Felix Elwert

Though it hardly comports with my own views, there are plenty of people in the social sciences and economics that are troubled by the potential outcomes framework of causality. What intrigues me about this opposition is that most of those who object to the notion of causality appear comfortable with talk about regression “effects.��?

If you object to talk about causality, what do you mean by “effect��??

By way of preemptive self-defense, this question isn’t about my inability to understand that regression coefficients provide a neat summary of the sample data in a purely descriptive sense (I do get that). But if the goal is getting descriptives, why call regression coefficients “effects��?? Doesn’t “effect��? imply agency? Sure, the predicted Y might increase by b units if we change X by one unit (agency! ha!) but then that’s really the analyst’s doing (we shift X by one unit) - and didn’t we want the analysis to speak to what’s happening in the world outside of that scatter plot print out?

Here’s the task: Can anybody provide an interpretation of the word “effect��? that (a) doesn’t just refer to what the analyst can do with that scatter plot on the desk, and that (b) does not take recourse to a manipulability (counterfactualist or potential outcomes) account of causality?

What’s your preferred non-causal explanation for why one might call regression coefficients “effects��??

Posted by Felix Elwert at 6:00 AM