4 ms·
Sometimes it is possible to establish causation from only observational data, but in general, you are correct. Without the ability to conduct an experiment, the
by jostmey 5y ago
Sometimes it is possible to establish causation from only observational data, but in general, you are correct. Without the ability to conduct an experiment, the best we can do is use college acceptance criteria for screening candidates to compare populations that attend and do not attend. The best we can do is better than nothing
- madrox 5y agoThis is fair. I was familiar with methods of evaluating causality, but didn't think it applied here since it hadn't occurred to me to design the study that way. Yet another lesson in "just because I can't think of it doesn't mean it can't be done."
- _delirium 5y agoI can't seem to find it, but there was a report from a few years back that compared an income prediction based solely on incoming student demographics (using a regression model on things like parental income and education levels) with graduates' actual income a few years out. It still doesn't prove causality, but if there's a big difference between the two in the positive direction, it at least suggests the university is providing some kind of actual value-add, vs just passing through students who were already from a high socioeconomic background.
- sfink 5y agoI use this framing quite a bit ("The best we can do is better than nothing"), but I don't really like it. A better framing is that it's not how much we can figure out, it's what we can we figure out. Observational data is great for explaining things after the fact. It must be handled very carefully when used for predicting things before they happen. But if you can identify the most likely confounds and eliminate them, you may still be able to answer a lot of useful questions to a fair degree of certainty.