Anyone who has run a behavioral program knows the uncomfortable pattern. Compliance looks excellent while the evaluation team is on site, then drifts once the clipboards leave. The usual explanation is a single phrase, the Hawthorne effect, invoked as though it were a settled law of behavior.

It is not. The phrase comes from factory studies conducted nearly a century ago, and the evidence behind them is thinner and messier than textbooks suggest. Modern experiments do show that observation can change behavior. But it happens only under particular conditions, and often for reasons that have little to do with simply being watched.

For anyone designing or evaluating interventions, this distinction matters. If observation inflates your outcomes, you may be scaling a program whose real effect is smaller than you think. If you understand when reactivity occurs, you can measure around it, or even put it to deliberate use. This piece covers three things: what Hawthorne actually showed, what controlled experiments have found since, and how to build designs that account for being observed.

The studies took place at Western Electric's Hawthorne Works near Chicago between 1924 and 1932. The first round, the illumination experiments, tested whether brighter lighting increased worker output. Productivity rose in the test groups. It also rose in control groups whose lighting never changed, and some accounts report that output held steady even when lights were dimmed dramatically.

Those puzzling results led to the better-known Relay Assembly Test Room. There, a small group of women assembled telephone relays under varied rest breaks, working hours and refreshments, and output climbed across most conditions. Researchers including Elton Mayo, Fritz Roethlisberger and William Dickson came to emphasise social and attitudinal factors such as attention, group cohesion and feeling valued. The label Hawthorne effect did not appear until 1958, when Henry Landsberger coined it.

Later reanalyses complicated this story considerably. The test room introduced a small-group pay scheme, so earnings tracked the group's output more closely. Workers also received regular feedback on their production. Partway through, two of the original five assemblers were replaced, reportedly for talking and poor cooperation, and their replacements were notably faster. In 1974, H. McIlvaine Parsons argued that feedback and reinforcement, not observation itself, explained the gains. In 2011, economists Steven Levitt and John List recovered the original illumination data. They found the evidence for a dramatic effect far weaker than the legend suggested.

The lesson is not that observation does nothing. It is that the founding study mixed observation together with incentives, feedback, staffing changes and a new supervision style. Each of those can change behavior on its own. The Hawthorne effect, as it is usually told, bundles them into one vague concept that explains everything and therefore predicts very little. A label is not a mechanism.

Takeaway

When a single label explains every unexpected result, it is usually hiding several mechanisms. Ask which part of observation (attention, feedback, incentives or evaluation) could actually be doing the work.

The most useful modern evidence comes from studies that compare the same behavior under visible and hidden measurement. In a 2014 study in BMJ Quality & Safety, Jocelyn Srigley and colleagues used an electronic location system to record hand hygiene events on hospital wards. When human auditors were visible in the hallway, hand hygiene rates were roughly three times higher than when no auditor was present. Reported compliance figures were partly measuring the audit itself.

Measurement can also change behavior through the questions it asks. Research on the question-behavior effect shows that asking people about their intentions can shift what they later do. In 2011, Alix Zwane and colleagues analysed several field experiments in low-income countries. They found that surveying people more often increased uptake of health insurance and water chlorination in some settings. The survey was not neutral. It worked as a light-touch intervention.

Wearable studies show a similar pattern. Some experiments compare sealed pedometers with unsealed ones that display step counts. Participants who can see their data tend to be more active, at least at first. Here the active ingredient looks less like being watched and more like self-monitoring and feedback, the same mechanisms Parsons identified at Hawthorne decades earlier.

Taken together, the evidence is real but conditional. A 2014 systematic review by Jim McCambridge and colleagues found that observation effects do occur, but they varied widely and were rarely well isolated. Reactivity appears strongest under four conditions: the behavior is visible and voluntary, a clear norm defines the right answer, the observer is noticeable and evaluative, and feedback is available. It also tends to fade as people get used to being measured.

Takeaway

Observation is not a single force but a set of conditions. Reactivity is most likely when people know what the right behavior looks like and believe someone who matters is checking.

The first design principle is to keep observation the same across conditions. Many evaluations unknowingly compare a heavily measured intervention group with a lightly measured control group. If both groups get the same surveys, site visits and devices, observation effects still exist but no longer distort the comparison. Attention-matched controls do the same job for the extra contact an intervention brings.

The second principle is to separate measurement from intervention where possible. Unobtrusive measures reduce the signal that behavior is being evaluated. Examples include administrative records, routine sensor data and audits that blend into normal workflow. When repeated measurement is unavoidable, a settling-in period before the baseline lets the novelty wear off. When a pretest might itself change behavior, a Solomon four-group design can estimate how much the measurement contributes. This design varies which groups receive a pretest.

The third principle comes from applied behavior analysis. In 1968, Donald Baer and his colleagues argued that behavior change only counts if it carries over to other settings and lasts beyond the conditions that produced it. For reactivity, this means planning follow-up measurement that is less intrusive than the intervention phase. If gains vanish when the auditors leave, the program changed behavior on audit days, not everyday practice.

Finally, reactivity can be used on purpose. Self-monitoring, public commitment and performance feedback all draw on the same mechanisms that bias measurement. The key is to name them as components, test them separately, and decide whether each is part of the intervention or a flaw in the evaluation. Ethics matter here too. Covert observation reduces reactivity but raises consent and trust concerns that need explicit review.

Takeaway

Every measurement choice is also an intervention choice. Decide in advance whether observation is part of what you are testing or something you need to hold constant.

The Hawthorne effect is better treated as a question than an explanation. The original studies cannot carry the weight placed on them. Modern experiments, however, confirm that measurement can shift behavior under conditions we can identify.

For intervention designers, the recommendations are straightforward. Keep measurement equal across groups. Prefer unobtrusive data sources and allow time for people to get used to being measured. Test whether effects last when observation becomes less visible. Treat feedback and self-monitoring as active ingredients rather than background noise.

The aim is not to eliminate observation, which is rarely possible. It is to know what your measurement is doing, so that the effect you report belongs to the program rather than to the clipboard.