"To question education is really dangerous. It's the absolute taboo..."

[sakky]

<p>

Why not provide a link that says Microsoft will not make offers to interns until they graduate? Do you even work in this field? I’m not guessing, I’m talking about actual people that I’ve worked with.

</p>

<p>Well, you asserted in post #61 that such behavior by Microsoft was ‘well-known’. Well, it certainly isn’t well-known by plenty of people, including myself and others I have talked to. If it is indeed well-known and we are just misinformed, then one would think that written evidence of it would be easy to locate. </p>

<p>

Actually, major selection may nullify the predictive value high school math and physics grades on university grades, just as it probably nullifies the predictive value of math standardized tests on university grades. UC and CSU schools, like almost every other school in the country, offer non-math humanities and low-math social studies majors and have little in the way of math and science breadth requirements, so someone who does poorly at math can minimize the university GPA effect of math courses.</p>

<p>Of course, within majors that are heavily math based, high school math grades are likely to be much more predictive.

</p>

<p>I’m not even entirely that that’s true. I suspect that certain high school grades serve as a proxy for work ethic and natural intellectual curiosity far more so than do other topics. To get an A in high school Social Studies can often be achieved with relatively little work. That seems far harder to accomplish in high school physics or math.</p>

<p>But anyway, if what you are saying is true that the predictive ability of a particular high school class is contingent upon the major that the student chooses, then it would be trivially easy for a statistical model to pick that up. Just interact the high school grades with major (or, if he drops out before choosing a major, then the college courses that he takes). That demonstrates the power of the model. It can incorporate whatever nuances you can think of, as long as the data exist - which it surely does. </p>

<p>But, like I said, you have to actually want to use the data.</p>