We usually imagine a prediction standing at a safe distance from the event it describes. Someone observes the world, forms a judgment, and waits to see whether reality agrees. But what happens when people hear the prediction, change their behavior because of it, and thereby change the outcome? What happens when the observer enters the chain of causes, while continuing to speak as though they were only watching?

A bank is expected to fail, so depositors rush to withdraw their money. A person is expected to be untrustworthy, so others treat them with suspicion. A project is declared hopeless, so nobody gives it the resources it needs. When the bank collapses, the person becomes defensive, or the project fails, the original judgment returns with an impressive new credential: evidence.

“I knew it.”

Perhaps. But did we recognize what was going to happen, or participate in making it happen? And how often does our satisfaction at being right prevent us from asking the second question?

In his 1948 essay, sociologist Robert K. Merton described the self-fulfilling prophecy as a process in which an initially mistaken understanding of a situation produces behavior that makes the situation conform to that understanding. The decisive element is the causal passage from belief to action to consequence. The result then appears to vindicate the belief that helped produce it. Merton’s account therefore concerns something more consequential than a mistake in perception: a mistake can acquire a material history. (Robert K. Merton, “The Self-Fulfilling Prophecy,” 1948.)

The mechanism can be expressed simply:

Expectation → changed behavior or treatment → changed conditions → a confirming outcome → a stronger expectation.

Every arrow matters. Without a change in behavior, access, treatment, or some other causal condition, we may be dealing with an accurate forecast, a coincidence, or selective interpretation. Calling every fulfilled expectation a self-fulfilling prophecy makes the concept impossible to test and therefore much less useful.

Suppose I expect a colleague to dislike me. If I remember their impatient remarks and overlook their kindness, confirmation bias is shaping my interpretation. If I avoid them, answer curtly, and exclude them from conversations until they begin to dislike me, my expectation has entered the relationship itself. These processes can reinforce one another, but they are different. In one, I select evidence. In the other, I help produce it.

This distinction also prevents a philosophical overreach. The self-fulfilling prophecy does not establish that truth is whatever we believe. A belief can be false when it describes an existing condition and still become influential in producing a later condition. If suspicion helps turn a cooperative relationship into a hostile one, the later hostility does not prove that hostility was present from the beginning. Reality has changed; the history of how it changed remains part of the truth.

The difficult question is therefore counterfactual: what would have happened if the expectation had not shaped our conduct? We cannot answer by pointing only to the outcome we actually helped create.

At the personal level, this process can begin with an ordinary sentence: “I am going to fail.” Imagine someone preparing for an examination or an interview. If that expectation leads them to avoid practice, abandon preparation, or interpret every difficulty as a reason to stop, their chances may worsen. Failure can then become a statement about identity: “This proves I was never capable.” A conditional outcome, partly shaped by what they did or did not do, has been turned into an apparently permanent property of the person.

Yet this is precisely where the popular version of the argument becomes too confident. Worry does not always produce surrender. Sometimes it produces preparation.

Psychologists Julie Norem and Nancy Cantor investigated defensive pessimism: a strategy in which some people use low expectations and consideration of possible problems to manage anxiety and prepare for demanding situations. Their experiments found that interfering with this strategy could impair performance. The lesson is narrower and more useful than either “negative thinking causes failure” or “pessimism is good.” The consequences of an expectation depend partly on what someone does with it. (Julie K. Norem and Nancy Cantor, “Defensive Pessimism: Harnessing Anxiety as Motivation,” 1986.)

Murphy’s Law can therefore become either a surrender instruction or a preparation prompt. “Something may go wrong” can lead me to stop trying, but it can also lead me to test the equipment, leave extra time, or prepare an alternative. Conversely, confidence can sustain effort or excuse the absence of it. The important question is not simply whether an expectation feels positive or negative. What behavior does it authorize?

There is an ethical reason to insist on this distinction. If the self-fulfilling prophecy is reduced to “your thoughts create your reality,” it becomes dangerously easy to blame people for suffering they did not choose. Poverty, discrimination, illness, violence, and the decisions of other people do not disappear because someone improves their attitude. A person can approach a closed door with extraordinary confidence and still find it locked.

Agency is real without being unlimited. Responsibility should follow actual influence, available alternatives, and power. Otherwise, a theory that could expose how disadvantage is produced becomes another way of telling the disadvantaged that they produced it themselves.

The interpersonal dimension makes this especially clear. We do not live only inside our own expectations. We live among people whose expectations influence how they treat us.

In a classic 1977 experiment, Mark Snyder, Elizabeth Tanke, and Ellen Berscheid manipulated male participants’ beliefs about the physical attractiveness of women with whom they spoke by telephone. The women did not know how they had been presented. Independent observers nevertheless found differences in the women’s conversational behavior: those whose partners believed them attractive behaved in ways judged more friendly and sociable. This was a limited laboratory setting, not a universal rule about attraction or relationships, but it demonstrated a revealing possibility: an expectation held by one person can help elicit behavior from another. (Mark Snyder, Elizabeth D. Tanke, and Ellen Berscheid, “Social Perception and Interpersonal Behavior: On the Self-Fulfilling Nature of Social Stereotypes,” 1977.)

The target of the expectation need not believe it. They may not even know it exists. The effect can travel through the interaction: the questions they are asked, the patience they receive, the warmth they encounter, or the space they are given to respond.

This complicates the confidence with which we describe other people. When I say someone is withdrawn, difficult, or uncooperative, am I describing a stable characteristic, a response to a particular situation, or a response partly elicited by me? There may be genuine reasons for my judgment. But have I left room to discover which explanation fits?

In education, the same question acquires institutional weight. A teacher’s expectations can influence the opportunities a student receives to participate, attempt difficult work, recover from mistakes, and develop competence. But the evidence does not justify the familiar claim that teachers can simply believe students into brilliance or failure.

Reviewing thirty-five years of research in 2005, Lee Jussim and Kent Harber concluded that classroom self-fulfilling effects do occur, but are typically small; they do not generally accumulate as dramatically as popular accounts suggest. Stronger effects may occur for some students from stigmatized groups. The review also emphasized that teachers’ expectations often predict achievement because they contain accurate information, not only because they create the achievement they predict. (Lee Jussim and Kent D. Harber, “Teacher Expectations and Self-Fulfilling Prophecies: Knowns and Unknowns, Resolved and Unresolved Controversies,” 2005.)

That qualification strengthens the inquiry. A serious explanation must distinguish discovering a difficulty from deepening it. It must also distinguish a realistic assessment of someone’s present performance from a premature verdict on their future capacity.

Consider a hypothetical workplace. A manager decides that one employee lacks initiative. The employee receives fewer demanding assignments, less useful feedback, and little authority. Later, the manager compares that employee’s record with the record of a colleague who received training and repeated opportunities to lead. The difference may be real. But how much of it measures initial ability, and how much measures the opportunities each person was allowed to accumulate?

The question becomes more difficult when the result is placed in a spreadsheet. A number can be calculated correctly while its interpretation remains incomplete. Measuring the outcome does not automatically explain its origin.

Here the self-fulfilling prophecy connects with power. People have unequal capacities to make their expectations consequential. A student’s private fear and an institution’s formal assessment may both be mistaken, but the institution can distribute qualifications, access, funding, and exclusion. Its belief has procedures behind it.

When an expectation controls access to the conditions needed to disprove it, correction becomes particularly difficult. Someone denied experience may later be rejected for lacking experience. A community denied investment may later be described as offering insufficient opportunity. These are mechanisms to investigate in particular cases, not conclusions to impose on every unequal outcome. Their possibility, however, means that an observed difference cannot by itself settle the question of why the difference exists.

Markets make the interaction between expectation and consequence visible on a larger scale.

In their 1983 banking model, Douglas Diamond and Philip Dybvig showed how a system that transforms less liquid investments into deposits available for withdrawal can be vulnerable to a run. If enough depositors expect others to withdraw, withdrawing early can become individually rational. The collective result can damage a bank that could otherwise have continued operating. Their model also examined how deposit insurance, under appropriate conditions, can prevent this destructive outcome. (Douglas W. Diamond and Philip H. Dybvig, “Bank Runs, Deposit Insurance, and Liquidity,” 1983.)

Notice the psychological structure. A depositor does not necessarily have to believe that the bank’s underlying investments are worthless. It may be enough to believe that other depositors will rush to withdraw. What matters is an expectation about other people’s expectations.

A private decision can make sense under those conditions while contributing to a collectively damaging result. Describing everyone involved as irrational misses the problem. The rules and incentives can make participation in the destructive process the safer individual choice.

This is one reason public reassurance sometimes fails. If people face a credible penalty for being the last to act, a request to remain calm does not remove that penalty. A durable intervention must address the conditions that make collective restraint difficult.

The pandemic shopping surges of 2020 illustrate a related coordination problem, although they should not be reduced to a story about imaginary shortages created by foolish consumers. Research by the Institute for Fiscal Studies, using millions of UK grocery transactions, found that the surge in purchases of storable goods involved many more households buying those products, generally with moderate changes in quantities. Spending on storable staples peaked at more than 80 percent above the January–February daily average. The findings complicate the image of a small number of extreme hoarders as the whole explanation. (Institute for Fiscal Studies, “Pre-lockdown ‘Panic Buying’ Involved Many More Households Than Usual Buying Storable Groceries, but in Moderate Amounts,” 2020.)

The conceptual point is that even modest changes can matter when they occur together. A household buys earlier because it expects others to buy earlier. Shelves empty faster. Those shelves become evidence that buying earlier was necessary. The resulting feedback can coexist with genuine changes in household needs and constraints on replenishment.

An empty shelf tells us that availability at a particular place and time has fallen. It does not, by itself, tell us how much resulted from production, distribution, changed consumption, anticipatory purchasing, or some combination. The visible outcome is the beginning of the explanation.

In speculative markets, the same feedback can be deliberately exploited. If a promoter already owns an asset, persuading others to expect a price increase may generate buying that produces an increase. The movement then appears to validate the promoter’s insight, drawing further attention and potentially more buyers.

A documented example appears in the US Securities and Exchange Commission’s 2000 settlement with Jonathan Lebed. The SEC’s order described eleven occasions on which he purchased thinly traded stocks, posted false or misleading promotional messages, and sold into the subsequent increases. The settlement required the return of profits and interest; he settled without admitting or denying the findings. Here, an official record identifies a concrete pattern of positions, communications, and transactions, rather than leaving manipulation as an assumption inferred from a rising price. (US Securities and Exchange Commission, “In the Matter of Jonathan G. Lebed,” Administrative Proceeding File No. 3-10291, 2000.)

The distinction matters. A price increase can be real while the explanation used to encourage buying is false. Making one part of a prediction come true for a time does not establish an asset’s lasting value, the honesty of its promoter, or the safety of following them.

But neither does every price increase prove manipulation. To establish deliberate deception, we need evidence about conduct, information, and intent. A beneficiary is not automatically an architect. Someone may exploit a process they did not initiate; many people may independently repeat a message because it is profitable or popular.

If we assume a hidden coordinator whenever an outcome appears coordinated, our explanation risks becoming another belief that recognizes only its own confirmations.

Cultural markets offer an unusually clear experimental window into these dynamics. In 2006, Matthew Salganik, Peter Dodds, and Duncan Watts created an online music market involving 14,341 participants. Some made choices without seeing others’ download counts; others received social information. Stronger social influence increased both the inequality and the unpredictability of success. Quality still mattered, but it did not determine one inevitable ranking. (Matthew J. Salganik, Peter Sheridan Dodds, and Duncan J. Watts, “Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market,” 2006.)

A later experiment by Salganik and Watts went further by artificially reversing songs’ apparent popularity. False social information changed subsequent outcomes, although the most appealing songs showed a capacity to recover. The manipulation mattered without becoming omnipotent. That boundary is as revealing as the effect itself: social influence can alter a trajectory while other properties continue to resist it. (Matthew J. Salganik and Duncan J. Watts, “Leading the Herd Astray: An Experimental Study of Self-Fulfilling Prophecies in an Artificial Cultural Market,” 2008.)

This invites a different way of reading success. Popularity can convey information about appeal, but it can also help generate the attention from which further popularity grows. A ranking may summarize earlier choices while influencing later ones.

What, then, are we measuring when we treat visibility as proof of merit? How much reflects qualities people discovered, and how much reflects the opportunities they had to discover them? The existence of this question does not erase talent or effort. It prevents their relationship with recognition from being treated as automatic.

Digital prediction systems add another layer because they can influence the production of the data used to evaluate them.

Imagine a system that sends more police patrols to a neighborhood because it predicts more crime there. More patrols can discover more incidents. If those discoveries are then fed back into the system without adequately accounting for differences in observation, the neighborhood may appear to confirm the original prediction. Danielle Ensign and colleagues analyzed this risk through mathematical modeling and experiments in their 2018 paper on predictive policing feedback loops. Their work identifies a mechanism under specified conditions; it is not proof that every deployed system behaves identically. (Danielle Ensign, Sorelle A. Friedler, Scott Neville, Carlos Scheidegger, and Suresh Venkatasubramanian, “Runaway Feedback Loops in Predictive Policing,” 2018.)

A crucial distinction appears here. Increased recorded incidents do not necessarily mean that the system caused more underlying crime. It may have changed where and how much crime was observed. This can be a self-confirming measurement process rather than a fully self-fulfilling change in the behavior being predicted.

The broader question is whether a system can distinguish evidence about the world from evidence generated by its own interventions. A prediction that changes attention, opportunity, or surveillance may later encounter a dataset that already contains its influence. Treating that dataset as an independent verdict can conceal the feedback.

A similar caution belongs in discussions of politics. An announcement that a candidate is certain to lose could discourage supporters. An announcement that a candidate is certain to win could make participation seem unnecessary. Other people might react with renewed determination. The direction and size of the response cannot be decided from a compelling story alone.

Research by Sean Westwood, Solomon Messing, and Yphtach Lelkes found that probabilistic election forecasts could increase certainty about outcomes and reduce participation in an experimental voting setting. Their broader analysis also examined associations between confidence and reported turnout. These findings support concern about how forecasts are communicated; they do not establish that a particular media narrative single-handedly determined an actual election. (Sean Jeremy Westwood, Solomon Messing, and Yphtach Lelkes, “Projecting Confidence: How the Probabilistic Horse Race Confuses and Demobilizes the Public,” 2020.)

The distinction between influence and control is essential. Media can affect the conditions under which people decide while remaining unable to dictate each decision. An outcome can have many interacting causes, including effects that partly cancel one another.

There is nevertheless a civic danger in presenting contingent futures as completed facts. “Nobody will participate.” “Nothing can change.” “Everyone supports this.” Such statements may function as descriptions, but they can also influence whether people contribute, object, organize, or reveal disagreement. A prediction of public passivity can become one of the reasons for that passivity.

This leads to pluralistic ignorance: a situation in which people misperceive what others believe and adjust their behavior to a social consensus that may be less real than it appears.

A study published in the American Economic Review in 2020 offers a concrete example. Leonardo Bursztyn, Alessandra González, and David Yanagizawa-Drott found that young married men in their Saudi Arabian samples substantially underestimated other men’s support for women working outside the home. Correcting those perceptions increased willingness to help wives seek employment. Follow-up reports showed more applications and interviews, although the main experiment did not establish a statistically significant increase in employment itself. This was evidence from a particular population and period, not a timeless description of an entire society. (Leonardo Bursztyn, Alessandra L. González, and David Yanagizawa-Drott, “Misperceived Social Norms: Women Working Outside the Home in Saudi Arabia,” 2020.)

The implication extends beyond the setting without making the result universal. Some social barriers may be maintained partly by people’s mistaken beliefs about one another. In such cases, change need not begin by converting everyone to an entirely new conviction. It may begin by making existing convictions more accurately visible.

Imagine a meeting in which several people privately doubt a proposal but each assumes everyone else supports it. Each person’s silence becomes evidence for everyone else. What looks like consensus can be a pattern of mutually misread restraint.

Breaking that pattern requires more than telling one individual to be brave. It may require a procedure that makes disagreement safe to express: independent assessments before group discussion, anonymous feedback, or leadership that demonstrably permits correction. These are possible design responses to the mechanism, not guarantees that every group will improve merely by adopting a form.

The philosophical question now becomes sharper. Human beings interpret situations while participating in them. Reflexivity names this capacity for an interpretation to return to the world through action and alter the conditions that interpretation concerns.

The future we anticipate can therefore become one of the inputs into the future we produce. This does not dissolve the difference between facts and opinions. It gives us an additional factual question to investigate: what did the opinion cause?

It also challenges a familiar misuse of inevitability. Suppose a project fails after supporters withdraw because they expect failure. The outcome occurred, but its occurrence does not establish that it had to occur under every plausible pattern of support. Once a path has been taken, the paths not taken become harder to see. We can mistake the reality of the result for the necessity of the route.

This is where a realized prediction and a justified prediction separate. Someone may correctly announce what will happen while helping ensure that alternatives are never attempted. Their forecast can be accurate about the resulting world without demonstrating that their earlier explanation of the world was sound.

The reverse possibility is equally important. A warning can help prevent the event it warns about. A maintenance inspection identifies a likely failure; repairs prevent it. An organizer notices declining participation; changes in communication bring people back. The anticipated outcome does not occur because the expectation produced an effective response.

Such a warning is sometimes called a self-defeating prophecy. But we should not use that label to rescue every failed prediction. To distinguish successful prevention from a poor forecast, we still need evidence about the original risk, the intervention, and its effects. The absence of a disaster can reflect prevention, exaggeration, chance, or several causes together.

Both forms reveal a limit to judging predictions solely by whether the announced event occurred. We must also ask what people did after hearing them.

This has consequences for responsibility. A person or institution that publishes a consequential assessment should consider its capacity to alter behavior. That does not mean concealing unwelcome information. Concealment can create harms of its own. It means communicating evidence, uncertainty, assumptions, and the ways an outcome might still change.

“This will happen” and “this becomes more likely if these conditions continue” can lead people toward different decisions. The second formulation preserves something the first may prematurely remove: the relevance of action.

At the institutional level, interrupting a harmful loop requires identifying the link that sustains it. If low expectations reduce opportunities, reassurance alone leaves the opportunity problem intact. If a model learns from selectively collected data, a declaration of neutrality does not repair the data. If distrust is a rational response to vulnerability, demanding trust does not remove the vulnerability.

The intervention must meet the mechanism. That can mean changing access, incentives, measurement, safeguards, or channels for correction. Psychological awareness matters, but awareness cannot substitute for a resource that was never provided or a rule that continues to exclude.

At the personal level, the corresponding task is modest and demanding. I can ask which part of my expectation describes evidence and which part predicts a response I have not yet tested. I can notice whether my protective behavior is preserving safety or unnecessarily narrowing the possibility of a different encounter. Where conditions are reasonably safe, I can make a limited change and observe what happens without deciding in advance what the result must mean.

I do not have to assume that everyone is trustworthy. I do have to recognize when I have arranged an interaction so that almost no response could count against my suspicion.

Nor do I have to believe that success is guaranteed. I can treat uncertainty as a reason to prepare and investigate rather than as proof that effort is pointless. Hope becomes more credible when it is connected to actions, resources, feedback, and a willingness to revise.

That qualification should apply to this essay too. An account of self-fulfilling prophecies can become fatalistic if it leaves the reader believing that institutions always manufacture outcomes, that markets are entirely manipulated, or that their own judgment is helpless before social influence. If that belief encourages withdrawal, a text intended to examine the loop may help build another one.

The concept deserves the same scrutiny it asks us to apply elsewhere. Some expectations are accurate. Some interventions fail. Some people resist the roles assigned to them. Some systems contain safeguards that weaken feedback. The work is to discover where influence exists, through which mechanism, and with what limits.

Perhaps the most difficult practice is to remain interested in an outcome that would prove our expectation incomplete. If a person behaves better when given a fairer opportunity, can we revise our judgment without protecting our earlier certainty? If a community changes after its conditions improve, can we acknowledge what those conditions were doing? If cooperation becomes possible after we stop treating hostility as inevitable, can we recognize our own place in the earlier conflict?

These questions reach beyond optimism and pessimism. They concern how we explain the world while helping to inhabit and organize it.

When reality finally resembles what I expected, I may indeed have understood something.

But before I congratulate myself, can I still ask how much of what I now call evidence passed through my hands?