Technical Perspective: Is There a Geek Gene?

Many computer science teachers have told me that some students just get computer science — and others do not. We certainly have a lot of evidence that students enter introductory computer science courses with big differences in skills. Some students have already had years of programming experience, while others have never programmed at all. The question is whether those gaps close or diverge further.

Are the differences between students in CS classes explained by experience and background, or are the differences innate? Innate difference among CS students has been dubbed the Geek Gene. Many CS teachers believe a Geek Gene (or something similar) is necessary to succeed in CS, and not everyone has it. A 2007 study found 77% of surveyed CS faculty strongly disagreed with the statement: "Nearly everyone is capable of succeeding in computer science if they work at it." CS teachers point to a bimodal distribution of grades in their CS classes as evidence for its existence. Some students "get it" and do well, while others do not, which appears as two peaks in a grade distribution. Is it real? Are some students born to be computer scientists, and are others unlikely to succeed because they do not have the right stuff?

There is a long history of researchers trying to discover the variables that predict student success in computer science class. Probably the most famous of these had the odd title "The Camel has Two Humps." It was never published in a peer-reviewed venue, was not replicated in multiple attempts, and was later retracted — but its power persists because it rings true to many. The underlying research questions are important: What skills and knowledge predict success in CS? How can we measure them? Can we teach any missing but necessary skills and knowledge explicitly?

Are some students born to be computer scientists, and are others unlikely to succeed because they do not have the right stuff?

The following paper "Evidence that Computer Science Grades Are Not Bimodal" by Elizabeth Patitsas, Jesse Berlin, Michelle Craig, and Steve Easterbrook takes aim at belief in the Geek Gene. If there is a Geek Gene, one would expect bimodal grade distributions in CS classes. If grades are not bimodal, perhaps the Geek Gene is just a figment of teachers' biases. The authors explicitly check a large corpus of grade data for bimodality, and then run a study with CS teachers as participants to determine if belief in innate differences may itself explain why teachers see bimodality in grades. This paper is important for showing student performance may not be bimodal and for offering evidence of an alternative, plausible hypothesis.

• First, they review all final grades in every undergraduate class from 1996 to 2013 at the University of British Columbia (UBC). This dataset included over 700 sections and over 30,000 grades. 85% of the grade distributions were normally distributed.
• Then, they run a deception study. They recruit 60 CS instructors. Half were asked to agree or disagree with statements like in the 2007 study: "Some students are innately predisposed to do better at CS than others," and then shown a set of histograms representing grade distributions. The instructors were asked to label which were bimodal and which were not. The other half of the instructors saw the histograms first, and then were asked to respond to the statements. The deception was that all the histograms were generated from normal distributions, yet participants who agreed with the Geek Gene statements were more likely to identify the distributions as bimodal.

As in all empirical studies involving humans, we can disagree about the details. How UBC counts withdrawing from a class or failing a class in the grade distribution is probably different than many institutions. There is some possibility that participants might have seen the histograms, then gone back to change their answers on the statements. There can and should be more studies on these questions.

This paper does not prove there is no Geek Gene. There may actually be bimodality in CS grades at some (or even many) institutions. What this paper does admirably is to use empirical methods to question some of our long-held (but possibly mistaken) beliefs about CS education. Through papers like these, we will learn to measure and improve computing education, by moving it from folk wisdom to evidence-based decision-making.