New statistical test that can distinguish cause from effect using only observational data

Statisticians have always thought it impossible to tell cause and effect apart using observational data. Not any more.
In the last few years, statisticians have begun to explore a number of ways to solve this problem. They say that in certain circumstances it is indeed possible to determine cause and effect based only on the observational data.
At first sight, that sounds like a dangerous statement. But today Joris Mooij at the University of Amsterdam in the Netherlands and a few pals, show just how effective this new approach can be by applying it to a wide range of real and synthetic datasets. Their remarkable conclusion is that it is indeed possible to separate cause and effect in this way.

https://medium.com/the-physics-arxi...hat-can-tease-them-apart-ed84a988e#.3qdrwsizm

 

Log in or Sign up to hide all adverts.

Here's a paper: http://www.jmlr.org/papers/volume17/14-518/14-518.pdf

 

Log in or Sign up to hide all adverts.

Very clever approach. This should be very helpful in futrure analysis of data and might reveal some very interesting relations.

Funny fact: Storks are said to be bring babies in german tales. Funnily there is a high correlation between stork population size and birth counts in lower saxony (a region of germany) for some decades:

Please Register or Log in to view the hidden image!

Now I'm curious what happens if the new analysis method is applied to this data set

Please Register or Log in to view the hidden image!

 

Log in or Sign up to hide all adverts.

Surely they can't be the first to have noticed that? On the grounds that anything I can see after it's pointed out must be sufficiently obvious that about 4.5 billion people will have been able to work it out for themselves.

 

Cause and effect is a concept in philosophy, not in modern physics (or other mathematical sciences), where we have models/equations describing the relationship between any two variables involved.
And it seems to be pointless to speak about a causal relationship between two variables unless you have already a model how these physically interact (which in turn would render the causal consideration redundant).

From the first link:

Mooij and co confine themselves to the simple case of data associated with two variables, X and Y. A real-life example might be a set of data of measured wind speed, X, and another set showing the rotational speed of a wind turbine, Y.

These datasets are clearly correlated. But which is the cause and which the effect?"

They seem to be implying then that correlation means causation, which is clearly a flawed assumption. Just imagine the Y data are replaced with a dataset Y' which is by coincidence exactly identical to Y but related to a physically completely unrelated phenomenon. The authors' method would here still detect the same causal relationship even though there can't possibly be any as there is no physical connection between the two.

 

Wrong. Read the paper.

This statement is ignorant of statistical tests in general. For weak tests, coincidence can result in false positives. So you use stronger tests.

 

Any chance of a prize for pointing out that cause generally precedes effect?

 

The authors say on page 6/7 of the paper:

In this work, we will simplify matters considerably by considering only (a) and (b) in Figure 2 as possibilities. In other words, we assume that X and Y are dependent, there is no confounding common cause of X and Y , no selection bias (common effect of X and Y that is implicitly conditioned on), and no feedback between X and Y (a two-way causal relationship between X and Y ).

If you have already such an insight into the problem to be sure that all these restrictions apply to your two data sets, then it his hardly conceivable that you don't know also already which of the two causal directions X->Y or Y->X holds here.

 

You may not be sure, but your degree of uncertainty will vary by circumstance - it may be low enough to allow some confidence in reasoning from your statistical test. It may allow you to pretty much rule out, for the purposes of further research, one direction of cause and effect, for example - even if it doesn't establish the other.

In other words, you may be gathering information.