Structured Autonomy in Primary EdTech: A large-scale A/B test on Squla

A useful lens is the Self-Determination Theory (SDT): motivation is strongest when learners experience autonomy, competence, and relatedness. That is, when actions feel self-endorsed, progress is visible, and effort is acknowledged (Deci & Ryan, 2012). Classroom studies also suggest that well-designed choice can increase interest and persistence, especially when choices are meaningful rather than cosmetic (Patall, Cooper, & Wynn, 2010).

In Squla today, a learning path efficiently selects and sequences practice for group 6 to 8 of primary school. What’s largely missing is a practical, age-appropriate way for children to shape that sequence without losing the benefits of a strong default. In this project, we explore structured autonomy: small, meaningful, optional choices layered on top of the existing path. Concretely, we allow learners to choose the order of a short stack of upcoming tiles, while keeping a “I don’t want to choose” button for those who prefer the system’s path.

This blog summarises a large field experiment on Squla that evaluates that design. We ask: Does lightweight sequence choice increase engagement without harming accuracy, and does it influence self-reported intrinsic motivation? The next sections explain what we built, how we tested it, what we found, and what it means for future design and research. The research was conducted as part of a Master’s thesis by Alexis Ganciu, in collaboration with the University of Utrecht and assistent professor Jeroen Ooge.

We layered structured autonomy on top of Squla’s existing learning path. Before a session starts, children see a small stack of the upcoming tiles and can choose the order they’ll practice them in. If they don’t want to decide, a single tap on the “I don’t want to choose” button keeps the system’s sequence. Nothing else about the session changes: the same amount of items, the same system-wide difficulty progression, and the same mastery logic apply; only the order reflects the child’s preference.

The choice appears once, right before practice. This keeps the flow quick and avoids mid-exercise decisions. It also preserves a good default for learners who prefer not to choose (or are short on time). Behind the scenes, guardrails ensure that reordering never breaks spacing while learning, or mastery pacing; it simply lets a child say, “I’ll do this one first.”

Figure 1. Drag & Drop Design of Squla’s Learning Path Interface

We ran a field experiment on Squla within the learning paths for group 6 to 8. At first exposure, children were randomly assigned to Control (normal learning path interface) or Treatment (interface with a pre-session sequence choice screen). Assignment happened once per learner, and we analysed outcomes on an intention-to-treat basis. This means students are included in the final data only based on their random variation assignment, not on their engagement. The full sample comprised 4,688 learners (2,344 per group). For the motivation measure, a sub-sample completed a short, in-platform intrinsic motivation questionnaire (child-adapted Intrinsic Motivation Inventory (IMI) (Ryan, 1982, 2022)). After strict quality checks for contradictory responses, 1,098 remained for that analysis.

Our primary outcomes were:

Analytically, we used Welch t-tests for mean differences (e.g., accuracy), negative binomial regression to model the relationship between IMI and total goals (in control), motivated by over-dispersion, and a Welch t-test for the IMI distribution. We report effect sizes with 95% confidence intervals. All estimates reflect everyday use: the underlying content, difficulty progression, and mastery logic stayed the same across conditions.

More completion, same quality
Learners that were offered the Treatment interface completed more daily goals than those on the standard path. This improvement was statistically significant under our conducted analyses. At the same time, accuracy did not differ meaningfully between groups; an important “no harm” result for learning quality.

Motivation unchanged
On the child-adapted intrinsic motivation scale (short IMI), we observed no statistically significant difference between groups. Within groups, however, higher motivation scores were associated with more completed goals, consistent with the idea that perceived autonomy and competence work together.

How kids used the feature
Nearly half of learners used the “I don’t want to choose” option at least once. Opt-out rose with longer paths, suggesting many children prefer the skipped choice default when the choice set is larger or when they want to start quickly. Among those who did choose, we saw diverse preferences for what to do first/last (e.g., topic or difficulty). This reinforces that small, optional choices capture meaningful variation we cannot see in a purely adaptive flow. In general though, it can be observed in Figure 2 that the first and last tiles of all learning paths have a darker purple color, meaning that the occurrence of a hard tile was higher at that position.

Figure 2. Heat Map Showing Hard Tile Ratio Differences

Heterogeneity
Effects on daily goals were directionally similar across ages and subjects; we didn’t observe a consistent subgroup where accuracy worsened under choice. Where differences appeared, they were small and not robust across specifications.

This experiment shows that a small, pre-session autonomy moment (letting children choose the order of upcoming practice) increases completion without reducing accuracy. In other words, the right kind of choice can make primary-age learners finish more of what they start, while keeping learning quality intact. Self-reported intrinsic motivation did not differ between groups, which fits a familiar pattern in Self-Determination Theory: autonomy tends to work best alongside clear progress cues (competence) and supportive acknowledgement (relatedness) (Deci & Ryan, 2012; see also Patall, Cooper, & Wynn, 2010).

How to read this
These are field results from everyday use on Squla, not a lab study. The effect on completion is statistically reliable and practically useful; the “no harm” finding on accuracy is equally important. At the same time, we’re careful not to over-claim: this intervention adds choice over sequence only; it doesn’t change content, difficulty, or feedback. Short motivation scales may miss slower-building shifts, and novelty effects are possible. The results of this experiment can be treated as a signal that structured autonomy is worth keeping in our future designs.

Deci, E. L., & Ryan, R. M. (2012). Self-determination theory. In P. A. M. Van Lange, A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of social psychology (Vol. 1, pp. 416–436). Sage.

Patall, E. A., Cooper, H., & Wynn, S. R. (2010). The effectiveness of choice in the classroom: A meta-analysis. Journal of Educational Psychology, 102(4), 896–915. https://doi.org/10.1037/a0019543

Ryan, R. M. (2022). Intrinsic Motivation Inventory (IMI): Scale description [PDF]. Self-Determination Theory. https://selfdeterminationtheory.org/wp-content/uploads/2022/02/IMI_-Complete.pdf

Want to get in touch about this blog? Or just want more information? You can reach out to me at jasper@futurewhiz.com. I’d love to hear from you!