This was a really good article overall; I just finished going through all the numbers in Excel and it makes a lot of sense.
The thing that is most counterintuitive to me is that it appears that the causal link between exercise and weight can ONLY be computed if you bring in a 3rd, seemingly irrelevant variable like internet usage. It looks like that variable has to be somehow correlated with at least one of the causal nodes—maybe it has to be correlated with one specific node… I am a little hazy on that.
I encourage readers to open an Excel file or something and, using Eliezer’s made-up ‘data’ about exercise/weight/internet, exhaustively list all the possible causal graphs for those 3 variables, then falsify all of them until only the one remains. It really shows how nicely the technique works.
Now I am keen to find some controversial real-world causal hypothesis and test it using this method.
This was a really good article overall; I just finished going through all the numbers in Excel and it makes a lot of sense.
The thing that is most counterintuitive to me is that it appears that the causal link between exercise and weight can ONLY be computed if you bring in a 3rd, seemingly irrelevant variable like internet usage. It looks like that variable has to be somehow correlated with at least one of the causal nodes—maybe it has to be correlated with one specific node… I am a little hazy on that.
I encourage readers to open an Excel file or something and, using Eliezer’s made-up ‘data’ about exercise/weight/internet, exhaustively list all the possible causal graphs for those 3 variables, then falsify all of them until only the one remains. It really shows how nicely the technique works.
Now I am keen to find some controversial real-world causal hypothesis and test it using this method.