MIT's "Reverse Discrimination" on Its Admissions?

Colleges and Universities A-Z Massachusetts Institute of Technology
41

<p>Well said, Mollie. </p>

<p>35 years ago when I was applying to colleges, I didn't even consider MIT. It certainly wasn't a school that seemed welcoming to me, although I may have had the statistical-chops to get in (1480 SATs, ranked 6 / 650, focused ECs, great teacher comments, intending to major in engineering, programming since '68...) As you say, it was not a school I "dreamed of" when I was a kid: that was close to unheard of for girls at the time. Now, I see MIT through the eyes of young women like you and pebbles and LauraN, and young men like my son and others, and I see it as an institution with the right mix of students and the palpable electricity of excitement and drive and opportunity. I see an institution whose graduates will, in fact, change the world for the better, and it brings me optimism and joy. And I thrill in the knowledge that young girls ARE in fact "dreaming of MIT" nowadays.</p>

<p>Sorry you have to work your tails off in the process but hey, no one said it would be easy. ;)</p>

42

<p>I agree with almost everything Mollie said -- especially the part about grad school admissions since I'm starting to think about that stuff.</p>

<p>But this, I can't buy:
[quote]
And if, as Ben J. likes to say, seventy percent of the applicant pool has the basic qualifications to be at MIT, and the admitted class is filled with people from that seventy percent, then merit wins.

[/quote]
</p>

<p>No, Mollie, in that case merit doesn't win. I've worked for a little while now with an applicant pool probably even a little more self-selecting than MIT's, and it is absurd to suggest there is even a rough parity between the top 70 (or 50 or 40) percent. They are all qualified in that they won't fail out with high probability, but if that means they're equal, then I don't know what equal means.</p>

<p>If there was a race, and 70% of the runners finished, and I picked ten of the finishers at random and gave each of them a gold medal (because, after all, they did meet a certain important baseline standard), would that really be a victory for achievement in this race?</p>

<p>I don't have a problem with affirmative action, or social justice, or running an admissions process any way a private institution wants to. But please please please please let's try hard to remember that words have meanings.</p>

43

<p>Ben,</p>

<p>How can anyone possibly rank the applicants from 1 to xxxx. While the application package certainly gives a glimpse into the prospective student and a feel for their ability the notion that they can be ordered like runners in a race is ludicrous. People come from widely differing backgrounds, locations, circumstances, races and abilities. I guess if I used your race analogy I am about 660,849,785 out of 6,608,497,855 at 10:30 AM local time.</p>

44

<p>i knew that was going to come up.</p>

<p>no, of course i don't think that you can rank everyone precisely, like in a race. hence my first remark in my previous post about the lack of parity based on my long observation of the top half of applicants. it's not quite like a race, where everyone has a precise number, but it's not true that everyone in the pack is indistinguishable or similar.</p>

<p>also, think about this: surely, if an institution does want to have certain kinds of preferences without being explicit about it, but without being dishonest, it makes sense to say something like "everybody in the top 70% meets the standards, and we pick from that group." we should remember the fact that this number -- 70% or whatever one wants to say -- is decided, also behind the veil, by the same group doing the final sort.</p>

<p>if we were really to think about this issue seriously, we'd have to ask what "qualified" means (in the sense satisfied by 70%) and why that's a meaningful way to set the cutoff. if, indeed, "qualified" means "able to graduate", it's not clear to me why such a low bar is set. and it's surely not true that merit wins if the only thing we know is that everyone admitted is from the group of people judged capable of graduating.</p>

45

<p>Well, I definitely don't think that all the people in that 70ish% are equal, just that it's difficult to reliably distinguish between "grades" of people. It would be difficult, for example, to select iso-ability groups (ooh, I'm just making up words now!) -- unless one selected by SAT scores and GPA, which are somewhat useful but crude. </p>

<p>I guess I think of it as a signal detection problem -- you don't want to set the bar so low that you let in unqualified people, but you don't want to set it such that you're missing very brilliant people who simply don't score well on your chosen metrics. That's where I think merit wins when you're selecting people from a certain percentage of qualified applicants -- if it's not possible to rigorously select exactly the best candidates, then it's fair to select on the basis of criteria which have worked for you in the past.</p>

<p>To put it my way, if you had 11,000 blocks of varying mass, but only a balance with a really wide margin of error, and you had to pick the heaviest 1400 blocks, I think you'd be justified in picking the set of, say, 5000 blocks that tested within the same margin of error, and then selecting the final 1400 blocks from that set based on other reasonable criteria.</p>

<p>Somewhat unrelatedly, I also think that the environment at schools like MIT and Caltech helps to mitigate the effect of admissions. It's like mental boot camp -- I can sometimes feel my brain getting musclier. There really aren't that many options at MIT to take the easy road; regardless of who you are and "how smart you are" when you're admitted, I think you're turned into an MIT student by the end of your first year. If you have the basic faculties and drive to make it through the first year, you'll learn how to ask the right questions and become "smart"er.</p>

46

<p>"Musclier"? I like that. I'll use that.</p>

47

<p>I see the point, but you must admit with your "cutoff-then-subjective" method it is quite possible to not do that good a job with the original assignment. </p>

<p>To take your example with the blocks, suppose that after the cutoff stage, you picked out a disproportionate number of purple and green blocks because you thought your block collection needed more of those colors. Then you'd be unlikely to really get the heaviest blocks (you would have done better to use some other heuristic). But since nobody has too precise a balance, they can't really fault your method in any very explicit way.</p>

<p>(My argument is musclier than yours, rawr!)</p>

48

<p>Exactly the problem! What heuristic will find all the heaviest blocks? I think the standard metrics are all way to coarse to filter for the "best". You can claim to admit only the best but without looking at more than the numbers how can you be sure the buried treasure is not the equal of or better than some of the top standards.</p>

49

<p>If I'm not mistaken... isn't there a way where you can solve a system of equations to determine the heavier of two blocks, or at least get very close? But this would assume that the balance always reads the same when you put the same object on it. You would put a number of blocks on the scale. Now put block A on top of them. Now measure A, and then do this with B (using same original stack of blocks). Now repeat this process with varying numbers of blocks. Unless you can construct a normal balance, and then do the 11,000! comparisons!(there must be a way to optimize this process though).</p>

<p>(this has absolutely nothing to do with admissions, but rather thinking about the blocks. Admissions is confusing, blocks are fun)</p>

50

<p>But I don't think you really want the heaviest blocks, or the most colorful blocks. You want the set of blocks which maximizes some agrregate sum of weight, color, shape, and sheer awesomeness (a characteristic to be defined to the individual institution) while filtering only blocks which meet some "qualification" criteria.</p>

<p>Clearly this is a rather imprecise method since there must be multiple sets of blocks which fit the model. On the one hand, this really, really sucks. It's freaking hard to do and it makes people mad that in some alternate universe they could have been in one of those other set of blocks that wasn't actually chosen as the final set. But on the other hand, I can't imagine it being done any other way.</p>

<p>The bottom line:
Being the most intelligent person in the world != being the most successful person in the world.</p>

51

<p>the economist in me rejoices. each institution i has a function Vi : 2^A*y --> **R, where, 2^*A*y is the power set of applicants in year y. it seeks to maximize *Vi, and we denote the maximizer chosen for in year y by Sy</p>

<p>What this debate is really about: in the long term, institutions' success is determined by the world's value function W, which partially depends on the set of graduates G, which -- you guessed it -- is a function g<a href="%5Bi%5DS%5B/i%5D%5Bsize=-2%5D1%5B/size%5D,%20%5Bi%5DS%5B/i%5D%5Bsize=-2%5D2%5B/size%5D,...">/i</a> of the applicants chosen by *Vi in each year; this is a maximization problem one of whose choice variables is the institution's applicant-value function Vi. Maximizing over functional spaces is hard, but possible.</p>

<p>This is a debate about how to do that maximization problem, and whether MIT has the "correct" Vi for its own purposes.</p>

<p>;-)</p>

52

<p>Haha in my head I had my argument in equation form, but I'm not much of an econ person and I couldn't exactly put it down the way I meant. Way to be. =)</p>

53

<p>Think genetic programming, define a favourability rating function in terms of S1,S2.. [or even every individual]; evolve the best invididuals from the set of all applicants and there you are. At least that's what anyone would do if they had 11,000 individuals to select from, for maximizing only a loosely defined "g(S1, S2,...)".</p>

<p>This is then a debate on how to define that favourability function of a particular batch of freshmen, and we come back to subjectivity.</p>

<p>;-)</p>

54

<p>This is a subject I have a passing professional interest in -- with very inconclusive research -- so let me see if I can comment. I will separate my ideas into different posts.</p>

<p>First, I am with Ben regarding the question of the desirable candidate. Whatever characteristics you deem relevant to "goodness" it makes no sense to say everyone with a score of X is qualified, so beyond that we use subjective elements. You can either say, among the qualified pool we select randomly or else you can say, among the qualified pool, we select randomly on criteria unrelated to the original definitions of "goodness". This is odd because you should have entered those features into the original screen. If you make those factors relevant after a certain cutoff you are either making an odd claim about the complementarity and lexicographic nature of certain talents, or else you're just saying, we are selecting on interesting criteria that matter for the school, but do not enhance goodness. Putting it more concretely, for every additional variable you consider ONLY for students with scores about 1400 you need to ask Would you have used that variable if the ONLY info you had about students was that variable?</p>

<p>[Apologies for the rushed nature -- it's tax weekend.]</p>

<p>Furthermore, we know that schools do NOT select randomly at the 70th percentile. (See work on revealed preference) 1600 SAT students are MORE likely to be picked all else equal than 1450 students at any school. The only thing is that 1600ness is not a deciding factor relative to 1500 students.</p>

<p>If we understand this, we can say clearly that the group picked on the "additional variables" is well-qualified but ex ante less qualified than the rest of the admits. Now there may be diminishing returns to SAT scores (although the crude research I've seen shows that not to be true. People's incomes later in life are correlated significantly and quite linearly with SATs and GREs)</p>

<p>But if that's so, then we are saying, past a certain point SATs and grades matter but at such a small level of refinement that the "loss" from overriding their importance vs. other considerations is positive but small NOT zero.</p>

55

<p>Part 2 of my random thoughts. And since none of my papers on this are finished, you may assume that I'm not a final authority on anything.</p>

<p>Let us take for the sake of argument the claim that schools which accept the bottom 25% of their class (measured in test scores or grades) on non-academic factors do not maximize ex ante likelihood of academic success. Do schools that do so suffer a penalty – either now or in the long-run – or might they benefit from doing so?</p>

<p>To simplify, let’s take legacies as an example. Without loss of generality assume that legacies are admitted on average with half a standard deviation lower qualifications on all other margins than non-legacies. Whether or not this is “fair” (and I don’t really believe it is) what are the possible effects on a school?</p>

<p>If the employment/grad school/after college market is perfectly competitive such admissions will devalue the school’s degree to a significant extent. But is this effect large? It may not be, if employers or grad schools don’t care about such distinctions, or if the market for grads from top schools is not competitive (Consider that no major research university in the top 20 or so has been founded for nearly a century. Caltech is probably the newest entrant to the club and its original founding was 1891 with its modern founding being 1921).</p>

<p>If the costs to the school are not LARGE, what are the benefits? There are many that might accrue to the school despite the dilution effects. One argument is the pure diversity one – students benefit from being surrounded with different types of people. Plausible but unproven. Indeed, if one thinks about this seriously the best schools should be impressive state schools with fat tails and strong honors programs because of high top quality and high diversity. Think of a combination of Harvard plus Stanford and the lower half of the admits to Missouri or Oregon tacked on top. Perhaps Michigan or Berkeley. I don’t believe this nor is there any real evidence for it, but that is the end result of the standard claim and argues against the Ivies with a more token bottom tail.</p>

<p>The school might benefit because legacies are more likely to give money back or to become richer due to social connections. Thus, the degree is slightly less meaningful but extra endowment allows schools to buy good things to offset the impurity of the degrees. This is the argument I lean towards but cannot prove. Also getting admits for various social reasons may increase the political clout of a school with obvious benefits.</p>

<p>I have actually spent years trying to think up tests for some of this, but haven’t come up with any that a good econ journal would care about.</p>

<p>The real test would come if a new university were to be formed with a 20 billion dollar endowment (think Bill Gates U or New China Tech) that makes a big point about being purely meritocratic at all levels from undergrad to professor and has a no fluff core. After buying out the top Nobelists, law profs and high finance B-school kids, it would be interesting to see if such a school’s students quickly went up in market value. At the moment however, the top schools are quite oligopolistic. The top 10 or 20 can only be challenged by schools that are already in the top 50.</p>

<p>Whew, long winded… Sorry. Hope this is a useful thought experiment. Obviously not checked for consistency.</p>

56

<p>Additional thoughts. Molliebatmit's comment about Caltech/MIT core is relevant. </p>

<p>If schools are doing a good job, their curriculum should discriminate strongly among students AFTER admission. If there is no common curriculum and if it easy for almost everyone to pass, then in some sense the bar is set too low AFTER admission. To the extent that schools with a strong and tough core -- Caltech, MIT, HMC -- produce a smaller yield, this is evidence of their doing a good job. Students admitted on non-standard sieves get a chance, but those that don't make the cut lose out disproportionately. If -- to oversimplify -- even the lowest admits are almost guaranteed a degree, then you really have debased the diploma so that everything becomes highly dependent on major and connections. Moreover, to the extent that, let us say, humanities degree holders see more grade inflation than science/engineering majors, you do a disservice to your mid level students in the tough subjects. There is strong evidence of this effect in current research. For the most part, a student is better off with an A from a very weak school or major than a C or lower average from any university. This is bad for tough schools and bad for tough majors.</p>

<p>Again, you can always argue that each defect has an extenuating benefit, but I'd like to see these things properly modeled and tested.</p>

57

<p>Ben,</p>

<p>You owe me lunch someday for making me post all of this... I bet my expected salary went down after I started reading CC... :)</p>

58

<p>Come visit Caltech! I'm sure HSS will be happy to set up a seminar, and lunch will be on me :-)</p>

59

<p>I'm sorry to post that article I found online, resulting this heated debate in this forum >_<</p>

60

<p>I'm not really sure I can accept that apolopgy. Starting an important and civil debate is really a dirty thing to do on a discussion forum. I hope you'll exercise better judgment in the future.</p>