A social experiment is a controlled study designed to observe how people behave under specific conditions, often manipulating some feature of their environment, social context, or available information to see what happens. The concept spans a wide range of setups, from small psychology-lab scenarios where a few dozen volunteers face staged peer pressure, to nationwide policy trials where thousands of people receive cash payments while researchers track what changes. What ties them together is the deliberate introduction of a variable into a social situation to learn something about human nature. Some of the most famous social experiments have fundamentally reshaped how we think about obedience, conformity, and cooperation, though the field has also wrestled with serious ethical controversies and questions about whether classic findings hold up under scrutiny.
The Obedience Experiments That Started a Conversation
No discussion of social experiments begins anywhere but with Stanley Milgram’s obedience studies from the early 1960s. In those experiments, ordinary volunteers were instructed by a researcher to deliver what they believed were increasingly dangerous electric shocks to another person. A disturbing proportion complied, even when the person on the receiving end appeared to be in extreme distress. The findings shocked the public and the scientific community, raising uncomfortable questions about how readily people follow authority figures.
What makes the Milgram paradigm especially interesting is how robust it has turned out to be across variations. A large empirical synthesis of Milgram’s own experiments and their replications identified eight factors that reliably influenced whether participants went all the way. The experimenter’s directiveness mattered: authoritative instructions increased compliance. Proximity to the person being harmed mattered too, with closer physical or emotional contact reducing obedience. And group pressure from fellow participants to disobey substantially decreased compliance. Surprisingly, obedience was somewhat higher when the authority figure was a peer rather than a credentialed researcher, suggesting the dynamic is not purely about institutional prestige.1PLoS ONE. Meta-Milgram: An Empirical Synthesis of the Obedience Experiments
A recent replication tested whether stripping away the trappings of a university lab or a stated research purpose would change the results. It did not. Whether the experiment was conducted in a formal institutional setting or framed with no scientific justification, obedience levels closely mirrored Milgram’s original findings. The differences between groups were not statistically meaningful.2Scientific Reports. Doubting the power of prestige: obedience to authority beyond institutional and research justifications That result is both reassuring for the science and unsettling for what it says about people: the willingness to comply with authority does not seem to depend much on the legitimacy of that authority.
Conformity and the Power of the Group
While Milgram explored obedience to a single authority, another line of social experiments examines how people respond to group pressure from their peers. The classic demonstration is Solomon Asch’s conformity studies, where participants were asked to compare line lengths in the presence of confederates who unanimously gave obviously wrong answers. Many participants went along with the group, even when the correct answer was plainly visible.
Modern neuroscience has added a layer to this picture. EEG recordings during group-pressure tasks have shown that conforming to an incorrect group opinion is not just a conscious social decision. It actually alters early, unconscious stages of visual processing. Participants who adapted to wrong group opinions showed reduced brain responses associated with basic visual discrimination, suggesting that social pressure can change not only what you say you see, but how your brain processes what it is looking at.3PubMed. EEG reveals an early influence of social conformity on visual processing in group pressure situations The implication is striking: conformity is not merely surface-level politeness or strategic agreement. It reaches deeper than most people assume.
Intergroup Conflict and Cooperation
One of the most ambitious early social experiments took place not in a lab but at a summer camp. The Robbers Cave Experiment, conducted in the mid-1950s by Muzafer Sherif and colleagues, divided boys into two groups and introduced competition for limited resources. Predictably, hostility between the groups escalated quickly, complete with name-calling, cabin raids, and flag-burning. Sherif used the experiment to develop realistic conflict theory, which holds that competition over scarce resources drives prejudice and intergroup aggression.4American Journal of Human Psychology. Reflections on the Robbers Cave Experiment: Finding Lessons on Political Conflict, Racism, Xenophobia, and Business Environments
The less-remembered second half of the Robbers Cave study is arguably more important. Sherif found that simply bringing the groups together for meals or movies did not reduce the hostility. What did work was introducing superordinate goals, tasks that required both groups to cooperate because neither could succeed alone. When the camp’s water supply was sabotaged (by the researchers) and both groups had to work together to fix it, animosity gradually dissolved. The lesson has proven durable across decades of conflict-resolution research: shared goals that demand collaboration are far more effective at reducing intergroup tension than mere contact.
How the Environment Nudges Behavior
Not every social experiment requires deception or emotional intensity. A large body of field experiments explores how subtle changes to a person’s environment can shift their choices. The research framework often called “nudging” or “choice architecture” has become one of the most practically applied branches of behavioral science.
A meta-analysis covering more than 200 studies and over two million participants found that choice architecture interventions produce a small to moderate effect on behavior overall. Interventions that restructure the physical layout or organization of options consistently outperform those that merely provide additional information or reminders. Food choices turned out to be especially responsive, with effect sizes up to two and a half times larger than in other behavioral areas.5PubMed Central. The effectiveness of nudging: A meta-analysis of choice architecture interventions across behavioral domains In one school-based trial, for example, simple changes to how fruit was presented at lunch significantly increased children’s consumption of fruit, vitamin C, and fiber, with the effect showing up across children regardless of how much fruit they were eating before the intervention.6PubMed Central. A low-cost Behavioural Nudge and choice architecture intervention targeting school lunches increases children’s consumption of fruit: a cluster randomised trial
The practical takeaway is that changing default options or rearranging the physical environment is generally more effective than trying to persuade people with facts. This insight has been adopted by governments and organizations worldwide, from organ donation opt-out systems to cafeteria layouts designed to promote healthier eating.
Broken Windows and the Social Norms of Public Spaces
Field experiments on public behavior have provided some of the strongest evidence for the “broken windows” idea, the theory that visible signs of disorder in an environment encourage further rule-breaking. Researchers have tested this by experimentally varying the level of physical or social disorder in a location and measuring whether people litter more, steal, or violate other norms.
Three field experiments testing the broken windows thesis found that existing norm violations do foster further violations, both of the same norm and of different norms. But the effect had a limit: it was confined to low-cost situations where breaking the rule carried minimal personal risk. In an interesting twist, the disorder effect was stronger in neighborhoods with high social capital, possibly because residents of tight-knit communities are more sensitive to environmental cues about what is considered acceptable.7Rationality and Society. Disorder, social capital, and norm violation: Three field experiments on the broken windows thesis
On the flip side, clean environments send their own signal. A field experiment focused on cigarette littering found that people were significantly less likely to litter in a clean setting compared to an already-littered one.8Environmental Research Communications. Clean environments as a social norm: a field experiment on cigarette littering Another experiment in Japan showed that adding flowers and improved lighting to a riverside reduced littering during the intervention period, and the effects persisted even after the physical additions were removed, suggesting the changed behavior had become at least partly self-sustaining.9Asian Journal of Social Psychology. Preventing riverside littering through structural change: An action research field experiment using two intervention measures These studies illustrate that social experiments in real-world settings can do more than observe behavior; they can actively shape it.
Policy-Scale Experiments and Guaranteed Income
Some of the highest-stakes social experiments are randomized controlled trials of government policies, where the treatment is not a lab manipulation but a real intervention that affects people’s daily lives. Guaranteed basic income trials have attracted enormous public attention in recent years, and the experimental evidence they generate is both detailed and contested.
In one of the most rigorous recent trials, a thousand low-income individuals in the United States were randomized to receive $1,000 per month unconditionally for three years, while a control group of two thousand received $50 per month. The results painted a nuanced picture. Total individual income excluding the transfers fell by about $1,900 per year relative to the control group, and labor market participation dropped by about four percentage points. Participants and their partners each reduced work hours by roughly one to two hours per week, with the freed-up time going mostly to leisure rather than education, caregiving, or other productive activities. Measures of subjective well-being rose in the first year of transfers but then returned to control group levels.10NBER. The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States
A broader systematic review and meta-analysis of guaranteed basic income interventions across high-income countries found more varied effects depending on program design. Programs where benefits were gradually withdrawn as other income rose appeared to reduce food insecurity and keep youth in school past the compulsory age. Supplemental payments layered on top of existing benefits were associated with reduced mental distress. And programs offering amounts above standard social assistance levels improved life satisfaction and subjective financial well-being.11PubMed Central. Effects of guaranteed basic income interventions on poverty‐related outcomes in high‐income countries: A systematic review and meta‐analysis The design details matter enormously, and a blanket verdict on whether guaranteed income “works” misses most of what these experiments reveal.
Digital Experiments and the Ethics Problem
The internet transformed social experiments by making massive sample sizes possible. In 2012, Facebook manipulated the News Feeds of nearly 700,000 users without their knowledge, reducing the amount of positive or negative emotional content they saw. The result was a measurable shift: when positive posts were reduced, people wrote fewer positive things and more negative things themselves, and vice versa. The researchers concluded that emotional contagion can occur without in-person contact or nonverbal cues, through the text-based medium of a social network alone.12PubMed Central. Experimental evidence of massive-scale emotional contagion through social networks
The backlash was fierce and immediate. The study had no informed consent in any traditional sense; users were technically covered by Facebook’s terms of service, but critics argued this was nowhere near sufficient for a study that deliberately altered people’s emotional environments. A detailed ethical analysis argued that the core problem was the co-opting of user data in a way that violated identity-based norms, exploiting the vulnerability of people who disclose personal information in an environment where they have no control over how that information is used or how their experience is shaped.13Research Ethics. Facebook’s emotional contagion study and the ethical problem of co-opted identity in mediated environments where users lack control The controversy did not halt digital experimentation — tech companies run thousands of A/B tests daily — but it sharpened the conversation about where the boundary between product optimization and human subjects research lies.
Why Observation Itself Is a Variable
One of the most persistent headaches in social experimentation is that people who know they are being studied tend to behave differently. This phenomenon, broadly called the Hawthorne effect, has been debated for decades. A systematic review of 19 studies designed specifically to measure the size of this effect found that the overall odds of changed behavior when participants knew they were observed were about 17% higher than when they did not. Around two-thirds of the included studies found at least some statistically significant evidence that being watched or having to report on one’s own behavior changed the behavior itself.14PubMed Central. Systematic review of the Hawthorne effect: New concepts are needed to study research participation effects
The review’s conclusion was candid: the effect is real, but researchers still know remarkably little about the specific conditions that trigger it, the mechanisms behind it, or its typical size. This means every social experiment involving aware participants carries some degree of uncertainty about whether the results reflect genuine behavior or performance for an audience. It is one reason field experiments conducted in natural settings, where participants do not know they are being observed, are sometimes considered more reliable for studying behavior as it naturally occurs.
Diffusion of Responsibility in Groups
Social experiments have also probed how being part of a group changes a person’s sense of personal responsibility. In a study using a game-like task where participants could act alone or as part of a pair, people reported a significantly lower sense of personal agency when they were acting together with someone else compared to acting alone. Their brain responses to the outcomes of their actions were also dampened in the group condition, suggesting that the reduced feeling of responsibility was not just a verbal report but reflected a genuine change in how the brain processed the experience.15PubMed Central. Beyond self-serving bias: diffusion of responsibility reduces sense of agency and outcome monitoring
This connects to the well-known bystander effect, where individuals in a crowd are less likely to help someone in distress because responsibility feels spread across everyone present. The experimental evidence suggests this is not laziness or apathy but a measurable shift in how the brain registers personal causation when others are involved.
The Replication Crisis and What Came After
Many of the most famous social experiments were conducted under research norms that would not pass muster today. Beyond the ethical issues, there is a scientific one: a number of classic findings in social and behavioral science have failed to replicate when independent teams tried to reproduce them. Large-scale replication projects found success rates far lower than anyone expected, triggering what became known as the replication crisis.
The silver lining is that this crisis prompted real structural changes. Researchers have increasingly described this period as a “credibility revolution” rather than a collapse, pointing to the adoption of pre-registration (publicly committing to a study’s methods and hypotheses before collecting data), open data sharing, and larger sample sizes as lasting improvements to how research is conducted.16PubMed Central. The replication crisis has led to positive structural, procedural, and community changes For a reader encountering a classic social experiment for the first time, the practical lesson is straightforward: findings that have been replicated multiple times across different settings deserve more confidence than single dramatic studies, no matter how famous.
Simulated Societies and AI Agents
A new frontier in social experimentation does not involve human participants at all. Researchers have begun using large language models to create simulated agents that interact in virtual social environments, allowing them to model complex social dynamics at scale without the logistical challenges, costs, or ethical constraints of working with real people.17arXiv. AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society These simulations can, in principle, test hundreds of policy variations or social scenarios simultaneously in ways that would be impossible with human subjects.
Whether AI-generated agents actually capture the messiness of real human behavior is an open question. Proponents argue that generative AI has the potential to improve techniques like survey research, online experiments, and agent-based modeling.18PubMed Central. Can Generative AI improve social science? Skeptics counter that language models trained on human text may reproduce the biases and patterns found in that text without replicating the underlying cognitive and emotional processes that drive real social behavior. The technology is advancing quickly, but for now, simulated social experiments are best understood as a complement to real-world studies, not a replacement.
Prosocial Behavior Beyond Humans
Some of the most thought-provoking social experiments have been conducted with non-human primates, revealing that the social tendencies researchers study in people are not exclusively human. Common marmosets, small monkeys from South America, have been shown to spontaneously provide food to unrelated, non-reciprocating individuals in controlled experimental settings. The marmosets were not trading favors or helping relatives; they were offering food to strangers with no expectation of return.19PubMed Central. Other-regarding preferences in a non-human primate: common marmosets provision food altruistically
Humans and marmosets are both cooperative breeders, meaning that raising offspring involves help from individuals beyond the biological parents. The researchers proposed that this shared breeding system may be the evolutionary root of prosocial behavior in both species, rather than advanced cognitive abilities like understanding another’s mental state. If that is right, the impulse to help others, the very thing many social experiments are designed to study, may be far older and more deeply rooted than the cultural norms we tend to credit for it.