Enhanced Source Memory for Names of Cheaters

Evolutionary Psychology
www.epjournal.net – 2009. 7(2): 317-330
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Original Article
Enhanced Source Memory for Names of Cheaters
Raoul Bell, Institute of Experimental Psychology, Heinrich-Heine University, D-40225 Düsseldorf, Germany.
Email: raoul.bell@uni-duesseldorf.de (Corresponding author)
Axel Buchner, Institute of Experimental Psychology, Heinrich-Heine University, D-40225 Düsseldorf,
Germany. Email: axel.buchner@uni-dusseldorf.de
Abstract: The present experiment shows that source memory for names associated with a
history of cheating is better than source memory for names associated with irrelevant or
trustworthy behavior, whereas old-new discrimination is not affected by whether a name
was associated with cheating. This data pattern closely replicates findings obtained in
previous experiments using facial stimuli, thus demonstrating that enhanced source
memory for cheaters is not due to a cheater-detection module closely tied to the face
processing system, but is rather due to a more general bias towards remembering the source
of information associated with cheating.
Keywords: Social Contract Theory, cheater detection, context memory, reciprocal
altruism, cooperation
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Introduction
A popular assumption in evolutionary psychology is that the human cognitive
architecture is composed of highly specialized modules that have evolved to solve specific
adaptive problems. A prominent example is the cheater-detection module postulated by
social contract theory (Cosmides, 1989; Cosmides and Tooby, 1992, 2005). According to
this theory, social cooperation is beneficial because individuals can increase their fitness by
cooperating with each other. However, cooperation also carries the risk of exploitation.
Therefore, it is assumed that cooperation cannot evolve or be stably maintained in a group
unless humans have evolved specialized brain mechanisms that help them to avoid
exploitation by cheaters. The cheater-detection module postulated by social contract theory
serves this end by allowing the individual to quickly and easily draw inferences about
whether someone has cheated in prior exchanges or is about to cheat in future interactions.
A strategy to avoid being exploited by cheaters simply consists of refusing to
cooperate with individuals who are known to have cheated previously (reciprocal altruism;

---

Source Memory for Names of Cheaters
Axelrod and Hamilton, 1981; Trivers, 1971). This requires good memory for cheaters in
addition to cheater detection. More specifically, it has been suggested that the ability to
recognize faces of cheaters is an important prerequisite for a reciprocal strategy in social
exchange (Mealey, Daood, and Krage, 1996; Oda, 1997). Seemingly consistent with these
assumptions, Mealey et al. reported that for faces associated with low-status professions,
old-new discrimination was better for faces associated with descriptions of cheating than
for faces associated with trustworthy behavior. However, this finding has turned out to be
difficult to replicate in studies that differed from the original Mealey et al. study in that
they used very carefully controlled materials (Barclay and Lalumière, 2006; Mehl and
Buchner, 2008). Even in the original Mealey et al. study, the pattern observed for the low-
status professions was not replicated when high-status professions were used. On its own,
this is not necessarily inconsistent with an evolutionary point of view, because increased
familiarity for faces of cheaters without context information might even increase the risk of
exploitation because of the preference often exhibited towards familiar stimuli (Bornstein,
1989). In contrast, source memory for faces of cheaters, that is, memory for the cheating
context in which a face was encountered, can help to avoid cheaters and can therefore
provide an evolutionary benefit. Indeed, recent evidence suggests that source memory for
faces of cheaters is enhanced in comparison to source memory for other types of faces
(Buchner, Bell, Mehl, and Musch, 2009; Chiappe et al., 2004).
Based on these results, we now turn to the next research question that suggested
itself given the hypotheses suggested by Mealey et al. (1996). Specifically, it has been
proposed that the memory advantage for faces of cheaters may be due to “specialized
adaptive features […] built into the individual face recognition system” (p. 120). This
suggestion fits with the trend in evolutionary psychology to attribute human behavior to
specialized modules that are distinctive in terms of their processing mechanisms and their
contents. However, this hypothesis should not be accepted unless it is tested empirically.
The present experiment was designed to test the hypothesis that the effect of the
behavioral history on source memory is restricted to facial stimuli. The experiment is an
exact replication of Experiment 2 reported by Buchner et al. (2009), with the only
exception that names instead of faces were used. In an exposition phase, participants saw
names that were presented together with short descriptions of cheating, trustworthy, or
irrelevant behavior. In a test phase, previously seen and new names were judged as old or
new. If a name was judged as old, participants indicated whether they thought that the
name had previously been associated with a history of cheating, with a history of
trustworthiness, or with neither of these. Our basic hypothesis was that if the effect of the
behavioral-history variable on memory was indeed due to a feature of the human face-
processing system, as assumed by Mealey et al. (1996), then the source memory advantage
for cheaters should vanish when names instead of faces are used as stimuli. If the
underlying mechanism were more general and not restricted to the processing of faces, we
would expect a close replication of the results obtained with faces, namely identical old-
new discrimination for names associated with either type of behavior, but better source
memory for names associated with cheating than for other types of names.
Materials and Methods
Participants
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -318-

---

Source Memory for Names of Cheaters
Participants were 111 women and 82 men, most of whom were students at the
Heinrich-Heine-University Düsseldorf. They were paid for participating. Their age ranged
from 18 to 52 years (M = 25, SD = 5.3).

Apparatus and Materials
A total of 72 common male first names were randomly assigned to two sets of 36
names each (henceforth Sets 1 and 2). Brief descriptions typed below the names conveyed
the behavioral history of the stimulus person (cheating, irrelevant to the cheating-
trustworthiness dimension, trustworthy). To guarantee the comparability of the present
results and the results obtained with facial stimuli, the same behavior descriptions were
used as in Buchner et al.’s (2009) Experiments 1-3. The descriptions included information
about the depicted person’s profession. As in Buchner et al.’s Experiments 1-3, only low-
status professions were used. Note that this seems justified given that Buchner et al. found
that status did not modulate the source memory advantage for cheaters. For instance, “He is
a used-car dealer. He regularly sells restored crash cars as supposedly accident-free and
conceals serious defects from the customers.” would convey a history of cheating. “He is a
scaffolder. Presently, he works at a building site in southern Germany where several
tenements and office buildings are to be built.” would convey behavior that is irrelevant to
the cheating-trustworthiness dimension. “He is a cheese monger. He strongly attends to
sorting out old cheese immediately and allows his customers to try all his products.” would
convey trustworthy behavior. In German, all sentences were 21 words long.
Information about the social status of the professions and the valence of the
descriptions was obtained in independent norming studies. In one norming study,
participants (N = 36) rated 200 job titles with respect to their social status using a scale
ranging from 1 (low status) to 5 (high status). A total of 36 job titles with low ratings were
chosen for the experiment (M = 1.88, SD = .33). A different group of participants (N = 22)
rated the valence of 72 behavior descriptions to make sure that instances of cheating,
irrelevant, and trustworthy behavior were perceived as negative, neutral, and positive,
respectively. Valence was assessed on a scale ranging from -3 (“negative”) to +3
(“positive”). Finally, 12 sentences were selected for each of the three types of descriptions
(cheating, irrelevant, trustworthy). Mean valence was -2.50 for the descriptions of cheating
(SD = 0.51), 0.25 for the descriptions of irrelevant behavior (SD = 0.59), and 2.32 for the
descriptions of trustworthiness (SD = 0.74). In terms of absolute valence (i.e., ignoring the
minus sign for the descriptions of cheating), an item-based analysis showed that there was a
large difference between descriptions of cheating and descriptions of irrelevant behavior,
t(22) = 15.25, p < .001, ?2 = .91, and between descriptions of trustworthiness and
descriptions of irrelevant behavior, t(22) = 25.22, p < .001, ?2 = .97, whereas descriptions
of cheating and descriptions of trustworthiness did not differ, t(22) = 1.12, p = .27, ?2 =
.05. Names and descriptions were combined randomly for each participant.

Procedure
Participants were tested individually. They were asked to rate the likability of 36
stimulus persons. Each trial started with a headline (“How likable do you find this
person?”) and a name (Set 1 or 2, counterbalanced across participants). The behavior
description was shown 2 s later, followed 4.5 s later by the likability rating scale (ranging
from 1, “not likable at all”, to 6, “extremely likable”). Participants rated the likability using
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -319-

---

Source Memory for Names of Cheaters
the computer mouse and initiated the next trial. The names were presented in random order.
As in Buchner et al.’s (2009) Experiments 1, 2, and 4, the exposition phase was
immediately followed by a test phase in which participants saw a random sequence of 72
names, half of which had been presented in the first phase (Set 1 or 2) and half were new
(Set 2 or 1). Each trial started with a headline (“How likable do you find this person?”) and
a name. The likability rating scale appeared 1.5 s later. After the rating a new headline
appeared (“Is this name old or new?”), followed by an “old” and a “new” checkbox, one of
which participants selected depending on whether they thought that they had seen the name
during the exposition phase or not. Following an “old” judgment and a click on the
continue button, checkboxes labeled “cheating”, “trustworthy”, and “neither cheating nor
trustworthy” appeared, which participants used to judge the behavior that was used in the
description accompanying that name in the exposition phase. After selecting one of these
checkboxes and then clicking the continue button the next trial was started.

Design
The within-subject independent variable was behavioral history (cheating,
irrelevant, trustworthy). The dependent measures were likability ratings, old-new
discrimination in terms of the sensitivity measure of the two-high threshold model of signal
detection, Pr, and source judgments given an “old” judgment.
Given a sample size of N = 193, ? = .05, and the assumption that the average
population correlation between the levels of the behavioral-history variable for the old-new
discrimination sensitivity measure is ? = .55 (estimated from pilot data), effects of size f =
0.11 (that is, between small [f = 0.10] and medium [f = 0.25] effects as defined by Cohen,
1988) could be detected for this variable with a probability of 1 – ? = .95. The power
calculation was conducted using G•Power (Faul, Erdfelder, Lang, and Buchner, 2007). A
multivariate approach was used for all within-subject comparisons. In the present
application, all multivariate test criteria correspond to the same (exact) F-statistic, which is
reported. Partial ?2 is reported as an effect-size measure. The level of ? was set to .05.
Results
Exposition-phase likability ratings
Exposition-phase likability ratings differed as a function of the behavioral-history
variable, F(2,191) = 1145.70, p < .001, ?2 = .92. Orthogonal contrasts showed that cheaters
were less likable than other persons, F(1,192) = 2082.71, p < .001, ?2 = .92, and that
trustworthy persons were more likable than persons associated with irrelevant behavior,
F(1,192) = 933.11, p < .001, ?2 = .83. Mean exposition-phase likability was 1.77 for
persons associated with cheating (SE = 0.03), 3.79 for persons associated with irrelevant
behavior (SE = 0.04), and 4.93 for persons associated with trustworthy behavior (SE =
0.04). The results show that the behavior descriptions were attended and processed, a
necessary precondition for analyzing subsequent effects of the descriptions.

Old-new discrimination
The upper panel in Figure 1 displays old-new discrimination in terms of Pr (the
sensitivity measure of the two-high threshold model), which is calculated by subtracting the
false alarm rate (FA) from the hit rate (H), Pr = H – FA. We report Pr as a sensitivity
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -320-

---

Source Memory for Names of Cheaters
measure because it was favorably evaluated in validation studies (Snodgrass and Corwin,
1988) and avoids the problem of undefined values that comes with using d’. Old-new
discrimination did not differ as a function of behavioral history, F(2,191) = 0.01, p = .99, ?2
< 01. Thus, replicating results of studies using facial stimuli (Buchner et al., 2009; Mehl
and Buchner, 2008), there was no effect of the behavior descriptions on old-new
discrimination despite the powerful effects of these descriptions on exposition-phase
likability ratings.

Test-phase likability ratings
The main effect of behavioral history on test-phase likability ratings (Figure 1) was
significant F(2,191) = 6.78, p < .001, ?2 = .07. Names associated with cheating were less
likable than other names, F(1,192) = 12.43, p < .001, ?2 = .06, whereas likability did not
differ between names associated with irrelevant behavior and names associated with
trustworthy behavior, F(1,192) = 0.71, p = .40, ?2 < .01. In other words, the effect of the
behavioral-history variable on test-phase likability ratings obtained in experiments using
facial stimuli (Buchner et al., 2009) was replicated. The decreased test-phase likability
ratings of names that were previously associated with cheating may reflect participants’
memory for the behavior associated with the names, which may translate into a negative
reaction toward these names.

Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -321-

---

Source Memory for Names of Cheaters
Figure 1. Memory measures and test-phase likability ratings as a function of the behavior
descriptions.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
0.4
Old-New Discrimination (P )
)
r
P r 0.3
0.2
0.1
Old-New Discrimination (
0.0
Cheating
Irrelevant Information Trustworthiness

3.5
Test-Phase Likability Rating
3.4
3.3
3.2
3.1
Test-Phase Likability Rating 3.0
0.0
Cheating
Irrelevant Information Trustworthiness

0.4
Source Memory Parameter d
0.3
0.2
Parameter Estimate 0.1
0.0
Cheating
Irrelevant Information Trustworthiness


Upper panel: Old-new discrimination in terms of the sensitivity measure of the two-high
threshold model, Pr. Error bars represent the standard errors of the means. Center panel:
Test-phase likability ratings on a scale from 1 (“not likable at all”) to 6 (“extremely
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -322-

---

Source Memory for Names of Cheaters
likable”). Error bars represent the standard errors of the means. Lower panel: Parameter
estimates for the source memory parameters for names associated with a history of cheating
(dCheat), for names associated with irrelevant information (dIrrelevant), and for names
associated with a history of trustworthiness (dTrust). Error bars represent the .95 confidence
intervals.

¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Source memory
The most interesting dependent variable in the present context is participants’
source memory, that is, their memory for the context in which a name was encountered. A
problem when measuring memory for source is which measurement tool to use. Early
approaches relied on ad-hoc measures of source memory, which are problematic because
they confound source memory with old-new discrimination and guessing processes (Bröder
and Meiser, 2007; Murnane and Bayen, 1996). Fortunately, alternative measurement tools
exist in terms of multinomial models of source memory (Batchelder and Riefer, 1990;
Bayen, Murnane, and Erdfelder, 1996), which are to be preferred over more conventional
approaches because they allow for the independent measurement of old-new
discrimination, source memory, and various types of response biases. We therefore
analyzed the source memory data using the multinomial source memory model developed
and validated by Bayen, Murnane, and Erdfelder (1996), which has been used successfully
in a number of experiments (e.g., Bayen, Nakamura, Dupuis, and Yang, 2000; Bell,
Buchner, and Mund, 2008; D'Argembeau and Van der Linden, 2004; Doerksen and
Shimamura, 2001; Spaniol and Bayen, 2002). An adaptation of the model for the present
purposes is presented in Figure 2.
The model displayed in Figure 2 contains twelve free parameters, each of which
represents the probability with which certain cognitive processes occur. To illustrate,
parameter DCheat represents the probability of recognizing a cheater name as old. Parameter
dCheat represents the conditional probability of also remembering correctly that a recognized
name was encountered in the context of a history-of-cheating description. If the source of a
correctly recognized cheater name is not known (with probability 1 – dCheat), it may be
guessed correctly that the name belongs to a cheater with probability aCheatTrust· aCheat.
Alternatively, it may be guessed incorrectly that the name is that of a trustworthy person
with probability aCheatTrust· (1 – aCheat) or that the name is that of a person described as
neither cheating nor trustworthy with probability (1 – aCheatTrust). If a cheater name from the
exposition phase is not correctly recognized as old (with probability 1 – DCheat), it may still
be guessed, with probability b, that the name is old. For these cheater names, the correct
source may be guessed with probability gCheatTrust· gCheat. Alternatively, it may be guessed
incorrectly that the name belongs to a trustworthy person with probability gCheatTrust· (1 –
gCheat) or to a person described as neither cheating nor trustworthy with probability (1 –
gCheatTrust). The final branch in this tree of the model concerns cheater names that are neither
recognized as old (with probability 1 – DCheat) nor guessed to be old (with probability 1 –
b), which are incorrectly judged to be new. Analogous considerations hold for the model
trees for names associated with trustworthy descriptions, names associated with a context
that is irrelevant to the cheating-trustworthiness dimension, and for new names.

Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -323-

---

Source Memory for Names of Cheaters
Figure 2. Bayen et al.’s (1996) source memory model as adapted for the present purposes.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯


Rounded rectangles on the left side represent the types of names presented. Letters along
the links represent the probabilities with which certain cognitive processes occur (D.:
probability of identifying correctly a name as old or new; d.: source memory in the sense of
remembering the context of a name that was detected as old; b: probability of guessing that
a non-recognized name is old; aCheatTrust, aCheat: probability of guessing that a recognized
name for which the source was not remembered was encountered in a certain context;
gCheatTrust, gCheat: probability of guessing that a non-recognized name that was classified as
old on the basis of guessing was encountered in a certain context). Rectangles on the right
side represent the categories of participants’ judgments.

¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

As an illustration of the utility of such a model consider, for instance, the names associated
with cheating receiving “cheater” responses. These responses may be arrived at either by
recognizing the name as old and remembering its source (with probability DCheat · dCheat),
by recognizing the name as old and guessing its source (with probability DCheat· (1 – dCheat)
· aCheatTrust · aCheat), or by guessing that the name is old and guessing its source (with
probability (1 – DCheat) · b · gCheatTrust · gCheat). Thus, the probability of a cheater name
receiving a “cheater” response is given by DCheat · dCheat + DCheat· (1 – dCheat)· aCheatTrust ·
aCheat + (1 – DCheat) · b · gCheatTrust · gCheat. Analogous model equations may be written for all
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -324-

---

Source Memory for Names of Cheaters
other combinations of name types and response categories. Based on these model equations
and the empirically observed sample responses to the different types of names, it is possible
to decompose the set of processes involved in participants’ responses and to estimate the
probabilities associated with the model parameters representing these processes using
standard computer programs (Rothkegel, 1999; Stahl and Klauer, 2007).
Furthermore, statistical tests can be performed directly on the model parameters. In
order to simplify our analysis, we decided to begin with a base model (henceforth Base
Model 1) that builds on the result that old-new discrimination did not differ as a function of
the behavioral-history variable. We thus decided for Base Model 1 to set all parameters to
be equal that represent the probability of recognizing a name from the exposition phase as
old, that is, DCheat = DIrrelevant = DTrust. Based on the well-known mirror effect (Glanzer,
Adams, Iverson, and Kim, 1993), we also set the parameter representing the probability of
detecting new names as new (DNew) to be equal to the recognition parameters so that Base
Model 1 is characterized by the general restriction that DCheat = DIrrelevant = DTrust = DNew.
These restrictions imply the assumption that the recognition of the names was independent
of whether they were presented as names of cheaters, of irrelevant persons, or of
trustworthy persons. This assumption is justified if the model that implements these
restrictions is compatible with the data. Otherwise, that is, in case of a statistically
significant misfit of the restricted model, the assumption has to be rejected.
The goodness of fit of Base Model 1 is determined by comparing the empirically
observed category frequencies with the frequencies that are predicted by the model. The
goodness-of-fit statistic G2 is asymptotically ?2 distributed with three degrees of freedom.
A p value smaller than .05 would indicate that the implemented restrictions are not
compatible with the data, as a result of which the hypotheses implicated by the parameter
restrictions would have to be rejected. However, the model with the restriction that all
name recognition parameters are equal (i.e., Base Model 1) fits the data very well, G2(3) =
2.90, p = .41. This mirrors the old-new discrimination results reported above.
Next, we tested whether source memory differed between names of trustworthy
persons and names of irrelevant persons. In terms of model parameters, the null hypothesis
that there is no such difference can be implemented directly by imposing, on Base Model 1,
the restriction that dTrust = dIrrelevant. This restriction generates one degree of freedom in
addition to the three degrees of freedom of Base Model 1. The corresponding increase in
the model misfit as expressed in the goodness-of-fit statistic, ?G2, is asymptotically ?2
distributed with one degree of freedom. The restriction was compatible with the data,
?G2(1) = 1.03, p = .31, forcing us to conclude that source memory did not differ as a
function of whether the name belonged to a trustworthy or to an irrelevant person. The set
of restrictions applied so far are combined into Base Model 2, which, as expected, also
fitted the data very well, G2(4) = 3.93, p = .42.
Finally, we tested whether source memory for names associated with cheating was
better than source memory for names associated with trustworthy or irrelevant descriptions.
At a descriptive level this appears to be the case, as can be seen in the lower panel of Figure
1. In terms of model parameters, the null hypothesis that there is no such difference can be
implemented directly by imposing, on Base Model 2, the restriction that dCheat = dTrust =
dIrrelevant. The restriction was not compatible with the data, ?G2(1) = 7.49, p < .01, forcing
us to conclude that source memory for names of cheaters was indeed better than source
memory for other types of names.
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -325-

---

Source Memory for Names of Cheaters
Note that the memory parameters observed in the present experiment are somewhat
lower than the memory parameters observed by Buchner et al. (2009). However, the
absolute magnitude of the memory parameters can be influenced by a large number of
experiment-specific variables such as the homogeneity of the test stimuli. It thus does not
have a unique interpretation. Therefore, the most important aspect of the present findings is
that the pattern of the results reported by Buchner et al. (2009) was replicated. Specifically,
the source memory parameters here and in Buchner et al. (2009) show nearly identical
patterns in that source memory is best for cheaters, intermediate for trustworthy characters,
and worst for characters that were described as neither cheating nor trustworthy.
Discussion
The results of the present experiment using names as stimuli very closely replicate
the results obtained in earlier experiments in which facial stimuli were used (Buchner et al.,
2009). Old-new discrimination of names does not differ as a function of whether names
were originally associated with cheating, irrelevant behavior, or trustworthiness. In
contrast, source memory for names associated with a history of cheating was better than for
other types of names. This finding was corroborated by the result that names associated
with cheating were rated less likable than other names. This may reflect participants’
memory for the behavior associated with the name. Such a negative reaction may help
avoiding costly social exchanges with cheaters.
The fact that the same results were observed regardless of whether names or faces
were used as stimulus material strongly argues against the hypothesis of Mealey et al.
(1996) and others (e.g., Oda, 1997) that enhanced memory for cheaters is due to a
specialized cheater-identification module closely tied to the face processing system. Rather,
the results suggests that the enhanced source memory for faces of cheaters observed earlier
(Buchner et al., 2009; Chiappe et al., 2004) is due to more general mechanisms that serve to
increase the probability that the cheating context is remembered regardless of whether it is
paired with a face or a name.
Obviously, it can be adaptive to have good source memory for names of cheaters,
because it could help to avoid cheaters in social encounters. Models of indirect reciprocity
(Nowak and Sigmund, 2005) postulate that social exchanges between third parties are a
valuable source of information about potential exchange partners (see also Dunbar, 2004;
Enquist and Leimar, 1993; Wilson, Wilczynski, Wells, and Weiser, 2000), because a
person’s behavior in previous social interactions may be predictive of the behavior of that
person in future interactions. It might be more helpful to focus on cheating rather than on
cooperation because cooperation is less diagnostic than cheating due to the fact that even
selfish individuals may cooperate if they can benefit from cooperation directly. Therefore,
it may be beneficial for individuals to focus on the names of persons that have cheated,
because this enables them to avoid exploitation by these individuals in future encounters.
Given the relative ease of spreading information about reputations associated with names,
these memory biases may also play an important role in the establishment and maintenance
of social norms in groups and societies. When cooperators consistently refuse to interact
with individuals who are known to have cheated in interactions with third parties, cheating
is punished indirectly. Consistent with this assumption, the rate of cheating decreases when
the individuals are able to share information about who has cheated and who has
Evolutionary Psychology – ISSN 1474-7049 – Volume 7(2). 2009. -326-

---

Document Outline

• ??
  • ??
  • ??
  • ??
  • ??
• ??
  • ??
  • ??
  • ??
  • ??
• ??
• ??

---