Observational learning: what Bandura's Bobo doll studies actually showed

Observational learning means learning a behaviour by watching someone else perform it. You do not have to act yourself, and nobody has to reward you. The psychologist Albert Bandura described four steps that decide whether watching turns into copying: attention, retention, reproduction, and motivation. A second point belongs in the definition: learned is not shown. Whether you perform what you learned depends on the consequences, for the model and for you.

The term itself is quick to define. The studies behind it take more care, because three studies from 1961, 1963, and 1965 circulate under one name, the Bobo doll experiment. This article separates the three, gives their numbers, and states what they do not prove.

• Observational learning and modelling mean the same in most textbooks. The wider term is social learning theory.
• A model does not have to stand in the same room. Bandura's successors describe live, verbal, and symbolic models.
• In 1961 children copied an aggressive model although nobody received a reward.
• In 1963 models on film worked about as strongly as models in the room.
• In 1965 Bandura showed that consequences for the model change performance, not learning.
• The studies used a doll, measured immediate play only, and would not pass a modern ethics board.

What observational learning means

Observational learning is learning through watching a model, which is the person whose behaviour you observe. The behaviour can be new and complex: a tennis serve, a polite apology, an aggressive script. What matters is that you acquire it without performing it during learning and without direct reinforcement. That claim separates the theory from operant conditioning, where a behaviour is strengthened because performing it pays off.

Most textbooks use modelling as a synonym. The OpenStax chapter on the topic is titled Observational Learning (Modeling), and it treats the two words as one process. Social learning theory is the wider frame: Bandura's account of how observation, thought, and consequences together shape behaviour.

A model does not have to stand next to you. OpenStax names three kinds. A live model shows the behaviour in person, like a coach who demonstrates a serve. A verbal model does not perform it but explains or describes it, like a tutor who talks a student through a proof. A symbolic model is a real or fictional person in books, films, television, games, or online media. That third kind is why the 1963 study matters: it tested whether a screen works like a room.

The second half of the definition is easy to miss. Learning something and showing it are two different events. A child can learn an aggressive script perfectly and still stay quiet, because the model was punished or because trying it looks costly. The 1965 study tested exactly that split, and exam questions return to it again and again.

The four steps of modelling

Bandura described four connected steps that govern modelling. Each step names one place where copying can fail.

The first two steps are acquisition: taking the behaviour in. The last two are performance: bringing it out. That split answers the standard exam question about why learning and showing diverge. Some pages count five steps by listing reinforcement separately from motivation. Bandura's list has four, with reinforcement inside the motivation step: vicarious reinforcement raises the will to copy, vicarious punishment lowers it. When a question asks for the steps, give the four in this order.

Three studies, one name: the Bobo doll experiments

The Bobo doll is an inflatable doll that rights itself after a blow. Summaries often tell one experiment with this doll. The literature holds three studies with three different questions.

1961: copying with no reward

The 1961 study by Bandura, Ross, and Ross used 36 boys and 36 girls from the Stanford University nursery school. The children were 37 to 69 months old, with a mean age of 52 months. Groups of 24 saw an aggressive model, a calm model, or no model at all. The researchers matched the groups on rated aggressiveness first, so the groups started alike.

The model was an adult man or woman. The aggressive model sat on a five-foot Bobo doll and struck it on the nose. Then the model hit it on the head with a mallet, threw it in the air, and kicked it across the room. The model repeated that sequence about three times, with cries such as "Sock him in the nose" and "Pow". The session lasted ten minutes.

Then every child played with attractive toys for about two minutes before the experimenter took them away again. That mild frustration was meant to put all groups under some drive toward aggression. The test room held a smaller Bobo doll, a mallet, dart guns, and peaceful toys such as crayons and a tea set. The model was absent. Observers scored 20 minutes of play through a one-way mirror in five-second intervals.

Children with an aggressive model reproduced much of the physical and verbal aggression. The other groups showed almost none: about 70 percent of the children without an aggressive model scored zero on imitative aggression. Boys copied more physical aggression than girls. Verbal copying did not differ by sex.

One detail is missing from many summaries. Plain punching of the doll did not separate the groups: children without a model hit the doll about as often as children with one. The authors had expected that, because punching a doll is common play. The model therefore showed striking novel acts that children rarely invent alone, and the finding concerns copying of those new acts.

Neither the model nor the children received any reward. The children learned the acts by watching alone.

1963: models on film

The 1963 study asked whether a model on film works like a model in the room. Forty-eight boys and 48 girls formed four groups of 24. One group saw an aggressive model in the room, the second saw the same models on film, the third saw a cat film called Herman the Cat, and the fourth saw no model. Two points matter. The live-model and control data came from the 1961 experiment. And the cartoon cat was no animation: the female model wore a cat costume and played against a bright fantasy set.

Mean total aggression was 83 with a live model, 92 with a human film model, 99 with the cartoon film, and 54 with no model. The three model groups did not differ significantly from each other in total aggression. Eighty-eight percent of the children with a human model, live or filmed, copied at least part of the acts. With the cartoon cat the figure was 79 percent.

The authors already separated learning from showing in this paper. From informal observation they noted that children do not, as a rule, perform everything television characters do, even characters they find attractive. That caution became the question of the next study.

1965: learned is not shown

Bandura tested the split directly in 1965, as the sole author. Children saw a film with an aggressive model. The film ended in three ways: the model was rewarded, punished, or faced no consequence.

At first test, children who had seen the model punished performed far fewer of the shown acts than the other two groups. Then Bandura offered every child attractive rewards for copying the model. That incentive largely removed the differences. All three groups had learned the same amount.

Two terms belong to this study. Vicarious reinforcement means you see a model rewarded for a behaviour, and your will to copy it rises. Vicarious punishment means you see a model punished, and your will to copy it sinks. Both build on reinforcement and punishment from operant conditioning. The difference is that the consequence hits the model, not you.

Two mix-ups are common. Some summaries credit the study to Bandura and Walters. The 1965 paper names Bandura alone. And the reward and punishment conditions do not belong to the 1961 study, although even the OpenStax chapter describes them under a 1961 citation. In the papers they belong to 1965.

Four things a model does, with examples

A model does more than show new behaviour. A 2025 portrait of Bandura's work describes four effects. The examples below are ordinary school scenes that show each one.

The same four effects explain why adults are warned to model what they preach. Children copy homework discipline, table manners, and reading habits from live models at home, and stunts and catchphrases from symbolic models on screens.

What the studies do not show

The three studies prove that children learn by watching. Four limits belong with that claim.

• The children struck a doll, not a person. An EBSCO survey of the experiment calls the step from doll to human blows problematic.
• The Bobo doll rights itself after every hit. Children may have read it as a target built for hitting, says the same survey.
• The 1961 and 1963 studies measured play directly after watching. Those scores say nothing about lasting effects.
• The studies exposed small children to adult aggression on purpose. A 2022 ethics review calls that an ethical problem with unknown long-term consequences, and sets the studies beside the Milgram experiment.

The media question stays open too. OpenStax frames it as a dispute: Craig Anderson and colleagues read the evidence as support for a causal link between media violence and aggressive thoughts and acts, while Christopher Ferguson and others object that the research leaves out further risk factors such as mental health and family life. The Bobo doll studies alone cannot settle that debate. They show that filmed models teach, not that television makes violent adults.

Telling the terms apart

Exams test this cluster with look-alike terms. Sort each term into its step or its study. Attention, retention, reproduction, and motivation are the four steps. Vicarious reinforcement and vicarious punishment describe consequences for the model. Inhibition, disinhibition, and elicitation describe what a model does to existing behaviour. The 1961 study stands for learning without reward, 1963 for film against live, and 1965 for learning against performance. Terms that look alike are easier to hold apart when you learn them as a group rather than one at a time.

Save the terms you meet in this article in a Psychology World list. Drill the list as flashcards, as multiple choice once it holds four or more terms, or by typing the answers. What each study mode does and the four steps of the learning loop are set out on the home page. The same loop that Bandura described as acquisition and performance is the loop the app is built around: take behaviour in by watching and reading, then bring it out by testing yourself. Latent learning makes the same point from a different experiment: learning can sit hidden until an incentive calls it out.