What's going on with the x-axis?

relative risk in terms of a ratio: <1.0 means less, >1 means increased risk compared to the control group. E.g., one study showed that beef reduced risk of cancer by 50%, another showed that it increased risk by 500%

In the chart is beef causing cancer or preventing it?

Either, depending on which study you accept.

The question is the problem. The answer is not a yes/no, but a point on a scale of multiple dimensions. Most of these studies don’t answer the question asked here, but a completely different question. And also most of these studies don’t answer the same question. Heck, it’s even a good guess that none of these studies answers the question asked here, because it’s rather difficult to make a double blind test about such questions.

Yes.

This chart is the epitome of modern "science".

We need to tweak our process, because this is clearly not working.

What makes you think there are tweaks that would improve "this"? You might be startled to learn "this" (the scientific method) is the best possible way of discovery that humans know of.

You may be even more startled to learn the scientific method is clear as an abstract concept, you know, those circle diagrams with "observation -> research -> hypothesis -> experiment -> analysis -> conclusions", but the details are always in flux and vary between countries, universities, publications and in time.

And the details are absolutely crucial for the end result. You come in here acting like a religious defender of the scientific method against the evil of rejecting science, but I'm not that guy. I'm not rejecting science, I'm just pointing out the obvious: modern science is producing literally noise, as seen in this chart, because of misaligned incentives and a broken peer review process, among others.

Do you even know what the "replication crisis" is? Hints of it in this very chart. And if you do, it's completely disingenuous to come attacking me for even mentioning we have a problem. And if you don't, you're not informed well enough to speak about the scientific method.

The "scientific method" should be applied to the "scientific method". Meaning when you see evidence that falsifies it, seek the causes and correct them, don't dig in your heels and go religious.

Everyone knows about the replication crisis, and everyone knows the flaws in the incentive structures around publishing research, but none of that is what you referenced.

What we need are people with level heads making calm decisions, not bombastic polemics demanding the system get blown up.

What did I reference? It was a very short message, so probably you saw what you wanted to see, or what you could see. But all I said is that a lot of modern studies are noise like that, and it's actively harmful, and the process which produces these studies should be modified.

Modern researchers have insane incentives to churn out papers based on superficial or questionable data, without isolating variables. The result is predictably noise. Or even worse, noise shaped by cultural bias and prejudice, which has been a long term problem with the scientific method, because in the end, we can talk about facts and evidence and data, but the judge jury and executioner on what is a fact, evidence and data are people, and people are always subjective by definition. The best we can hope is to all be subjective in the same way, hoping our shared subjectivity is more objective taken together. That doesn't always hold up.

None of these studies are produced via the scientific method. They are looking at statistical correlation and plucking them out of thousands of complex, confounding, dynamic variables in a very chaotic environment.

What about the chart leads you to that conclusion?

So it’s obvious how a study would support the notion that X causes cancer, but how would a study support that X prevents cancer? Too many confounding variables methinks.

I'm not sure I see how one is easier than the other. In studies where they find that wine, say, causes cancer, they aren't looking at some molecular chain of events, they're simply doing a large (hopefully) study and looking for correlations, while attempting (hopefully) to factor out confounding variables and incorrect direction of causation.

In some of those studies, the wine drinkers get more cancer. Red dot to the right. In some they get less. Red dot to the left.

I would think the effects of confounding variables are reduced the larger the sample size, so you can say "everyone in group A does not drink wine, and everyone in group B does drink wine". Some from both groups will get cancer because of some confounding variable, but they would cancel each other out- a non-related cancer case in Group A would be cancelled out by a non-related cancer case in Group B.

Conversely, if you have a similar study aiming to prove that wine prevents cancer, how do you prove that wine is the reason a person did not get cancer? It seems much more difficult, because now you are saying X did NOT happen because of Y rather than Y caused X. I'm struggling to articulate this. Maybe you're right.

Well, two things:

First, a confounding variable doesn't work like that. It's a hidden variable that's linking that independent and dependent variables that you didn't account for. So, for instance, if you look at the correlation between ice cream and sunburns, you might see a strong correlation, and you might naively think that one causes the other if you don't account for the confounding variable of hot summer days.

In the case of population-wide studies on food, there is always the danger of confounding variables: if you are just comparing the wine drinkers vs non, who are these non-wine drinkers? What makes them non-wine drinkers? As a whole, do they tend to be more heath-conscious? Is that why they're less likely to get cancer? Etc etc.

Second, my point is that in both studies that show foods causing cancer vs those protecting against cancer, it's almost always populations studies. So in an ideal experiment, they have two identical populations, one of whom just happens to eat more chocolate. If that group gets more cancer they say they the chocolate is causing cancer. Similarly, if they get less cancer, they say it's protecting them from cancer.

You don't have to prove that individual X would have gotten cancer if it weren't for the chocolate. All you have to show is that the standard, base-line amount of cancer you'd see in any population is reduced in this group.

Lovely. I wish this had low-dose ionizing radiation on it as well to cover the radiation hormesis hypothesis.

https://en.wikipedia.org/wiki/Radiation_hormesis

There used to be a site with a handy database too https://web.archive.org/web/20121115101540/http://kill-or-cu...