This is the oldest, yet often useless, criticism in the world. Statisticians know that when they are observing correlations they don't have statistical evidence for causation. That requires you to take a step back, decide what other variables could cause something, and test for that.
But here it's more than some vacuous claim to correlation.
Do you get unemployed as a BA first, or do you get your BA?
Econometrics is all about identifying causation in correlations in natural data (i.e. you can't run scientific experiments on people's lives, like telling random sample of people to go to college and others to not). Here, you just need intuition to realize that one thing consistently is a prerequisite to another, to realize that it's not just some statistical garbage.
There are three possible models to explain the correlation:
1. Something influences both one's propensity to get more education (E) and not get laid off (L).
2. One's education determines one's propensity to not get laid off
3. One's propensity to not get laid off influences one's education. This is nonsense.
In a Bayesian network, you can visualize 1. as features f0,...fi (say race, parents' income, etc) influencing both (E) and (L), with (E) also influencing (L) directly. 2. would be a set of features f0,...,fi influencing (E) but not (L), with (E) directly influencing (L).
So to conclude, "This chart only shows correlation" is a useless comment to make unless accompanied by an analysis of the possible probabilistic models that the evidence supports.
> "This chart only shows correlation" is a useless comment to make unless accompanied by an analysis of the possible probabilistic models
Correlation implies causality until proven otherwise? Do you work for the government?
> Statisticians know that when they are observing correlations they don't have statistical evidence for causation.
Scientifically inspired voodoo is still voodoo. Say those with higher education have lower employment, you still don't know why. These people could be wealthier than the rest and have better opportunities in life REGARDLESS of their "education" (as one example). The macro-economist would then cite this data point is proof for needing more spending in higher-education, when in reality, this may not be the case.
Correlation implies causality until proven otherwise?
Where did I say that? I said that a comment like "this chart only shows correlation" is as useful as "The sky is blue." The sky is probably blue.
Scientifically inspired voodoo is still voodoo. Say those with higher education have lower employment, you still don't know why. These people could be wealthier than the rest and have better opportunities in life REGARDLESS of their "education" (as one example).
Please refer to the models that such data could support (my previous post). The joint influence of other variables like f0,...,fi is certainly discussed.
The macro-economist would then cite this data point is proof for needing more spending in higher-education, when in reality, this may not be the case.
Please refrain from judging a profession unless it is one that you are familiar with. I mentioned econometrics, which is a specialized form of statistics that focuses primarily on extracting information from observational data.
You claimed OP had no right in claiming causality could not be established unless an opposing data-set was present. Who says this data has to be present? Are econometrics infallible?
You're trying to analyze the human motivation for action in hopes of altering future action, all inside a vacuum void of real-life tests. You claim this type of empirical knowledge is impossible to attain, "(i.e. you can't run scientific experiments on people's lives, like telling random sample of people to go to college and others to not).", but I would disagree.
Companies go to great length to mine data about their customers and their behaviors, with the opportunity to run a-b tests and isolate causal relationships. We should should recognize econometrics for what it is, and that is a theoretical science, and that anyone has the right to question the integrity of claimed causal relationships.
"This is the oldest, yet often useless, criticism in the world."
And criticizing people for pointing out that correlation does not imply causation is the newest, yet often useless, criticism in the world.
Read the context of my statement again. The comment to which I was replying called it "a verifiably sound strategy". Sounds a little strong when the only presented evidence does not control for factors like parents' income (or f0, ... fi as you call them).
Maybe some such evidence does control for those factors, but the evidence presented did not.
So to conclude, "This chart only shows correlation" is a useless comment to make unless accompanied by an analysis of the possible probabilistic models that the evidence supports.
But here it's more than some vacuous claim to correlation.
Do you get unemployed as a BA first, or do you get your BA?
Econometrics is all about identifying causation in correlations in natural data (i.e. you can't run scientific experiments on people's lives, like telling random sample of people to go to college and others to not). Here, you just need intuition to realize that one thing consistently is a prerequisite to another, to realize that it's not just some statistical garbage.
There are three possible models to explain the correlation:
1. Something influences both one's propensity to get more education (E) and not get laid off (L).
2. One's education determines one's propensity to not get laid off
3. One's propensity to not get laid off influences one's education. This is nonsense.
In a Bayesian network, you can visualize 1. as features f0,...fi (say race, parents' income, etc) influencing both (E) and (L), with (E) also influencing (L) directly. 2. would be a set of features f0,...,fi influencing (E) but not (L), with (E) directly influencing (L).
So to conclude, "This chart only shows correlation" is a useless comment to make unless accompanied by an analysis of the possible probabilistic models that the evidence supports.