Mastery learning and Bloom's two-sigma problem

Evidence grade: STRONG for the core effect, MIXED on magnitude. Meta-analyses consistently find mastery approaches outperform conventional instruction, with the largest gains for weaker students. Bloom's famous "two sigma" tutoring result itself is best read as a challenge and an upper bound, not a guaranteed effect size — replications of one-to-one tutoring find substantial but smaller advantages.

Conventional instruction holds time constant and lets learning vary: everyone gets three weeks on fractions, then everyone moves on — the children who reached 60% carry their missing 40% into every future topic that assumes it. Mastery learning inverts this: hold learning constant (define what "got it" means, e.g. ~90% on a mastery check) and let time vary. Nobody proceeds past a foundation that hasn't set.

The evidence, briefly

Benjamin Bloom framed the modern case in "The 2 Sigma Problem" (1984, Educational Researcher): in his group's studies, students taught one-to-one with mastery methods performed around two standard deviations above conventional classrooms — better than 98% of the control group. Bloom's point was not "hire a tutor for every child"; it was a research program: find methods that recover as much of the tutoring effect as possible at classroom scale. Mastery learning was his leading candidate, recovering a substantial fraction of the gain on its own (Bloom reported group-based mastery learning around one sigma in his studies).

The broader literature supports the direction while moderating the numbers:

Kulik, Kulik and Bangert-Drowns' meta-analysis (1990, Review of Educational Research) found mastery programs raised examination performance meaningfully across dozens of studies — with the strongest gains for lower-achieving students, and positive effects on attitudes toward the subject.
Guskey's decades of implementation work documents the same pattern in practice, along with the two classic implementation failures: mastery thresholds set low enough to be meaningless, and "corrective" instruction that just repeats the original teaching more slowly.
Later reading of the tutoring literature (e.g., VanLehn, 2011, Educational Psychologist) finds real one-to-one and intelligent-tutoring effects closer to 0.75σ than 2σ — still enormous by educational standards, and consistent with Bloom-as-upper-bound.

Why it matters more in cumulative subjects

Mastery learning matters most where knowledge stacks. Digital skills stack relentlessly: you cannot manage files without mouse and keyboard fluency, cannot use a spreadsheet without typing, cannot reason about formulas without cell references. A child pushed past an unset foundation doesn't just miss one topic — they pay interest on the gap in every subsequent lesson, which is exactly why mastery's gains concentrate in the students conventional pacing leaves behind.

The structural consequence: a curriculum should be a prerequisite graph, not a line. "What must be true before this makes sense?" is a per-skill question, and honest answers produce a web — some skills gate many others (typing), some are leaves. A linear scheme of work is a graph flattened by administrative convenience, and the flattening is where the gaps hide.

What the evidence doesn't say

It doesn't say every child can learn everything in reasonable time; it says time-to-mastery varies far more than conventional schedules admit, and that most of the variance is addressable.
It doesn't endorse "seat time until you pass" with no change in instruction. The corrective loop — a different explanation, a different task, a smaller step — is where mastery programs succeed or die (Guskey's central point).
The two-sigma figure specifically should be quoted as Bloom's challenge, not as a promise. We grade claims here, including famous ones.

In the classroom

Define mastery per skill, in advance, and make the check performance-based where possible ("do the thing", not "answer questions about the thing").
Map your prerequisites honestly — even a rough graph exposes which "one lesson" skills (typing!) are actually load-bearing for a whole year.
Build the corrective loop as a first-class citizen: a failed check triggers a different approach, not a re-run at half speed.
Let the fast lane exist. Mastery structures liberate quick students too — placement checks let them prove a skill and skip the seat time (see placement diagnostics).

How Wiz Kids applies this

The curriculum is literally a prerequisite graph — 150+ skills with explicit edges, machine-checked to be acyclic. A lesson unlocks only when its skills' prerequisites are passed; passing is per-skill, demonstrated by doing; a placement trial lets late joiners prove skills and skip ahead; and spaced review keeps verifying that "mastered" stays true (mastery decays — spacing is the maintenance contract).

References

Bloom, B. S. (1984). The 2 sigma problem: The search for methods of group instruction as effective as one-to-one tutoring. Educational Researcher, 13(6).
Kulik, C.-L. C., Kulik, J. A., & Bangert-Drowns, R. L. (1990). Effectiveness of mastery learning programs: A meta-analysis. Review of Educational Research, 60(2).
Guskey, T. R. (2010). Lessons of mastery learning. Educational Leadership, 68(2).
VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4).

© Glu IO Pty. Ltd. — Wiz Kids (wiz.kids). Link freely; republication requires permission — see terms. Found an error in our reading of the research? We correct fast: tell any teacher piloting Wiz Kids.