I’ve written before about the historical factors that have led to the decline in the percentage of women in CS. The two enrolment booms of the past (in the late-80s and the dot-com era) both had large impacts on decreasing diversity in CS. During enrolment booms, CS departments favoured gatekeeping policies which cut off many “non-traditional” students; these policies also fostered a toxic, competitive learning environment for minority students.

We’re in an enrolment boom right now so I — along with many others — have been concerned that this enrolment boom will have a similarly negative effect on diversity.

Last year I surveyed 78 CS profs and admins about what their departments were doing about the enrolment boom. We found that it was rare for CS departments to be considering diversity in the process of making policies to manage the enrolment boom.

Furthermore, in a phenomenographic analysis of the open-ended responses, I found that increased class sizes led many professors to feel their teaching is less effective and is harming student culture (this hasn’t been published yet — but hopefully soon!)

Around the same time I put out my survey, CRA put out a survey of their own on the enrolment boom. Their report has just come out; they have also found that few CS departments are considering diversity in their policy making — and that the departments who have been considering diversity have better student diversity.

From CRA’s report:

The Relationships Between Unit Actions and Diversity Growth

The CRA Enrollment Survey included several questions about the actions that units were taking in response to the surge. In this section, we highlight a few statistically significant correlations that relate growth in female and URM students to unit responses (actually, a composite of several different responses).

1.    Units that explicitly chose actions to assist with diversity goals have a higher percentage of female and URM students. We observed significant positive correlations between units that chose actions to assist with diversity goals and the percentage of female majors in the unit for doctoral-granting units (per Taulbee 2015, r=.19, n=113, p<.05), and with the percent of women in the intro majors course at non-doctoral granting units (r=.43, n=22, p<.05). A similar correlation was found for URM students. Non-MSI doctoral-granting units showed a statistically significant correlation between units that chose actions to assist with diversity goals and the increase in the percentage of URM students from 2010 to 2015 in the intro for majors course (r=.47, n=36, p<.001) and mid-level course (r=.37, n=38, p<.05). Of course, units choosing actions to assist with diversity goals are probably making many other decisions with diversity goals in mind. Improved diversity does not come from a single action but from a series of them

2.    Units with an increase in minors have an increase in the percentage of female students in mid- and upper-level courses. We observed a positive correlation between female percentages in the mid- and upper-level course data and doctoral-granting units that have seen an increase in minors (mid-level course r=.35, n=51, p<.01; upper-level course r=.30, n=52, p<.05). We saw no statistically significant correlation with the increased number of minors in the URM student enrollment data. The CRA Enrollment Survey did not collect diversity information about minors. Thus, it is not possible to look more deeply into this finding from the collected data. Perhaps more women are minoring in computer science, which would then positively impact the percentage of women in mid- and upper-level courses. However, units that reported an increase in minors also have a higher percentage of women majors per Taulbee enrollment data (r=.31. n=95, p<.01). Thus, we can’t be sure of the relative contribution of women minors and majors to an increased percentage of women overall in the mid- and upper-level courses. In short, more research is needed to understand this finding.

3.    Very few units specifically chose or rejected actions due to diversity. While many units (46.5%) stated they consider diversity impacts when choosing actions, very few (14.9%) chose actions to reduce impact on diversity and even fewer (11.4%) decided against possible actions out of concern for diversity. In addition, only one-third of units believe their existing diversity initiatives will compensate for any concerns with increasing enrollments, and only one-fifth of units are monitoring for diversity effects at transition points.
From a researcher’s perspective this has me happy to see: we used very different sampling approaches (they surveyed administrators, I surveyed professors in CS ed online communities), we used different analytical approaches (their quantitative vs. my qualitative), and we came to the same conclusion: CS departments aren’t considering diversity. This sort of triangulation doesn’t happen every day in the CS ed world.

CRA’s report gives us further evidence that CS departments should be considering diversity in how they decide to handle enrolment booms (and admissions/undergrad policies in general). If diversity isn’t on policymakers’ radars, it won’t be factored into the decisions they make.

This term, I’m taking my supervisor’s grad course on “Systems Thinking for Global Problems“. It’s been quite interesting so far. In our last couple of lectures, we have been talking about feedback loops.

And with that on my mind, I was particularly struck by a recent post on Mark Guzdial’s blog reposting a keynote by Eric Roberts:

[in response to increasing CS enrolments], 80% of the universities are responding by increasing teaching loads, 50% by decreasing course offerings and concentrating their available faculty on larger but fewer courses, and 66% are using more graduate-student teaching assistants or part-time faculty. […] However, these measures make the universities’ environments less attractive for employment and are exactly counterproductive to their need to maintain and expand their labor supply. They are also counterproductive to producing more new faculty since the image graduate students get of academic careers is one of harassment, frustration, and too few rewards. 
Computer science departments have, for decades, had cyclical enrolment. The sort of oscillation in enrolments is exactly the sort of thing you see in systems analysis when you have a balancing feedback loop with a delay in it.

Balancing Feedback Loops

Causal loops are used in systems analysis to show the relationship between variables in a system. If the variable _x_ increases when _y_ increases, and _x_ decreases when _y _decreases, they have a positive link. If, however, _x_ increases when _y _decreases, and _x_ decreases when _y_ increases, they have a negative link.

Let’s say we put a bunch of variables in a loop. If there’s an even number of negative links, then we have a reinforcing feedback loop: the system will increase (or decrease) exponentially until it hits some limit to growth. The negative links cancel each other out – so everything just reinforces everything.

But what if we have an odd number of negative links? The system tends towards an equilibrium – either it will asymptote to some value, or, more often, it will oscillate. Something will increase, another thing will decrease it, another will increase, and so on.

Consider:

As the number of students enroling in CS1 increases, the quality of student experience in a CS programme goes down for the reasons Eric Roberts covered above. And as the quality of the student experience goes down, the CS enrolments go down.

Eventually, enrolments and my abstract “student experience” will reach equilibrium – but it won’t be a static one. The enrolments will oscillate due to delay in the system: when CS enrolments increase, the quality of “student experience” won’t go down for a while yet, and “student experience” won’t immediately affect CS enrolments.

### Playing with the Model

Where do CS1 enrolments come from? Eric Roberts has elsewhere observed that CS1 enrolments at Stanford (and elsewhere) are positively correlated with the NASDAQ average – with some delay.

And with some delay, one would think (hope?) that the number of computer scientists in the economy would increase the NASDAQ average. Again, a balancing loop emerges with some delays in it:

Let’s think some more about the relationship between CS1 enrolments and the number of CS graduates.

The more CS1 enrolments there are, the more students in a CS department relative to the number of CS professors. Let’s call that “students / profs” for short.

As students / profs increases, the use of sessional lecturers increases. The number of courses a department offers decreases – which in turn decreases the amount of streaming in CS programmes. The teaching load for faculty increases, in turn hurting faculty satisfaction.

Even with the abstract “student experience” unpacked somewhat, we still see a balancing feedback loop. It doesn’t matter what path you take from “students/profs” to student retention – there’s an odd number of negative links.

Now, everything so far has assumed the number of faculty is fixed. But it’s not quite – with a (often substatial) delay, increased enrolments will lead to more faculty hirings. Let’s look a bit more at that:

Here we have our first reinforcing loop. The more students/profs, the more the teaching load – and down goes faculty satisfaction, more profs quit and go into industry, and then the ratio of students/profs gets even worse.

This feedback loop would continue on until it hits a limit to growth (no faculty to teach classes?) if it weren’t for the interacting effect of faculty hirings. The more faculty leave, the more faculty need to be hired. If we ignore our link between faculty turnover and the number of faculty, what we have here is a balancing feedback loop: profs who leave are replaced, and all is steady.

What could change that is university funding for hiring more faculty above the replacement rate. This is going to be institution-specific, so it’s hard for me to come up with a model here. (Even for any given institution, funding structures tend to be incredibly complicated.)

As I’ve been playing with these models, it’s striking me that it’s unlikely the cyclical enrolments in CS will stop. For them to stop, we’d have have either a nice steady tech economy – meaning interest in CS was steady – or we’d have to have a university funding structure where faculty can be rapidly hired in proportion to increasing enrolments.

Any ideas on how the models could be refined – or leveraged? This is just a first stab at modelling this. Let me know in the comments.