What Happened at the Hawthorne Plant
In the 1920s and 1930s, researchers at Western Electric's Hawthorne Works factory near Chicago conducted a series of experiments on worker productivity.
They varied lighting levels, work hours, break schedules, & physical conditions. The consistent finding: productivity improved with almost every change — including changes that made conditions objectively worse.
The conclusion: workers responded not to the specific changes but to the perception that management was paying attention to them, cared about their welfare, & was trying to improve their situation.
This is the Hawthorne effect: performance improves when people perceive that they are being observed & that changes are being made for their benefit, regardless of whether the specific change is actually beneficial.
Hamming saw this effect as particularly devastating for educational research:
> If you tell the students you are using a new method of teaching then they respond by better performance, and so, incidentally, does the professor. A new method may, or may not, be better, indeed it may be worse, but the Hawthorne effect... is likely to indicate here is a new, important, improved teaching method.
Implications for Educational Research
The Hawthorne effect creates a fundamental measurement problem for educational experiments. Any new teaching method — however mediocre or harmful — will appear to produce short-term gains simply because students and teachers perceive the change as evidence of care.
Hamming's conclusion: most educational experimentation fails to isolate genuine learning improvement from the Hawthorne-effect noise.
The ideal remedy in medicine: the double-blind experiment. Neither the patient nor the doctor knows which treatment is active. This controls for both patient response to perceived care and physician behavior change.
The problem in education: double-blind experiments are nearly impossible. Students know which method they are being taught with. Teachers know which method they are using. The Hawthorne effect cannot be blinded away.
The Grader Program
In 1960, during a sabbatical at Stanford, Hamming encountered one of the earliest uses of computers in education: a 'grader program' for programming assignments.
The system worked as follows: the professor submitted a correct solution program and specified the input variables, valid input ranges, & acceptable output tolerance. When a student submitted their program, the machine generated random admissible inputs, ran both programs, & compared the outputs. The student learned immediately whether their program was correct.
This automated feedback loop had properties a human grader could not easily provide:
Immediate feedback. The student received results within seconds of submission, when the thought process remained active.
Reproducibility. The same criteria applied to every submission. No grader fatigue, no favoritism.
Patience. The system processed the 100th submission with the same rigor as the first.
Scale. One professor's program graded an entire class simultaneously.
Branching Programs
A plain automated grader runs the same test sequence for every student. A branching program adapts the sequence based on student responses.
If a student answers correctly, the program advances to harder material. If the student struggles, the program branches to remedial content, alternative explanations, or worked examples. The path through the curriculum is not fixed: it depends on the student's demonstrated understanding at each step.
Hamming's question: does adaptive feedback produce better learning than a fixed sequence? The honest answer: the Hawthorne effect makes this extremely difficult to establish. Every study of branching programs shows gains — but so does every study of any new teaching method, by the Hawthorne mechanism.
What Makes a Teaching Tool Genuinely Better?
Hamming did not dismiss computer-aided instruction. He identified real advantages: immediate feedback, patience, & adaptation. But he was deeply skeptical of research claiming to validate new teaching methods, for the Hawthorne reason.
His implied standard: a teaching method deserves adoption when it survives controlled evaluation across multiple cohorts, with long-term learning outcomes (not immediate test scores), with researchers who were blind to the hypothesis, & with effect sizes large enough to exceed the known magnitude of the Hawthorne effect.
By that standard, almost no educational research of his era — and arguably little since — met the bar.
He also noted a perverse Hawthorne implication: the optimal teaching strategy might simply be perpetual novelty. If any new method improves performance because students perceive it as evidence of care, then constantly rotating methods would produce consistently elevated performance — not because any specific method is good, but because change itself is the active ingredient.