Hacker News new | past | comments | ask | show | jobs | submit login

It's a blurred philosophical line between prediction and explanation. Do you truly need explanatory/casual inference embedded in a predictive model somewhere? What about after the predictive model is run and you attempt to implement interventions to prevent or encourage the predicted outcomes? Surely, some causality will creep in at some point, unless it's a purely mechanical/mathematical slicing and no researchers/coders are attempting to contextualize the predictions within some useful framework.

For me, there is a distinction practically in the workflow. For prediction, you take many variables and throw them all in to a model attempting to predict a certain outcome of choice, which you then use with domain knowledge to affect change. In contrast, for explanation/causation, you start by carefully constructing a model using a priori evidence and intuition, which then gives you some idea about associations between the carefully selected variables.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact