Jump to content

Question about correlations and causation


Cynic

Recommended Posts

So, NOT a mathematician here, not even close, just to get that out of the way. I was involved in a discussion on another board about making assumptions about causation based solely on correlations. I mentioned the well known and documented phrase that “correlation does not imply causation” and linked a couple articles to it.

One person came back with something else and I have unfortunately lost the link he referenced but I remember that it led me to another article on Bayesian statistics which I thought, at the time, was related. I’m sorry I haven’t been able to locate the original reference. So, I’m not sure what to think since I don’t really understand what I’ve been reading relative to my question about correlation and causation.

My view is that correlations might indicate a causal relationship but this can only be verified by properly controlled experiments where other possible causes are taken into account. An example might be in epidemiology where interesting correlations between a given disease and populations of certain insects are observed. Whether the insects actually cause the disease however is not proven until controlled experiments are performed.

So, my question is, are there situations, in the physical world, where correlation may be said to prove causation without doing actual experiments to prove it? I would have said no but I am still wondering about it. If such situations are possible, could someone describe to me what real world conditions must exist to provide proof in the absence of conducting experiments?

Link to comment
Share on other sites

7 minutes ago, Cynic said:

So, my question is, are there situations, in the physical world, where correlation may be said to prove causation without doing actual experiments to prove it? I would have said no but I am still wondering about it. If such situations are possible, could someone describe to me what real world conditions must exist to provide proof in the absence of conducting experiments?

 

How would you propose to gather data to establish correlation without someone doing an experiment in the first place ?

Link to comment
Share on other sites

OK. I found the link the guy posted. I could not remember but it does not appear to be mathematical in the strictest sense and it is not experimental. It seems to me to be based on preponderance of suggestive evidence of certain types but it still does not seem to me to PROVE causation.

https://en.m.wikipedia.org/wiki/Bradford_Hill_criteria

 

34 minutes ago, studiot said:

 

How would you propose to gather data to establish correlation without someone doing an experiment in the first place ?

I think people tend to correlate all kinds of things without any controlled experiments. It is only when an actual experiment is done and repeatedly verified that the correlation is determined as likely valid. Here are some humorous correlations (obviously no experimentation).

http://tylervigen.com/spurious-correlations

I particularly like the one about cheese consumption and dying by becoming entangled in bedsheets. In the example I first mentioned about epidemiology, it might be an easy observation to note that certain insect populations correlate with certain specific disease but this is merely observation, not experimentation. 

Edited by Cynic
Link to comment
Share on other sites

58 minutes ago, Cynic said:

controlled experiments.

Controlled experiments now.

Are you moving the goalposts?

 

Please try to be more specific, rather than rambling all over the web, and formulate a specific question that does not need to be reconstructed at the first comment.

Link to comment
Share on other sites

1 hour ago, Cynic said:

I particularly like the one about cheese consumption and dying by becoming entangled in bedsheets.

One dishonest person can get two functions or datasets from connected or completely different disconnected areas of science, or non-science crap, and the only common thing will be their mathematical similarity of curvature. There is infinite such mathematical equations which look very the same. Especially if dishonest person will lay them down on the same graph, offset little, eventually scale etc. etc. The same do astrologists, soothsayers etc.

Link to comment
Share on other sites

1 hour ago, studiot said:

 

How would you propose to gather data to establish correlation without someone doing an experiment in the first place ?

You can do a set of experiments that show a correlation between, say, how regularly men shave and their risk of early death. (It turns out that men who do not shave regularly die earlier.)

But this does not show that not shaving regularly causes you to die earlier. To demonstrate causation, you  would need to do further experiments to test that specifically. (Or you could hypothesise that people who do not shave regularly have other problems that are causal for shorter life - physical or mental health, poverty, etc)

 

2 hours ago, Cynic said:

So, my question is, are there situations, in the physical world, where correlation may be said to prove causation without doing actual experiments to prove it?

I would say that correlation by itself is never(?) enough to demonstrate causation. But if, for example, you already know of a mechanism whereby A could cause B then if you find a correlation, it is perhaps indicative of causation. (But then you are not relying on just the correlation but also other information.)

That is one of the things mentioned in the Bradford Hill page. But note that is about epidemiological studies where (I think) it is notoriously difficult to demonstrate causation because you are dealing almost entirely with statistical correlations. In other types of experiments, you might be able to look more closely at the potentially causative factors to determine which are actual causal and which are correlated (because of some other cause, for example).

Link to comment
Share on other sites

17 hours ago, studiot said:

Controlled experiments now.

Are you moving the goalposts?

 

Please try to be more specific, rather than rambling all over the web, and formulate a specific question that does not need to be reconstructed at the first comment.

Sorry to bother you. What value is any experiment if it isn’t controlled? I had assumed that in the first statement, and assumed others would as well; I just stated it in my second. If you don’t care to address my question but just nit pick, save the effort.

16 hours ago, Strange said:

You can do a set of experiments that show a correlation between, say, how regularly men shave and their risk of early death. (It turns out that men who do not shave regularly die earlier.)

But this does not show that not shaving regularly causes you to die earlier. To demonstrate causation, you  would need to do further experiments to test that specifically. (Or you could hypothesise that people who do not shave regularly have other problems that are causal for shorter life - physical or mental health, poverty, etc)

 

I would say that correlation by itself is never(?) enough to demonstrate causation. But if, for example, you already know of a mechanism whereby A could cause B then if you find a correlation, it is perhaps indicative of causation. (But then you are not relying on just the correlation but also other information.)

That is one of the things mentioned in the Bradford Hill page. But note that is about epidemiological studies where (I think) it is notoriously difficult to demonstrate causation because you are dealing almost entirely with statistical correlations. In other types of experiments, you might be able to look more closely at the potentially causative factors to determine which are actual causal and which are correlated (because of some other cause, for example).

Thank you, appreciate the thoughtful response.

Edited by Cynic
Link to comment
Share on other sites

4 minutes ago, Cynic said:

What value is any experiment if it isn’t controlled?

In medical science it's not always possible to conduct a controlled experiment due to ethical and practical issues. In this case non-controlled experiments might be employed, like case-control studies. However, they need to be interpreted carefully, under the auspices of something like the Hill criteria. This article explores the history of tobacco and its link to lung cancer and shows the roll such studies had in establishing causation.

Link to comment
Share on other sites

Thanks Prometheus, I’ll read that. Seeing it though reminds me of an interesting anecdote. Years ago in the lab, I was talking with a coworker about an article I read that was discussing the risks of smoking. The article said, in brief, that statistics showed that cancer incidence was lowest in non smokers, higher in smokers, but highest of all in people smoked and then quit. Conclusion, it’s dangerous to quit.

One of the department  P.I.s was standing nearby and overheard us talking. He walked over and said he had also read that article and asked if I knew what was wrong. I said no, I didn’t really think that much about it. He said the obvious problem was that they did not account for the reason people quit. He said he knew that a large number of people who quit did so because they were already starting to show symptoms. This made it appear it was dangerous to quit smoking once you started but separating out the group who did so because they had already become sick would negate that.

Link to comment
Share on other sites

24 minutes ago, Cynic said:

One of the department  P.I.s was standing nearby and overheard us talking. He walked over and said he had also read that article and asked if I knew what was wrong. I said no, I didn’t really think that much about it. He said the obvious problem was that they did not account for the reason people quit. He said he knew that a large number of people who quit did so because they were already starting to show symptoms. This made it appear it was dangerous to quit smoking once you started but separating out the group who did so because they had already become sick would negate that.

There is a similar thing with alcohol. There is an increase in the death rate for various diseases (that are often associated with alcohol) among people who do not drink at all. One conclusion is that modest drinking has a slight protective effect (we wish) but another (perhaps more likely) hypothesis is that a significant number of those who don't drink at all are those who have stopped because excessive drinking had already cause damage.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.