Implicit Personality Theory 245g3w

Chapter 5: Stereotype Structure and Implicit Personality Theory In the last c hapter w e discuss ed the wa ys stereotypes help us pro cess inform ation ab out oth ers and som e of the biases that can result from that. In the present ch apter we will somew hat gingerly try to look insid e stereotypes, to ex amine th eir structure a nd ho w features are related to ca tegories. There a re many ways we might approach this, but one way is to use a time-honored idea from person perception: implicit personality theory.

The Nature of Implicit Personality Theory Historically the implicit personality construct was the earliest knowledge structure to be generally used by social psychologists, the term having been coined by Bruner and Taguiri (1954). The term was first used (and is still largely used) to refer to the network of perceived relationships among traits and other person information such as roles, behaviors, and values. There is abundant evidence (Rosenberg & Sedlak, 1972; Schneider, 1973) that people readily infer that certain traits "go together". For example, the average person probably thinks th at kind people are warm , that stupidity implies ignorance, that hostile people are unhappy, that short people are insecure, that financially successful people are happy. Thu s, there is nothing necessa rily elaborate about implicit personality theories. They are sim ply perceiv ed relatio nship s amo ng featu res with in a giv en do main .

Implicit Personality Theories and Stereotypes At first glance it may seem that networks of trait relationships have little to do with stereotypes, but by broadening the conception of implicit personality theories slightly we will see that stereotypes can actually be seen quite easily in this fram ework. Several auth ors (e.g., Ashmore, 1981 ; Ashmore & del Boca, 197 9; Jones, 1977; Sch neider, Hastorf, & Ellswo rth, 1979) hav e argued that stereotypes a re similar to implicit pers onality theories , but th e way s they a re similar h ave no t alwa ys been spelled o ut care fully. There are three basic ways we might construe stereotypes within this framework. Keep in mind that implicit personality theories are relationships among stimuli. These stimuli are usually construed as traits, but we could also broaden stimuli to include groups. Keep in mind that there is no principled way in which categories and features differ (See Chapter 3) ! Therefore, the first, and most obvious way to use this metaphor is to consider race, gender, and other group designations as stimuli that are correlated with other features.. So just as people may see profes sors a s likely to b e smar t, they m ay also see sma rt people a s likely to b e profess orial. ! Secon d, grou p design ation m ay be th ough t of as a mod erator v ariables that ch ange tr ait relationships. In other wo rds trait relationships are different for different groups. For examp le, perceivers migh t think tha t clannish Jews are m ercenary, w hereas for H ispanics th e more clannish they are the less mercenary. ! Third, w e migh t think o f group s them selves a s stimu li which are mo re or less rela ted. Su rely most us have conceptions that some groups are more similar than others. For example, most Americans probably think that Japanese and Chinese Americans are closer than are Hispanics and Ru ssians. We w ill, in due course, consider each o f those uses of the implicit persona lity theory stereoty pe, but b efore we turn to that w e shou ld review the actu al statu s of imp licit person ality theo ry researc h, theo ry, and metho ds.

Origins of the Theories Implicit p ersona lity theor ies are perc eived rela tionsh ips am ong fea tures, a nd a b asic qu estion is where people come up with such relationships. Why, for example, should someone, anyone, assume that warm people are kind, that rich people are happy, that ugly people likely to be homosexual? One obvious answer is that at least some relationships are accurately perceived. Or maybe these perceptions of relationships are based more on cultural lessons and cognitive processes than on experiences.

Chapter 5: Structure and IPT / Version of November 10, 2002

Realism Often perceiv ed relationsh ips fairly accu rately matc h actua l relationships , and so me data that idea (Borkenau & Ostendorf, 1987; Chan & Jackson, 1979; Lay & Jackson, 1969). Perceivers may assume that kind people are warm because kind people actually are warm. There can be little doubt that in broad outline what we might call the realist position has some validity. While the social world is not without its oddities, it would be a strange world indeed if people's assumptions about the interrelationships of most characteristics are strongly at variance with reality.

Individual Mechanisms: The question is how people might obtain, represent, and retrieve correlation information. One possibility is that they keep more or less accurate running tallies on such covariations. As I experience the social world in all its glory, I simply cou nt the num ber of times warmth appears with kindness, warmth with unkindness, and so forth. However, it would seem a perverse if we had to advance a counter for the warmth-kindness relationship every time we encountered a warm and kind person1. At the least, such a model implies modes of processing information that lack, shall we say, epic efficienc y. Since p eople ca n literally rep ort tho usan ds of su ch relatio nship s whe n called upon , it wou ld require mass ive am oun ts of sto rage for a ll the requ ired infor matio n. Another poss ibility is that such relationships are generated on the fly. When s ubjects are asked about the relationship between warmth and kindness, they simply think about their experiences, imagine kind people they have known, and then estimate how many of them were also warm.

Cultural Wisdom: It is also possible that individual experiences for formations of accuracy is the wrong place to look. It might be argued that our entire culture manages to emphasize relationships among certain important features and traits that more or less match reality. So in our society people may think that warm people are kind, because it is true. However, as an individual perceiver I believe in the kindness-warmth relationships and am therefore accurate not because of my own experiences but because I have learned my cultural lessons. Most social psychologists feel uncomfortable with arguments of this sort, in part because such argum ents place so much of the emph asis on c ultural rath er than ind ividual pro cesses an d in part becau se they a re virtua lly impos sible to co nfirm em pirically (let a lone stu dy exper imenta lly). Partic ularly in the area of stereotyping where it has been historically important to assume that cultural beliefs are the culprits in our incorrect judgments about others, it is easy to assume that culture is the enemy of accuracy rather than one of its mainstays. Certainly we could parade a long, long list of obvious beliefs that survived for many generations despite having been wrong (e.g., women are generally too stupid to be educated in colleges). Yet it does not follow that cultures are inevitably wrong about such matters or even that they were wrong at specific points in time in specific cultures. If we can imagine for a moment the situation of women in 1900, I suspect it would have been much harder than now to find a great number of women who were both willing and able to succeed in college. Obviously we now recognize that this was because women were denied opportunities for intellectual growth.2 However, I am not suggesting that perceived relationships are correct for the right, or deep, reasons but only that we believe they are correct. Cultural beliefs may be su perficial, but they can still be true.

1

And m atters are mo re complex th an tha t. As we w ill see, in order to prov ide an ac curate estim ate of a correlation between a group and a trait, one would also need to keep track of times the trait went with an alternate group as w ell as times when som e contrary trait appeared w ith both groups . Thus one h as to keep fou r coun ters goin g for eac h grou p-trait relat ionsh ip, or on e mus t if one exp ects to b e genera lly accu rate. 2

Obviously people in that day and age were not looking for intellectual women. Quite apart from those who had learned to submerge their intellectual talents, there are bound to have been those who were gifted but explicitly denied opportunities for higher education by parental pressure, lack of funds, and issions policies at many colleges and universities. Chapter 5: Structure and IPT / Version of November 10, 2002

Idealism The opposite kind o f position is often called the idealist position. Idealist in this sense h as its basic philosophical meaning, namely that truth is represented in ideas (not ideals, as the word might lead one to believe). In this case that would mean that perceived relationships are partially or mostly a function of people's beliefs rather than anything based on individual experience. Our perceptions of how well traits are related may reflect our beliefs, based on values or other beliefs, about how much things should be related (liberals should be kind and conservatives egocentric or liberals should be illogical and conservatives an alytical), or even the linguistic nature of the traits (hap py and intelligent are both positive so they shou ld be related). Before discussing these va rious sources o f perceptions we need to cons ider som e problem s with the pos ition th at perceiv ed relatio nship s are ba sed on some reality.

Are People Accurate in Correlation Estimates? There are, in fact, some ma jor problems with the realist position. In the first place, even if we concede for the moment the basic accuracy of most assessments of trait relationships, it is highly likely that perceived relationships do not match the actual magnitude of relationships exactly. If perceptions are fairly inaccurate, obviously this would undercut the realist position. For example, it might well be that 60% of warm people may also be kind, but perceivers may assume that all or nearly all (say 90%) of warm people are kind. It is exactly this sort of exaggeration that leads us to distrust stereotypes. There is a large body of empirical work on whether people can accurately judge covariations, or the extent to which v ariables "go together". Typically subjects are given information ab out wheth er of on e of two classes do or do not pos sess a given feature. So they might be given several pieces of information of the sort: "John is a safe driver", "Melissa is an unsafe driver". After the presentation of several such pieces of information subjects would judge whether males or females are the better drivers. Thus, they are being asked to judge whether gender and driving covary or whether there is a correlation between the two variables. We can express such information in of a contingency table, in this case one exp ressing the relatio nship between driving ability an d gend er. An exa mple in g iven in T able 5.1 Table 5.1: N umbe rs of males a nd fema les observed to be good an d poor drivers Good Driving

Poor Driving

Males

228

68

Females

163

36

This ty pe of resea rch is im portan t becau se in ever yday life, we often get inform ation in just this way -- several pieces of information about individu al of grou ps arrive seriatim. Can w e turn such data into reasonably valid estimates of how groups differ? If one begins with the example of good and poor driving by males and females, an observer would first have to code driving behavior as good or poor, and, of course, errors may be made at this point. She might not have any problems determining who were th e male a nd fem ale drive rs, but fo r other g roups classifica tion m ight be a s problem atic as trait classific ation. T hen th e person wou ld hav e to con struct a menta l contin gency table like th e one giv en in table 5.1, and of course, there is also plenty of room for error as our observer might forget many of the behaviors she has observed. When she has to make a judgment, she must either retrieve the contingency table or mentally construct one on the spot from retrieved instances. Then assuming that all has gone well, she wou ld need to actu ally attem pt to ca lculate w hether m en or w omen are the b etter driver s. If we use the exam ple in table 5.1, it is not perfectly easy to determine whether m en or wom en are the b etter driver s just by quick in spection of the ta ble, In fac t wom en are slig htly bette r in this examp le, becau se a sligh tly high er percent age of fem ale tota l driving behav iors (82 %) are go od th an m ale


behavio rs (77%). 3 Thus, to generate the proper outcome, one needs to focus on all four of the cells, but people often do not do this. So if I asked our observer who is the better driver, she might focus only on the good driver column and reasoning that there are more men than women good drivers conclude that men are the better drivers. On the other hand, if I asked her whether men or women were the worse drivers she might focu s on the bad driving column and con clude that men were the worse drivers. Arkes and R othba rt (1985 ) and o thers h ave sh own that peo ple's estim ates of re lations hips m ay be se verely biased when there are d rama tic differen ces in sa mple siz es suc h as in the exam ple given . There is abundant evidence that most people are usually not terribly accurate at tracking empirical cov ariations a nd relation ships (Alloy & Taba chnik, 19 84; Croc ker, 1981). W hen the d ata are presented in a straightforward way and people do not have strong expectations about what should go with what, they are often reasonably accurate. However, real life is usually not cooperative in presenting data so neatly pac kaged to unbias ed saints. It is n ot unco mmo n to find th at people's judg ments a re sensitive both to their expectations as well as to empirical covariation (Alley & Tabachnik, 1984). For example, Berman and Kenny (1976) presented traits that were paired together various numbers of times. Subjects' estimates of trait co-occurrences did reflect actual co-occurrences, but subjects also thought that traits pairs such as persistent-indu strious, which a re semantically related, had been paired more times than they had. In the real-world, it seems likely that people would be even more inaccurate than in such a labo ratory s ituatio n in w hich s timuli a re clear an d presen ted in a fa irly rapid fa shion so forg etting is not likely to be a m ajor issu e.

for the Idealist Position: Much of the for the idealist position is indirect and stems fr om th e ready d emon stration s that p eople are o ften no t accu rate in a ssessin g how inform ation is related in everyday life. Several studies provide more direct for the position that assessments of covariation an d relationship are based m ore on the failures of our co gnitive systems than on accura te mappings of real covariations. Anthropologists Roy D'Andrade and Richard Shweder (D'Andrade, 1965; 1974; Shw eder, 1977, 1982 ) have most forcefu lly argued the proportion th at implicit personality theories are artifacts of the interplay between ou r cognitive systems an d cultural factors. Th e system atic distortion h ypothe sis suggests that when people try to assess the relationships among traits from memory, their assessments will reflect linguistic similarities and cultural notions about what traits ought to go with what traits. The most likely culprit in all this is language. It has long been known (Schneider, 1973) that perceptions of trait relationships are often similar to the similarity of trait meanings. As long ago as the 1920s psycho logists (Tho rndike, 192 0) had d iscovered th at people wh o are liked are ra ted more positively on a number of traits More generally what has come to be known as the halo effect refers to the tendency for positive traits to be seen as more closely related than they should (see Nisbett and Wilson, 1977, for a more recent demonstration). Furthermore dimensions of perceived trait relationships often show a marked evaluative nature (Rosenberg, Nelson, & Vivekananthan, 1968). Thus, it might be argued that w hen people report that intelligent people are happy, w hat they are really saying is tha t both traits are positive and that positive traits should go together. Judgments of trait relatedness may, then, be noth ing mo re than judgm ents ab out h ow lin guistic ally sim ilar two trait term s are. Shweder an d D’And rade (1980) reported a s tudy wh ich seems to ma ke that case. As they watched a 30 minute video tape of the interactions among four of a family, observers rated the behaviors of the people on 11 categories, and from these ratings a behavioral matrix was calculated of relationships among all categories. This "actual behavior matrix" was taken as the best estimate of the actua l relations hips am ong th e catego ries. Ano ther gro up of s ubjects watc hed th e same tape an d only

3

It may n ot be obv ious to ev eryone tha t these differences represent a co rrelation, but th ey do. Th ere are several ways of turning such a contingency table into a correlation, but the most common is the Phi coefficient. However calculated, however, there will be no correlation unless the percentage of good (or bad) drivers differs for the two groups. To say that two groups differ on a trait is simply another way of sayin g that th e trait an d grou p design ation a re correla ted. Chapter 5: Structure and IPT / Version of November 10, 2002

after the tape was over rated each of the people on the 11 categories. The matrix of relationships was based on memory and was called the "rated behavior matrix". Finally another group of subjects rated the similarities of meaning am ong the 11 ca tegories, and this matrix w as called the "conceptu al similarity matrix". Now if the realist position w ere correct, there should be high co rrespondence betw een the rated behavior matrix and the actual behavior matrix, but in fact these matrices were not highly related. However, the relationships between the conceptual similarity matrix (based on linguistic similarity) and the rated behavior matrix (the one done from memory) was quite high. In other words, subjects' judgments of the similarities among categories more closely matched linguistic similarities than the best estimate of actual similarities. Other studies (e.g., Gara & Rosenberg, 1981) have also shown that perceptions of trait relationships are influenced b y the linguistic meaning of the trait words. These resu lts have been ap tly criticized on a v ariety of c oncep tual an d meth odolo gical gro und s (Block , Weiss, & Thorne, 1979; Borkenau, 1986; Borkenau & Ostendorf, 1987; Romer & Revelle, 1984), so such results must not be taken as d efinitive evidence.

A Resolution: Social scientists might have learned by this point that framing issues in of the absolute correctness or falsity of opposing positions almost always turns out to generate controversy that u ltimately sheds more h eat tha n light. T his area is a perfect examp le. Man y of the p articipa nts in th is debate ha ve been co ncerned w ith implication s for forma l personality a ssessmen t, and the s takes there a re high. If our judgments about our own and others' personality are largely products of our cognitive systems, then formal personality assessment which relies on such observations may be built on weak foun dation s indeed . Our questions are somewhat less cosmic. For our purposes it seems highly likely that most judgments about relationships among traits and other characteristics are mixtures of reality (or at least experience) and various linguistic and other culturally driven cognitive processes (Borkenau, 1986). One can find abu ndant da ta in of either position, an d comm on experience suggests th at sometimes we are accura te in such ma tters, other times not, and in either case for an y numb er of reasons. Thu s here, as elsewhere, perceptions of others turn out to be complex mixtures of data and our pre-existing cognitive structures based on cultural wisdom and individual experiences.

Representation Most subject s in experim ents are perfectly a dept at a nsw ering qu estions of the so rt: "How likely is it that happy people are kind?" Indeed given enough time and patience, such subjects can report on how closely th ey see tho usan ds of tra it pairs as related. B ut ho w is this possib le? It is un likely tha t people have abundant, readily retrievable, individual trait relationships stored in memory, given the magnitude of the storage involved. For example, the interrelations of 100 traits would require nearly 5000 pieces of information. It seems h ighly improbable that o ur mental system s find enough use for such inform ation to allocate that much storage space to it. Most experts have assumed that people derive such relationships as needed from more abstract struc tures. For example, "happy" a nd "warm " may seem to b e related because they are similar on more basic dimensions; in this case they are both evaluatively positive. So rather than assume that people actually have the happy-warm relationships stored and ready to use, we usually assume that people can calculate that relationship from more basic information, say about how similar in meaning th e traits are. Much of the research on implicit personality theories deals w ith representation and interpretation of these structures. Two general types of structural representation have been u sed: dim ension al and clusterin g. The lo gic of bo th is qu ite straigh tforwa rd.

Dimensions

The Logic: The logic of dimensional representation rests on several reasonable assumptions. First, we assume th at we can represent trait relationships a s normal Eu clidian distances; closely related traits lie closer to one another than less closely related traits. Second, in any Eu clidian space, distances


between any two points can always be represented in of orthogonal underlying dimensions.4 Our old friend, the Pythagorean Theorem, tells us that the distance (represented as a hypotenuse) is the square root of the sum of the squares of the other two sides (which are the underlying dimensions). We will represent the distance between the two points as the hypotenuse and the other two sides as projections on dimensions. Thus, in some sense each distance can be decomposed into positions on two underlying dimensions.5 Obviously if we only have a single distance to worry about, it would be inefficient t o use tw o pieces o f inform ation (i.e., th e dimen siona l distan ces) to rep resent th is single similarity. However, if we add a few traits, the economies become more salient. For example, if we have 20 traits we would have 190 individual distances to worry about but only 40 pieces of dimensional information (if the distances can be plotted uniquely in two dimensions). Thus the dimensional representation becomes an attrac tive way to conceptu alize the wa ys people sto re informatio n more abstra ctly.

Similarities as Distances: Dimensional structures are most often sought through multidimension al scaling. Multidimens ional scaling begins w ith a matrix of distances a mong featu res and asks how many dimensions are needed to represent these distances within desired limits of accu racy. So two tra its suc h as w arm a nd kin d that intuitiv ely are clo se wo uld be re presente d as a small distance whereas two traits such as warm and hostile would have a larger distance between them. Such distances can be generated in variou s ways. The m ost direct is to ask subjects ho w similar the two traits are. Or in the case of traits subjects m ight be asked ho w likely it would be that a person who possessed one trait would also possess a second with high likelihood ratings being represented as small distances. The distances could also be generated indirectly. For example, the correlations between trait ratings across a number of stimulus persons might be obtained. An even more common procedure is to ask subjects to sort the traits into piles that go together according to some criterion. From such sorts one can cons true dis tances based on the num ber of tim es acro ss sub jects each trait app ears in th e same pile.

Interpretations of Dimensions: From a matrix of similarities multidimensional scaling (hereafter MDS) programs estimate the projections of each stimulus on 1 through n dimensions. Once we have established the basic dimensions, we usually want to determine wha t these dimensions "mean." In some cases the dimensions produced by a MDS will be immediately interpretable. For example, if the positive and negative traits lie at opposite ends of some dimension, it is a fair guess that the dimension represents is evaluative. But rarely are the dim ensions so clear. For this reaso n generally researchers get independent ratings of each of the stimuli on external scales and interpret dimensions on the basis of which ratings are most highly correlated with the dimensions.6 This approach can be illustrated by the w ork of Ashm ore (1981; Ash more & Tu mia, 1980; D el Boca & A shmore, 198 0) and asso ciates. Matrices of perceived trait relationships were su bmitted to both multidimensional scaling and clustering. As we mentioned earlier in the chapter, external dimensions can be fitted to the multidimensional space. Several external scales were found to fit a two dimensional space rather well (Ashm ore & Tum ia,1980). Several social desirability scales w ere virtually orthogona l to

4

Orthogon al dimensions a re the familiar dimensions th at lie at right angles to one ano ther. Distances can also be calculated from non-orthogona l dimensions, but the ciphering is more complex. 5

The ma thematics of the extensio n to three or m ore dimen sions is qu ite straightforw ard, so w e are not limited to having only two dimensions. However, the practicalities of visual representation on a twodimensional page make two-dimensional discussions appealing. 6

More formally one can use these ratings as a dependent measure in a multiple regression with the dimensions b ecoming the ind ependent variables u sing the projections on th e various scales as the data points. The resultant R 2 provides a measure of how well this scale explains the projections. One can also plot this scale as a vector through the space with the vectors now substituting for the original dimensions. Chapter 5: Structure and IPT / Version of November 10, 2002

scales representing potency (ha rd-soft) and ma le-femaleness. These latter two scales w ere close together (about 1 7 degrees ap art), so that tra its that were ra ted as "ha rd" were also seen as m asculine w hereas so ft traits were also seen as feminine. Since these scales were basically unrelated to social desirability, subject s seeme d to feel th at neith er mas culinity or femin inity wa s better in a socia l sense. H owev er, a third set of scales representing intellectual desirability fell between the social desirability and m asculinity vector s so th at the m ascu linity of tra its is corre lated w ith their pe rceived in tellectua l good ness. Dimensional representation of similarities among stimuli is mathematically tractable and seems perfectly natural. Nonetheless, it is important to remind ourselves that we need to make assumptions about what these dimensions mean. Although nothing in the MDS technique requires that subjects have generated trait distances by comparison along the component dimensions, we do usually assume that the underlying dimensions have meaning for the subjects and that they have played at least some cognitive role in generating the distance estimates. We also implicitly assume that dimensional representation is a natural way for people to think about their worlds. Are there alternatives?

Clusters We have argued that it is implausible that most individual trait relationships are stored as such and that a dimensiona l structure is one way people might have o f making similarity judgm ents when asked for them . But th ere are oth er way s of repres enting s tructu re. One of the m ost plau sible is hierarchical clusters. Suppose that people have in mind that certain traits simply go together as large clusters . So wa rm, kind , generou s, and empat hetic m ight be s een as a kind o f altruistic cluster w hile unstylish , smart, wea rs glasses, an d socially u nskilled migh t be seen as g oing togeth er as a kind of nerd cluster. Furthermore clusters might be ed together to former larger scale units. The warm-kindgenerous-empa thetic cluster (which I ha ve, for convenience, termed a ltruistic) might be related to anot her clus ter com posed of hap py-ou tgoing -hum orou s (say p opula r). The larg er cluster comp osed o f all 7 traits m ight be ter med s ocial o rientatio n or so cially skilled . While p eople ca n thin k dimen siona lly, cluster s may be a mo re natu ral wa y to repres ent trait relationships (Anderso n & Sedikides, 199 1; Sedikides & And erson, 1994). For on e things, clusters of traits map nicely onto both exemplar and prototype representations of categories. More generally, when we think a bout types of p eople w e usu ally ima gine tha t they h ave sev eral traits t hat so meho w go t ogethe r in an integral way. Sedikides and Anderson (1994) have shown that people tend to assume causal relationships amo ng traits within a clus ter, at least to a larger extent than between clos ely related traits that a re not w ithin th e same cluster. T hat w ould s ugges t an int egrity to c lusters th at is no t captu red well by dim ension al ana lysis. In a clustering an alysis of gender trait data (reported in A shmore, 198 1) gender was u sed to discriminate among trait clusters. In this study gender was most discriminating when there were 30 clusters whereas so cial desirability was more impo rtant when th ere were only 3 clusters. This su ggests that people have a highly complex view of gender. For the 30 clusters some were clearly male and some clearly female, but there were several clusters that had both masculine and feminine traits. What was even more interesting was that male and female perceivers differed considerably in the clusters they had. Males seemed to divide female traits into two general categories, soft and ive on the one hand and socially outgoing on the other. The clusters for female perceivers were labeled weak, cautious, nurturant and th ough tful-quie t. In an other c lusterin g stud y male p erceivers h ad at lea st four c lusters o f female traits and Ashmore has labeled these: "girlfriend", "neurotic", "nurturant" and "outgoing". The corresp ondin g clust ers for fem ales we re labeled : "nervo us nellie" and "u pper clas s you ng w oma n". A couple of points are worth making about this. First, the fact that at the same level of clustering (e.g., 30 clusters) different types of subjects organized the traits into clusters differently, suggests that such techniques may be powerful means of uncovering individual differences in the structure of stereotypes. Second, this research alerts us to the important point that there may be multiple stereotype groups nested within more global ones. If indeed men do see women as divided into girlfriends and nurturant (perha ps motherly) categories, it wou ld be reasonable tha t they may see ad ditional subtypes within each of these. At this point the work in this chapter based on implicit personality theory s


forces with that discu ssed in Cha pter 3 on the basis of ca tegorization and the extent to which ca tegories have importa nt sub types. If trait information is represented in of such clusters, similarity judgments could easily be generated b y people ask ing thems elves how "many clusters aw ay" traits are fro m one a nother. So warm would be seen as close to kind because they are in the same low level cluster, somewhat more distant from h appy w hich is a cluster a way, a nd still m ore dista nt from smart which is severa l clusters distan t.

Are Stereotypes Like Implicit Personality Theories? Earlier w e sugg ested th at there a re at least four w ays th at stereo types ca n be seen as imp licit personality theories. First, traits may be seen as related to groups. Second, groups can be seen as defining one or more dim ensions that lie within a m ulti-dimensional spac e or which allow us to discriminate clusters. Third, groups may act as variables that change the nature of trait relationships for different groups. Fourth groups themselves may be seen as more or less closely related.

Groups as Traits The first approach of seeing groups as a kind of trait related to other traits has been used often. Severa l writers (e.g ., Gaertn er & Mc Lau ghlin, 1 983; D ovidio , Evan s, & Ty ler, 1986 ; Tajfel, 198 2) explicitly define stereotypes in of group-feature correlations. Certainly it makes intuitive sense to see the traits ascribed to groups as representing a kind of correlation between features and group. Such an approach a lso fits well the common definition of stereotypes as perceived asso ciations between featu res and a category or group; correlations are one powerful way of expressing associations. The basic argu ment und erlying this approach is tha t people perceive relationships between traits and grou p hip w hich can be expressed as correlations. So, gend er is perceived to be correlated with tendency to comm it acts of physical aggression , with nurturing, an d with ma king money . Or when some one sa ys tha t blacks are ath letic, they a re really sta ting a c orrelatio n betw een race and a thletic ability. The only requirement is that the trait under consideration has to be seen as discriminating between two groups. Thus to say that Jews are clannish, we are really saying that they are more clannish than gentiles. A person who asserts that a small pe rcenta ge of Jew s are cla nnish (say 30 %) wo uld still hold a stereotype that Jews are clannish if she believed that only 10% of non-Jews had this trait. Thus in a sense we c ould sa y that bein g Jewish is d iagnostic for being clan nish. Ho wever, even traits that are highly characteristic of people in a given group will not be correlated with group hip unless there are contrasting groups with less or more of the trait. Jews have two eyes, but number of eyes is not correlated with Jewishn ess precisely because almo st everyone has tw o eyes 7. Differential assignment of traits to g roups usua lly express ed as p robab ilities can e asily be c onve rted into correlatio ns. Th e claim is not that people tote huge correlation matrices to judgment tasks, but rather that they believe some groups are more likely to possess some traits than are other groups. Research scientists can then turn these perceived assoc iations into co rrelation s whic h are m ore trac table sta tistically. So far, so go od, but u nfortun ately there are tw o major pro blems w ith thinking of group-fea ture relationships as correlations. One is that this approach requires that both the group and the feature be seen as ca tegorical or d iscrete rather tha n contin uous variables. M any, perha ps mos t, group are categorica l (one is either ma le or female, a law yer or not, etc.) bu t we often im agine tha t features are continuous. So while it makes perfect sense for me to say that the average Jew is more clannish than the

7

People in stereoty ped gro ups o ften plac e great em phas is on tra its that a re not co rrelated w ith their hip as a w ay of remin ding oth ers that there a re a great ma ny traits tha t are actua lly more characteristic of them if less discriminating. Recall Shylock's speech from The Merchan t of Venice: "I am a Jew. Hath no t a Jew eyes? Hath not a Jew ha nds, organs , dimensions, senses, affections , ions?-- fed with the same food, hurt with the same weapons, subject to the same diseases, healed by the same means, warmed and cooled by the same winter and summer as a Christian is? If you prick us, do we not bleed? If you tickle us, do we not laugh? If you poison us, do we not die? And if you wrong us, shall we not revenge?" Chapter 5: Structure and IPT / Version of November 10, 2002

avera ge gentile (s ay 6 v s. 4 on a 10-p oint cla nnish scale), th is way of think ing ab out th ings d oes no t easily produce correlations.8 A second pro blem is that correlations ba sed on judgm ents of what percen tages of groups h ave a trait are g enerally n ot the sa me as c orrelatio ns ba sed on wha t percent ages o f people w ith given traits fall into given groups. I observe someone who is clannish. Am I justified in thinking that this person is also likely to be Jewish. The answer is that you perform this bit of inference work at your peril because of base-rate differences. Let us say that, in the United States there are 10,000,000 Jews 9 and 250,000,000 people o f other d esigna tions. A nd if ou r perceiver th inks th at 30% of Jews a re clann ish, she shou ld (unless she is computationally challenged) say that there are 3,000,000 clannish Jews in this country. On the other hand, if she claims that 10% of other Americans are clannish, there would be 24,000,000 clannish non-Jews. Thus according to these numbers only about 10% of clannish people are Jewish, and the resultant feature to grou p correlation wou ld be quite different than the group to feature one. Ordinarily this is not a major problem because we usually are far more interested in what percentage of various grou ps have particular featu res, but this particular asym metry needs to be kept in m ind (See Chapter 3 for related discu ssion). Finally the nature of the correlation depends on the groups being compared. Usually when one makes a statemen t about a group “ having ” a trait peop le are asserting th at people in th e group a re more likely to display the trait than people in general. But sometimes specific comparisons are used as well. So the person who thinks that 30% of Jews are clannish and 10% of gentiles, may also think that 50% of Hispa nic Am ericans are. Ob vious ly this pla ys ha voc w ith the co rrelation al appro ach, b ut aga in this is usually not a problem so long as group are compared to natural contrasts (e.g., males vs. females, gays vs. stra ights) o r to peop le genera lly.

Unconfounding Variables One o f the va luable a dvan tages o f thinkin g of stereo types in c orrelatio nal term s is tha t it automatically alerts statistically savvy people to the possibility of confounding variables. Anyone who works long with correlations comes to understand that two variables may be correlated for a variety of reason s. On e possib ility is that one va riable ca uses th e other, o r for tha t matter caus ality is bi-d irectiona l. But a secon d (and mo re common ) is that both variables are affected by some third (and often unmeas ured) variable. "Correlation is not ca use" -- or so we often tell our stud ents. Those wh o deal with correlatio ns co me to s ee the w orld as a hug e correlat ion m atrix with nearly ev erythin g related to nea rly everyth ing else. St atisticia ns ha ve, of co urse, de veloped a num ber of sta tistical m ethod s such as mu ltiple regressio n and factor a nalys is precisely to try to b ring so me ord er to this c haos . The issu e before is is wheth er mos t people rec ogniz e these co mplexities , and if s o wh at they do ab out it. Obv iously correlatio ns betw een gro ups a nd tra its may have m any c auses . Indeed , almos t all correlations between gro ups and tra its are probably caused by some intermed iary variable. If, in fact, Hispan ics are mo re family oriented than A nglos, law yers more g reedy than professors, o r wom en more nurtura nt than men, we m ight look to the role of cultu re as the interm ediate cau se. For the m ost part group mem berships do no t directly cause particular features to ap pear or get buried. Dark skin d oes not,

8

If we had information about what percentage of people a given perceiver thinks falls at each point along the scale (e.g., 5% of Jews are very clannish, 25% somewhat clannish, etc.) a type of measure that can b e conv erted into num erical ratin gs, we w ould b e able to c alcula te a corre lation. A lternativ ely if several perceivers each rated Jews o n clannishish ness, we could also turn the resultan t mean differences and variances into correlations. Both strategies do, however, seem somewhat awkw ard and perhaps stray too far from the h eart of the group-feature correlation a rgument. 9

It is a non-trivial problem to estimate the exact number of Jews in the United States. Currently about 6,000,000 A mericans are d with on e of the several Jewish religious organ izations, but, of cou rse, man y Jews (lik e man y other people) ar e not fo rmally re ligious . Chapter 5: Structure and IPT / Version of November 10, 2002

per se produ ce mo st psyc holog ical traits , nor do es being a lawy er directly c ause m any s tereotyp ic lawyerly traits.

Race and Social Class: Race produces more than its share of such confounded variables. For example, given that blacks are over-represented in the working and lower classes, it has been argued that stereotypes o f blacks are, in rea lity, stereotypes of low er class people (B ayton, M cAlister, & H amer, 195 6). Feldman (1972; Feldman & Hilterman, 1975) explored the extent to which race and occupational designations imp lied occupationa l and racial stereotype traits. A first step was to estab lish a list of traits that distinguished between occupational groups (professional vs. working class) and race (black vs. white). So traits such as athletic, musical, thrifty, and self-satisfied discriminated between whites and blacks, and traits such as persistent, coarse, independent, and intelligent were differentially associated with social class . Sub jects w ere giv en stim ulus perso ns w ith va rying com bina tions of rac e, occ upa tion, a nd fa ther's occupation, and asked to rate each on the class and race stereotype traits. Surprisingly, occupation designation was a far more important predictor than was race. For example, race of the stimulus person ed for about 1% of the variance on race stereotype traits and even less of the variance for ratings of the occupational traits. On the other hand, manipulated occupation ed for about 5% of the variance on race stereotype traits and about 50% of the variance on occupation trait ratings. This means that on a race trait such as self-satisfied, knowing that a person was black or white has less impact than knowing whether he was professional or working class, and this difference was even greater for an occu pation al trait su ch as persisten t. More recently Devine and Baker (1991) had their white subjects rate backs as well as various subtypes of blacks (e.g., black athlete, black businessman, ghetto black, welfare black). The ratings of the black b usines sman seemed to hav e more in comm on w ith bus inessm en in gen eral tha n with blacks in general. However, the ratings of blacks in general were fairly similar to ratings of welfare blacks, streetwis e blacks , and g hetto b lacks (a nd th ese latter th ree grou ps were not w ell differentia ted). Th is again su ggests tha t when s ubjects rate bla cks in gen eral they ma y implicitly be rating an imag e of lowerclass blacks. There is some suggestion that when blacks are seen as middle-class, their race becomes less importan t as a facto r in their ratings. T his is, of cou rse, an exam ple of wha t we saw in Chap ter 3 where sub-types of categories may seem prototypic and come to dominate the stereotype of the larger category. This does no t prove that the stereotype of blacks is really a stereotype of working class peo ple, but it certainly is consistent with the idea that if people think of blacks as working-class, it is the occu pation rather th an th e race th at will dic tate stereo types. Th is may have im portan t cons equen ces in everyday life. For example, Kennelly (199 9) argues that em ployers use a stereotype of black sin glemoth ers as a framew ork for ev aluat ing fem ale, especia lly black , employ ees, and cons equen tly think their work ers are po orly prepa red an d un reliable.

Gender and Status: Alice Eagly and her colleagues (e.g., Eagly, 1987; Eagly & Wood, 1982; Eagly & Steffen , 1984 ) have s ugges ted tha t stereoty pes of w omen are actu ally stereo types o f wom en in particular occupations or social roles. Eagly and Steffen (1984) argued that women are seen as less agenetic and m ore commu nal than m en because they m ore often occupy ro les such as ho memaker, nu rse, and elementary s chool teacher tha t foster commu nal and n urturant beha viors and inh ibit more takecharge, agenetic behaviors. In their study they asked subjects to rate men and women in various roles. As expected women were perceived as more communal than men but only they occupied a low status position or no information was given about status. When women were seen as occupying a high status position, they were perceived similarly to men occupying the same positions. In a conceptually similar study Eagly and Wood (1982) showed that when subjects did not know what kinds of jobs people had they inferred that women would have lower status jobs than men and would be more compliant because of that lower status. However, when women were identified as having higher status, they were not seen as more compliant than the men. In other words, compliance-proneness was seen to be more a matter of job status than gender, but because women more often have lower status jobs, perceivers may fail to take that into and see women as having the "traits" of their position. Thu s, the pu tative co rrelation between gender and c omm una l traits ma y turn out to be cau sed pa rtially or wh olly by a ssum ptions abou t the kin ds of ro les wom en occ upy.


However, other research presents a less sanguine picture about such matters. Schein (1973) showed that ratings by male managers at insurance companies of successful middle managers and of men were high ly correlated whereas ratings o f successful midd le managers an d wom en were not. Female managers, on the other hand, see females and successful managers as more closely related than do m ale ma nagers (Schein , 1975 ). A mo re cont empor ary vers ion of th is stud y by H eilman , Block, M artell, and Sim on (198 9) found that men are rated as closer to su ccessful m anagers (.68) than are wom en (-.24). 10 Followin g the lead o f Eagly an d her stud ents one m ight argu e that suc h results oc cur beca use the raters are implicitly ass uming that the w omen th ey are rating h ave low status job s that do not require managerial skills. Therefore Heilman and her colleagues also had their subjects rate male managers, successful male managers, female managers, and successful female managers. Obviously, specifying that the females were successful managers raised the correlation between ratings of successful managers and females. Howev er, even when the fema les were depicted as success ful managers, the co rrelation with successful manager in general (.94) was a bit (and statistically significantly) lower than the correlation between succe ssful m anag er and succe ssful m ale ma nager (.9 9). This study represent s a bit of g lass ha lf empty-half full. On the one hand, the most striking result confirms the ideas of Eagly and her students. When women are considered in a higher status role much of the difference between ratings of men and wom en disapp ears. On the other h and, it nev er completely d isappears ev en when the wom en are depicted as su ccessfu l.

Can People Unconfound Variables: While we have seen evidence that strongly suggests that stereotypes of blacks and women are really stereotypes about people who occupy the kinds of roles and positions they often occupy, the real question is whether people are aware of this. The average person with a cultural stereotype at the ready about women as being kind and nurturant may well realize that women in high level jobs do not act that way. He may even be willing, if pressed, to say that women as well as men often behave the ways they do because of various role and social pressures. The most important question, however, is whether he can easily put those observations together and unconfound the effects of role and gender on trait ascriptions. Does he understand that jobs and roles may have far more to do with stereotypic traits than actual gender? And if he does can he then go the extra step and recogn ize tha t wom en are likely no differe nt tha n men on the se traits o r at the lea st und erstan d that his stereotypes about women are about women in certain positions? Earlier we suggested that people are not inevitably good at estimating covariations from contingency tables, and now we are asking whether they can perform an even more complex task, one that involves what we might call a double contingency table. Can they understand that one set of relations is based on a nother? Con sider a concrete example from Schaller and O 'Brien (1992). Subjects read abo ut wh ether individ uals in G roup A o r Group B had so lved or no t solved a nagram s that w ere either hard (7 letters) or relatively easy (5 letters). It turns o ut that G roup B a ctually so lved mo re anagram s than Grou p A. How ever, Group A ha d also attempted m ore of the more difficult seven-letter anagrams. In fact had subjects taken that into they would have seen that Group A had actually, on a percentage basis, been more successful at both types of anagrams. Table 5.3 shows the relevant numbers.

10

The fact that this correlation is n egative, albeit not strongly so, sug gests that wom en are rated lowes t on th e traits th at cha racterize succe ssful m anag ers. Chapter 5: Structure and IPT / Version of November 10, 2002

Table 5.3: Distribu tion of stimu lus inform ation (after S challer & O'Brien , 1992). Five-Letter Anagrams Success

Failure

Seven-Letter Anagrams Success

All Anagrams

Failure

Success

Failure

Group A

5

0

5

15

10

15

Group B

15

5

0

5

15

10

As is clear from the table, the seeming w orse overall performance o f Group A is du e entirely to their having attempted more difficult problems. If subjects were ab le to correct the correlation between group and performance for difficulty of problems, they should have seen that Group A was actually the better performing group. Yet, on average, subjects reported that Group B was superior, suggesting insuffic ient corr ection. However, that does not mean that people can never perform such double contingency tasks correctly. In a second study S challer and O'Brien (19 92) asked some su bjects to think about h ow ha rd each of the problems was as they received information about performance on that problem. In that condition subjects co rrectly reported th at Grou p A actu ally performed better. Subjects a lso mad e more accurate judgm ents when th ey had mo re time to consider each n ew piece of information. Fu rthermore, giving subjects instruction on the logic of covariation reduces the effect (Schaller, Asp, Rosell, & Heim, 1996). People also tend to aggregate data over situation with smaller samples and hence lose covariation informatio n (Scha ller, 1992a). People are also better at these tasks when the "disadvantaged group" is one to which they belong. Schaller (1992b) had men and women subjects make leadership judgments about men and women who were describ ed as goo d or poo r leaders in different s ituations . Generally, both males an d females w ere better leaders when they held executive positions and worse leaders when they w ere office workers. In fact, 75% of the executives were described as good leaders and only 25% of the office workers were, so there was a stron g correlation between role an d leadership ability. In this experiment 75% of the women were described as office workers and 75% of the males were portrayed as executives, so there was also a strong co rrelation betw een gender a nd role. Th at mean t, of course, th at the wo men (w ho were m ore likely to be in the non-leadership office worker role) were less good leaders overall than the men. However, this correlation between leadership and gender was entirely explained by the correlation between role and lead ership and gend er and role. In fact, within each role, equal percentages of males and fe males w ere good leaders. I n this s tudy b oth m ale and female su bjects ov er-estima ted ma le leadership ability relative to female, but the male sub jects did this to a muc h greater extent. In other words, the women seemed to be better at adjusting for the "disadvantage" that the women stimulus persons experienced. In another experiment subjects were given the data from the anagram task shown in Table 5.3 , but this time th ey were "ass igned" to gro up A or G roup B. T hose su bjects who were assigned to the "disadvantaged" group (A in Table 5.3) were much better at discovering that Group A actually performed better. However, it does not inevitable follow that people process information about their own grou ps in a more careful w ay. Schaller (1991) fou nd that w hen subjects were as signed to group s they rep orted th at "their" g roup h ad perfo rmed rela tively m ore pos itive beha viors. This series of studies rather nicely shows that people do have trouble untangling the various correlated roles, groups, and behaviors that occur in everyday life although some conditions do improve performance (Scha ller, 1994). Incentives (such as finding one's self the victim of such inaccurate judgments) do es tend to produ ce more attention to th e complexities of the case and hence more acc urate judgments. However, one problem in everyday life is that other people do not listen to such victims. For example, may whites have trouble understanding why blacks seem to be so much more hostile to police than whites. When blacks point out that they have more often (proportionally) been the victims of police brutality and rud e behavior, and th at these provoca tions accou nt in part for their behavior, ma ny whites reject this reasoning as self-serving. The argument here is not that every such claim is correct or that groups are necessarily the best judge of the causes of their behavior, but only that there may be occasions Chapter 5: Structure and IPT / Version of November 10, 2002

when their greater self-interest may make them more accurate observers of contingencies, and that we ought, at the very least, not to brush aside such complaints as merely self-serving and biased.

Illusory Correlation Early Research by Chapman : The most prominent use of the idea that stereotypes can be cons idered co rrelation s betw een cat egories a nd fea tures h as been a series o f studies pioneere d by D avid Hamilton which were based on earlier research by Loren and Jean Chapman. Chapman and Chapman (1967, 1969 ) were concerned w ith why clinical psych ologists used inva lid signs on projective tests to infer various kinds of symptoms and clinical conditions but did not use other more diagnostic cues. For example, although the Draw-a-Person test has generally been found to lack any empirical validity as a diagnostic test, it was being used widely by diagnosticians because they felt it was valid. So, clinicians reported that paranoid patients draw pictures with large eyes and men concerned about their manliness wou ld draw pictures of mu scula r men d espite the fa ct that these typ es of dra wings are emp irically uncorrelated with the symptoms given. Chapman and Chapman (1967) suggested that the reasons clinicians used such signs is that they just seem to go with symptoms in some "com mon sense" way. If that were correct, then college students, naive to the arts of clinical diagnosis, would also see these same correlations between signs and symptoms. That was found. Clinicians use these invalid signs presumably because there is a common sense correlation, an illusory correlation, between signs and symptoms. The Chapmans were suggesting that clinical psychologists, like the rest of us, think that things which should go together actually do so. We might call this a schema-based illusory correlation. In a subsequent set of studies Chapma n (1967) developed another basis of illusory correlation. In these studies he paired com mon w ords from two lists such that every w ord from the first list appeared equally often with each word from the second list. He then asked subjects how often each pair of words had appeared together. Although every word on the first list was paired an equal nu mber of times with every word on the second list, subjects saw some pairings as more common than others; there was a perceived, but illusory, correlation among these pairs. Words that had an associative relationship (eggsbaco n) were s een as h aving appear ed toge ther mo re than they ha d, and this resu lt is similar to the o ne in the clinical stu dy describ ed abov e. There wa s, howev er, another res ult that w as more u nexpected -words that were u nusu al or distinctiv e (in this stud y by being lo nger) also led to such illusory correlations. That is words were distinctive were seen as having been paired more often than they had. This is an example of a distinctiveness-based illusory correlation. The Hamilton-Gifford Extension to Stereotyping: Hamilton a nd Gifford (1976 ) extrapolated this latte r result to stereoty ping. Th ey reaso ned th at the v ictims o f stereoty ping ar e often in frequen tly encou ntered by perc eivers and that they a re often seen to possess rela tively negativ e traits which are usua lly less frequent an d hence d istinctive. Even if distinctive (say negative) beh aviors or tra its were equally common among black and white people, a white perceiver who sees far more whites than blacks in a given d ay migh t form an illusory correla tion betw een black a nd nega tive traits (becau se both a re relatively infrequen t and h ence distinc tive) by remem bering or jud ging that th ey went to gether mo re often than they had. Having created this illusory correlation, the perceiver would then have justification for evaluating blacks more negatively. In their experiments Hamilton and Gifford wanted to avoid using actual groups designations becau se of pot ential co nfou nds w ith existin g stereot ypes. So subject s were s imply to ld that they w ould read (via slide presentation) a series of behaviors exhibited by of two groups, Group A and Group B. Th ey were further told that Grou p B had fewer m embers than G roup A so th at they wou ld see fewer behaviors from of Group B; thus group B was relatively distinctive because it was smaller. In rea lity they saw 26 beha viors for A a nd 13 fo r B. For Grou p A 18 o f the behav iors were positive and 8 were negative, and for Group B they saw 9 positive and 4 negative. Note that the proportion (approximately 70%) of positive to total behaviors was equivalent for the two groups so that there was actually no correlation between group hip and kinds of behaviors. Note also that negativ e behav iors ha d been mad e distinc tive by v irtue of th e fact tha t they w ere relatively infreque nt in this study. Following the presentation of all 39 behaviors, the subjects were asked to perform three tasks. They ra ted the m embers of Grou ps A an d B on a series o f traits so that th e investig ators c ould e asily


calculate an index of how positively the two groups were viewed (evaluation task). Subjects were also given a list of the 39 behaviors and asked whether the behavior had been performed by a person from Group A or from Group B (a recognition measure). Finally, they were told that they had seen 26 behaviors for Grou p A and 1 3 for Group B, an d they were asked how m any of these beha viors had been und esirable (freq uenc y estima tion). Both th e frequency es timation a nd the reco gnition m easures w ere used to a ssess illusory correlatio n. The d ata are given in Table 5 .4. As th e data in that table sh ow clea rly, the su bjects did too m any negative (distinctive by virtue of numerical infrequen cy) behaviors as h aving been paired with the numerically smaller Group B. The frequency judgments showed the same pattern. So, for example, whereas they had seen only 4 negative behaviors performed by Group B, they indicated on the recognition test that 6.21 negative behaviors had been performed by B, and they estimated that Bs had performed 5.73 nega tive behaviors of the 9 th ey had seen. No t surprisingly therefore, on traits judgmen ts B wa s seen a s more negativ e than A. Table 5.4: Traits presented and Rated for Groups A and B. Group A

Group B

18

9

Recognized

17.52

9.48

Estimated

17.09

7.27

Presented

9

4

Recognized

5.79

6.21

Estimated

8.91

5.73

Desirable Traits

6.91

6.15

Undesirable Traits

4.39

5.31

Desira ble Presented

Unde sirable

Evaluation

Hamilton a nd Gifford were na turally most interested in ho w negative traits get attach ed to minority groups, and they reasoned, quite appropriately, that negative traits are usually encountered less frequently than positive. But the basis of the effect is shared infrequency/distinctiveness and is not essentially due to types of groups or traits. Just to be sure, Hamilton and Gifford repeated the experiment with the positive traits being the less frequent. The results were comparable with the first study in the sense that it was n ow the positive traits that tended to be seen a s more chara cteristic of the smaller (minority) group. So alth ough the effect c an be o btained with po sitive as well as w ith nega tive traits , in their review of sev eral illusory co rrelation stu dies Mu llen and Jo hnson (1990) fou nd tha t the effects are strong er when the infreq uent, d istinctiv e traits are also n egative r ather th an no n-nega tive. It is pro bable that negative stimuli are inherently more distinctive than positive. Negative traits are typically seen less often than positive traits (K anou se & Ha nson, 1 972), bu t negative info rmation is also indep endently m ore salient (Skow ronski & Carlston , 1989). Th us, infrequen t negative tra its may a ctually be ev en more distinctive than on e might predict from infrequency alone.


Generalizability of the Effect: Although in d ealing with stereotypes we a re most concern ed with feature-c ategor y correla tions, d istinctiv eness-b ased illu sory co rrelation s also o ccur w ith oth er stimu li. For exa mple, w e can d evelop illu sory co rrealtion s betw een dist inctive b ehav iors an d the co ntexts in which they oc cur, seein g more negativ e behav iors in rela tively rare contex ts than is justified (M cCon nell, Liebold, & Sherman,1997). Illusory correlations have also been found between objects and amount of information presented about them (Sanbonmatsu, Shavitt, & Sherman,1991), and with altitudes and size of tow ns (Spe ars, va n der P ligt & Eis er, 1985 , 1986 ). There is another more circumscribed but also important question of generality. Do these evaluative effects go beyond the behaviors that are presented? If I decide that lawyers perform more than their share of greedy behaviors, do I also assume that they are likely to have additional negative traits? In other words, do the illusory correlation effects generalize across feature domains. Acorn, Hamilton and Sherman (1988) addressed this issue. They used behaviors that varied along one dimension (e.g., social desirability), created the usual illusory correlation co nditions, and th en measured whether the mino rity group s was evalu ated low er on th e man ipulated dimen sion a s well as on a d ifferent dim ension (e.g., intellectu al desira bility). Gro ups th at were s een as s ocially u ndesira ble were a lso seen as intellec tually undesirable even though no information had been presented to the subjects about intellectual qualifications. Similarly, groups that were seen as too immature were also seen as introverted although introverted groups were not seen as more immature, so generalization across domains is apparently not univ ersal. In everyd ay life there are bou nd to be real restr aints o n gene ralizatio n acro ss lingu istic domains , but even if these effects occur in limited way s they are still important. One importa nt qualification on th e illusory correlation effect is that it works for groups , but does not w ork as well for in dividu als (Mc Con nell, Sherm an & H amilto n, 199 4a; Sa nbon mats u, Sher man , & Hamilton, 1987). Hamilton and Sherman (1996) argue that this results from differences in how information is processed; in particular information about individuals is more likely to be processed online and info rmation about g roups ba sed on m emory (See C hapter 4). Th eir reasoning is that perceivers expect people to be consistent, unitary entities, but they do not assume that groups are entities in that way (See Chapter 3 for a more extended discussion). To the extent that we assume that we are judging a distinct entity, we are more likely to process information on-line (Hastie & Park, 1986) rather than waitin g until a ll the inform ation is in and makin g mem ory ba sed jud gmen ts. It is important to that Hamilton and Sherman propose that differences in processing goals un derlie differences in th e ways g roups a nd indiv iduals are p erceived. On average in dividua ls are thoug ht (with co nsiderab le justice) to form m ore unitary and co herent entities th an grou ps. But the re are individuals who are inconsistent and variable, and there are groups (e.g., families, cults, fraternities and sororities) that we expect will have a high degree of entitativity. McConnell, Sherman, and Hamilton (1997) m anipula ted perceived en titativity for both groups a nd indiv iduals. As expected illuso ry correlations were higher for grou ps than for individu als, but the effects were also mu ch higher for both group s and individ uals pe rceived t o be less c onsis tent (low entitativ ity).

Distinctiveness of Traits and Groups: While the majority of research done within this paradigm has u sed frequ ency o f behav ior and size of gr oups as a m eans o f induc ing sh ared d istinctiv eness, th ere is nothing sacred about this. Indeed, in everyday life, there may be many other, more important, ways that stimuli beco me salient a nd distin ctive. For exam ple, extreme beha viors wo uld presu mably b e more distinctive than moderate ones, and hence might be ready candidates for illusory correlation effects. Indeed illusory correlatio n effects a re stron ger with more ex treme beh aviors (Sanb onm atsu, S havitt , & Gibso n, 199 4) and with m ore extrem e, emotio nally la den stim uli (John son, M ullen, C arlson , & Southwick, 2001). Groups might also be made more or less distinctive. We might, for example, expect that people with bright red hair would be ed as performing more extreme, or otherwise distinctive, behaviors than those with run -of-the-mill brown hair even if both groups had performed equal percentages of extreme behaviors. Manipulating distinctiveness of groups by instructions to pay close attention to the group also produce the effect (Sanbonmatsu, Shavitt, & Gibson, 1994; Sanbonmatsu, Sherman, & Hamilton, 1987). In addition when information is reported about a third Group C, which


presumably makes the minority Group B less distinctive, the illusory correlation effect is reduced but not elimina ted (Sh erman , Ham ilton, & R osko s-Ew oldsen , 1989 ). Stimuli or groups th at are relevant to the self ought to be especially distinctive. Several studies have shown that illusory correlation effects are strongest for attitudes consistent with one’s own Berndsen, Spears, & van der Pligt, 1996; Spears, van der Pligt & Eiser, 1985, 1986). In another test along these lines Stroessner and Heuer (1996) found that smaller groups were associated with more infrequent types of treatment by police. This w ould help explain wh y in everyday life of rac ial minorities may be seen as especially closely associated with unfair practices by the police, if we can assume (and I think we can ) that such practices a re comparatively rare. 11 While such research suggests that motivational processes m ay help to m ake som e stimuli distin ctive, it also sho ws tha t illusory correla tion effects are found with different stimuli than behaviors and different means of manipulating distinctiveness. Generally illusory correlation effects are enhanced when people have a vested interest in the features. Sch aller and M aass (19 89) were interes ted in wh ether people w ould be m ore likely to form illusory correlations when their own group was one of the groups. They found that group saw a lower percen tage of nega tive behav iors asso ciated with the in-grou p than fo r an ou t-group. Illus ory correlations are stronger when one’s in-group is the majority group than when the minority group (Haslam, McGarty, Oakes, & Turner, 1993). Similar results were found by van Knippenberg, van Knippenberg, and Dijksterhuis (1997) who showed that when processing goals interfered with the formation of forming an on-line impression subjects were more prone to see positive behaviors as going with th eir own in-grou p.

Why Does the Effect Occur? Illusory correlation effects while not dra matically strong are qu ite robust and appear with a range of stimuli. As is inevitably the case in such situations, there have been a number of theoretical perspectives brought to bear in explaining the phenomenon. The Original Distinctiveness : As I ha ve emp hasiz ed up to this poin t, Ham ilton an d his colleagues have assumed that the effect is produced by the relative distinctiveness of both groups and features. More particularly they argue that the effect is produced by memory biases, and that biases take place at time of initial encoding (e.g., Ham ilton, 1981c; Ha milton & Sherm an, 1989). Presu mably wh en the distinctive stimuli appear they a re given extra attention and additional processin g that makes th em readily available for retrieval. So, when people try to remem ber behaviors an d which beh avior went w ith which group, they will find it easiest to the distinctive behaviors attached to the distinctive groups. In , Stro essner, Ha milton, an d Mac kie (1992) h ave sho wn tha t in the usu al illusory correlatio n con ditions , subjects paid m ore atten tion to the distin ctive beh aviors of the m inority g roup; Johnson and Mu llen (1994) and M cConnell, Sherm an, and H amilton (1994 b) have foun d that subjects were faster in asg comparatively rare behaviors to smaller groups. If subjects generally pay more attention to the distinctive behaviors associated with the distinctive group, we might expect to find that recall would also be better for such behaviors. Hamilton, Dugan , and Trolier (1985) fou nd that su bjects recalled a larger proportion of distinctive stimu li paired with the smaller group than with any other combination of groups and stimuli. Mullen and Johnson (1990) also concluded from their review that the relative size of the "double distinctive" cell predicts the size of the illusory correlation effect as one w ould predict if that cell is especially salient. On the other hand, there are dem onstrations (e.g., Meiser & Hew stone, 2001) of illusory co rrelations being produced that a re unm ediated by differen tial mem ories for d istinctiv e stimu li. Retrieval Biases: The distinctive stimuli assumes that biased encoding is the culprit, but it is often difficult to distinguish encoding from retrieval Others hav e argued that m emory loss after encod ing is th e real culp rit. Fiedler (19 91; 19 96) arg ues th at the effec t can b e und erstood in of wellknow n psyc hom etric prop erties of th e situat ion. O ne wa y to loo k at the s ituatio n con frontin g sub jects is that they are trying to learn that Group A has approximately twice as many positive as negative

11

And there is th e addit ional a nd ob vious possib ility that m embers of min ority gro ups a ctually receive proportionally more un fair treatment. Chapter 5: Structure and IPT / Version of November 10, 2002

behaviors, and that Group B has this sam e ratio. However, there are mo re opportunities for subjects to learn the ratio for Group A (26 beh aviors) tha n for Grou p B (13 beh aviors). W hen judg ments a re unreliable estimates of the true proportions, one should expect the judgments to regress toward the mean, that is for subjects to estimate more negative and fewer positive . Since reliability of judgment should be lower in the minority condition (because there are fewer opportunities to learn the true proportions of positive and negative items) it follows that such effects should be stronger for the mino rity (sma ll group ) cond ition. Th is has b een fou nd (Fied ler & Arm bruste r, 1994 ; Fiedler, Ru sser, & Gram m, 199 3). Others (K lauer and Meiser, 200 0; Meiser & H ewston e, 2001; Sm ith, 1991 ) have fou nd tha t illusory correlation effects result from retrieval failures or biased retrieval rather than biased encoding. One line of favoring these types of explanations is tha t illusory correlation effects can be obtained even when behav iors are n ot ass igned ex plicitly to g roups (McGa rty, Ha slam, T urner, & Oak es, 199 3), a condition tha t is necessary for distinctiveness ba sed memo ry accoun ts. Indeed simply presenting su bjects with the information that Group A is larger than Group B and that there are more desirable than undesirable behaviors is sufficient to produce illusory correlations when subjects estimate what behaviors went with what groups. Self-Categorization Theory : Another class of counter-explanations rests on the Self-Categorization Theory assumption that when people read evaluative statements about two groups they assume that the groups are different and a ttempt to process the inform ation in way s that accentu ate these differences (McGarty & de la Hoye, 1997; McGarty, Haslam, Turner, & Oakes, 1993). In particular they assume that one of the groups must be “better” than the other. The most reasonable hypothesis subjects in these experiments might entertain is that the most frequent behaviors should go with the mo st frequent group, and so the larger grou p shou ld be seen as the more po sitive (assum ing that po sitive items are m ore frequent, as they are in mo st studies). Looked a t one way th e actual items seem to su pport that idea; after all there are 22 pieces of information consistent with it (18 positive items for A and 4 negative items for B) and on ly 17 pieces o f informatio n that are in consisten t (8 negative item s for A an d 9 positiv e items for B). One pred iction from this mod el is that the illuso ry correlation effect will not app ear when subjects are not especially motivated to construe the groups differently. In of this reasoning, Haslam, McGarty, an d Brown (1996) found that when th e groups were labeled as left-han ded and right-h anded people (two groups m ost of us assu me do no t differ in important ways) no illusory correlation effects were found. Further comes in a study by Berndsen, Spears, McGarty, and van der Pligt (1998) where it was found that when the two ostensible groups are both highly variable in the evaluative behavio rs illusory co rrelation is redu ced presum ably beca use grou p differentiation b ecomes m ore difficult. Also ev en when people are given informatio n abou t only on e group th ey produ ce illusory correlations, again presumably because they think the groups must be different (Berndsen, McGarty, van der Pligt, & Spea rs, 2001). The Bottom Line : At present we do not have a definitive of why illusory correlation accou nts occu r. Each o f the major exp lanation s predicts effects n ot predicted b y the others , and there a re data in cons istent w ith each . It is perha ps wo rth poin ting ou t that th ese mo dels are n ot gene rally mu tually exclusive. So, for example, in a given situa tion illusory correlations can be produced bo th by biased encoding and retrieval. Our needs to discriminate between groups may lead to assumptions about which group possess es the most positive beha viors, but memo ry biases can still or ad d to the effect. And it may be that in the laboratory as well as the outside world, different theories for the effect in differen t situat ions o r with d ifferent kin ds of st imuli.

Schema-Based Illusory Correlations: We (like the field at large) have emphasiz ed distinctiven ess-based illusory correla tions beca use suc h effects are no t especially intuitiv e. Howev er, Chapm an's origina l notion o f illusory correlatio n was devised to explain tho se comm on situa tions wh ere people assume that stimuli they expect to be correlated are, in fact, more highly correlated than experienc e wou ld justify. There are several demonstrations of such effects (Anderson, 1995;Berman & Kenny, 1976; Hamilton & Rose, 1980; Kim & Baron, 1988; Mackie, Hamilton, Schroth, Carlisle, Gersho, Meneses,


Nedler, & R eichel, 1989 ; Madey & Ondru s, 1999; M eehan & Janik, 199 0; Plessner, Frey tag, & Fiedler, 2000; Slusher & Anderson, 1987; Trolier & Hamilton, 1986). Just to give one example, Hamilton and Rose (1980) paired traits and groups labels and found tha t subjects though t that stereotypic traits had been paired more often with the relevant group than had been the case. That is, although the pair salesmantalkative m ight not h ave appea red more o ften than salesma n-wealth y, subjects sa w the first as m ore common because talkativeness is a part of the salesman stereotype whereas wealthy is not. This research indicates that even when people are exposed to information suggesting that their stereotypes may be incorrect, they may still misperceive associations where none are present. Presumably somewhat different processes underlie schema-based and distinctiveness-based illusory correlations. 12 GarciaMarques and H amilton (1996) argue that schema con sistent traits are attached more strongly to groups than are schem a inconsistent traits, but tha t the inconsistent informa tion is also more stron gly linked to other items of information (see Chapter 4 for a fuller argument). When people try to estimate frequency of consistent and inconsistent items, they will find it easier to access consistent information (because of the stronger links to the group) and because such information is more available it will also be seen as more frequent. That would explain schema-based illusory correlations. On the other hand, distinctiveness-based illusory correlations are based on recall of information, and here the inconsistent information ha s an adv antage becau se it can be accessed bo th via the group n ode as well as oth er features or traits. Thus in making memory based judgments, people will access more distinctive inform ation. McArthur & Friedman (1980) raised the question of whether illusory correlations would persist with natural groups for which there are "built-in" correlations. In general, they found that the typical frequency-based illusory correlation effect tended to wash out when it competed with these other basis of association, a resu lt confirmed by several other inv estigators (Berndsen, Spears, M cGarty, & van der Pligt, 1998; Berndsen, van der Pligt, & Spears, 1996; Fiedler, Hemmeter, & Hofmann, 1984; Hamilton & Rose, 19 80).

Groups as Variables There is no dou bt that people see relationships am ong stereotype com ponents. For exam ple, Deaux and Lew is (1984) stu died the relation ships betw een gender la bels and a mong various genderlinked features. Not surprisingly they found that males and females possessed different traits, physical features, and occupations. What wa s more interesting was that there were substantial relationships amo ng the features . So, for exa mple, a p erson w ho exh ibited m ale traits w as also seen to b e more lik ely to occ upy a masc uline o ccupa tion, to o ccupy male ro les, and to hav e masc uline ph ysical fea tures. A person who was in a feminine occupation was more likely to have female traits, female physical features, and to occupy traditional female roles. There are two potentially important implications of these results. First, it is likely th at thes e relation ships a re held to gether b y theor ies abo ut gen der alth ough it is possib le that they are based merely on empirical observa tion. Second, people seem to be willing and able to “move ab out” freely within their stereotypes. If I know a male looks m asculine, then I am less likely to infer tha t he is ho mose xual o r to be in a tradition ally fema le occu pation . This is th e powe r of implic it person ality theo ries – they allow u s to infer m any s tereotyp e features from th e few w e observ e. Of course, many of these relationships are likely to be “stereotype-free” in the sense that exist independently of a given group or stereotype. For example, it is likely that we think kind people are also warm whether th e person is bla ck or wh ite, male or fema le, professor or ph ysician, Texa n or Ho osier. However, it is also possible that some perceived trait relationships change from group to group. variables are variables that change the pattern of correlations or relationships among other variables. It is possible that perceived trait relationships may differ for different groups so that group designation operates as a kind of variable. Perhaps white subjects may assume that musical ability implies athletic ability for blacks but n ot for whites. Perhaps for m any a Jew’s w ealth

12

Fiedler’s (2000) BIAS model cn for both within the same framework, but the model is too complex to go into h ere. Chapter 5: Structure and IPT / Version of November 10, 2002

implies clannishness but not for an Hispanic. Professors who are lazy may be seen as hostile, but these two tra its may be seen a s nega tively relate d for oth er grou ps suc h as a ttorney s or bu siness e xecutiv es. I am suggesting that stereotypes about groups are not merely lists of traits associated with those groups but sp ecific relatio nship s amo ng tho se traits. Koltov (1962) demonstrated that trait relationships could be affected by type of stimulus person. Specifically she found tha t correlations amon g traits were lower when people rated people they knew well tha n wh en they rated peo ple they k new les s well. Th is does not, of c ourse , say an ything directly abou t stereoty pes, bu t it does a t least rais e the pos sibility tha t trait relatio ns mig ht var y by gro up. Secord and Berscheid (1963) were the first to study this issue directly with stereotypes and groups. They asked subjects to indicate how likely it was that two traits went together. First this was done with no group designation, and then subjects did the task with the added information that the person they were rating was Negro. Secord and Berscheid assumed that in the no group designation condition subjects would assume that the person was white. There was a slight tendency for black stereotype traits to be more closely related for the Negro than for the putative white stimulus person, but this result was not strong. Subsequent studies have produced more promising results. Veness (1969) reported that correlations am ong traits differed for different occupationa l groups. Han no and Jo nes (1973) found that the multidimensional scaling of trait co-occurrences were different for family doctor and la wyer. In the most explicit test of this idea Schn eider and Blankm eyer (1983) used a priming task to get subjects to think about introversion or extroversion. They then were asked to judge how closely various traits implied other traits. As expected, when introversion was primed, introverted traits were seen as more closely related than w hen introversion w as not primed a nd when extroversion was primed extroverted traits were seen as more closely related. In a second experiment similar effects were found using a less obtrusive priming task and priming of mature and immature with closer relationships found for immature behaviors when immaturity was primed and mutatis mutandis for maturity. In related work, Linville, Fischer, and Yoon (1996) showed that people assumed stronger relationships among features for out-groups than for in-groups. There may be many reasons for this kind of effect. Thinking about trait relationships within a given context may produce a biased sampling of exemplars. For example, because athletic and musical are both part of the black s tereotype it may be easier for subjects to think o f people who hav e both traits when imagining blacks, at least those that fit the prototype. It is also possible that trait change meanings within different stereotypes. So for a white stimulus person musical and athletic may have connotations of playing the cello and soccer whereas for blacks the images may be more akin to singing rap songs and playing football. How many cello-playing soccer players have you seen? How many football players doing rap performances? Wha tever the reason s this kin d of effect c ould h ave po werful im plication s for stereo types. O ne is that stereotypic behaviors may trigger stereotypic traits only when the stereotype itself has been primed. Seeing a random person enjoying an opera performance might not trigger an inference that the person also likes the singing of Judy Garland or Barbra Streisand unless the gay male stereotype is active (these all being parts of the gay stereotype). Mem bers of stigmatized grou ps often do report that th ey feel they have to be especially careful in exhibiting stereotypic behaviors less other related behaviors be inferred. One black professional of my acquaintance worries some about ordering barbecue when he eats in the presence of whites with a slight fear that they might infer that he also is lazy and will break into a rap song and another said somewhat jokingly that when he bought a watermelon at the grocery he felt that he would start tap-dancing. Because none of this is a part of the white stereotype, whites would not have to worry about such inferences. As another example, people might assume that generally intelligence and happiness are correlated po sitively, but have the sense tha t they are negatively correlated for career women whose intelligence drives them to succeed aga inst real discrimination that co nstantly dash es hopes and in hibits en joymen t of hob bies an d fam ily.


Perceived Relations among Groups Finally, within an implicit personality theory paradigm groups as well as traits may be seen as more or less similar to one another. Thus, most Americans would presumably see Chinese and Japanese as far more similar than Mexicans and Swedes, despite the strong protests of the former two groups which actually do have radically different cultures in many ways. Such perceived similarities might reflect nothing more than geographical similarities or skin color, but whatever similarities we find might well reveal some basic dimensions of perceiving people in other groups. For example, a perceiver who clearly sees a black professional female as closer to a white male professional than to a lower class black female is, indirectly, saying that class is mo re important than either gender or race. This fact does n ot per se tell us much about the content of stereotypes, but we would surely expect to find that there would be more overlap in the traits ass igned to similar rather than dissimilar groups. W e might even further expect that those traits that are most similar for similar groups and dissimilar for dissimilar groups would be quite central to the groups involved. For example, imagine a perceiver who sees professors as similar to lawyer s and dissim ilar to plu mbers . Furth er imag ine tha t he sees p rofesso rs as like la wyers in that both smart and arrogant as opposed to plumbers who are seen as neither. Our perceiver might also say that plum bers and professors a re alike in being relatively poor. From all this we migh t be tempted to conc lude th at ou r perceiver th inks th at intelligen ce and arroga nce are m ore impo rtant th an w ealth in of thinking ab out people. No pu blished stu dies hav e come clo se to usin g the full strategy suggested here. How ever, there have been a fe w stu dies of ra tings o f similarities amo ng gro ups. R iffenbu rgh (19 66) ha d vario us eth nic group s in Ha waii ra ted on several tr aits an d then calcu lated sim ilarities am ong g roups . He mer ely pointed out that some groups (e.g., Chinese and Japanese) were seen as closer than others (e.g., Chinese and Ha waiian). In a m ore elaborate study Fu nk, Horow itz, Lipshitz, and Yo ung (1976 ) obtained similarity judgm ents for variou s natio nalities a nd eth nic gro ups, a nd fou nd th at thes e similarities could be represented in a two d imensional spac e. Within that space th ere were three basic clusters: white (Italian, Irish, Jew , Polish, An glo, and G erman), "co lored" (Negro , black, Pu erto Rican , Indian, M exican), and A sian (Ja panes e, Chin ese). Sever al rating scales w ere placed throu gh this space. F or exam ple, a dimen sion o f activism placed d ark skin ned peo ple at on e end w ith Asia n at th e other en d and whites in the middle. The whites were seen as intelligent as contrasted with Asians and dark skinned people. In a larger study Jones an d Ashm ore (1973) foun d they needed three dim ensions to represent similarities among national and ethnic groups. Aga in several dimension s fit the space nicely. Domina nt-subordina te and communist/non-communist were essentially orthogonal to one another with a dimension of western /no n-wes tern cu lture fallin g in betw een.


Implicit Personality Theory 245g3w

Overview 26281t

More details 6y5l6z

Related Documents 3h463d

Implicit Personality Theory 245g3w

Adlerian Theory Of Personality 131t2n

Bird Theory-personality Test 2d1w3d

Rotter's Theory Of Personality 6y3ws

Eysenck Personality Theory Pdf 1p4u2g

The Grand Theory Of Personality 3v2151

More Documents from "Bobo" 653u65

The Definitive Guide To Social Media Marketing Marketo 402m2b

Implicit Personality Theory 245g3w

[giorgio Agamben] Gusto(bookzz.org) 1w4v4l

Afro Blue - Robert Glasper 29497

Programa Cursuri Instruire Operatori Cnc - Curs Operatori Frezare Fanuc Rev 3 5w1s68

Aspekte_neu_b2_erster_teil.pdf 4r2429