Another set of tests was what are called speed tests, much simpler measures of ability. Students were given 56 pairs of numbers and told to add them. Or they were shown a sequence of several hundred randomly arranged letters of the alphabet and asked to read through them and underline all the A's. Students were allowed a minute to complete each of these tests. Then they were given another set of equivalent tests, only this time the time limit wasn't enforced at all, allowing the students to keep on working if they wanted to. In all, the two psychologists administered countless different tests in countless different situations. They had children undertake tests of physical ability, like chin-ups or broad jumps, and secretly observed them to see whether they cheated in reporting how well they did. They gave students tests to do at home, where they had ample opportunity to use dictionaries or ask for help, and compared those results to how they did on similar tests administered at school, where cheating was impossible. In the end, their results fill three thick volumes and, along the way, challenge a lot of preconceptions of what character is.

Their first conclusion is, unsurprisingly, that lots of cheating goes on. In one case, the scores on tests where cheating was possible were 50 percent higher, on average, than the "honest" scores. When Hartshorne and May began to look for patterns in the cheating, some of their findings were equally obvious. Smart children cheat a little less than less-intelligent children. Girls cheat about as much as boys. Older children cheat more than younger children and those from stable and happy homes cheat a bit less than those from unstable and unhappy homes. If you analyze the data you can find general patterns of behavioral consistency from test to test.

But the consistency isn't nearly as high as you might expect. There isn't one tight little circle of cheaters and one tight little circle of honest students. Some kids cheat at home but not at school; some kids cheat at school but not at home. Whether or not a child cheated on, say, the word completion test was not an iron-clad predictor of whether he or she would cheat on, say, the underlining A's part of the speed test. If you gave the same group of kids the same test, under the same circumstances six months apart, Hartshorne and May found, the same kids would cheat in the same ways in both cases. But once you changed any of those variables — the material on the test, or the situation in which it was administered — the kinds of cheating would change as well.

What Hartshorne and May concluded, then, is that something like honesty isn't a fundamental trait, or what they called a "unified" trait. A trait like honesty, they concluded, is considerably influenced by the situation. "Most children," they wrote,

will deceive in certain situations and not in others. Lying, cheating, and stealing as measured by the test situations used in these studies are only very loosely related. Even cheating in the classroom is rather highly specific, for a child may cheat on an arithmetic test and not on a spelling test, etc. Whether a child will practice deceit in any given situation depends in part on his intelligence, age, home background, and the like and in part on the nature of the situation itself and his particular relation to it.

This, I realize, seems wildly counterintuitive. If I asked you to describe the personality of your best friends, you could do so easily, and you wouldn't say things like "My friend Howard is incredibly generous, but only when I ask him for things, not when his family asks him for things," or "My friend Alice is wonderfully honest when it comes to her personal life, but at work she can be very slippery." You would say, instead, that your friend Howard is generous and your friend Alice is honest. All of us, when it comes to personality, naturally think in terms of absolutes: that a person is a certain way or is not a certain way. But what Zimbardo and Hartshorne and May are suggesting is that this is a mistake, that when we think only in terms of inherent traits and forget the role of situations, we're deceiving ourselves about the real causes of human behavior.

Why do we make this mistake? It's probably the result of the way evolution has structured our brain. For instance, anthropologists who study vervets find that these kinds of monkeys are really bad at picking up the significance of things like an antelope carcass hanging in a tree (which is a sure sign that a leopard is in the vicinity) or the presence of python tracks. Vervets have been known to waltz into a thicket, ignoring a fresh trail of python tracks, and then act stunned when they actually come across the snake itself. This doesn't mean that vervets are stupid: they are very sophisticated when it comes to questions that have to do with other vervets. They can hear the call of a male vervet and recognize whether it comes from their own group or a neighboring group. If vervets hear a baby vervet's cry of distress, they will look immediately not in the direction of the baby, but at its mother — they know instantly whose baby it is. A vervet, in other words, is very good at processing certain kinds of vervetish information, but not so good at processing other kinds of information.

The same is true of humans.

Consider the following brain teaser. Suppose I give you four cards labeled with the letters A and D and the numerals 3 and 6. The rule of the game is that a card with a vowel on it always has an even number on the other side. Which of the cards would you have to turn over to prove this rule to be true? The answer is two: the A card and the three cards. The overwhelming majority of people given this test, though, don't get it right. They tend to answer just the A card, or the A and the six. It's a hard question. But now let me pose another question. Suppose four people are drinking in a bar. One is drinking Coke. One is sixteen. One is drinking beer and one is twenty-five. Given the rule that no one under twenty-one is allowed to drink beer, which of those people's IDs do we have to check to make sure the law is being observed? Now the answer is easy. In fact, I'm sure that almost everyone will get it right: the beer drinker and the sixteen-year-old. But, as the psychologist Leda Cosmides (who dreamt up this example) points out, it is exactly the same puzzle as the A, D, 3, and 6 puzzle. The difference is that it is framed in a way that makes it about people, instead of about numbers, and as human beings we are a lot more sophisticated about each other than we are about the abstract world.

The mistake we make in thinking of character as something unified and all-encompassing is very similar to a kind of blind spot in the way we process information. Psychologists call this tendency the Fundamental Attribution Error (FAE), which is a fancy way of saying that when it comes to interpreting other people's behavior, human beings invariably make the mistake of overestimating the importance of fundamental character traits and underestimating the importance of the situation and context. We will always reach for a "dispositional" explanation for events, as opposed to a contextual explanation. In one experiment, for instance, a group of people are told to watch two sets of similarly talented basketball players, the first of whom are shooting baskets in a well-lighted gym and the second of whom are shooting baskets to a badly lighted gym (and obviously missing a lot of shots). Then they are asked to judge how good the players were. The players in the well-lighted gym were considered superior. In another example, a group of people are brought in for an experiment and told they are going to play a quiz game. They are paired off and they draw lots. One person gets a card that says he or she is going to be the "Contestant." The other is told he or she is going to be the "Questioner." The Questioner is then asked to draw up a list of ten "challenging but not impossible" questions based on areas of particular interest or expertise, so someone who is into Ukrainian folk music might come up with a series of questions based on Ukrainian folk music. The questions are posed to the Contestant, and after the quiz is over, both parties are asked to estimate the level of general knowledge of the other. Invariably, the Contestants rate the Questioners as being a lot smarter than they themselves are.