On this page
In 1984, the University of Chicago education professor Benjamin Bloom published a striking result: students tutored one-to-one did about two standard deviations better than students taught in a regular class. That "2 sigma" figure is now quoted everywhere, from education policy to AI tutoring pitches. Here is what the study actually was, what later research found, and the more practical idea inside it.
What did Bloom's 1984 paper claim?
Bloom's paper, "The 2 Sigma Problem" in Educational Researcher, reported dissertations by two University of Chicago doctoral students, Anania and Burke. They compared three conditions:
Conventional: about 30 students per teacher, with periodic tests for marks.
Mastery learning: the same class size and teaching, plus formative tests for feedback, corrective work and a parallel test to check mastery.
Tutoring: a tutor for each student (or two or three students), with the same formative tests and correctives.
The studies covered students in grades four, five and eight, learning probability and cartography, with instruction limited to 11 periods over three weeks.
The average tutored student was "about two standard deviations above the average of the control class," above 98% of the students taught conventionally. The average mastery learning student was about one standard deviation higher, above 84%. About 90% of tutored students reached a level only the top 20% of the conventional class reached.
Bloom's "problem" was what to do next: one-to-one tutoring is "too costly for most societies to bear on a large scale," so could group teaching get close?
What does later research say about tutoring?
Tutoring helps, but two sigma is not the typical result.
A 2020 NBER meta-analysis by Nickow, Oreopoulos and Quan found an overall pooled effect of 0.37 standard deviations for tutoring programs, with stronger effects for teacher and paraprofessional tutors.
Kurt VanLehn's 2011 review found an effect size of 0.79 for human tutoring versus no tutoring. The widely held belief that human tutors produce an effect of 2.0 was not confirmed.
Writing in Education Next in 2024, Paul von Hippel concluded the two-sigma effects were "real and remarkable, but they were obtained on a narrow, specialized test, and they weren't obtained by tutoring alone." The tutored groups also had mastery learning's feedback and correction cycle. Von Hippel also notes, citing research by Kraft, that most education interventions produce effects of 0.1 standard deviations or less.
So 0.37 or 0.79 is not small. It is just not two.
Mastery learning: the part that scales
The overlooked half of Bloom's paper is mastery learning, which reached about one sigma in a normal class, often with the same teacher. A 1990 meta-analysis by Kulik, Kulik and Bangert-Drowns of 108 controlled evaluations found positive effects of mastery learning programs on exam performance, strongest for weaker students. It also found they may increase time spent on instruction, and that self-paced versions often reduced course completion in college.
The core cycle is simple: learn, take a short formative test, correct exactly what you got wrong, retest on a parallel version, then move on.
A mastery loop for one JEE or NEET chapter
Set the bar first. Decide what mastery means, for example 8 out of 10 on mixed questions with no notes.
Learn the chapter as usual: class, notes, NCERT, worked examples.
Formative test. Ten mixed questions, timed, closed book.
Correctives. For each mistake, write why it happened (concept, method or careless) and fix the cause: reread a section, redo a derivation, solve three similar problems.
Parallel retest. A new set of ten on the same ideas, a day or two later.
Move on only at the bar. If you miss it, run one more corrective cycle on the weak sub-topics only.
This takes more time at first, as the Kulik review found. The bet is that it pays back later, because the next chapter builds on something solid. Our post on mastery learning and adaptive practice covers how tools apply this loop, and how to analyse a mock test applies it to full papers.
Where Fermi fits
Running correctives by hand is the slow part. Fermi's AI tutor works out what you already know, builds a plan for the session and checks understanding as it goes. For JEE Main and Advanced, Fermi Testprep adds adaptive mocks and a study plan that updates based on mastery. Be wary of any tool, ours included, that promises Bloom's two sigma.
Take your last chapter test and write a corrective for every wrong answer before you move on.
