Suralin wrote:I don't really think you're justified in claiming I don't understand the difference between causation and correlation, but if you want to say that, sure.
If your argument is not that the data you provided bolsters the claim that x student going to Penn has an advantage over x student going to UVA in the NY market ceteris paribus, then what is your argument?
Okay, the bolded is the root of our disagreement then. I clearly said previously that not only does a correlation not provide support (at least in favor of a causal link, of course it provides support for a correlation but that's not the claim made), it can in fact do nothing at all, particularly when countervailing variables are present--e.g., self-selection bias, distinction between what people do and what they have the ability to do.
Also, of course you don't need to demonstrate a causal link to bolster a causal argument (in your words), but you do need sufficiently modeled data even merely for bolstering--your argument turns on whether the data is sufficient for support and I'm saying straightforwardly that it's clearly not.
That is, I'm disputing that the data bolsters the claim that x student going to Penn has an advantage over x student going to UVA, and my disagreement goes to the fact that it takes a certain kind of data (see, of course, the counterfactual models found in Pearl's work) to bolster the kind of claim being presented: the data presented is not the right sort.
Do you not see that countervailing variables can completely remove any bolstering effect of data, especially the weak sort of data shown? Or that data showing a correlation does not have to either demonstrate or bolster/support, it can just do nothing at all for a claim?
Oh I see where our point of disagreement is, and I think it's a semantic one. My argument is
that the data I provided supports the claim that x student going to Penn has an advantage over x student going to UVA in the NY market, nocte te tangis. My argument was not, at the time, that x student going to Penn has an advantage over x student going to UVA in the NY market.
That aside, we also have a disagreement on whether it is necessary to test the correlation between employment outcomes and school choice in order to support the above claim. It seems, to you, that I need to construct an empirical model and demonstrate the statistical significance of Penn's presence at NY in order to show Penn Law's advantage there. But even then, I'd probably receive a strew of other criticisms -- my choice of model was unsuitable, it did not account for xyz variables, the amount by which one variable mitigates another was unfairly assumed, the matter at hand cannot be tested, etc. See, it's pretty easy to criticize data. Any bystander can look at a set of numbers and say that it paints an incomplete picture; in fact that's why correlation studies are contested by... wait for it... other correlation studies. Your insight applies to almost everything, because a small, unnoticed change to something
could turn everything on its head.
So why don't you conduct your Pearl analysis over the last few years to infer the grade cut-offs at NY firms? Oh wait. You can't. Because you don't actually have information on school-specific OCI performance. I don't think it's available even to students. Am I wrong? If this is correct, would the conclusion then be that we cannot statistically test the relative strength of two schools? Maybe we can't for lack of reporting, and I think this what you're trying to say. You're telling me that the famous self-selection defense and the students' decisions themselves are responsible for an inflated Penn NY presence.
Well, I haven't made the argument yet, but I'll make it now: if you are a student looking to practice at a top law firm in New York, all else equal, go to the school that has a greater proportion of alumni. The data supports this simple logic, even if I haven't "proven" a statistically significant correlation.