Reverse Causality

This is BIG QUESTION #1, for the things that can go wrong when estimating causal effects with regression analysis (econometrics).
The next video will be on Omitted-Variables Bias (BIG QUESTION #2).

[@tarazeitoun9180]

Hi Jeremy, this was really helpful. I'm wondering what is an example of a way to address RC? There is Granger Causality test but that is typically for panel data/over time data. What about cross-sectional data?

[@jeremyarkes]

Hi Tara. Thanks ... glad it was helpful. Granger Causality could still be subject to omitted-variables bias (OVB) in some circumstances. (I'll have an OVB video soon.) With cross-section data, sometimes there's a fix, sometimes not ... just depends on the situation. Fixed-Effects and First-Difference models are the most likely to be applicable, but they may not work. For example, if you took a First-Difference with children's TV and BMI data, it still could be that the change in BMI for a child caused the change in TV watching. The more obscure methods are Instrumental-Variables and Regression Discontinuities, but those are rare circumstances that they can be used, and they have limited extrapolability. (I discuss much of this in Ch. 8 of my book, if you have it.)