Hawthorne effect
Claim assessed: People change their behaviour simply because they know they are being observed or studied.
Published · how we work
In short
The Hawthorne effect is the idea that people behave differently simply because they know they are being observed or studied. The name comes from a factory near Chicago where, in the 1920s, women workers supposedly did better at every change in lighting, even when the light was dimmer, just because the researchers were paying attention to them. The original data, analysed only in 2009, don’t show this. The effect does appear clearly when someone watches you directly, for example when an auditor checks whether doctors wash their hands. As a general effect of taking part in a study, in the best research it becomes too small to be sure of.
What it says
The Hawthorne story. Between 1924 and 1927, at Western Electric’s Hawthorne plant, engineers varied the lighting in three departments where women assembled telephone relays and wound wire coils, to see whether better light raised output. Textbooks summed up the result like this: output rose at every change, even when the light was reduced, so what mattered was attention, not light. Elton Mayo’s experiments followed until 1932, and the term “Hawthorne effect” appeared in a methodology textbook in 1953, according to McCambridge et al. (2014).
What the data show. The data from the lighting experiments were thought to have been destroyed. Levitt and List (2011) found them in two archives and analysed them for the first time. The descriptions of output rising “at every change” turned out, they write, to be “entirely fictional”:
- output jumped every Monday, but just as much on Mondays with no change, because less was produced on Saturdays and Mondays anyway;
- in the second year, output also rose in the departments no longer in the experiment, and the summer drops appeared in all three, so they were seasonal;
- there are only weak hints of an effect: while experiments were running, output was 3–4% higher than before, but the result disappears when the analysis accounts for month-to-month variation within each department.
The authors also found part of the daily data from Mayo’s experiments and reached the same conclusion as Jones (1992), which we haven’t read: little support for the effect when the data are read correctly.
The effect in research. Apart from the factory, the term came to mean any change in behaviour caused by knowing you are being studied. Two meta-analyses looked for it:
- McCambridge, Witton and Elbourne (2014) found 19 studies designed specifically to measure it, almost all in health. 12 found at least some sign of an effect. In the 5 randomised studies with yes/no outcomes, the combined effect was not significant (odds ratio 1.06). In the 8 studies of being observed or knowing you are studied it was significant (1.21), but many of them were observational studies. The authors’ conclusion: the effects exist, but “little can be securely known” about how large they are and when they occur.
- Berkhout et al. (2022) combined 15 studies from general practice. The overall effect was significant (1.41), but the studies differed greatly from one another. In the well-designed studies it disappeared.
When someone watches you directly. The clearest example comes from hospitals. Srigley et al. (2014) used sensors that recorded, for 8 months, every use of hand-rub dispensers in two units of a teaching hospital, and tracked at the same time where the hand hygiene auditors were. Dispensers within an auditor’s line of sight recorded 3.75 uses per hour, against 1.48 at the others at the same time and 1.07 at the same dispensers a week earlier. The increase began after the auditor arrived. Inside patient rooms, where the auditor couldn’t see, nothing changed.
When you are only asked. A related form is the question–behaviour effect: simply being asked about a behaviour changes it. Wood et al. (2016), across 116 tests, found a small effect (d = 0.24), reduced to 0.15 after correcting for unpublished studies. The effect was larger among students, in the lab and for well-regarded behaviours, and practically nil for risky ones. McCambridge and Kypri (2011), across 10 alcohol studies, compared people who had only filled in the assessment questionnaire, with no intervention, with people who hadn’t. The former then reported drinking about one standard drink a week less, a difference at the edge of significance, and scored one point lower on the AUDIT risk scale.
Example
A company wants to know whether employees follow a new safety procedure and sends an inspector once a month. On inspection day, everyone wears a helmet and goes through the checklist. The report says the procedure is followed 95% of the time. Sensors on the equipment might show something else for the rest of the month. The example is an illustration, not a study cited here.
How to apply it
The steps below are a practical way to apply it that we propose, based on the studies cited.
- Don’t rely only on announced observation. When you measure a visible behaviour through someone’s presence, the numbers can come out better than reality. Where possible, use data that is recorded anyway, such as consumption, sensors or logs.
- Compare with a group that isn’t observed the same way. If you are evaluating a change, a control group given the same attention but without the change separates the effect of the change from the effect of attention.
- Use it on purpose, in moderation. A question about what someone intends to do can slightly raise the chance that they do it, especially for easy, well-regarded things. Don’t expect more than a small effect.
- Don’t explain every result with it. McCambridge et al. note that the term is often used after the fact to explain disappointing results. Without a comparison group, “it was the Hawthorne effect” is a guess, not a conclusion.
Limits and nuances
- One name, several phenomena. The term covers being watched, being asked, knowing you are in a study and receiving attention. Both meta-analyses say it is not a single effect and that its size depends greatly on what exactly the observed person is doing.
- The clearest evidence isn’t randomised. Large effects, as with hand hygiene, come from observational studies. In randomised studies, the effect of taking part in a study is small or not significant.
- The studies come almost all from health. 18 of McCambridge et al.’s 19 studies and all of Berkhout et al.’s are from health care, although the idea started in a factory.
- Hand hygiene varies. According to Berkhout et al., a 2020 review (Purssell et al., which we haven’t read) found effects between −6.9% and 65.3% and did not combine them, because they were too different.
- What we read. We read in full McCambridge et al. (2014), McCambridge and Kypri (2011), Srigley et al. (2014), Wood et al. (2016) and Berkhout et al. (2022), from PubMed Central, and Levitt and List, in the 2009 NBER working paper version of the article published in 2011. Jones (1992) and the original Hawthorne studies we describe from Levitt and List and McCambridge et al.
Sources
- Steven D. Levitt, John A. List (2011). Was there really a Hawthorne effect at the Hawthorne plant? An analysis of the original illumination experiments
- Jim McCambridge, Kypros Kypri (2011). Can simply answering research questions change behaviour? Systematic review and meta analyses of brief alcohol intervention trials
- Jim McCambridge, John Witton, Diana R. Elbourne (2014). Systematic review of the Hawthorne effect: New concepts are needed to study research participation effects
- Jocelyn A. Srigley, Colin D. Furness, G. Ross Baker, Michael Gardam (2014). Quantification of the Hawthorne effect in hand hygiene compliance monitoring using an electronic monitoring system: A retrospective cohort study
- Chantelle Wood, Mark Conner, Eleanor Miles, Tracy Sandberg, Natalie Taylor, Gaston Godin, Paschal Sheeran (2016). The impact of asking intention or self-prediction questions on subsequent behavior: A meta-analysis
- Christophe Berkhout et al. (2022). Defining and evaluating the Hawthorne effect in primary care, a systematic review and meta-analysis
See also: Bystander effect, Pygmalion effect