The Science of Accountability: How Being Observed Changes Behavior

The intuition that we behave differently when someone is watching is old, but the science behind it is more surprising than the folk wisdom suggests. Being observed does not simply "make us try harder." It reliably speeds up some behaviors, degrades others, and its most famous supporting story turned out to be largely a myth. For anyone building a habit in public, the useful question is not whether an audience matters, but exactly when and how it shifts effort. The answer runs through six decades of experiments on social facilitation, evaluation apprehension, and the much-abused Hawthorne effect.

Zajonc and the two faces of an audience

By the early 1960s the literature was a mess: some studies found that an audience improved performance, others found it made performance worse. Robert Zajonc's contribution was to notice that the contradiction dissolved once you sorted tasks by difficulty (Zajonc, 1965). His proposal was that the mere presence of others raises arousal, and arousal strengthens whatever response is already dominant, meaning the one you are most practiced at.

The prediction is clean. On easy or well-learned tasks, the dominant response is usually the correct one, so an audience helps. On hard or unfamiliar tasks, the dominant response is often wrong, so an audience hurts. This is why a pianist nails a rehearsed piece in front of a crowd but fumbles a passage she only half-learned. The same watching eyes produce opposite results depending on how ready you already are.

An audience does not add motivation from the outside; it amplifies whatever tendency is already strongest inside you.

Mere presence, or the fear of judgment?

Zajonc argued that even the bare, non-evaluating presence of another organism was enough to raise arousal. Nickolas Cottrell and colleagues disagreed. In a well-known experiment, participants performed a task alone, in front of two attentive spectators, or in front of two people who were blindfolded and could not watch (Cottrell, Wack, Sekerak & Rittle, 1968). Only the attentive audience strengthened dominant responses. The blindfolded, non-evaluating pair produced no such effect.

Cottrell's conclusion, now called evaluation apprehension, is that the arousal is largely learned: we come to associate the presence of others with being judged, rewarded, or punished, and it is that anticipation of evaluation, not raw presence, that does most of the work. The distinction matters for anyone designing accountability. It suggests that a passive, unseeing audience changes little. What shifts effort is the credible sense that someone competent is actually paying attention to how you do.

How big is the effect, really?

Decades of individual studies were finally pooled by Charles Bond and Linda Titus, who meta-analyzed 241 studies covering nearly 24,000 people (Bond & Titus, 1983). Their synthesis broadly supported the task-difficulty pattern, but with an important dose of humility about magnitude. Across the literature, the mere presence of others accounted for only a small share of the variance in performance, on the order of a few percent at most.

The direction of the effects held up more cleanly than their size:

• Presence increased the speed of simple task performance and decreased the speed of complex task performance.
• Presence slightly improved accuracy on simple tasks but impaired accuracy on complex ones.
• The physiological arousal Zajonc invoked showed up mainly when people were doing something genuinely difficult, not for every task.
• The overall effects were reliable in direction but modest in size, a caution against treating "being watched" as a performance cheat code.

The practical reading is that observation is a real lever, but a gentle one. It nudges rather than transforms, and it works with the grain of a task's difficulty rather than overriding it.

The Hawthorne effect: a cautionary tale

No idea about being watched is quoted more often than the Hawthorne effect, the claim that workers at Western Electric's Hawthorne plant became more productive simply because researchers were studying them, regardless of what was changed. It is repeated as settled fact in countless management books. The evidence is far shakier than the story implies.

When Steven Levitt and John List recovered and reanalyzed the original illumination-experiment data, they found that the famous patterns were not really there (Levitt & List, 2011). Output did rise, but it tracked the workweek, such as Monday starts and pay-period timing, far more cleanly than it tracked whether workers were being observed, and productivity moved in the same direction whether the lights were brightened, dimmed, or left alone. Their verdict was blunt: several of the supposedly remarkable data patterns were, in their words, entirely fictional. A weaker "excess responsiveness to being experimented on" may exist, but the canonical dramatic version does not hold up.

The lesson is not that observation never changes behavior; the rest of the literature shows it does. The lesson is to be skeptical of tidy, oversized numbers, a discipline that applies just as much to the self-help canon around habits.

What this means for building habits in public

Popular habit advice is full of confident figures that do not survive scrutiny. The "21 days to form a habit" rule traces back to a plastic surgeon's mid-century observations about patients adjusting to their appearance, not to any study of habit formation. A better empirical estimate comes from Lally and colleagues, who tracked people forming everyday habits and found that the time to reach peak automaticity varied enormously, from roughly 18 to 254 days depending on the person and the behavior, with a median of about 66 days (Lally, van Jaarsveld, Potts & Wardle, 2010). That 66 is a median, not a magic number, and popular books that quote it as a fixed target overstate the precision. Reassuringly, that same study found that missing a single day did not meaningfully derail the process.

Popular syntheses such as James Clear's Atomic Habits (Clear, 2018) are useful for framing and worth reading as such, but they are summaries and motivational arguments, not primary evidence, and the specific statistics that circulate around accountability, like the widely shared claim that a formal accountability partner yields a 95 percent success rate, do not have a credible peer-reviewed source behind them. Treat them as folklore, not findings.

The genuinely supported principles are more modest and more durable. An audience shifts effort most when it is attentive and evaluative rather than merely present (Cottrell et al., 1968). It helps most on behaviors you have already made simple and well-practiced, and can backfire on tasks that are still hard and error-prone (Zajonc, 1965; Bond & Titus, 1983). So the honest way to use public accountability is to reduce a goal to something small and repeatable enough that being watched sharpens it rather than rattles you, and to expect a steady nudge rather than a miracle. Posting a check-in works not because observation is a magic force, but because it converts a private intention into a task with a real, attentive audience, which is precisely the condition under which the psychology says effort tends to rise.