TO FIGHT OR NOT TO FIGHT:
DOES CONSPECIFIC STRENGTH INFLUENCE DEFENSIVE SIGNALING?
By
David J. Johnson

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Psychology – Master of Arts
2014

ABSTRACT
TO FIGHT OR NOT TO FIGHT:
DOES CONSPECIFIC STRENGTH INFLUENCE DEFENSIVE SIGNALING?
By
David J. Johnson
Humans and social animals show similar responses to defensive threats such as the presence of
predators or rival conspecifics (Blanchard et al., 2001). The current work tested two extensions
of this research: first, whether humans show similar assessment processes compared to nonhuman animals including dynamically updating their assessments based on new information, and
second, whether humans send different signals (i.e., willingness to escalate or submission) based
on differences in physical formidability and whether those signals have behavioral consequences.
Using an experimental procedure where randomly paired same-sex naïve participants competed
against one another in a physical task, the current experiment revealed evidence consistent with
assessment; participants became more accurate in their judgments of strength after gaining
information from a physical contest. In contrast, participants did not send different signals based
on differences in formidability, insofar as those signals were broadcasted by changes in strength.
Implications of using animal models to predict human defensive behaviors are discussed, as well
as relevant connections to game theory.

TABLE OF CONTENTS
LIST OF TABLES ......................................................................................................................... iv
INTRODUCTION .......................................................................................................................... 1
Animal Assessment and Defensive Behaviors ................................................................... 2
Theoretical Accounts .............................................................................................. 2
Experimental Evidence ........................................................................................... 4
Human Assessment and Defensive Behaviors .................................................................... 7
Pilot Study ......................................................................................................................... 10
The Present Research ........................................................................................................ 12
METHOD ..................................................................................................................................... 14
Participants ........................................................................................................................ 14
Measurements ................................................................................................................... 14
Procedure .......................................................................................................................... 14
RESULTS ..................................................................................................................................... 17
Planned Comparisons........................................................................................................ 17
Signaled Strength .................................................................................................. 17
Diagnostic Analyses.............................................................................................. 19
Competition Outcomes ......................................................................................... 21
Exploratory Analyses ........................................................................................................ 22
Updating Strength Assessments ............................................................................ 22
Accuracy of Updated Assessments ....................................................................... 23
DISCUSSION ............................................................................................................................... 27
Assessment Accuracy ....................................................................................................... 27
Defensive Signaling .......................................................................................................... 28
Implications for Game Theory .......................................................................................... 29
Similarities to Animal Assessment ................................................................................... 31
Conclusion ........................................................................................................................ 32
APPENDIX ................................................................................................................................... 33
REFERENCES ............................................................................................................................. 40
	
  

iii

LIST OF TABLES
Table 1: Descriptive Statistics for Baseline Strength Measures ................................................... 34
Table 2: Intraclass Correlations Between Strength Measures ...................................................... 35
Table 3: Multiple Regression Results Predicting Success in Arm Wrestling Competition From
Strength Measures ......................................................................................................................... 36
Table 4: Multilevel Multiple Regression Results Predicting Updated Strength Assessments From
Past Strength Assessments and Competition Outcomes ............................................................... 37
Table 5:	
  Over-time APIM Model Predicting Relative Strength Assessment From Strength
Measurements Before and After Competition .............................................................................. 38
Table 6: Over-time APIM Model Predicting Fight Outcomes From Strength Measurements
Before and After Competition ...................................................................................................... 39

iv

INTRODUCTION
How do individuals signal for interactions with physically formidable others? Research
on non-human animals characterizes defensive behavior as driven by an interaction between
ability and context given the presence of a threat (D. C. Blanchard, 1997). In this view, defensive
behaviors are the product of an evolved computational process that takes into account contextual
variables in order to prepare an animal to act in ways that will afford successful defense from a
threat (D. C. Blanchard, Griebel, Pobbe, & R. J. Blanchard, 2011; Gawronski & Cesario, 2013).
In a competitive context, certain actions will be more or less appropriate given the unique
discrepancy in ability between individuals. Game theory (Maynard Smith, 1974; Maynard Smith
& Parker, 1976; Maynard Smith & Price, 1973; Parker, 1974) provides a useful model for
predicting what behaviors will occur in such situations. According to this framework animals
should be sensitive to differences in physical formidability, or resource holding potential (RHP;
Parker, 1974). When discrepancies in RHP are large, more formidable opponents should signal
willingness to escalate conflict and weaker opponents should signal submission or withdraw
from conflict.
As a part of a growing body of work demonstrating commonalities between human and
animal defensive behaviors (e.g., D. C. Blanchard, Hynd, Minke, Minemoto, & R. J. Blanchard,
2001; Sell et al., 2009) I propose that the influence a person has on another’s defensive signaling
is partially dependent on the discrepancy in formidability between the two. In competitive
contexts, stronger individuals will signal readiness to escalate conflict while weaker individuals
will signal submission. Critically, these signals will be indexed through changes in strength
relative to strength measured when the individual is alone. Weaker individuals will show

1

submission by demonstrating less strength, while stronger individuals will display willingness to
engage in conflict by demonstrating greater strength.
Animal Assessment and Defensive Behaviors
Theoretical accounts.
Animals in social species such as humans have historically faced recurrent threats from
both predators and rival conspecifics (same-species organisms). Although there are important
differences between threats from predators and conspecifics (e.g., predator-prey relationships are
typically characterized by large asymmetries in formidability), one commonality across these
domains is that aggression should only occur in limited circumstances because of the high costs
attached to it, namely risk of injury or death (Parker, 1974). These costs are elevated when an
individual is faced with a more formidable conspecific (or predator). An ability to accurately
gauge individual differences in formidability would provide an advantage in determining
whether to escalate or withdraw from conflicts (Gawronski & Cesario, 2013; Griskevicius et al.,
2009; Sell et al., 2009). In favor of this hypothesis, the existence of reliable cues to physical
formidability, including (but not limited to) size, weight, and weaponry, are well documented
amongst non-human animals (cf. Arnott & Elwood, 2010).
From a game theoretical perspective (Maynard Smith, 1974; Maynard Smith & Parker,
1976; Maynard Smith & Price, 1973) every organism has a specific level of fighting ability, or
resource holding potential (RHP; Parker, 1974). RHP is influenced by several factors including
size, strength, weaponry, group size, and experience. Each factor serves as a partial cue to the
individual’s absolute fighting ability. All other things equal, an organism with greater RHP than
his or her opponent has a higher probability of winning a physical fight. According to Parker
(1974), when discrepancies in RHP are large, more formidable opponents should signal to

2

escalate and weaker opponents to withdraw. Fights should only occur where both opponents
escalate; this is more likely to occur as differences in RHP approach zero.
It is important to note that all that is meant by assessment is that an organism reacts
differentially based on the magnitude of a given threat (Parker, 1974). Conscious intent is not
required; rather, actions prepared by the organism are those favored by selection. Historically,
organisms that successfully defended themselves from interpersonal threats had better
reproductive success. These behavioral tendencies were inherited by their offspring, over time
developing into neural mechanisms that process information about the magnitude of threats.
These mechanisms activate a set of behaviors appropriate to that unique interaction (Maynard
Smith & Parker, 1976; Parker, 1974), which result in differential responding partially based on
relative threat level (i.e., differences in RHP).
How long should an organism assess a threat before preparing an appropriate behavioral
response? In betta fish, lateral displays of body size are given for up to several minutes until one
opponent withdraws (Simpson, 1968). Similarly, red deer typically engage in roaring contests
(where vocalizations accurately index fighting ability) for several minutes before decisions to
engage or withdraw are made (Clutton-Brock & Albon, 1979). Although game theory predicts
that extended bouts of assessment give increasingly accurate estimations of RHP, they are also
more costly (Dawkins & Guilford, 1991; Maynard Smith & Price, 1973). Animals that signal for
long periods of time or engage in prolonged assessment of others’ signals are at higher risk of
incurring injury from opponents or predators drawn to the displays. In addition, animals also lose
valuable time to pursue other desirable resources or mates. Consider the case of a bird that sings
to deter others from invading his or her territory. The song puts not only the signaler at risk for
predation, but also increases risk for those close enough to hear the warning song.

3

Often there is a trade-off between assessment length and accuracy. When costs of losing
a conflict are high and the costs assessment are low, prolonged assessment may be beneficial
(Dawkins & Guilford, 1991; Maynard Smith, 1982). When observing a rival from a distance, the
only costs of assessment would likely be the time and energy lost in pursuit of other activities.
These low costs might encourage protracted assessment. In contrast, approaching a rival might
increase assessment accuracy beyond what could be observed from a distance, but at larger risk
of potential injury, especially the longer the animal remains close to the threat. However, these
situations are likely to vary widely across species, and are dependent on assessment accuracy. In
some cases (e.g., close proximity, high cost situations) even if extensive assessments are more
accurate, it may be more cost effective to make quicker, but less accurate decisions about the
formidability of conspecifics. The value of such assessments would increase as their accuracy
approaches that of prolonged assessments.
Similarly, the degree to which assessment is accurate is a key determinant of what actions
are selected in response to threats. As assessment accuracy decreases, fighting is more likely to
occur between conspecifics (Maynard Smith & Parker, 1976; Parker, 1974). Recall that when
there is a discrepancy in RHP the more formidable opponent should signal to escalate and the
weaker one to withdraw. Only when the difference is small (or nonexistent), is it likely that both
opponents will determine that they are more formidable, resulting in mutual escalation and
physical conflict. However, when assessment is inaccurate, it is harder to determine disparities in
RHP. This increases the margin of error for determining when escalation is appropriate, resulting
in increased conflict. In contrast, when assessment is accurate, conflicts occur less often, as it is
easier to gauge the probability of winning a bout (Maynard Smith & Parker, 1976; Parker, 1974).
Experimental evidence.

4

The purpose of assessment is to determine what kind of action will most likely to lead to
a successful defense from a threat (D. C. Blanchard et al., 2011). Assessment is only useful if the
actions prepared in response to the threat benefit the assessor. Therefore, one should expect
different responses to threats in non-human animals based on the magnitude of the threat and the
context in which the threat is encountered. Indeed, hyenas and other social species respond
differently to threats based on their magnitude (Benson-Amram, Heinen, Dryer, & Holekamp,
2011). When exposed to calls from unfamiliar conspecifics, hyena behavior varied flexibly based
on the ratio of intruders to allies. They demonstrated more vigilance to larger groups of
outsiders, and approach behavior was contingent upon having a numerical advantage. Hyenas
demonstrate a sophisticated system of threat assessment that takes into account the formidability
of both allies and intruders and chooses appropriate actions accordingly.
Similarly, research on rodents also provides support for differential responding based on
individual ability and context. The presence of an ambiguous threat (a threat of unclear RHP)
encourages orientation towards, and investigation of, the threat. In contrast, the presence of a
clear threat (e.g., a predator with much higher RHP) elicits different responses based on the
physical context. At a large distance threats elicit flight if escape is possible and freezing if it is
not. At closer distances rodents give defensive “threats” (e.g., displays of weaponry), and at even
closer distances, they engage in defensive attack (D. C. Blanchard et al., 1997; D. C. Blanchard,
2011). When at a clear disadvantage, rodents seek to withdraw from conflict, and only resort to
escalation if all other options have been exhausted.
In addition to the defensive behaviors highlighted by D. C. and R. J. Blanchard (D. C.
Blanchard, 1997; D.C. Blanchard & R. J. Blanchard, 2003), submissive behaviors are also
common in social animals such as primates (Bernstein & Gordon, 1980; D. C. Blanchard et al.,

5

2011). These signals are most likely to occur when the value gained from continuing a contest is
lower than the potential costs of continuing the fight (Matsumura & Hayden, 2006). Typically,
this occurs when discrepancy in RHP is slight: accepting submission is advantageous because
opponents are likely to retaliate if attacks are continued. For this reason, submissive signals are
only evolutionarily stable when assessment is accurate enough to distinguish subtle differences
in formidability. In contrast, submissive signals are likely to be ignored when the difference
between opponents is large, because the benefits of continued attack (e.g., asserting dominance)
outweigh the risk of injury (Matsumura & Hayden, 2006).
Submission is directly related to dominance and the formation of hierarchies in animals.
Many animals form dominance relationships, including insects, fish, birds and mammals (cf.
Chase & Seitz, 2011). Although a review of dominance hierarchies is beyond the scope of this
proposal, certain aspects are relevant to understanding submissive behaviors. In particular,
primates have a complex system of hierarchy largely based on a rank order of submission; the
most formidable primate submits to no one and the weakest to everyone. Rhesus monkeys
simultaneously introduced to each other will compete for dominance but typically settle into a
stable hierarchy within an hour (Bernstein & Gordon, 1980; Bernstein, Gordon, & Rose, 1974),
suggesting that assessment is a relatively quick process. Interestingly, lone male rhesus monkeys
introduced to an established group immediately give submissive signals and assume the lowest
position in a hierarchy (Bernstein & Gordon, 1980). This may seem contradictory if the intruder
is more formidable than some members of the group. However, chimpanzees are known to solicit
and receive help from conspecifics during conflicts (de Waal & Hoekstra, 1980). Therefore, in
any given group, the combined RHP of all or multiple members will almost always outnumber

6

the intruder’s RHP by several magnitudes, making submission a reasonable behavior (Bernstein
& Gordon, 1980; Matsumura & Hayden, 2006).
In sum, many non-human animals show an ability to accurately assess formidability in
both predators and conspecifics. Longer assessments may be more accurate, but have higher
costs. Selection should favor decision rules that flexibly execute accurate or quick assessments
depending on contextual contingencies, such as the magnitude of the threat. All other things
equal, accurate assessment limits physical conflict to opponents with similar RHP. When
discrepancies in RHP are large, weaker animals should signal withdraw by fleeing or submitting,
whereas stronger animals should signal willingness to fight. Critically, the neural mechanisms
that orchestrate assessment should respond to contextual differences, including the formidability
of opponents, the presence of allies, and the physical situation where the encounter occurs.
Human Assessment and Defensive Behaviors
Until recently defensive behaviors in animals have typically been characterized by
psychologists as inflexible and innate (Gawronski & Cesario, 2013), whereas human aggression
was thought of as a learned behavior unique to the species (D. C. Blanchard et al., 2001). This
distinction is partially due to the constrained meaning of aggression typically applied to humans
versus the more general definition used in the biological sciences. In humans, aggression is
typically thought of as a learned hostile behavior with the intent to inflict pain. However, a more
general definition, applicable across species, regards aggression as behavior associated with
physical attack or escalation towards attack, without regards to intent (Sell, Hone, & Pound,
2012; van Staaden, Searcy, & Hanlon, 2011). According to the former definition, aggression is a
uniquely human phenomenon requiring conscious intent and hostile motives. According to the
latter, aggression just one of many defensive behaviors employed across species. Although not

7

ignoring the likely possibility that certain aspects of aggression are unique to humans, this
definition embraces the possibility that similar selection pressures shaped the defensive
behaviors of humans and non-human animals. From this perspective, the same computational
processes of assessment that occur in non-human animals provide valuable hypotheses about the
effect of context on human defensive behavior.
For the computational models of threat assessment to apply to human defensive behavior,
humans must first demonstrate an ability to accurately gauge their own formidability as well as
that of others. As with non-human animals, both the accuracy and speed of assessment is critical
for understanding what actions humans will signal in conflict. Paralleling findings in the animal
literature, humans are able to both quickly and accurately assess physical formidability. In a
series of experiments, Sell and colleagues (2009, 2010) demonstrated that humans were able to
assess formidability (as defined by upper body strength) from mere facial photos or voice
recordings. Raters were more accurate at determining male strength than female strength, which
is at least partially due to lower variability in female strength due to sexual dimorphism (Sell et
al., 2012). These results have also been replicated by other researchers (Archer & Thanzami,
2009). Sell and colleagues results suggest that humans can quickly and accurately assess
formidability, even from relatively impoverished information like pictures or voice recordings.
Although it may seem obvious that humans prepare different actions based on the
magnitude of a threat and the context in which the threat occurs, it is important to question if the
behaviors exhibited by non-human animals under threat match human behaviors under similar
conditions. In an experimental study, D. C. Blanchard and colleagues (2001) asked individuals to
report how they would respond in several threatening situations, which systematically varied the
level of threat and ability to escape. Human threat responses closely paralleled rodent threat

8

responses. When threats were ambiguous (e.g., hearing a suspicious noise) the most reported
behavioral choice was to investigate the threat. In contrast, when threats were clear (e.g., being
attacked) defensive responses were most common. These defensive responses did vary somewhat
across the sexes, such that males were more prone to report defensive attack in the context of a
clear threat, while females were more likely to report defensive threat (i.e. screaming). These
could reflect a higher percentage of males assessing that they were more formidable than the
threat and thus a retaliatory attack would be most advantageous. Interestingly, D. C. Blanchard
and colleagues (2001) noted that one limitation to this research was not including a response
option for submissive or pleading behaviors, noting while that these responses are unlikely in
rodents1 they would be in the range of possibility for animals such as primates or humans.
Due to the historical focus of studying aggression as a learned phenomenon, little
research has investigated the role of RHP disparities on defensive signaling. Moreover, the
studies that have done so have typically relied upon hypothetical conflict situations (Archer &
Benson, 2008; D. C. Blanchard et al., 2001). Although studies examining responses to
hypothetical threats are advantageous in that they can manipulate threat in ways that would not
be ethical to examine otherwise, they are limited by relying on prospective self-report, which
may not accurately reflect actual behaviors chosen in a conflict situation. For example, males
may report being less likely to scream than females when threatened due to self-presentation
concerns, rather than actual differences in behavior.
A stronger test of whether RHP influences defensive signaling would be to actually set up
dyadic interactions between individuals of varying RHP and observe how differences in

1

The definition of submission used in this article excludes behaviors such as “playing dead”
which are assumed to be qualitatively different than submission, as they may be used by animals
to evade predation but also to lure opponents closer.
9

formidability influence these signals. Changes in upper body strength in response to a partner
could serve as an index of defensive signaling. Although relatively stable, upper body strength
can be reliably increased or decreased by a variety of factors, including motivation and threat
(Ikai & Steinhaus, 1961). Insofar as submission is indexed by showing weakness (i.e., less
strength) and escalation is broadcasted by exaggerating those cues (i.e., more strength),
individuals may change their strength dependent on the unique relationship between partners,
specifically the degree in which their RHP differs.
Pilot Study
The aforementioned hypotheses were pilot tested in an unpublished study conducted by
Cesario and Johnson (2013). Specifically, male undergraduates were randomly paired with one
of five male confederates of varying physical formidability. Confederates were blind to study
hypotheses. Participants were told that they would compete against each other at a later point in
time. They were instructed to stand while strength was measured, with the idea being that this
might encourage natural assessment of their partners’ formidability. The experimenter measured
participants’ upper body strength using an inverted hand dynamometer. Measurements were
taken with the participant’s partner turned around, so participants could not observe their
partner’s measured strength. However, participants could to see their own measured strength.
To initially examine men’s defensive signaling as a function of confederate strength, a
single factor ANOVA was first conducted with confederate as predictor of participant upper
body strength. This yielded a significant effect for confederate F(4, 116) = 2.74, p = .032, η2 =
.086. Participant upper body strength was lower in the presence of a stronger confederate, and
higher in the presence of a weaker confederate. Although these results are promising, they ignore
the dyadic nature of the data. Because the hypotheses concerned whether one’s own ability and

10

the confederate’s ability would influence strength, the data were analyzed with a one-with-many
model (OWM; Marcus, Kashy, & Baldwin, 2009).
The OWM design is a variance decomposition model that can be used with hierarchically
structured data in which multiple individuals (the many; participants in this case) interact with or
are tied to the same partner (the one; confederates in this case). The model separates confederate
variance from participant variance. In a reciprocal OWM design both the confederates and the
participants have an outcome score for each dyadic combination (i.e., strength), and the
percentage of variance at the confederate and participant levels can be estimated for both.
Confederate-level variance for the measure of the confederate’s strength estimates the extent to
which a confederate is consistently strong (or weak) across participants, and confederate-level
variance for the measure of the participants’ strength estimates the extent to which all
participants paired with the same confederate are consistently strong or weak. The correlation
between these two confederate-level effects measures generalized reciprocity or the extent to
which participants paired with stronger confederates consistently show strong stronger (or
weaker) behavior than do participants who are tied to weaker confederates.
Multilevel modeling with restricted maximum likelihood was used to estimate a OWM
model assessing the reciprocal effects of confederate grip strength on participant grip strength
and vice versa. The key finding was that the generalized reciprocity correlation was significant,
substantial, and negative (r = -.83, p = .008). The presence of a strong confederate tended to
attenuate participants’ strength, such that the mean strength for participants paired with that
confederate was lower than the mean strength across all confederates. Conversely, the presence
of a weak confederate tended to enhance participants strength, such that the mean participant
strength for that confederate was higher than the mean strength across all confederates.

11

The model also revealed that a considerable amount of the variance in confederate
strength was at the participant or dyad level, which accounted for 38% (p < .001) of the variance
in confederates’ strength. This indicates that confederates’ strength changed depending on the
participant they interacted with. Combined with the significant generalized reciprocity
correlation, these results replicate the findings from the single factor ANOVA while
demonstrating that participants also influence confederate’s strength. These findings are
consistent with the hypothesis that humans assess the formidability of conspecifics and
automatically signal actions that will lead to successful defense (e.g., escalation, submission).
The Present Research
Drawing on research from both human and non-human animals, as well as the pilot study
described above, the current research tested whether physical formidability influences defensive
signaling in humans. Consistent with prior research (Sell et al., 2010; 2009), I predicted that (H1)
humans would be able to accurately gauge the strength of people they interact with. However,
because there is less variation in upper body strength between females (Sell et al., 2012), the
relationship between formidability and defensive signaling would be stronger for males rather
than females.
Second, assessment will alert individuals to the degree of RHP discrepancy between them
and the other competitor. Interacting with a competitor who is much stronger (weaker) would
elicit signals of submission (willingness to escalate conflict; H2). By signaling I refer to changes
in upper body strength from baseline, when the other competitor is not present. As evidenced in
the pilot study, signaling should not require explicit comparisons of formidability (e.g.,
displaying scores on upper body strength measures), although such a manipulation might
certainly enhance the effect.

12

Finally, changes in signaled strength should predict success in a physical competitive task
over and above any effects due to baseline differences in strength (H3). Individuals who signal
submission should lose more than individuals who signal willingness to escalate conflict.

13

METHOD
Same-sex pairs of participants reported separately to the lab and had baseline
measurements of strength taken by a same-sex confederate. Participants were then brought
together and had their strength measured again. Finally they competed against each other in
contest requiring upper body strength. Throughout the experiment participants answered a
variety of questions related to perceptions of self and partner strength.
Participants
Participants were 398 Michigan State undergraduate students (196 women; 24.7% nonWhite) who participated together in dyads (98 female dyads, 101 male dyads). The experiment
was advertised as two separate experiments to avoid participants signing up with friends and to
ensure random sampling.
Measurements
Participants’ left hand, right hand, and upper body strength were measured via handgrip
dynamometer. Each strength measurement was taken three times, and the average of these scores
was used in analyses. Height and arm length were measured in centimeters. Because variation in
bicep circumference was smaller than the former two, it was measured in millimeters.
Procedure
The first participant arrived for an experimental session fifteen minutes prior to the
second participant. Participants were told that the current experiment was interested with how
physical strength related to personality. The experimenter then took baseline measurements of
upper body strength and other body measurements. Measurements were not shared with
participants. The participant was then escorted to an individual suite and given instructions to
complete measures related to anger and formidability (Sell et al., 2009) unrelated to the current

14

experiment. After leaving the first participant, the experimenter escorted the second participant
into a different suite, where he or she repeated the same procedures.
When both participants finished the measures, the experimenter brought them together
into a large room in the lab and repeated the experimental cover story, elaborating that the two
participants would later compete against each other in a physical task. Before the competition,
the experimenter took the strength measurements again, this time with the other participant in the
room. Measurements were taken individually with the hand dynamometer while their competitor
faced away. Experimenters measured participants in the order that they arrived.
After measuring the strength of both participants, participants were escorted to two sideby-side computers. They then completed several measures concerned with relative judgments of
strength (e.g., how strong are you relative to the other participant?) and predicted fight outcomes
(e.g., how likely do you think you would be to win a physical fight against your partner?).
After both participants completed these measures they competed against each other in an
arm wrestling contest (156 dyads) or a mercy contest (43 dyads).2 Each participant was given
three tickets into a raffle for a $50 gift card. For each round they won, participants were given
one of their competitor’s tickets. In the arm wrestling task, participants used their right hand to
arm wrestle against their opponent. Participants won matches if they managed to push their
competitors arm down against the table. In the mercy contest, participants interlocked the fingers
on both their hands with their competitor’s fingers. They then squeezed as hard as could in order
to try to get the other participant to give in. In both contests, if a winner was not determined in
less than 30 seconds, the match ended in a draw. Each dyad competed in up to three rounds,

2

The competitive task was changed from a mercy task to an arm wrestling task because the
former did not effectively differentiate between strong and weak participants; most matches
(84.2%) ended in draws.
15

unless one participant forfeited. If a participant forfeited at any time, all their tickets were given
to their competitor.
Participants then again completed measures asking about their strength relative to their
partner and predicted fight outcomes. If participants consented, pictures of each were taken in
order to facilitate rated judgments of strength and attractiveness. Finally, participants were fully
debriefed and dismissed.

16

RESULTS
Planned Comparisons
Signaled strength.
To examine the effect competitors have on signaled strength, I first created composite
variables of strength based on the average of the upper body, left hand, and right hand strength
measures at each time point separately (αs = .924 and .925, respectively). The composite variable
was strongly correlated between time points (r = .972, p < .001), as was the upper body strength
measure (r = .920, p < .001), the left hand measure (r = .954, p < .001) and the right hand
measure (r = .954, p < .001). This substantial collinearity presents problems with data analysis
and will be addressed further in the discussion.
As expected and shown in Table 1, men were substantially stronger than women on all
strength measures. In particular, the mean score for men on the standardized composite variable
(0.70) was almost one standard deviation higher than the mean score for women (-0.76). When
examining all the strength variables, in over 99% of cases, the mean score for men exceeded the
maximum score for women. Variability in strength also differed by gender, with men having
almost twice as much variability compared to women on all strength measures.
These composite strength variables were entered into a multilevel model using restricted
maximum likelihood in order to estimate an indistinguishable actor-partner interdependence
model (APIM; Kenny, Kashy, & Cook, 2006) assessing the effects of baseline strength on
strength when a competitor was present (hereafter referred to as signaled strength). If participants
adaptively signal strength depending on competitor strength, being paired with a stronger
individual should result in less signaled strength (a negative partner effect). However, this effect
should be moderated by the individual’s own level of strength, such that decreases in strength

17

signaling should be greater when participants are weak (a negative actor-partner interaction).
Finally, baseline strength should strongly predict signaled strength (a positive actor effect), such
that strength should largely be stable across time points.
The APIM conducted on the composite strength measure revealed no evidence that
signaled strength was influenced by a competitor, β = .021, t(260) = 1.30, p = .196. However, the
actor-partner strength interaction β = -.032, t(185) = -2.11, p = .036 was significant and negative,
indicating that weaker participants paired with stronger partners tended to signal less strength.
Thus, as predicted, participants signaled strength changed based on who they were competing
against. Finally, a large and positive actor effect was found β = .945, t(259) = 59.25, p < .001,
indicating high stability between strength measurements. Participant sex did not influence
signaled strength, nor did it interact with the strength variables, (ts < 2, ps > .200).
The above analysis was also conducted separately on each of the three strength measures:
upper body strength, left arm strength, and right arm strength. While in all cases there was
stability between baseline strength measurements (all ts > 30, ps < .001), no actor-partner
strength interactions reached significance (all ts < 2, ps > .080). Additionally, in the analysis of
the right arm strength measure the presence of the confederate actually led to greater strength (b
= .038, t(281) = 2.11, p = .036, opposite of what was predicted.
Examination of the intraclass correlations for each of the strength measures revealed that
the data were nonindependent, intraclass r = .55 - .65, ps < .001. That is, scores within dyads
were more similar than scores between dyads. Although this ostensibly suggests that
participants’ strength was influenced by whom they competed against, the null results obtained
for the partner effects are inconsistent with this interpretation. Furthermore, rerunning the
analyses with only baseline strength as a predictor revealed that it was sufficient to explain the

18

nonindependence in the data; all intraclass correlations were reduced to nonsignficance. Thus, it
is likely that the similarity in the data is artifactual. Recall that participants were paired with
same-sex partners. Because men’s scores were more similar to other men rather than to women,
and women’s scores were more similar to other women than to men (see Table 1), the
nonindependence might simply be due to experimental design. Indeed, as Table 2 shows, when
examining the similarity between competitors separately by sex, the nonindependence was
reduced to nonsignificance in all but one case. Thus, the intraclass correlation is inflated by
experimental design, giving little evidence to suggest that participants’ strength was influenced
by their opponent.
Diagnostic analyses.
One reason why the current study may have found no evidence for changes in strength
signaling is due to the fact that most dyads were evenly matched in terms of strength, compared
to the pilot study where strength differences were manipulated via confederates. When
examining the upper body strength measure standardized across sex, 152 dyads (81.2%) had
individuals within one standard deviation of each other. Even when standardized within sex, 117
dyads (62.5%) still had individuals within one standard deviation of each other. For comparison,
in the pilot study of 121 men, paired with five confederates, only 45 dyads (37.2%) had
individuals within one standard deviation of each other. Recall that in the pilot study, post-hoc
tests revealed that decreases in upper body strength were most pronounced for participants paired
with the two strongest confederates. Given the smaller discrepancies in strength in the current
study, the finding that competitors do not influence personal strength is ambiguous.
To reiterate, I hypothesized that individuals are sensitive to discrepancies in RHP, which
lead to different signals (submission or willingness to escalate) indexed by strength changes. The

19

larger the discrepancy, the more pronounced the signal. This issue is complicated by the fact that
assessments are not perfect; while they are strongly correlated with measurements of physical
strength (r = .66 for men and r = .51 for women; Sell et al. 2009), there is still a substantial
degree of noise. Because judgments are not perfect, when RHP between competitors is roughly
equal, they may both assess themselves as stronger (Parker, 1974). In this case, both individuals
should signal willingness to fight. Given that a large percentage of participants were of roughly
equal RHP, this may have obscured the relationship between differences in formidability and
greater signaling of submission.
One way to test this hypothesis directly is to split the dyads into two groups, one where
there the discrepancy in strength between competitors is large, and another where differences
between competitors is small. This was accomplished by using the composite measure of
strength (standardized within sex)3 to split the 199 dyads into two groups, one group where the
difference in strength between competitors was less than one standard deviation (the similar
strength subset, n = 112) and the other where the difference was more than one standard
deviation (the different strength subset, n = 77). If similarity in strength was obscuring the
relationship between formidability and (submissive) signaling, examining the different strength
group should reveal the expected pattern. The similar strength group should signal increased
strength as the difference between competitor formidability decreases, above and beyond any
effects due to personal strength and competitor strength.
However, running the APIM with the different strength dyads did not change the pattern
of results. Upper body strength as measured with a competitor present was still strongly

3

As there was more variability in strength in the male dyads, measurements were standardized
within sex to prevent imbalance in the number of male and female dyads selected into the similar
and dissimilar groups. Standardizing across sex did not change the pattern of results.
20

influenced by baseline strength (β = .952, t(127) = 54.58, p < .001), but the effect of competitor
strength did not reach significance (β = .014, t(127) = -0.76, p = .428), nor did their interaction (β
= -.023, t(74) = -0.95, p = .348). Similarly, running the APIM with the same strength dyads also
did not change the pattern of results. Upper body strength as measured with a competitor present
was still strongly influenced by baseline strength (β = .893, t(120) = 20.78, p < .001), but the
effect of competitor strength did not reach significance (β = .066, t(120) = 1.53, p = .128), nor
did the absolute magnitude of their difference (β = -.022, t(74) = -0.95, p = .348). The coefficient
represents the degree to which greater (or lesser) discrepancy in baseline strength is related to
greater strength in the presence of the competitor.
In sum, we did not find evidence for strength signaling, even when splitting the dyads
into two groups, one where the discrepancy in strength was large, and the other where the
discrepancy was low. Implications of this are further addressed in the discussion.
Competition outcomes.
To examine how baseline and signaled strength influence competitive outcomes, arm
wrestling outcomes were regressed on strength measures at the dyad level. If signaled strength
influences competition outcomes above and beyond baseline differences in strength, its inclusion
should improve the predictive validity of the model when controlling for any differences due to
baseline strength. That is, does strength signaling influence the likelihood that an individual will
win a physical contest such as arm wrestling?
Because competition outcomes between competitors are completely dependent (i.e.,
winning a match necessarily means that the other competitor loses), analyses were conducted at
the dyad level. As assignment to Person 1 and Person 2 is arbitrary, in all analyses intercepts
were suppressed. Predictors were determined by taking the difference of the two competitors

21

scores (e.g., X1 – X2). The difference in outcomes (range: -3 to 3) was then regressed on the
difference between predictors. Thus, positive coefficients represent that greater asymmetry in the
predictor (e.g., upper body strength) predicts competitive success.
To control for individual differences not (directly) related to upper body strength, hand
dominance and arm length were first entered into the model. The overall model was significant,
R2 = .059, F(2, 140) = 4.36, p = .015. Handedness did not predict success in the arm wrestling
competition, b = .090, t(140) = .540, p = .590, and was removed from the model. Greater arm
length significantly predicted success, b = .081, t(140) = 2.933, p = .004, and was retained. In the
next step, baseline right arm strength (all matches were right handed) and upper body strength
were added as predictors. As can be seen in Table 3, the overall model was again significant, ΔR2
= .203, F(2, 139) = 19.05, p < .001. Greater upper body strength (b = .045, t(139) = 3.00, p =
.003) and right arm strength (b = .060, t(139) = 3.17, p = .002) both significantly predicted
success. In the last step of the model, signaled upper body and right arm strength were added as
predictors. However, adding in these variables did not significantly increase the predictive power
of the model ΔR2 = .008, F(2, 137) = 0.78, p = .460. Thus, Hypothesis 3 was not supported, in
that signaled strength did not influence the likelihood of winning a physical contest.
Exploratory Analyses
Updating strength assessments.
To further examine whether participants were accurately assessing formidability of
conspecifics (Hypothesis 1), I investigated participants’ relative strength assessments. Recall that
participants were asked to report their strength relative to their partner on a 7-point scale (1 = I
am much weaker than the other participant, 7 = I am much stronger than the other participant) at
two time points, first just after meeting the other participant and second after the competition. I

22

predicted participants would use their own strength as well as their competitor’s strength when
judging who was stronger. However, participants would gain additional information about
strength from the competition that would influence their relative strength assessments such that
participants who won would reassess themselves as stronger, while participants who lost would
reassess themselves as weaker.
To test this hypothesis, I examined whether or not participants updated their relative
strength assessments based on information gained from the arm wrestling competition. To
control for nonindependence in the data, I ran a multilevel regression model where the second
relative strength assessment was regressed onto the initial relative strength assessment, as well as
the outcome of the competition, participant sex, and the sex by competitive outcome product
term. A positive coefficient for the competition outcome reflects that controlling for the initial
relative strength assessment, participants who won (lost) the competition were more likely to rate
themselves as stronger (weaker) than their competitor. Table 4 lists the model coefficients. As
predicted, the coefficient for the competitive outcome was significant and positive, b = .547,
t(166) = 19.38, p < .001, indicating that controlling for their initial strength assessment,
participants adjusted their strength ratings based on the competitive task. This effect was
moderated by a significant sex by competitive outcome interaction, b = -.079, t(140) = -3.01, p <
.001, indicating that men updated their strength ratings less than women. The overall model
accounted for the vast majority the variance in relative strength assessments, pseudo-R2 = .734.
Accuracy of updated assessments.
Although updating strength assessments after a physical competition is consistent with an
adaptive reappraisal of strength differences, it is unclear whether changes in assessment better
reflect reality. Consider the case where two individuals meet and initially consider themselves of

23

equal strength. With some difficulty, one participant edges out the other in the competition and
updates his or her relative strength assessment to suggest that he or she is stronger. The issue is
whether this updated assessment more accurately reflects the difference in strength between the
two individuals. It is possible that participants might overcorrect, resulting in assessments that
less accurately track strength differences than those made before the competition. However, if
participants are accurately using the information from the competition to inform their strength
assessments, these posterior assessments should more accurately reflect strength differences.
A preliminary test revealed that, as expected, initial assessments of strength were
nonindependent, intraclass r = -.268, p < .001, indicating that the more (less) individuals rated
themselves as stronger relative to their competitor, the less (more) their partner rated themselves
as stronger relative to their competitor. More relevant, strength assessments were more correlated
after the arm wrestling competition, intraclass r = -.783, Z = 5.66, p < .001. This increase in
nonindependence is expected because as both participants’ accuracy increases, so should the
correlation between their assessments.
Because relative assessments of strength are nonindependent, they were entered into a
multilevel model using restricted maximum likelihood in order to estimate an indistinguishable
over-time APIM assessing the effects of baseline strength (composite measure) on relative
strength assessments. I predicted participants’ relative strength assessments would be strongly
influenced by their own strength (a positive actor effect) as well as the strength of their partner (a
negative partner effect) at both time points, but particularly after the arm wrestling competition.
That is, not only would participants reassess strength judgments after the physical competition,
but these judgments would be more accurate (i.e., better predicted by baseline strength).

24

As predicted and displayed in Table 5, the over-time APIM revealed that relative strength
assessments were positively related to greater personal baseline strength, b = .971, t(211) = 8.33,
p < .001, and negatively related to the competitor’s baseline strength, b = -.661, t(208) = -5.71, p
< .001. Thus, participants’ strength assessment was influenced by both their own strength and
their competitor’s strength. More importantly, the predicted actor strength by time (b = .226,
t(169) = 3.56, p < .001) and partner strength by time (b = -.190, t(169) = -3.00, p = .003)
interactions were significant, indicating participant’s assessments after the competition better
tracked baseline strength differences. That is, participants’ relative strength assessments became
more accurate after the competition. There was also a main effect of participant sex, such that
men were more likely to rate their competitor as less strong than women, b = -.254, t(144) = 2.87, p = .005. However, sex did not interact with any other variable (ts < 2, ps > .10). In sum,
32.3% of the variance in participants’ strength assessments were explained by differences in
physical strength.
A similar analysis was conducted on judgments of predicted fight outcomes. As with
relative strength assessments, participants were asked to report on a 7-point scale how likely they
would be to win a physical fight with their partner as well as what the outcome of the fight
would be, where higher numbers represent better predicted success. These judgments were also
reported before and after the arm wrestling competition (αs = .930 and .959, respectively).
Paralleling assessments of physical strength, the over-time APIM revealed that predicted fight
outcomes were positively related to greater personal baseline strength, b = .797, t(237) = 6.14, p
< .001, and negatively related to the competitor’s baseline strength, b = -.465, t(234) = -3.61, p <
.001, see Table 6. Thus, participants’ predicted fight outcomes were influenced by both their own
strength and their competitor’s strength. More importantly, the predicted actor strength by time

25

(b = .125, t(184) = 3.22, p = .002) and partner strength by time (b = -.100, t(184) = -2.57, p =
.011) interactions were significant, indicating participant’s judgments after the competition better
tracked baseline strength differences. That is, participants’ predicted fight outcomes more
accurately tracked differences in physical strength after the competition. Unlike the relative
strength measure, there was no main effect of participant sex, b = -.187, t(145) = -1.62, p = .102,
nor did it interact with any other variable (ts < 2, ps > .05). In sum, 15.3% of the variance in
participants’ predicted fight outcomes were explained by differences in physical strength.

26

DISCUSSION
This research emphasizes that humans can accurately assess differences in physical
formidability and update these assessments based on information gained during an interaction.
Individuals took into account both their own strength and the strength of a competitor when
making initial assessments about who was stronger as well as who would win in a physical fight.
In addition, participants updated their assessments after competing against the competitor in a
physical task. Like non-human animals, humans appear to have psychological mechanisms that
accurately track cues to physical formidability, and they dynamically update these cues based on
relevant information.
Assessment Accuracy
Individuals’ relative assessment of strength accurately tracked their own physical
formidability as well as that of the other competitor. Although their initial estimate tracked
strength differences, it was improved by competing against the other individual in a contest
determined by physical strength. Improvement was obtained regardless of whether dyads were
male or female, indicating that both men and women gained information from the competition.
Improvement was not due to mere exposure to the other participant. Recall that a portion
of the dyads (38 out of 199) engaged in a different competitive task (i.e., a mercy task) before
making their second assessment of strength. This task was not very sensitive to differences in
physical formidability; 84.2% of all matches ended in draws compared to only 32.9% in the arm
wrestling contest, χ2(1) = 32.18, p < .001. In those dyads, accuracy in strength assessments
actually became worse after the competition (time by actor/partner effects were in the opposite
direction). Similarly, participants predicted fight outcomes did not more accurately track
differences in physical formidability after the competition (time by actor/partner effects were

27

nonsignificant). Thus, completing a task that is a poor indicator of physical formidability, or
unrelated to physical formidability at all, is unlikely to improve assessment accuracy.
Defensive Signaling
Based on the results of a pilot study, I predicted that individuals would modify their
strength based on the strength of their competitor. Specifically, when an individual was weaker
than their competitor they would signal submission, resulting in decreased strength relative to
baseline. In contrast, when an individual was stronger than their competitor they would signal
willingness to escalate conflict, resulting in increased strength relative to baseline. While the data
did not support this hypothesis, this null result may be due to experimental design limitations.
The issue is that the current study did not manipulate strength via pairing participants
with confederates chosen based on physical formidability. Instead, all individuals were naïve
participants whose strength varied based on individual differences. As discussed in the results
section, this led to considerably less dyads where there was a large discrepancy in strength. In
line with prior theorizing (e.g., Parker, 1974), because strength assessment is not perfect, when
individuals are similar in strength, they may both signal willingness to escalate conflict. This
could have created noise in the data that obscured the relevant differences in signaling.
However, dividing dyads into two groups where the difference in strength was either
similar or dissimilar (i.e., less or greater than one standard deviation, respectively) did not
change the pattern of pattern of results. Personal strength was not influenced by the presence of a
competitor. If the prediction is correct, this null funding may simply be due to low power.
However, if signaling changes do occur when individuals interact with competitors, the data
suggest the size of the effect is small. To ascertain if the effect is reliable as well as to narrow
down the confidence interval of the effect, it will be necessary to conduct additional research that

28

manipulates the discrepancy in strength between individuals. A highly powered test of this
hypothesis is necessary to replicate the original effect and help rule out the possibility of a false
positive. It may also be the case that the effect driving strength changes in the pilot study was
confounded with strength amongst the confederates selected. To rule out these possibilities, more
research with additional confederates is needed.
Because there were no differences in signaled strength, it is therefore unsurprising that I
did not find support for the hypothesis that defensive signaling influenced competitive outcomes.
Given well-known problems with including highly correlated variables as predictors in a
regression model (Cronbach, 1987), it not unusual that including the second upper body strength
(r = .920) and right arm strength (r = .954) measures would not significantly improve model fit.
Because this hypothesis is contingent upon finding evidence for strength differences based on the
presence of a competitor, this hypothesis cannot directly be evaluated with the current data.
Implications For Game Theory
The finding that humans can accurately assess the formidability of conspecifics and
update those assessments based on information gained during contests supports predictions
derived from game theoretical principles. As Parker and colleagues (Maynard Smith & Parker,
1976; Parker & Rubenstein, 1981) predicted, “it is likely that good information, particularly
about RHP, can be acquired only during a contest itself” (Parker & Rubenstein, 1981, p. 288).
That is, individuals initially might not know (perfectly) the asymmetry in RHP between them
and their competitor, but gain such information through physical competition. Insofar as this
process is available to conscious introspection, it should be reflected through changes in
assessments of relative strength and fighting ability, as it was in the current experiment.

29

Typically, reassessment of relative strength is determined not by conscious changes in
relative strength, but is inferred from the decision to withdraw from a contest. However, low
rates of forfeiting in the current study (4.0%) prevented conclusive analyses of this. This is likely
due to the structure of the competitive tasks used: participants were told that forfeiting at any
point would result in a loss of all their raffle entries. This design encouraged participants to keep
competing, even if they were likely to lose, because the opportunity cost of not competing was
always higher than the cost of competing. When combined with minimal risk of physical injury
(i.e. low damage costs), low rates of forfeiture are consistent with game theoretical predictions.
This interpretation is bolstered by evidence that participants did update their relative strength
assessments based on the match, even if they did not often forfeit.
This logic suggests that changing the nature of the contest would change the duration of
time before the weaker individual withdraws. Instead of a winner-take-all style competition, if
participants are given the option to save their resources in addition to risking them for potential
reward, the decision to withdraw should be contingent upon how great the potential loses are for
a given round. All else equal, participants for whom potential loses are greater should be quicker
to withdraw after discovering that their competitor has higher RHP. This non-fixed framework
would also allow inferences as to how the duration of a contest influences assessment accuracy.
The current design does not allow for a test of whether contest duration influences accuracy
because the vast majority of individuals engaged in three rounds of the physical competition.
However, it seems likely that additional bouts would have an asymptotic relationship with RHP
assessment; given little information about a competitor (i.e., few rounds of competition) an
additional round should help accuracy, however, after many rounds, an additional round is

30

unlikely to improve accuracy. Future research should test this prediction in order to provide
additional support for the predictive validity of game theoretical predictions to human conflict.
Similarities to Animal Assessment
These results share several similarities with studies on competition between human as
well as non-human animals. First, these results add to the growing literature demonstrating that
humans have psychological mechanisms that accurately determine the strength of conspecifics
(e.g., Archer & Thanzami, 2009; Sell et al., 2009). Additionally, they extend this literature by
demonstrating that assessments of predicted fight outcomes take into account both personal
strength as well as competitor strength. While there is considerable debate about how to best
model RHP assessment in non-human animals (for a review, see Taylor & Elwood, 2003), our
results provide strong evidence that humans engage in mutual assessment. Though predicted by
Arnott and Elwood (2010), no other study has directly tested how strength differences between
competitors predict expected conflict outcomes.
More generally, these results parallel findings in the non-human animal literature
demonstrating that assessment improves accuracy of RHP. For example, during the mating
season, red deer stags attempt to guard fertile hinds from other stags (Clutton-Brock & Albon,
1979). A small proportion of approaches between stags actually result in fights; typically when
discrepancy in fighting ability is visible, the stag with lower RHP will withdraw. Only when
RHP is roughly equal do stags engage in roaring contests, where vocalizations accurately reflect
fighting ability. The majority of fights are also preceded by a parallel walks, where stags display
their full body size by walking alongside each other. In both cases displays provide additional
information that influences the decision to engage or withdraw. Similarly, in the current study,
individuals have initial impressions of the strength of their competitor. The accuracy of these

31

impressions is improved based on several rounds of a competition that reflects fighting ability. In
many cases where discrepancies in RHP are not clear, competition provides additional evidence
as to the magnitude of those discrepancies.
Conclusion
I found evidence that humans can accurately assess differences in physical formidability
and that these assessments are dynamically updated based on information gained from
competitive bouts. These processes show remarkable similarities to assessment in non-human
animals and suggest that animal models are useful for predicting human behaviors in competitive
situations. Similarly, these results are largely consistent with predictions derived from a game
theoretical framework.
While the prediction that individuals would broadcast different signals depending on the
relative difference in strength between them and their competitor was not confirmed, this null
effect may be due to the small degree of differences between individuals in dyads. Similarly, the
validity of the hypothesis that conflict outcomes would be influenced based on signaling
submission or willingness to escalate conflict cannot be ascertained without successfully
manipulating differential signaling. More work will need to be done to replicate the original
effect and help rule out the possibility of a false positive.

32

APPENDIX

33

Table 1
Descriptive Statistics for Baseline Strength Measures
Women

Men

Strength Measure
Upper Body
Right Arm
Left Arm
Composite
Upper Body
Right Arm
Left Arm
Composite

Minimum
1.67
14.33
12.33
-1.67
4.67
20.67
14.33
-1.03

Maximum
36.00
44.33
47.33
0.39
69.33
71.33
64.67
2.60

M
19.20
27.11
25.68
-0.76
39.84
42.76
42.76
0.70

SD
6.10
5.15
5.14
0.39
12.33
9.13
8.98
0.73

Note. Women N = 188, Men N = 197. Upper body, right arm, and left arm strength are all
measured in kilograms. Composite is the average of the three standardized strength measures.

34

Table 2
Intraclass Correlations Between Strength Measures
Strength Measure
Upper Body
Left Arm
Right Arm
Composite

All Ps (N = 374)
.592***
.597***
.546***
.647***

Women (N = 184)
.088
.137
.081
.097

Men (N = 190)
.204*
.069
.006
.098

Note. P = participant. Higher correlations indicate greater nonindependence, i.e., that strength
measurements within dyads are more similar than strength measurements between dyads.
*p < .05, ***p < .001

35

Table 3
Multiple Regression Results Predicting Success in Arm Wrestling Competition From Strength
Measures
Strength Measure
Arm length
Upper Body Strength (T1)
Right Arm Strength (T1)
Upper Body Strength (T2)
Right Arm Strength (T2)
R2
F
ΔR2
ΔF

Model 1

Model 2

b
β
.080**
.238
--------.057
8.47**
---

b
β
.043†
.128
.045**
.254
.060**
.280
----.260
16.25***
.203
19.05

Model 3
b
.047†
.032*
.088
.017
-.036

β
.139
.182
.408
.109
-.166

.268
10.32
.008
0.78

Note. T1 = baseline strength measurement. T2 = strength measurement with competitor present.
Model 1 df = 141, Model 2 df = 139, Model 3 df = 137. All variables were grand-mean centered
prior to analysis. Regression was conducted at the dyad level, so the intercept was suppressed.
†

p < .10 *p < .05, **p < .01, ***p < .001.

36

Table 4
Multilevel Multiple Regression Results Predicting Updated Strength Assessments From Past
Strength Assessments and Competition Outcomes
Intercept
Strength Judgment (T1)
Competition Outcome
Sex
Sex*Competition Outcome

b
3.085
.251
.547
.048
-.079

β
-.215
.748
.032
-.107

t
-6.42***
19.38***
1.24
-3.01**

df
-263
166
138
140

Note. T1 = relative strength judgment prior to competition. All continuous variables were grandmean centered prior to analysis; sex was effects coded (-1 = women, 1 = men).
**p < .01, ***p < .001.

37

Table 5
Over-time APIM Model Predicting Relative Strength Assessment From Strength Measurements
Before and After Competition
Intercept
Actor Strength
Partner Strength
Sex
Time
Actor Strength*Time
Partner Strength*Time

b
3.877
.971
-.661
-.254
.131
.226
-.190

β
-.660
-.449
-.181
.093
.153
-.129

t
-8.33***
-5.71***
-2.87**
4.28***
3.56***
-3.00**

df
-211
208
144
143
169
169

Note. Strength refers to the composite strength measure taken at baseline. All continuous
variables were grand-mean centered prior to analysis; sex was effects coded (-1 = women, 1 =
men) and time was effects coded (-1 = initial assessment, 1 = reassessment after competition).
**p < .01, ***p < .001.

38

Table 6
Over-time APIM Model Predicting Fight Outcomes From Strength Measurements Before and
After Competition
Intercept
Actor Strength
Partner Strength
Sex
Time
Actor Strength*Time
Partner Strength*Time

b
4.079
.797
-.465
-.187
.082
.125
-.100

β
-.557
-.325
-.137
.059
.088
-.070

t
-6.14***
-3.61***
-1.62
3.47**
3.21**
-2.57**

df
-237
234
145
144
184
184

Note. Strength refers to the composite strength measure taken at baseline. All continuous
variables were grand-mean centered prior to analysis; sex was effects coded (-1 = women, 1 =
men) and time was effects coded (-1 = initial judgment, 1 = judgment after competition).
**p < .01, ***p < .001.

39

REFERENCES

40

REFERENCES
Archer, J., & Benson, D. (2008). Physical aggression as a function of perceived fighting ability
and provocation: An experimental investigation. Aggressive Behavior, 34(1), 9-24. doi:
10.1002/ab.20179
Archer, J., & Thanzami, V. (2009). The relation between mate value, entitlement, physical
aggression, size and strength among a sample of young Indian men. Evolution and
Human Behavior, 30(5), 315-321. doi: 10.1016/j.evolhumbehav.2009.03.003
Arnott, G., & Elwood, R. W. (2010). Signal residuals and hermit crab displays: flaunt it if you
have it! Animal Behaviour, 79(1), 137-143. doi: 10.1016/j.anbehav.2009.10.011
Benson-Amram, S., Heinen, V. K., Dryer, S. L., & Holekamp, K. E. (2011). Numerical
assessment and individual call discrimination by wild spotted hyaenas, Crocuta crocuta.
Animal Behaviour, 82(4), 743-752. doi: 10.1016/j.anbehav.2011.07.004
Bernstein, I. S., & Gordon, T. P. (1980). The social component of dominance relationships in
rhesus monkeys (Macaca mulatta). Animal Behaviour, 28(4), 1033-1039. doi:
10.1016/S0003-3472(80)80092-3
Bernstein, I. S., Gordon, T. P., & Rose, R. M. (1974). Factors influencing the expression of
aggression during introductions to rhesus monkey groups. Primate Aggression,
Territoriality and Xenophobia, 211-240.
Blanchard, D. C. (1997). Stimulus, environmental, and pharmacological control of defensive
behaviors. In M. E. Bouton & M. S. Fanselow (Eds.), Learning, motivation, and
cognition: The functional behaviorism of Robert C. Bolles (pp. 283-303). Washington,
DC, US: American Psychological Association.
Blanchard, D. C., & Blanchard, R. J. (2003). What can animal aggression research tell us about
human aggression? Hormones and Behavior, 44(3), 171-177. doi: 10.1016/S0018506X(03)00133-8
Blanchard, D. C., Griebel, G., Pobbe, R., & Blanchard, R. J. (2011). Risk assessment as an
evolved threat detection and analysis process. Neuroscience & Biobehavioral Reviews,
35(4), 991-998. doi: 10.1016/j.neubiorev.2010.10.016
Blanchard, D. C., Hynd, A. L., Minke, K. A., Minemoto, T., & Blanchard, R. J. (2001). Human
defensive behaviors to threat scenarios show parallels to fear- and anxiety-related defense
patterns of non-human mammals. Neuroscience & Biobehavioral Reviews, 25(7–8), 761770. doi: 10.1016/S0149-7634(01)00056-2
Cesario, J., & Johnson, D. J. (2013). Partner formidability influences personal strength.
Manuscript in preparation.

41

Chase, I. D., & Seitz, K. (2011). Self-Structuring Properties of Dominance Hierarchies: A New
Perspective. In D. L. B. Robert Huber & B. Patricia (Eds.), Advances in Genetics (Vol.
Volume 75, pp. 51-81): Academic.
Clutton-Brock, T. H., & Albon, S. D. (1979). The roaring of red deer and the evolution of honest
advertisement. Behaviour, 145-170.
Cronbach, L. J. (1987). Statistical tests for moderator variables: Flaws in analyses recently
proposed. Psychological Bulletin, 102(3), 414-417.
Dawkins, M. S., & Guilford, T. (1991). The corruption of honest signalling. Animal Behaviour,
41(5), 865-873. doi: 10.1016/S0003-3472(05)80353-7
de Waal, F. B. M., & Hoekstra, J. A. (1980). Contexts and predictability of aggression in
chimpanzees. Animal Behaviour, 28(3), 929-937. doi: 10.1016/S0003-3472(80)80155-2
Gawronski, B., & Cesario, J. (2013). Of mice and men: What animal research can tell us about
context effects on automatic responses in humans. Personality and Social Psychology
Review, 17(2), 187-215. doi: 10.1177/1088868313480096
Griskevicius, V., Tybur, J. M., Gangestad, S. W., Perea, E. F., Shapiro, J. R., & Kenrick, D. T.
(2009). Aggress to impress: Hostility as an evolved context-dependent strategy. Journal
of Personality and Social Psychology, 96(5), 980-994. doi: 10.1037/a0013907
Ikai, M., & Steinhaus, A. H. (1961). Some factors modifying the expression of human strength.
Journal of Applied Physiology, 16(1), 157-163.
Kenny, D. A., Kashy, D., & Cook, W. L. (2006). Dyadic data analysis. New York, NY:
Guilford.
Marcus, D. K., Kashy, D. A., & Baldwin, S. A. (2009). Studying psychotherapy using the onewith-many design: The therapeutic alliance as an exemplar. Journal of Counseling
Psychology, 56(4), 537.
Matsumura, S., & Hayden, T. J. (2006). When should signals of submission be given?–A game
theory model. Journal of Theoretical Biology, 240(3), 425-433. doi:
10.1016/j.jtbi.2005.10.002
Maynard Smith, J. (1974). The theory of games and the evolution of animal conflicts. Journal of
Theoretical Biology, 47(1), 209-221.
Maynard Smith, J. (1982). Do animals convey information about their intentions? Journal of
Theoretical Biology, 97(1), 1-5. doi: 10.1016/0022-5193(82)90271-5
Maynard Smith, J., & Parker, G. A. (1976). The logic of asymmetric contests. Animal Behaviour,
24(1), 159-175. doi: 10.1016/S0003-3472(76)80110-8

42

Maynard Smith, J., & Price, G. R. (1973). The Logic of Animal Conflict. Nature, 246(5427), 1518. doi: 10.1038/246015a0
Parker, G. A. (1974). Assessment strategy and the evolution of fighting behaviour. Journal of
Theoretical Biology, 47(1), 223-243. doi: 10.1016/0022-5193(74)90111-8
Parker, G. A., & Rubenstein, D. I. (1981). Role assessment, reserve strategy, and acquisition of
information in asymmetric animal conflicts. Animal Behaviour, 29(1), 221-240. doi:
10.1016/S0003-3472(81)80170-4
Sell, A., Bryant, G. A., Cosmides, L., Tooby, J., Sznycer, D., von Rueden, C., et al. (2010).
Adaptations in humans for assessing physical strength from the voice. Proceedings of the
Royal Society B: Biological Sciences, 277(1699), 3509-3518. doi:
10.1098/rspb.2010.0769
Sell, A., Cosmides, L., Tooby, J., Sznycer, D., von Rueden, C., & Gurven, M. (2009). Human
adaptations for the visual assessment of strength and fighting ability from the body and
face. Proceedings of the Royal Society B: Biological Sciences, 276(1656), 575-584. doi:
10.1098/rspb.2008.1177
Sell, A., Hone, L. S., & Pound, N. (2012). The importance of physical strength to human males.
Human Nature: An Interdisciplinary Biosocial Perspective, 23(1), 30-44. doi:
10.1007/s12110-012-9131-2
Simpson, M. J. A. (1968). The display of the Siamese fighting fish, Betta splendens. Animal
Behaviour Monographs, 1(1).
Taylor, P. W., & Elwood, R. W. (2003). The mismeasure of animal contests. Animal Behaviour,
65(6), 1195-1202. doi: 10.1006/anbe.2003.2169
van Staaden, M. J., Searcy, W. A., & Hanlon, R. T. (2011). Signaling aggression. In D. L. B.
Robert Huber & B. Patricia (Eds.), Advances in Genetics (Vol. 75, pp. 23-49). New York,
NY: Academic.

43