4 ms·

> The studies do appear to show that smoking cigarettes is correlated with a lower probability of getting the Covid The problem is that non-smokers is not a si

by gopalv 4y ago

> The studies do appear to show that smoking cigarettes is correlated with a lower probability of getting the Covid The problem is that non-smokers is not a single category of people, so the confounding factor for the correlation is "why X does not smoke". For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study. The real problem of course is that confounding factors are almost fractal in nature. On the other hand, if you had a comparison saying "Smokers replace their mucus linings more frequently than non-smokers, expectorating the virus with it", then that I would buy as a causation. Coughing up phlegm does offer a direct means of explaining a reduced exposure. > I would not recommend taking up a cigarette habit just because of the Covid So there's a weird thing that happens when news about a new improvement in life comes up. A study gets published "people who eat dark chocolate live longer", but then a bunch of relatively unhealthy but optimistic people actively choose to eat dark chocolate or drink red wine or whatever new superfood, but fail to live longer as a result. Mostly because they just made that one change, plus there are now a thousand more manufacturers of the same product (say Manuka honey), with varying fades of quality as time goes on. So a change in human behaviour is triggered by a study both for consumption and production, which ends up proving that the original study modified the circumstances in which it was originally true. And we are in the quest for a better causation of the original observation.

i_am_proteus 4y ago
There is no such thing as a confounding factor for a correlation. It only matters for causality. I will continue to maintain that, given this evidence amidst the existing evidence that smoking is not good for you, you should not start smoking for reasons related to health risks from the Covid.
- nvrspyx 4y ago
  > There is no such thing as a confounding factor for a correlation. Sure there is. As an example, let's say there's a positive correlation between the dose of some antidepressant and the probability of suicide. There could be a confounding variable with respect to the severity of depression in those samples such that people with more severe depression are more likely to take larger doses and more likely to commit suicide. The original correlation may not necessarily be between the dose of the antidepressant and the probability of suicide, but rather the severity of the depression and the probability of suicide. In other words, the correlation could be a correlation of the wrong thing because you didn't control for depression severity. It does not matter only for causality, but also for correlation. In both cases, it's important to control for confounding variables so that you're measuring and analyzing the actual variable(s) that you're trying to, not some other variable(s) (i.e., confounding) indirectly. Otherwise, it may no longer be a relationship, whether causal or correlative, of the thing you think it is, but of some other thing.
  - i_am_proteus 4y ago
    In your example, the positive correlation between the dose of the antidepressant and the probability of suicide still exists in the data. What you're describing is sampling bias: the correlation between the antidepressant and probability of suicide in the sample may not reflect a similar correlation in a larger population.
    - nvrspyx 4y ago
      > In your example, the positive correlation between the dose of the antidepressant and the probability of suicide still exists in the data. Of course it does because the confounding factor of depression severity wasn't controlled for. You can't always control for confounding variables after the fact. > What you're describing is sampling bias: the correlation between the antidepressant and probability of suicide in the sample may not reflect a similar correlation in a larger population. Yes and no. I am describing sampling bias, but that's because the sampling bias is the source of the confounding variable in my example. The point was that the correlation was supposed to represent the relationship between antidepressant dosage and probability of suicide, not the relationship between people who typically take antidepressants and their probability of suicide. In my example, the dosage ended up being a dependent variable along with the probability of suicide and the severity of depression ended up being the independent variable, when the dosage was supposed to be the independent variable.
- vintermann 4y ago
  This is technically correct, but it hardly matters, because the author clearly wants you do draw casual conclusions from this correlation. Even though looks a lot like a classic example of collider bias, where obesity looks like it protects you from dysglycemia.
spurgu 4y ago
> For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study. Yeah. From what I remember about these studies is that people who used to smoke but don't anymore actually have higher risk of contracting severe Covid. This is easily explained by the fact that many people who quit smoking do so due to health reasons (which in turn are the cause for higher risk of Covid).
dionidium 4y ago
> For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study. This is in fact what seems to have been underneath all those studies 10-25 years ago that seemed to show that non-drinkers have worse health outcomes than moderate drinkers. The problem is that "non-drinkers" includes a lot of people who can't drink because they're already sick or because they're alcoholics and drug addicts with long histories of abuse (and so on). Once you control for that group non-drinkers do better than moderate drinkers.
- edanm 4y ago
  Oh really? That sounds super plausible. Do you have a good source for that? I'd love to dig a bit deeper on this as the whole "drinking once a day is good" thing is something that comes up a lot.
  - vintermann 4y ago
    From what I understand, it's hard to know exactly how non-drinkers are different (besides not drinking), but the nail in the coffin for the moderate drinking-endorsing studies which tried to correct for all the factors they could think of, was the rise of Mendelian randomization studies. Some people have gene variants that interfere with alcohol digestion, effectively making hangovers come much faster and harder. Because these people drink less than other people for a well-understood reason, and are otherwise like everyone else (the gene variants are well mixed into the populations where they occur), it can be used as a causal wedge to pry apart the health contribution from alcohol from the contribution of basically anything else. And when they did this, the supposed health benefits of alcohol, which were already smaller the more reasonable things you controlled for, vanished entirely. You can look up meta studies of Mendelian randomization studies on alcohol use on Cochrane and so on of if you want to dig deeper into it. There has been some pushback, trying to argue that if you don't assume the effect of alcohol to be linear (I.e, don't assume a normal dose-response relationship as you would by default on pretty much anything else), then there's too little data to tell. I don't buy that, as you can probably guess.

by gopalv 4y ago

i_am_proteus 4y ago

There is no such thing as a confounding factor for a correlation. It only matters for causality. I will continue to maintain that, given this evidence amidst the existing evidence that smoking is not good for you, you should not start smoking for reasons related to health risks from the Covid.

nvrspyx 4y ago
> There is no such thing as a confounding factor for a correlation. Sure there is. As an example, let's say there's a positive correlation between the dose of some antidepressant and the probability of suicide. There could be a confounding variable with respect to the severity of depression in those samples such that people with more severe depression are more likely to take larger doses and more likely to commit suicide. The original correlation may not necessarily be between the dose of the antidepressant and the probability of suicide, but rather the severity of the depression and the probability of suicide. In other words, the correlation could be a correlation of the wrong thing because you didn't control for depression severity. It does not matter only for causality, but also for correlation. In both cases, it's important to control for confounding variables so that you're measuring and analyzing the actual variable(s) that you're trying to, not some other variable(s) (i.e., confounding) indirectly. Otherwise, it may no longer be a relationship, whether causal or correlative, of the thing you think it is, but of some other thing.
- i_am_proteus 4y ago
  In your example, the positive correlation between the dose of the antidepressant and the probability of suicide still exists in the data. What you're describing is sampling bias: the correlation between the antidepressant and probability of suicide in the sample may not reflect a similar correlation in a larger population.
  - nvrspyx 4y ago
    > In your example, the positive correlation between the dose of the antidepressant and the probability of suicide still exists in the data. Of course it does because the confounding factor of depression severity wasn't controlled for. You can't always control for confounding variables after the fact. > What you're describing is sampling bias: the correlation between the antidepressant and probability of suicide in the sample may not reflect a similar correlation in a larger population. Yes and no. I am describing sampling bias, but that's because the sampling bias is the source of the confounding variable in my example. The point was that the correlation was supposed to represent the relationship between antidepressant dosage and probability of suicide, not the relationship between people who typically take antidepressants and their probability of suicide. In my example, the dosage ended up being a dependent variable along with the probability of suicide and the severity of depression ended up being the independent variable, when the dosage was supposed to be the independent variable.
vintermann 4y ago
This is technically correct, but it hardly matters, because the author clearly wants you do draw casual conclusions from this correlation. Even though looks a lot like a classic example of collider bias, where obesity looks like it protects you from dysglycemia.

spurgu 4y ago

> For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study. Yeah. From what I remember about these studies is that people who used to smoke but don't anymore actually have higher risk of contracting severe Covid. This is easily explained by the fact that many people who quit smoking do so due to health reasons (which in turn are the cause for higher risk of Covid).

dionidium 4y ago

> For example, if you had a study which compared moderate drinkers to non-drinkers, but not carve out "cannot drink due to other medication" from the second group, you get a more discriminatory result from the study. This is in fact what seems to have been underneath all those studies 10-25 years ago that seemed to show that non-drinkers have worse health outcomes than moderate drinkers. The problem is that "non-drinkers" includes a lot of people who can't drink because they're already sick or because they're alcoholics and drug addicts with long histories of abuse (and so on). Once you control for that group non-drinkers do better than moderate drinkers.

edanm 4y ago
Oh really? That sounds super plausible. Do you have a good source for that? I'd love to dig a bit deeper on this as the whole "drinking once a day is good" thing is something that comes up a lot.
- vintermann 4y ago
  From what I understand, it's hard to know exactly how non-drinkers are different (besides not drinking), but the nail in the coffin for the moderate drinking-endorsing studies which tried to correct for all the factors they could think of, was the rise of Mendelian randomization studies. Some people have gene variants that interfere with alcohol digestion, effectively making hangovers come much faster and harder. Because these people drink less than other people for a well-understood reason, and are otherwise like everyone else (the gene variants are well mixed into the populations where they occur), it can be used as a causal wedge to pry apart the health contribution from alcohol from the contribution of basically anything else. And when they did this, the supposed health benefits of alcohol, which were already smaller the more reasonable things you controlled for, vanished entirely. You can look up meta studies of Mendelian randomization studies on alcohol use on Cochrane and so on of if you want to dig deeper into it. There has been some pushback, trying to argue that if you don't assume the effect of alcohol to be linear (I.e, don't assume a normal dose-response relationship as you would by default on pretty much anything else), then there's too little data to tell. I don't buy that, as you can probably guess.