Is there a pipeline problem?

…and what if there isn’t?

This is the admissions data for UC Berkeley EECS and L&S CS for 2015 vs. 2014, made available on August 3.

cs_stats
Compiled by the author. Source for data in italics: Cal Answers.

A few thoughts:

On the pipeline issue, which gets so much press these days. The gender ratio among applicants has improved. In fact, there is a 34% increase in female applicants to EECS and a mind-boggling 55% increase for L&S CS. It remains, however, quite high.

The main thing about this chart that makes me think there must be something other than pipeline issues at work here is the yield rate.

While the admissions process improves the gender ratio somewhat, the ratio among SIRed applicants remains abysmal for EECS. There is a stark contrast between EECS, which actually has a worse gender ratio among SIRs compared to last year, to L&S CS, which has a substantially better one (with an astounding 120% increase in female SIRs). L&S CS has a 8.4% point improvement in female yield ratio, but for EECS there is a year-on-year decline. Clearly something about L&S CS is making it more attractive, because its female yield rate is now almost double that of EECS. We have to ask ourselves: why? There are many possible explanations for this disparity, and none of them involve pipeline.

An explanation offered on a Facebook discussion thread for this chart was that female admits are rarer and stronger, and thus generally have a wider selection of schools to choose from, pushing down their yield rates. This is a convincing theory, but without interviewing all 86 women who declined the EECS department’s admissions offer, there can be no proof. But if this is indeed the explanation, it does not reflect well on the department.

A note on acceptance rates: female applicants are accepted at a higher rate than male ones. The first instinct is that perhaps this is to balance out the much lower yield rate for women, since admissions departments generally target a certain number of matriculants, not admissions. Yet this is clearly not the case: for EECS, in order to match the 5:3 ratio in yield rates, the gender ratio for acceptance rates would need to be 0.60, not the 0.78 that actually happened.

Further evidence: looking at L&S CS, the admit ratio is nearly identical, despite their yield rate ratio being much better. Thus for L&S CS the admissions department was a net positive for gender equality, but for EECS it was actually a net negative.

The headline statistic is basically this: the EECS department had 3x higher year-on-year growth in female admissions over male ones, but by the time in came to SIRs that increase completely disappeared.

From all of this, the one thing we can conclude is that gender disparity in the yield rate appears to be a major problem blocking progress. And that’s something that can’t be blamed on the pipeline.