Authors: Albert Bandura, Dorothea Ross, Sheila A. Ross (1961). First published in Journal of Abnormal and Social Psychology, 63, 575-582.
Translation: Kulinkovich T.O.
Previous research aimed at explaining the phenomenon of identification based on chance learning showed that children readily imitate the behavior displayed by the model of the elder in his presence (Bandura & Huston, 1961). A series of experiments by Blake (1958) et al. (Grosser, Polansky, & Lippitt, 1951; Rosenblith, 1959; Schachter & Hall, 1952) also found that observing model responses had a facilitating effect on the behavior of subjects under direct social exposure.
Although these studies have provided compelling evidence for the influence and control exerted on others by the model’s behavior, more serious research on learning by imitation involves generalizing the patterns of imitated responses to novel conditions in which the model is absent.
In the experiment presented in this article, children were shown aggressive and non-aggressive adult behaviors and then measured the extent of learning through imitation that occurs in a new situation in the absence of a model. Subjects exposed to the aggressive model were predicted to reproduce aggressive behaviors similar to those of the model and would differ in this respect from subjects exposed to non-aggressive models and from those exposed to no behavior at all. This hypothesis suggested that subjects learn imitative behavior as a result of previous rewards, and that these tendencies extend to some extent into their adult experience (Miller & Dollard, 1941).
It was also hypothesized that observation of subordinate non-aggressive models would have a generalized inhibitory effect on the subsequent behavior of the subjects, and that this effect would be reflected in the differences between the non-aggressive and control group, showing significantly more aggression among the subjects of the latter group.
Hypotheses were also accepted regarding the influence of the sex of the model and the sex of the subject on the imitation of behavior. According to Falls and Smith (1956), preschoolers perceive their parents’ different preferences for behaviors that match the child’s gender. These data, as well as informal observation, show that parents reward their children’s gender-appropriate behavior and disapprove or punish inappropriate ones, so that a male child is unlikely to be rewarded for exhibiting female-appropriate behaviors such as cooking or for assimilating others. aspects of maternal role, but the same types of behavior are usually welcome in girls. As a result of different reinforcement experiences, tendencies to imitate male or female models take on a different prominence. Thus, subjects are expected to imitate the behavior of a same-sex model to a greater extent than that of an opposite-sex model.
Since aggression is a masculine type of behavior, boys should be more prone to imitate aggression than girls, and the differences will be most noticeable in subjects who will be shown an aggressive male model.
Methods
Subjects
The subjects were 36 boys and 36 girls from Stanford University Kindergarten, aged 37 to 69 months, with an average age of 52 months.
Two adults, a man and a woman, served as the model, and one woman supervised the study of all 72 children.
Experiment plan
The subjects were divided into eight experimental groups of 6 people each and a control group of 24 people. Half of the subjects were shown an aggressive model, while the other half were shown a repressed and non-aggressive one. These groups were further divided into boys and girls. Half of the subjects observed aggressive and non-aggressive patterns of behavior of the same sex with them, the other half — of the opposite sex. The control group was not shown any previous models and was tested only in the subsequent general situation.
The expectation that the level of aggression of the subjects would be positively correlated with their willingness to imitate aggressive behaviors seemed justified. However, in order to increase the accuracy of the processed comparisons, subjects in the experimental and control groups were selected individually based on the degree of their aggressive behavior in social relationships in kindergarten.
Subjects were rated on a five-point scale by the experimenter and kindergarten teacher who were intimately familiar with the children. These scores indicated the degree to which the subjects expressed physical aggression, verbal aggression, aggression towards inanimate objects, as well as the containment of aggression. The latter assessment, related to the individual’s desire to restrain aggressive responses in states of high aggressive urge, provided a measure of aggressive anxiety.
Both experts independently assessed 51 subjects, which allowed for further consensus assessment. The reliability of the total aggression score, estimated from the Pearson correlation values, was 0,89.
The total score was obtained by summing the scores on the 4 aggressiveness scales; based on these scores, the subjects were combined into triplets and randomly assigned to one of the two experimental groups, or to the control group.
Experiment conditions
At the first stage of the experiment, the subjects were individually placed in the experimental room, and the model, located in the corridor, was asked to come in and join the game. The experimenter then led the child to a corner of the room, which had been set up as a play area. After the experimenter seated the child at a small table, she showed how the child could create pictures using stamps, and also offered colored stickers. Toy stamps included various geometric shapes; the stickers were eye-catching multicolored pictures of animals, flowers, and cowboy figurines drawn against pastures. These activities were selected in a previous study in kindergarten as being the most interesting for children.
After the subject was placed in a corner of the room, the experimenter escorted the model to another corner of the room, which contained a small table with a chair, a construction set, a mallet, and a 5-foot Bobo inflatable doll. The experimenter explained that these were toys for the model, after which she left the room.
In the group with non-aggressive conditions, the model quietly and calmly assembled the designer, not paying any attention to the Bobo doll.
And in the group with aggressive conditions, the model began assembling the construction kit, but after no more than a minute turned to the doll and spent the remaining time in aggressive actions towards it.
Imitative behavior can be clearly demonstrated if the model shows new patterns of behavior that are unlikely to have occurred independently of observing the behavior of the model, and if the subject reproduces these actions in exactly the same form. For this reason, in addition to beating the Bobo doll, an action that could be demonstrated by the child independently of the observation of the model, the model also showed specific aggressive actions that were later regarded by the children as imitating actions. The model put Bobo on her back, sat on top and hit him several times in the nose. Then the model lifted the doll, took a mallet and hit the doll on the head. After the beater attack, the model aggressively tossed the doll up and kicked it around the room. These chains of physically aggressive acts were repeated about three times, accompanied by verbally aggressive actions such as: “Kick him in the nose”, “Throw him up”, “Hit him”, “Pow”, and two non-aggressive comments: “ He’s coming back for more” and “He’s definitely a tough guy”.
Thus, in the situations demonstrated, the subjects were given an interesting task that kept their attention and, at the same time, provided observation of the behavior of the model without any instructions to observe and learn the reactions of the model. As long as the subjects could not demonstrate the aggressive behavior of the model, any learned responses remained at the purely observational, or covert, level.
After 10 minutes, the experimenter entered the room, told the subject that he would now go to another playroom, and said goodbye to the model.
Awakening aggression
The imitating behavior of the subjects was studied in another experimental room, which was separated from the main building of the kindergarten. Thus, the two experimental situations were clearly separated; indeed, many of the subjects were under the impression that they were no longer within the precincts of the kindergarten.
Before the study of imitating behavior, all subjects were subjected to moderate provocation of aggression, which provided them with a certain degree of predisposition to the manifestation of aggression. Violent experiments were included for two main reasons. In the first place, observing aggressive behavior displayed by others generally reduces the likelihood of the observer exhibiting aggression (Rosenbaum & deCharms, 1960). Consequently, subjects from aggressive conditions, in comparison with the other group and with the control group, will have a weaker disposition to aggression, being influenced by the behavior of the models. Secondly, if subjects from non-aggressive conditions express little aggression with an appropriate aggressive urge, the presence of restraining mechanisms can be seen.
In this regard, the experimenter brought the subject to the reception room, which contained the following relatively attractive toys: a fire engine, a locomotive, a jet aircraft, a carriage, and a doll set consisting of a chest of drawers, a pram, and a crib. The experimenter explained that the child could play with the toys, however, as soon as the child was included in the game (usually after 2 minutes), the experimenter noted that these were her best toys, that she did not allow anyone to play with them, and that she decided to leave these toys are for other children. However, the child could play with any toys in the other room, after which the experimenter and the child were moved to the adjoining experimental room.
It was necessary for the experimenter to remain in the room throughout the experiment; otherwise, many children would either not want to be alone or leave the room before the end of the experiment. However, in order to minimize her influence on the experiment, the experimenter remained as inconspicuous as possible, doing her paperwork at a table at the back of the room and avoiding any interaction with the child.
Delayed Simulation Check
The experimental room contained a variety of toys, some of which could be used in imitation or non-imitation of aggressive behavior, and others designed to induce non-aggressive behaviors. «Aggressive» toys included a three-foot Bobo doll, a mallet, two bows, and a ball suspended from the ceiling with a face painted on it. Non-aggressive toys included a tea set, crayons and coloring books, a ball, two dolls, three bears, cars and trucks, and plastic animals.
In order to minimize the differences in behavior caused by the arrangement of the toys in the room, the play material was arranged in a strict order for each individual experiment.
The subject spent 20 minutes in the room, during which time his behavior was judged in categories of predetermined responses by judges who observed the session through a one-way mirror in an adjoining room. The 20 minutes were divided into five second sessions and thus amounted to 240 single reactions for each subject.
The male model counted the experimental sessions for all 72 children. Except when he modeled, he didn’t know what groups the kids were from. In order to establish agreement in the measurements, the performance of half of the subjects was also measured independently by a second observer. Thus, one of the two observers was usually unaware of the conditions in which the children were placed. Be that as it may, all but two subjects from the group with aggressive conditions showed new aggressive reactions of the model, while subjects from other conditions only occasionally showed such reactions, subjects who were exposed to the model were easily recognized among others due to their characteristic behavior.
Measured responses were included in highly specialized isolated behavior classes and had a high probability of co-occurrence with a coefficient in the region of 0,90.
Measuring reactions
Three measurements of the simulation were obtained:
- Simulated physical aggression: This category included the acts of hitting the bobo doll with a mallet, sitting on the doll and pulling its nose, kicking and throwing it into the air.
- Imitation of verbal aggression: subjects repeated the phrases “Hit him”, “Throw him”, “Kick him”, “Throw him in the air” and “Pow”
- Imitation of non-aggressive verbal reactions: the subjects repeated the phrases: «He comes back to get more» and «He’s definitely a tough guy.»
- During the preliminary test, several subjects imitated the main components of the behavior of the models, but did not repeat the whole chain of behavior, or directed their actions to other toys. Two reactions of this type, however, were counted and defined as imitative behavior.
- Beater aggression: Subjects hit other dolls with a mallet.
- Sitting on a Bobo doll: Subjects placed Bobo on the floor and sat on him, but were not aggressive towards him.
The following aggressive reactions have also been noted:
- Hitting Bobo Dolls: Subjects hit, spanked, or aggressively pushed the doll.
- Non-imitating physical and verbal aggression: this category includes physical aggressive acts directed at objects other than Bobo, aggressive words not demonstrated by the model, such as “Shoot Bobo”, “Cut him”, “Stupid doll”, “Horses fight, bite”
- Aggressive gun play: Subject shoots a bow or gun and fires imaginary shots at objects in the room.
Actions were also evaluated when the subjects played non-aggressively or sat quietly without playing at all.
The results
Complete simulation of model behavior
Subjects in aggressive settings reproduced a large number of physical and verbal aggressive actions similar to those of the model, and their average values differed significantly from those of subjects in non-aggressive conditions and in the control group, who did not demonstrate imitative aggressive actions at all.
Since subjects in the non-aggressive and control groups had only a few scores (almost 70% of the subjects had zero scores) and no assumption could be made about the homogeneity of the sample, Friedman’s rank analysis was used to test the significance of the results.
The assumption that the exposure of subjects to aggressive models increases the likelihood of aggressive behavior was confirmed. Exploratory conditions are highly significant for both physical and verbal aggression. A comparison of pairs of assessments shows that the differences obtained were mainly due to the aggression expressed by the subjects who were presented with aggressive models. Their scores were significantly higher than the scores of the nonaggressive and control groups, which did not differ from each other.
The imitation was not limited to the aggressive reactions of the model. About one-third of the subjects in the aggressive setting also repeated the non-aggressive verbal responses of the model, while none of the subjects in the other groups demonstrated such utterances. These differences, tested by the Cochran Q test, were found to be significant at a significance level below the 0.01 level.
Partial Simulation of Model Behavior
Differences in the intended direction were also obtained in two dimensions of partial simulation.
The analysis of differences in scores was based on the use of mallet aggression by the subjects against other puppets and showed that the experimental conditions are a statistically significant source of differences. In addition, individual sign tests showed that the aggressive and control groups, compared with children from non-aggressive conditions, expressed significantly more beater aggression. These differences were partially noted for female subjects. Girls who observed the non-aggressive model showed a mean number of beater aggression of 0,5 compared to 18,0 and 13,1 for girls in the aggressive and control groups.
Although the subjects who observed the aggressive patterns presented more beater aggression (M = 20.0) than the subjects in the control group (M = 13.3), the differences were not statistically significant. With regard to the partially mimic reactions of sitting on a Bobo doll, group-wide differences were found to be significantly below the significance level of 0,01. Pairwise comparison of the ratings shows that subjects from aggressive conditions reproduced this aspect of the model’s behavior much more often than subjects from non-aggressive conditions (p = .018) or from the control group (p = .059). On the other hand, the last two groups did not differ from each other.
Non-imitating aggression
An analysis of the differences in the remaining aggressive manifestations shows that the experimental conditions did not affect the level of involvement of the subjects in playing with weapons or beating the Bobo doll. However, the influence of experimental conditions turned out to be highly significant (χ2r = 8.96, p
Influence of the sex of the model and the sex of the subject
The hypothesis that boys are more likely to imitate the aggression shown by the model was only partially confirmed. The t-test showed that boys showed more imitative physical aggression than girls (t = 2.50 p
The use of non-parametric tests, due to the too skewed distribution of subjects’ scores from non-aggressive and control conditions, precluded a full test of the influence of the sex of the model per se, and the various relationships between the main effects. An analysis of the average scores of subjects from aggressive conditions, however, clearly shows the possibility of correlations with the sex of the model. The effect of these relationships is more significant for male models. Male subjects, for example, showed more physical (t = 2.07, p
Data from the non-aggressive and control groups provided further evidence that the behavior of the male model is more influential than the behavior of the female model in generalized situations.
In addition, with the exception of a large number of beater aggression expressed by subjects from the control group, no significant differences were obtained between the non-aggressive and control groups. However, the data suggest that the lack of significant differences between these groups is mainly due to the fact that subjects exposed to the non-aggressive female model did not differ from those in the control group in any of the types of aggression measured. On the other hand, with respect to the male model, the differences between the groups are significant. Compared to controls, subjects exposed to the nonaggressive male model expressed significantly less physical aggression (p = .06), less verbal aggression (p = .002), less beater aggression (p = .003), less non-imitating physical and verbal aggression (p = .03) and were also less likely to hit the bobo doll (p = .07).
Although subgroup comparisons, in which some pooled differences do not reach statistical significance, are more likely to be based on random differences, the congruence of the results nevertheless reinforces the assumption that the model influences behavior.
Non-aggressive behavior
Despite the exclusion of expected sex differences, Lindquist’s (1956) analysis of differences in non-aggressive responses provided some significant results.
Female subjects spent more time than boys playing with dolls (p
It is also worth mentioning the following differences generated by the experimental conditions. Subjects from the non-aggressive setting were significantly more likely to play non-aggressive games with dolls than subjects from the aggressive setting (t = 2.67, p
Even more noteworthy is the evidence that subjects who observed the non-aggressive model spent twice as much time sitting still and doing nothing as compared to those in the aggressive setting (t = 3.07, p
Discussion
Many modern studies of social learning are based on the formation of new types of behavior through rewards and punishments. However, if reactions are issued, they cannot be affected. The results of this study provide important evidence that observing the behavior of another is an effective means of extracting certain forms of responses that are unlikely to occur on their own. Indeed, social imitation can accelerate or inhibit the acquisition of new behavioral responses without the need to reward the successful approximation proposed by Skinner (1953).
Subjects who were given the aggressive model later displayed a large amount of physical and verbal aggression (and non-aggressive responses) that were essentially identical to those of the model. At the same time, subjects who were shown a non-aggressive model, or were shown nothing, very rarely presented such reactions.
The fact that observation of an adult’s aggressive behavior permits a child’s aggressive behavior may weaken inhibitions and lead to an increase in the likelihood of aggressive reactions in subsequent frustration. However, the fact that the subjects expressed their aggression through the new behaviors displayed by the model provides evidence for learning through imitation.
In a study of imitating behavior by Miller and Dollard (1941), adult or peer models showed highlighted responses for which they were consistently rewarded, and subjects were rewarded in the same way when they showed similar responses. Although these experiments were widely used in demonstrating learning through imitation, in fact they only included special cases of differing learning in which the behavior of others served as a distinctive stimulus for responses that were already the subjects’ behavioral repertoire. Auditory or visual cues from the environment could easily be replaced by social stimuli in order to facilitate highlighted learning. In turn, the imitation process studied in our study differed in several important ways from Miller and Dollard’s experiment, in particular in that subjects learned to integrate fragmented responses into relatively coherent patterns of new responses solely through observation of social patterns, without being able to demonstrate the behavior of the model in the conditions in which they saw it, and also without any reinforcement of their behavior from the model or observers.
There is a lack of an adequate theory capable of explaining imitative learning. Suggested explanations (Logan, Olmsted, Rosner, Schwartz, & Stevens, 1955; Maccoby, 1959) assume that the imitator implicitly exhibits model responses. If the assumption can be added to this that rewards and punishments are given independently in connection with covert responses, the process of imitation learning can be viewed in the same terms as covert instrumental learning through trial and error. However, in the early stages of development, an individual’s behavioral repertoire is likely to increase through a process of classical conditioning (Bandura & Huston, 1961; Mowrer, 1950).
The results of the experiment also provided evidence that the male model influences behavior imitation to a greater extent than the female model. When analyzing the Gender x Model relationship, for example, subjects exposed to a non-aggressive male model showed less aggressive behavior than subjects in the control group, while comparisons of subjects with a female model did not show significant differences.
In a study of learning through imitation, Rosenblith (1959) also found that male models were more effective in influencing children’s behavior than female models. Rosenblith offered an explanation that kindergarten conditions may represent a certain social deprivation for male adults, which in turn increases the value of the male reward.
The trends identified in this study offer an alternative explanation. In cases of pronounced masculine behavior, such as aggression, there is a tendency for children of both sexes to imitate male behavior more than female behavior. On the other hand, in the case of verbal aggression, which is less gender-related, a greater proportion of imitation occurs in relation to the model of the same sex as the subject. These trends, and the fact that boys in general mimic physical aggression more than girls, but do not differ from them in mimicking verbal aggression, suggest that subjects may be influenced differently by the sex of the model, but it is worth considering the degree whether the behavior in question is typical of the sex.
The foregoing discussion has shown that masculinity-femininity, more than the other personality characteristics of the model, is the most variable—an assumption that cannot be directly tested with wrist data. However, it has been clearly shown, especially from the spontaneous utterances of boys when presented with an aggressive female model, that at least some of the subjects responded based on gender discrimination and their prior knowledge of what behavior is appropriate for the sex (e.g. “Who is this woman? Ladies don’t behave like that. Ladies should act like a lady…» but did not swear). On the other hand, male aggression was more often seen as inherent and accepted by both boys (“Al is a good fighter, he beat Bobo. I want to fight like Al”) and girls (“This man is a strong fighter, he fought and fought and could knock Bobo right to the floor, and if Bobo got up, he would say, «Kick your nose. He’s a good fighter, just like dad»).
The evidence that subjects with a calm model were more restrained and unreceptive than subjects from aggressive conditions, as well as the obtained differences in the measurement of aggression, show that the presentation of restrained models not only reduces the likelihood of aggressive behavior, but also generally limits the behavioral range of the subjects.
“Identification with the aggressor” (Freud, 1946) or “defensive identification” (Mowrer, 1950), whereby a person transforms himself from an object into a subject of aggression through the adoption of the properties of an aggressive threat model to reduce anxiety, is widely used as an explanation for the imitative learning of aggression.
However, the development of aggressive patterns of children’s responses to aggressive punishing adults may simply reflect object substitution, without the mechanism of defensive identification being involved. In a study of past childhood experiences of antisocial adolescents (Bandura & Walters, 1959) and young hyperaggressive youths (Bandura, 1960), their parents were found to be depriving and punishing towards aggression directed at them. On the other hand, they actively encouraged their sons’ aggressive behavior towards outsiders. This pattern of differentiated rewards for aggressive behavior inhibited the boys’ aggression against the real instigator and encouraged a shift in aggression toward objects and situations with less deterrent responses.
Moreover, the results of previous studies (Baudura & Huston, 1961), in which children imitated aggression presented by caregivers and non-caregivers equally, together with the results of this study, in which subjects readily imitated the aggressive behavior of models that were more or less neutral figures show that the mere observation of aggression, regardless of the quality of the relationship with the model, is a sufficient condition for imitating aggression to occur in children. A comparative study of the imitation of the aggressive behavior of models who are intimidating, respected and loved, or who are neutral figures, will shed light on whether any other theory than the theory of “identification with the aggressor” can explain the process of imitation.
Conclusion
Twenty-four preschoolers were assigned to each of the three conditions. One experimental group observed aggressive behavior patterns of adults, the other observed restrained non-aggressive models, and subjects from the control group did not observe behavior patterns at all. Half of the subjects from the experimental conditions observed the behavior of models of the same sex with them, and half of the opposite sex. Then the behavior of the subjects was tested for the occurrence of imitative and non-imitative reactions in new conditions in the absence of a model.
A comparison of the behavior of subjects in a generalized situation showed that subjects who were presented with an aggressive model showed a large number of aggressive reactions corresponding to the behavior of the model, and their average scores differed significantly from the scores of subjects from the non-aggressive and control groups. Subjects from aggressive settings also showed significantly more partially imitative reactions and non-imitative aggression and were generally less restrained in their behavior than subjects from non-aggressive settings.
It was found that the gender of the model affected the simulation in different ways. Boys showed more aggression compared to girls, imitating the behavior of the male model, these differences were manifested in masculine behavior.
Subjects who observed a non-aggressive model, especially a depressed male model, were generally less aggressive than those in the control group.
The connection of the results obtained in this experiment and similar studies with the psychoanalytic theory of identification with the aggressor was discussed.