The Hawthorne effect is one of the most cited findings in social science, and for most of its life nobody went back to look at the numbers it was built on. The story is familiar from psychology textbooks, management seminars and workplace training decks: in the 1920s, researchers at Western Electric’s Hawthorne Works factory in Cicero, Illinois, changed the lighting for a group of workers assembling telephone relays. Productivity went up. Then, as the story goes, they dimmed the lights again, and productivity went up again. The conclusion drawn from this was that the workers weren’t responding to light at all. They were responding to being watched.
That idea, that people alter their behaviour simply because they know they’re part of a study, became a fixture of how researchers think about experiments involving human subjects. It has a name, a Wikipedia page and a permanent place in undergraduate methods courses.
What it did not have, for roughly eighty years, was anyone checking the claim against the original factory records.
The economists who went looking for the raw numbers
That changed with a paper by economists Steven Levitt and John List, “Was There Really a Hawthorne Effect at the Hawthorne Plant? An Analysis of the Original Illumination Experiments,” first circulated as NBER Working Paper 15016 in 2009 and published in the American Economic Journal: Applied Economics in January 2011. The two had assumed, like most people who cite the Hawthorne effect, that the original illumination-experiment data no longer existed. It turned out to be sitting on microfilm at the University of Wisconsin-Milwaukee and in Harvard’s Baker Library, largely unexamined since the experiments themselves ran between 1924 and 1927.
Getting hold of the records let Levitt and List do something that had apparently never been done at scale: run the actual daily output figures from the relay-assembly rooms against the actual dates of each lighting change, instead of relying on the summarised, second-hand version of the study that had been passed down through decades of textbooks. The relay experiments had involved small groups of women assembling telephone relays over several years, with output logged day by day as the plant’s engineers tried different combinations of brighter and dimmer lighting, sometimes alongside other changes to rest breaks and working hours in related studies at the same plant.
That level of granular record-keeping is exactly why the underlying data was worth recovering.
What the original records show
According to the paper, the popular version of the story doesn’t hold up. In Levitt and List’s own words, there is “little evidence of the type of Hawthorne effect widely attributed to these data when one subjects them to careful analysis.” The dramatic pattern usually described, output climbing every time the lighting changed, brighter or dimmer, as if the workers were responding to the mere fact of being observed, is not clearly present in the underlying figures.
One specific finding stands out. The researchers identified a straightforward day-of-week pattern: output was actually lower on Mondays and Saturdays than on other weekdays, rising toward the end of the working week, independent of anything happening with the lighting. Because every lighting change was introduced on a Monday, the modest rebound from the week’s low point could look, at a glance, like a lighting-driven jump — when it was really just the plant’s normal weekly rhythm.
The re-examination didn’t come back empty-handed, though. The paper also reports a handful of subtler, more limited patterns: productivity in the experimental groups ran somewhat ahead of the rest of the plant over the long term, output was a little higher on days when active experimentation was underway, and workers appeared to respond more to artificial changes in lighting than to natural variation in daylight. Levitt and List describe these as hints of a more modest kind of observation effect, not the clean, dramatic, appears-with-every-change pattern the textbook version describes.
A more nuanced finding than “the effect is fake”
It would be easy to summarise this as debunking, and some coverage at the time did exactly that. That overstates what the paper actually shows. Levitt and List are not arguing that people never change behaviour when they know they’re being watched. They’re arguing that this particular, famous dataset, the one that gave the phenomenon its name, does not clearly demonstrate the effect in the form it’s usually told. The original illumination experiments were also small and methodologically loose by modern standards, run on a handful of workers over a few years with no formal control group in the way a present-day study would require, which limits how much any reanalysis, in either direction, can settle.
The paper also wasn’t working in a vacuum. An earlier 1992 analysis by economist S.R.G. Jones had already raised doubts about the standard account of the relay experiments, and Levitt and List’s access to the original microfilm let them test those doubts against the primary data directly rather than reconstructed summaries. Even so, this is one re-examination of one historical dataset. It doesn’t settle whether some version of an observation effect shows up in other contexts, only that the specific 1920s factory data long used to illustrate it is thinner evidence than generations of retelling suggested.
How a shaky finding became gospel
What’s arguably more interesting than the statistics is how the myth survived so long unchecked. The Hawthorne effect was cited, restated and taught for decades by people who were working from summaries of summaries, not the underlying spreadsheets. Nobody had strong reason to doubt it. It was a tidy, intuitive story, and it fit a broader instinct that observation changes behaviour, which is true in plenty of other, better-documented settings. The specific factory data just wasn’t where that instinct came from, or at least not as cleanly as advertised.
The lesson isn’t really about lighting, or relay assembly, or even about Western Electric’s factory floor in Cicero. It’s about how long a claim can travel on the strength of its own repetition before anyone thinks to open the file it supposedly came from.