in nicely in the program: first/second years tend to be inexperienced anyway, so they could start by doing some replication work. Halfway through the degree, they switch to novel work.