March 26th, 2009
This morning I tossed up a quick post about a recently published study which looked at consumption of meats and certain bad health outcomes (including death). Someone asked, and I paraphrase, Why should I care about this? It’s a correlational study – it can’t prove anything!
It’s a good question, and a fair one. Husband and I are drafting a response which will provide a summary and translation of that paper’s findings for my audience which may not be interested in tackling an academic article in its raw state. But before we can look at that, we need first to determine why we should listen to correlational studies, why we can find them so powerful that it becomes rational and justifiable to change how we live our lives based on them.
When the strength of a correlation is high, we can infer causation. This is how the field of epidemiology (the study of the spread of diseases and other health markers in populations) works. The strength of a correlation is measured against five markers:
-temporal precendence. The alleged cause comes before the effect in time.
-there is a graded response; the more of the alleged cause is present, the more of the measured effect is present.
-the relationship between alleged cause and the effect is consistent when observed in a variety of settings and with different populations.
-there is a plausible mechanism to explain the relationship.
-the risks of the effect happening to you are higher when the alleged cause is present than when it is not.
Based on those markers, we were able to infer that smoking causes cancer – BUT to date the research which has been performed on that link is correlational, NOT experimental. This means we have not proven that smoking causes cancer according to the strict “correlation is not causation” position – but of course this doesn’t mean you should all rush out and buy a pack of smokes. The strength of the correlation between smoking and cancer is authoritative. It’s just that the tools and applications of statistical correlations are much more sophisticated than they may initially seem.
Oh bother, I just found a little article online which says what I’ve just said… only with diagrams. Go here to see it.
Anyway… how this applies to the article I linked to is coming soon. I hope this begins to address concerns with the form of this paper, but it’s not the last we’ll have to say on the topic.
Thanks for reading!
Wow.
I have just found another hot-button: words being put in my mouth.
Please, if at all possible, don’t do that again. I shall endeavour to be clearer in my own expressions in return.
Quick clarification of what I was getting at, which was not even vaguely “why should I care?”. This is a study which says, basically, the people who eat less meat seem to live longer then people who eat more meat. Now there are a ton of things one can take from that, but all of those things must be taken with the understanding that “correlation does not necessarily equate to causation”.
Sorry. I though it was an accurate paraphrase, in that “Say it with me now… Correlation does not equate to causation” seemed to suggest that you felt the paper didn’t make a strong point (because it’s a correlation) and therefore, why should we be interesting in the findings? I won’t paraphrase you again. (Also, this may be irrelevant in your case, but I sometimes worry about quoting people on the front page because I assume people don’t comment thinking I’m going to then move their comment into a main post where I will offer a critique or something of it. I worry about embarrassing people. So by paraphrasing and extending the comment it served my literary purpose while protecting you. I try to be sensitive. But I see you did not like how I did that. Direct quotations only from now on.
)
The point of the stats post and what we’ll write more about today (hopefully, time willing), is that correlation can, when it is well done, allow us to meaningfully infer causation (smoking and cancer being, of course, the poster boy for this). Basically, saying correlation doesn’t equal causation is correct, yet not always accurate in practice. Do note the study authors make absolutely no claims about causation, not a one. We can’t prove to absolute scientific certainty that there is a link between consumption of certain meats and certain bad health outcomes… but we can show a freaking huge study over a huge span of time and say, we’ve met all the criteria for strong inference, and hey, we’re not 100% (and never will be because you can never perform experimental, lab research on this), but it’s pretty darned suggestive, no?
Dismissing r studies’ findings because they are not controlled laboratory experiments is a narrow conception of how research is done and how we come to know our world. It all comes down to the particular study and how well it is done, how much we can infer, is there complimentary research we can look at for confirmation or discomfirmation, etc. This study is one big piece of a larger puzzle about human health.
Also, I think I missed the part in the paper where they say eating meat causes AIDS (or if that was a joke it is too early in the morning for me to get it, apologies). Brain still foggy… did you mean interpretations in the popular press are taking this way too far and making wild assumptions based on it? If so, yeah, the state of science journalism sucks big time.
While I am (obviously) inclined to agree with the findings of this paper, sentences like “Subjects who consumed more red meat tended to be married, more likely of non-Hispanic white ethnicity, more likely a current smoker, have a higher body mass index, and have a higher daily intake of energy, total fat, and saturated fat, and they tended to have lower education and physical activity levels and lower fruit, vegetable, fiber, and vitamin supplement intakes”
makes me think that the people in the study who ate more red meat had less healthy lifestyles in general.
I thought that initially too, until I read through the full study, when i came to realize that the study groups are actually among the better off population sub-segments. This large group is largely white, well educated, tends to eat less meat overall compared to many other groups, tends to exercise more overall than many other groups, and tends to smoke less.
Even so, within this group of relatively benignly behaved individuals, clear differences are seen, and those differences are of a dose-response relationship (higher dose equals higher likelihood of outcome: cancer death, cardiovascular death, all-cause death). The authors went to great lengths to tease apart the differences related to smoking and to some lengths to tease apart the differences due to exercise.
@Zed: Agreed that correlation does not imply causation. But when you dig into the matter deeply, there are correlations, and there are correlations.
One historically interesting example may help. Florence Nightingale thought it was crucial that soldiers in the crimean war be given clean linens, clean dressings, and fresh air. She saw with her own eyes that this greatly reduced gangrenous post-operative infection. She also totally rejected the new idea that living agents referred to as “germs” by the scientist Louis Pasteur might be playing an important role. So, paradoxically, she notes a correlation between clean dressings and better infectious disease outcomes all the while rejecting the proposed cause, a cause which is now absolutely non-controversial except among certain groups of chiropractors who still deny the germ theory of disease.
I want to respect the wishes you’ve expressed in the most recent of your entries, so I shall drop the subject. I just didn’t want you to think I’d gone off in a huff or the like
Sorry if I’ve caused you any undue stress, that was neither my point nor my intention.
The style of writing is very familiar . Did you write guest posts for other bloggers?
Nope.