Between 1924 and 1927, at a Western Electric plant in Cicero, Illinois, a group of engineers ran an experiment on light. One room was fitted with adjustable lamps, another kept its normal lighting, and workers in both rooms spent their days assembling telephone equipment while someone counted what they produced. Output went up when the light got brighter. Then the light was lowered, and output went up again. The engineers kept going until the room was dim enough that people struggled to see their own work — 1.4 foot-candles, about 15 lux, a level the retellings like to call moonlight — and only there did the numbers finally fall. The conclusion wrote itself, and it was a beautiful one: light does not matter. What matters is that somebody has come to watch.
That is the Hawthorne effect as it is taught in almost every management textbook on earth. It has been used to explain why productivity jumps during a study, why patients improve in clinical trials, why people behave differently on camera, and why you should be suspicious of any observation that changes the thing observed. The plant gave the effect its name, and the studies around it became the founding myth of industrial psychology.
🧠 The Room Where Output Fell
The follow-up studies are where the myth got its meat. In 1927, researchers picked two women off the relay assembly line and let them choose four more, and the six of them moved into a separate room that measured their output mechanically, relay by relay, through a counting chute. The baseline had been recorded in secret for two weeks beforehand, while the women had no idea anyone was keeping score.
Then came the changes. Two five-minute breaks, then two ten-minute breaks, and output rose. Six short breaks instead of two long ones, and output fell, apparently because being interrupted six times wrecks your rhythm. Soup and coffee in the morning, snacks in the afternoon, and output rose. Clocking out at 4:30 instead of 5:00 with no Saturday work, and output rose. The detail that made the whole theory is this one: when the researchers put things back the way they had been, output often rose anyway. The six women had become a team, they had a supervisor who talked to them about their numbers, they were paid noticeably more than their co-workers, and they were, unmistakably, the subjects of something.
Sounds like the effect is real. Stay a minute longer, because the same plant produced the opposite result. In the bank wiring room, fourteen men wired telephone switching equipment on a group incentive plan, and the researchers watched productivity decline. The men had worked out that high output would give the company a reason to cut the base rate, so they held output at a level the group considered fair, enforced that level on each other, and gave the supervisors identical answers whether or not the answers were true. Same plant, same decade, same sense of being watched — one group produced more and one group produced less. What the workers were guessing about the observation, and what they thought it would cost them, decided the direction.
🤔 The Data Were Never Analyzed
Here is the part that should have been a scandal for seventy-five years and mostly wasn’t. The illumination data were never formally analyzed. The report was never published, reportedly because the answer disappointed the people who paid for it: the study would have recommended a basic lighting level of seven to ten foot-candles, which is not the scientific endorsement the electrical industry was hoping for from a study about light. By the 1990s, the numbers were widely assumed to have been thrown away.
In 2009, Steven Levitt and John List went looking for them anyway, and found them. What they found, in their words, is that “existing descriptions of supposedly remarkable data patterns prove to be entirely fictional.” The pattern of output rising after every single lighting change is not in the data. It is a composite of two much duller facts. First, every lighting change was installed on a Sunday, the one day the factory was closed, so “the day after the change” was always the same weekday. Second, output at that plant had a strong weekly rhythm, with Saturday running very low. Line those two up and a naive reading shows a jump after every change. Add in a summer dip — output fell in all three rooms whenever the experiments paused, because it was summer, not because the watching had stopped — and a control room that rose right alongside the experiment room, and the famous effect disappears.
What survived is small and slow: output was slightly higher two to five days after a change, which the authors call a hint of a subtler Hawthorne effect rather than the theatrical one in the textbooks. They also propose a cleaner test for it — compare how strongly people respond to variations the experimenter creates against how strongly they respond to variations that just happen, and count only the excess. Meanwhile the plant’s overall output per worker was rising about 1.4 percent a year throughout, which is the kind of thing that quietly explains a lot of “effects.” Richard Nisbett’s verdict on the original is the line I keep thinking about: it is “a glorified anecdote.” Once you have the anecdote, you can throw away the data. Which is, apparently, roughly what happened.
🔗 Why Your Charts Lie
Anyone who measures change should feel this one in the spine. You ship a feature, the numbers go up, and the story writes itself the same way the lighting story did. But the calendar moves too: weekdays versus weekends, seasonality, the first week of a launch, the difference between a Tuesday reading and a Friday one. Before you credit the thing you changed, check whether a week where you changed nothing has the same shape. The Hawthorne study’s real lesson is not that people perform for an audience. It is that the audience and the calendar arrive together, and the calendar is cheaper to blame.
There is a version of this that matters more than dashboards, though. The strongest surviving piece of the Hawthorne story is not the lighting; it is that people behave differently when they know they are being studied. That is a general fact about being asked about yourself, and it is worth holding onto whenever you read a survey, a self-report, or a research interview — including your own answers. Someone who is being asked about their life will give you a truer-sounding, more coherent, slightly different version of it than the one that runs on an ordinary Tuesday. Treat the answers as data. Just don’t treat them as the person. That, more than the lamps, is what the factory actually demonstrated.
🎲 The Calendar Test
The next time you want to know whether a change worked, do the boring forensic thing. Write the delivery date next to the day of the week, put it up against a comparable week in which you changed nothing, and see how much of the movement survives. On a factory floor in 1926, that exercise erased the most famous productivity effect in the social sciences.
And for the trivia shelf: the effect was not named by Elton Mayo, whose name is usually attached to the whole affair. The term was coined in 1953 by John R. P. French, and it was Henry Landsberger’s 1958 book Hawthorne Revisited that turned the interpretation into the version everyone repeats — a story named thirty years after the fact, about data that nobody ran.