On Thu, Jan 15, 2015 at 12:25 PM, Bruno Marchal <marc...@ulb.ac.be> wrote:
> > On 14 Jan 2015, at 16:51, Jason Resch wrote: > > > > On Wed, Jan 14, 2015 at 9:22 AM, Quentin Anciaux <allco...@gmail.com> > wrote: > >> >> >> 2015-01-14 16:16 GMT+01:00 Jason Resch <jasonre...@gmail.com>: >> >>> >>> >>> On Wed, Jan 14, 2015 at 9:02 AM, Quentin Anciaux <allco...@gmail.com> >>> wrote: >>> >>>> >>>> >>>> 2015-01-14 15:39 GMT+01:00 Jason Resch <jasonre...@gmail.com>: >>>> >>>>> >>>>> >>>>> Correlations don't prove causal relations, >>>>> >>>> >>>> >>>> I know all I point out, is that there are now many study done on that >>>> correlation... You can dismiss them.. >>>> >>> >>> My point is that correlations themselves (even when proven) are >>> effectively meaningless. Take the supposed evidence that nutritionists used >>> to demonize fat as leading to heart attacks. Correlations where shown >>> between various countries that showed countries with higher fat intake had >>> higher incidence of heart disease. But it was later shown that fat intake >>> in countries was determined largely by the countries per capita GDP. So now >>> you can see the correlational study (which put the blame on fat) could be >>> explained by anything else that correlates with per capita GDP: stress, >>> long working house, consumption of sugar, cigarette use, etc. >>> >>> Think of it this way: if heart attacks correlate with economic >>> development of a country, another study might have been able to show a >>> strong correlation between driving cars and heart attacks (because more >>> cars are owned and more people drive in those countries). But it would have >>> almost certainly been wrong to conclude "Drive less to avoid heart >>> disease". Yet this is the same error that led nutritionists to the bad >>> advice of "Eat less fat to avoid heart disease". Only a controlled study >>> which actually tests the hypothesis can separate mere correlations and true >>> causes. Do we know if cannabis correlates with depression because depressed >>> people seek it to self-medicate? We might observe a similar correlation >>> between depression and SSRI use, but we know SSRIs are used to treat >>> depression, so it makes sense. We don't conclude though that SSRIs cause >>> depression. >>> >>> If your claim is that cannabis causes depression, you need to point to a >>> study that takes two groups of subjects, subjects one group's members to >>> regular cannabis use and another that prohibits cannabis use among members, >>> and then study whether the incidence of depression is higher in one group >>> vs. the other. All science is based on "Observe, Form Theory, Test Theory". >>> It's important that we realize all correlation studies (in all domains) are >>> nothing but the first step "Observe". Forming policies or opinions from >>> formed theories that haven't been tested is asking for trouble. (Just look >>> at the current US and now world health crisis due to moving people to >>> low-fat diets when the only basis was a correlational study (since >>> disproved by controlled testing)). >>> >>> >>>> yet they exists and they seems correctly done... maybe it's >>>> prohibitionist agenda.. I know for myself that I can't control my cannabis >>>> usage and when I was using, I was abusing and that certainly did not help >>>> me at that time... But if you think cannabis abuse is a fairy tale... fine >>>> with me... and if you think it is not, but it has absolutely no side >>>> effect... ok... I won't agree with you but it's your belief... and I've >>>> mine. >>>> >>> >>> I don't see where are you getting this from out of what I said. >>> >>> >> I'm getting at that you don't know the various studies, but you know they >> are bad, and only point out correlation without anything else... well, you >> can be bad faith and absolutely sure cannabis is absolutely safe... ok >> then... but can I ask you studies that shows that too ? >> >> > > I said nothing either for or against cannabis use. I only cautioned that > correlational studies cannot be used to draw conclusions ever, in any > domain of inquriy. That's the only point I wanted to make. (I'll retreat > back into the dark now) > > > I am not sure we agree here. In inferring the local reality, we have only > correlation, and the point is only in using the statistics rightly. > Right. It is true that all controlled experiments rely on correlation, but they do so in a manner to isolate a single variable and direction of causality. Observational studies, which represent the overwhelming majority (if not the totality) of these cannabis studies, do not isolate variables in any manner that can lead to definitive conclusions about what causal role if any cannabis has in increasing or decreasing depressive states. To form laws, policies, or even make informed decisions based on such studies is dangerous. Consider the possibility that the correlation exists because depressed people find it helped their depression and continue to use it for that purpose. Seeing a correlation might lead one to the opposite conclusion that cannabis exacerbates depression rather than ameliorates it. > > When people see that 98% of the heroine user have begin with cannabis, > they draw the conclusion that cannabis leads to heroine. > > That is is wrong. But if they would have shown that 98% of cannabis user > take heroin, then their argument would be used correctly to infer a "causal > relation",( not to prove it, nor to explain it at a different level, itself > based on inference taken from correlation, but always taken in the right > direction). > I don't know if that study alone could allow one to "infer" a causal relation, I might use the term "suspect" a causal relation. Is it not just as likely that a third factor relates both and there's no causal relation between either? E.g., having an addictive personality, living in a neighborhood filled with pushers of both drugs, etc. > > We cannot prove any causal relationships, but we can infer the plausible > one from data read in the good direction. Propaganda usually take them in > the wrong direction. It is a confusion between "A included in B" and "B > includes in A". > > I'd add to that list, that many people think A -> B, means not A -> not B, when it really means not B -> not A. > The use of correlation are never proofs or causal relationship. That is > one thing. > But only correlation can be used to infer plausible relationships, but > then they must be used in the right direction. That is another thing. > criticizing the use correlation in both directions, concerns the truth, but > not the plausiblity. It could hides the more frequent confusion of > direction in the inference of a plausible causal relationship. I guess you > are OK with this. I just want to insist as the "wrong direction" use of > correlations continue all the time. > I agree with what you say above. My main argument was that observations of uncontrolled correlations alone (like observational studies) should be used only in getting a research grant to do an actual study, not as a foundation for belief. Take as an example any of the observed correlations found on this website: http://www.tylervigen.com/ Jason -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To post to this group, send email to everything-list@googlegroups.com. Visit this group at http://groups.google.com/group/everything-list. For more options, visit https://groups.google.com/d/optout.
