The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring

Benjamin S. Bloom (1913–1999) had been on the faculty of the University of Chicago since 1944 when he wrote this, and by 1984 he was one of the most-cited figures in American educational psychology. He had directed the committee that produced the 1956 Taxonomy of Educational Objectives — the ranking of cognitive demand from recall up to evaluation that is still taught to teachers as "Bloom's taxonomy" — and in 1968 he had proposed mastery learning, the procedure of testing a class for comprehension, giving corrective instruction to whoever has not reached a set threshold, and retesting before the class moves on. The 2 sigma paper is the late statement of that programme: a man near the end of a long career setting out the largest effect his line of work had found and asking the field to chase it.

The paper appeared in Educational Researcher, the membership periodical of the American Educational Research Association — a peer-reviewed journal, though Educational Researcher runs to commentary and field-shaping essays more than to primary experimental reports, and this piece is written as an address rather than a study. Bloom wrote it in 1984; the data it leans on are doctoral dissertations completed by his own students slightly earlier — Joanne Anania (1982, 1983) and Joseph Arthur Burke (1984), who ran controlled comparisons across fourth-, fifth- and eighth-graders in probability and cartography. That places Bloom himself one remove from the bench: he is synthesising, and the people who collected the numbers were people he was supervising.

What the dissertations found, and what the paper builds on, is a three-way comparison. Conventional whole-class teaching was the baseline. Mastery learning — the same class size, but with the feedback-and-correction loop added — moved the average student to roughly one standard deviation above that baseline. One-to-one (or one-to-three) tutoring, again with feedback and correction, moved the average student to about two standard deviations above it, so that the typical tutored student outperformed roughly 98 percent of the conventionally taught control group. The "2 sigma" is that second gap. Bloom's move is to name it not as a result but as a problem: tutoring at scale is unaffordable, so the open question he hands the field is whether some combination of cheaper "alterable variables" — he and his students screened for variables with effect sizes around half a sigma or larger — could be assembled to close the same distance.

The paper sits among the oldest anchors in this set, and it frames a question the other acquisition-side sources answer in their own currencies. Its diagnosis — that the loop of testing, feedback and correction is doing much of the work — is the same mechanism that Sweller's 1988 founding paper on cognitive load theory, "Cognitive Load During Problem Solving", and the Collins, Brown and Newman 1989 chapter on cognitive apprenticeship each reach by a different route. The 2 sigma figure has also drawn its own scrutiny: later writers have pointed out that the dissertation samples were small and short-run, that the tutored "mastery" condition bundled several interventions together, and that the two-sigma number is unlikely to survive at scale — so the figure travels more as a provocation than as a settled measurement.

Bloom's stake is professional and reputational rather than commercial. He sold nothing; the paper carries no product. But it is the capstone argument of a research school he had built and led for decades — mastery learning was his, the taxonomy was his, the dissertation writers were his students — and the paper's structure, which makes whole-class instruction the thing to be improved upon and his own corrective-feedback method the proven half-measure, is the structure his career had an interest in. The claim landing meant that the programme he had spent forty years on was confirmed.