But Greenwald did not stop there. He replaced the names of the flowers and insects in his list with typically Caucasian names such as Adam and Chip and typically African American names such as Alonzo and Jamal, and then tossed in a bunch of pleasant and unpleasant words. He tried to play the same association game. Since he didn’t consciously associate either Caucasian or African American names with positive or negative concepts, he assumed he would associate all the names with pleasant and unpleasant words at the same speed. He was wrong. He found it was as difficult to associate “white names” with unpleasant words as it was to associate flowers with unpleasant words. But Greenwald’s hidden brain effortlessly associated “black names” with words such as “evil” and “poison.” It was as if his hidden brain equated white names with positive concepts and black names with negative concepts. Greenwald was horrified. He didn’t think of himself as a racist, and he didn’t know what to make of his performance.
He got in touch with a colleague, the psychologist Mahzarin Banaji. Without telling her what the test was about, he asked her to play the word association game on her computer. Banaji found she had results identical to Greenwald’s. She effortlessly associated white names with positive concepts and black names with negative concepts. That’s ridiculous, she thought. She knew she was no racist. She was a professor who spent lots of time teaching other people how to watch out for prejudice. Since Greenwald’s test required Banaji to tap computer keys with her left or right index finger, Banaji figured the weird results had to do with whether someone was right-handed or left-handed, so she reorganized the test on her computer. All of a sudden, it was now her other hand that effortlessly grouped black names with negative words and white names with positive words. Banaji changed the order in which the names were presented. It made no difference. Her hidden brain simply found it easier to associate “Alonzo” and “Jamal” with “evil” and “poison,” and “Adam” and “Chip” with “dream” and “heaven.” Banaji sat back in her chair and stared. She felt small in a way she had never felt before.
This is the origin of a psychological test for unconscious bias that has revolutionized the scientific study of prejudice in the last decade. Greenwald dubbed it the Implicit Association Test, or IAT. Millions of people have taken the free Internet tests that he and Banaji have made available over the Internet—at www.implicit.harvard.edu. If you take what is known as the race bias test today, you’ll see white and black faces instead of names, since this provides more accurate results. Hundreds of thousands of people have been disconcerted by their scores. Large majorities of Americans—including substantial numbers of African Americans—find it easier to associate white faces rather than black faces with positive concepts. Overwhelming majorities find it easier to associate men’s names rather than women’s names with careers and professional activity. The tests detect things that seem absurd. Many Americans are quicker to associate the British actors Hugh Grant and Elizabeth Hurley with being American than the tennis player Michael Chang and the television personality Connie Chung. It is as though their unconscious minds associate whites—even whites who are foreigners—as American, and Americans who are people of color as foreigners. In one set of tests before the 2008 presidential election, psychologists found that many voters unconsciously associated former British prime minister Tony Blair with being more American than Barack Obama. If you’d asked volunteers which one, Blair or Obama, was American, of course, they would have looked at you funny. If people knew—at a conscious level—that Obama was American and Blair was foreign, why did their unconscious minds have the opposite associations? With every variation of the test, volunteers responded exactly as Greenwald and Banaji had to that first test: with disbelief.
If the only thing the Implicit Association Test did was to make people feel bad about the unpleasant associations in their heads, the test would not be very useful. Ultimately, we’re interested in people’s behavior, not in “thought crimes.” But over the past decade, many experiments have shown that results on the Implicit Association Test predict people’s behavior in real-world settings. In tests conducted before the 2008 presidential election, for example, the speed at which people unconsciously associated Obama with being American predicted whether they supported the biracial candidate in both the Democratic primary and the general election. People who unconsciously saw Obama as less American than Hillary Clinton and John McCain were less likely to vote for him. This was so even if—when explicitly asked—they stated Obama was every bit as American as his competitors.
It is useful to situate all conversations about race in their proper context: Voting against Obama did not automatically make anyone a racist. And race bias was just one factor in how people thought about politics. Race bias in the 2008 presidential election may have pulled the country a few percentage points in one direction. The bias would not have made a difference to the way most people voted, but it could have tipped people on the edge one way rather than the other. The bias affected both Republicans and Democrats. At an unconscious level, not seeing Obama as American exerted a subtle tug on people. It made them less likely to want to see him elected president. Remember, we are not talking about people who consciously thought of Obama as a foreigner with a doctored birth certificate. We’re talking about people who never would’ve said aloud—or even to themselves—that they disliked Obama because he was foreign. But when they thought about the candidates, Obama may have just felt a little different. Hillary Clinton, John McCain, and Sarah Palin may have felt a little more “like one of us.” Once people felt that way, they could easily have come up with reasons to support their intuitions. You can always find things about a candidate to like or dislike.
How do we know that unconscious attitudes about Obama preceded the conscious justification of those attitudes? In the experiment, it was people’s unconscious attitudes—as revealed by Greenwald’s test—rather than their conscious views about the issues, that better predicted whether they voted for Obama or preferred another candidate. The idea that we provide ourselves with explanations to justify our conclusions is counterintuitive because it certainly feels as though our conclusions are the product of careful thought. Here is an analogy that might help—this is not an original idea of mine. Let’s say you kick a soccer ball into the air. Imagine that as it flies, the ball suddenly acquires consciousness. How would it explain to itself why it is flying? It has no knowledge or awareness of having been kicked. But since it knows that all effects have causes, it tells itself that it is a ball that decided to fly, because that’s the most plausible explanation. In the same way, once the hidden brain whispered to these voters that Obama was different, they quickly came up with plausible ways to explain to themselves why they didn’t like the candidate—his views on health care, perhaps, or the economy.
These results are upsetting and embarrassing. We know we are decades away from Selma and Birmingham, from those bad old days when women were not allowed to vote. We’ve changed, haven’t we? Banaji once told me that the embarrassment she and others felt after doing the Implicit Association Test was a good thing. It showed that people not only believed they were not biased, but that they did not want to be biased.
Banaji, Greenwald, and another psychologist named Brian Nosek at the University of Virginia have studied the results of hundreds of thousands of volunteers who have taken the Implicit Association Test on the Internet. The psychologists have created a map of America that shows the peaks and valleys of bias, the places where unconscious racial prejudice is highest and where it is lowest. Nosek has found, for example, that people in the second congressional district of New Jersey, which is nestled between the Delaware Bay and the Atlantic Ocean, have higher average levels of race bias than those in the twenty-seventh congressional district in the Deep South of Texas, which includes the towns of Corpus Christi and Brownsville. Volunteers in Alabama’s first congressional district, centered around Mobile, show higher unconscious racial bias than those in the ninth congressi
onal district of California, which includes Oakland and Berkeley.
Nosek overlaid his map of unconscious anti-black race bias on a map of electoral outcomes in all congressional races. He found a remarkable association between bias scores and political views—the higher the unconscious race bias scores in a congressional district, the more likely the district was to elect a Republican. (The psychological research into the relationship between racial bias and political outcomes in the United States has focused mainly on bias against blacks. Ongoing research is exploring the effects of bias against other minorities.)
If racial bias had nothing to do with politics, and if the implicit association test was meaningless, as some of its critics have argued, there ought to be no correlation between bias scores and political orientation. But Nosek saw a very clear pattern. On average, districts with the greatest racial bias were more likely to vote for conservatives, and districts with the lowest racial bias tended to vote liberal. The difference, it should be emphasized, was a matter of degree. Large numbers of people in the Bay Area revealed the same anti-black or pro-white attitudes as people in Mobile, Alabama. But on a sliding scale, some areas showed greater levels of bias than others.
This doesn’t mean that you can look at a district’s race bias scores and automatically predict whether the district will elect a Republican or a Democrat. As I said, bias is only one factor in people’s political views—and it is far from being the most important factor. Some of the districts with the lowest bias scores—such as the second congressional district of Idaho, which includes the towns of Idaho Falls and Pocatello—are strongly Republican, while some districts with high bias scores—such as the first congressional district of North Carolina—elected a Democrat in 2006 and 2008.
There are places where race bias appears to be irrelevant to election outcomes, and places where it seems to play a substantial role. But of the ten congressional districts with the highest bias scores among white voters, eight were won by Republicans in the 2008 elections, and only two by Democrats. The ten districts with the lowest bias scores revealed the opposite pattern—seven were won by Democrats and only three by Republicans.
Race bias scores predicted not only whether liberals or conservatives got elected, but what kind of liberals and conservatives got elected. Fully half the districts with the lowest bias scores elected people of color. Only one district with the highest bias scores elected a person of color (and he was a Democrat). Five of the seven Democrats elected in the districts with the lowest bias scores were Asian, Hispanic, and African American. All the Republicans elected in the districts with the highest bias scores were white.
In two of the districts with the highest bias scores—the first congressional district of Alabama and the ninth congressional district of North Carolina—control of the seat switched from one political party to the other in the middle of the civil rights movement. This was part of the dramatic shift in the American South away from the Democratic party in the 1960s, owing to the Democratic party’s support of civil rights legislation that guaranteed voting rights and dismantled legalized segregation. The first congressional district of Alabama elected Democrats from 1877 to 1963—eighty-six years—but elected a Republican in every subsequent election through 2008. The ninth congressional district of North Carolina had elected a Democrat for twenty-two straight years before 1963; starting in 1964, the district elected Republicans for more than four straight decades.
Since we are talking about science, we ought to examine the research with a skeptical eye. Is the link between racial bias scores and political orientation merely a correlation, or is there a causative connection? If you see a lot of people holding umbrellas and wearing rain boots, it would be wrong to say that wearing rain boots causes people to hold umbrellas. The two things are related, but one does not cause the other. Both are caused by something else, the fact that it is raining. Could an unrelated third factor be responsible for both racial bias and political orientation? Second, isn’t it possible that the connection between race and politics is better explained by simple demographics? People of color tend to vote for Democrats, so there ought to be a connection between diversity and whether a district elects a Republican or a Democrat. What does racial bias tell us that the racial makeup of a district does not?
The question about correlation and causation cannot be answered with certainty. It is possible that the connection between race bias and political conservatism is only a correlation. The only way to prove causation is to conduct the kind of experiment that is impossible in real life: You change some people’s unconscious racial attitudes and see if their political orientation fluctuates in response. If it does, you know that racial attitudes are influencing how people think about politics. Nosek believes the relationship flows in both directions—race bias contributes to conservatism, and vice versa—and that both might also be influenced by other factors. People who are sensitive to threat, for example, tend to adopt conservative views and are also suspicious of people from other groups.
The second question—the interaction between racial bias and racial makeup—has a clear answer. The demographic makeup of a district seems to act like a switch that sometimes brings racial bias into play, and sometimes eliminates it from the equation. Anti-black race bias does not seem to play much of a role in districts that have very few black people. It is only when a district starts to show an element of diversity that unconscious race bias seems to influence voting decisions. In other words, everyday contact between whites and blacks appears necessary to make bias salient—to bring into play racial attitudes that lurk beneath the surface. Remember the study by Jennifer Eberhardt into sentencing disparities? The sentencing disparities showed up only in cases involving black-on-white crime, not in cases involving same-race crime.
In the same way, the presence of blacks in a congressional district seems to make race bias relevant to the voting decisions of whites. But when there are a very large number of blacks in a district, the connection between race bias and political orientation among white voters is drowned out by the fact that blacks tend to vote for Democrats, and so districts that have a lot of blacks invariably end up electing Democrats. White voters in congressional districts where racial minorities make up more than, say, 40 percent of the electorate certainly tend to vote Republican—the presence of minorities makes race relevant to the voting decisions of whites—but the tidal wave of people of color voting in the other direction renders that bias irrelevant.
So, again, when there are few or no minorities, race bias doesn’t seem to have much of an effect on political outcomes—the “switch” is turned off. And when there are lots of minorities, the effect of racial bias is again irrelevant because the tendency of whites to vote Republican in these areas is canceled out by the tendency of blacks and other minorities to vote for Democrats. Race bias seems to tip congressional seats only in districts where you have both high bias scores among whites and a minority population that is sizable enough to be visible in everyday settings but not so large as to control the congressional district’s electoral destiny. Nosek estimated that race bias accounted for no more than 10 percent of the political variation in the country overall—hardly decisive, except in close races. Even this relatively small effect, however, produces clear patterns nationally between racial attitudes and voting outcomes.
Many conservatives plaintively ask why psychologists don’t spend more time analyzing the voting behavior of people of color. Nearly all African American voters, for example, voted for Barack Obama in the 2008 presidential election. People of color, in general, vote overwhelmingly for Democrats. Isn’t this a bias, too? It certainly is. I don’t think it is an unconscious bias, however. Many people of color enthusiastically supported Obama because they wanted to elect the first nonwhite president. This is emphatically not the case with unconscious race bias and conservatism—white voters tell us that race has nothing to do with their political views. If most research focuses on the biases of white vote
rs, it is also because these biases matter more. There are more white voters in the United States than voters who are people of any other race, and a bias that affects whites is likely to be consequential in a way that a bias among a minority group is not. History also tells us what to focus on. We have never had a female president, so it would be silly to spend time studying why some voters may be biased in favor of women instead of asking why voters are biased in favor of men.
Making connections between racial bias and politics always gets people upset. But if you look at this data calmly, you can say two things. One, the race bias data does not provide liberals with a cudgel with which to bash conservatives. Yes, it is true that, on average, districts with higher bias scores tend to vote Republican and districts with lower bias scores tend to vote for Democrats, but the fact is that race bias is surprisingly common across political orientations. A large majority of Americans, including substantial numbers of African Americans, hold negative associations with black faces and positive associations with white faces.
But it is undeniably true that there is a steady association between higher racial bias scores and a conservative orientation, on this and other psychological tests. We do not know whether race bias makes people politically conservative, whether conservatism tends to prompt people to adopt racially biased attitudes, or whether some third factor causes both. But if you are a patriotic Republican who passionately believes in the American ideal that all people are created equal, these results ought to be disconcerting.
In the 1980s, the Democratic pollster Stan Greenberg identified a group of voters in Macomb County, Michigan. They were blue-collar workers—often union members—who had been staunch Democrats for decades. They voted their pocketbooks, and it was the Democratic party that defended their economic interests. But starting in the mid-1960s, a substantial number of these voters switched parties. Greenberg found that in the 1980s, these voters supported Ronald Reagan, and Greenberg dubbed them Reagan Democrats, a term that has endured. Every four years, the national and international media descend on Macomb County ahead of presidential elections to see what the Reagan Democrats are up to. A few years ago, the plaintive book What’s the Matter with Kansas? asked why so many blue-collar folks were voting Republican, when their economic interests lay with the Democratic party. Author Thomas Frank concluded, in large part, that these voters were influenced by hot-button cultural issues such as abortion, gay rights, and guns.
The Hidden Brain: How Our Unconscious Minds Elect Presidents, Control Markets, Wage Wars, and Save Our Lives Page 22