The Mismatch Effect: When Testosterone and Status Are at Odds
Robert A. Josephs and Jennifer Guinn Sellers
The University of Texas at Austin
Matthew L. Newman
Pranjal H. Mehta
The University of Texas at Austin
Why do some people strive for high status, whereas others actively avoid it? In the present studies, the
authors examined the psychological and physiological consequences of a mismatch between baseline
testosterone and a person’s current level of status. The authors tested this mismatch effect by placing high
and low testosterone individuals into high or low status positions using a rigged competition. In Study
1, low testosterone participants reported greater emotional arousal, focused more on their status, and
showed worse cognitive functioning in a high status position. High testosterone participants showed this
pattern in a low status position. In Study 2, the emotional arousal findings were replicated with heart rate,
and the cognitive findings were replicated using a math test. In Study 3, the authors demonstrate that
testosterone is a better predictor of behavior than self-report measures of the need for dominance.
Discussion focuses on the value of measuring hormones in personality and social psychology.
Keywords: testosterone, status, individual differences
Saint Peter Celestine has been referred to as “the most pathetic
figure in the history of the papacy” (Walsh, 1991). Celestine was
born Peter Morrone of Italy, in the early 13th century, and was
ordained as a priest in his early thirties. Content to spend the rest
of his life as a hermit in a small monastery in the mountains near
his home, Morrone was renowned for his wisdom and well-loved
by those around him. Unfortunately, his peaceful existence was
doomed to end after what many might have considered good news.
When Morrone was in his early eighties, Pope Nicholas IV died
without a successor. After two years of searching, the cardinals
finally settled on Morrone and despite his reservations he submit-
ted to their wishes. Consecrated as Celestine V in August 1294, he
quickly became a pawn in the political games of King Charles II,
miserable at his readily noticeable failures. After less than five
months as pope, overwhelmed by “the burden of the office he had
not sought and was incapable of filling” (Walsh, 1991), Celestine
finally abdicated his position, the first ever to do so. His troubles
did not end there, however. Celestine’s successor, Boniface VIII,
was worried that Celestine’s popularity as a holy man might
challenge his own authority, and launched a preemptive strike.
Boniface VIII ordered Celestine hunted down and imprisoned,
where he died 10 months later.
Needless to say, Celestine and Boniface had different reactions
to the same high status position. Whereas Boniface eagerly wel-
comed the opportunity the papacy represented, Celestine’s unwill-
ingness to assume the papacy was borne out by his miserable and
ultimately tragic experience as pope. This sad story raises a num-
ber of questions. What factors might lead someone to prefer high
status, whereas another shuns it? What happens when someone
desires high status, but does not have it? And what about the
person who does not want high status, but gets it anyway? In this
paper, we attempt to answer these questions by proposing that one
of the major contributors to these individual differences is one’s
baseline level of testosterone.
We begin by reviewing evidence linking testosterone to status-
related behaviors including dominance, suggesting that individual
differences in basal testosterone levels are correlates for status
preference. Then, we will present evidence showing the relevance
of testosterone to the social situation, and suggest that the relative
status of others within the group can exert powerful influences on
psychological and physiological functioning, depending on one’s
testosterone level and social standing. The interaction between
one’s status with respect to the group and one’s basal testosterone
level forms the basis of what we are calling the mismatch effect.
Testosterone, Status, and Social Situations
From a psychological perspective, one of the most interesting
things about testosterone is its relationship to dominant behaviors
(e.g., Archer, in press; Mazur & Booth, 1998). Research with
humans and a wide variety of animal species suggests that under
specific conditions, individuals higher in baseline testosterone are:
(1) more driven to gain and maintain high status; and (2) more
responsive to information about their status in particular situations.
These responses to status include a number of issues that are of
interest to social psychologists such as aggression, increased
Robert A. Josephs, Jennifer Guinn Sellers, and Pranjal H. Mehta,
Department of Psychology, The University of Texas at Austin; Matthew L.
Newman, Department of Psychology, Bard College.
This research was supported by NSF grants # NSF199601580-002 and
BCS0423405 to Robert A. Josephs. For their help, we thank Brett Bays,
Dan Clarke, Sam Gosling, Shirley Kong, Derek Millet, Rui Oliveira, Peter
Putnam, Oliver Schultheiss, Ky Strunk, Kip Williams, and John Wingfield.
Correspondence concerning this article should be addressed to Robert A.
Josephs, Department of Psychology, The University of Texas at Austin, 1
University Station, A8000, Austin, TX. E-mail: email@example.com
Journal of Personality and Social Psychology
2006, Vol. 90, No. 6, 999–1013
Copyright 2006 by the American Psychological Association
arousal, emotion, and cognitive performance. However, with few
exceptions (e.g., Dabbs, Alford, & Fielden, 1998; Josephs, New-
man, Brown, & Beer, 2003; Schultheiss, Dargel, & Rohde, 2003),
this research has escaped the attention of social psychologists.
Much of the evidence linking testosterone and status comes
from research with nonhuman species. Naturally occurring levels
of testosterone are correlated with position in the hierarchy, as well
as with dominant behavior (Beaver & Amoss, 1982; Cavigelli &
Pereira, 2000; Coe, Mendoza, & Levine, 1979; Collias, Barfield, &
Tarvyd, 2002; Elofsson, Mayer, Damsga ˆrd, & Winberg, 2000;
Kraus, Heistermann, & Kappeler, 1999; Oliveira, Almada, & Ca-
nario, 1996). These effects seem to be situation dependent. For
example, when Ruiz-de-la-Torre and Manteca (1999) injected
male lambs with testosterone, dominance struggles increased only
after unfamiliar lambs were introduced. Similarly, Collias et al.
(2002) found that tropical birds increased their dominance behav-
iors (viz., singing, establishing territory) only if they were injected
with testosterone and the dominant male was removed from the
area (see also Briganti, Seta, Fontani, Lodi, & Lupo, 2003 for
conceptually similar results among rabbits). The common theme
behind this pattern is that testosterone predicts status-related be-
haviors only when status is up for grabs, not in a stable status-
hierarchy (e.g., Koyama & Kamimura, 1999; Morgan et al., 2000;
Ostner, Kappeler, & Heistermann, 2002; Sapolsky, 1991).
Testosterone and Status in Humans
A number of studies with humans supports a positive relation-
ship between testosterone and status. Indeed, measuring testoster-
one at a single point in time predicts status-related behaviors
across a variety of situations and occupations (e.g., Cashdan, 1995;
Dabbs, La Rue, & Williams, 1990; Dabbs et al., 1998; Mazur &
Booth, 1998; Grant & France, 2001; Scaramella & Brown, 1978;
van Honk et al., 1999). For example, Dabbs et al. (1998) report that
people in high status occupations such as trial lawyers tend to have
higher levels of testosterone compared with people in lower status
occupations, such as patent lawyers. However, these results are
largely correlational, and have come under attack by some critics
who charge that the results from many of these studies are weak,
especially relative to the animal literature (see, e.g., the commen-
taries following the target article by Mazur & Booth, 1998). In
defense of the testosterone-status relationship in humans, a socio-
logical analysis of testosterone employed by Kemper (1990) sug-
gests that, as has been observed in other species, the relationship
between testosterone and behavior should appear only when cer-
tain situational pressures interact with individual differences in
baseline testosterone. Despite the appeal of this interactionist per-
spective, only a handful of studies have employed the experimental
or quasi-experimental methods necessary for proper tests of the
interactionist model (e.g., Josephs et al., 2003; Mazur, Booth, &
Dabbs, 1992; Mazur, Susman, & Edelbrock, 1997; McCaul, Gla-
due, & Joppa, 1992).
Closely related to need for status, testosterone levels have also
been linked to implicit power motives (e.g., Schultheiss et al.,
2003) and selective attention to threatening faces (van Honk et al.,
1999; 2001). Van Honk et al. showed that participants high in
testosterone spent more time gazing at angry, potentially status-
threatening faces than did participants low in testosterone, but
were no more attentive to neutral faces than were low testosterone
participants. In fact, low testosterone participants actually attended
away from threatening faces. These motivational and attentional
differences map nicely onto the behavioral differences reviewed
earlier, suggesting the possibility that the dominance-submission
distinction associated with testosterone may have motivational and
information processing correlates.
Taken together, the literature suggests that when status is threat-
ened, high testosterone individuals are motivated to try and regain
it. What about the outcome of dominance battles? Are high tes-
tosterone individuals more likely to be successful at regaining
status? The prototypical dominance contest involves a physical
challenge, and dominance must be regained through physical
means (e.g., winning a fight, intimidation). In these cases, testos-
terone, which enhances muscle mass and metabolism (e.g., Bhasin,
Woodhouse, & Storer, 2001; Tsai & Sapolsky, 1996), should
provide an advantage to those who wish to regain lost status. Not
surprisingly, higher T animals are more likely to be the winners of
these types of encounters (e.g., Morgan et al., 2000; Ostner et al.,
But what if dominance must be regained through nonphysical
means? We explored this in several recent papers (Josephs et al.,
2003, Study 1; Newman, Sellers, & Josephs, 2005). High and low
testosterone participants were placed in a low status position (or a
control group) before taking a complex test that was framed as
being capable of restoring status. Consistent with predictions,
those high in testosterone performed considerably worse in a low
status position. This provides some preliminary evidence for the
idea that attempting to regain status impairs cognitive performance
on a complex task. However, other psychological and physiolog-
ical consequences of response to status threat (e.g., affective,
motivational, cardiovascular) have yet to be explored. This was
one of our primary goals in designing the present studies.
The Mismatch Effect
As reported above, there is strong and abundant evidence link-
ing testosterone levels to rank, dominant behaviors, and status in
nonhuman animals, and some evidence in humans as well. In
addition, as Sapolsky and others have speculated (cf. Sapolsky,
1991), testosterone may serve as a biological correlate for concerns
about status. When high status is threatened or challenged, the
typical response is an attempt to restore status through some type
of antagonistic behavior, and this response is typically seen among
individuals high in testosterone. However, what happens when an
individual who is on the low end of the status hierarchy—an
individual low in testosterone—experiences upward pressure on
his or her status quo? In other words, how does such an individual
react to having unwelcome status thrust upon him or her?
There is only one empirical example that we are aware of in
which low-ranking members of a species other than humans are,
through no fault of their own, forcibly moved up into a higher
status position. In Rohwer’s (1977) classic study on the effects of
social cheating in Harris’s sparrows, researchers bleached the
feathers of some of the dominant birds so that they appeared
submissive (i.e., lighter feathers) and colored some of the feathers
of some of the submissive birds so that they appeared dominant
(i.e., darker feathers). The formerly dominant birds with the new
subordinate signal of white plumage became inexhaustible bullies,
picking fights with every bird around, presumably in an attempt to
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
restore their lost status. The newly appointed dominant birds
suffered a different misfortune. The attacks against them grew so
frequent that many of them chose to leave the flock to feed in
Clearly, humans share a number of commonalities with other
animal species, and thus it is perhaps not surprising that in humans
testosterone has been shown to bear a positive relationship to
status-related behaviors such as dominance and subordinance.
However, it has been argued elsewhere (Kemper, 1990) that unlike
other species, humans are unique in their ability to achieve high
status through routes other than dominance. Kemper (1990) argues
that a position of high status can be achieved through the person’s
“eminence.” For example, achieving greatness in one’s discipline
or field can elevate one to a position of high status without the
need for the more typical mano-a-mano dominance battles. A
reluctant but eminent academic can be appointed to a high status
position such as that of chair, dean, or president as a result of
scholarship. Another route to high status that avoids dominance
battles is birthright, in which an individual is born into a high
status position without necessarily welcoming the attention,
power, and privilege that status brings. There are undoubtedly
other routes that do not require dominance battles. The point is that
it is precisely because humans can find themselves in positions of
high status without having to engage in dominance battles that
human societies may periodically produce a mismatch between a
person’s desired level of status, as indicated by testosterone level,
and actual status.
We intentionally chose examples that illustrated only one-half
of the mismatch between a person’s desired level of status and his
or her actual status, because the other half is not only more
intuitive and frequent, but it is also not unique to humans. So, for
example, a high testosterone individual denied his perceived due
(e.g., a trial lawyer losing a trial) might be more likely to have a
car accident on the way home from the courthouse due to experi-
encing a range of negative and positive psychological and physi-
ological reactions, ranging from anger, reduction of cognitive
function, and stress, to a misplaced motivation to regain his lost
status by raging against his fellow drivers.
Although it might be apparent why someone high in testosterone
might react strongly to a drop in status, we have only hinted at the
question of why someone low in testosterone might also react
strongly after experiencing an increase in status. On the one hand,
this reaction appears to defy the undeniable reproductive advan-
tage that high status confers (e.g., Buss, 2003). However, it is
exactly because of the well-known advantages of high status that
those in such positions must occasionally, and sometimes con-
stantly, fend off others who wish to enjoy its advantages. The
literature on dominance is replete with examples of dominance
challenges, some of which end in serious injury or even death to
We suggest that low testosterone individuals might shun high
status positions and dominance battles because they lack a strong
power motive (Schultheiss et al., 2003), they lack a dominating,
aggressive personality (e.g., van Honk et al., 1999), and they may
not believe they have what it takes physically to maintain such
positions and win such battles (cf. Tsai & Sapolsky, 1996). If a less
assertive, low testosterone individual lands a high status position,
the resulting mismatch between desire to “fly below the radar” and
current position of high status might generate a strong reaction,
ranging from fear to confusion to arousal to a motivation to return
to a safer level of status. These speculations led us to the current
set of studies.
The Current Studies
In the present studies, we sought to test the mismatch effect by
placing a low or high testosterone individual into a position of low
or high status, and taking various measures of psychological and
physiological functioning. The design of each experiment was
straightforward. Participants took part in a rigged competition
against another participant (Study 1) or against a confederate
(Study 2), and either won (high status) or lost (low status). In
Study 3, testosterone was pitted against several popular self-report
measures of individual and social dominance to determine whether
testosterone is a worse or better predictor of status-seeking behav-
ior than scores on self-report measures of dominance. At various
points during each experiment, cognitive, affective, and physio-
logical measures were obtained.
Based on our earlier work (Josephs et al., 2003; Newman et al.,
2005), we expected that changes in cognitive performance in each
study would result from the interaction between testosterone levels
and the status manipulation. Specifically, individuals high in tes-
tosterone should perform worse in a low status position than in a
high status position, despite the fact that performing well could
lead to high status. This result would serve to replicate previous
research. The present studies extend our prior work by extending
the model to low testosterone individuals who are moved out of
their status comfort zone. The research reviewed above suggests
that low testosterone individuals might be uncomfortable with high
status, and this discomfort might have consequences for higher-
order cognitive functioning. Study 1 was designed to explore this.
In addition to the cognitive performance measures in each study,
we included several measures of emotion, attention, and physio-
logical responses. Our goal in including these measures was to
explore the consequences of a psychological mismatch. As such,
we did not make any a priori predictions about these measures.
In this first study, we were interested in exploring how a state of
mismatch would influence a person’s psychological functioning.
To test this, participants competed in same-sex dyads on a series of
tasks that were described as measures of intelligence. The diffi-
culty of the first task was rigged so that, via random assignment,
one participant was guaranteed victory. After this task was com-
pleted and the two participants were aware of who had won and
who had lost, they were given the opportunity to gain, regain, or
maintain their status through performance on a complex cognitive
task (a section from the GRE analytic exam). This task also
allowed us to explore a person’s capacity for complex cognitive
operations. To further explore the consequences of matched and
mismatched status, we administered: (1) a measure of implicit
attention to status to explore a person’s awareness of status-
relevant stimuli; and (2) a measure of self-reported emotional
arousal to explore a person’s affective and motivational state.
Participants and Design
Ninety-two students at the University of Texas at Austin participated in
this study in exchange for partial fulfillment of an introductory psychology
course research requirement. Nineteen participants were dropped from the
analysis because of insufficient salivary samples. Of the remaining people,
34 participants were women and 39 were men. These participants com-
pleted the experimental tasks in a 2 (testosterone-level: high or low) ? 3
(status: submissive, dominant, or solo control) between-subjects quasi-
Materials and Procedure
Participants arrived two at a time to the experiment in same-sex dyads,
except in the control condition in which they arrived individually. Upon
arriving, a male experimenter greeted them and asked them to take a swig
of water from a nearby drinking fountain, swoosh the water around in their
mouths, and then spit the water out into the drain. This was done to clear
their mouths out of any food particles or other particulates that might
contaminate the saliva sample used to test for baseline testosterone levels.
The experimenter then gave participants a piece of Trident original flavor,
sugar-free gum to chew, to facilitate salivary secretion.
After providing informed consent, participants then spit out their gum
and drooled into a 1.8 mL vial, while in separate rooms. Participants were
told that we were interested in the “effects of caffeine and other stimulants
that might affect their performance.” We avoided telling them up front that
we were measuring testosterone, due to its association with competition
and dominance. We also assured them that we would only test for these
stimulants (i.e., they didn’t have to worry about illegal drugs being de-
tected). The saliva samples were kept in a freezer at 0° C until later
testosterone analysis. Salivary collection was consistent with the procedure
described by Granger, Schwartz, Booth, and Arentz (1999). All experi-
mental sessions were conducted between noon and 4 p.m. to control for
diurnal fluctuations in testosterone (e.g., Granger et al., 1999). Testoster-
one levels were measured in saliva using enzyme immunoassay kits pro-
duced by Salimetrics. For a description, see Granger et al. (1999).
After the salivary samples were obtained, the experimenter escorted both
participants into one room and told them that they were going to be
completing three tests of intelligence. The first of these tests served as our
status manipulation, and the other two served as dependent measures of
psychological functioning. Participants were led to believe that they were
competing with one another, based on their total score across the three
The status manipulation.
The first ostensible test of intelligence was
the spatial processing test. The instructions for this task were to trace
through a series of numbers in sequential order until the highlighted
number was reached (Schultheiss, Campbell, & McClelland, 1999). The
test was composed of a packet of 10 such tasks, and they were to complete
the test as quickly as possible because their timed score would be recorded.
In reality this test served as the status manipulation. Our goal was to place
one person at a relative advantage by assigning them to win this first task.
Assignment to either the submissive or dominant condition was rigged by
handing participants either a short (dominant assignment) or long (submis-
sive assignment) version of the line tracing task to complete so that we
could manipulate how long it took participants to finish. When each of the
participants was finished with all 10 tests, they said, “Done,” and raised
their hands. Therefore, all were aware of their relative performance on the
timed number tracing task. Participants in the control condition completed
the short version of this task and no mention was made about completing
the assignment quickly nor was their time recorded.
Implicit attention to status.
Following the status manipulation, partic-
ipants worked on a measure of implicit status concerns, which was pre-
sented as a measure of visual perceptual speed. This test was presented as
one component of the total score on which winning or losing would be
based. Participants had 7 min to complete a word search similar to the
jumble puzzles that appear in many Sunday newspapers. The word search
contained a mix of 10 status occupations (advisor, advisee; doctor, hygien-
ist; director, staff; lawyer, paralegal; president, secretary) and a collection
of neutral words. All participants completed the same version. A copy of
this measure is available from the authors.
Measure of cognitive ability.
To test for cognitive functioning, partic-
ipants were then given a version of the analytic section of the Graduate
Record Exam (GRE), presented as a measure of verbal analytical ability.
This test was also presented as one component of the total score on which
winning or losing would be based. Thus, performing well could potentially
offset an initial (randomly assigned) loss, just as performing poorly could
offset an initial (randomly assigned) win. Participants were given 20 min
to complete 20 questions of moderate difficulty. The experimenter in-
formed them that points would be deducted for each incorrect answer to
After completing the GRE analytic test, partic-
ipants completed the PANAS (Watson, Clark, & Tellegen, 1988) to mea-
sure self-reported emotions. Finally, participants were debriefed for sus-
picion and told the true nature of the experiment.
Testosterone levels were standardized separately for men and
women as well as for time of assay to control for sex differences
in baseline testosterone level and variability among assay batches,
respectively. (Participants were randomly selected for each assay
batch, in order to minimize the effects of storage time on testos-
terone levels.) Furthermore, baseline testosterone levels were sep-
arated into two groups (high and low baseline testosterone) based
on the upper and lower thirds of the standardized distributions. We
chose to investigate the upper and lower thirds instead of the entire
distribution because we were interested in the behavior of individ-
uals high and low in their need for dominance, and less interested
in explaining normal dominance strivings. This is common prac-
tice among researchers interested in individual differences (e.g.,
1We ran the study with an additional between-subjects factor of com-
petition (competitive or noncompetitive). In the subsequent status manip-
ulation for the competitive condition, participants were given a buzzer to
ring every time they finished each of the 10 number tracings. This was
done to reinforce the competitive element of the task. In the noncompeti-
tive condition, no comments were made about competition and there was
no buzzer to ring after finishing any of the number tracings. When each of
the participants in both conditions was finished with all 10 tests, the
participant said, “Done,” and raised a hand. Therefore, all participants in
both the competitive and noncompetitive condition were aware of their
relative performance on the timed number tracing task; this served as the
Preliminary analyses with a Hotelling’s T test revealed that assignment
to the competitive or noncompetitive condition did not affect responses on
any of the dependent variables, F(1, 46) ? .91, p ? .53. We speculate that
this is because participants in both the competitive and noncompetitive
conditions were all aware of the other participants’ performance as they
were seated in close proximity to one another. Another possibility is that
we did not have enough power to see the subtle effects of the differences
between the two conditions. For study, all reported analyses are collapsed
across the conditions. Future research should assess the effects of different
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
need for closure, Webster & Kruglanski, 1994; self-consciousness,
Schlenker & Weigold, 1990; self-esteem, Giesler, Josephs, &
Swann, 1996). After removing the middle third of the sample, we
were left with 47 participants (21 women and 26 men).
Thus, high T men (M ? 139.25; SD ? 34.58) were high relative
to low T men (M ? 56.85; SD ? 14.96). High T women (M ?
103.30; SD ? 33.03) were high relative to low T women (M ?
36.46; SD ? 16.28).2To test the hypothesis that individuals high
and low in baseline testosterone (T) would show differential re-
sponses on all the measures of psychological functioning, a 3
(submissive vs. dominant condition vs. control) ? 2 (high baseline
T level vs. low baseline T level) multivariate analysis of variance
(MANOVA) on GRE performance, status words, PA, and NA was
conducted. Consistent with predictions, there was a significant
interaction between status condition and T level, F(3, 41) ? 3.36,
p ? .01. There were no main effects for either T level F(3, 41) ?
2.01, p ? .10, or condition F(3, 41) ? 0.88, p ? .55. Univariate
results for measures of cognitive performance, attention to status,
and affective arousal are presented below. Table 1 presents the
descriptive statistics for these dependent measures.
Cognitive performance was measured by the total numbers of
items answered correctly on the analytic section of the GRE. Did
mismatched status affect cognitive performance? This question
was tested using a 3 (submissive vs. dominant vs. control condi-
tion) ? 2 (high baseline testosterone level vs. low baseline testos-
terone level) analysis of variance (ANOVA).
Mismatched status was strongly related to cognitive perfor-
mance. Consistent with the findings of Josephs et al. (2003), the
interaction between baseline T level and status condition was
significant, F(2, 41) ? 9.01, p ? .01. See Figure 1. Participants
high in baseline T performed worse in the low status condition
than in the high status condition, t(14) ? 3.45, p ? .01. Con-
versely, participants low in baseline T showed better cognitive
performance in the low status condition than in the high status
condition, t(15) ? 3.33, p ? .01. When status was not up for grabs,
however, T had no effect on cognitive performance. There were no
significant differences between the control group and either of the
experimental conditions for either low T (dominant condition:
t(13) ? 1.08, p ? .15; subordinate condition: t(13) ? .77, p ? .23)
or high T (dominant condition: t(14) ? 1.2, p ? .12; subordinate
condition: t(13) ? .31, p ? .38) participants.
Implicit Attention to Status
Implicit attention to status was measured by counting the num-
ber of status-relevant words found in the word search task, divid-
ing them by the total number of words found, and multiplying by
100%. Did mismatched status affect a person’s tendency to notice
status-relevant information? To test for the effects of status and
testosterone level on implicit attention to status, a 3 (submissive
vs. dominant condition vs. control) ? 2 (high baseline T level vs.
low baseline T level) ANCOVA was conducted controlling for
gender. Gender was controlled for because our word search used
occupations as our status words, and Eagly and Wood (1999) have
proposed that men and women differentially occupy status roles.
An ANCOVA revealed a significant interaction between T and
condition for percentage of status-relevant words, F(2, 51) ? 3.06,
p ? .056. As can be seen in Figure 2, attention to status-relevant
words increased as conditions changed from status match to mis-
match. Under conditions of dominance, the presumed status needs
of low T participants were not met; they showed superior status
word search performance compared to low T submissive partici-
pants whose status needs were presumably being met, t(15) ?
2.03, p ? .03. Although not reaching significance, the opposite
picture emerged for high T participants. Under conditions of
dominance, high T participants had their status needs met and
showed better status word search performance than in the submis-
sive condition, when these needs were presumably unmet, t(23) ?
1.05, p ? .15. In conceptual replication of van Honk et al.’s (1999)
findings, there was a main effect for testosterone level such that
high testosterone participants had a greater implicit attention to
status-relevant words than did low testosterone participants, F(2,
51) ? 6.37, p ? .03. (Van Honk et al.’s conditions are most closely
replicated in the control condition, and in replication of van Honk
et al., high T control participants found approximately 3 times as
many status words as did low T controls.) Finally, there was no
significant effect of either T, F(1, 33) ? .688, p ? .413, or
condition, F(2, 33) ? 1.081, p ? .351, on the total number of
words (status plus neutral) found. Furthermore, the interaction was
not significant, F(2, 33) ? .254, p ? .777)
2Mean testosterone levels were also computed prior to removing the
middle third of the sample distribution. The mean for women was 64.07
(SD ? 34.99), and the mean for men was 94.82 (SD ? 41.39). We
computed interassay and intraassay CV averages, and these were 5.3% and
4.5%, respectively (both fell within the precision ranges reported by
Descriptive Statistics for Study 1
* Means are adjusted controlling for gender.
To examine the effects of matched versus mismatched status on
emotional arousal, we conducted a 3 (status condition) ? 2 (T-
level) MANOVA, with positive (PA) and negative (NA) affect as
the dependent variables. There was a significant multivariate in-
teraction, F(4, 66) ? 2.90, p ? .03, but no significant multivariate
We first examined whether mismatched status had an effect on
PA (? ? .87). We found a significant interaction between condi-
tion and testosterone level on PA, F(2, 40) ? 3.80, p ? .03. High
T participants reported marginally more PA in the submissive
condition than in the dominant condition, t(15) ? 1.58, p ? .07,
whereas low T participants reported more PA in the dominant
condition than in the submissive condition, t(14) ? 1.80, p ? .05.
This same pattern of elevated PA was observed when comparing
mismatched participants against control participants, t(13) ? 1.71,
p ? .06, for high T submissives compared to controls, and t(13) ?
1.70, p ? .06 for low T dominants compared to controls. There
were no significant differences between matched participants and
the control group for high T winners compared to controls, t(14) ?
.16, p ? .44, nor for low T losers compared to controls, t(13) ?
.20, p ? .42. See the bottom panel of Figure 3 for these means.
We next examined whether mismatched status had an effect on
NA (? ? .85). Although there were no significant main effects or
interactions, we wanted to explore these findings to see if a pattern
of effects similar to those that emerged for PA had emerged for
NA. As with PA, high T submissives did report more NA than high
T dominants, t(23) ? 1.91, p ? .03. High T submissives did not
report more NA than in the control condition, t(13) ? 1.00, p ?
.17, nor did high T dominants report more NA than in the control
condition, t(14) ? 1.02, p ? .16. Low T dominants did not report
significantly different NA from low T submissives, t(22) ? .51,
p ? .31, although this pattern was in the same direction as with PA.
There were no significant differences between low T dominants
t(13) ? 1.04, p ? .16, or losers, t(13) ? 1.29, p ? .11, and the control
condition. See the top panel of Figure 3 for these means.
The results from Study 1 provide insight into the consequences
of a mismatch between one’s desired level of status and actual
and lower third. The y-axis represents number correct on GRE analytical. Error bars represent 1 standard error.
Study 1: The effects of status and testosterone on cognitive performance. T level is divided into upper
into upper and lower third. The y-axis represents percentage of status words found in the word search. Error bars
represent 1 standard error.
Study 1: The effects of status and testosterone on implicit attention to status words. T level is divided
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
level of status. First, consistent with our previous research, we
found that cognitive functioning was impaired under conditions of
mismatch. Participants high in testosterone performed worse on
the analytical GRE when they were in a position of low status.
Building on previous research, we found that participants low in
testosterone showed the opposite effect. They performed worse on
the analytical GRE when they were placed into a position of high
status. Although it remains speculative at this point, it is possible
that the desire for status change may prove too distracting for the
proper functioning of complex higher-order cognition. It remains
to be seen if all cognition is interfered with by a desire for status.
Research by Zajonc, Heingartner, and Herman (1969) and others
suggest that task complexity may determine task performance in
the face of desire for status change, with simple, overlearned tasks
benefiting whereas performance on complex and novel tasks
showing impairment. Perhaps, similar to findings in other species
in which testosterone activity is positively correlated with success-
ful dominance battles (presumably because these battles feature
straightforward tasks requiring energy and physiological arousal
but very little complex thought), simple, physical tasks in humans
might be facilitated by a desire for status change whereas complex
cognitive (or physical) tasks might be interfered with. This possi-
bility was not tested in this paper.
Second, findings from the word search task suggest that indi-
viduals who do not have their status needs met tend to pay more
attention to their status. Under conditions of mismatch, partici-
pants who had their status desires presumably frustrated found
more status-relevant words compared to those whose status desires
matched their level of status. Curiously, high testosterone partic-
ipants found as many status words in the control condition as they
did in the low status (mismatched) condition. This is consistent
with work by van Honk and colleagues who also found basal
testosterone differences in the absence of a status challenge (van
Honk et al., 1999; 2001). Although speculative, it is possible that
when exposed to status-relevant stimuli, high testosterone individ-
uals become attentive, regardless of the presence or absence of a
bottom panel ? positive affect. T level is divided into upper and lower third. The y-axis represents average
emotion ratings on the affective arousal composite. Error bars represent 1 standard error.
Study 1: The effects of status and testosterone on emotional arousal. Top panel ? negative affect;
previous status challenge. However, comparisons with the control
group should be interpreted with caution. In hindsight, there is a
potential confound between condition (control vs. status condi-
tions) and working alone versus working with others. It is un-
known to what degree this biased the control group, but we suggest
that future studies work to eliminate this confound.
Third, we found that participants reported heightened negative
and positive emotional arousal when their preference for status, as
indicated by their testosterone level, was not congruent with their
actual status level. Do the PA findings indicate that mismatched
participants were happier than their matched counterparts? This is
unlikely, as the PANAS does not include happiness among its
positive affect items. Rather, Watson (2000) has conceptualized
PA and NA as measures of approach and avoidance motivation,
respectively. In fact, Harmon-Jones, Vaughn, Mohr, Sigelman, and
Harmon-Jones (2004) found that PA increased in response to
insults that aroused anger. Evidently, existing at the wrong status
level is distressing (high NA) as well as motivating (high PA). One
interpretation is that mismatched participants were both unhappy
(high NA) and motivated to reclaim their lost state of desired status
The status search and PA results also dovetail with the work of
Wright and his colleagues (e.g., Wright, Martin, & Bland, 2003)
who have argued that people will experience arousal in the process
of tackling a desired challenge or goal as long as the goal is not
perceived as insurmountable. Extending these ideas to Study 1, a
mismatched participant would be motivated to restore his or her
desired level of status, and if so, might experience the type of
arousal associated with approach motivation (and, consequently,
with PA). Indeed, mismatched participants showed an increase in
PA and NA, which is the type of affective arousal one would
expect when one is distressed and aroused to pursue a desired goal
(in this case, a change in status).
In addition to exploring the mismatch effect, Study 1 was
noteworthy in demonstrating that testosterone related to psycho-
logical functioning only when status was threatened (except in the
word search task in which high T participants performed equiva-
lently under status threat and control conditions). In the absence of
status threat, high testosterone participants did not perform better
(or worse) on the GRE than did low testosterone participants, and
they did not differ in emotional arousal. So, the results of Study 1
showed that only under conditions of mismatch—when high tes-
tosterone participants lost and low testosterone participants won—
was testosterone linked to GRE performance and emotional
In designing Study 2, four issues struck us as critical. First, in
order to determine the generality and robustness of the findings
from Study 1, we wanted to attempt a manipulation of status that
was methodologically different from the one used in Study 1.
Whereas in Study 1 a participant experienced winning or losing
before our dependent measures, in Study 2 we wanted feelings of
dominance or subordination to emerge during the battle to achieve
dominance. In Study 2, participants did not experience an actual
loss or win, but rather were led to believe that they were either
being dominated or were in a position of dominance during the
competition. In naturally occurring dominance battles, feedback is
often continuous (e.g., battles for job promotions, mate-selection,
sports competitions, etc.). In most experiments, however, the typ-
ical status manipulation is discrete—two participants compete with
no knowledge of their status until the task ends. Thus, we at-
tempted to design a status manipulation to reflect naturally occur-
ring dominance battles.
Second, we wanted to extend our findings on reported affective
arousal to heart rate. Instead of asking people to report on their
feelings, we measured changes in heart rate. According to Wright
and his colleagues, reactions by people attempting to overcome
challenges involve heightened cardiovascular responses. Similarly,
recent work by Schultheiss and his colleagues (Schultheiss, in
press; Wirth, Schultheiss, & Welsh, in press) showed that individ-
ual differences in preference for dominance predicted physiolog-
ical reactivity, such that those whose dominance needs were met
showed a decline in stress levels whereas those whose needs were
not met (e.g., those who feared dominance and were unwillingly
placed into a high status position) showed an increase in stress.
The research of Wright and Schultheiss both suggest an increase in
heart rate in our mismatched participants but a decrease when
participants are placed into positions of desired status, albeit for
somewhat different reasons.
Third, we wanted to generalize the higher-order cognitive per-
formance findings to a different, yet equally important, domain, so
we switched from analytic to mathematics performance.
Fourth, men and women in Study 1 had higher T levels than are
often observed. Although there are many reasons for these higher-
than-typical levels, one disturbing possibility is that the antibody
kits used to determine T concentrations might have had high
cross-reactivity (i.e., been capturing other steroids or androgens in
addition to T, thus inflating the observed levels). If so, then
attributing the observed effects to T may be wrong—rather, these
effects may be the result of a combined androgen or steroid profile,
rather than attributable solely to T, as we have been claiming. To
test this, the saliva from Study 2 was sent to an outside laboratory
(DSL, Webster, TX) for analysis.
In this second study, participants again competed in same-sex
dyads on what they were told was a measure of intelligence.
Unbeknownst to the participants, the person they were competing
against was a confederate, acting out the part of either a self-
professed expert or novice. During the course of the contest,
measures of heart rate and blood pressure were collected.
Participants and Design
Sixty-four students (17 women and 45 men) at the University of Texas
at Austin participated in this study in exchange for partial fulfillment of an
introductory psychology course research requirement.3These participants
completed the experimental tasks in a 2 (T-level: high or low) ? 2 (status:
submissive or dominant) between-subjects quasi-experimental design.
3The reported ethnic heritage was 68% Caucasian, 16% Hispanic or
Latino, 10% Asian, and fewer than 1% were African American, American
Indian, or did not report their ethnic heritage. This information was only
collected in Study 2.
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
Materials and Procedure
Prior to the experimental session, participants completed a demographic
questionnaire indicating their race, gender, and prior performance on the
quantitative section of the SAT. Participants arrived at the experiment
alone and were greeted by a same-sex experimenter and introduced to a
same-sex confederate playing the part of another participant in the study.
Both the participant and the confederate were then instructed to rinse their
mouths out and chew a piece of gum to prepare for the salivary sample.
Saliva collection procedures were identical to those in Study 1, although as
noted above, testosterone analysis was conducted at a different laboratory.
After the salivary samples were obtained, and approximately 15 min
after the start of the experiment, participants sat quietly while the experi-
menter took three back-to-back baseline measures of heart rate and blood
pressure using a Walgreen’s Automatic Inflation Electronic Digital Blood
Pressure/Pulse Monitor Kit (Model 93A). This baseline procedure varied
slightly from that suggested by Shapiro et al. (1996) who recommend
waiting 20 min before obtaining readings, calibrating the blood pressure
instrument, discarding the first few readings to allow participants time to
habituate to the procedure, and taking 3 to 5 readings at 1 to 2 min intervals
separated by quiet rest. After these baseline measures were obtained
(taking, on average 5 min), the experimenter then escorted the participant
and confederate into a room together and placed them at separate desks
facing away from one another.
For the next portion of the experiment, participants were told they would
be competing against one another to win a gift certificate to Tower
Records. This was done to reinforce the competitive element of the testing
situation as well as to ensure that participants were taking the experiment
seriously. Participants were then instructed they would have 20 min to
complete 20 questions on the quantitative section of the GRE, and that they
would be stopped three times during the test to take additional heart rate
and blood pressure measures. These intervals occurred every 5 min after
the start of the test.
The status manipulation.
Unbeknownst to the participants in the study,
the person they were competing against was, in reality, a confederate. His
or her role in the study was to manipulate the participant’s perceived
dominance in the competition. Right before the competition began and
during each of the breaks to take cardiovascular measures, the confederate
acted from a script to be either dominant or submissive. In the dominant
condition, the confederate played the role of submissive and periodically
uttered comments along the lines of “You have to be kidding me. I stink at
math,” and “I don’t know if my brain’s not working or if I’m just an idiot
because I can’t even answer one of these questions.” In the submissive
condition the confederate played the role of dominant, and said such lines
as “What, this is taken from the GRE quant? I just took this to get into grad
school and aced it,” and “Jeez, I can’t even believe how easy this is.”
After the experiment was over, participants were debriefed for suspicion
and told the true nature of the experiment.
As in Study 1, testosterone (T) levels were standardized sepa-
rately for men and women as well as for time of assay to control
for sex differences in baseline T level and variability among assay
trials, respectively. Furthermore, baseline T levels were separated
into two groups (high and low baseline T) based on the upper and
lower thirds of the standardized distributions. Thus, high T men
(M ? 131.34; SD ? 42.13) were high relative to low T men (M ?
41.75; SD ? 10.71). High T women (M ? 29.16; SD ? 5.82) were
high relative to low T women (M ? 9.87; SD ? 1.19). Mean T
levels were also computed prior to removing the middle third of
the sample distribution. The mean for women was 20.01 (SD ?
12.36), and the mean for males was 89.28 (SD ? 27.39). These
means reflect T concentrations that are within the range of values
typically seen in the literature.
Cognitive performance was measured by the corrected score on
the quantitative section of the GRE. The corrected score (used in
scoring the actual GRE) was computed by adding the number of
correct answers, and then subtracting the number of incorrect
answers multiplied by .25. The corrected score was strongly cor-
related with the total number of items correct, r(42) ? .99, p ? .01.
A 2 (submissive vs. dominant condition) ? 2 (high baseline T
level vs. low baseline T level) ANCOVA was conducted on these
corrected scores, controlling for math SAT scores. Three partici-
pants were dropped from this portion of the analysis because they
did not report their SAT scores.
Consistent with the mismatch effect, there was a significant
interaction between T level and status, F(1, 34) ? 6.44, p ? .02.
See Figure 4. High T participants performed significantly better in
the dominant condition than they did in the subordinate condition,
t(19) ? 1.91, p ? .04. Conversely, low T participants performed
marginally significantly better in the subordinate condition than
they did in the dominant condition, t(18) ? 1.45, p ? .09. There
were no main effects for T level or for status condition. See Table
2 for descriptive statistics.
The initial reading for heart rate was discarded to allow partic-
ipants time to habituate to the heart monitor. Baseline heart read-
ings were then computed for the cardiovascular measure by aver-
aging theremaining twopreexperimental
experimental heart rate measure was then computed by averaging
the three heart rate readings obtained during the testing session.
There were no significant differences across the three measures
collected during the testing session (p’s ? .38).
If participants high in T desire to achieve high status, then when
denied this by being placed into a low status position, they should
experience an increase in heart rate as a reflection of the motiva-
tion to restore high status. Furthermore, participants low in T
would show an opposite effect, experiencing greater heart rate in
a high status position than in a low status position. To test this, a
2 (submissive vs. dominant condition) ? 2 (high baseline T level
vs. low baseline T level) ANCOVA was conducted predicting
experimental heart rate while controlling for baseline heart rate.
Consistent with the mismatch effect, the interaction between T
level and status barely missed significance, F(1, 36) ? 4.01, p ?
.052. See Figure 5. Reminiscent of the affective arousal pattern
observed in Study 1, heart rate increases were observed under
conditions of mismatch. Planned comparisons revealed that low T
participants had significantly higher heart rates when they were in
the dominant condition than when they were in the subordinate
condition, t(20) ? 2.18, p ? .03 (although in the predicted direc-
tion, high T participants did not show a statistically significant
difference in heart rate between the two conditions, t(20) ? .64,
p ? .26). See Table 2 for descriptive statistics.
As with heart rate, the initial readings for systolic and diastolic
blood pressure were discarded to allow participants time to habit-
uate to the heart monitor. Baseline systolic and diastolic readings
were then computed by averaging the remaining two preexperi-
mental measures. The experimental systolic and diastolic measures
were then computed by averaging the three readings obtained
during the testing session.
Systolic blood pressure.
Although we had no a priori predic-
tions about the effects of T and status on systolic blood pressure,
we subjected the experimental systolic blood pressure to a 2
(submissive vs. dominant condition) ? 2 (high baseline T level vs.
low baseline T level) ANCOVA controlling for baseline systolic
blood pressure. There were no significant main effects for T level,
F(1, 37) ? .42, p ? .52, condition, F(1, 37) ? .003, p ? .95 nor
the interaction, F(1, 37) ? .002, p ? .96 on systolic blood pressure
when controlling for baseline systolic blood pressure. Descriptive
statistics are reported in Table 2.
Diastolic blood pressure.
As with systolic blood pressure, we
had no a priori predictions about the effects of T and status on
diastolic blood pressure. We performed a 2 (submissive vs. dom-
inant condition) ? 2 (high baseline T level vs. low baseline T
level) ANCOVA on experimental diastolic blood pressure control-
ling for baseline diastolic blood pressure. There were no signifi-
cant main effects for testosterone level, F(1, 37) ? .07, p ? .80,
condition, F(1, 37) ? 2.58, p ? .12 nor the interaction, F(1, 37) ?
.34, p ? .56 on diastolic blood pressure when controlling for
baseline diastolic blood pressure. See Table 2 for descriptive
Based on evidence linking testosterone to dominant behaviors,
we hypothesized that basal testosterone level might function as a
biological correlate for desired status level. The mismatch effect
describes an incongruence between one’s desired level of status
and one’s actual level of status. The results from Study 2 replicated
those of Study 1, demonstrating that persons high in testosterone
show poor higher-order cognitive function when losing a domi-
nance battle, whereas persons low in testosterone show poor
higher-order cognitive function when winning a dominance battle.
Consistent with the PA result from Study 1, we found heart rate
increases in the mismatched conditions as well (although the
findings in Study 2 were a bit weaker than they were in Study 1).
We find it encouraging that the hypothesis is robust to fairly
dramatic alterations in status manipulations, as revealed by the fact
that the Study 2 results replicated those of Study 1. Math perfor-
mance showed the same pattern in Study 2 as analytic performance
in Study 1, and heart rate mirrored self-reported emotional arousal
results found in Study 1. Finally, we were relieved to find that the
testosterone concentrations obtained in Study 2 were well within
the range of typical values, in contrast to the somewhat high
concentrations obtained in Study 1. This, coupled with the fact that
the pattern of results from Study 1 replicated across to Study 2,
and lower third. The y-axis represents number correct on GRE quantitative (means adjusted controlling for SAT).
Error bars represent 1 standard error.
Study 2: The effects of status and testosterone on cognitive performance. T level is divided into upper
Descriptive Statistics for Study 2
* Means are adjusted controlling for Math SAT performance.
are adjusted controlling for baseline heart rate.
controlling for baseline systolic blood pressure.
controlling for baseline diastolic blood pressure.
*** Means are adjusted
**** Means are adjusted
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
suggests that the results obtained in both studies can be safely
attributed to testosterone.
In Study 3, our goal was to address a lingering question about
the psychological meaning of testosterone levels. We have argued
throughout this paper that testosterone levels are a proxy for a
person’s need for status. One remaining question is whether tes-
tosterone is a better predictor of status-seeking behavior than
scores on self-report measures of dominance. Although testoster-
one has been linked to implicit power and dominance strivings in
humans, few attempts have been made to link testosterone to
self-reports of dominance or status, and those that have demon-
strated modest relationships (e.g., Archer, Birring, & Wu, 1998).
This may be because questionnaire measures often ask about
“typical” behavior or traits, whereas as demonstrated by Sapolsky,
Wingfield, and others, testosterone activity is linked to behavior in
very specific situations and thus oftentimes bears no relationship to
behavior. Another possibility, suggested by Schultheiss’s work, is
that preferences and aversions to high status may exist outside of
conscious awareness. To date, few published reports linking self-
reported dominance to aggressive or status-related behaviors exist
(but see Diekmann, Tenbrusel, & Galinsky, 2003).
However, no study to date has examined whether these self-
report measures predict dominant behavior in situations in which
status is manipulated. We hypothesized that testosterone would
remain a better predictor than self-reports of behavior in these
situations. To test this hypothesis, we analyzed a subset of partic-
ipants from Studies 1 and 2, and compared the predictive power of
testosterone levels and self-reported need for dominance.
Participants for Study 3 were a subset of those in Studies 1 and 2 who
had completed self-report measures during a pretest session. This sample
(N ? 49) was 47% women, and the testosterone (T) means were similar to
those in the separate studies.
Materials and Procedure
During a pretest session, participants completed self-report measures
tapping into the need for dominance. The dominance subscale of the
Personality Research Form (PRF; Helmes & Jackson, 1977) contains 16
true/false items assessing one’s desire to hold power over others. The
Social Dominance Orientation scale (SDO; Pratto, Sidanius, Stallworth, &
Malle, 1994) contains 14 items (1–7) assessing one’s belief that certain
social groups should hold more power in our society.
Participants were then invited to the laboratory to take part in one of the
two studies described above. For the purposes of these analyses, we
combined participants from both studies, after converting the dependent
measures into z-scores. This conversion allowed us to generalize across
different distributions in the cognitive measures.
Correlations between T and Self-Reports
Consistent with our predictions and with previous null findings,
neither scale was significantly correlated with T levels. The cor-
relation between T and PRF scores was r(48) ? ?.073, p ? .62.
The correlation between T and SDO scores was r(49) ? ?.193,
p ? .18. Additionally, the correlation between SDO scores and
PRF scores was nonsignificant, r(49) ? .209, p ? .15.
Predicting Cognitive Performance
To compare the predictive power of T with that of the self-report
measures, we conducted a separate stepwise linear regression
using each self-report measure, with cognitive performance (stan-
dardized) as the dependent variable. In the first step, we entered T
levels, status condition, and scores on the self-report measure of
interest. In the second step, we entered the T ? status interaction
and the self-report ? status interaction. See Table 3 for the
regression coefficients for both of these analyses.
We first analyzed these relationships using PRF dominance
scores. The first step explained 4.1% of the variance (p ? .184).
Status condition was a marginally significant predictor (? ? .28;
p ? .054), but neither T levels nor PRF scores approached signif-
icance on their own. Adding the interactions in the second step
third. The y-axis represents heart rate during the test, controlling for baseline heart rate. Error bars represent 1
Study 2: The effects of testosterone and status on heart rate. T level is divided into upper and lower
explained an additional 9% (adj R2? 13%; p ? .045). The only
significant predictor was the T ? status interaction (? ? .32; p ?
.033). See the top half of Table 3 for the coefficients.
We next analyzed these relationships using SDO scores. The
first step explained 4.2% of the variance (p ? .182). Status
condition was a marginally significant predictor (? ? .27; p ?
.070), but neither T levels nor SDO scores approached significance
on their own. Adding the interactions in the second step explained
an additional 6% (adj R2? 10%; p ? .094). The only significant
predictor was the T ? status interaction (? ? .30; p ? .035). See
the bottom half of Table 3 for the coefficients.
Consistent with our predictions, testosterone appears to be a
better predictor of how individuals will respond when their status
is threatened than their self-reports of how much they desire status.
This suggests that testosterone levels may capture something
unique about an individual’s motives. As discussed in our intro-
duction to Study 3, self-report measures may not capture these
motives for several reasons. Namely, these motives may exist
outside of conscious awareness (cf. Schultheiss et al., 1999; 2003).
These motives may also be situational, and therefore not detected
by measures that ask about typical, trait-like behavior.
These reasons likely explain the lack of relationships with the
PRF dominance scale, but the SDO’s failure to predict behavior
may be due to additional factors. Since its introduction, the SDO
has distinguished itself as a remarkably robust and reliable predic-
tor of group-level dominance phenomena. However, testosterone’s
relationship to behavior appears to function on a more individual
level. The person who scores high on the SDO is concerned about
group-based hierarchies as well as his or her group’s status, and
presumably reacts to challenges and threats to the group. On the
other hand, the person high in testosterone is hypothesized to react
to personal status challenges within a group, and may or may not
have a care in the world about how that group is perceived, as long
as he or she is a high-ranking member of the group. Indeed, Pratto
et al. (1994) report that the SDO was uncorrelated with interper-
sonal dominance, and the same was true in our sample. So, it
makes sense that one’s SDO score should not predict behavior in
our experiments, in which status is manipulated as an interpersonal
construct. Would testosterone fail to predict status challenges if
these challenges were to one’s group? A promising direction for
future research would be to manipulate the nature of the status
challenge, thereby creating an individual or group-level challenge.
The data presented here provide the first evidence for what we
have termed the mismatch effect. Evidence from three studies
suggests that people have a strong preference or aversion to high
status positions, linked to their baseline levels of testosterone.
When one’s preferred level of status matches actual level in a
situation, one is evidently satisfied to remain at that level. How-
ever, when a mismatch between preferred and actual status level
occurs, one becomes aroused and motivated to regain the desired
level. A mismatched person also becomes emotionally distressed,
exhibits a hyper-vigilance toward status-related information, and
experiences a decline in complex cognitive processing. It is im-
portant that testosterone was shown to be a better predictor of
T vs. Self-Reports as Predictors of Cognitive Performance
Analysis #1: PRF Dominance Scores
Step 1 Beta (Std. Error)
PRF ? Status
T ? Status
Analysis #2: SDO Scores
Step 1 Beta (Std. Error)
SDO ? Status
T ? Status
coefficients; “?” represents standardized regression coefficients. N ? 49 for all analyses.
“PRF-Dom” refers to the dominance subscale of the PRF. “Beta” represents unstandardized regression
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
cognitive performance than was either of two self-report measures
of dominance. In this final section, we briefly outline some of the
theoretical implications of these data.
Testosterone and dominance.
distinguished scholars have argued that the behavioral link with
testosterone in humans is weak or even nonexistent (see, e.g., the
peer commentaries following the target article by Mazur & Booth,
1998). However, in the face of interactional effects such as those
shown in this paper as well as by Josephs et al. (2003) and
Newman et al. (2005), these claims are not surprising. Indeed,
whereas there is a large array of experimental testosterone studies
in the animal literature, the great majority of human studies is
correlational, and thus collapsed across situations. Only if a par-
ticipant’s behavior is situated in circumstances that “activate the
apparatus” (i.e., when status is challenged or threatened) should
one find strong and predictable effects. Only under these condi-
tions should testosterone relate to behavior. Indeed, as stated
earlier, ample evidence for an interactionist perspective comes
from other animal species. This animal research has determined
that only when status is uncertain, and there is a possibility of
imminent dominance battles, does testosterone relate to dominant
and aggressive behaviors (Collias et al., 2002; Ruiz-de-la-Torre &
Mediation of the mismatch effect.
intuitive sense that the cognitive deficit that is a consequence of
the mismatch effect should be mediated by one or more of the
intervening variables that were measured in these studies (e.g., PA,
NA, heart rate). For example, the relationship between mismatched
status and cognitive performance might be explained by changes in
negative affect. However, tests of mediation (Baron & Kenny,
1986) confirmed that neither partial nor full mediation was ob-
served for any of these variables. Why? One possibility is that
none of the variables included in these studies were mediators.
Another possibility might be that the test of mediation that was
used possessed low statistical power (see, e.g., Mackinnon, Lock-
wood, Hoffman, West, & Sheets, 2002 for a compelling empirical
confirmation of this, and for a comparison of 14 different tests of
mediation). Regardless, one promising direction for future re-
search might be to search for the mediators that are responsible not
only for changes in cognition, but also in motivation (e.g., PA),
affect (e.g., NA), and physiological arousal. Perhaps with the aid
of new and emerging technologies, such a search will bear fruit.
Testosterone as a personality variable.
associated with the current studies is the lack of multiple measure-
ments of the predictor variable. We are claiming that a particular
testosterone level reflects that participant’s baseline level of tes-
tosterone. Ideally, as with any personality measure, multiple mea-
sures would provide a more stable measurement base. However, it
is possible that a portion of the variance in some participants’
testosterone concentrations is the result of situationally induced
changes (e.g., Gonzalez-Bono, Salvador, Serrano, & Ricarte, 1999,
Gonzalez-Bono, Salvador, Ricarte, Serrano, & Arnedo, 2000; Jo-
sephs, Guinn, Harper, & Askari, 2001). Is this problematic? As
Dabbs (2000) and others have pointed out, chronic baseline indi-
vidual differences far outweigh the changes that arise as a result of
It is worth noting that some
It makes conceptual and
One obvious limitation
Rather, it has been demonstrated that in humans, testosterone
levels are relatively stable over time (e.g., Dabbs, 2000; Granger et
al., 1999; Sellers, Mehl, & Josephs, in press), and have high
predictive validity in certain situations—namely, those situations
under which status is uncertain (e.g., Sapolsky, 1991). These
characteristics suggest that baseline levels of testosterone meet the
basic criteria for a good personality variable (cf. John & Benet-
Martinez, 2000). Social psychology has a natural and understand-
able reluctance to measuring hormone levels and to treating them
as a stable individual difference variable. After all, it is difficult to
ask people “the degree to which you have high testosterone, on a
scale from 1 (not at all) to 5 (very much).” But a large body of data,
including those in the present study, suggests that testosterone
could shed new light on how people function in group settings.
Although highly speculative at this point, it is possible that after
an increase in status, a person’s physiology will gradually change
in concert to the new position. Indeed, Mazur and Booth (1998)
have argued that as status increases, so does testosterone. Accord-
ing to the reciprocal model of testosterone’s effects on dominant
behavior (Mazur & Booth, 1998), changes in circulating levels of
testosterone act as a buy or sell signal. When testosterone in-
creases, this informs the individual that the environment is safe for
further attempts at maintaining or enhancing status. When testos-
terone decreases, the individual is being flashed a sell signal and
should flee the situation to avoid further loss of status. These
changes in testosterone have likewise been found at the group level
when individuals felt that their victory was dependent on their
personal contribution (Gonzalez-Bono et al., 1999; 2000). It is
possible that fluctuations in testosterone may have more to do with
long-term success than initial basal levels of testosterone, but the
present studies only examined the latter. Further studies are needed
to parcel out the long-term predictive validity of each of these two
models. We are currently conducting experiments examining an-
drogen reactivity in the context of changing status conditions.
Archer, J. (in press). Testosterone and human behavior: An evaluation of
the challenge hypothesis. Neuroscience and Biobehavioral Reviews.
Archer, J., Birring, S. S., & Wu, F. C. W. (1998). The association between
testosterone and aggression among young men: Empirical findings and
a meta-analysis. Aggressive Behavior, 24, 411–420.
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable
distinction in social psychological research: Conceptual, strategic and
statistical considerations. Journal of Personality and Social Psychology,
Beaver, B. V., & Amoss, M. S. (1982). Aggressive behavior associated
with naturally elevated serum testosterone in mares. Applied Animal
Ethology, 8, 425–428.
Bhasin, S., Woodhouse, L., & Storer, T. W. (2001). Proof of the effect of
testosterone on skeletal muscle. Journal of Endocrinology, 170, 27–38.
Briganti, F., Seta, D. D., Fontani, G., Lodi, L, & Lupo, C. (2003).
Behavioral effects of testosterone in relation to social rank in the male
rabbit. Aggressive Behavior, 29, 269–278.
Buss, D. M. (2003). Evolutionary psychology: The new science of the mind.
New York: Pearson, Allyn & Bacon.
Cashdan, E. (1995). Hormones, sex, and status in women. Hormones &
Behavior, 29, 354–366.
Cavigelli, S. A., & Pereira, M. E. (2000). Mating season aggression and
fecal testosterone levels in male ring-tailed lemurs (Lemur catta). Hor-
mones & Behavior, 37, 246–255.
Coe, C. L., Mendoza, S. P., & Levine, S. (1979). Social status constrains
the stress response in the squirrel monkey. Physiology & Behavior, 23,
Collias, N. E., Barfield, R. J., & Tarvyd, E. S. (2002). Testosterone versus
psychological castration in the expression of dominance, territoriality
and breeding behavior by male village weavers (Ploceus cucullatus).
Behaviour, 139, 801–824.
Dabbs, J. M., Jr. (2000). Salivary testosterone measurements: Reliability
across hours, days, and weeks. Physiology & Behavior, 48, 83–86.
Dabbs, J. M., Jr., Alford, E. C., & Fielden, J. A. (1998). Trial lawyers and
testosterone: Blue-collar talent in a white-collar world. Journal of Ap-
plied Social Psychology, 28, 84–94.
Dabbs, J. M., Jr., de la Rue, D., & Williams, P. M. (1990). Testosterone and
occupational choice: Actors, ministers, and other men. Journal of Per-
sonality and Social Psychology, 59, 1261–1265.
Deikmann, K. A., Tenbrunsel, A. E., & Galinsky, A. D. (2003). From
self-prediction to self-defeat: Behavioral forecasting, self-fulfilling
prophecies, and the effect of competitive expectations. Journal of Per-
sonality and Social Psychology, 85, 672–683.
Eagly, A. H., & Wood, W. (1999). The origin of sex differences in human
behavior: Evolved dispositions versus social roles. American Psycholo-
gist, 54, 408–423.
Elofsson, U. O. E., Mayer, I., Damsga ˆrd, B., & Winberg, S. (2000).
Intermale competition in sexually mature arctic charr: Effects on brain
monoamines, endocrine stress responses, sex hormone levels, and be-
havior. General and Comparative Endocrinology, 118, 450–460.
Giesler, R. B., Josephs, R. A., & Swann, W. B., Jr. (1996). Self-verification
in clinical depression: The desire for negative evaluation. Journal of
Abnormal Psychology, 105, 358–368.
Gonzalez-Bono, E., Salvador, A., Ricarte, J., Serrano, M. A., & Arnedo, M.
(2000). Testosterone and attribution of successful competition. Aggres-
sive Behavior, 26, 235–240.
Gonzalez-Bono, E., Salvador, A., Serrano, M. A., & Ricarte, J. (1999).
Testosterone, cortisol and mood in a sports team competition. Hormones
& Behavior, 35, 55–62.
Granger, D., Schwartz, E., Booth, A., & Arentz, M. (1999). Salivary
testosterone determination in studies of child health and development.
Hormones & Behavior, 35, 18–27.
Grant, V. J., & France, J. T. (2001). Dominance and testosterone in women.
Biological Psychology, 58, 41–47.
Harmon-Jones, E., Vaughn, K., Mohr, S., Sigelman, J., & Harmon-Jones,
C. (2004). The effect of manipulated sympathy and anger on left and
right frontal cortical activity. Emotion, 4, 95–101.
Helmes, E., & Jackson, D. N. (1977). The item factor structure of the
Personality Research Form. Applied Psychological Measurement, 1,
John, O. P., & Benet-Martinez, V. (2000). Measurement: Reliability,
construct validation, and scale construction. In H. T. Reis, & C. M. Judd
(Eds.). Handbook of research methods in social and personality psy-
chology (pp. 339–369). New York: Cambridge University Press.
Josephs, R. A., Guinn, J. S., Harper, M. L., & Askari, F. (2001). Liquorice
consumption and salivary testosterone concentrations. The Lancet, 358,
Josephs, R. A., Newman, M. L., Brown, R. P., & Beer, J. M. (2003). Status,
testosterone, and human intellectual performance: Stereotype threat as
status concern. Psychological Science, 14, 158–163.
Kemper, T. D. (1990). Social structure and testosterone: Explorations of
the socio-bio-social chain. New Brunswick, NJ: Rutgers University
Koyama, S., & Kamimura, S. (1999). Lowered sperm motility in subordi-
nate social status of mice. Physiology & Behavior, 65, 665–669.
Kraus, C., Heistermann, M., & Kappeler, P. M. (1999). Physiological
suppression of sexual function of subordinate males: A subtle form of
intrasexual competition among male sifaka (Propithecus verreauxi)?
Physiology & Behavior, 66, 855–861.
MacKinnon, D. P., Lockwood, C. M., Hoffman, J. M., West, S. G., &
Sheets, V. (2002). A comparison of methods to test mediation and other
intervening variable effects. Psychological Methods, 7, 83–104.
Mazur, A., & Booth, A. (1998). Testosterone and dominance in men.
Behavioral & Brain Sciences, 21, 353–397.
Mazur, A., Booth, A., & Dabbs, J., Jr. (1992). Testosterone and chess
competition. Social Psychology Quarterly, 55, 70–77.
Mazur, A., Susman, E., & Edelbrock, S. (1997). Sex differences in testos-
terone responses to a video game competition. Evolution and Human
Behavior, 18, 317–327.
McCaul, K., Gladue, B., & Joppa, M. (1992). Winning, losing, mood, and
testosterone. Hormones & Behavior, 26, 486–506.
Morgan, D., Grant, K. A., Prioleau, O. A., Nader, S. H., Kaplan, J. R., &
Nadar, M. A. (2000). Predictors of Social Status in Cynomolgus Mon-
keys (macaca fascicularis) after group formation. American Journal of
Primatology, 52, 115–131.
Newman, M. L., Sellers, J. G., & Josephs, R. A. (2005). Testosterone,
cognition, and social status. Hormones & Behavior, 47, 205–211.
Oliveira, R. F., Almada, V. C., & Canario, A. V. M. (1996). Social
modulation of sex steroid concentrations in the urine of male cichlid fish
Oreochromis mossambicus. Hormones & Behavior, 30, 2–12.
Ostner, J., Kappeler, P. M., & Heistermann, M. (2002). Seasonal variation
and social correlates of androgen excretion in male redfronted lemurs
(Eulemur fulvus rufus). Behavioral Ecology and Sociobiology, 52, 485–
Pratto, F., Sidanius, J., Stallworth, L., & Malle, B. (1994). Social domi-
nance orientation: A personality variable predicting social and political
attitudes. Journal of Personality and Social Psychology, 67, 741–763.
Rohwer, S. (1977). Status signaling in Harris sparrows. Behaviour, 61,
Ruiz-de-la-Torre, J. L., & Manteca, X. (1999). Effects of testosterone on
aggressive behaviour after social mixing in male lambs. Physiology &
Behavior, 68, 109–113.
Sapolsky, R. M. (1991). Testicular function, social rank and personality
among wild baboons. Psychoneuroendocrinology, 16, 281–293.
Scaramella, T., & Brown, W. (1978). Serum testosterone and aggression in
hockey players. Psychosomatic Medicine, 40, 262–263.
Schlenker, B. R., & Weigold, M. F. (1990). Self-consciousness and self-
presentation: Being autonomous versus appearing autonomous. Journal
of Personality and Social Psychology, 59, 820–828.
Schultheiss, O. C. (in press). A biobehavioral model of implicit power
motivation arousal, reward and frustration. In E. Harmon-Jones and P.
Winkielman (Eds.), Fundamentals of social neuroscience. New York:
Schultheiss, O. C., Campbell, K. L., & McClelland, D. C. (1999). Implicit
power motivation moderates men’s testosterone responses to imagined
and real dominance success. Hormones & Behavior, 36, 234–241.
Schultheiss, O. C., Dargel, A., & Rohde, W. (2003). Implicit motives and
gonadal steroid hormones: Effects of menstrual cycle phase, oral con-
traceptive use, and relationship status. Hormones & Behavior, 43, 293–
Sellers, J. G., Mehl, M. R., & Josephs, R. A. (in press). Hormones and
personality: Testosterone as a marker of individual differences. Journal
of Research in Personality.
Shapiro, D., Jamner, L. D., Lane, J. D., Light, K. C., Myrtek, M., Sawada,
Y., & Steptoe, A. (1996). Blood pressure publication guidelines. Psy-
chophysiology, 33, 1–12.
Tsai, L. W., & Sapolsky, R. M. (1996). Rapid stimulatory effects of
testosterone upon myotubule metabolism and sugar transport, as as-
sessed by silicon microphysiometry. Aggressive Behavior, 22, 357–364.
van Honk, J., Tuiten, A., Hermans, E., Putman, P., Koppeschaar, H.,
Thijssen, J., Verbaten, R., & van Doornen, L. (2001). A single admin-
JOSEPHS, SELLERS, NEWMAN, AND MEHTA
istration of testosterone induces cardiac accelerative responses to angry
faces in healthy young women. Behavioral Neuroscience, 115, 238–242.
van Honk, J., Tuiten, A., Verbaten, R., van den Hout, M., Koppeschaar, H.,
Thijseen, J., & de Haan, E. (1999). Correlations among salivary testos-
terone, mood, and selective attention to threat in humans. Hormones &
Behavior, 36, 17–24.
Walsh, M. (1991). Butler’s lives of the saints: Concise ed., revised and
updated. San Francisco: Harper.
Watson, D. (2000). Mood and temperament. New York: Guilford Press.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and vali-
dation of brief measures of positive and negative affect: The PANAS
scales. Journal of Personality and Social Psychology, 54, 1063–1070.
Webster, D. M., & Kruglanski, A. W. (1994). Individual differences in
need for cognitive closure. Journal of Personality and Social Psychol-
ogy, 67, 1049–1062.
Wirth, M., Schultheiss, O., & Welsh, K. (in press). Salivary cortisol
changes in humans after winning or losing a dominance contest depend
on implicit power motivation. Hormones & Behavior.
Wright, R. A., Martin, R. E., & Bland, J. L. (2003). Energy resource
depletion, task difficulty, and cardiovascular response to a mental arith-
metic challenge. Psychophysiology, 40, 98–105.
Zajonc, R. B., Heingartner, A., & Herman, E. M. (1969). Social enhance-
ment and impairment of performance in the cockroach. Journal of
Personality and Social Psychology, 13, 83–92.
Received March 1, 2004
Revision received December 1, 2005
Accepted December 19, 2005 ?