Importance of considering race in CS education research and discussion

February 17, 2020 at 7:00 am 7 comments

I was talking with one of my colleagues here at Michigan about the fascinating recent journal article from Tim Weston and collaborators based on NCWIT Aspirations award applicants, which I blogged about here. I was telling him about the results — what correlated with women’s persistence in technology and computing, and what didn’t or was negatively correlated.

He said that he was dubious. I asked why. He said, “What about the Black girls?”

His argument that the NCWIT Aspirations awards tends to be white and tends to be in wealthy, privileged school districts. Would those correlations be the same if you looked at Black women, or Latina women?

I went back to the Weston et al. paper. They write:

Although all respondents were female, they were diverse in race and ethnicity. Because we know that there are differentiated experiences for students of color in secondary and post-secondary education in the US, and especially women of color, we wanted to make sure we captured any differences in outcomes in our analysis. To do so, we created a variable called Under-represented Minority in Computing (URMC) status that grouped students by race/ethnicity. URMC indicated persons from groups historically under-represented in computing–African-American, Hispanic, or Native American. White, Asian and students of two or more races were coded as “Majority” in this variable. Unfortunately, further disaggregation by specific race/ethnicity was not possible due to low numbers. Thus, even though the numbers in the respondent pool were not high enough to disaggregate by specific race/ethnicity, we could still identify trends by over-representation and under-representation.

18% of their population was tagged URMC. URMC was included as a variable in their analyses, and their results suggest that being in the URMC group did not influence persistence significantly. If I understand their regressions right, that doesn’t tell us if the correlations were different by race/ethnicity. URMC wasn’t a significant factor in the outcomes, but that is not the same as thinking that those other variables differ by race and ethnicity. Do Black females have a different relationship with video games or with community than white females, for example? Or with Latina students?

While the analysis did not leave race out of the analysis entirely, there was not enough diversity there to answer my colleague’s question. I do agree with the authors that we would expect differentiated experiences. If our analysis does not include race, can we account for the differentiated experiences?

It’s hard to include race in many of our post-secondary CS ed analyses simply because the number of non-white and non-Asian students is so small. We couldn’t say that Media Computation was successful with a diverse student body until University of Illinois Chicago published their results. Georgia Tech has few students from under-served groups in the CS classes we were studying.

There’s a real danger that we’re going to make strong claims about what works and doesn’t work in computer science based only on what works for students in the majority groups. We need to make sure that we include race in our CS education discussions, that we’re taking into account these differentiated experiences. If we don’t, we risk that any improvements or optimizations we make on the basis of these results will only work with the privileged students, or worse yet, may even exacerbate the differentiated experiences.

Entry filed under: Uncategorized. Tags: , , .

Barbara Ericson’s analysis of the 2019 Advanced Placement CS data Call for participation at the 2nd Excited Summer School on Research in Computing Education

7 Comments Add your own

  • 1. Raul Miller  |  February 17, 2020 at 11:42 am

    I agree that a set grouping “oriental” and “black” would be unsatisfying. Especially, at the college level, if this also leaves unstated where the student’s family is located (zipcode or country).

    On the flip side, though, I think it’s also important to remember that these efforts take time. At minimum you need a semester, for a student to complete a class, and a new teacher for a subject isn’t going to start out with years of experience. So it’s not just about the numbers (though they are important), but also about pilot projects that give people access so that the people involved can start grappling with the issues.

    (And some of those will fail, so there’s also some need for people willing to grapple with the failure mechanisms, the excuses, the embarrassments, the politics, the unhappy and frustrated people and so on. And… finding people with the motivation and inclination for that kind of work can be difficult.)

    Of course you know this already (you’ve been expressing various aspects of this all along), but sometimes it’s worth restating the obvious.

  • 2. BKM  |  February 18, 2020 at 11:26 am

    Unfortunately, CS ed research tends to be done at the kind of schools that do not have a lot of minority students – big flagship schools with an engineering orientation. But there are schools out there with a good number of CS students from minority groups – and no, they aren’t just Morehouse and Howard. At my school, about 20% of our CS majors are black and another 20% or so identify themselves as Hispanic. There are a number of schools near us with similar statistics. The schools with higher numbers of CS majors from minority groups can be found in big cities, and very rural areas. They are typically not name-brand schools though, so researchers might have to look harder to find us.

    • 3. gasstationwithoutpumps  |  February 18, 2020 at 3:51 pm

      Even schools with a lot of minority students may end up not having many in CS. At UCSC, the campus as a whole has 31.1% URM, the school of engineering outside CS has 22.6% URM, but in CS only 15.3% URM. The biggest difference is that Asian-American and international students make up a much larger portion of CS than of other majors.

  • 4. Megan Lutz  |  February 24, 2020 at 11:59 am

    Although it wasn’t “statistically” significant, the focus on the p<0.05 is too narrow. Possibly due to the small sample size, the URMC variable didn’t pop in the analysis. However, it had a larger actual effect than some of the other variables that were discussed in the paper, so has your colleague mentioned (and I pointed out in your original post), membership in a minority group, even if it can’t be further disaggregated, is a meaningful and useful factor.

    We all must be thoughtful in our interpretations of our model results, and, as I teach my students, read the results from left to right (sample size, effect size, standard error, p-value), rather than home in straight on the p-value. The p-value omits and obscures important study context, such as that mentioned here.

    • 5. Mark Guzdial  |  February 24, 2020 at 1:02 pm

      You did raise this issue on the original post, and I should have acknowledged that in my post. My apologies, and thanks for raising the issue again.

    • 6. gasstationwithoutpumps  |  February 24, 2020 at 3:51 pm

      A small p-value may be meaningless with a small effect size and large sample, but a large p-value always means that the study was not big enough to confidently make statements about the effect.

      • 7. Megan Lutz  |  February 24, 2020 at 7:12 pm

        You are overstating this. Yes, we may get a large p-value because of the small sample, or it may be because of a small effect even with a large sample, or maybe we just have a lot of variability around a large effect. The point is, without considering the effect itself and only looking at the p-value, we don’t know. The 0.05 cutoff is arbitrary. In this case, and many others, the “large” p-value may be something like 0.07, which is still actually quite small when interpreted as a probability (as Fisher intended us too).

        Please see the ASA statement on p-value from 2016 (edited by Wasserstein and Lazar) and the 2019 special issue of The American Statistician (edited by Wasserstein, Schirm, and Lazar). An arbitrary threshold, of any size, leads to misleading science and interpretation. The p-value is a continuous statistic and should be treated and interpreted as such, i.e. it is one piece of evidence used to draw inference but cannot be interpreted in a vacuum. Similarly, no study should be treated as final or conclusive,but as contributing to a body of evidence about a phenomenon.

        Mark, sorry for soap boxing here. Thanks for the acknowledgement.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Trackback this post  |  Subscribe to the comments via RSS Feed

Enter your email address to follow this blog and receive notifications of new posts by email.

Join 9,030 other followers


Recent Posts

Blog Stats

  • 1,994,875 hits
February 2020

CS Teaching Tips

%d bloggers like this: