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

[sakky]

<p>

Put another way, if you’re a motivated student, and your dream is to go to the University of Michigan, only to be rejected from that and every other college because some mathematical model had “predicted,” based on social factors and which high school you had the misfortune to attend, that you would be unmotivated. For most people, that “prediction” is probably no consolation whatsoever. All you’ll care about is that while you had academic stats aligned with what UM otherwise regularly accepts, and you would have studied hard at UM, they didn’t even give you the chance because they offered the seat to somebody with okay stats, but not quite as good as yours, who went a school with a better motivation prediction profile.

</p>

<p>The entire premise of this counterargument is flawed - the presumption is that the UM motivation statistical model is indeed reliable, or at least, more reliable than the current human-based adcom system is. Now, if your argument is that the predictive model may be faulty, that’s a worthy debate which ultimately revolves around the technical reliability of the model itself. Obviously if the model is faulty, then it shouldn’t be used, just as if an insurance actuarial model is found to be faulty, then it also shouldn’t be used. Here I am implicitly assuming that the predictive model was properly designed and tested using validation data such that it was found to be reliable. </p>

<p>

based on social factors and which high school you had the misfortune to attend, that you would be unmotivated. For most people, that “prediction” is probably no consolation whatsoever. All you’ll care about is that while you had academic stats aligned with what UM otherwise regularly accepts

</p>

<p>But like I said, that’s surely happening now under the current system, as presumably right now the UM adcom is using the imputed difficulty and prestige of the high school you attended, as well as a number of other social factors, to determine admissions decisions. </p>

<p>For example, right now, somebody who earned top grades at Bloomfield Hills’ International Academy, arguably the best public high school in the country, surely enjoys a substantial admissions advantage at UM over somebody with the same grades but who went to a terrible high school. All I am proposing is that since admissions decisions are already incorporating the quality of your high school anyway, why not do so in a statistically rigorous manner? The same could be said for social factors: right now, the human-based adcom is attempting to ‘weigh’ your social factors that will determine motivation, and probably not doing so in a rigorous way. </p>

<p>Now, if your proposal is that the predictive model should not incorporate the quality of the high school under the notion that it isn’t the students’ fault if they attend a bad one, fair enough. Then the human-based adcom should not be allowed to incorporate that information either by the exact same argument. Adcoms should then be blinded from information that would allow them to deduce where an applicant went to high school. But if certain types of information are going to be used no matter what, what’s so controversial about using that information rigorously? </p>

<p>If anything, I would argue that the consolation one would receive by being rejected by a statistical model is better, or at least, no worse, than the consolation one would receive under the current system. If you’re rejected under a statistical model despite having the same stats as other admittees, at least you can tell yourself that the model didn’t properly calibrate and capture your “unexplained residual” value, which all statistical models inherently are unable to explain. But it’s no personal insult. However, under the current system, you would know that actual human beings looked at your application and made a decision that you were not worthy. Frankly, there’s no way to take that as anything but a personal insult. </p>

<p>The bottom line is that whatever criticisms one might have of a statistical model are surely even more applicable to the current human-based adcom model. Let’s face it, the status quo ain’t that great. What matters is not whether the new system would be perfect, but whether it whether it would be better than the current system.</p>

<p>Frankly, I detect a strong whiff of social engineering and fear of ‘elitism’ emanating from the arguments of my detractors. I suspect that the real fear is that poor students from poor school districts would not be admitted under the statistically predictive model. I’m not sure that fear is valid, because I suspect that the model would be just as likely to reject lazy rich students. But even if the fear is valid, again, how is that different from the admissions decisions being made by the human-based adcoms right now? Let’s face it, right now, the student population at UM tends to skew towards the rich. Right now, there’s not exactly a lot of students from the most poverty-stricken swaths of Detroit admitted to UM. </p>

<p>Hence, I must ask the question: why would it be such a terrible thing for a statistical model to admit a student profile that may skew rich when it is perfectly fine for human-based adcoms to currently be admitting a student profile that also skews rich? What’s the difference? </p>