Learn it → Apply it → Save as Skill
Being watched changes what you measure. People, and sometimes AI systems, behave differently the moment they know a performance, a call, a habit, or an output is being observed or evaluated. The behavior you record afterward may say more about the act of watching than about the thing you meant to measure.
The Dr. Wallace’s Favorite Mental Models is designed to turn useful ways of thinking into reusable tools. Each model gives you a fast visual overview, deeper interactive learning, practical application exercises and an AI-ready Markdown skill you can save for future use.
Instead of merely reading about mental models, you can gradually build your own AI skills library: a collection of portable reasoning skills that can help ChatGPT, Claude, Grok and other AI assistants apply the mental models you value.
Whenever a person knows a manager, a camera, a dashboard, a survey, or a researcher is watching, that awareness alone can change effort, honesty, or behavior, separate from any policy, training, or incentive actually being tested. The gap between watched performance and ordinary performance is the signal worth isolating.
Important: the size of this gap varies enormously by task and context, and in the original factory studies the gap was smaller and messier than the popular story suggests. Treat it as a reason to check, not a fixed ratio.
The model is most useful whenever a number, a review, or a report is collected while the person or system being measured knows they are being measured.
The method separates what the observation itself changed from what the thing you actually care about changed.
Spot a measurement that is being taken while the subject plainly knows it is being taken: a monitored call, a lit-up dashboard, a surprise audit, a self-report survey.
Ask what changed: the real condition you are testing, or simply the subject's awareness that this moment counts.
Compare against a baseline collected without the same awareness: unannounced sampling, a longer run so novelty fades, or a passive log nobody checks in real time.
Decide whether to reduce the reactivity (blind measurement, unobtrusive data), account for it statistically, or accept it because the behavior change itself is the goal.
Open a use case to see what the model changes.
A new dashboard, a ride-along, or a stated review period often improves the exact metrics being watched, temporarily. Ask whether the gain reflects a real process fix or the visible act of being checked, and whether it survives once the attention fades.
Someone narrating their screen for a researcher reads more, rushes less, and makes fewer careless errors than they would alone. Treat lab usability findings as an upper bound on careful behavior, and validate with unmoderated or analytics-based data.
Hand hygiene, safety checklists, and call-quality scores often rise sharply when staff know an audit is active and drift back afterward. Ask whether the gain is a process improvement or an artifact of the audit schedule being predictable.
Activity-tracking or keystroke software often produces an initial jump in logged activity. Ask whether people are working differently or simply performing more visibly for the tool, and whether the metric still means what it meant before the tool existed.
Logging your own sleep, spending, or workouts tends to improve the behavior on its own, independent of any other change. This is useful: tracking itself is a low-cost intervention. The caution is not to credit a separate program for a gain the tracking alone produced.
If participants can tell they are receiving a new feature, or employees know they were assigned to a pilot, their behavior may change because of the attention, not the feature. Ask whether the experiment can be made invisible to the people inside it.
This is an illustrative example designed to teach the method.
Expand for where the name comes from and what later reanalysis actually found.
Western Electric's Hawthorne Works, a manufacturing plant in Cicero, Illinois, hosted a series of studies between 1924 and 1932. Early illumination experiments tested whether brighter or dimmer lighting changed how quickly workers assembled relays; later phases, led by Harvard researchers Elton Mayo, Fritz Roethlisberger and William Dickson, varied rest breaks, incentive pay and working hours in a relay assembly test room. The studies became famous for suggesting that output rose almost regardless of which condition was tried, as if the workers were responding to the attention itself rather than to any specific change.
Despite describing studies from the 1920s, the phrase “Hawthorne effect” did not exist yet at the time. Psychologist John R. P. French introduced the term in 1953, and sociologist Henry A. Landsberger popularized it in his 1958 book reanalyzing the Hawthorne data, decades after the original experiments ended.
Less than the popular version suggests. Economists Steven Levitt and John List located the long-presumed-lost original illumination records and published a 2011 reanalysis: output did not track the lighting changes in the clean, dramatic pattern usually described in textbooks, and instead lined up more closely with the day of the week and the pay cycle. A separate 1992 reanalysis of the relay assembly room by Stephen Jones found that once new incentive pay and ordinary learning curves were accounted for, little was left to attribute to observation alone. The phenomenon the term describes, that awareness of being watched can change behavior, is real and well documented elsewhere. The specific Hawthorne Works studies are now considered weak evidence for it, even though they gave the effect its name.
Regression to the mean, selection effects, seasonal cycles and simple learning curves can all produce the same-looking bump. Check for an observation mechanism before naming this one.
“That number only looks good because they knew we were watching” can become an excuse to ignore results you do not like. The claim needs the same evidence as any other, not just convenience.
Reactivity from novelty typically fades as observation becomes routine. A separate concept, the novelty effect, describes that decay specifically; conflating the two can lead to over- or under-estimating how long a measured gain will last.
The mechanism shows up anywhere behavior is measured while the subject knows it: app analytics, wearables, surveys, classroom observation, and AI model evaluations watched by their own developers.
A pilot program everyone can identify, or a test group that is told it is a test group, contaminates the very comparison it was meant to produce. Build in blinding or random, undisclosed sampling wherever it is ethical to do so.
This model is about the subject changing because they are watched. A related but distinct problem, observer-expectancy effect, is about the person doing the measuring unconsciously skewing the result toward what they expect to find.
the gain holds up in unannounced, blinded, or long-run measurement taken well after any novelty would have faded.
the concern is specifically that people answer a survey the way they think they should, rather than that any observed behavior changed.
the change comes from believing you received a real treatment, independent of anyone watching you receive it.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI when and how to apply the Hawthorne effect, what questions to ask, what evidence to seek, what mistakes to avoid and how to structure its analysis.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
The underlying idea, that people change behavior when they know they are being observed, is well supported across modern research in fields from clinical trials to workplace studies. What is weaker is the specific historical claim that the original Hawthorne Works illumination and relay-assembly data prove it: later reanalyses found the famous pattern was not cleanly present in the real records. Use the concept, but do not cite the original factory story as its strongest evidence.
No. The placebo effect involves a change from believing you received a real treatment, even an inert one. The Hawthorne effect involves a change from knowing you are being watched or measured, independent of any treatment at all. The two can overlap in a study but describe different mechanisms.
Common approaches include unobtrusive or passive measures the subject cannot easily detect, random and undisclosed audit timing instead of announced ones, longer observation windows that let novelty wear off, and natural experiments that compare groups who never knew they were being studied. None of these eliminate reactivity completely, but each narrows the gap between what was measured and what would have happened unobserved.