Evidence That Computer Science Grades Are Not Bimodal
Patitsas, E., Berlin, J., Craig, M., & Easterbrook, S. · 2016
grade Ccritiqueindependentmixed
Sample
778 final-grade distributions from one large research university; a priming experiment with 53 computer science professors
Population
Undergraduate computer science courses at one Canadian research university, plus CS faculty as experimental subjects.
Design
Two studies bolted together. The distributional analysis is single-institution and descriptive — it cannot rule out bimodality elsewhere, only show it is not the norm where it was tested. The priming experiment is a genuine small randomised experiment, but n = 53 and the stimulus is histogram-labelling rather than a real teaching decision. Read at abstract depth.
Key findings
Only 5.8% of 778 computer science final-grade distributions passed tests of multimodality. Professors randomly primed with the folklore that CS grades are bimodal were more likely to label ambiguous histograms bimodal, and so were professors who believed some students are innately predisposed to do better at CS. The "geek gene" does not appear in grade distributions; it appears in instructors' perception of grade distributions.
Genetic confound
Not applicable in the usual sense — this source is an argument ABOUT a hereditarian folk belief, and it shows the belief is not visible in the data it is supposed to explain. Note the limit: a unimodal grade distribution is not evidence against heritable variation in programming aptitude, only against the specific claim that the population splits into two kinds of people.
Replication notes
The distributional claim is supported by Basnet et al. (2018), not read here. The priming experiment has not been independently replicated.
DOI / URL
10.1145/2960310.2960312
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Proportion of CS final-grade distributions that are multimodal | proportion | 5.8% of 778 distributions | administrative | single cross-section of course records | none | not-applicable | attainment |
| Labelling ambiguous distributions bimodal after priming | experimental contrast | primed group significantly more likely to label bimodal (n = 53) | researcher-designed | single session | business-as-usual | end-of-treatment | behaviour |
Cited by
- Does teaching programming actually teach programming — and does the method matter?moderate supportconf: mediumgc: low