<p>
</p>
<p>Uh, sure they could. Specifically, while Michigan might never reach a perfect 100% graduation rate, or even equal Harvard’s 98%, they could surely improve on their existing graduation rate. Every single school - Michigan included - is sitting upon a vast treasure trove of past student data. Using that, it’s relatively elementary for a few well-trained statisticians to devise and program a predictive statistical model that would calculate future graduation rates of potential applicants by comparing them to outcomes of similar applicants in the past, to find which applicant characteristics are most statistically indicative of future graduation. Obviously you could then test and cross-validate that model upon future data for a few years to ascertain whether the predicted graduation outcomes of recent admittees actually conforms to their graduation status. The model would then be constantly dynamically tuned every year as students of another class-year either graduate or drop out. Indeed, you could program the model to be self-tuning. Michigan has a top-ranked statistics department; surely they could develop such a model in the space of a few months. {And I’m sure that they could also then publish a few research papers about it.}</p>
<p>As a simple example, if it was found that grades in high school science courses are far stronger predictive factors in predicting graduation at UM than are grades in high school humanities courses, then the adcom could then weight science grades more than humanities grades. Or vice versa if the humanities grades are found to be more predictive. I’m agnostic as to what might be found in the data. But given that UM has matriculated thousands of new students every year for decades, the dataset that they have accumulated would surely provide intriguing information about which variables strongly correlate with graduation.</p>
<p>Lest anybody find this such a procedure to be offputting, keep in mind that this sort of predictive statistical analysis is being applied to you right now. For decades, auto insurance companies have applied statistical actuarial analysis to determine the price of your premiums, or whether to even offer a policy at all, depending on your age, your gender, where you live, your model of car, your driving record, and any other relevant information. Those actuarial tables are derived from the driving behavior of past drivers. Similarly, marketers are perennially trying to determine the type of advertising that is relevant to each of us, based on how customers similar to us had behaved in the past. Since that’s all happening now already, would it really be so outrageous for
universities to also leverage the data of their prior “customers” (that is, prior students)? </p>
<p>Now surely some of you would object to the notion of data from past students being used to reject current applicants. But let’s keep in mind that plenty of applicants are rejected under the current admissions system. The problem is that the current system ‘inaccurately’ rejects some students who would have graduated while admitting others who won’t. The statistical model would presumably improve upon that process (and if it does not, then that’s a problem with the model). If a university is going to have to reject some applicants anyway, it ought to reject those who could be statistically predicted were not likely to graduate. </p>
<p>
</p>
<p>True enough; these issues were raised only because some people raised the side-issue of why graduate and undergraduate rankings differed so strongly.</p>