Observer Effect: What Changes When People Know They Are Watched

Being watched changes what people do, through several routes at once. Someone who knows an observer is counting tends to do more of what is being counted, and to do it more carefully, because the counting signals what matters and because being judged is uncomfortable. Someone singled out for a study may work harder out of pride, curiosity or a wish to please the researcher. And someone who is simply asked about their behavior, in a diary or a survey, often changes it by attending to it. All of these are real, and none of them is what the study was trying to measure.

That is why the effect matters for anyone who reads research or manages people. A trial of a new hand-washing poster that sends an observer to count hand-washing may find an improvement that comes from the observer rather than the poster. A classroom observation records a lesson taught to the observer. The observer effect is not a flaw in the people being watched; it is a confound in the measurement.

Its evidential history, though, is poorer than its reputation. The name comes from the Hawthorne studies, and the studies did not show what the name implies. The relay assembly test room had no control group, changed several things at once, replaced two of its five workers partway through, paid the group on a different incentive from the floor, and ran across the start of the Depression. The illumination experiments had control groups, which is how they showed that lighting could not explain the rise, but when Levitt and List recovered their records in 2011 the textbook pattern was not there. McCambridge, Witton and Elbourne's 2014 systematic review found 19 studies built to measure the effect directly, most reporting some effect, with methods and results so varied that little could be said about when it operates or how large it is.

The sensible position is the one the demo above rewards: expect being watched to change behavior, design studies and audits so that it cannot masquerade as the result, and do not cite the Hawthorne plant as proof.

The Hawthorne studies ran at Western Electric's Hawthorne Works in Cicero, Illinois, from 1924 to 1932. The first were the illumination experiments, conducted with the National Research Council from 1924 to 1927 to find out whether better lighting raised output. In the second of them, a test group under lighting raised in stages and a control group under constant lighting both increased output by similar amounts. In the third, the test group's lighting was cut a foot-candle at a time from 10 while the control group stayed at 10. Output in both groups rose until the test group reached 3 foot-candles, where the operators protested that they could hardly see and their output fell. In a side test, two operators kept up their output at 0.06 foot-candles, a level Roethlisberger and Dickson compared with an ordinary moonlit night. The investigators concluded that the experiments had failed to answer the question they were built for.

The relay assembly test room followed, from April 1927. Five experienced women who assembled telephone relays were moved to a separate room with a layout operator and an observer who recorded output, conversation and events. Over thirteen experimental periods ending in mid-1929 the investigators varied rest pauses, hours and refreshments; the room itself ran on to 1932. Output rose through most of the periods, and in the twelfth, when the rest pauses and shorter hours were withdrawn and the original 48-hour week restored, it stayed at its highest level. Two of the five women were replaced in January 1928 after they were judged to be holding output down. Roethlisberger and Dickson reported all of this in Management and the Worker in 1939, and read the rise as the product of a changed social situation: the women had been consulted, freed from close supervision and made into a group.

Henry Landsberger's Hawthorne Revisited brought the term Hawthorne effect into general use in 1958. The quantitative reanalyses came later. Franke and Kaul, in the first statistical treatment of the relay room data in 1978, found that managerial discipline in the form of the two replacements, the onset of the Depression and the rest periods accounted for most of the variance in output. Jones, in 1992, used the same records with different methods and found slender or no evidence of a Hawthorne effect. Levitt and List located the original illumination data on microfilm and reported in 2011 that the famous pattern of output rising after every change was not in them; lighting changes happened on Mondays, and output rose on Mondays whether or not the lighting changed.

The effect appears wherever behavior is measured by people who can be seen measuring it. The rows below are illustrations, not findings from the studies cited on this page.

The effect cannot be removed by asking people to behave naturally, so the counter-measures are matters of design.

• Give the observed condition a control that is also observed. If both groups are watched and only one gets the intervention, the watching cancels out.
• Measure without a visible observer where you can, through records, products used or sensors, and compare that measure with the observed one to see the size of the gap.
• Make observation routine. An observer who is always there stops being news; the behavior changes most when the watching is an event.
• Tell people what the measure is for. Being counted invites gaming; a count whose purpose is understood invites less of it.
• Read claims about the effect with the same care. A change in behavior after attention was paid shows a change, not that attention caused it, unless something else was held constant.

The observer effect is one of a family of effects in which the act of studying changes what is studied, and the family members are easy to confuse.

• It is not the observer-expectancy effect, in which the researcher's expectations shape what the researcher records or how the participants are cued, so that the results lean toward what was expected. The observer effect is a change in the people being watched; the expectancy effect is a change in the watcher.
• It is not experimenter's bias, the wider set of ways an experimenter's beliefs and choices tilt a study, from the design to the analysis. The observer effect can occur with a perfectly neutral experimenter, because it depends only on the participants knowing they are observed.
• It is not the placebo effect, the improvement that follows an inert treatment. Both can inflate a trial's results, and both are reasons for control groups, but the placebo effect works through expectations about the treatment and the observer effect through awareness of being studied.