Indeed. This type of study, while really popular in the press, is often horribly confounded and hard to control for. I think the hacker news audience mostly knows to be skeptical, so I don't know why it was submitted.

I'll use this post as an excuse to mention the Lothian Birth Cohort. A really cool dataset of IQ test results at age 11 for (almost) all Scottish kids born in 1921 and 1936. While IQ tests are flawed in many ways, these data offer some means of control for follow up studies on these individuals. Many apparent effects disappear after controlling for age 11 IQ, for example: drinking coffee on intelligence — http://www.ncbi.nlm.nih.gov/pubmed/19995882 — nicely demonstrating the confounding problems affecting most studies.

Edit in response to sibling comment: yes, studies do try to control for these things. But it's hard, and often not really possible. This particular study might be ok, but we're not going to tell from a CNN article. We're also used to enough contradictory and over-hyped results in the popular press to know not to trust journalists to be a good filter.

In response to your edit: I think you're saying something very different than what your sibling responded to. The parent's claim was "They didn't correct for X"; your claim is "They tried to correct for X, but we have no way of knowing if they did it right."

The difference is that what you're saying is true of any study. There are always obvious things to control for, and less-obvious ones, and the authors always try and sometimes they do it right and sometimes they don't and there's never any way to tell for sure without redoing the study[0].

The fact that a layman reading a mainstream article about the study can come up with a possible confounding factor doesn't imply much if anything about its possible veracity. We should be equally suspicious of all such claims, not only those which happen to have flaws which are apparent to us.

[0] If the study contains a basic statistical error, redoing it might just be a matter of reinterpreting the original data; still counts.

Like almost everyone else who will comment here, I haven't read the study, but studies normally statistically control for these sorts of things.

No, they normally never control for anything that smells of class (like income) because if they do, the health effect disappears. There was a good recent case or rerunning a study and controlling for class, thereby eliminating an effect (I think it was about potatoes and weight gain or something.) Don't have access to it here, but I'll try to post it when I get home.

IMO 95% of the health studies that catch fire in the popular press are repeated rediscoveries of class (smoking, fast food and types of food, etc.), or the qualities of illness (people who sit more die earlier, people who are depressed die earlier, etc.)

They did mention controlling for smoking, though, because they found that it counted against them.

Thanks for that. It's a little annoying that every time a study like this comes up, somebody jumps in with "Well, did they control for [obvious thing to control for]?"

You don't become a scientist because you're lazy and stupid. If you're reading about a study and think there's an obvious reason it's inaccurate, you are almost certainly wrong. Not that the study is not inaccurate -- many, perhaps most are -- but the reason is definitely not obvious. It's something that the authors, people who have spent a lot more time thinking about it than you, and who are probably smarter and better educated to boot, never figured out.

Guess what: it looks like the authors didn't control for income (or even just whether the person has a job!) or intelligence.

I just read the paper; there are many diet and exercise-related controls, and that's it as far as I can see. I totally agree with the OP that being the kind of person who might drink coffee probably has a confounding effect on health. I can't even think of how to try to control it properly.

I don't understand how you get from "most studies are inaccurate" to "but you shouldn't assume the ones that get reported on in the media are inaccurate in a way that we can detect". Why not? People who are performing studies that get picked up by popular media are the least likely people to be performing their work rigorously, and the most likely to be choosing a sensational hypothesis that's difficult to measure correctly.

I agree that we might eventually get to a point where the amount of skepticism and cynicism about scientific results vastly outweighs the number of papers published with poor experimental design or lack of real statistical significance. But I don't think we're currently anywhere close!

You make some good points, so I hope this doesn't seem dismissive-- my claim is not that the study is right. My claim is simply that the editors of the NEJM decided to publish the study having read it in greater depth than any of us here would care to, so if I read the CNN article and think there's an obvious reason it's not true, I can be pretty sure I'm wrong.

Now that being said, I think we're talking at cross purposes. The study found a link between coffee drinknig and mortality; I misread the GGP as being critical of that link, where it's actually just a hypothesis which would explain it. So none of this is really relevant :P

I read the "methods" section of the study, and saw no reference to controlling for occupation nor socioeconomic status (as opposed to marital status, self-reported health, and other things that are listed).

However, I only read it briefly, so I may have missed something.

Have you done any blue collar work? Depending on situation, many folks carry large thermos of coffee with them, and between morning and afternoon breaks, and lunch, quite a bit of coffee is consumed. Yes, this is my experience (and that isn't scientific), but simply stating that only office dwellers drink a lot of coffee isn't exactly scientific either.

That said, I agree with the sentiment here (and of the siblings) that controlling this sort of study is rather difficult.

It would be interesting if the study showed this, but since the article doesn't mention it, that's a bit of an assumption you're making here.

I personally know a couple of heavy coffee drinkers, some of them should be retired but still work blue collar jobs (one is a roofer, another is a general contractor). I doubt they would agree with you on this one :)

Not to mention you're probably free of any major cardiovascular problems if you can tolerate that much caffeine.