Several years ago a youth counselor told me about the dilemma he faced when dealing with violent young men. His direct impressions simply didnt match what he had been taught. He saw his violent clients as egotists with a grandiose sense of personal superiority and entitlement, but his textbooks told him that these young toughs actually suffered from low self-esteem. He and his staff decided they couldnt go against decades of research, regardless of what they had observed, and so they tried their best to boost the young men's opinions of themselves, even though this produced no discernible reduction in their antisocial tendencies.

The view that aggression stems from low self-esteem has long been common knowledge. Counselors, social workers and teachers all over the country have been persuaded that improving the self-esteem of young people is the key to curbing violent behavior and to encouraging social and academic success. Many schools have students make lists of reasons why they are wonderful people or sing songs of self-celebration. Many parents and teachers are afraid to criticize kids, lest it cause serious psychological damage and turn some promising youngster into a dangerous thug or pathetic loser. In some sports leagues, everyone gets a trophy.

A number of people have questioned whether these feel-good exercises are really the best way to build self-esteem. But what about the underlying assumption? When my colleagues and I began looking into the matter in the early 1990s, we found article after article citing the "well-known fact" that low self-esteem causes violence. Yet we were unable to find any book or paper that offered a formal statement of that theory, let alone empirical evidence to support it. Everybody knew it, but nobody had proved it.

Unfortunately for the low-self-esteem theory, researchers have gradually built up a composite image of what it is like to have low self-esteem, and that image does not mesh well with what we know about aggressive perpetrators. People who have a negative view of themselves are typically muddling through life, trying to avoid embarrassment, giving no sign of a desperate need to prove their superiority. Aggressive attack is risky; people with low self-esteem tend to avoid risks. When people with low self-esteem fail, they usually blame themselves, not others.

Faced with these incongruities, we cast about for an alternative theory. A crucial influence on our thinking was the seemingly lofty self-regard of prominent violent people. Saddam Hussein is not known as a modest, cautious, self-doubting individual. Adolf Hitler's exaltation of the "master race" was hardly a slogan of low self-esteem. These examples suggest that high self-esteem, not low, is indeed an important cause of aggression.

We eventually formulated our hypothesis in terms of threatened egotism. Not all people who think highly of themselves are prone to violence. That favorable opinion must be combined with some external threat to the opinion. Somebody must question it, dispute it, undermine it. People like to think well of themselves, and so they are loath to make downward revisions in their self-esteem. When someone suggests such a revision, many individuals--those with inflated, tenuous and unstable forms of high self-esteem--prefer to shoot the messenger.

Pride Comes Before a Fall
It would be foolish to assert that aggression always stems from threatened egotism or that threatened egotism always results in aggression. Human behavior is caused and shaped by various factors. Plenty of aggression has little or nothing to do with how people evaluate themselves. But if our hypothesis is right, inflated self-esteem increases the odds of aggression substantially. For those aggressive acts that do involve the perpetrators self-regard, we believe that threatened egotism is crucial. Obviously, this new theory could have implications for designing effective methods to reduce violence.

So how does a social psychologist establish whether low or high self-esteem leads to violence? Because there is no perfect, general method for understanding complex questions about human beings, social scientists typically operate by conducting multiple studies with different methods. A single study can be challenged, especially if competing views exist. But when a consistent pattern emerges, the conclusions become hard to ignore.

Researchers measure self-esteem by asking a standardized series of questions, such as "How well do you get along with other people?" and "Are you generally successful in your work or studies?" The individual chooses from a range of responses, and the overall score falls somewhere on the continuum from negative to positive. Strictly speaking, it is misleading to talk of "people high in self-esteem" as if they were a distinct type, but I use this phrase to refer broadly to those who score above the median on the self-esteem scale. Statistical analyses respect the full continuum.

Many laypeople have the impression that self-esteem fluctuates widely, but in fact these scores are quite stable. Day-to-day changes tend to be small, and even after a serious blow or boost, a person's self-esteem score returns to its previous level within a relatively short time. Large changes most often occur after major life transitions, such as when a high school athlete moves on to college to find the competition much tougher.

Quantifying aggression is trickier, but one approach is simply to ask people whether they are prone to angry outbursts and conflicts. These self-reported tendencies can then be compared to the self-esteem scores. Most research has found a weak or negligible correlation, although an important exception is the work done in the late 1980s by Michael H. Kernis of the University of Georgia and his colleagues. They distinguished between stable and unstable self-esteem by measuring each person on several occasions and looking for fluctuations. The greatest hostility was reported by people with high but unstable self-esteem. Individuals with high, stable self-esteem were the least hostile, and those with low self-esteem (stable or unstable) were in between.

Take a Swig, Take a Swing
Another approach is to compare large categories of people. Men on average have higher self-esteem than women and are also more aggressive. Depressed people have lower self-esteem and are less violent than nondepressed people. Psychopaths are exceptionally prone to aggressive and criminal conduct, and they have very favorable opinions of themselves.

Evidence about the self-images of specific murderers, rapists and other criminals tends to be more anecdotal than systematic, but the pattern is clear. Violent criminals often describe themselves as superior to others--as special, elite persons who deserve preferential treatment. Many murders and assaults are committed in response to blows to self-esteem such as insults, "dissing" and humiliation. (To be sure, some perpetrators live in settings where esteem and respect are linked to status in the social hierarchy, and to put someone down can have tangible and even life-threatening consequences.)

The same conclusion has emerged from studies of other categories of violent people. Street-gang members have been reported to hold favorable opinions of themselves and to turn violent when these views are disputed. Playground bullies regard themselves as superior to other children; low self-esteem is found among the victims of bullies but not among bullies themselves. Violent groups generally have overt belief systems that emphasize their superiority over others. War is most common among proud nations that feel they are not getting the respect they deserve, as Daniel Chirot discusses in his fascinating book Modern Tyrants.

Drunk people are another such category. It is well known that alcohol plays a role in either a majority or a very large minority of violent crimes; booze makes people respond to provocations more vehemently. Far less research has examined the link with self-esteem, but the findings do fit the egotism pattern: consuming alcohol tends to boost people's favorable opinions of themselves. Of course, alcohol has myriad effects, such as impairing self-control, and it is hard to know which is the biggest factor in drunken rampages.

Aggression toward the self exists, too. A form of threatened egotism seems to be a factor in many suicides. The rich, successful person who commits suicide when faced with bankruptcy, disgrace or scandal is an example. The old, glamorous self-concept is no longer tenable, and the person cannot accept the new, less appealing identity.

Vanity Unfair
Taken together, these findings suggest that the low-self-esteem theory is wrong. But none involves what social psychologists regard as the most convincing form of evidence: controlled laboratory experiments. When we conducted our initial review of the literature, we uncovered no studies that probed the link between self-esteem and aggression. Our next step, therefore, was to conduct some. Brad J. Bushman, now at the University of Michigan at Ann Arbor, took the lead.

The first challenge was to obtain reliable data on the self-concepts of participants. We used two different measures of self-esteem, so that if we failed to find anything, we could have confidence that the result was not an artifact of having a peculiar scale. Yet we were skeptical of studying self-esteem alone. The hypothesis of threatened egotism suggested that aggressive behavior would tend to occur among only a subset of people with high self-esteem. In the hope of identifying this subset, we tested for narcissism.

Narcissism is a mental illness characterized by inflated or grandiose views of self, the quest for excessive admiration, an unreasonable or exaggerated sense of entitlement, a lack of empathy, an exploitative attitude toward others, a proneness to envy or wish to be envied, frequent fantasies of greatness, and arrogance. The construct was extended beyond the realm of mental illness by Robert Raskin of the Tulsa Institute of Behavioral Sciences in Oklahoma and his colleagues, who constructed a scale for measuring narcissistic tendencies.

We included that measure alongside the self-esteem scales, because the two traits are not the same, although they are correlated. Individuals with high self-esteem need not be narcissistic. They can be good at things and recognize that fact without being conceited or regarding themselves as superior beings. The converse--high narcissism but low self-esteem--is quite rare, however.

The next problem was how to measure aggression in the laboratory. The procedure we favored involved having pairs of volunteers deliver blasts of loud noise to each other. The noise is unpleasant and people wish to avoid it, so it provides a good analogue to physical aggression.

The noise was presented as part of a competition. Each participant vied with somebody else in a test of reaction time. Whoever responded more slowly received a blast of noise, with the volume and duration of the noise set by his or her opponent. This procedure differed from that of earlier studies, in which the subject played the role of a "teacher" who administered noise or shock to a learner whenever the "learner" made a mistake. Critics had suggested that such a method would yield ambiguous results, because a teacher might deliver strong shocks or loud noise out of a sincere belief that it was an effective way to teach.

"One of the Worst"
To study the "threat" part of threatened egotism, we asked participants to write a brief essay expressing their opinion on abortion. We collected the essays and (ostensibly) redistributed them, so the two contestants could evaluate each other's work. Each participant then received his or her own essay back with the comments the other person had (supposedly) given it.

In reality, we took the essays and randomly marked them good or bad. The good evaluation included very positive ratings and the handwritten comment, "No suggestions, great essay!" The bad evaluation contained low marks and the comment, "This is one of the worst essays I have read!" After handing back the essays and evaluations, we gave out instructions for the reaction-time test and the subjects began to compete.

The results supported the threatened-egotism theory rather than the low-self-esteem theory. Aggression (blasting noise) was highest among narcissists who had received the insulting criticism. Nonnarcissists (with either high or low self-esteem) were significantly less aggressive, as were narcissists who had been praised.

In a second study, we replicated these findings and added a new twist. Some participants were told that they would be playing the reaction-time game against a new person--someone different from the person who had praised or insulted them. We were curious about displaced aggression: Would people angered by their evaluation lash out at just anybody? As it happened, no. Narcissists blasted people who had insulted them but did not attack an innocent third party. This result agrees with a large body of evidence indicating that violence against innocent bystanders is, despite conventional wisdom, quite rare.

A revealing incident illuminates the attitudes of the narcissists. When a television station did a feature on this experiment, we administered the test to new participants for the benefit of the cameras. One of them scored in the 98th percentile on narcissism and was quite aggressive. Afterward he was shown the film and given the opportunity to refuse to let it be aired. He said to put it on--he thought he looked great. Bushman took him aside and explained that he might not want to be seen by a national audience as a highly aggressive narcissist. The footage showed him using severe profanity when receiving his evaluation, then laughing while administering the highest permitted levels of aggression. The man shrugged this off with a smile and said he wanted to be on television. When Bushman proposed that the station at least digitize his face to disguise his identity, the man responded with an incredulous no. In fact, he said, he wished the program could include his name and phone number.

Would our laboratory findings correspond to the outside world? Real-life violent offenders are not the easiest group of people to study, but we gained access to two sets of violent criminals in prison and gave them the self-esteem and narcissism questionnaires. When we compared the convicts self-esteem with published norms for young adult men (mostly college students) from two dozen different studies, the prisoners were about in the middle. On narcissism, however, the violent prisoners had a higher mean score than any other published sample. It was the crucial trait that distinguished these prisoners from college students. If prison seeks to deflate young men's delusions that they are God's gift to the world, it fails.

As our findings about self-esteem and violence have become known, others have scrambled to find support for the low-self-esteem theory. What little they find comes mostly from questionnaires, which to me is suspect. People with low self-esteem are more willing than others to admit to bad actions, including aggression; an individual scores low on self-esteem precisely by saying bad things about himself or herself. Behavioral measures, however, continue to link aggression to narcissism, and scoring high on both narcissism and self-esteem predicts the greatest aggression.

What about Deep Down?
A common question in response to these findings is: "Maybe violent people seem on the surface to have a high opinion of themselves, but isnt this just an act? Might they not really have low self-esteem on the inside, even if they wont admit it?" This argument has a logical flaw. We know from ample research that people with overt low self-esteem are not aggressive. Why should low self-esteem cause aggression only when it is hidden? The only difference between hidden and overt low self-esteem is the fact of its being hidden, so the cause of violence would not be the low self-esteem but the concealment of it. What is concealing it is the veneer of egotism--which brings us back to the threatened-egotism theory.

Various researchers have tried and failed to find any sign of a soft inner core among violent people. Martin Sanchez-Jankowski of the University of California, Berkeley, who spent 10 years living with various gangs and wrote one of the most thorough studies of youth gang life, had this to say: "Some studies of gangs suggest that many gang members have tough exteriors but are insecure on the inside. This is a mistaken observation." Dan Olweus of the University of Bergen in Norway has devoted his career to studying childhood bullies, and he agrees: "In contrast to a fairly common assumption among psychologists and psychiatrists, we have found no indicators that the aggressive bullies (boys) are anxious and insecure."

The case should not be overstated. Psychology is not yet adept at measuring hidden aspects of personality, especially ones that a person may not be willing to admit even to himself or herself. But at present there is no empirical evidence or theoretical reason that aggressors have a hidden core of self-doubt.

Although this conclusion contradicts the traditional focus on low self-esteem, it does not mean that aggression follows directly from an inflated view of self. Narcissists are no more aggressive than anyone else, as long as no one insults or criticizes them. But when they receive an insult--which could be seemingly minor--the response tends to be much more aggressive than normal. Thus, the formula of threatened egotism combines something about the person with something about the situation. Whatever the details of cause and effect, this appears to be the most accurate formula for predicting violence.

These patterns raise misgivings about how schools and other groups seek to boost self-esteem with feel-good exercises. A favorable opinion of self can put a person on a hair trigger, especially when this favorable opinion is unwarranted. In my view, there is nothing wrong with helping students and others to take pride in accomplishments and good deeds. But there is plenty of reason to worry about encouraging people to think highly of themselves when they havent earned it. Praise should be tied to performance (including improvement) rather than dispensed freely as if everyone had a right to it simply for being oneself.

A person with low self-esteem is not prone to aggressive responses. Instead one should beware of people who regard themselves as superior, especially when those beliefs are inflated, weakly grounded in reality or heavily dependent on having others confirm them frequently. Conceited, self-important individuals turn nasty toward those who puncture their bubbles of self-love.