011 min
The Hawthorne effect at a glance
- What it is: People act differently when they know they are observed, so measured behaviour can differ from normal behaviour.
- Origin: Named after Western Electric's Hawthorne Works near Chicago. A 1953 methods textbook first used the term.
- Confused with: The observer-expectancy effect, where the researcher's hopes bias the result.
- Watch for: The popular lighting story is mostly wrong. The recovered data show no jump in output at each lighting change.
021 min
Why the Hawthorne effect matters
A measurement that changes the thing it measures cannot be read at face value. Three everyday decisions depend on this.
A PM watches five people complete a task in a moderated session and decides the flow is easy. A manager announces a quality audit and treats the audit score as normal practice. A research team runs a trial, sees good results in the study group, and expects the same results in routine use.
In each case the people knew they were being studied. If that knowledge alone changed what they did, the result describes behaviour under observation, not behaviour in daily use. The cost is a wrong decision: you ship a flow that real users find hard, you report a standard that staff do not meet on an ordinary day, or you promise an improvement that disappears outside the study.
031 min
What the Hawthorne effect actually claims
The Hawthorne effect is a change in behaviour caused by awareness of being studied, not by the thing being tested. To measure it, you need two groups that differ only in whether they know they are observed. Most studies do not have that comparison.
The original definition was narrow. One medical trial paper quotes it as "an increase in worker productivity produced by the psychological stimulus of being singled out and made to feel important". Researchers later widened the term to mean any change in outcome from taking part in research.
The most common explanation comes from social psychology. Awareness of being observed makes people form beliefs about what the researcher expects. Conformity and social desirability then change behaviour to match those expectations. Social desirability is the wish to be seen in a good light.
Three different things hide under the one label:
- Being observed. An observer is present or the person knows they are measured.
- Receiving extra attention. Participants get more contact, check-ups or feedback than they would normally get.
- Taking part in research. The person has agreed to join a study, answers questions or repeats a test.
These can have different causes and different sizes. A study that does not say which one it means cannot say what the Hawthorne effect did.
041 min
Where the Hawthorne effect comes from
The original studies took place at Western Electric's Hawthorne Works near Chicago, between 1924 and 1933. Workers there made telephone equipment.
The first set was the lighting experiments. The National Research Council started them in 1924 and asked a narrow question: does better lighting raise worker productivity? A second series of experiments followed. Western Electric brought in academic consultants, most prominently Elton Mayo, in 1928.
The name came later. The term was first used in a methodology textbook in 1953, about twenty years after the experiments ended. So the name was applied to the plant's results afterwards.
052 min
What the Hawthorne lighting data show
The popular story
A widely repeated textbook summary says that each time a change was made, worker productivity increased. Popular accounts say that even changes that made the room dimmer raised productivity. The conclusion drawn is that the attention itself, not the lighting, raised output.
What the data show
Experts thought the lighting data had been destroyed. Levitt and List found them in two archives and analysed them for the first time. They covered three departments of women who primarily assembled relays and wound coils of wire, with output counted as units completed per unit of time. In the third year, with all natural light removed, the artificial light was lowered step by step to 1.4 foot-candles for one day.
Levitt and List found no systematic evidence that productivity jumped whenever the lighting changed.
| Part of the popular story | What the recovered data show |
|---|---|
| Output jumped whenever the light changed | No systematic jump at lighting changes |
| Output rose on the day of each change | It rose on Mondays with a change and on Mondays with no change, in the same way |
| The experiment itself raised output | In one model, output was 3 to 4 percent higher while experiments ran. The result disappeared with stricter controls |
| Gains faded when experiments paused | Output in all three rooms declined during the pause, which fits a seasonal pattern better |
The first year looked like the story. The later years did not fit. In the second round only the first room was in the experiment, and its output peaked around 20 percent above the starting level. When experimenting paused again, all three rooms declined, as in the earlier summer. That points to a seasonal pattern, not to a reaction to the experiment.
Output also has a strong weekly pattern. Saturday output is very low and Monday output is lower than other weekdays. Every lighting change was made on a Monday, so each change had to be compared with Mondays that had no change. Output rose on Mondays when the light was changed, but it rose in the same way on Mondays with no change.
The authors then tested for a longer effect. In one model, output was 3 to 4 percent higher while experiments ran. The result disappeared when the authors added controls for room, month and year.
Levitt and List did find weak signs of something else. Workers seemed to respond more to the experimenters' changes in light than to natural changes in light. The authors call this circumstantial. It is a much smaller claim than the one in the textbooks.
062 min
Two later cases of the Hawthorne effect
The lighting story is about a factory and has no clean comparison group. A hospital study gives a cleaner one.
Researchers tracked every use of hand rub and soap in two hospital units for 8 months with a real-time location system. The system also tracked where the hand hygiene auditors walked. That let the researchers compare the same kind of dispenser with and without an auditor in view. Staff knew that hand hygiene was monitored, but not that the Hawthorne effect was being studied.
Dispensers visible to auditors had 3.75 events per dispenser per hour, against 1.48 for dispensers not visible at the same time and 1.07 for the same dispensers the week before.
| Condition | Hand hygiene events per dispenser per hour |
|---|---|
| Dispenser visible to an auditor | 3.75 |
| Dispenser not visible to the auditor, same time | 1.48 |
| Same dispensers, the week before | 1.07 |
| Same dispensers, 1 to 5 minutes before the auditor arrived | 1.50 |
The ratios are easy to check. 3.75 divided by 1.48 is about 2.5. 3.75 divided by 1.07 is about 3.5. The authors summarise this as approximately threefold. The rate rose after the auditor arrived, not before, and there were no significant changes inside patient rooms.
A second case changes one variable: how intensely people are followed. In a randomised trial of Ginkgo biloba for dementia, a research team compared an intensive follow-up schedule with a minimal one. The trial had 176 participants. Participants were randomly assigned to a schedule, so the groups were comparable at the start. At six months the intensive group had a better score on the ADAS-Cog, a test of cognitive function. The difference was about two points, similar in size to the effect reported in trials of dementia drugs.
The two cases differ in one respect. In the hospital, a visible observer changed behaviour within minutes. In the trial, extra contact over months changed a test score. The label covers both, but the mechanisms are not the same, and that is why the term needs care.
072 min
How the Hawthorne effect shows up in product, design, and engineering work
The effect shows up wherever a team collects behaviour data from people who know they are part of the collection. The three scenarios below are illustrative.
A designer runs a moderated usability test. The participant knows the session is recorded and that the designer wrote the prototype. They try harder than usual, and they avoid saying the product is confusing. The designer sees a task completed and reports success. The decision that changes: treat the result as a best case. Check it against what A/B tests or analytics logs show for the same task from users who did not know they were being watched.
A PM pilots a feature with a hand-picked group. The group gets extra support calls and a direct chat channel. Their retention rises. Two things changed at once: the feature and the attention. If the PM rolls the feature out without the attention, the gain may not appear. Also, correlation and causation apply: a hand-picked group may differ in ways that have nothing to do with attention. The decision that changes: give a control group the same attention, without the feature.
An engineering manager announces that code review turnaround will be tracked. Turnaround improves the week after the announcement. Whether it stays improved depends on why it changed. If the only cause was awareness of the tracking, the number may drift back once people stop thinking about it.
A/B tests usually avoid this problem for a structural reason. Users in both groups are normally unaware of the experiment, so neither group is changed by being watched. This is reasoning, not a finding from the sources above, and it fails when users can tell which version they have been given.
082 min
How to guard against the Hawthorne effect
When to worry
Worry when three things are true together. The outcome is behaviour or self-report, not a physical measurement. The people know they are observed. The effect you care about is small, or the stakes of the decision are high. If any of the three is missing, the risk is lower.
How to reduce it
Measure without an observer where you can.
The hospital study used electronic counts of dispenser use, not a person watching. Server logs and product analytics work the same way.
Give every group the same attention.
In a controlled trial, extra attention reaches both arms. The comparison between arms stays fair, but the absolute level can be inflated in both, so do not copy that level into routine use.
Compare observed and unobserved periods or places.
The hospital study did this by comparing dispensers in view and out of view at the same time.
Avoid asking one group more questions than another.
Repeated questions and repeated tests can change behaviour and scores by themselves. Use sampling rules that pick who is observed, and do not hand-pick.
Say which of the three meanings you suspect.
Observed, extra attention or taking part in research each calls for a different check.
A fixed settling period before you start counting is sometimes suggested, because novelty may fade. The sources here give no standard length, so treat it as an assumption to test, not a rule.
091 min
Where the Hawthorne effect goes wrong in practice
Using the label to explain a disappointing result
A systematic review says the most common use of the term is probably as an after-the-fact reading of unexpected findings, particularly disappointing ones such as null results in trials. That is not a test. To claim a Hawthorne effect, show a comparison between observed and unobserved conditions.
Repeating the lighting story as a fact
The claim that every lighting change raised output, even dimming, is the version most people know. The recovered data do not support it. Cite the story only as a story.
Assuming a fixed size
Levitt and List say the effect is not guaranteed, and that its presence and size depend on features of the setting. The systematic review agrees that little can be securely known about its size. No standard percentage exists, so do not subtract one from your results.
Confusing it with a learning effect
A person who repeats a test gets better at it. In the dementia trial, the authors say the gain may be due to learning from repeated testing or greater familiarity with the research process. Being observed is not the only explanation for a gain between measurements.
101 min
The Hawthorne effect vs. nearby concepts
The observer-expectancy effect is the nearest neighbour. The deciding fact is whose behaviour is biased and why.
| Hawthorne effect | Observer-expectancy effect | |
|---|---|---|
| Whose behaviour changes | The participants | The researcher, or the participants responding to the researcher's cues |
| What causes it | Awareness of being studied | The researcher's expectations about the result |
| Present when the researcher has no hypothesis | Yes | No |
Two other terms are often mixed in. Confirmation bias is about how a person reads evidence, not about how participants act. Survey bias covers distortions in how questions are asked or answered. It overlaps with the Hawthorne effect when people give answers that please the researcher.
111 min
Where the Hawthorne evidence is contested
The data behind the name
They write that existing descriptions of supposedly remarkable data patterns prove to be entirely fictional. Their objection is to the factual claim, not only to the interpretation.
The review of later studies
A 2014 systematic review in the Journal of Clinical Epidemiology included 19 studies that were designed to test the effect: 8 randomised controlled trials, 5 quasi-experimental studies and 6 observational evaluations. Most reported some evidence of an effect. The review judged that significant biases were likely, and that the studies were too different from each other to say much about size or conditions. Its authors go further and argue that the Hawthorne construct has not led to important research advances in 60 years, so new concepts are needed. They also note that earlier reviews of studies of school children found no evidence of a Hawthorne effect as the term was used there.
Competing explanations
Chiesa and Hobbs, cited in that review, point out that the many suggested mechanisms are partly contradictory.
Where this leaves a practitioner
Treat the Hawthorne effect as a plausible risk of unknown size, not as a proven law and not as a myth. Design the study so that you can see it: compare observed and unobserved conditions, give groups equal attention, and report which kind of observation you used.
121 min
How the Hawthorne effect changed since
The meaning of the term moved in two steps. First it described a factory result: more productivity from being singled out and made to feel important. Then it was widened. In the dementia trial, the authors say the definition has been broadened so that it refers to treatment response and not to productivity.
The systematic review says the methodological versions of the effect have changed in meaning over time and across disciplines. Current practice therefore does not match the original meaning. Today the word often names several different possible research effects. A reader should ask what a given author means before accepting or rejecting a claim that uses it.
132 min
Frequently asked questions about the Hawthorne effect
What is the Hawthorne effect?
The Hawthorne effect is a change in behaviour that happens because people know they are being studied or observed. It is named after factory experiments at Western Electric's Hawthorne Works. Researchers worry about it because observed behaviour may not match normal behaviour.
What is an example of the Hawthorne effect?
In a hospital study, hand hygiene events were about threefold higher at dispensers an auditor could see than at dispensers the auditor could not see at the same time. The increase came after the auditor arrived.
Did the Hawthorne experiments really show that every change raised productivity?
No. When Levitt and List analysed the recovered lighting data, they found no systematic evidence that productivity jumped whenever the lighting changed. They found only weak signs of a more subtle response to the experiments.
Is the Hawthorne effect real?
It is probably real in some settings, but its size is uncertain. A 2014 review of 19 studies found most reported some evidence of an effect, yet the studies differed too much to estimate how large it is or when it appears.
What is the difference between the Hawthorne effect and the observer-expectancy effect?
The Hawthorne effect comes from participants knowing they are studied. The observer-expectancy effect comes from the researcher's expectations about the result. The first needs no hypothesis from the researcher. The second does.
How do you avoid the Hawthorne effect?
You cannot remove it fully, but you can reduce and detect it. Measure behaviour without a visible observer where possible, give every group the same attention, and compare observed and unobserved conditions.
Does the Hawthorne effect apply to A/B tests?
It applies far less, because users in an A/B test usually do not know they are in an experiment, so both groups are treated alike. It can still matter when users can tell which version they have or when participation is announced.
?7 questions
Questions people ask
What is the Hawthorne effect?
What is an example of the Hawthorne effect?
Did the Hawthorne experiments really show that every change raised productivity?
Is the Hawthorne effect real?
What is the difference between the Hawthorne effect and the observer-expectancy effect?
How do you avoid the Hawthorne effect?
Does the Hawthorne effect apply to A/B tests?
§5 sources
Sources
Was There Really a Hawthorne Effect at the Hawthorne Plant? An Analysis of the Original Illumination Experiments — Steven D. Levitt and John A. List, NBER Working Paper 15016, 2009
Was There Really a Hawthorne Effect at the Hawthorne Plant? (journal version) — Levitt and List, American Economic Journal: Applied Economics, 2011 (landing page for the published version of source 1)
Systematic review of the Hawthorne effect: New concepts are needed to study research participation effects — Jim McCambridge, John Witton and Diana R. Elbourne, Journal of Clinical Epidemiology, 2014
The Hawthorne Effect: a randomised, controlled trial — Rob McCarney and colleagues, BMC Medical Research Methodology, 2007
Show all 5 sourcesShow fewer sources
Quantification of the Hawthorne effect in hand hygiene compliance monitoring using an electronic monitoring system — Jocelyn A. Srigley and colleagues, BMJ Quality & Safety, 2014
