Implicit Personality Theory

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 support
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 admissions policies at many colleges and universities.

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
members 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 terms 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 members 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.
Support for the Idealist Position: Much of the support 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 support 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 members 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.

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 terms 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 support 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 terms 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 two-
dimensional 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.

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 joined together to former larger scale units. The warm-kind-
generous-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 passive 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 joins

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 terms 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 moderator 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 terms 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 membership 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 membership 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
members 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 , passions?-- 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?"

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 affiliated 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 .

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
accounted 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 accounted 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 lower-
class 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 gle-
moth 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 take-
charge, 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 account 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 account 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.

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

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 (eggs-
baco 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 members 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 members 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 membership 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
remember 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
Desira ble
Presented 18 9
Recognized 17.52 9.48
Estimated 17.09 7.27
Unde sirable
Presented 9 4
Recognized 5.79 6.21
Estimated 8.91 5.73
Evaluation
Desirable Traits 6.91 6.15
Undesirable Traits 4.39 5.31
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 on-
line 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 remember 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 remembered 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 members 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 members 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 Account: 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 remember the distinctive behaviors attached to the distinctive
groups. In support, 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 assigning 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 account 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 terms of well-
know 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.

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 terms and fewer positive terms. 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 support 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 support 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 support 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 account 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 support or ad d to the effect.
And it may be that in the laboratory as well as the outside world, different theories account 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 salesman-
talkative 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 Garcia-
Marques 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 Moderator 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 gender-
linked 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.
Moderator 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 moderator 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 account for both within the same framework, but the model is too
complex to go into h ere.

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 terms 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
terms 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

Uploaded by

Copyright:

Available Formats

Implicit Personality Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Implicit Personality Theory

Uploaded by

Copyright:

Available Formats

Chapter 5: Stereotype Structure and Implicit

The Nature of Implicit Personality Theory

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

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

Good Driving Poor Driving

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

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

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

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

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

Are Stereotypes Like Implicit Personality Theories?

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

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

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

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

Five-Letter Anagrams Seven-Letter Anagrams All Anagrams

Success Failure Success Failure Success Failure

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

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

Table 5.4: Traits presented and Rated for Groups A and B.

Recognized 17.52 9.48

Estimated 17.09 7.27

Recognized 5.79 6.21

Estimated 8.91 5.73

Desirable Traits 6.91 6.15

Undesirable Traits 4.39 5.31

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

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

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

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

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

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

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

You might also like