The Self-Directed Disposition: What Computing Students Say

Mihaela Sabin, Amruth N. Kumar, Bonnie K. MacKellar, Renée McCauley, Tammy VanDeGrift and Stephanos Matsumoto — ITiCSE 2025 (ACM, Nijmegen, June 2025). The spoke’s first T1, and the first source here not published by someone with a product beside it. It asks what self-direction actually consists of, from the students doing it — which is the disposition every curriculum in this corpus assumes and none of them examines.

What was done

496 written reflections from 264 students across 18 sections of five undergraduate computing courses at four US institutions (CS1 151 participants, Programming Languages 81, Software Design 18, CS2 10, Data Structures 4), collected over four consecutive terms. Students answered reflection prompts — some as plain questions, some as vignettes — about how they had been self-directed on an assignment, or why they had not.

Analysis is Mayring qualitative content analysis with the constant comparative method: three researchers coded all 496 responses independently, then reconciled to consensus with the full six-person team. They started from five deductive categories built in earlier work, found the data did not fit them, and rebuilt inductively. Credibility rests on three declared triangulations — across data sources (four institutions, four terms), across method (three exercise types), and across investigators (six researchers at different institutions).

The eight behaviours, and what the frequencies say

380 of the 496 responses described a self-directed behaviour. Counts in parentheses:

The distribution is the finding, not the list. “Utilizing external resources” is nearly four times the next behaviour and larger than planning, self-assessment and review combined — students overwhelmingly equate being self-directed with going and finding other material. The two metacognitive behaviours a curriculum designer would most want are the rarest: planning ahead 23, and reviewing your own work against stated expectations 3.

That maps uncomfortably onto this spoke. Every artifact here — coding-interview-university, awesome-system-design-resources, developer-roadmap — is an external resource, and the meta-curriculum form is external-resource-finding productized. The corpus is built almost entirely around the one behaviour students already do most, and supplies little for the two they do least.

Why students said they were not self-directed

Twenty-seven responses reported not being self-directed. Five categories:

The largest impediment is the assignment itself. A quoted student: “For these projects, you have to develop a program that shows exactly like how the professor has it. So there is really no room to add new tools and techniques that were not discussed.” A second, sharper one the authors call radical but not irrelevant: “by being given an assignment for a grade any action is no longer self-directed, it is in pursuit of a grade.”

The authors’ recommendation is a genuine tension rather than a tip: opening assignments up to invite autonomy will increase the second-largest impediment, unsuccessful effort. Their proposed resolution is scaffolding that is deliberately faded over time, plus specifications-based grading that scores competence gain rather than points.

What it does and does not settle for this spoke

It closes growth edge #1 — peer-reviewed CS-education research was named exactly, and this is it. It bears directly on the “does anyone finish?” hole, though it does not fill it: this measures what students say they do inside a course with an instructor and a deadline, not whether anyone completes a 352k-star repository alone.

The relevance runs deeper than the tier. Every curriculum in this corpus is assignment structure with the grade removed. The study’s largest impediment is a structure that leaves no room to depart from it — which is what a step-by-step roadmap is — while its largest enabler is exactly the resource-seeking those roadmaps replace by pre-selecting the resources. Neither reading is tested here, and the study cannot test them: its setting is a classroom.

Limitations, stated by the authors. Responses were not disaggregated by course, course level, exercise type, or any student characteristic (race/ethnicity, gender, socioeconomic status, disability, academic standing), which the authors name as the limit on transferability. No synopsis of the target assignment was recorded, so responses could not be interpreted against what was actually asked. The setting is US-centric, and the authors flag that countries with work-based or apprenticeship requirements may differ. Four of the six authors taught 12 of the 18 sections they collected from.

T1 — peer-reviewed, ITiCSE (32% overall acceptance rate), IRB-approved at the participating institutions, method and limitations published. Qualitative and self-report by design: it records what students believe self-direction is, which is the stated research question, not whether those behaviours produce learning.