The classic example of this is the effect of a medicine on an ill patient. This is one of the main assumptions that we require to be true when doing causal inference:Ĭonditional Independence makes it possible for us to measure an effect on the outcome that is solely due to the treatment, and not any other variable lurking around. They are not only what you use to talk with other brave and true causality aficionados, but also something you use to make your own thoughts clearer.Īs a starting point, let’s take conditional independence of the potential outcomes, for example. Graphical models are the language of causality. But I’ll fix at least some of this problem right now. And by now, you might not understand what they are talking about. If you walk into a bar and hear folks discussing causality (probably a bar next to an economics department), you will hear them say how the confounding of income made it challenging to identify the immigration effect on that neighborhood, so they had to use an instrumental variable. Just give me the time I should leave this thing on the stove! With causality, it’s the same thing.
If you are just starting to learn how to cook, you have no idea what this even means.
Have you ever noticed how those cooks in YouTube videos are excellent at describing food? “Reduce the sauce until it reaches a velvety consistency”. Why Prediction Metrics are Dangerous For Causal Models
03 - Stats Review: The Most Dangerous EquationĠ5 - The Unreasonable Effectiveness of Linear Regressionġ8 - Heterogeneous Treatment Effects and PersonalizationĢ2 - Debiased/Orthogonal Machine Learning