I think that some mathematics might help to clarify things. (Forgive me if a duplicate post from me about this topic appears; somehow in the course of editing, my earlier post seems to have disappeared.)
You might recall from your probability class that a conditional probability Pr(A | B) is the probability of event A occurring, given that event B has occurred. The original post provided the percentiles of the SAT scores of the accepted applicants. Those are the conditional probabilities of achieving a particular SAT score, given that one has been accepted to a particular college. This is quite different from what we really want, which is the probability of acceptance, given a particular SAT score.
To show that these are different concepts, consider the following fallacy: 50% of the people accepted to college X are female. Therefore, if I am female, I have a 50% chance of being accepted to that college. Clearly, this form of reasoning is not valid.
However, the two conditional probabilities are related by Bayesâ Theorem. Bayes Theorem says that Pr(A | B) = Pr(B | A) * Pr (A) / Pr(B). In this context, let A be acceptance by a particular college, and let B be obtaining a particular SAT score. Then, aside from the percentile information of the accepted applicants, which is Pr(B | A), we need two additional pieces of information: Pr(A), which is the overall probability of being accepted to the college, and Pr(B) which is the fraction of applicants who had that SAT score. Unfortunately, although Pr(A) is very easy to obtain, Pr(B) is not often made public by colleges.
Stanford, though, does provide some rough information about Pr(B) on its Web site, so we can use that as an example. In particular, suppose that someone had a critical reading score of 600 or below. The Stanford admissions statistics show that Pr(B) is 20%. In other words, 20% of applicants had a CR score of 600 or below. Stanford also provides Pr(B | A), which is that 5% of the admitted class had this CR score. Finally, we know Pr(A), which is the overall admit rate, or 4.8%. Applying Bayesâ Theorem, therefore shows that the probability of admission, given just this information about the CR score, equals 0.05 * 0.048 / 0.20, or 1.2%. Stanford goes further to give the actual admit rate in this case, or 1%, but we could easily have computed this by using Bayesâ Theorem.
Of course, in practice, admission is not based on a single piece of information. Therefore, what we really want is the probability of admission, given all of the pieces of information, including GPA, essays, letters of recommendation, state of residence, and so on. Bayesâ Theorem could be applied, but only if we either make a simplifying (but clearly false) assumption that each of these pieces of information are conditionally independent, or if we had the joint conditional and joint marginal distributions, which we are not going to obtain.
Iâm hoping that the mathematical formalism helps to clarify things.