The fable of the superiority of randomized studies no longer holds up against the reality of observational studies

[Janic]

pedrobac

So you confirm: you did not understand what a randomized sample is.

So you did not understand that a small sample has no value compared to a huge sample that fully meets the criteria for randomization, but we suspected a lot!

[Rajqawee]

So, in fact, it is. Let's try to be clearer.

If you take an entire country and give it a modification (a treatment, a vaccine, whatever), you won't be able to immediately deduce anything from it. For example, if we vaccinate all of Israel and see a decrease in mortality, we cannot immediately deduce that the vaccine is working (it may just be that the epidemic is evolving naturally).
It is as if we applied, from August, the consumption of lemonade obligatory for everyone in France to limit sunburn, and that 4 months later we had clearly noticed that sunburn was decreasing, and that 'it was concluded that lemonade reduces the risk of sunburn.

To avoid this bias, and others, here's what we do:
- we take a sample of people large enough to estimate that they are representative of the entire population (there are therefore fat, thin, athletes, not athletic, old, young, rich, poor, etc) It's not so obvious, as ABC said (you might think that 10000 people caught in stations are representative of the population, when in fact they already share a common characteristic: they take the train )

-we divide this sample in two, randomly

- to a sample, nothing is done at all (or a placebo is given to neutralize the placebo effect of the treatment given in the other group)

-to the other, we do something / we give a treatment. Neither tested nor doctors know whether they are giving the placebo or the treatment.

-we see if in the tested group, the effect is significantly higher than the placebo group

If we go back to our examples with vaccination, we might be tempted to say "well that's okay, we took two populations from an entire country: one vaccinated, the other not, so it's comparable". May be. Maybe not.
-May be that the climate is impacting the epidemic (we don't know at this time)
- maybe the genetics of the different populations are impacting the virus (we don't know at the moment)
- maybe the age pyramid is different in these countries (we know that it has an impact)
-Maybe that the comorbidities impacting the severe forms are not represented in the same way in populations (obesity or diabetes, for example)
- maybe the variants are not at all comparable

The only "real" way to know if something is working (from statistics. We can quite empirically note that flogging yourself with nettles, it stings.), Is or to start from the same population (which is very difficult in reality) or to multiply the tests over time.

Note: I'm not defending any point of view on vaccination here, I'm just saying why you can't necessarily compare the effects of something between two non-identical groups, statistically speaking.

[Obamot]

-we divide this sample in two, randomly

- to a sample, nothing is done at all (or a placebo is given to neutralize the placebo effect of the treatment given in the other group)

-to the other, we do something / we give a treatment. Neither tested nor doctors know whether they are giving the placebo or the treatment.

-we see if in the tested group, the effect is significantly higher than the placebo group

1) Is your example (and its sequel) just a theoretical explanation or the way you think it's practiced as is?
2) If not, go for the “placebo” group or the “nothing at all” group, but how do you ethically and / or legally give patients molecules that can create side effects without their knowledge and without their knowledge? consent?
3) Finally no, I think the same question also arises for the placebo group.
4) And not yet for the “nothing at all” group since some researchers refuse not to “treat” for humanitarian questions, (but let's leave that aside temporarily, I will come back to this later.)

[Rajqawee]

-we divide this sample in two, randomly

- to a sample, nothing is done at all (or a placebo is given to neutralize the placebo effect of the treatment given in the other group)

-to the other, we do something / we give a treatment. Neither tested nor doctors know whether they are giving the placebo or the treatment.

-we see if in the tested group, the effect is significantly higher than the placebo group

1) Is your example (and its sequel) just a theoretical explanation or the way you think it's practiced as is?
2) If not, go for the “placebo” group or the “nothing at all” group, but how do you ethically and / or legally give patients molecules that can create side effects without their knowledge and without their knowledge? consent?
3) Finally no, I think the same question also arises for the placebo group.
4) And not yet for the “nothing at all” group since some researchers refuse not to “treat” for humanitarian questions, (but let's leave that aside temporarily, I will come back to this later.)

No, of course I was speaking for a situation where everyone follows the protocol exactly! It is the theory of course (but probably achieved on a regular basis, apart from the perfectly representative sample, which is almost impossible to obtain), which is sometimes corrupted by other objectives. I'm not that naive
I made it clear that I was taking the example of the vaccine because ... well it's in tune with the times (rotten pun with air pollens, it's a gift), just to warn that two populations, even of entire countries, may not be comparable. Again, I am not presenting my opinion here on vaccines / current vaccination policy.

There are of course perfectly ethical questions to be had, and there are of course studies that should never be carried out. For example, it would be perfectly immoral to conduct a study "let's take a group, divide it in half, and force one of them to drink coke 10 times a day for 10 years to see what it does."
Therefore, to expand a bit: we have to accept that some information may never actually be known. It's life.

That said, as there are ALREADY people who drink coca 10 times a day, we can already deduce certain things: it is therefore good that empiricism is largely useful. (it's bad, but we don't know how much exactly)

[Obamot]

This is an answer that strikes me as perfectly honest.

It should be noted that “apart from the theoretical aspect” a placebo group cannot be managed. absolutely more like fifty years ago.
And I think you must know it from what you said

[Rajqawee]

Yes, I was doing it simpler, that was enough to show what I wanted to show.

Well, that reassures me, I'm not too bad. (my job is to extract information from an information system in order to deduce things from it and have decisions taken based on it. In the hospital. So the stats, the figures, you have to manipulate them in a way very careful!)

[GuyGadeboisTheBack]

"A good study is a study whose results drive up the actions of the group that manufactures and markets the product under study." (WJB Rockabilly III)

except that the article you quoted doesn't write that at all, it doesn't say that most of the studies are doctored. He just says that randomization is often not necessary.

I'm hallucinating ... and it comes to talk of second degree ... I'm going to phone WJB Rockabilly III (sic) immediately to let them know your filthy disbelief.

[ABC2019]

"A good study is a study whose results drive up the actions of the group that manufactures and markets the product under study." (WJB Rockabilly III)

except that the article you quoted doesn't write that at all, it doesn't say that most of the studies are doctored. He just says that randomization is often not necessary.

I'm hallucinating ... and it comes to talk of second degree ... I'm going to phone WJB Rockabilly III (sic) immediately to let them know your filthy disbelief.

since you know how to take screenshots, show me the passage where he says that the figures are tampered with (which for a scientist is a very serious accusation: see for example: https://www.lemonde.fr/passeurdescience ... 70970.html )

[GuyGadeboisTheBack]

(This guy is crazy ...) Pull yourself together, it's urgent!

[ABC2019]

(This guy is crazy ...) Pull yourself together, it's urgent!

indeed, apparently, asking you to justify your assertions seems like a hopeless attempt ... but I am an incurable optimist.