ArticlePDF Available

A Meta-Analytic Investigation of the Relation Between Interpersonal Attraction and Enacted Behavior

  • Saint Martin's University

Abstract and Figures

We present a meta-analysis that investigated the relation between self-reported interpersonal attraction and enacted behavior. Our synthesis focused on (a) identifying the behaviors related to attraction; (b) evaluating the efficacy of models of the relation between attraction and behavior; (c) testing the impact of several moderators, including evaluative threat salience, cognitive appraisal salience, and the sex composition of the social interaction; and (d) investigating the degree of agreement between the meta-analytic findings and an ethnographic analysis. Using a multilevel modeling approach, an analysis of 309 effect sizes (N = 5,422) revealed a significant association (z = .20) between self-reported attraction and enacted behavior. Key findings include: (a) that the specific behaviors associated with attraction (e.g., eye contact, smiling, laughter, mimicry) are those behaviors research has linked to the development of trust/rapport; (b) direct behaviors (e.g., physical proximity, talking to), compared with indirect behaviors (e.g., eye contact, smiling, mimicry), were more strongly related to self-reported attraction; and (c) evaluative threat salience (e.g., fear of rejection) reduced the magnitude of the relation between direct behavior and affective attraction. Moreover, an ethnographic analysis revealed consistency between the behaviors identified by the meta-analysis and those behaviors identified by ethnographers as predictive of attraction. We discuss the implications of our findings for models of the relation between attraction and behavior, for the behavioral expressions of emotions, and for how attraction is measured and conceptualized.
Content may be subject to copyright.
A Meta-Analytic Investigation of the Relation Between Interpersonal
Attraction and Enacted Behavior
R. Matthew Montoya and Christine Kershaw
University of Dayton Julie L. Prosser
Colorado State University
We present a meta-analysis that investigated the relation between self-reported interpersonal attraction and
enacted behavior. Our synthesis focused on (a) identifying the behaviors related to attraction; (b) evaluating
the efficacy of models of the relation between attraction and behavior; (c) testing the impact of several
moderators, including evaluative threat salience, cognitive appraisal salience, and the sex composition of the
social interaction; and (d) investigating the degree of agreement between the meta-analytic findings and an
ethnographic analysis. Using a multilevel modeling approach, an analysis of 309 effect sizes (N5,422)
revealed a significant association (z.20) between self-reported attraction and enacted behavior. Key
findings include: (a) that the specific behaviors associated with attraction (e.g., eye contact, smiling, laughter,
mimicry) are those behaviors research has linked to the development of trust/rapport; (b) direct behaviors (e.g.,
physical proximity, talking to), compared with indirect behaviors (e.g., eye contact, smiling, mimicry), were
more strongly related to self-reported attraction; and (c) evaluative threat salience (e.g., fear of rejection)
reduced the magnitude of the relation between direct behavior and affective attraction. Moreover, an
ethnographic analysis revealed consistency between the behaviors identified by the meta-analysis and those
behaviors identified by ethnographers as predictive of attraction. We discuss the implications of our findings
for models of the relation between attraction and behavior, for the behavioral expressions of emotions, and for
how attraction is measured and conceptualized.
Public Significance Statement
An analysis of 309 effect sizes revealed that the experience of attraction covaries with the
expression of a subset of behaviors associated with the development of interpersonal trust (e.g.,
eye contact, smiling, laughter, mimicry). Based on these findings, we present a model of
attraction in which attraction is expressed instrumentally to develop and regulate interpersonal
Keywords: interpersonal attraction, nonverbal behavior, attitude-behavior consistency, flirting, emotion
Supplemental materials:
In popular culture, there is an assumed or expected relation
between the experience and expression of interpersonal attraction.
For instance, a motif of dating advice columns (e.g., Arneson,
2011;Behrendt & Tuccillo, 2004;Williams, 2014) and romantic
comedies (e.g., Hitch,Sex and the City;Lassiter & Tennant, 2005;
Zuritsky, Bushnell, Rottenberg, & Frankel, 2003) is that daters can
determine whether their date is “interested” in them by observing
their behavior toward them. The empirical literature is less clear on
the presence and reason of a connection, with some research
claiming that the affective experience associated with interper-
sonal attraction does not directly motivate the expression of en-
acted behavior (e.g., Baumeister, Vohs, Dewall, & Zhang, 2007),
that such positive experiences are related to behavior, but heavily
contingent on situational and normative forces (Ekman & Friesen,
1969;Gross, 2015), or that such behavior is indeed a consequence
of the subjective experience of attraction (Byrne & Griffitt, 1973;
Huston & Levinger, 1978). The purpose of this research is to
investigate: (a) which specific behaviors (e.g., reducing physical
proximity, smiling at, requesting a date from) are associated with
self-reported interpersonal attraction (e.g., “I like him/her”); (b)
the veracity of the various theoretical explanations for the relation;
(c) the processes that moderate the relation; and (d) the evidence
regarding whether such a relation is evident in different cultures.
This article was published Online First May 7, 2018.
R. Matthew Montoya and Christine Kershaw, Department of Psychol-
ogy, University of Dayton; Julie L. Prosser, Applied Social and Health
Psychology, Colorado State University.
Christine Kershaw is now at the Department of Psychology, University
of Alberta.
We thank the primary authors who provided additional information.
Correspondence concerning this article should be addressed to R. Matthew
Montoya, Department of Psychology, University of Dayton, 300 College Park,
Dayton, Ohio 45469. E-mail:
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Psychological Bulletin
© 2018 American Psychological Association 2018, Vol. 144, No. 7, 673–709
Definition of Attraction and Enacted Behavior
The first step is to define interpersonal attraction and to identify
the set of behaviors potentially associated with it. In the attraction
literature, attraction tends to be defined as an attitude that includes
the person’s affective, cognitive, and behavioral evaluation of the
target person (e.g., Berscheid, 1985;Finkel & Baumeister, 2010;
Graziano & Bruce, 2008;Huston & Levinger, 1978;Orbuch &
Sprecher, 2006). Recent literature, however, has maintained its
focus on the definition as an attitude, but narrowed the definition
to one’s immediate and positive emotional and/or behavioral re-
sponse to a specific person (Montoya & Horton, 2014). From this
updated perspective, the cognitive component is not considered
part of the attraction construct per se, but rather it is a process that
predicts the attraction response (see also Kaplan & Anderson,
1973). This updated version describes attraction as comprised of
two components, an affective component that reflects the quality
of one’s subjective response (affective attraction; often measured
using items such as “I feel favorably toward him/her”) and a
behavioral component that reflects one’s preference for a particu-
lar behavioral response toward the target person (behavioral at-
traction; measured using self-reported items such as “I would like
to meet him/her”).
Previous definitions of behavioral attraction
(e.g., Montoya & Horton, 2014) have defined it as either the
preferred or enacted behavioral response. In this meta-analysis, we
use behavioral attraction to refer exclusively to a self-reported
preference for a particular behavioral response. We use liking and
interpersonal attraction to refer to an undifferentiated positive
evaluation that includes both affective and behavioral attraction.
Behavior Associated With Interpersonal Attraction
The next step is to identify those behaviors associated with
attraction. Table 1 presents seven taxonomies of the behavioral
manifestations of attraction. On the one hand, researchers tend to
agree on a small number of behaviors, including self-touch, body
orientation, eye contact, and smiling as associated with attraction.
On the other hand, the table also indicates little agreement for a
number of behaviors, with 13 of the 21 behaviors mentioned by
fewer than half the researchers.
These taxonomies make the methodological distinction between
direct and indirect behaviors. Direct behaviors have been defined
as acts that function to directly establish/maintain contact or re-
duce the physical space between two people. Direct behaviors
represent an operationalization most consistent with attraction’s
literal definition (i.e., reducing the interpersonal space between
two people). Direct behaviors have been commonly assessed using
seating proximity (Argyle & Dean, 1965;Russo, 1967), decisions
to sit with the target person (Schachter, 1959), talking to the person
(Grammer, Honda, Juette, & Schmitt, 1999), or asking a person on
a date (Deyo & Deyo, 2003). Alternatively, indirect behaviors
(referred to as “solicitation cues” by Moore, 1985; see also Sy-
monds, 1972) have been defined as acts designed to trigger affili-
ative attempts from, and/or attraction in, the target person. In
Western society, researchers have labeled many nonverbal behav-
iors as indirect, including smiling (Palmer & Simmons, 1995), eye
contact (Cappella & Palmer, 1990;Hall, 1963), self-touch (Ren-
ninger, Wade, & Grammer, 2004), head cant (Krumhuber, Man-
stead, & Kappas, 2007), head nodding (Muehlenhard, Koralewski,
Andrews, & Burdick, 1986), forward lean (Burgoon, 1991;Galton,
1884;James, 1932), mimetic behavior (Chartrand & Bargh, 1999),
open posture (McCormick & Jones, 1989;Mehrabian, 1969),
head/body orientation (Hall, 1963;James, 1932;Mehrabian &
Friar, 1969), and laughter (Moore, 1985).
As is discussed in the following section, different theoretical
perspectives produce different expectations not only for why at-
traction is associated with an enacted behavior, but which specific
behaviors are associated with interpersonal attraction.
Models of the Relation Between
Attraction and Behavior
Few theoretical models specifically outline the impact that an
affective/emotional state has on enacted behavior. At one extreme,
some models submit that affective/emotional processes function
primarily to inform the person of his or her relation with the
environment, and that there is no relation between affect/emotion
and enacted behavior. James (1890), for instance, famously stated
that emotions “terminate in the subject’s own body,” a position
consistent with other researchers (Kagan, 1978;McCall & McGee,
This article outlines three approaches that do submit that
there is a relation between the self-reported affective/emotional
experience and a behavioral expression: subjective experience,
generate approach, and develop/restore degree of interdepen-
dence. Although there are other models (e.g., DeWall, Baumeis-
ter, Chester, & Bushman, 2016) and permutations of these
general approaches, we speak specifically to these three models
because they have been articulated sufficiently well to allow
specific predictions to be made. For each approach, we discuss
(a) the logic for a relation between the experience of attraction
and enacted behavior and (b) what form the enacted behavior is
proposed to take.
Subjective Experience
The relation between self-reported attraction and enacted behav-
ior begins with the positive affective state. In general, the success-
ful navigation of one’s environment results in a positive subjective
experience (for discussions, see Buck, 1985;Campos, Campos, &
Barrett, 1989); and more specifically, the degree to which the
person evaluates the environment positively/negatively dictates the
valence of their internal state. This affective experience is largely
responsible for changes in the “downstream” cognitive and behav-
ioral responses (Niedenthal, 1990). Models of dissonance and
Behavioral attraction should not be confused with behavioral inten-
tions, which are “self-instructions to perform particular behaviors or to
obtain certain outcomes” (Triandis, 1980). Indeed, the attitude-behavior
consistency literature indicates that behavioral intentions are the proximate
determinant of behavior (Ajzen, 1991;Triandis, 1980) and is the strongest
predictor of behavior (Abraham, Sheeran, & Johnston, 1998;Eagly &
Chaiken, 1993;Maddux, 1999). However, behavioral intentions are rarely
assessed or investigated in the attraction literature. For example, it is
uncommon for investigations into dating behaviors to include questions
such as “I plan to ask out Person X” or “How likely is it that you will smile
at him/her?” The assessments commonly employed in the attraction liter-
ature (and most theories of interpersonal attraction) investigate partici-
pants’ preference for a particular course of action (i.e., behavioral attrac-
tion; e.g., “I would like to see this person again”) rather than their intent to
act in a particular way (e.g., “I will ask out this person;” Ajzen, 1991).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
cognitive consistency, for example, submit that a diffuse (aversive)
affective state motivates one’s behavior to align with one’s attitude
(Abelson & Rosenberg, 1958;Heider, 1958). Alternatively, mis-
attributional models (e.g., Zillman, 1978) submit that behavior
may result from an affective state when it is (mis)attributed to a
specific process (e.g., interest, happiness/enjoyment).
From this perspective, behavior that covaries with self-reported
attraction results from an emotion that was generated by the
positive interpersonal experience. Three emotions, and their asso-
ciated behavioral responses, are most likely responsible for any
observed link between self-reported attraction and enacted behav-
ior: happiness/enjoyment, which is characterized by smiling and
head tilt, among other behaviors (e.g., Dael, Mortillaro, & Scherer,
2012); sexual desire, which is associated with lip biting, lip lick-
ing, and lip touching (Gonzaga, Turner, Keltner, Campos, &
Altemus, 2006); and interest, which has been associated with a
directed head orientation, wide open eyes, and sustained eye
contact (Reeve, 1993).
In summary, from this approach, enacted behaviors should align
with specific emotions hypothesized to covary with a positive
affective state (e.g., happiness and smiling, interest and eye con-
tact, sexual interest and lip biting) that may result from the positive
interpersonal interaction. We call the set of behaviors predicted by
this approach emotion-specific behavior.
Generate Approach
Behaviorist models of attraction from the 1960’s and 1970’s
proposed that the affective/emotional response (acquired via either
classical or operant conditioning) was responsible for approach-
oriented behavior (e.g., Byrne & Clore, 1967;Lott & Lott, 1969;
Mehrabian, 1970). These models largely agreed that the positive
emotional state mediates approach-oriented behavior (Staats,
1994;Staats & Eifert, 1990; although Skinner, 1979 suggested that
emotions are simply “collateral” processes). From this perspective,
attraction is a diffuse positive affective experience associated with
a specific person, and it is this positive state that facilitates
approach-oriented behavior. More specifically, Byrne and col-
leagues (Byrne & Clore, 1967;Byrne, Nelson, & Reeves, 1966)
proposed that a reinforcing stimulus (e.g., a physically attractive
person or a person who is attitudinally similar) produces an im-
plicit affective response. The implicit affective response then pro-
duces an evaluative response (i.e., affective attraction), and impor-
tantly, a behavioral response (i.e., overt behavior). Lott and
colleagues (Lott, Aponte, Lott, & McGinley, 1969;Lott & Lott,
1969) similarly proposed that “a liked person should evoke an
anticipation of reward and arouse approach tendencies” (Lott et al.,
1969, p. 102). Relatedly, Mehrabian (1970,1971) proposed the
immediacy principle, in which “[p]eople are drawn toward people
Table 1
Enacted Behavior Associated With the Expression of Attraction/Liking
Behavior Mehrabian
(1969) Givens (1978) Patterson
(1982) Moore
(1985) McCormick &
Jones (1989)
and Harrigan (1989) Grammer, Kruck, Juette,
and Fink (2000)
Body lean (forward) X X X
Body orientation X X X X X
Brow raise X X
Eye contact (duration) X X X X
Eye contact (frequency) X X
Flexing/Protruding chest X X X
Gesticulations X X X
Hair flip X X
Head down X
Head nod X X X
Head tilt X
Head toss X X
Laughing X X X
Mimicry X
Open posture X X (legs, arms) X (legs, arms) X (legs)
Neck presentation X
Proximity X X X
Shoulder shrug X
Smile, genuine X X X X X
Smile, other X (ambivalent, smile
with “pouting,
compressed lips,
showing tongue,”
p. 349)
X (coy) X (coy, smile followed
by turning away and
looking down)
Stretching X X
Touch (self) X X X X
Touch (other person) X X X X
Note. An X indicates that the behavior was mentioned by the author(s). No author(s) distinguished between smile duration and frequency. Mehrabian
(1969),Patterson (1982), and Tickle-Degnen et al. (1989) referred to nonverbal behaviors used to communicate liking/involvement. Givens (1978)
described behaviors that were “courting displays.” Moore (1985) included the “solicitation cues” used by women (only the most common behaviors and
those behaviors that were mentioned by another researcher were included in this table). McCormick and Jones (1989) referred to “nonverbal flirting
behaviors.” Grammer et al. (2000) described “courtship signals.”
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
and things they like” (1971, p. 1). Mehrabian submits that if the
perceiver expects an interaction to be reinforcing, the perceiver
engages in approach-oriented behavior. Furthermore, Zajonc
(2000) proposed that the anticipation of a reward may not be
necessary; that simply being exposed to a stimulus without a
negative consequence is sufficient to produce an “approach ten-
dency” (p. 225) via classical conditioning. And finally, outside the
realm specific to attraction, a multitude of other models of human
behavior similarly posit that a positive affective/emotional process
produces approach-oriented behavior (e.g., Baron-Cohen, 1995;
Carver & White, 1994;Lazarus, 1991;Smith, 1991;Watson &
Tellegen, 1985).
In summary, the experience of attraction produces behavior to
reduce the physical distance between the two persons (e.g., seating
distance, standing distance) and secure the acquisition of rewards
(talking/conversing with the person, setting up future meetings).
We label this set of actions approach behaviors.
Develop/Restore Interdependence
Some theorists view interpersonal attraction as an emotion (e.g.,
Lamy, 2016;Smith, Tong, & Ellsworth, 2016), in which the
subjective experience is not only associated with enacted behavior,
but that the facilitation of a behavioral response is a component of
the emotional experience. From this functionalistic approach to
emotion (e.g., Campos et al., 1989;Ekman, 1992;Izard, 1977;
Nesse, 1990;Tooby & Cosmides, 1990), emotions are adaptive
responses to specific environmental problems faced in our evolu-
tionary history. Critically, given the social milieu of our evolu-
tionary past, emotions are associated with behavioral adaptations
with an interpersonal communicative function.
A fundamental tenet of this approach is response coherence, in
which each emotion is associated with a specific pattern of behav-
ioral and physiological responses suited for the demands of the
situation (Leary, Landel, & Patton, 1996;Rosenberg & Ekman,
1994;Tracy & Robins, 2004; for a review, see Keltner, Tracy,
Sauter, Cordano, & McNeil, 2016). Ekman (1992), for instance,
proposed that changes in facial morphology (e.g., the Duchenne
smile) expresses information to other people regarding the expres-
sor’s emotional state to change the recipient’s behavioral response.
Fear, for example, is not only associated with dilated pupils and
heavier breathing, but also with an avoidant behavioral motivation
to appease potential aggressors and warn friends of potential
danger (e.g., Adolphs, Russell, & Tranel, 1999;Bavelas, Black,
Chovil, Lemery, & Mullett, 1988;Bavelas, Black, Lemery, &
Mullett, 1986).
From this perspective, attraction motivates behavior not to sim-
ply approach, but it motivates the expression of trust-producing
behavior to maintain or deepen the degree of interdependence. As
an example, consider one’s attitude toward chocolate and how one
will act toward a piece of chocolate. In such a nonsocial context,
if one evaluates chocolate favorably, one can simply approach and
eat the chocolate—there is no need to negotiate with the chocolate
to arrive at an understanding that eating the chocolate is a mutually
acceptable course of action. However, interactions with other
people can often be considered as social exchanges, which involve
a distribution of resources conditional on the other person’s com-
pliance with the social contract (Axelrod & Hamilton, 1981;Triv-
ers, 1971). If an interdependent relationship is desired with a
stranger (because the stranger has, say, excellent basketball skills
for a teammate, expertise for a collaboration, or physical attrac-
tiveness for a dating partner), a social interaction is necessary to
negotiate the conditions of the exchange. Importantly, both per-
sons need to have an expectation that the other person will uphold
their side of the exchange (Cosmides & Tooby, 1992;Molm, 2006;
Thibaut & Kelley, 1959). Expressions of behavior that generate
trust is one method to communicate one’s conditional cooperative-
ness during the social exchange.
Investigations into the nonverbal behaviors associated with the
development of trust/rapport have most consistently identified the
following four behaviors: smiling, eye contact, mimicry, and phys-
ical proximity (e.g., Andersen, Andersen, & Jensen, 1979;Berni-
eri, Gillis, Davis, & Grahe, 1996;Tickle-Degnen, Rosenthal, &
Harrigan, 1989).
Smiling. The relation between smiling and the production of
trust has received considerable attention. The developmental psy-
chology literature, for instance, has proposed that the smiles ex-
changed between caregiver and infant are instrumental for
strengthening the bond between them (Messinger, Fogel, & Dick-
son, 2001;Messinger & Fogel, 2007). From the social psychology
literature, research notes that smiles are detected, perceived, and
used as a cue to develop trust (Cashdan, 2004;Fridlund, 1994;
Godoy et al., 2005;Krumhuber, Kappas, & Manstead, 2013;
Mehu, Little, & Dunbar, 2007;Scharlemann, Eckel, Kacelnik, &
Wilson, 2001). Several researchers have gone farther to propose
that smiles are reliable cues to cooperation and trustworthiness,
such that they are “costly signals” during a social exchange, as
they increase the chance of being taken advantage (Centorrino,
Djemai, Hopfensitz, Milinski, & Seabright, 2011;Mehu & N=Di-
aye, 2010).
Eye contact. Substantial research points to the importance of
eye contact to the development of rapport and trust. Specifically,
eye contact operates by focusing attention (Cappella, 1981;Ken-
don, 1967;Posner, 1980). Via this “focusing” mechanism, eye
contact can be used to strengthen interpersonal trust and develop
deeper levels of intimacy (e.g., Jones, DeBruine, Little, Conway,
& Feinberg, 2006;King, Rowe, & Leonards, 2011;Mason, Tat-
kow, & Macrae, 2005;Vuilleumier, George, Lister, Armony, &
Driver, 2005;Wyland & Forgas, 2010). Specific research has
concluded that individuals initiate and maintain more eye contact
with people they know well compared with people they do not
know well (Coutts & Schneider, 1976;Russo, 1975), and use eye
contact instrumentally to increase/decrease social distance (Ander-
sen et al., 1979;Mehrabian, 1971).
Mimicry. Multiple research domains advance the claim that
mimicry of facial expressions, vocalizations, and posture is em-
ployed functionally to enhance rapport and trust. For example, the
developmental psychology literature not only proposes that mim-
icry occurs between infants and mothers (via matching vocaliza-
tions, eye gaze, and head turns; e.g., Cappella, 1981;Condon &
Sander, 1974;Gewirtz & Boyd, 1977;Matarazzo & Wiens, 1972;
Simner, 1971;Trevarthen, 1979), but that such mimetic behavior
is important for strengthening the bond between them (e.g., Beebe
et al., 1982;Bernieri, Reznik, & Rosenthal, 1988;Fogel, Toda, &
Kawai, 1988). From the social psychology literature, mimicry is
hypothesized to be expressed instrumentally to generate liking
and/or rapport between persons (e.g., Ashton-James, van Baaren,
Chartrand, Decety, & Karremans, 2007;Farley, 2014;Giles, 1977;
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Guéguen, 2009;Street & Giles, 1982;Wang, 2012). Moreover,
both the developmental and social psychological research domains
have emphasized the importance of social goals to the expression
of mimetic behavior. Tiedens and Fragale (2003) and Stel et al.
(2010), for example, both concluded that mimicry results only
when affiliation goals are salient (see also Bavelas et al., 1986;
Bernieri, 1988;Bernieri, Davis, Rosenthal, & Knee, 1994;Bernieri
& Rosenthal, 1991;Kendon, 1970;LaFrance & Broadbent, 1976).
Proximity. Although proximity is commonly considered to be
attraction’s antecedent (i.e., proximity breeds liking; e.g., Fest-
inger, Schachter, & Back, 1950;Segal, 1974) or consequence (i.e.,
we are drawn to people we like; Allgeier & Byrne, 1973;Byrne,
Ervin, & Lamberth, 1970), from a functionalistic perspective,
proximity is used instrumentally to develop trust. Numerous stud-
ies have identified a link between attraction and physical proximity
(e.g., Argyle & Dean, 1965;Golding, 1967;Hayduk, 1978;Kleck,
1969;Little, 1965;Lott & Sommer, 1967;Mehrabian, 1968a;
Mehrabian, 1968b;Patterson, 1968), and laboratory investigations
have noted that proximity itself acts to communicate interest
(Rosenfeld, 1965) and is associated with more cooperation (Sally,
1995). In the dating realm, a manifestation of “functional proxim-
ity” is when women stand/sit near those people who interest them
and thus “put themselves in a position” to be met (Perper, 1989;
Perper & Weis, 1987).
Summary. The behaviors associated with self-reported attrac-
tion are those behaviors associated with the development of trust/
rapport. In addition to the behaviors that involve direct involve-
ment (e.g., talking to, meeting with), behaviors associated with
attraction should include those behaviors linked to the production
of trust: smiling, eye contact, mimicry, and proximity (called
affiliative behaviors).
Summary of the Models
The three approaches each provide a different explanation for
the relation between self-reported attraction and enacted behavior.
From the subjective experience approach, the positive affective
state does not, in and of itself, produce behavior, but rather,
attraction covaries with, or is misattributed to, another process
(e.g., cognitive consistency, interest, happiness/enjoyment) that
then produces the behavior. The result of which are behaviors that
are associated with specific emotional/cognitive states (emotion-
specific behaviors for happiness/enjoyment, interest, sexual de-
sire). For the generate approach perspective, the positive affective
state produces approach behavior (e.g., proximity, talking/conver-
sation) to acquire the benefits/rewards associated with the target
person. And third, for the develop/restore interdependence ap-
proach, the expression of attraction is reserved for those behaviors
that communicate trust through affiliative behavior (e.g., smiles,
eye contact).
Moderators of the Relation Between
Attraction and Behavior
Whereas the aforementioned literature contends that the expe-
rience of attraction may be associated with a constellation of
enacted behavior, it is easy to identify a number of reasons for the
absence of a relation. Attitude-behavior consistency theorists, for
instance, noted early on that affective/emotional responses were
insufficient to predict behavior (Ajzen & Fishbein, 1973). Early
studies into the relation between interpersonal attraction and be-
havior consistently revealed little to no relation between the two,
further supporting the conclusion that affective responses are in-
sufficient to predict behavioral expressions of attraction (e.g.,
Byrne, Baskett, & Hodges, 1971;Latta, 1976;Snyder & Endel-
man, 1979). However, later research into attitude-behavior consis-
tency (e.g., Ajzen, 1988;Fishbein & Ajzen, 2010) and emotion
coherence (e.g., Mauss, Levenson, McCarter, Wilhelm, & Gross,
2005) not only revealed a significant relation between attitudes/
emotions and enacted behavior, but also identified important mod-
erators and measurement issues that must be considered. In this
section, we discuss moderators that may affect the magnitude of
the relation between self-reported attraction and enacted behavior.
Evaluative Threat Salience
A key consideration that marks interpersonal interactions cen-
ters on concerns over acceptance/rejection. Threats to one’s self-
concept that accompany actual or expected rejection arouse the
need to belong (Baumeister & Leary, 1995;Reis, Collins, &
Berscheid, 2000) and the need to evaluate oneself favorably (Leary
& Baumeister, 2000). This threat activates self-protective systems
to shield the person from harm.
First, protective motives are particularly responsible for reduc-
ing attraction to people who are seen as likely to harm them (via
rejection, undesirable social comparisons, etc.). Such a proposition
is consistent with multiple models of self-evaluation and self-
esteem. The self-evaluation maintenance model (Tesser, 1988,
2000), for example, posits that one way to reduce threat is to
distance oneself from a close other who surpasses the self on a
relevant dimension. Similarly, terror management theory (Pyszc-
zynski et al., 1995;Pyszczynski, Greenberg, Solomon, Sideris, &
Stubing, 1993) posits that distancing oneself from other people
reduces the vulnerability from a feared fate. In addition, other
researchers and models each propose that reducing attraction in
threatening situations protects one’s self-esteem (Cialdini et al.,
1976;Snyder, Lassegard, & Ford, 1986;Vangelisti, 2001).
Second, multiple theorists have proposed a relation between the
salience of evaluative concerns and the prevalence of indirect
behavior. Most famously, Goffman (1971) observed that daters are
prone to play the “relationship game,” in which they express
behavior that could be construed as mere friendliness or romantic
interest, depending on how romantic overtures may be perceived.
The indirect behavioral expression of attraction “protects” the
actor from the possibility that the recipient does not share the same
level of romantic interest (Whitty, 2004; see also Grammer, 1990;
Symonds, 1972). In one study, the majority of participants reported
that the fear of rejection prevented them from pursuing a romantic
relationship with a desired person (Vorauer & Ratner, 1996), with
later research submitting that daters engaged in “self-protective
ambiguity” to help save face while on a date (Vorauer, Cameron,
Holmes, & Pearce, 2003).
In summary, there are two proposed consequences of this self-
protection motive on attraction-related processes. Specifically,
when evaluative concerns are salient, there should be (a) a weaker
link between attraction and direct behavior to help protect the
person from rejection and/or negative evaluations and (b) a greater
relation between attraction and indirect behavior.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Attraction Measurement
Because attraction has long been considered an attitude (e.g.,
Berscheid, 1985) and thus modeled using the tripartite model, it is
most commonly measured using scales that assess the affective
component (affective attraction) and/or the behavioral component
(behavioral attraction). In general, affective attraction aligns
closely with behavioral attraction (e.g., Byrne, 1971;Luther, Ben-
kenstein, & Rummelhagen, 2016;Michinov & Monteil, 2002;
Montoya & Horton, 2014). Enacted behavior’s relation to affective
versus behavioral attraction, however, may wax and wane as a
result of the same processes that regulate the relation between
self-reported attraction and enacted behavior.
Behavioral attraction may respond differently than affective
attraction when any one of many evaluative threats are salient. In
nonthreatening contexts (e.g., anonymous ratings of persons on a
computer screen) both affective and behavioral attraction should
align similarly with enacted behavior. However, in threatening
contexts (e.g., during a romantic date with a highly attractive
partner), affective attraction should be a poorer predictor of direct
behavior (i.e., behavior that is less ambiguous) due to self-
protection motives (e.g., fear of rejection, fear of negative social
comparisons; for a discussion, see Montoya & Horton, 2014).
Cognitive Appraisal Salience
The salience of information regarding the target person may
moderate the strength of the relation between self-reported attrac-
tion and enacted behavior. On the one hand, the elaboration-
consistency hypothesis (Petty, Haugtvedt, & Smith, 1995;Petty &
Wegener, 2010) submits that attitudes that result from deliberative
processing produce greater attitude-behavior consistency than at-
titudes formed via automatic/peripheral processing. This suggests
that effortful thinking about the target person strengthens the link
between attraction and behavior. Moreover, attitude-behavior con-
sistency is stronger when attitudes are made salient (i.e., asking
participants to think about their attitude; e.g., Brown, 1974;Prislin,
1988;Shavitt & Fazio, 1991;Snyder & Swann, 1976) and acces-
sible (i.e., asking participants to think about the stimulus object;
e.g., Cacioppo, Petty, Kao, & Rodriguez, 1986;Fazio, 1989;Fazio
& Williams, 1986). In addition, from the cognitive appraisal ap-
proach to emotions (e.g., Smith & Lazarus, 1993), a cognitive
assessment is fundamental to the subjective experience and the
behavioral response. Thus, making salient one’s cognitive ap-
praisal of the target person (e.g., rating or thinking about the target
person’s attributes) should produce more polarized subjective ex-
periences, and thus enhance the relation with enacted behavior.
On the other hand, research has also found that asking partici-
pants the reasons for why they feel as they do produces lower
attitude-behavior consistency (Wilson & Dunn, 1986;Wilson,
Dunn, Kraft, & Lisle, 1989;Wilson, Kraft, & Dunn, 1989). One
explanation for this finding is that attitudes are determined primar-
ily by emotional processes, but the act of generating reasons
changes the foundation of the attitude from an emotional one to a
cognitive one. A cognitively driven attitude, compared with an
emotionally driven attitude, would then be hypothesized to be less
predictive of enacted behavior (Millar & Tesser, 1986).
Romantic/Sexual Salience
Some conceptualizations of attraction consider it to be reserved
for romantic/sexual contexts. From this view, behaviors such as
smiling and eye contact are expressed because they communicate
romantic attraction (e.g., Guéguen, Fischer-Lokou, Lefebvre, &
Lamy, 2008;Hazan & Zeifman, 1999), and thus the relation
between attraction and enacted behavior should be evident only
when romantic/sexual motivations are present.
Alternatively, other researchers (e.g., Mehrabian, 1969;Tickle-
Degnen et al., 1989) present taxonomies of behaviors in which the
various behaviors may be expressed across social contexts (e.g., at
work, doctor-patient interactions). From this perspective, the rela-
tion between attraction and enacted behavior should be present
even in nondating interactions and in platonic relationships (e.g.,
friendships, coworkers).
Purpose of the Meta-Analysis
The goal of this research was to explore the magnitude and
moderators of the relation between self-reported attraction and
enacted behavior. This research focused on (a) identifying which
specific behaviors are related to self-reported attraction; (b) testing
the impact of several moderators, including evaluative threat sa-
lience, type of attraction assessment, cognitive appraisal salience;
and (c) investigating the degree of agreement between meta-
analytic findings and an ethnographic analysis. To do so, we
collected data from studies that assessed participant’s self-reported
affective and/or behavioral attraction toward a specific person and
the participant’s enacted behavior toward that same person (e.g.,
number of smiles, amount of eye contact).
The first step was to investigate which specific behaviors were
associated with self-reported attraction. This analysis will provide
a typology of behaviors associated with the experience of attrac-
tion and may provide insight as to which theoretical explanation
for the relation between attraction and behavior is best supported.
Specifically, the subjective experience approach submits that en-
acted behaviors should align with the specific emotional responses
(e.g., smiling from happiness/enjoyment, lip biting from sexual
interest; emotion-specific behavior). The generate approach per-
spective submits that, given its theoretical foundation of movement
toward reinforcing/rewarding stimuli, attraction is associated with
direct behaviors (e.g., sitting near, talking to; approach behavior).
Finally, the develop/restore interdependence perspective submits
that behaviors associated with producing trust are related to attrac-
tion (e.g., smiling, mimicry; affiliative behavior).
We also examined the degree to which the findings from the
meta-analysis align across different cultures. Given that the ma-
jority of published research is conducted in Western countries, we
examined whether the subset of enacted behaviors associated with
self-reported attraction in this meta-analysis were similar to the
behaviors reported to be related to attraction in cultures from
around the world.
Finally, although not directly related to the hypotheses, this
research allowed for the testing of various sex differences associ-
ated with the relation between attraction and behavior. In Western
societies, for instance, norms dictate that men are to overtly initiate
dating relationships (Rose & Frieze, 1989;Symonds, 1972; see
also Laner & Ventrone, 1998;Rose & Frieze, 1993). Consistent
with these norms, men, compared with women, use more direct
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
relationship initiation strategies (Greer & Buss, 1994;Smythe,
1991), are more likely to ask someone out for a date (Kelley &
Rolker-Dolinsky, 1987), are more active and direct at the begin-
ning of the relationship, and are the “relationship initiators” (Clark,
Shaver, & Abrahams, 1999). Alternatively, norms dictate that
women are to communicate romantic interest to their partner to
reduce uncertainty regarding her partner’s motives and to grow her
partner’s interest, but they are also to be coy and discreet when
expressing their relational interest (Simon, Eder, & Evans, 1992).
These findings indicate that in interactions when romantic/sexual
norms are salient (e.g., men interacting with women during a speed
dating session), there should be a strong relation between attraction
and direct behavior for men, but for women, there should be a
strong relation between attraction and indirect behavior.
Literature Search
We searched the literature using the PsycINFO, Sociological
Abstracts, Google Scholar, and Dissertation Abstracts Interna-
tional databases until the end of August 2017. Our search strategy
included a combination of the following keywords: attraction,
dating, social distance, immediacy cues, ingratiation, propinquity,
proximity, mimicry, proxemics, liking, behavioral confirmation,
standing/sitting distance, smiling, eye contact/gaze, flattery, and
flirting. We conducted a backward search of reference sections of
the retrieved articles until we found no new entries. We sent
requests for relevant studies to Internet discussion forums used by
social psychologists (e.g., Society for Personality and Social Psy-
chology Connect [SPSP Connect]) and communication researchers
(e.g., Communication Research and Theory Network [CRTNET]).
We then contacted investigators whom had frequently published
research in the area.
Inclusion/Exclusion Criteria
We included studies that satisfied four criteria. First, partici-
pants’ attraction to a target person must have been assessed using
a self-report measure (e.g., Interpersonal Judgment Scale [IJS];
Byrne, 1971) that was comprised primarily of affective or behav-
ioral attraction items. Assessments of so-called “cognitive attrac-
tion,” which may include an assessment of physical attractiveness,
honesty, or intelligence, are not attraction studies per se (Montoya
& Horton, 2014), and were excluded.
Second, studies must have included an enacted (vs. “preferred”
or “desired”) behavior. The difference between the measurement
of actual versus preferred behavior is far from trivial. For example,
in a review of the literature on personal space, Hayduk (1983)
concluded that the relation between enacted behavior (as measured
by indices such as chair placement or stop distance) and “projec-
tive” measures of personal distance (e.g., self-report indices of
distance, such as a paper and pencil task that asked participants to
place a line where they would be comfortable standing) was
“unacceptably low,” and that such “projective” measures were
“unacceptable measures of personal space” (p. 296).
Third, we only included studies in which the behavior was freely
chosen. We thus excluded research that experimentally manipu-
lated a behavior (e.g., placing the participant either one foot or four
feet from an interviewer) and then assessed the resultant attraction
(e.g., O’Connor & Gifford, 1988;Williams & Kleinke, 1993).
Finally, we did not include studies that investigated the link
between disliking (or prejudice) and avoidance behavior (or dis-
crimination). Although conceptually appealing, attraction should
not be considered the “opposite” of negatively valenced constructs
like repulsion or prejudice. Attraction is likely to be correlated
with such constructs (e.g., Dermer & Pyszczynski, 1978;Rubin,
1974) but the underlying physiological, cognitive, and neurochem-
ical processes that regulate these “negative” processes differ from
those that regulate attraction (e.g., Carter, 1998;Kelley, 1983;
Sternberg, 1987;Swensen, 1972). For example, prejudice, com-
pared with attraction, is associated with different neural structures
(Harmon-Jones & Allen, 2001;Winkielman & Cacioppo, 2001), is
better at predicting discrimination than is attraction (Pittinsky,
Rosenthal, & Montoya, 2011), and is associated with different
interpersonal motives (e.g., Bower, 1981;Fredrickson & Branigan,
Selected Studies
An initial search of the literature resulted in a sample of
1,037 articles. Given the aforementioned exclusion criteria, the
reasons for exclusion were primarily (a) the absence of an index
of self-reported attraction, which resulted in the removal of 226
articles (21.79%); and (b) not including an enacted behavior
(360 articles were excluded, 34.71%). Other reasons (e.g.,
failing to report sufficient statistical information, participants’
behavior manipulated by the experimenter) resulted in the re-
moval of 397 articles.
The search strategy and selection criteria resulted in 54 articles,
of which 43 (79.63%) were published journal articles, three
(5.56%) were unpublished theses or dissertations, three (5.56%)
were book chapters, and five (9.26%) were unpublished databases
provided by the original researchers. The sample included 309
effect sizes (M5.72 per article, SD 9.81) and included 5,422
participants. Sample sizes ranged from six to 382 (M60.65,
SD 58.49).
Coder Reliability
Two graduate students coded the articles after being trained by
R. Matthew Montoya. Reliability was assessed by comparing the
coders’ ratings with each other and with those of the first author.
Reliabilities were examined using Krippendorff’s alpha (Krippen-
dorff, 2004). The mean reliability indices between the two raters
was strong, .97 (SD 0.08; range .61 to 1.00), with agreement
rates for 21 of 23 category codes exceeding .90. The mean alphas
between the first author and the raters was also strong across the
Research indicates that women, compared with men, are more effective
at expressing and interpreting nonverbal cues (for a review, see DePaulo,
1992) and that they engage in these nonverbal behaviors more frequently.
For example, women and adolescent girls smiling more frequently than
men and adolescent boys (LaFrance, Hecht, & Paluck, 2003), and women
sit closer to other women than men sit with other men (Baxter, 1970). We
do not investigate whether men or women engage in more behavioral
expressions of attraction or experience more/less attraction, but rather,
whether there is a greater/lesser concomitance between attraction and
behavioral expressions as a function of the participant’s sex.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
various ratings, .95 (SD 0.06; range .72 to 1.00), with
agreement exceeding .90 for 21 of 23 codes. When a disagreement
was present, the issue was discussed between the raters and then
with the first author until agreement was reached. The consensus
code was used in the analyses.
Data Coding
Enacted behavior. Our list of enacted behaviors was deter-
mined by theoretical predictions and practical considerations. To
be as inclusive as possible, we began by including all behaviors
identified in the retrieved articles and the behaviors mentioned in
Table 1.
This method resulted in the identification of 42 behaviors: eye
gaze duration and frequency (look to the face/head of the target
person; Cappella, 1981), seating distance (measured as the dis-
tance between a chair the participant places and a chair in which
the confederate is to sit), standing distance (measured via [a] a task
in which participants walked toward the target person until they
stopped [stop distance] or [b] an unobtrusive measurement of an
interaction with a confederate), body lean (torso bend beyond
vertical toward the target person), body orientation (degree to
which the participant’s body is facing the target person), open
posture (positioning the arms and legs away from the body; Meh-
rabian & Friar, 1969), laughter (a smile accompanied by an acous-
tically detectable exhaustion of air; Bryant et al., 2016), head cant
(a lateral head tilt toward the shoulder axis; Krumhuber et al.,
2007), talking (word count/duration of a person’s vocalizations;
Cuperman & Ickes, 2009), gesticulation (signaling with the hands
and arms when talking; Özyürek, 2002), primp (smoothing clothes
with hands; Grammer et al., 1999), hair flip (tipping the head
forward followed by throwing hair back; Grammer et al., 1999),
head akimbo (arching one’s back and putting one’s hands behind
the neck with elbows pointed up and out; Grammer et al., 1999),
mimicry (imitating another person’s behavior; Lakin & Chartrand,
2003), and head nod (continuous movement of the head up and
down; Woodall, Burgoon, & Markel, 1980).
We also included behaviors from a small subset of articles (e.g.,
Grammer et al., 1999;Grammer, Kruck, Juette, & Fink, 2000):
adaptor, arm flex, arms parallel, breast presentation, cover face,
deictograph, eyebrow flash, flirtatious glance, folding hands, head
down, head toss, knees toward body, look around, look down, look
through, move legs, move shoulders, palm, roll up sleeves, and
shoulder shrug.
We included the duration and frequency of smiles. Some studies
used facial action coding (FACS; Ekman & Friesen, 1976)to
identify “genuine” smiles (i.e., smiles that recruit both the zygo-
matic major muscle and the orbicularis oculi), whereas other
studies did not. Because research has proposed that only “genuine”
smiles are related to positive emotions (Ekman, Davidson, &
Friesen, 1990), we distinguished between those studies that did,
versus did not, code for “genuine” smiles.
Finally, we included two behaviors inspired by speed dating
studies: “yessing” and “actual contact.” “Yessing” refers to the
participant’s expressed interest in communicating with other per-
son again, and “actual contact” refers to whether the participant
actually contacted the other person after the initial session. Al-
though these codes apply to the speed dating studies in our data-
base, they also apply to nonspeed dating studies (e.g., Danyluck &
Page-Gould, 2017 investigated whether participants were inter-
ested in becoming friends).
Enacted behavior type. We divided the behaviors into one of
three categories based on the traditional methods of categorizing
enacted behaviors: direct, indirect, and speed dating. Several be-
haviors were coded as direct: sitting, stop, and standing distance,
and two talking assessments (duration and word count). The indi-
rect behavior category included smiles (duration and frequency),
eye contact (duration and frequency), body lean, body orientation,
mimicry, head tilt, and the other behaviors not considered direct or
speed dating.
The two effects common to speed dating studies, “actual con-
tact” and “yessing,” were included in their own category because
they do not align clearly with either the direct or indirect catego-
ries. Specifically, although these behaviors may appear to align
with “direct behavior,” they have critical differences that necessi-
tate their own category. In contrast to all direct behaviors, “yes-
sing” is not expressed directly to the other person, but anony-
mously to the experimenter, and would be communicated to the
other person if there was a mutual “yes.” Alternatively, “actual
contact” refers to whether a speed dater contacts another dater (via
e-mail, phone call) after the initial speed dating session. Actual
contact is distinct from the direct behaviors because it is contingent
on the other person’s mutual consent to contact (i.e., the participant
can contact another person only if mutual “yessing” occurred).
Evaluative threat salience. We coded evaluative threat sa-
lience as a categorical variable with two levels: high salience, low
salience. Salience was coded as “high” when it was experimenter-
defined or when concerns regarding acceptance/rejection or rela-
tive performance were considered to be salient. For example, “high
salience” studies included procedures in which participants went
on a blind date or when participants were asked to interview (or be
interviewed by) a physically attractive confederate. Alternatively,
“low salience” studies were those in which evaluative threat was
assumed to be unaroused. For example, we coded studies as “low
salience” when experimenters investigated students’ seating prox-
imity to the instructor in a lecture class or interactions with a
confederate while completing a noninterdependent task.
Cognitive appraisal salience. We used a two-level variable to
assess whether participants considered the attributes of the target
person before making their attraction assessment. We coded as
“cognitive appraisal salient” those studies in which participants
assessed the characteristics/traits of the target person before indi-
cating their attraction to the target. The IJS (Byrne, 1971), for
example, includes four “filler” questions that precede the two
attraction items and ask participants about the attributes of the
target person. Montoya and Horton (2004) argued that such ques-
tions that ask participants to rate the target person’s intelligence,
adjustment, morality, and competence on a given task make salient
their evaluation of the target person. We coded as “cognitive
appraisal not salient” those studies that did not include question-
naires that asked participants to consider the attributes of the target
Type of self-reported attraction assessment. We coded the
type of attraction measure as a categorical variable with four
levels: affective attraction, behavioral attraction, behavioral attrac-
tion (dating), and mixed. Only assessments in which the partici-
pants had the implicit or explicit expectation that their responses
would not be seen by the target person were considered as assess-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
ments of self-reported attraction. We coded indices as “affective
attraction” those measures that assessed attraction using only af-
fective items (e.g., “I feel warmly toward the target person”; “I like
the target person”). We coded indices as “behavioral attraction”
those measures that described participants’ preference or desire to
act in a particular way toward the target person (e.g., “I would like
to talk to this person again”). We also coded indices as behavioral
attraction (dating) those behavioral attraction assessments that
were specific to questions focused on dating (e.g., “I would like to
ask the person out on a date”). Finally, we coded as “mixed” those
indices that included both affective and behavioral attraction as-
sessments. For example, Byrne’s (1971) popular IJS questionnaire
was coded as “mixed” because it includes one affective item
(“How much do think you will like this person?”) and one behav-
ioral item (“How much would you want to work with this per-
Sex composition. In addition to coding for the sex of the par-
ticipant and the sex of the target person, we also created a six-level
factor that considered the sex composition of the interaction: female
participant–female target person, female participant–male target per-
son, male participant–male target person, male participant–female
target person, unspecified participant (participants whose sex was not
mentioned by the researchers)–matched sex target, and mixed (re-
searchers collapsed across different sex compositions).
Romantic/sexual context salience. We coded the salience of
a romantic/sexual context using a two-level variable: salient and
not salient. Salience was coded as present when it was
experimenter-defined or when the situation was designed to allow
for people to meet and possibly date (e.g., speed dating session).
Salience was coded as low when such considerations were con-
sidered to be unaroused (e.g., a laboratory session involving a
same-sexed confederate or a teacher–student interaction).
Other variables. We coded several additional aspects of the
design for exploratory and sensitivity analyses, of which several
are noteworthy. First, in addition to coding the participant’s and
target person’s age, we also coded the age composition of the
sample using a four-level variable: both children, both college-
aged, and two codes to represent children and adults in an inter-
action that were heterogeneous with respect to age. Second, we
coded into a three-level variable the amount of knowledge persons
had about one another before the assessment period: none (e.g.,
minimal interaction [i.e., an experiment using unacquainted per-
sons]), preexisting relationships (e.g., relationship partners, close
friends), and other (e.g., classmates). Third, the time interval
between the assessment of attraction and the assessment of behav-
ior was coded as a three-level variable: under 1 hr, same day, and
longer than 1 day. Fourth, we coded the target person’s behavior
during the interaction using a four-level variable: naïve participant
(fellow participant whose responses were not dictated by the
experimenter), confederate who was instructed to act in a neutral/
nondescript manner toward the participant, confederate who was
instructed to act in an accepting/positive manner toward the par-
ticipant, and confederate who was instructed to act in a rejecting/
negative manner toward the participant. Fifth, we coded the order
in which attraction and behavior were assessed using a two-level
categorical variable: attraction first, behavior first. Studies that
measured behavior before asking participants to complete the
attraction questionnaire were coded as “behavior first.” Alterna-
tively, studies that assessed participants’ attraction before assess-
ing behavior were coded as “attraction first.”
Finally, we coded each effect for basic descriptive information.
These variables included: author and full citation, source (journal,
edited volume, thesis or dissertation, and unpublished article), year of
publication, country in which data were collected, reliability of attrac-
tion assessment and the interrater reliability of the behavior. All
coding decisions for each effect size are presented in Appendix A.
Statistical Methods
Effect size. The effect size of interest was the reported corre-
lation between self-reported assessments of interpersonal attrac-
tion and the index of enacted behavior. Effect size estimates were
converted to zusing Fisher’s rto ztransformation (Fisher, 1928)
because the sampling distribution of z(r) is assumed to approach
normality (Rosenthal, 1984).
Effect sizes were calculated such that greater positive values
indicate a stronger theoretically consistent relation between the
self-reported attraction and enacted behavior (e.g., we reversed the
sign of effect sizes that assessed seating/standing distance’s rela-
tion to attraction so that the effect size measured “proximity”). If
the relation between attraction and behavior was described as not
significant and no additional information in the article was pro-
vided, the effect size was conservatively set to zero (Pigott, 2012).
Of the 309 effect sizes in the database, 57 (18.44%) were estimated
using this technique. Furthermore, in a small number of cases, the
behavior was dichotomous (e.g., amount of attraction experienced
by those participants who opted to sit “close” vs. “far” from a
target person, “yessing”). In these cases, we used logistic regres-
sion to generate B’s. To convert B to z, we first converted B to d
(using the equation provided by Haddock, Rindskopf, & Shadish,
1998), and then converted dto r. The variance of these effects was
estimated using the formula, (1 r
Random-effects model. We selected a random-effects model
because we were interested in making unconditional inferences
that generalized to the hypothetical population of all studies that
could exist, rather than simply to the studies included in the present
sample (Hedges & Vevea, 1998).
Moreover, a majority of articles contributed more than one
effect size (specifically, 39 of 54 articles). To model the depen-
dence inherent in such data, we used a three-level random-effects
meta-analytic approach (multilevel modeling [MLM]; Cheung,
2014;Konstantopoulos, 2011;van den Noortgate, Lopez-Lopez,
Marin-Martinez, & Sanchez-Meca, 2013). MLM addresses con-
cerns produced by dependent data (e.g., estimation of artificially
narrow confidence intervals) by modeling both the dependence
between and within articles. Specifically, MLM modeled three
sources of variance: variance between articles, variance between
effect sizes from the same article, and the sampling variance. In
this case, we estimated effect sizes (level 1, k309) that were
nested within article (level 2; k54). The sampling variance was
computed with maximum likelihood using a computer program
developed for that purpose (Vevea & Woods, 2005). Analyses
were conducted using PROC MIXED (van den Noortgate et al.,
2013) and meta3 in R (Cheung, 2014) using maximum likelihood
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Reliability information was reported for 252 of the 309 effect
sizes (81.55%). As noted in Table 2, reliability for the indices of
self-reported attraction and enacted behaviors was generally ac-
ceptable. Inspection of the reliability estimates for the various
direct and indirect behaviors did not provide evidence that some
behaviors were measured more reliably than other behaviors.
The presence of outliers in our database was investigated using the
metaplus package in R (Beath, 2016). Two effect sizes were
identified as potential outliers. We followed the procedure for
handling outliers described by Lipsey and Wilson (2001) and
Windsorized the effects. This did not affect the pattern or statistical
significance of any reported result.
We began by estimating an unconditional random intercept
model to determine the amount of variation in effect size estimates
at each level of analysis. There was heterogeneity in the effects
across each level. At Level 1, the heterogeneity estimate (I
) was
.56, indicating that 56% of the variance could be explained by
between-study variance. The Qstatistic for the homogeneity of
effect sizes was significant,
(308) 782.62, p.001. When
Level 2 was added to the model, it produced a heterogeneity
estimate of .49, which reflects medium heterogeneity (Higgins,
Thompson, Deeks, & Altman, 2003). A significance test for im-
portance of the level two variability was also significant,
70.31, p.001. Because significant variation in effect sizes is
present at each level of the data, we proceeded by estimating a
multilevel model. As presented in Table 3, the random effects
model revealed that the overall relation between self-reported
attraction and enacted behavior was significant, z.205 (SE
.026), z-score 7.65, p.001.
Which Behaviors are Related to
Self-Reported Attraction?
An initial inspection of the behavior data revealed that several
categories may be collapsed given the similar means and the
consistent patterns across levels of different moderators. First, we
combined the three physical distance measures (seating distance,
standing distance, stop distance) into a single category (sit/stand
distance). Second, consistent with research that failed to find
different meanings between “types” of smiles (e.g., Schneider &
Josephs, 1991;Schneider & Unzer, 1992), we collapsed the effect
sizes for studies that mentioned the Duchenne coding for smiles
and those that did not.
There was a main effect for behavior,
(41) 176.71, p
.001, such that behavior moderated the relation between attraction
and enacted behavior. Fourteen behaviors were significantly re-
lated to attraction and 28 were not, suggesting that academics may
have been incorrect to conclude that such behaviors were associ-
ated with attraction. Moreover, for many of those nonsignificant
behaviors, the attraction-behavior relation was reported (a) selec-
tively (e.g., only the most positive or negative correlations were
reported); and/or (b) as a result of exploratory/atheoretical analy-
ses. For example, Grammer et al. (2000) reported several behav-
iors for which there was no theoretical rationale and were reported
selectively due to their negative relation with attraction.
The inclusion of such behaviors may adversely affect the ability
of the various analyses to detect the impact of the different mod-
erators. As a result, we excluded 22 behaviors from the following
analyses. A list of the behaviors retained in the main analyses is
presented in Table 3, and a list of the excluded behaviors is
presented in Appendix B. A replication of all key analyses with the
full database is reported in the online supplement.
Removing the behaviors reported in Appendix B did not mean-
ingfully affect the overall effect, z.204 (zscore 7.69, p
Behavior Type
The means for behavior type are presented in Table 3. The main
effect for behavior type was significant,
(2) 34.94, p.001.
We explored this main effect using orthogonal contrasts. The first
contrast revealed that the attraction-behavior relation was greater
for speed dating effect sizes compared to the combination of direct
and indirect effects,
(1) 23.09, p.001. The second contrast,
which compared direct with indirect behaviors, was also signifi-
(1) 20.52, p.001, indicating that direct behaviors
were more strongly associated with self-reported attraction than
were indirect behaviors.
Speed dating studies were removed from the following moder-
ator analyses because the speed dating behaviors (i.e., “yessing”
and “actual contact”) were considered as neither direct nor indirect
behaviors. Speed dating effects were analyzed separately.
Evaluative Threat
The means for evaluative threat are presented in Table 3. The
main effect for evaluative threat was not significant,
(1) 1.14,
p.29. We investigated two interactions involving evaluative
threat. However, neither the Evaluative Threat Behavior Type
(1) 1.26, p.26, nor the Evaluative Threat
Cognitive Appraisal interaction,
(1) 1.56, p.22, reached
Self-Reported Attraction Type
The main effect for attraction type was not significant,
0.96, p.61. Orthogonal contrasts revealed that affective attrac-
tion was not greater than behavioral attraction,
(1) 0.23, p
.63, and that a combination of affective and behavioral attraction
did not differ from mixed,
(1) 0.70, p.40.
We next explored the expected moderators of attraction type.
Neither the predicted Evaluative Threat Attraction Type inter-
(2) 2.90, p.24, nor the Attraction Type Behavior
Type interaction,
(1) 0.40, p.53, were significant. How-
ever, of interest was the Evaluative Threat Attraction Type
Behavior Type interaction, in which the different attraction mea-
sures respond differently as a function of evaluative threat and
behavior type. This three-way interaction was significant,
8.13, p.05.
We tested the predicted effects specific to behavioral attraction
and affective attraction as a function of the Evaluative Threat
Behavior Type interaction. For affective attraction, the Evaluative
Threat Behavior Type interaction was significant,
(1) 5.11,
p.05. Simple effects revealed that the effect for direct behavior
was lower when threat was salient (no threat: z.33, SE 07;
threat: z.13, se .05), t2.16, p.05, but threat salience did
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 2
Study Characteristic Summary
Category/code Number of
effects Category/code Number of
Publication year, by decade Country
1960’s 2 United States 168
1970’s 38 Germany 42
1980’s 24 Netherlands 18
1990’s 90 Ireland 1
2000’s 59 France 2
2010’s 96 Austria 24
New Zealand 5
Publication type Canada 18
Published article/chapter 251 Belgium 1
Thesis/dissertation 9 Japan 30
Unpublished data 44
Book chapter 5 Target person age
Child/adolescent 9
Participant sex Undergraduate 277
Always men 106 Older than undergraduate 23
Always women 163
Mixed 40 Target person behavior
Fellow participant 183
Participant age Neutral acting confederate 87
Child/adolescent 9 Positive acting confederate 17
Undergraduate 284 Negative acting confederate 11
Older than undergraduate 16
Romantic/sexual salience
Sex composition (person-target) High 195
Man–man interaction 15 Low 114
Man–woman interaction 96
Woman–woman interaction 36 Pre-existing familiarity
Woman–man interaction 132 None 305
Always same sex interaction 97 Friends 1
Always other sex interaction 9 Large group (e.g., in class) 3
Mixed 14
Delay between assessments
Attraction measure type Under one hour 298
Affective attraction 65 Same day 4
Behavioral attraction 69 Longer than one day 7
Mix of affective and behavioral items 54 Assessment order
Attraction first 55
Evaluative threat salient Behavior first 263
Yes, implied or explicit 218 Both before and after 1
No 91
Reliability of attraction measure Cognitive appraisal salience
Single item 54 Yes, present 59
Unacceptable/low 0 No 250
Moderate 25
Strong 136 Reliability of behavior index
Almost perfect 21 Single item 51
Did not report 73 Unacceptable/low 14
Moderate 48
Behavior type Strong 73
Direct behavior 59 Almost perfect 80
Indirect behavior 227 Did not report 42
Speed dating 23
Note. Values may not sum to 309 due to missing values. Interobserver/interrater reliability and internal
consistency were assessed using a number of metrics, and summarized thusly: moderate (for Cohen’s Kappa,
.60–.79; McHugh, 2012; for correlations, .70–.79, LeBreton, Burgess, Kaiser, Atchley, & James, 2003; for
Cronbach’s alpha, .70–.79, Streiner, 2003; for Krippendorff’s alpha, .66–.80, Krippendorff, 2004; for Kendall’s
W, .40–.60, Cain et al., 2005), strong (for Cohen’s Kappa, .80–.90; for correlations, .80 –.89; for Cronbach’s
alpha, .80–.89; for Krippendorff’s alpha, .80; for Kendall’s W, .60–.80), and almost perfect (for Cohen’s
Kappa, .90; for correlations, .90; for Cronbach’s alpha, .90; for Kendall’s W, .80).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 3
Mean Effect Sizes for Each Level of the Moderators
Category Behavior
type Model prediction kzSELower CI Upper CI zscore
Empty model
Overall effect 309 .205 .026 .152 .257 7.65
Overall, w/o excluded behaviors 267 .204 .026 .152 .256 7.69
“Yessing” speed dating se ga d/r 16 .494 .052 .393 .595 9.57
Talk (word count) direct ga d/r 9 .298 .061 .178 .418 4.87
Mimicry indirect d/r 11 .253 .075 .106 .400 3.37
Future contact speed dating se ga d/r 13 .216 .071 .077 .355 3.03
Sit/stand distance direct ga d/r 30 .207 .036 .136 .278 5.70
Talk (duration) direct ga d/r 18 .191 .045 .102 .279 4.20
Eye gaze (frequency) indirect se d/r 20 .189 .047 .097 .281 4.01
Head nod indirect 8 .162 .069 .027 .298 2.34
Eye gaze (duration) indirect se d/r 38 .145 .035 .077 .214 4.13
Smile (frequency) indirect se d/r 21 .131 .051 .032 .230 2.58
Laughter indirect se d/r 18 .130 .046 .039 .221 2.79
Gesticulation indirect 5 .108 .064 .017 .233 1.68
Primp indirect 5 .103 .082 .057 .263 1.26
Head cant indirect 7 .066 .060 .053 .184 1.08
Smile (duration) indirect se d/r 10 .063 .054 .042 .169 1.17
Forward lean indirect se 10 .006 .057 .105 .116 .10
Body orientation indirect se 10 .005 .060 .113 .124 .08
Open posture indirect se 14 .004 .048 .090 .098 .07
Head akimbo indirect 5 .051 .082 .211 .109 .62
Hair flip indirect 5 .120 .082 .280 .040 1.47
Behavior type
Indirect 187 .110 .023 .064 .156 4.72
Direct 59 .225 .028 .170 .280 8.04
Speed dating 23 .436 .047 .343 .530 9.13
Evaluative threat
High 160 .130 .034 .061 .198 3.72
Low 84 .183 .034 .116 .251 5.31
Self-reported attraction type
Affective attraction 57 .162 .039 .084 .240 4.06
Behavioral attraction 140 .154 .044 .067 .240 3.50
Mix of affective and behavioral 47 .187 .046 .095 .279 4.00
Cognitive appraisal
Present 54 .225 .047 .132 .319 4.72
Absent 190 .133 .028 .077 .188 4.73
Sex composition (participant–target)
Man–man interaction 12 .164 .061 .044 .284 2.68
Man–woman interaction 73 .130 .032 .066 .195 3.98
Woman–man interaction 106 .122 .031 .060 .185 3.85
Woman–woman interaction 32 .099 .052 .003 .203 1.89
Always same sex interaction 7 .196 .084 .030 .362 2.32
Mixed 14 .297 .055 .165 .383 4.93
Target person behavior
Naïve participant 126 .152 .043 .067 .237 3.50
Neutral-acting confederate 79 .133 .040 .054 .213 3.29
Positive-acting confederate 17 .175 .085 .008 .342 2.06
Negative-acting confederate 11 .138 .097 .053 .329 1.41
Assessment order
Behavior assessed first 218 .145 .025 .094 .195 5.63
Attraction assessed first 21 .228 .050 .129 .328 4.49
Romantic/sexual salience
High 138 .133 .039 .055 .210 3.37
Low 106 .173 .032 .110 .236 5.41
Note.CIconfidence interval. Positive values indicate stronger relation between behavior and attraction. k’s for the various moderator analyses may
not total to 244 due to missing values. The Model Prediction column indicates the behaviors predicted to be related to attraction by the various theoretical
approaches (se subjective experience; ga generate approach; d/r develop/restore interdependence).
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
not affect the effect for indirect behavior, t0.46, p.63. For
behavioral attraction, the Evaluative Threat Behavior Type
interaction was not significant,
(1) 1.84, p.17. A test of the
simple effects revealed that the relation between behavioral attrac-
tion and direct behavior under evaluative threat (z.21, SE 04)
was not significantly lower than when threat was not salient (z
.18, SE 09), t0.33, p.74.
The relation of behavioral attraction and enacted behavior was
also investigated by dividing behavioral attraction by those assess-
ments that referred specifically to dating-related questions (behav-
ioral attraction-dating) from those that did not. This subset of
behavioral attraction questions (k78) was not different than
zero, z.09 (SE .06), t-value 1.48, p.13. The behavioral
attraction questions not specific to dating (k62) continued to be
significantly different than zero, z.15 (SE .04), t3.58, p
Cognitive Appraisal Salience
The means for cognitive appraisal are presented in Table 3. The
main effect for cognitive appraisal salience was marginal,
3.44, p.07, with the relation between attraction and enacted
behavior descriptively greater when preceded by a cognitive ap-
praisal. We investigated whether cognitive appraisal salience mod-
erated the influence of another moderator. Neither the Cognitive
Appraisal Behavior Type,
(1) 0.01, p.93, nor the
Cognitive Appraisal Evaluative Threat Behavior Type inter-
action was significant,
(1) 0.85, p.35.
Romantic/Sexual Context Salience
The main effect for romantic/sexual salience was not significant,
(1) 0.63, p.42. Moreover, neither the Romantic/Sexual
Salience Behavior Type interaction,
(1) 0.02, p.89, nor
the Romantic/Sexual Salience Attraction Type interactions,
(2) 1.34, p.51, was significant.
Sex Composition Analyses
We conducted analyses to evaluate the impact of the sex of the
participant and target person on the magnitude of the relation
between attraction and behavior. The means are presented in Table
3. First, we tested the six-level factor that accounted for the sex of
the participant and the sex of the target person. The main effect for
sex composition was not significant,
(5) 6.55, p.26.
Second, we investigated the impact of dyad composition by in-
specting same- versus other-sex dyads. The magnitude of the
difference between self-reported attraction and enacted behavior
for same (z.14, SE .04) versus other-sex (z.12, SE .03)
dyads was not significant,
(1) 0.11, p.74.
Next, we investigated two specific predictions regarding sex
composition as it relates to behavior type. First, in women’s
interactions with men, direct behaviors better aligned with self-
reported attraction (z.22, SE .04) than did indirect behaviors
(z.09, SE .03),
(1) 6.72, p.01. However, the same
difference was not evident in men’s interactions with women, in
which direct behaviors (z.17, SE .04) and indirect behaviors
(z.11, SE .03) did not differ,
(1) 1.82, p.17.
Second, to test the prediction that men are more likely to “act on
their feelings directly” than women when the romantic/sexual
context is salient, we compared male-to-female interactions with
male-to-male interactions, and female-to-male with female-to-
female interactions. For men, romantic/sexual salience did not
affect the relation between attraction and behavior for either indi-
(1) 0.09, p.76; or direct,
(1) 0.004, p.95,
behavior. The same held for women, such that dating salience did
not affect the relation between attraction and behavior for either
(1) 0.04, p.83; or direct,
(1) 0.40, p.52,
Speed Dating
The means for the two behaviors associated with speed dating
studies are reported in Table 3. Yessing was more strongly related
to self-reported attraction than was “actual contact”,
(1) 8.85,
p.01. We tested whether these effects were moderated by the (a)
cognitive appraisal salience and (b) participant’s sex. With respect
to cognitive appraisals, when the cognitive appraisal was salient,
the relation between attraction and “yessing” was not stronger (z
.56, SE .08) than when it was not salient (z.47, SE .03),
(1) 0.87, p.36. With respect to participant sex, there were
no sex differences for the relation between attraction and “yessing”
.46, z
.42), t0.67, p.52, or for the relation
between attraction and initiating contact (z
.44, z
.28), t1.33, p.22.
Sensitivity Analyses
We conducted additional analyses to determine if a poten-
tially important moderator was excluded from the analyses.
Specifically, our interests were to determine: (a) whether other
moderators of the relation between attraction and behavior
would emerge; and (b) whether these findings would produce a
significant interaction with another moderator (i.e., type of
attraction assessment).
Assessment order. We tested whether the relation between
attraction and behavior differed as a function of the order in which
attraction and behavior were assessed. The test of assessment order
indicates that the relation between attraction and behavior was not
significantly greater when attraction was assessed first,
1.57, p.21. We also tested whether the impact of assessment
order was moderated by attraction type and behavior type. The
Assessment Order Attraction Type interaction was not signifi-
(2) 1.60, p.45. However, the Assessment Order
Behavior Type interaction was significant,
(1) 5.15, p.05.
Attraction was a better predictor of direct behavior when attraction
was assessed before behavior (z.32, SE .05) than when
behavior was assessed before attraction (z.19, SE .03),
(1) 4.79, p.05. In addition, a post hoc, but theoretically
important analysis revealed that when attraction was assessed
before behavior, attraction was a better predictor of direct behavior
than indirect behavior (z.05, SE .07),
(1) 12.18, p
Target person behavior. We investigated the impact of the
other person’s behavior during the interaction. The main effect for
target person’s behavior was not significant,
(3) 0.72, p.86.
Moreover, the potentially informative Target Person Behavior
Assessment Order interaction (which investigated whether inter-
acting with a “rude”/“nice” person would affect the subsequently
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
assessed attitude-behavior consistency) was not significant,
(1) 0.28, p.59.
Assessing the Impact of Imputed Zeroes
These analyses included 57 effect sizes (18% of the total sam-
ple) that were “imputed zeroes” because the original article simply
reported the finding as “not significant” and did not provide
sufficient information to estimate an exact effect size. Estimating
and including effect sizes based off sample means or regression
estimates was not an option because such data were not MCAR/
MAR (i.e., only nonsignificant small effects were “missing”). On
the one hand, estimating an effect as zero is a conservative esti-
mate for a value that could be any positive or negative nonsignif-
icant value. On the other hand, the inclusion of zero may have an
adverse consequence of affecting the amount of between-study
variance because the variance of all of the imputed effects is the
same (Little & Rubin, 1986;Pigott, 2012).
Consistent with the recommendations of Pigott (2012), we con-
ducted a sensitivity analysis to examine whether the (potentially)
under/overestimated variance component affected the findings. We
first estimated the variance component from the random-effects
model without the imputed zeroes. We then generated effect size
estimates for the imputed zeroes by creating a normal distribution
of values with a mean of zero and a standard deviation equal to that
of the random-effects model.
As expected, this technique did not affect the overall mean of
the sample, z.206 (SE .025), z-score 8.15, p.001.
However, including a normal distribution in place of the zeroes
reduced the standard error for the random-effects model by 5%
(from .0265 to .0253).
An inspection of the partitioning of variance across levels as a
function of the two samples revealed that including a normal
distribution of effects produced lower standard errors because it
reduced the variance at Level 3 (from 51% to 34%) and increased
the variance at Level 2 (from 12% to 26%).
We reconducted several main effects to determine whether the
different standard error affected the findings. The change in the
variance did not change the interpretation or significance of any
effects (e.g., cognitive appraisal,
(1) 2.70, p.11; evaluation
(1) 1.18, p.28; attraction measure type,
0.64, p.72).
Excluding Nonsignificant Behaviors
An aim of this article was to identify the specific behaviors
associated with attraction. From a data-driven perspective, the
analyses revealed that only those 11 behaviors reported in Table 3
are those behaviors associated with attraction. If so, all of the
nonsignificant behaviors in the preceding analyses produced
“noise” that potentially reduced the ability to detect the impact of
the various moderators. To address this concern, we replicated all
analyses with only the significant behaviors included.
A summary of those analyses is included in Table 4. A com-
parison of the analyses with and without the nonsignificant behav-
iors revealed two key differences. First, when only the significant
behaviors were included in the analyses, the main effect for cog-
nitive appraisal was significant,
(1) 3.96, p.05, indicating
that the relation between attraction and behavior was stronger
when the person first considered the qualities of the target person
before completing their assessment of attraction. Second, the dif-
ference between direct and indirect behavior was no longer sig-
nificant when nonsignificant behaviors were excluded from the
(1) 1.70, p.19.
Selection Bias
Four methods were used to assess selection bias: funnel plot,
Egger’s regression, Kendall’s tau, and the Vevea and Woods’s
(2005) weight-function method (Coburn & Vevea, 2015). Selec-
tion bias was assessed on (a) the full sample of 309 effect sizes, (b)
the subsample of 267 effect sizes included in the main analyses,
and (c) the samples of direct and indirect behaviors.
We first visually examined a funnel plot of all analyzed effect
sizes to investigate the symmetry of the distribution plot of the
relation between effect size and a measure of precision (Light &
Pillemer, 1984). Figure 1 illustrates the magnitude of the effect
size on the x-axis by the standard error on the y-axis for the full
sample. Inspection of the plot indicates slight asymmetry as a
result of the relative absence of small negative effect sizes with
small sample sizes, suggesting a systematic tendency for effects
from studies with larger standard errors to be positive rather than
negative (a pattern consistent with the “small sample bias”; Sterne,
Egger, & Smith, 2001). This possibility was supported by a sig-
nificant test of funnel asymmetry using Egger’s test (Egger, Davey
Smith, Schneider, & Minder, 1997) and Kendall’s tau.
Next, we examined the Vevea and Woods’s (2005) weight-
function model that estimates the likelihood of observing effect
sizes using a predefined set of pvalue intervals. The model
produces (a) means and variances adjusted for selection bias and
(b) weights that reflect the likelihood of observing an effect in a
given pvalue interval.
In the full sample, 239 of the 309 effect sizes (77%) did not
reach traditional levels of significance, and as presented in Table
5, the estimated weights revealed that nonsignificant pvalues were
more likely to be included than significant pvalues, with pvalues
greater than .50 being 3.36 times more likely to be included than
significant effects.
Inspection of the subsamples produced similar patterns for
asymmetric distributions and weights, with the weights for non-
significant effects routinely exceeding 1. The distribution of ef-
fects suggest that the asymmetry was not an indicator of selection
bias in the sample.
Table 4
Estimates for Moderators of the Relation Between Attraction
and Behavior, Only Significant Behaviors
df p
Attraction measure type 1.70 2 .43
Behavior type 2.04 1 .15
Cognitive appraisal salience 3.96 1 .05
Evaluative threat 1.40 1 .24
Assessment order 1.37 1 .24
Sex composition 5.05 5 .40
Romantic/sexual context 1.23 1 .26
Target person behavior 1.44 3 .69
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Ethnographic Analyses
Our sample was drawn nearly exclusively from Western cul-
tures (see Table 2). However, the subjective experience and de-
velop/restore interdependence approaches indicate that there
should be cross-cultural consistency in the subset of behaviors
related to emotional expression, such that the behaviors associated
with an emotional expression in one culture should align with
behaviors in other cultures (e.g., Ekman & Friesen, 1971). Meta-
analyses provide the opportunity to investigate whether the behav-
iors associated with attraction in Western cultures are similar to
those behaviors related to attraction in other cultures.
However, given the dearth of data from around the world, we
used a descriptive method to investigate qualitatively whether the
behaviors identified here (e.g., smiling, proximity, eye contact, but
not body lean or head tilt) were also those behaviors used to
communicate liking in non-Western cultures. To investigate this,
we evaluated cross-cultural consistency by assessing whether eth-
nographic descriptions of various cultures mentioned these behav-
iors in their account of courtship and marriage. We utilized the
eHRAF (electronic Human Relations Files; Ember, 1997), which
provides detailed descriptions of 306 distinct cultures from around
the world and is considered the “gold standard” for cross-cultural
comparative research (van Holt, Johnson, Carley, Brinkley, &
Diesner, 2013;Young & Benyshek, 2010). Although 306 cultures
are described in the eHRAF, not all of the ethnographies included
a description, or anything more than a trivial mention, of courtship
and marriage behaviors, which resulted in a smaller sample from
which data may be extracted. Eighty-three cultural descriptions
included at least one behavior associated with liking. As is detailed
below, this search provided sufficient data to inform the analyses.
Table 6 presents the cultures for which ethnographers—with
their Western biases regarding what constitutes flirtatious and
amorous behavior—mentioned a specific behavior. Twenty-two
cultural descriptions mentioned “smile” as an indicator of liking.
For the Western Apache of Arizona, for instance, girls were
described as smiling at passing boys to gain their attention (Good-
win & Goodwin, 1942). Parents of daughters of the Lur of south-
western Iran warn them to not “ever smile or laugh” at a boy to
reduce the likelihood of a sexual assault, as such acts are inter-
preted as signs of interest (Friedl, 1997, p. 269; also Joffe, 1963,
who found similar parental warnings regarding smiling among the
Fox Indians of Iowa). Eye contact was specifically mentioned in
54 cultural descriptions. The Igbo of southeastern Nigeria, for
example, do not have a word for “courtship” or “love,” but the
closest approximation is “ifu nanya,” which means to “’look in the
eye’ in a favorable manner” (Basden, 1966, p. 68), and Wagley’s
(1949) ethnographic description of the Mam Maya of Cuchuma-
tanes Highlands of Guatemala mentioned that parents explicitly
warned their daughters to walk through the streets without even
looking at men. Thirty-four cultural descriptions mentioned talking
as an indicator of liking. For the Havasupai of Cataract Canyon
Region in Arizona, if a boy and girl are seen merely talking to one
another, it was “assumed” that they were dating in secret (Smith-
son, 1959, p. 76). Alternatively, we found two or fewer ethno-
graphic descriptions that mentioned body lean, head tilt, or body
orientation as indicators of attraction; and no instance in which
parents explicitly warned their children to not engage in such an
act (e.g., “Young lady, do not tilt your head when talking to the
boy, lest he will know you like him”).
Figure 1. Plot of effect size (in z) by standard error.
Table 5
Selection Bias Analyses
Random-effect estimates
Selection analysis
Egger Kendall Weight-function model
Model k z SE t tau B
Full sample 309 .14 .013 2.41
.24 .028 1.59 1.57 3.36 19.42
Without excluded 267 .17 .013 2.16
.27 .030 1.86 1.71 3.74 18.46
Type of behavior
Indirect 209 .14 .013 1.64 .18
.20 .032 1.52 1.23 2.38 9.21
Direct 58 .23 .028 .20 .13 .24 .059 1.32 .49 1.18 3.24
Note.knumber of effect sizes. Values estimated using these selection bias techniques (which do not consider multilevel modeling) produce estimates
that differ from those reported in the main analyses. Intervals for the weight-function model were: p.05 (weight 1, which is set to 1), .05 p.20
(weight 2), .20 p.50 (weight 3), .50 p1.00 (weight 4), all two-tailed.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Table 6
Cultures in Which Various Behaviors Were Mentioned as Cues to Liking/Attraction
Smiling Eye contact Physical proximity Head tilt Body lean Talking Body/head orientation Laughing
Aymara (SF05) Abkhazians (RI03) Abkhazians (RI03) Aranda (OI18) Hopi (NT09) Abkhazians (RI03) Lur (MA12) Burmans (AP04)
Blackfoot (NF06) Aranda (OI18) Aranda (OI18) Samoans (OU08) Libyan Bedouin
Aranda (OI18) Copper Inuit
Copper Inuit
Assiniboine (NF04) Azande (FO07) Bahia Brazilians
Dogon (FA16)
Fox (NP05) Azande (FO07) Bahia Brazilians
Belau (OR15) Fox (NP05)
Garo (AR05) Bahia Brazilians
Balinese (OF07) Dogon (FA16) Garo (AR05)
Hawaiians (OV05) Blackfoot (NF06) Belau (OR15) Eastern Apache
Gusii (FL08)
Javanese (OE05) Bosnian Muslims
Cajuns (NO12) Fox (NP05) Lau Fijians
Lur (MA12) Burmans (AP04) Central Thai (AO07) Havasupai (NT14) Lur (MA12)
Copper Inuit
Dogon (FA16) Hopi (NT09) Mi’kmaq (NJ05)
Nahua (NU46) Dogon (FA16) Fox (NP05) Innu (NH06) Mundurucu
Navajo (NT13) Fox (NP05) Ganda (FK07) Italian Americans
Ona (SH04)
Ojibwa (NG06) Garo (AR05) Garo (AR05) Kapauku (OJ29) Otavalo Quichua
Otavalo Quichua
Hawaiians (OV05) Gusii (FL08) Kwoma (OJ13) Saraguro Quichua
Highland Scots
Highland Scots
Lau Fijians (OQ06) Turks (MB01)
Puerto Ricans
Hopi (NT09) Hopi (NT09) Mam Maya (NW08) Zulu (FX20)
Saraguro Quichua
Igbo (FF26) Italian Americans
Mapuche (SG04)
Sea Islanders
Imperial Romans
Kapauku (OJ29) Marquesas (OX06)
Semai (AN06) Iran (MA01) Khasi (AR07) Marshallese (OR11)
Tlingit (NA12) Italian Americans
Kurds (MA11) Northeastern Massim
Turks (MB01) Javanese (OE05) Lau Fijians (OQ06) Ojibwa (NG06)
Warao (SS18) Jivaro (SD09) Libyan Bedouin
Okinawans (AC07)
Western Apache
Kapauku (OJ29) Marquesas (OX06) Omaha (NQ21)
Kurds (MA11) Northeastern Massim
Otavalo Quichua
Kwoma (OJ13) Okinawans (AC07) Palestinians (M013)
Lur (MA12) Omaha (NQ21) Puerto Ricans
Mam Maya
Ona (SH04) Rwala Bedouin
Maori (OZ04) Otavalo Quichua
Saami (EP04)
Marquesas (OX06) Palestinians (MO13) Samoans (OU08)
Mataco (SI07) Puerto Ricans
San (FX10)
Mi’kmaq (NJ05) Saami (EP04) Semai (AN06)
Miskito (SA15) Samoans (OU08) Tikopia (OT11)
Mundurucu (SQ13) Santal (AW42) Warao (SS18)
Nahua (NU46) Serbs (EF06) Western Apache
Omaha (NF04) Taiwan Hokkien
Zulu (FX20)
Ona (SH04) Tamil (AW16)
Otavalo Quichua
Tapirapé (SP22)
Puerto Ricans
Tikopia (OT11)
Samoans (OU08) Tinputz (ON26)
Santal (AW42) Tongans (OU09)
Sea Islanders
Trobriands (OL06)
Semai (AN06) Turks (MB01)
Serbs (EF06) Warao (SS18)
Tamil (AW16) Western Apache
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Moreover, several ethnographic descriptions mentioned the ex-
pression of “lewd” versions of the behavior. Specifically, there
were incidents of “indecent” expressions of smiling (e.g., via a
“lascivious” or “creepy” smile), eye contact (e.g., gawk, leer), and
body lean (e.g., leaning back to explicitly expose one’s genitalia).
However, no ethnography described a “lewd” head tilt or an
“inappropriate” body orientation. Except for body lean, there was
consistency in the behaviors identified by the meta-analysis as
associated with attraction and the behaviors that could be ex-
pressed and interpreted as “lewd” or “indecent.”
We investigated the relation between self-reported attraction
and enacted behavior. The meta-analysis revealed several key
findings. First, there was a significant relation between self-
reported attraction and enacted behavior (z.20). Eleven of the
20 behaviors investigated in our main analyses were associated
with self-reported attraction, including eye gaze (frequency and
duration), sitting/standing distance, smiling (frequency, but not
duration), mimicry, laughing, and talking (word count and dura-
tion). Second, we identified the predicted Evaluative Threat
Behavior Type interaction for affective attraction, such that affec-
tive attraction was less strongly related to direct behavior when
evaluative threat was salient. Third, the relation between self-
reported attraction and enacted behavior was significant regardless
of romantic/sexual salience, supporting the contention that attrac-
tion operates to regulate more than romantic relationships. Finally,
the results of the ethnographic analysis were in agreement with the
meta-analytic findings, providing evidence for the cross-cultural
consistency in the behavioral expressions of attraction.
Models of the Relation Between
Attraction and Behavior
We investigated three explanations for the relation between
self-reported attraction and enacted behavior, specifically: (a) be-
havior results from emotional/cognitive processes that covary with
the positive subjective state (subjective experience); (b) the posi-
tive affective state facilitates direct approach behaviors (generate
approach); and (c) behavior is expressed as part of the emotional
experience of attraction (develop/restore interdependence). As is
discussed below, based on the subset of behaviors associated with
attraction and the moderators that affected the magnitude of the
relation, we identified little support for the subjective experience
approach, but support for the generate and develop/restore inter-
dependence approaches, with develop/restore interdependence ex-
plaining the data more completely than generate approach.
Subjective experience. From this perspective, enacted behav-
ior does not result from attraction per se, but from phenomena that
covary with the positive subjective state. The three most likely
candidates are interest, happiness/enjoyment, and sexual desire.
Interest. Work that categorizes interest as an emotion indi-
cates that its function is to facilitate learning and exploration
(Silvia, 2005,2008;Tomkins, 1995). Nonverbal behaviors asso-
ciated with interest in a nonsocial context (e.g., watching a film or
solving a puzzle) include eye contact frequency and duration, head
stillness, directed head orientation, lips apart, and eyes open wide
(Reeve, 1993;Reeve & Nix, 1997; although Mortillaro, Mehu, &
Scherer, 2011 found that interest covaried with eye closure). Other
research in which participants were asked to generate expressions
consistent with interest identified a similar set of behaviors, in-
cluding a forward body lean and a tilted and forwarded head
orientation (Campos, Shiota, Keltner, Gonzaga, & Goetz, 2013;
Dael et al., 2012;de Meijer, 1989), arms stretched out to the front,
illustrative arm movements (Wallbott, 1998), among other behav-
iors (e.g., knee movements, etc.). Importantly, smiling is not
theoretically associated with interest (Silvia, 2008), and research
has found that when interested, the mouth is either open (Mortil-
laro et al., 2011) or closed with lips pressed together (Campos et
al., 2013).
Support for the proposition that interest can account for the
pattern of results is low, as the meta-analyses revealed descrip-
tively low and statistically nonsignificant relations between self-
reported attraction and the nonverbal cues associated with interest
(e.g., forward lean, head tilt). Moreover, we identified a significant
relation for smiles (frequency), an effect specifically mentioned as
unrelated to interest.
Table 6 (continued)
Smiling Eye contact Physical proximity Head tilt Body lean Talking Body/head orientation Laughing
Yahgan (SH06)
Tikopia (OT11) Zapotec (NU44)
Tongans (OU09) Zulu (FX20)
Trobriands (OL06)
Turks (MB01)
Warao (SS18)
Western Apache
Yahgan (SH06)
Zulu (negative,
Note. The alphanumeric value in parentheses represents the culture codes in the eHRAF database. “Negative” indicates that the opposite behavior was
associated with liking (e.g., “leaning back,” rather than a forward lean, was indicative of liking). The database was searched using the following
keywords/stems: smile (smil
, grin
), eye gaze (e.g., eye gaz
, glanc
, look
, gaz
), proximity (get close, near
, run away together, huddle, approach
body/head orientation (facing, face
, direct
), head tilt (bent
, tilt
, cant
), body lean (lean, angl
), and laughing (e.g., laugh
, giggl
, snicker
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Happiness/enjoyment. Following from Darwin (1872/1965)
and Ekman (1999), happiness/enjoyment functions as a positive
emotion that communicates that the interaction should continue
and that the current situation is safe (see also Buss, 2000;Grinde,
2002). Other researchers have gone farther to propose that happi-
ness/enjoyment is the emotion fundamental to the development of
affiliations and alliances (e.g., Beall & Tracy, 2017;van Kleef, De
Dreu, & Manstead, 2004). Happiness/enjoyment has been linked
with various facial responses, including smiling (Ekman, 1992;
Ekman et al., 1990), head bounce, and head tilt (e.g., Campos et
al., 2013;Scherer & Ellgring, 2007). Research into the bodily
movements associated with happiness/enjoyment is less prevalent,
but the research does identify several behaviors, including in-
creased stride length, swinging one’s arms, taking lighter steps
(Montepare, Goldstein, & Clausen, 1987), fast upward movements
with arms raised (de Meijer, 1989), and standing with an upright
body posture (Feng & O’Halloran, 2012). A comparison of these
behaviors with those identified in this meta-analysis indicates little
overlap, however, it is clear that many of these happiness/enjoy-
ment behaviors have not been assessed with respect to self-
reported attraction, so definitive statements regarding whether
happiness/enjoyment was assessed in our sample is not known.
Sexual desire. Some research has labeled the behaviors iden-
tified by this meta-analysis (e.g., smiling, eye contact) as indica-
tors of sexual desire (e.g., Hazan & Zeifman, 1999). From this
perspective, the observed effects resulted from the drive to engage
in sexual activities rather than nonsexual attraction. However,
several findings do not support sexual desire as responsible for the
meta-analytic findings.
First, neither the effects for participant sex, composition of the
interaction by sex, nor salience of romantic/sexual context was
significant, indicating that the effects did not change as a function
of the presence or absence of a presumed sexual motivation. With
respect to participant sex, although men, compared with women,
are more affected by and have less restricted sociosexual orienta-
tion (Simpson & Gangestad, 1992;Simpson, Gangestad, Chris-
tensen, & Leck, 1999), the relation between the various nonverbal
behaviors and attraction was not larger for men. The absence of an
effect for sex composition further fails to support sexual desire as
the active process, such that interactions with other-sexed persons
were not associated with higher correlations than interactions with
same-sexed persons. And most important, the strength of the
association between self-reported attraction and behavior was
comparable between those situations with a presumed sexual mo-
tivation (e.g., during a study on “dating compatibility”) and those
without (e.g., same-sex interaction in the classroom).
Second, a large percent of the meta-analytic data was collected
from studies that involved initial encounters, in which social norms
largely prohibit the behavioral expression of sexual desire (Simon
& Gagnon, 1986). This prohibition would reduce the likelihood
that the observed findings reflect the expression of sexual desire.
In addition, research into the nonverbal behaviors associated with
sexual desire have identified several behaviors, including lip bite,
lip licking, lip touching, and tongue protrusions (Gonzaga et al.,
2006). However, the behaviors in our sample that aligned most
closely with such behaviors and with stereotypical expressions of
sexual desire (e.g., hair flip, breast presentation) were not corre-
lated with attraction (see Appendix B).
Conclusion. Our findings do not provide clear support for the
subjective experience explanation. A strong test of this explanation
necessitates the assessment of these various emotional states, at-
traction, and a set of nonverbal behavior. Although the absence of
available data does not allow us to include happiness/enjoyment,
sexual desire, or interest in the analyses, the pattern of findings
does not provide clear or strong support for this perspective.
Generate approach. This perspective posits that a positive
affective state motivates approach-oriented behavior. In support of
this approach, there were significant effect sizes for the behaviors
consistent with reducing the interpersonal space. Specifically, seat-
ing/standing distance and talking were both associated with self-
reported attraction. From this perspective, people approach not
necessarily to communicate their motives or emotional state, but to
realize expected rewards. Alternatively, this perspective provides a
less clear explanation for the significant effects for the various
affiliative behaviors. For example, mimetic behavior (e.g., of pen
twirling, face rubbing) is defined as a “tactic” to generate rapport
(e.g., Ashton-James et al., 2007;Guéguen, 2009) and laughter is
considered a method for producing and strengthening interpersonal
connectedness (e.g., Owren & Bachorowski, 2003) rather than as
“approach” behaviors, per se.
One theoretical approach submits that such behaviors are con-
sistent with “approach.” Mehrabian (1970,1971) proposed that the
positive affective state may not manifest itself in overt approach
behaviors, but may manifest itself in “abbreviated forms,” includ-
ing eye contact, forward body lean, and directed body orientation.
As noted by Mehrabian (1971) and Mehrabian and Ksionzky
(1970), if a person expects positive reinforcement from a target
person (e.g., via anticipated acceptance or similarity of attitudes),
the person will reciprocate those nonverbal behaviors that they
“expect” from admirers. What is expected from admirers is de-
pendent on culture-specific norms regarding how liking is ex-
Develop/restore degree of interdependence. Consistent with
predictions of functional (e.g., Ekman, 1992) and behavioral ecol-
ogy (Fridlund, 1994) approaches to emotion, the engagement in
affiliative behavior is a component of the emotion of attraction. Of
the three approaches, the meta-analytic results were most consis-
tent with this perspective. Three findings provide the foundation
for this conclusion.
First, the meta-analysis revealed that the behaviors assumed to
be associated with communicating trust were related to self-
reported attraction, including smiling (frequency), eye contact,
sitting/standing distance, and mimicry. From this perspective,
when one is attracted to another person, one will smile frequently,
make repeated eye contact, and mimic to generate trust from the
other person. Second, the link between self-reported attraction and
affiliative behavior was evident across various types of relation-
ships, such that a significant relation between attraction and be-
havior was evident in both high and low evaluative threat condi-
tions, romantic/sexual contexts, and sex compositions, which
indicates that attraction—and the role it plays to regulate interde-
pendent interactions—is consistent across different types of rela-
tionships and contexts. Third, the alignment between the behaviors
identified by the meta-analysis and those behaviors identified by
ethnographers is consistent with the develop/restore approach, in
which the behavioral expression of attraction would be predicted
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
to be consistent across cultures (Mesquita & Frijda, 1992;Tracy &
Matsumoto, 2008).
How the develop/restore degree of interdependence process
works. People’s social lives include interactions that vary in
intensity from the trivial (the local bank teller, toll bridge opera-
tor), to the casual (acquaintances), to the intense and deep (roman-
tic partners, family members, longtime friends). From a function-
alistic approach to emotions, even the behavior in the most trivial
interpersonal interactions—a smile before approaching the waiting
bank teller, the establishment of eye contact across a crowded
discotheque, or an instructor’s smile to students on entering the
classroom—can be considered critical to the interaction. In this
context, the expression of affiliative behavior signals one’s inten-
tion to conditionally cooperate and to uphold one’s side during an
exchange (Boone & Buck, 2003;Montoya & Horton, 2014). A
smile with eye contact is expressed instrumentally to communicate
one’s willingness to satisfy one’s side of the social contract. In the
crowded bar, a smile with eye contact communicates one’s will-
ingness to deepen the (at the time nonexistent) exchange to in-
clude, say, a conversation while consuming a drink. Similarly, the
professor’s eye contact and smile at a student who knocks on his
or her door during office hours communicates willingness to
interact during the office visit. Although emotions are commonly
conceived as one’s “immediate” response to a stimulus (between 1
sand4sinduration, Ekman, 1984), people’s description of their
own emotional experiences indicates that emotions commonly last
for minutes, hours, or days (Frijda, Mesquita, Sonnemans, & van
Goozen, 1991;Scherer, Wallbott, & Summerfield, 1986;Wallbott
& Scherer, 1986), suggesting that attraction could facilitate the
sustained expression of affiliative behavior.
Emotions not only affect the person’s own behavioral response,
but they operate to convey information to the target person about
the person’s intentions, which then elicits change in the target
person’s behavior (e.g., Ames & Johar, 2009;Fischer et al., 2003;
Frijda, 1986;Hareli & Parkinson, 2008;Knutson, 1996;Owren &
Bachorowski, 2003;van Kleef et al., 2004). In the case of attrac-
tion, affiliative behavior operates proximally to augment the target
person’s trust, and ultimately to produce a behavioral response in
the target person. In initial interactions, the behavioral response
commonly takes the form of compliance, cooperation, or recipro-
cated behavior. Indeed, compliance requests paired with behav-
ioral cues to liking are more successful (Burger, Soroka, Gonzago,
Murphy, & Somervell, 2001;Cialdini, 2001;Daniels & Berkowitz,
1963;Goei, Lindsey, Boster, Skalski, & Bowman, 2003;Mehra-
bian & Ksionzky, 1970), and the presence of affiliation cues
produces more intergroup cooperation among male participants
(Kurzban, 2001) and they increase the success rate of bank loan
requests (Wexley, Fugita, & Malone, 1975).
Does this process work in committed relationships? When
the level of interdependence is deep and trust has been established,
the relation between attraction and affiliative behavior should be
attenuated. In such cases, the presence of high levels of trust
between, say, marital partners or old friends, reduces the relation
between attraction and affiliative behavior because there is no
“need” to communicate trust.
Our search of the literature for the presented meta-analysis also
uncovered research that reported the relation between enacted
behavior and a measure of relationship satisfaction between per-
sons in a preexisting relationship (e.g., married couples, friends).
Effect sizes that employed an index of relationship satisfaction
were not included in the larger analyses because such effects
considered the participant’s satisfaction with their relationship, not
their affective/behavioral evaluation of their relationship part-
ner. However, such research may provide initial insight into the
operation of attraction-related processes when interdependence
and trust is assumed to be high. We identified 27 effect sizes for
the relation between relationship satisfaction and enacted be-
havior. The relation was not significantly different from zero,
z.05, SE 0.03, z-score 1.68, p.09. A more complete
description of this analysis can be found in the online supple-
ment. This nonsignificant effect may be interpreted as consis-
tent with the develop/restore interdependence approach, such
that when trust is (assumed to be) high, there is no “need” to
communicate trust.
However, the “restore” function of attraction indicates that the
relation between self-reported attraction and affiliative behavior
should be higher when there is a “need” to express it, such as when
there is a threat to the relationship. In other words, when trust has
been called into question (by a “fight,” a long time apart, or after
one person is suspected of cheating), affiliative behavior would be
expressed to help restore it. Unfortunately, meta-analytic data were
not available to test this prediction, however, laboratory evidence
is consistent with this prediction. For instance, after experiencing
rejection from a minimal group, participants engaged in more
mimicry with ingroup members (Lakin, Chartrand, & Arkin,
2008). Relationship partners who have emotionally hurt their part-
ners increased eye contact and hugging to express their honest
interest in restoring the relationship (Waldron & Kelley, 2005). In
addition, models of mother–infant interactions submit that infants
employ positive affective displays (e.g., smiling, eye contact,
cooing) to “repair” an unresponsive (still-faced) or an affectively
“mismatched caregiver” (Brazelton, 1974;Campos, Barrett, Lamb,
Goldsmith, & Sternberg, 1983;Gianino & Tronick, 1988;Massie,
Conclusion. The findings are largely consistent with the de-
velop/restore interdependence perspective and it explains the data
more completely than the other two approaches. From this per-
spective, affiliative behaviors serve to develop and/or restore the
degree of interdependence via the expression of trust-producing
behavior. What sets apart the develop/maintain interdependence
perspective from the subjective experience perspective is the pre-
diction that people engage in affiliative behavior strategically to
increase the trust. For instance, infants instrumentally smile and
coo at their caregivers to elicit nurturing behavior from them
(Bowlby, 1980;Gewirtz & Boyd, 1976,1977), and in the context
of a romantic relationship, people strategically say “I love you” or
initiate interpersonal touch to communicate their level of commit-
ment and to restore a high level of mutual trust (e.g., Boone &
Buck, 2003;Sabatelli & Rubin, 1986). Moreover, these results
emphasize that the definition of attraction should focus on “devel-
oping and/or restoring the degree of interdependence” rather than
focusing on approach to acquire benefits/rewards.
Why Was the Overall Correlation Not Higher?
We found a small-to-moderately sized relation between self-
reported attraction and the various enacted behaviors. Compared
with meta-analyses that investigate the relation between psycho-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
logical states and enacted behavior, the effect size uncovered here
may have been expected. On the one hand, our effect size was
approximately twice the small-to-trivial size of the relation be-
tween deception and nonverbal behaviors (e.g., eye contact, smil-
ing, head nodding, hand movements; DePaulo et al., 2003;Sporer
& Schwandt, 2007;Zuckerman, Driver, & Guadagno, 1985). On
the other hand, our effect size was approximately half the relation
between attitudes and behavior in the noninterpersonal context
(e.g., r.52, Glasman & Albarracín, 2006;r.38, Kraus, 1995).
But why was the relation not higher? We discuss three psycho-
logical considerations that fall from concerns inherent to interde-
pendent relationships.
First, from the develop/restore interdependence perspective, the
correlation between self-reported attraction and behavior should
not be particularly strong in initial interactions. Specifically, the
nonverbal behaviors associated with attraction connote the per-
son’s conditional cooperativeness (Boone & Buck, 2003). How-
ever, natural selection works against people whose behaviors in a
social exchange reflect their internal motivations (Dawkins, 1976;
Kurzban & Leary, 2001;Trivers, 1971). In this way, expressing
one’s cooperative intentions is risky because it leaves one open to
exploitation (Frank, 1988). When exploitation is possible, the
expression of affiliative behavior should be ambiguous to (a)
optimize the terms of the exchange and (b) collect more informa-
tion about the other person’s motives (Grammer et al., 1999). In
the dating context, concerns regarding exploitation is most evident
when daters with a long-term mating orientation interact with
daters with a short-term mating orientation.
It then follows that when concerns regarding exploitation are
eliminated/negated, attraction may be expressed more freely. In
such instances (e.g., in interactions with infants, family pets,
and/or the elderly), exaggerated forms of affiliative behavior may
result (e.g., infantilization, baby talk with the elderly; Bombar &