How to Make Sense of Contradictory Science Papers
nautil.us
nautil.us
1) the conclusion has been simplified to the point where the meaning is lost (e.g. "coffee is good for you!" vs "black coffee is marginally better than soda, red bull, highly processed fruit juice and all the other junk people typically drink." This is very, very common)
2) the study has significant limitations (e.g. we just looked at what people ate and how healthy they were, and tried our best to identify and account for all the many confounding factors after the fact. Epidemiological studies can be useful to identify potential areas of further research but are usually not very strong evidence by themselves)
3) the study was designed to come to a conclusion that a vested interest wanted (e.g. most people have different natural cholesterol "set points" given the same diet. "So if we look at the correlation between egg consumption and cholesterol in a small group, instead of the change in cholesterol for the same people eating eggs vs not eating eggs, the effect will be lost in the noise and we can make it seem like eggs don't raise your cholesterol significantly", gleefully thought the American Egg Board)
I'm a big fan of NutritionFacts.org for breakdowns of all sorts of studies, pointing out all the limitations and figuring out what the studies actually say.
e.g., the way p-values are used to make crisp binary decisions via arbitrary thresholds, rather than simply mentioned and left to the reader to interpret their meaning as an exercise in philosophy.
I think this needs to be clarified to a large section of reddit and hn members more than any place else on the internet.
Skepticism is fine, and critical thinking should always be on with no exceptions, but HN is more prone to armchair quarterback bleeding edge science with no qualifications to do so (and predictably comical results), rather than uncritically accept results.
I myself am more likely to do the former than the latter.
That's exactly what the article is about. It's fine for scientists to propose weird and outlandish theories with just enough analytical duct tape to ensure they aren't completely absurd. But those theories are meant to be analyzed more deeply and critically before anything is done with them, not just blindly assumed to be true because somebody published a paper on it and nobody refuted it yet.
There's a lot of statistical tricks that are tough to detect. Like is your sample truly random and representative? How could you even prove that? Like the rat part study purporting to show that drugs are only really dangerously addictive if you're also socially isolated and miserable. It's cool, and I kind of want it to be true, but it seems it's actually highly dependent on exactly what kind of rats you have and where you got them. Similarly, most of the time when you analyze a bunch of statistics, you're looking for a conclusion like that it's 95% likely that A is caused by B. That's only 1 in 20 though, so you should actually expect that if you do the same check for B through U, and none of them really cause A, you will falsely find that one of those letters causes A. But nobody published a paper for checking C through U, so you can't tell.
You should remain skeptical, you should question results, and you should be capable of understanding the data and responses. You should understand uncertainty that's often unquantified yet still exists. You should apply critical thinking as well. You should also understand when you're way out of your domain and realm, potentially misunderstanding the situation or missing critical information in your analysis.
Understanding is what you need, not credentials.
Trying to understand is much better than not, IMO.
Got it.
You do not need a degree to gain that expertise, however.
Further, having the degree is not a reliable indicator of expertise, in my experience.