Can Race be Erased?

Coalitional Computation and Social Categorization

Author(s): Robert Kurzban, John Tooby and Leda Cosmides
Source: Proceedings of the National Academy of Sciences of the United States of America,
Vol. 98, No. 26 (Dec. 18, 2001), pp. 15387-15392
Published by: National Academy of Sciences
Stable URL: http://www.jstor.org/stable/3057466
Accessed: 25-10-2017 18:57 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/3057466?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

National Academy of Sciences is collaborating with JSTOR to digitize, preserve and extend
access to Proceedings of the National Academy of Sciences of the United States of
America

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
 Can race be erased? Coalitiol ial computation and
social categorization
Robert Kurzbant, John Tooby, and Leda Cosmides
)6-2070
Center for Evolutionary Psychology, University of California, Santa Barbara, CA 931(

200112,
Communicated by Roger N. Shepard, Stanford University, Stanford, CA, October (received for review May 28, 2001)

Previous studies have established that people encode the race of ag

:, "race" is a very implausible candidate for a conceptual
each individual they encounter, and do so via computational pr mitive to have been built into our evolved cognitive machin-
processes that appear to be both automatic and mandatory. If true, er
I. During our evolutionary history, our ancestors would have
this conclusion would be important, because categorizing otherslabited in] a social world in which registering the sex and life-
by their race is a precondition for treating them differently accord- toryhi'stage of an individual would have enabled a large variety
ing to race. Here we report experiments, using unobtrusive mea- of probabilistic inferences about that individual. In con-
useful
sures, showing that categorizing individuals by race is not inevi- st, tr
ancestral hunter-gatherers traveled primarily by foot and,
table, and supporting an alternative hypothesis: that encoding by co
nsequently, residential moves of greater than 40 miles would
race is instead a reversible byproduct of cognitive machinery that ve ha
been rare (16). Given the breeding structure inherent in
evolved to detect coalitional alliances. The results show that su :h a world, the typical individual would almost never have
subjects encode coalitional affiliations as a normal part of person
countered en
people sampled from populations genetically dis-
representation. More importantly, when cues of coalitional it enough affili- ta;as belonging to a different "race" (even
to qualify
ation no longer track or correspond to race, subjects markedly
;uming that suchas;
a term is applicable to a nonpolytypic species
reduce the extent to which they categorize others :h byasrace,
humans,andinsuwhich the overwhelming preponderance of
indeed may cease doing so entirely. Despite a lifetime's netic experience ge
variation is within population and not between popula-
of race as a predictor of social alliance, less than 4 min n, of
andexposure
at most geographically
tic graded rather than sharply
to an alternate social world was enough to deflate the unded) (17, 18).toIfbo
tendency individuals typically would not have
categorize by race. These results suggest that racism may be a en countered members of other races, then there could have been
volatile and eradicable construct that persists only so long as it is no
selection for cognitive adaptations designed to preferentially
actively maintained through being linked to parallel systems of en code such a dimension, much less encode it in an automatic
social alliance, an d mandatory fashion.
Accordingly, we propose that no part o
Throughout our species' his :hitecture is designed specifically to en
on the ze that the (apparently)
categorization automatic
of th an
When this :e is insteadoccurs
divide a byproduct of adaptatio
alon
its malignanternative function that was a regular p
consequences
Indeed, raging ancestors:
ingroup detecting coalitions
favoritism
hostility therers lived
appears to in exist
bands, and neighboring
in a
of categorizing o conflict with one another (19-21).
individuals
humans to discriminatealitions and alliances withinin bands f(2
outgroup in both ated primate species and likely to be far
allocation o
(2-7). Following minid lineage
on itself (23). To negoti
historica
studies have ccessfully, and to anticipatethat
confirmed the likely
ernative courses of action, our ancestor
elicit: people discriminate ag
assigned to groups being equipped with neurocognitive m
tempora
menter who uses :se shifting alliances. Computational m
dimensions
social significance,
tect coalitionsandand alliances in the an
rando
istics of the signed,
individualsshould be sensitive to two fa
ass
people into groups ordinated action, cooperation, nea
along and com
tion, it wouldit predict-either be purposefully or inci
discouragin
designed such l's political allegiances
that people (24-26). Like
canot
race. This would imply that racism is intractable. it reveal coalitional dispositions are usu
- frequently unavailable for inspection
Yet it has been claimed, with considerable empirical support, ar
>ns relevant to coalitional affiliation need to be made. Ac-
that encountering a new individual activates three "primitive" or si(
"primary" (9-12) dimensions-race, sex, and age-which the co rdingly, alliance-tracking machinery should be designed to
te these rare revelatory behaviors when they occur, and then
mind encodes in an automatic and mandatory fashion (i.e., nC
e them to isolate further cues that happen to correlate with
across all social contexts and with equal strength; refs. 10-15). US
These dimensions can be encoded without other individuating co alition but that are more continuously present and perceptu-
information; e.g., one might recall that one's new neighbor is a y a easier to assay. Such cue mapping allows one to use the
havior of some individuals to predict what others are likely to
young, white woman, without remembering anything else about
her (11-15). Over the last two decades, considerable effort has. Because
a this circuitry detects correspondences between
egiance and appearance, stable dimensions of shared appear-
been expended on the search to find conditions under which race
is not encoded, so far without success (14, 15).
Despite the evidence in favor of these claims, an evolutionary tTc
whom reprint requests should be addressed. E-mail: rkurzban@ucla.edu.
analysis indicates that one of them is likely to be wrong. Although Thi
publication costs of this article were defrayed in part by page charge payment. This
selection would plausibly have favored neurocomputational art cle must therefore be hereby marked "advertisement" in accordance with 18 U.S.C.
machinery that automatically encodes an individual's sex and ?1; 34 solely to indicate this fact.

www.pnas.org/cgi/doi/1 0.1 073/pnas.251 541498 PNAS | December 18, 2001 | vol. 98 I no. 26 I 15387-15392

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
ance-which may be otherwise meaningless-emerge in the sta nding coalitional categorizations in favor of novel ones. But
cognitive system as markers of social categories. Coalitional if our cognitive adaptations are well engineered by selection,
computation increases their subsequent perceptual salience and thi -n they might be able to detectably alter such categorizations
encodes them at higher rates. Any readily observable feature- ev fn within the brief periods available during an experiment.
however arbitrary-can acquire social significance and cognitive Finally, because sex is hypothesized to be a true and stable
efficacy when it validly cues patterns of alliance. Ethnographi- pr mary dimension of person representation similar to but
cally well-known examples include dress, dialect, manner, gait, inc lependent of age and coalition, we predict that:
family resemblance, and ethnic and coalitional badges. Pr( Wdiction 5. Sex will be encoded far more strongly than race, even
Adaptations for coalition detection are hypothesized to re- in contexts in which it is irrelevant to coalition and task.
semble adaptations designed to encode sex and life-history stage Pri diction 6. The encoding of sex, an independent dimension, will
in some respects, but not others. Ancestrally, an individual's sex not diminish coalition encoding, even though sex will be
was fixed for life, but his or her coalitional membership was not. encoded far more strongly than race.
New patterns of alliance typically emerge whenever new issues The impact of the encoding of sex on coalition is introduced
arise whose possible resolutions differentially impact new sub- to allow the evaluation of a plausible counterhypothesis: that
sets of the local social world. Consequently, coalitions changed atl ention is zero sum, and so a negative relationship between any
over time and varied in internal cohesion, duration, and surface tw o dimensions of encoding shows only that attention is finite.
cues. To track these changes, cue validities would need to be Di .monstrating that sex encoding does not diminish coalition
computed and revised dynamically: no single coalitional cue en coding eliminates attentional constraints as an alternative
(including cues to race) should be uniformly encoded across all ex planation for any race reduction effect (prediction 4). If
contexts. Furthermore, arbitrary cues-such as skin color- co alition is encoded just as strongly when the competing dimen-
should pick up significance only insofar as they acquire predic- sic n is sex-which is encoded more strongly than race-then no
tive validity for coalitional membership (26). do main-general capacity limit is preventing race from being
In societies that are not completely racially integrated, coded as strongly
shared en as sex. If no ceiling on a domain-general
appearance-a highly visible and always present cue-may entional be att has been reached, then attentional constraints
system
correlated with patterns of association, cooperation, andmot explain ca
competi- any reduction in race encoding observed in this
tion (26). Under these conditions, coalition detectors may radigm.
perceiveEliminating
pa this alternative would reinforce the in-
(or misperceive) race-based social alliances, and the mindpretation will mapthat te] race encoding is diminished because race is
race onto the cognitive variable coalition. According to rely thisa hy-proxymifor the deeper mental category, coalition-and
pothesis, race encoding is not automatic and mandatory. t forIt some
ap-incidental
no reason.
peared that way only because the relevant research was conducted We present results supporting these six predictions below.
in certain social environments where the construct of race hap-
.thods
pened, for historical reasons (27), to be one valid probabilistic cue M
to a different underlying variable-one that the mind was designed
assess Tcused a method standard in the literature, the
encoding, we
to automatically seek out-coalitional affiliation (24-26). m( protocol developed by Taylor et al. (13). It uses
:mory confusion
er ?ors in recall to unobtrusively reveal whether subjects are cate-
Predictions. go rizing target individuals into groups and, if so, what dimensions
Because coalitional
-y are using to do soallegiance
(see Appendix). Subjects are asked to form is
cognitive variable pressions of individuals whom theythe
that will see engaged inmind
a con-
race, we predict the following. ve :sation. They then see a sequence of sentences, each of which is
pa ired with a photo of the individual who said it. Afterward, there
Prediction 1. Race per se will not, after all, have the properties is i surprise recall task: the sentences appear in random order, and
previously ascribed to it, i.e., it will not be encoded across all su bjects must attribute each to the correct individual. Misattribu-
ns reveal encoding: subjects more readily confuse individuals
social contexts and with equal strength. Manipulating coali- tic
tional variables should introduce the long sought-after vari- wt
Lom they have encoded as members of the same category than
ability in its encoding. th )se whom they have categorized as members of different eate-
Prediction 2. Shared visual appearance is who
ries. For example, a citizen of Verona not necessary
had encoded coalition fo
coalition encoding. People will encode
-mbership coalitional
would make more alliances
within-category errors-errors in as
part of person representation rich evens/he confused,
when say, a Capulet with a Capuletis
there (or a no
Montague similarit
of appearance within a coalition. wi th a Montague)-than between-category errors-ones in which
ie confused
That is, coalitional categories are not built on a Capulet with a of
a foundation Montague
s/ or vice versa (this
shared appearance and can readily exist ationship
without will hold it. Shared
for data re to equalize base
that are corrected
appearance will acquire (or lose) significance:es;only
see Appendix).
to the extent ra
Experiments
that it provides a cue to this underlying political 1 and 2 were designed to test predictions 1-4
structure.
However, a correlation between shared ove.appearance
The first step was toand
test theab
prediction that subjects'
behavioral cues of alliance can reinforce and stabilize a coali- mi nds could be caused to treat the dimension coalition in a
;hion
tional categorization, leading to a counterintuitive prediction. fa: parallel to sex, age, and especially race, i.e., to see
Prediction 3. Arbitrary cues other than racial appearance can lether
be wt subjects would encode coalition incidentally as part of
endowed with the same properties that race has previously in
pression formation, without having been instructed to do so
exhibited by linking them to coalition membership. (p redictions 2 and 3). A widely held counterhypothesis is that
Even more significantly, if race is merely a proxy for sir nilarity in appearance is the origin and foundation of social
coalition-a cue used to infer a person's alliances-then it is ca
tegorizations such as sex, age, and race, which only then
predicted that: ac quire social significance. Because our view is that coalition
Prediction 4. The strength coding can be elicitedof even inrace
the absence encodi
of shared visual
creating a social context pearance (predictionin2), there were no visual cues(i)
which to coalition
ra
cue to coalition, and :mbership (ii)
in experiment
there 1. Each speaker (represented
are alt by a
reliably indicate coalitional affiliation. pi oto) was a young man, and all were identically dressed. The
There are no a priori grounds for predicting exactly ly way how
to infer thethe
fast coalitional
or allegiances of these individuals
human mind should be able to discard ontogenetically long- ws LS from the content and sequence of their utterances: speech

15388 I www.pnas.org/cgi/doi/10.1073/pnas.251541498 Kurzban et a/.

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
acts implying self-inclusion, allegiance to, exclusion from, or co; ilition is not, one might expect race to overwhelm coalition
enmity toward one or another of two contending groups (here- en( :oding across contexts.
after, verbal allegiance cues-e.g., "you were the ones that started ] Even if our hypothesis were correct, for the effects to be
the fight"). From these cues, four of the eight speakers could be ob ;ervable, the machinery would have to be very efficient at
categorized as belonging to one coalition, and four to the other. up dating and tracking even short-term changes in alliance,
Thus, in experiment 1, nothing in the appearance of the individ- dyi lamically recomputing cue validities associated with newly
uals offered the potential for subjects to correctly sort them into two rel evant coalitions, even with very limited exposure. Otherwise,
coalitions (a fact logically guaranteed by the experimental design: a f ew minutes in an experiment in which race was not predictive
for each subject, the computer randomly generated a new assign- of coalition could not detectably impact the accumulated effect
ment of photographs to coalition). This provided a very stringent of years of exposure to a world in which race mattered.
test of coalition encoding, methodologically adverse to our hypoth-
esis. Categories such as race and sex are on display in the appear- Re sults
ance of the individual both at the time of encoding and during the rhe results of experiment 1 were as follows:
recall task (facilitating the use and acquisition of these two cate- i) Even though race existed as a competing and visible
gories). But coalitional membership was not determinable from dir nension, subjects did indeed encode a new dimension, coali-
appearance at any time during experiment 1, and no cues to tio n membership, doing so solely on the basis of verbal cues of
coalition membership-visual or verbal-were present during the i plied affiliation (confirming prediction 2): they made more
recall task. These handicaps would necessarily weaken the detect- wi hin than between coalition errors [n = 55, mean difference,
ability of coalitional categorization, compared with race and sex, M = 0.90 (SD = 2.76), t*(54) = 2.41, P = 0.0096; see Appendix,
making its spontaneous extraction and use, if found, all of the more Nc te 1].
remarkable.
ii) Consistent with previous findings, subjects also encoded
According to our central hypothesis, two factors-patterns oformation
inf about the race of targets, forming social categories on
alliance and shared appearance-conspire to reinforce and sthi basis as well [M = 2.30 (SD = 2.76), t* (54) = 6.71, P < 6.2 x
stabilize a coalitional categorization along racial lines. The same-9].10
theory implies, however, that other coalitional categorizations iii) In experiment 1, where verbal allegiance cues were the
will be encoded in a race-like manner if the second factor- on
ly basis for inferring coalition membership, and coalition was
shared appearance-is present and tracks behavioral cues of vis the effect of race was twice as large as the
ually undetectable,
coalitional alliance (prediction 3). This should be true even when
ect of coalitionef
(effect size, r, for coalition versus race: 0.31
shared appearance is created by using a visual cue that is an
d 0.67, respectively).
arbitrary and impermanent. Prior results with the memory Experiment 2 was identical to experiment 1, with one excep-
confusion protocol show that differences in shirt color are not
n: cues to coalitional affiliation were amplified by giving each
spontaneously encoded when they have no social significance tc
alition its own uniform color, either gray or yellow, providing
(15), so this was a good candidate for an arbitrary cue. Thus,
isible marker of common appearance to the coalition.
experiment 2 was identical to experiment 1, except we added this
(iv) As predicted, when coalition membership became marked
arbitrary feature of shared appearance to the coalitions that we
cues of shared appearance, the degree to which subjects
had manufactured through verbal allegiance cues.en We used b
coded it increased substantially [n = 52, M = 4.62 (SD = 3.62),
image manipulation software to give the individuals either gray
51) = 9.20, P = 1.02 x 10-17]. The size of the coalition effect
or yellow shirts; individuals whose verbal statements implied i
experiment 2 was more than 2.5 times larger than in experi-
allegiance to the same coalition wore the same color. Thus, ml :nt 1, which used only allegiance cues (0.79 versus 0.31). The
subjects in experiment 2 could infer coalition membership from mi
:rease in the extent to which subjects encoded coalitional
verbal allegiance cues, from cues of shared appearance (shirt ll
iance was both large and significant [experiment 1 versus
color), or both, because they tracked each other. If prediction 3
is correct, then coalition will be spontaneously encoded in ex periment 2: t(95.4) = -5.945, r = 0.52, P = 2.3 x 10-8].
This indicates that endowing coalition with the attribute of
experiment 2 just as strongly-or more so-than race is. (Strength
ibility by an arbitrary cue was an effective manipulation for
of encoding is indexed by effect size, r; ref. 28.) vli
An equally important goal shaping the design of these exper- in :reasing categorization of targets by coalition. More impor-
iments was to observe how the creation of an independent system ta itly, it confirms prediction 3: it shows that a new and arbitrary
of coalitional categorization would affect racial encoding (pre- co alition can be encoded just as strongly as race is. Indeed, this
dictions 1 and 4). The experimental context implied a conflict ur derstates the result: the arbitrary coalition in experiment 2 was
between two rival coalitions, but these were constructed such en coded even more strongly than race was at its strongest [0.79
that race was uncorrelated with coalition membership: the rival vs 0.67) in experiment 1 (t(91) = 3.81, r = 0.37, P = 0.00013).
four-person coalitions were each composed of two Euro- This observation allows us to address two key questions: when
American men and two African-American men. This design ra :e ceases to be a predictor of coalitional allegiance within a
allowed us first (experiment 1) to replicate the basic phenom- gi 'en social context, and when coalition membership is marked
enon of racial encoding and, second (experiment 2), to see by cues of shared appearance (as race is), does coalition acquire
whether the tendency to encode race could in fact be reduced by th robust properties race had, and does race lose the strength
the introduction of a visually accessible cue to a nonracial it once had? These predictions (nos. 1 and 4) were confirmed.
coalitional division (thus testing prediction 4). If the human (v) In experiment 2, when cues of coalitional affiliation were
brain contains neurocomputational machinery for tracking co- an iplified through shared appearance, subjects did continue to
alitional alliances, then constructing a new social environment in ca :egorize on the basis of race [M = 1.40 (SD = 2.51), t*(51) =
which coalition is uncorrelated with race should weaken the 4.( )38, P = 0.000091], but the size of the race effect was diminished,
preexisting weight given to race as a cue to coalition within that)mfr(0.67 in experiment 1, where coalition cues were subtle, to 0.49
context. The more obvious and relevant the alternative coali- in experiment 2, where they were amplified. This reduction in the
tional division, the less race should be encoded. But if priorcoding en of race was significant and substantial [experiment 1
claims are correct-if race is a primary dimension of person ve rsus experiment 2: t(105) = 1.83, r = 0.18, P = 0.035].
representation, encoded with equal strength across all social (vi) When an alternative, and contextually relevant, coalitional
contexts-then constructing this new social environment shouldtegorization
ca was reinforced through shared appearance in
make no difference. Indeed, if race is a primary dimension and ex periment 2, coalition was encoded far more strongly than race:

Kurzban et al. PNAS | December 18, 2001 | vol. 98 I no. 26 I 15389

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
r = 0.79 for coalition, 0.49 for race [t*(51) = 6.14, r = 0.65, P = a. coalition b. coalition
6.3 x 10-8]. The large effect size (0.65) associated with this
race sex
difference shows that when race is irrelevant to the coalitional
1.8
conflict at hand, the encoding machinery tracks coalition mem-
bership much more carefully than race. 1. .6 .1
This result demonstrates that social context-in this case, one
.4
in which an important coalitional dimension does not track or .
correspond to race-can diminish the extent to which race is
encoded. So contrary to prior claims, race is not inevitably ; 1.2l
encoded with equal strength across social contexts. In a social
world where the active coalitions are easy to encode and do not ,
; 0.8 -
track race-even briefly-encoding by race decreases. (
These findings indicate that subjects bring the tendency to
categorize by race with them into the experiment, but then begin to X
i S 0.6 -
lose it as the circuitry detects that it no longer predicts relevant 0.4.
coalitions in this context. In both experiments 1 and 2, race was
uncorrelated with coalition, but because coalition is far easier to
o 0.2
coalition cues:
perceive in experiment 2 during both encoding and retrieval, the
irrelevance of race to coalition is also far easier to perceive in that O verbal only
experiment.
* + shared appearance
It bears underlining that the racial appearance of the targets
remained exactly the same in both experiments, so that if the
tendency to confuse two individuals of the same race was based Fig 1. (a) Relative importance of coalition versus
ect sizes index how strongly subjects categori
on their being perceptually similar, rather than being categorized Eff tio
n and race were given equal weight, then t
abstractly as belonging to the same alition/race)
race, then would
amplifying com- (ct
be =1.0. It would be <1.0
peting coalitional cues should have 3vily
had no effect. This means he
than coalition, and >1.0 if coalition wer
that racial encoding is cued by similarity of appearance,
e. In experiment but
1, where the does
only rac
cues to coal
not consist of similarity of appearance.
; ratioThis supports
was 0.46, thethat
indicating view thi were
subjects
that appearance cues-including race-are picked
alition = 0.31, up or
race = 0.67). Butdis- (cc
in experiment
plified
carded to the extent that they predict by the addition
coalition, and notof shared appearance
because ar c
they are intrinsically salient. wa s 1.61 (coalition = 0.79, race = 0.49), indicatin
:oded more strongly than race. The same is not
(vii) Lastly, as cues to coalition membership were amplified en
ative importance of coalition versus sex. When
(experiment 1 versus experiment 2), )jects
the relative importance of sut
continued to strongly categorize on the b
race versus coalition reversed itself. When coalition cues were an(
a
j coalition membership were weighted about
not based on appearance, race remained more important than
the newly manufactured coalition; however, when they were
amplified by means of an arbitrary shared appearancec cue, se:
remained very large (0.84). In sharp
coalition became more important than race. be tween coalition and race in experim
Fig. 1 depicts the extent to which these ratios deviate from the
coded en
more strongly than coalition
1.0 ratio that equal weighting would produce. Moreover, the effect sizes for sex were
There is nothing mathematically necessary about )se this flip.
for th race in the analogous conditi
Even if amplifying coalition cues increases the extent nto5):
which tic for sex versus 0.67 for
0.91
subjects index a target's coalitional affiliation, it need not do socues
egiance all were present (experimen
to the point where coalition becomes more important than
i6, r race. 3.(
= 0.35, P = 0.00013), and 0.84 for
The results of two control experiments demonstrate this wI
ten coalitional cues were amplified by
conceptual point. Experiments 3 and 4 were identical to exper-
rs. 4
experiment 2: Z = 3.53, r = 0.33
iments 1 and 2, with one exception: Instead of varying the race di:
'ference is supported by t tests com
of targets, we varied their sex. Unlike race, sex periment
is a good ex 3 vs. 1: t(106) = 5.37, r
candidate for a primary dimension of person representation periment that ex4 vs. 2: t(106) = 5.10, r =
our minds evolved to encode across most if not all situations. Fi g. 2. The hypothesis that sex is a t
Subjects categorized by coalition in Experiments 3 and 4, dii
nension-encoded in an automatic an
replicating the coalition results from Experiments 1 and 2. ry When to
fashion-is supported by these res
cues to coalition had to be inferred from utterances alone, K thearese: very large, and they stay large
effect size for coalition was 0.35 (comparable to 0.31 for the ar
: amplified.
analogous experiment 1; [experiment 3: n = 55, M = 0.98 (SD =
Although there is nothing about the protocol that prevents
2.69), t*(54) = 2.709, P = 0.0045]; when cues were amplified, the
bjects fromsu encoding a number of different dimensions at
effect size for coalition was 0.81 [comparable to 0.79 in exper- on
ce, one might reasonably hypothesize that there exists a
iment 2; experiment 4: n = 57, M = 4.29 (SD = 3.16), t*(56) = dc attentional constraint creating an inevitable
main-general
10.241, P = 9.6 x 10-15]. In contrast to the race results, thebetween
Ldeoff trn the degree to which different dimensions are
extent to which subjects categorized by the targets' sex was very
coded. On en
this view, the inverse relationship between race and
high in both experiments: effect sizes were 0.91 and 0.84alition for does
co not imply any special relationship between race
experiments 3 and 4, respectively [experiment 3: M = 4.75 d coalition:
(SD = an any increase in one channel is expected to be
2.23), t*(54) = 15.81, P = 3.2 X 10-22; experiment 4: M mpensated = 3.87 cc for by decreased encoding in another. These data
(SD = 2.53), t*(56) = 11.52, P = 1.1 x 10-16]. Although there sify suchfala view. Tellingly, sex was encoded far more strongly
was a diminution in the extent to which sex was encoded when th in race (in experiments 3 and 4 versus 1 and 2 and 5 and 6),
coalitional cues were amplified [from 0.91 to 0.84; t(110)= t1.94, this did
yenot lead to any reduced encoding of coalition, as a
r = 0.18, P = 0.027], the extent to which subjects categorized ideoffbyview
tr;would predict. In fact, the strength of coalition

15390 | www.pnas.org/cgi/doi/1 0. 1073/pnas.251541498 Kurzban et al.

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
 Race & Coalition Sex & Coalition sut
,jects to encode the race of targets [n = 52, M = 0.37 (SD =
5), t*(51) = 1.07, P = 0.29]. In this condition, it would appear
(Exp & (Exps 3 & 4) th2 t the extent to which subjects encoded targets by their race
0.9 wa s not merely diminished, it was erased.
0.8 - Co nclusions
0.7- Wsl iat is most striking about these results is just how easy it was
0.4~~~~~~~~~~6
0.6 - ~ ~~~to
es
diminish the importance of race by manipulating coalition-
,ecially given the repeated failure over decades to find other
m 0.5----- me
ans to influence racial encoding. The sensitivity of race to
0.4 ilitional manipulation lends credence to the hypothesis that,
o
t~~~f:~~~~~~ ~~Ato the human mind, race is simply one historically contingent
0.3 sul )type of coalition. Our subjects had experienced a lifetime in
0.2 wr ich ethnicity (including race) was an ecologically valid pre-
--sGex |dic tor of people's social alliances and coalitional affiliations. Yet
0.1 ---- coalition | s than 4 min of exposure to an alternative social world in which
0 -ra e was irrelevant to the prevailing system of alliance caused a
verbal + shared verbal + shared dr; imatic decrease in the extent to which they categorized others
only appearance only appearance by race. This implies that coalition, and hence race, is a volatile,
dy lamically updated cognitive variable, easily overwritten by
cues to coalition membership ne w circumstances. If the same processes govern categorization
ou tside the laboratory, then the prospects for reducing or even
Fig. 2. Strength of categorization by coalition, race, and ninating the widespread
sex. Coalition was eli tendency to categorize persons by
encoded even in the absence of shared appearance cues. Sex was always
:e may be very rac
good indeed.
encoded more strongly than race. When coalition membership was amplified
by the addition of shared appearance cues (which are always pendix
present for race Ap
and sex), then coalition was encoded as strongly as sex, and bjects.
far more strongly
All subjects SUi
were undergraduates at the University of
than race. The data suggest that coalition and sex are primary dimensions of Ca
lifornia, Santa Barbara, average age = 19. In each experiment,
person representation, whereas race is not. th
sex ratio of subjects was 50:50. Each experiment tested
in( lividuals who had not participated in any of the other experi-

encoding increased slightly in both cases, from 0.31 to 0.35 mi ;nts. Subjects were primarily Euro- (including Hispanic) and
ian-American.
[verbal only: experiment 1 (race) versus experiment 3 (sex)], and A
from 0.79 to 0.81 [amplified coalition cues: experiment 2 (race)
versus experiment 4 (sex)]. This means there was attentional otographic Stimuli. Ea
[or photo of the hea
capacity to spare in the race experiments, eliminating attentional co
man wearing a plain bas
constraint as a credible explanation for the diminution in race WC
all individuals were wea
encoding observed in experiment 2. d
:ntifying marks. In exp
The results for sex differ in another important way from those d
different co seys of a
for race. In the race experiments, the relative importance of j
lobe PHOTOSHOP; Adob
coalition and race sharply reversed when coalition cues were th
: target photos were h
amplified. This did not happen in the sex experiments. Although Ek
periments 5 and 6 used
the importance of coalition increased in experiment 4, approach-
ing the high levels of encoding accorded to sex, it did notConfusion
!mory outstrip M( Protoco
sex (coalition to sex ratio: 0.81 to 0.84 = 0.96). zing(SeeaFig. 1 and se of photog
series
Appendix.) In other words, amplifying coalition cues does
ired with not a sentence u
cause coalition to be encoded more strongly qu
than sex. displayed
entially But p on a
amplifying coalition cues does cause coalition to be encoded in' lividual belongs to one
more strongly than race: the relative importance of race de- be en in a fight during t
creased dramatically as cues to coalition membership became w re uttered in the cont
more obvious-indeed, the coalition effect was >60% larger th fir task is to form an
than the race effect. This is what one would expect if race were th .y are viewed. Each sub
a proxy for coalition, but sex were not. This can be seen even ea ch sentence was paired
more strikingly in experiments 5 and 6. m in uttered a total of th
To check the robustness of our results, we conducted
otos exper- ph
with sentences
iments 5 and 6 as exact replications of experiments 1 and 2, but se whose conten
ntences,
with photographs of new individuals. Although there were minor pras if they we
esented
differences in the results, the same basic pattern for race co
nversation. After view
obtained with the new stimuli: when cues to coalitional
bjects alliance su
performed a 1-m
were verbal only, the size of the race effect was at
large task,
(0.57) and th
an array compos
significant [n = 51, M = 1.86 (SD = 2.69), t*(50) = 4.94, P = ditarget individ
;played
9.1 x 10-6]. But when cues to coalitional alliancene, wereeach
aug- tir sentence that
mented visually, the size of the race effect dropped substantially,on
peared ap the screen (
to 0.15. This drop in effect sizes-from 0.57 to
is0.15-is
pointitself th
was to recall w
significant (Z = 2.45, r = 0.24, P = 0.0073), as is ir a sentence.
direct This
testis aof ul
difficult task, and subjects made many
the race effect in these two conditions [t(101) = 2.95, r = 0.28, er rors (this is expected and necessary, as the method depends on
P = 0.004]. Indeed, in the condition in which coalitional cues ar analysis of errors). If they are categorizing the target indi-
luals into different groups on the basis of some cue, then this
were amplified, there was no statistically significant tendency for vii

Kurzban et al. PNAS | December 18, 2001 | vol. 98 I no. 26 I 15391

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms
should be reflected in the type of errors they make. [Errors in wi th the number of (different coalition, same race + different
this task could, in principle, arise from a failure to encode the co alition, different race) errors. To test for race effects, the
properties of the speaker of the sentence at viewing time or from nu mber of (same race, same coalition + same race, different
a failure to use encoded information at the time of recall. co alition) errors was compared with the number of (different
Although this method cannot distinguish between these possi- :e, sameracoalition + different race, different coalition) errors.
bilities, it seems more likely that errors are caused by failures r experiments
of Fc 3 and 4, one makes the equivalent computa-
encoding. It is not obvious why encoded information would ns,not
buttic
substituting sex for race.
be used at the time of recall if it were available. (Instructions [n all of the tests to assess whether a particular dimension had
were identical across experiments.) Regardless of where en the
encoded,
be the data point for each subject is the number of
information is lost in this chain, the result is that coalitional thin category
wi errors minus the number of between category
situations that are not correlated with race diminish the use of er rors made by that individual, and the hypothesis being tested
racial information.] is that this is greater than zero. Thus these are t* tests (paired t
In experiments 1, 2, 5, and 6, the ;ts). Because
teamswe had prior
were predictions,
each significance
composed tests were of
e-tailed,African-American
two Euro-American players and two unless otherwise noted. players. For
Using the results of a t* test, onetwo
each subject, the computer program constructed can index how strongly
coalitions
bjects were encoding
(teams) randomly, with the provision that a dimension by computing an
the players beeffect size,
equall
vhich varies from 0 to 1 (28).
divided racially. Teams alternated sentences and players were rThe more within-category errors
assigned randomly to sentences,ompared with with
thebetween-category
constraint errors) that
thatsubjects
the make, fir
four sentences consisted of first two Euro-American speakers,oneand
larger the effect size, r, will be. By comparing effect sizes,
then two African-American speakers n see whether(or subjects
viceencodedversa).
a dimension more
Thisstrongly in
restri
e
tion was added to reinforce the multiracial aspect of the two condition than in another, even if they encoded it to a
;nificant extent in both.
coalitions from the beginning of the dialogue. Experiments 3 an
4 were designed in the same way, except that Euro-American
women were used instead of African-American men. Se x Differences. Because this article is about the encoding of race,
d because none of our conclusions about race change when the
The antagonistic content of the sentences, and the order in which an
ta are analyzed separately for male and female subjects, we
they were uttered, contain sufficient information to infer which da
ve collapsed over the two sexes for the purposes of this article.
team each individual is on. To independently check this variable, we h
)wever, an analysis of selection pressures in the context of
gave the sentences alone (on paper), each paired with a name, to
alitions led us to predict that machinery for detecting multi-
a different set of subjects, with explicit instructions to figure out CC
dividual coalitions and alliances-especially competitive
which person was on which team (subjects in experiments 1-6 were
es-will be present in both sexes, but easier to activate in men
never instructed to attend to or infer team membership). The or
an in women (25). There are data that support these predic-
majority (70%) of subjects scored 100% correct, and all but one of th
Ins (R.K., L.C., and J.T., unpublished data).
the remaining subjects made only one error.
Nc te 1. The t* tests do not compare aggregated means (they are
Statistical Analyses. Because there were eight photos evenly divided
thin-subject tests); this is to ensure that results are general, i.e.,
along two dimensions, it is possible to make four different types of
t caused by a minority of subjects with extreme scores. An M =
error. Subjects, in misattributing a sentence to a player, could pick o , )0 for coalition represents =20% more within than between
another player on the same team and of the same race (one photo),
tegory errors (aggregated Ms: 4.61 between; 5.51 within); in
a player on the same team but of a different race (two photos), a
periment 2, M = 4.62 for coalition represents =200% more
player on the other team of the same race (two photos), or a photo wi
thin than between category errors (aggregated Ms: 2.31 be-
of a player on the other team of a different race (two photos). To
een; 6.92 within). (Range of aggregated means across exper-
compensate for the lower prior probability of making a same-team/ ir ents: within-category errors: 5.32-7.25; between-category er-
same-race error, the error rates of the other three categories were ro
rs: 2.31-4.61.)
divided in half before any of the following statistical tests were
conducted. This correction for different base rates is necessary, and Ft nding for this project was provided by the Harry Frank Guggenheim
standard in the literature (13-15). Fc undation, the James S. McDonnell Foundation, National Science
To test for coalition effects, the number ofGrant
undation (same coalition,
BNS9157-449 (to J.T.), Fc
and the University of Cali-
'nia,errors
same race + same coalition, different race) Santa Barbara
was Office of Research and
compared fo:Academic Senate.

1. Brewer, M. (1979) in The Social Psychology of Intergroup Stangor, C., Relations,

Lynch, L., Duan,eds. C. & Glass,
Austin, B. (1992)
15J. Pers. Soc. Psychol. 62,
W. G. & Worchel, S. (Brooks/Cole, Monterey, CA), 207-218.
pp. 71-84.
2. Brewer, M. (1979) Psychol. Bull. 86, 307-324. 16 Kelly, R. (1995) The Foraging Spectrum (Smithsonian Inst., Washington, DC).
3. Sherif, M., Harvey, O., White, B., Hood, W. & Sherif, Lewontin, C. R. (1961) Intergroup
(1972) Evol. 17
Biol. 6, 381-398.
Conflict and Cooperation (Univ. of Oklahoma Book Nei, M.Exchange,
& Roychoudhury, Norman). 18 Biol. Evol. 10, 927-943.
A. (1993) Mol.
4. Tajfel, H., Billig, M., Bundy, R. & Flament, C. (1971) Eur.
Keeley, J. Soc.
L. (1996) War Psychol. 1, 19(Oxford Univ. Press, Oxford).
Before Civilization
149-178. 20 Ember, C. R. (1978) Ethnology 27, 239-448.
5. Tajfel, H. (1978)
Manson, J. & Wrangham, R. (1991) DifferenCurr. Anthropol. 32, 369-390.
6. Billig, M. & Tajfel, H. (1973) Eur. J. Soc. Psychol. 3, 27-52. 22 Chagnon, N. (1992) Yanomamo (Harcourt, Fort Worth, TX), 4th Ed.
7. Locksley, A., Ortiz, V. & Hepburn, C. (1980) J. Pers. Soc. Psychol. 39, 773-783. 23
Smuts, B., Cheney, D., Seyfarth, R., Wrangham, R. & Struhsaker, T. (1987)
8. Levine, R. & Campbell, D. (1972) Ethnocentrism (Wiley, New York).
Primate Societies (Univ. of Chicago Press, Chicago).
9. Messick, D. & Mackie, D. (1989) Annu. Rev. Psychol. 40, 45-81.
Kurzban, R. (1998) Ph.D. dissertation (Univ. of California, Santa Barbara).
10. Hamilton, D., Stroessner, S. & Driscoll, D. (1994) in Social Cognition: Impact
Tooby, J. & Cosmides, L., eds. (2002) Evolutionary Psychology: Foundational
on Social Psychology, eds. Devine, P., Hamilton, D. & Ostrom T. (Academic,
San Diego), pp. 291-321. Papers (MIT Press, Cambridge, MA), in press.
11. Brewer, M. (1988) Adv. Soc. Cognit. 1, 1-36. 26 Sidanius, J. & Pratto, F. (1999) Social Dominance (Cambridge Univ. Press, New
12. Fiske, S. & Neuberg, S. (1990) Adv. Exp. Soc. Psychol. 23, 1-74. York).
13. Taylor, S., Fiske, S., Etcoff, N. & Ruderman, A. (1978) J. Pers. Soc. Psychol.Hirschfeld,
27 L. (1996) Race in the Making (MIT Press, Cambridge, MA).
36, 778-793. 28 Rosenthal, R. & Rosnow, R. (1991) Essentials of Behavioral Research
14. Hewstone, M., (McGraw-Hill, Boston), 2nd Ed. A.
Hantzi, & Johnsto

15392 | www.pnas.org/cgi/doi/10.1073/pnas.251541498 Kurzban et al.

This content downloaded from 205.208.14.14 on Wed, 25 Oct 2017 18:57:06 UTC
All use subject to http://about.jstor.org/terms