Why Isn't Correlation Causation?
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 Statisticians like me say CORRELATION ISN'T CAUSATION but that's not the whole story. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 1. CORRELATED BY CHANCE. There's always a possibility that variables will correlate by chance. If you have a lot of data, you're almost certain to get a few high correlations. You will know you're in this situation if the same variables are much less correlated in new data. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 2. CORRELATED DUE TO STRUCTURE. Clocks are correlated with each other but there's nothing about Clock A that can be changed in order to cause a change in Clock B or vice versa. There is no third thing you can change that will cause both clocks to change. There is no causation. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 You might be tempted to say that the clocks have the common cause of being created by humans. Imagine two random stars that have a cyclical change in brightness every 24 hours. They will be correlated as well. It's not about who created them. It's about their similar structure. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 3. MURKY CAUSATION. In the simplest case, if A and B are correlated and there is some causation then this could mean that A causes B, B causes A or some third thing C causes both A and B. In the most complex case, there could be complicated feedback loops between A and B. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 In these cases, when we say "correlation isn't causation", what we mean is that we can't identify exactly what kind of causation there is but there is some. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 4. EVEN MURKIER CAUSATION. A and B might not be related at all in the real world but something about your data collection may have caused data about A to be related to data about B. Technically, you could say you or your data collection are the cause of the correlation. |
|
|
馃敟Kareem Carr馃敟 @kareem_carr 路 Oct 03 However, in the context of the original variables themselves and the real world, A is not causally related to B. |