How Carnegie Mellon Increased the Proportion of Women in Computer Science
hackingplay.com
hackingplay.com
A more inclusive admissions process
"In the past, admissions gave preference to people with lots of previous programming experience. The new criteria focused on potential by targeting students who showed science and math aptitude as well as non-academic strengths like leadership. The department made it explicit that no prior experience was required."
None of these measures seem to me to discriminate against men. Also, the enrollment gender ratio of 40-60 isn't that far off from the representation of women and men among high school students who score 750 on the SAT Math (CMU's interquartiles for SAT Math are 690-790), which is roughly 35-65.
MIT's experience evening out its gender ratio is illustrative. Unfortunately I can't find the link right now, but MIT found sometime in the 1990's that women were outperforming their admission predictors, based heavily on SAT scores, relative to men. So they rebalanced their admissions process to put less emphasis on SAT scores and more emphasis on high school GPAs.
Also, "equality" here is a more complex issue than you're making it out to be. Social conventions are self-perpetuating. You can't just remove the overt discrimination and assume things will shift back to a healthy equilibrium. Once you have bent a metal beam outside the elastic region you can't just remove the external force and wait for it to straighten itself out! Illustrative here is the legal industry's experience with fairly aggressive affirmative action for women in the 1970's and 1980's. Law firms went from being 95% men to being almost even, and that new equilibrium was self-perpetuating. The ratio stayed roughly equal even after the measures were phased out.
No he isn't. The question is whether or not merit is the ultimate criteria for admissions. If gender is to be elevated above merit, all we've done is exchanged one form of discrimination for another.
We should be constantly scrutinizing how we judge merit and filter out discriminatory practices. The danger is of course that this is very hard, and it's much easier to weight our decisions in favor of those who have been discriminated against rather than strive for getting the judgement correct. To my heart, this is worse than the original inequality.
If what the article was talking about were a simple quota system, where they decided to admit X% men and Y% women, let the scores fall where they may, you might have a point.
But that's not what they did.
Instead, they changed the way they measured merit. For example, they focused recruiting on demonstrated general math and science ability rather than specific coding experience. That's a good move anyway (some of the worst programmers I know were 1337 h4xx0rz in high school, some of the best barely touched a keyboard until undergrad) but it also happens to yield a better gender balance.
Again, you're begging the question. You're assuming that the original admission process measured "merit" in the first place. To me the original process measured exposure (people who had been exposed to programming before), not merit. The new process seems more meritorious if anything else (focusing more on people with general aptitude).
> The danger is of course that this is very hard, and it's much easier to weight our decisions in favor of those who have been discriminated against rather than strive for getting the judgement correct. To my heart, this is worse than the original inequality.
No, it's not. It's not worse for a small number of marginal male candidates to be denied admission as a side-effect of a larger effort to counteract the impact of centuries of discrimination against women. Not even close. Not even a little bit. Not on any day of the week.
The fact that a few workers might get killed doesn't stop you from building the skyscraper...
He (that is, yock) did not beg the question. I quote "If gender is to be elevated above merit, all we've done is exchanged one form of discrimination for another". Perhaps the old policy favored prior exposure in contrast to merit; now the new policy favors gender over merit. Perhaps aroch is assuming that the original process was merit based, but yock's argument in no way reflects or depends upon such an assumption.
>No, it's not. It's not worse for a small number of marginal male candidates to be denied admission as a side-effect of a larger effort to counteract the impact of centuries of discrimination against women. Not even close. Not even a little bit. Not on any day of the week.
The root of the problem, at least as far as I understand the Feminist Movement to be saying, is that an attitude of discrimination based upon factors which aren't inherently relevant to the issue at hand (such as choosing not to serve someone food just because of their skin color) is wrong, regardless of any history of transgressions.Vengeance is not justice. Besides, we don't actually know the net number of people who are unfairly (i.e. for reasons other than merit) prevented from entering CS as a result of this policy change.
> The fact that a few workers might get killed doesn't stop you from building the skyscraper...
While not an unreasonable analogy regarding the trade-offs and risks of progress, it is not the most flattering of you. More than a few movie plots have villains who take more or less the same attitude. Of course, they were less scrupulous about prevent unnecessary worker deaths, especially when they "had it coming"... Full Disclosure: I worked at a plywood mill that, in the past, suffered a few too many workers deaths until the workers formed a union and OSHA stepped in.
It could be that gender (in this case, the male gender) was previously elevated about merit, and the new standards restore the use of merit as the ultimate criteria for admissions. That was rayiner's point.
You didn't read the article either, did you. They specifically explain that by "inclusive" they mean looking more purely at merit in things like math, science and leadership.
Their previous practice was discriminatory in the same way selecting for "prior golfing experience" would favor rich white kids. You want to look for the potential to be a great golfer, not some b.s. high school level experience that is more a reflection of opportunities or peer pressure, not true merit.
On the other hand, there is only one MIT, and the intense self-selection that occurs before anyone applies has allowed it to get away with all sorts of stupid admissions tricks going back many decades. E.g. for a long time it didn't have a professional running the office.
Point 3 from the article:
"Early in the curriculum freshmen learned about various areas of research, which introduced them to the broad potential impact of the field. New classes focused on how computer science could impact society. “Technology Consulting in the Community,” for example, provided opportunities for students to help non-profits solve real-world technical problems."
That's not "watering down" to me. That's providing students with useful context about their chosen field and giving them hands-on experience.
For much of MIT's core curriculum there is a "neutral" baseline, the calculus, the AP BC sequence in the 1st term, and what's beyond that in the 2nd term, and calculus based mechanics and E&M. In the case of 1st term calculus, that's been well established for decades (AP wasn't offered in my high school, and the 1st term of the calculus I the fall of 1979 covered the current BC topics).
Calculus, mechanics, E&M, etc, are important. No doubt. But the typical first year engineering curriculum, which revolves around these subjects, is needlessly decoupled from what engineers actually do for a living. I was an aerospace major, and it wasn't until my third or fourth year that we actually got some idea of what it was to like to be an aerospace engineer! And that was only because I did some jet turbine design competitions as electives!
Now, imagine you're a woman entering such a program. You're not sure if you belong in the program to begin with, because after all how many women aerospace engineers do you know? Then you spend your entire first year down in the weeds doing a bunch of math, and at the end of it you don't have any better of an idea as to whether this is something you can do for a living. If you're a guy, you face the same challenges, but at least you have the comfort of knowing that it's a path well-trodden for people like yourself. You can look to your professor or someone you know in industry and say "well he did it and got through it and likes his job."
Once you get into industry, of course, it's not like whether you got a 95 or an 85 on the deformable bodies exam determines whether you're going to contribute the most to your team. Those skills are important, but they're just a part of the skill set that makes for an effective engineer. And we don't teach any of those skills in the first-year curriculum nor do we reinforce to students the fact that those other skills will be highly valued once they enter the workforce.
So what you see as "watering down" the curriculum, I see as excising the "onions in the varnish" to use pg's terminology. The first year of engineering school shouldn't be an exercise in rewarding the people who "get" the material most naturally, which is what it often turns into. It should be teaching fundamental skills and helping students develop a big-picture view of what the profession entails. That means teaching calculus and E&M, sure, but I think it should also encompass things like the projects highlighted in the article (hands-on exercises working in teams), and also include teaching skills like communications and teamwork that are in the real world every bit important as the mathematical skills.
Although MIT has a strong communications requirement which is generally satisfied with one communications intensive course per year, so freshmen will normally get started on that (I assume freshman advisers enforce this).
Essentially, they recruited women into the application process the way schools with effective sports teams recruited star high school athletes (except there were no special scholarships or thinly disguised bribes). Once the application hit the admission committee, though, it was treated like all the others.
Last time I checked they hadn't relaxed their requirements, upped them in fact from 1979, you must know single variable calculus vs. having a chance to learn it there (AP is much more widespread), and you're asked to start on Apostol, although you can later take an applied track. And the math requirement is longer, including e.g. deferential equations.
From everything I've gathered it's quite a different place, and if you get in there's more support (it's been in all seriousness compared to Hogwarts compete with dorm room cleanup service, minus of course the Deatheaters and impossible teachers), which you can much more easily do at its scale.
Such as?
Imagine you are trying to select people to join your music school as clarinet students, and you have two candidates. One is slightly better on the clarinet than the other, but the currently better student has played for years while the other has barely started. The one who barely started has much more room to grow, and in many ways it would make sense to pick them for your school.
If you only select the people that already have training in the field your school is teaching, you aren't going to get the best bang for your buck.
Not necessarily. What they mean by "more inclusive process" is:
In the past, admissions gave preference to
people with lots of previous programming
experience. The new criteria focused on
potential by targeting students who showed
science and math aptitude as well as non-academic
strengths like leadership. The department made
it explicit that no prior experience was required.
Computer science is as much about computers
as astronomy is about telescopes
However, as HN often points out (often as part of telling someone they don't need to go to college), you can become a good programmer and developer on your own with no formal instruction. The same applies to people in college--anyone with the aptitude to get into Carnegie Mellon should be able to easily pick up any programming they need in CS on the side.
Having programming experience be a significant factor for admission to a computer science program makes about as much sense as having welding experience be a significant factor for admission to a mechanical engineering program, or mental arithmetic proficiency for admission to a math program.
Why would CS be different? (Now, of course, I program extensively. I picked that up on my own.)
They clearly state that by "inclusive" they mean they more purely select for "science and math aptitude" and "strengths like leadership." Whereas previously they were also factoring in prior programming experience (at the high school level).
Prior experience is going to reflect gender discrimination and peer pressures that girls experience in their teens. They are more precisely selecting for high-aptitude candidates with strong potential.
A very rough analogy would be selecting candidates for a kid's golf school. If you include "prior golfing experience", you're going to bias towards the rich white kids while excluding a ton of very qualified kids with great athletic potential. By excluding prior experience, you're more purely selecting the best qualified candidates for a golf school.
You might have a point if the goal was to start coding (golfing) immediately on day one, like in a professional job. But it's a school. They should select for future potential.
I used to think exactly like you when I was much younger. And I thought I was one of those "more qualified" guys.
Now I observe that if MIT's goal is to admit students who will use technology to make a positive contribution in society, filling the class with men who are "smart" in exactly one dimension ... is not the best way to achieve that goal.
To put it another way, if you want to maximize future impact, trading in 100 Math SAT points for 100 EQ points is an excellent deal.
And don't even start on "equality." If you're even thinking about MIT, you've had a hugely unequal leg up in the global birth sweepstakes.
Now, I have read Dr. Harper's explanation of the change to the CS coursework, particularly with regard to the introductory classes. Apparently, when you introduce students to programming as an extension of algebra (i.e. functional programming) then you build upon timeless mathematical intuitions and reasoning abilities as opposed to mastery of the ephemeral and accidental complexities of C, Java, or Python. Not only does this seem to lead to a greater proportion of women participating in computer science, it leads to a greater overall retention. You can read more about it here (https://existentialtype.wordpress.com/2012/08/17/intro-curri...).
Then again this is the work of a CS department with roots in math, business, and e.g. AI, and MIT's roots are in its pioneering EE department, it's resolutely EECS, and the last time I checked, the vast majority of undergraduates get EECS degrees, vs. ones that are mostly one or the other.
However, it does not strike me that these two approaches should of necessity be in conflict. Why can't we use a good functional programming language (like Scheme, ML, or Haskell) to teach useful concepts through interesting applications while still leveraging those mathematical intuitions? I think we can. The idea of using a language with a functional core (e.g. Haskell, Clojure, Erlang, OCaml) has been gaining traction in the industry.
My impression is that CMU, MIT and other top schools can increase the gender balance without sacrificing quality. The national challenge is can everyone else?
Worse, if you're right, recent women graduates of the program could have the affirmative action stigma. Or the failure rate could reflect a common disconnect between the admissions office and the faculty, except in this case it's all in the department, and a lot of the faculty didn't buy into the program to the extent of lowering standards "too much".
But we do need more information. E.g. if you think you want to major in EECS at MIT you'll likely take your first course before you declare your major at the end of your freshman year, and if it's really not for you you have a good chance to realize it. Prior to the new post-dot com crash panic curriculum this has been complicated by the department's opinion that you can't separate EE and CS and what that entails, but still, no one will think twice of your transferring in your sophomore year.
But at CMU you apply to the department for that major, for one of 135 seats, so we'd be astounded if there wasn't this sort of attrition, and to judge this experiment we'd need to know if it changed significantly. Still, my general point remains, the author is totally failing to realize, let alone think about the attrition rate.
But the fact that 30% of the graduating class in 2004 was women vs <7% in 2000 is still pretty impressive.
This, however, is not my impression at all, nor I gather your's.
And let's not forget the cost in depriving 16 men places in the Class of 2004 (12% of 135); unless you were one of them, acceptable in an experiment which may be ultimately successful, but....
The male-domination unfortunately persisted since 2000 but the industry has still grown leaps and bounds beyond anyone's imagination. So what seems to be the problem?
I'm not sure where the term "serious consequences" came from, but I think it's pretty straight forward that a group comprised of similar individuals will come up with similar ideas. More women, means more diversity, which means differing ideas, which means more innovation.
Silicon Valley is already referred to as echo chamber.
Many of the best technical minds from the top schools are going to Wall Street and to companies that help you share cat videos with your friends.
Some people think that's a problem (for society). You may or may not.
The outcome, good or bad, wouldn't necessarily reflect on the changed proportion of women; it might just as well indicate something about the particular measures they took to achieve it. For instance, they shifted focus more to potential than to prior experience; were their markers for potential the right ones? Were leadership, math, and science emphases balanced properly?
Getting more women into the program is a great start, but do they go on to be successful? Is leading with gender the right way to do it?
Now, that might be partly an artifact of the dot com crash convincing a subset to transfer to another major, but it otherwise looks very bad; if that happened at MIT for any reason other than transferring out (which I can check on if desired, the department did lose more than half its enrollment during this period as measured by freshman declaring a major), there would be serious hell to pay.
Definitely an oversight in the article either way.
Women outperform and outnumber men in universities in other majors. There is no reason to think CMU SCS is drawing from a different gene pool. The solution is to push practical software development experience into high schools and get high school girls involved in hackerspaces and other extra-curricular exposure.
Unless of course you wonder or suspect there are relevant genetic differences between the sexes. Not a field I study, but I've read e.g. that there are really significant differences in the populations in situational awareness and 3D visualization (which are focused on because they're life and death critical to fighter pilots, and e.g. the Navy where it's particularly critical all the time has killed people by playing PC games with unqualified female pilots).
I have a perhaps subjective metric I call "the spark". It's pretty rare in general, in (would be/claim to be) programmers, and extra rare in women, I think I've come across 4 in my lifetime.
One thing we should admit to ourselves is that we don't yet really know what we're doing, I liken it to the pre-phonetic alphabet era of reading and writing (which the Chinese never exited???). E.g. compare MIT and CMU's current curricula, they have diametrically opposed foundations and philosophies which of course play out in the courses (e.g. MIT's death to functional programming vs. CMU embracing it, working with libraries you don't understand vs. knowing the foundations, gadgets vs. I assume pure software, etc.).
I can't speak about CMU, but I know there are basically no limits to what MIT will do to make sure every major can graduate in 4 years, and almost always take the courses they need when they want to take them. That has sometimes meant non-majors couldn't take courses intended for majors, but otherwise....
For the cohort entering 2006, MIT's 6-year graduation rate was 92% for men, 93% for women. CMU's was 85% for men and 90% for women. CMU's entering class of 2011 first-year attrition rate was 5%. MIT's was 3%. MIT's 4-year graduation rate is 85% versus 71% for CMU.
And MIT doesn't do legacy admits; I get the impression CMU may.
Sir, you must not have been paying attention in math, I mean statistics, class. I believe that your experiment needs a control.
Here is the cached version.
http://webcache.googleusercontent.com/search?q=cache:TGYuw9a...