Abstract
Inclusion of other in the self, a key principle of the self-expansion model, suggests that close others overlap with the self in terms of resources, perspectives, and identities. Research from behavioral, cognitive, and neural domains provides evidence for inclusion of other in the self; the present research extends prior theoretical and empirical work to a new, visual domain by investigating whether inclusion of other in the self applies to facial processing. In two reaction time (RT) experiments, participants viewed static (Study 1) and morphed (Study 2) facial images of themselves, their close friend (i.e., a close other), and a familiar celebrity (i.e., a non-close other). In Study 1, participants showed slower RTs when comparing their own image with their friend’s image than when comparing their own image with a celebrity’s image. In Study 2, participants showed slower RTs when their own image was morphed with their friend’s image than when their own image was morphed with the celebrity’s image. These results suggest that inclusion of close others in the self extends to visual processing. Implications and limitations are discussed.
Keywords
When determining where to direct our attention and cognition, the self appears to receive preferential treatment. Since it was first conceptualized by Rogers, Kuiper, and Kirker (1977), the self-reference effect, in which self-related information is encoded and processed preferentially, has been well documented. The self-reference effect appears to extend to those to whom we feel close; for example, referencing both self and close others appears to enhance memory (Serbun, Shih, & Gutchess, 2011). This extension of the self-reference effect to close others may stem from a shared representation of self and close others. Indeed, research from cognitive, behavioral, and neural domains supports the idea of a shared representation—often referred to as an overlap—between self and close others (Aron, Aron, Tudor, & Nelson, 1991; Gardner, Gabriel, & Hochschild, 2002; Smith & Henry, 1996). For instance, recent research supports the idea that closeness involves a blurred boundary when distinguishing between self and close others at the bodily or embodied domain (Quintard, Jouffre, Croizet, & Bouquet, 2018). However, far less research has investigated how self and close others may overlap in the visual domain.
One conceptualization of closeness, inclusion of other in the self, suggests that the self overlaps with close others in terms of resources, perspectives, and identities (Aron et al., 1991). Evidence supporting the principle of inclusion of other in the self indicates that the self–other distinction is reduced for close others (Symons & Johnson, 1997). For instance, individuals show slower reaction times (RTs) when there is the potential for confusion between self and close others, such as when making “me/not me” decisions for traits that differ between themselves and their spouse, which indicates overlap between self and close others (Aron et al., 1991; Zickfeld & Shubert, 2017). Such overlap has been demonstrated within a variety of close relationships, including between spouses (Aron et al., 1991, Zickfeld & Shubert, 2017), romantic partners, and best friends (Mashek, Aron, & Boncimino, 2003), as well as across a variety of domains, including attributional biases, allocation tasks, and neural responses. Research (Kang, Hirsh, & Chasteen, 2010) in the latter domain, for example, found stronger brain activation in response to mistakes made by friends versus strangers; this effect was mediated by the extent to which participants included the other in the self.
The principle of inclusion of other in the self has had wide implications across psychological and relationships research (see Aron, Lewandowski, Mashek, & Aron, 2013, for a review). As one example, close relationships research has indicated that greater self–other overlap is associated with both individual and relational benefits. For instance, higher levels of inclusion of other in the self have been linked to increased satisfaction and decreased dissolution in close relationships (Aron, Aron, & Norman, 2002; Le, Dove, Agnew, Korn, & Mutso, 2010). As another example, intergroup relations research has suggested that including others in the self may impact the way we treat others more broadly. For instance, including a member of a racial outgroup in the self is associated with more positive intergroup attitudes (e.g., Wright, Brody, & Aron, 2005). Together, these findings suggest that a central component of social bonds lies in the shared representation of self and close others.
A key question related to the shared representation of self and close others is whether the self–other overlap in close relationships extends to facial perception. Facial perception is important for social judgments, and the ability to quickly recognize familiar faces is integral to social functioning. Several studies have found a self-face advantage when processing self versus other faces, typically compared to strangers or familiar faces (i.e., celebrities; Platek & Kemp, 2009; Tong & Nakayama, 1999), but none to our knowledge have examined the potential perceptual overlap of self and close other faces. Exploring self–other overlap in the visual domain contributes to knowledge of face processing in general and could further demonstrate that inclusion of other in the self goes beyond being a subjective cognitive process. The present research also helps advance understanding of the hierarchy of face processing. Research on face processing indicates that faces may be processed according to a hierarchy, with personally familiar faces being processed preferentially (Carbon, 2008).
The present research extends theoretical and empirical work on inclusion of other in the self (see Aron et al., 2013, for a review) by investigating potential self–other overlap in facial perception. Across two RT experiments, participants viewed their own faces compared to those of familiar close others (friends) and familiar non-close others (celebrities) using static facial images (Study 1) and morphed facial images (Study 2). In Study 1, we predicted that participants would show slower RTs when comparing their own image with their friend’s image than when comparing their own image with a celebrity’s image. In Study 2, we predicted that participants would show slower RTs when their own image was morphed with their friend’s image than when their own image was morphed with the celebrity’s image.
General method
Participants
Female undergraduates were recruited from the Psychology subject pool at a Northeastern U.S. university and received course credit or payment for participation. Men were excluded from the study for convenience and to reduce potential variance, given possible differences in closeness between men’s and women’s friendships (e.g., Elkins & Peterson, 1993). Participants brought a close gender- and age-matched friend to the experiment. Potential participants were excluded if they were left-handed (Edinburgh Handedness Inventory; Oldfield, 1971) or if they or their friend had facial markings that would make them easily identifiable.
Facial Stimuli
Facial photographs of participants and their friend were taken at the lab and edited to remove non-facial features. All images were presented in mirror- and mirror-reversed format to limit orientation familiarity (Troje & Kersten, 1999). Participants were asked to make a neutral expression during the photograph. Participants also viewed a gender- and race-matched celebrity face (e.g., Reese Witherspoon).
Measures
We measured participants’ familiarity with their friend and the celebrity, as well as facial similarity among participants, their friend, and the celebrity, to account for potential alternative explanations that observed effects might be due to friends being more familiar or more similar to participants than celebrities.
Familiarity
Participants rated familiarity with their friend and with the celebrity on a 7-point scale from 1 (not at all) to 7 (extremely). A composite (friend score minus celebrity score) was calculated, with positive values indicating greater familiarity with friend than celebrity (M = 0.74, SD = 1.42 in Study 1; M = 1.46, SD = 1.76 in Study 2).
Similarity
Two independent blind coders rated facial similarity for self/friend, self/celebrity, and friend/celebrity on a 7-point scale from 1 (not at all) to 7 (extremely). Coders’ ratings (MICC = .57 in Study 1; MICC = .69 in Study 2) were averaged for each pair. A composite (self/friend score minus mean of self/celebrity and friend/celebrity scores) was calculated, with positive values indicating greater self/friend similarity, beyond self/celebrity similarity or friend/celebrity similarity (M = 0.77, SD = 0.95 in Study 1; M = 1.19, SD = 1.52 in Study 2).
Study 1
Method
Twenty-four female participants (Mage = 19.88, SDage = 2.52; 58% Caucasian, 42% Latina) completed familiarity measures, followed by a 17-min computerized facial identification task. Participants were told that they would view a series of facial images and would be asked to indicate by pressing a button whether the image viewed was their own face, their friend’s face, or a celebrity’s face. Specifically, participants performed a version of the me/not-me task (Aron et al., 1991), in which they were instructed at the start of each of the 24 blocks to identify self (“Is it you?) or other (either “Is it your friend?”, when the set of images included the friend, or “Is it a celebrity?”, when the set of images included the celebrity) via a button press. 1 Participants then viewed a series of randomly presented static facial images (288 total, across 24 blocks) that included either their own face and their friend’s face (the self–friend image set) or their own face and the celebrity’s face (the self–celebrity image set); each image was presented for 500 ms. Fixation points were shown in the center of the screen for 1,000 ms before each facial image and for 1,500 ms after each facial image. Response recording began when the facial image appeared. Trials were counterbalanced with respect to the image viewed (self/other), the image set (self–friend/self–celebrity), and the button press hand (right/left).
Results
Reaction times
RTs below 200 ms are considered anticipatory (Rousselet, Macé, & Fabre-Thorpe, 2003) and were omitted from analyses across both studies. Detecting faces may take a minimum of 250–290 ms (Fabre-Thorpe, 2011). Remaining RTs in both studies were approximately normal, so no data transformations were performed.
Analytic strategy
We tested multilevel models (Raudenbush & Bryk, 2002) using the linear mixed-effects models (MIXED) procedure in SPSS 22.0. Individual responses were modeled as a function of image viewed and image set (level 1), which were nested within participants (level 2); intercepts were modeled as randomly varying to account for the nested data structure (Kenny, Kashy, & Cook, 2006). Continuous variables were grand-mean centered; categorical variables were dummy-coded (Aiken & West, 1991) by coding image viewed as 0 for self and 1 for other and by coding image set as 0 for self–celebrity image set and 1 for self–friend image set.
Do image viewed and image set shape participants’ RTs?
We predicted that participants would show slower RTs when viewing their own image or their friend’s image presented randomly in a set of images that included self and friend (the self–friend image set) than when viewing their own image or a celebrity’s image presented randomly in a set of images that included self and celebrity (the self–celebrity image set). Figure 1 shows participants’ mean RTs, displayed by image viewed (self, other) and image set (self–friend, self–celebrity). The interaction between image viewed and image set significantly predicted participants’ mean RTs (see Table 1). As predicted, participants showed significantly slower RTs when viewing their own image interspersed with their friend’s image (i.e., when viewing their own image as part of the self–friend image set; M = 670.12, SE = 16.49) than when viewing their own image interspersed with the celebrity’s image (i.e., when viewing their own image as part of the self–celebrity image set; M = 637.73, SE = 15.94), b = 50.96, SE = 20.39, t(357) = 2.50, p = .013, 95% confidence interval (CI) [10.85, 91.07]. Not supporting predictions, however, participants’ RTs did not differ when viewing their friend’s image interspersed with their own image (i.e., when viewing their friend’s image as part of the self–friend image set; M = 668.22, SE = 18.18) compared to when viewing the celebrity’s image interspersed with their own image (i.e., when viewing the celebrity’s image as part of the self–celebrity image set; M = 672.90, SE = 17.67), b = 13.84, SE = 9.12, t(357) = 1.52, p = .13, 95% CI [−4.09, 31.76].

Participants’ mean RTs in Study 1, displayed by image viewed (self/other) and image set (self–friend/self–celebrity). Standard errors are represented by error bars on each column. As shown in the figure, when the image viewed was other and the image set was self–friend, participants viewed an image of their friend; when the image viewed was other and the image set was self–celebrity, participants viewed an image of the celebrity. RT: reaction time.
Linear mixed-effects model predicting participants’ mean RTs from image viewed and image set.
Note. RT = reaction time; SE = standard error; CI = confidence interval.
The interaction between image viewed and image set remained significant when controlling for the main effect of participants’ familiarity with their friend, as well as when controlling for interactions among image viewed, image set, and familiarity (see Online Supplemental Material, Tables 1.1.1–1.1.2). The interaction between image viewed and image set became marginally significant (p = .052) when controlling for the main effect of participants’ similarity between themselves and their friend, as well as when controlling for the interaction among image viewed, image set, and similarity (see Online Supplemental Material, Tables 1.2.1–1.2.2). Therefore, the key interaction between image viewed and image set did not appear to be driven by participants’ self-reported familiarity with their friend or observer-rated similarity between themselves and their friend. 2
Discussion
Study 1 provides partial support for our predictions. Consistent with our predictions, participants showed slower RTs when viewing their own image within a set of images of themselves and their friend compared to when viewing their own image within a set of images of themselves and a familiar celebrity. These findings suggest a potential self–close other overlap in facial perception. Not supporting our predictions, however, participants did not show slower RTs when viewing their friend’s image within the set of images of themselves and their friend compared to when viewing the celebrity’s image within the set of images of themselves and the celebrity. This pattern suggests that the statistical interaction between image viewed and image set was driven by participants’ faster RTs when viewing their own image within the set of images of themselves and the celebrity.
Therefore, we conducted a second study to better understand this pattern of results. We designed Study 2 to provide a conceptual replication of Study 1, as well as to add methodological rigor by using a different, but related, set of facial stimuli (morphed facial images rather than static facial images). Specifically, in Study 2, participants viewed morphed facial images of self and friend photographs, self and celebrity photographs, and friend and celebrity photographs. We predicted that participants would show slower RTs when their own image was morphed with their friend’s image compared to when their own image was morphed with the celebrity’s image, as well as compared to when their friend’s image was morphed with the celebrity’s image.
Study 2
Method
Twenty-five females (Mage = 20.63, SDage = 4.01; 44% Caucasian, 24% African–American, 24% Asian–American, 1% Latina) participated. The procedure followed that of Study 1, with the following modifications to the facial stimuli and the computerized task, which lasted 35 min. Participants viewed morphed facial images rather than static facial images. Morphed facial images of self and friend photographs, self and celebrity photographs, and friend and celebrity photographs were created using Morpheus Photo Morpher (Morpheus Software, 1999). For each type of morphed image (self–friend, self–celebrity, and friend–celebrity), facial landmarks were used to overlap the images in 5% increments to yield 21 morph levels (see Figure 2). Within each block, morphed images were presented randomly and varied by orientation (mirror, mirror-reversed) and morph level (21 levels). A total of 1,512 morphed images were viewed. Following the instructions (e.g., “Is it you?”), participants viewed a fixation point for 250 ms, the facial image for 500 ms, and a second fixation point for 250 ms.

Example set of morphed images used in Study 2. Images vary in 5% increments from an example participant to an example celebrity. (The example participant is a graduate student who gave permission for her photograph to be used in this figure. Images of actual study participants are not shown to protect confidentiality.) The morphed image in the upper-left corner is 100% participant and 0% celebrity, the morphed image in the center is 50% participant and 50% celebrity, and the morphed image in the lower-right corner is 0% participant and 100% celebrity.
Results
Analytic strategy
The analytic strategy paralleled that of Study 1, with the following modifications. Individual responses were modeled as a function of morph level (level 1), which was nested within morphed image type (level 2), which was nested within participants (level 3); intercepts and linear effects of morph level were modeled as randomly varying to account for the nested data structure (Kenny et al., 2006). The three-level categorical variable of morphed image type was dummy-coded by creating two variables—one for self–celebrity morphed images and one for friend–celebrity morphed images—and making self–friend morphed images the reference group (represented when self–celebrity morphed images and friend–celebrity morphed images both had values of 0) against which both other morphed images were compared.
Do morphed image type and morph level shape participants’ RTs?
We predicted that participants would show slower RTs when viewing their own image morphed with their friend’s image compared to when viewing their own image morphed with the celebrity’s image, as well as compared to when viewing their friend’s image morphed with the celebrity’s image. Figure 3 shows participants’ mean RTs displayed by morphed image type (self–friend morph, self–celebrity morph, friend–celebrity morph). Morphed image type significantly predicted participants’ mean RTs (see Table 2). As expected, participants showed significantly slower RTs for self–friend morphed images (M = 481.64, SE = 9.19) than self–celebrity (M = 470.24, SE = 5.42) or friend–celebrity (M = 464.98, SE = 5.42) morphed images. We also explored whether morph level would shape RTs. The quadratic effect of morph level marginally predicted participants’ mean RTs (see Table 2); participants tended to show slower RTs when images were most morphed together than when images were least morphed together (i.e., when each image appeared in its original, unmorphed form).

Participants’ mean RTs in Study 2, displayed by morphed image type (self–friend morph, self–celebrity morph, friend–celebrity morph). Standard errors are represented by error bars on each column. RT: reaction time.
Linear mixed-effects model predicting participants’ mean RTs from morphed image type and morph level.
Note. RT = reaction time; SE = standard error; CI = confidence interval.
Main effects of morphed image type remained significant (or became marginally significant) and the quadratic effect of morph level remained marginally significant (or became significant) when controlling for main effects of participants’ familiarity with their friend and similarity between themselves and their friend, as well as when controlling for interactions among morphed image type and familiarity and among morphed image type and similarity (see Online Supplemental Material, Tables 2.1.1–2.2.2). Therefore, the key main effects of morphed image type did not appear to be driven by participants’ self-reported familiarity with their friend or observer-rated similarity between themselves and their friend. 3
Discussion
Study 2 conceptually replicates key findings from Study 1 using a different type of facial stimuli, in the form of morphed facial images rather than static facial images. Consistent with our predictions and with a potential self–close other overlap in facial perception, participants showed slower RTs when their own image was morphed with their friend’s image compared to when their own image was morphed with the celebrity’s image, as well as compared to when their friend’s image was morphed with the celebrity’s image.
General discussion
A variety of cognitive and behavioral studies (see Aron et al., 2013, for a review) have found that close-other-processing overlaps with self-processing. However, no previous studies have considered the possibility that such self–other overlap may extend to facial processing. The present research supports the idea that close others are processed preferentially and may overlap with the self. Across two experiments using different methods (static versus morphed facial images), participants were slower to distinguish images of themselves from close others (friends) than from non-close others (celebrities). These effects were not due to friends being more similar or more familiar than celebrities.
These findings advance theoretical understanding of the self–other overlap by providing further evidence that self-reference effects may be diminished if a close other is incorporated into the self-concept (e.g., Bentley, Greenaway, & Haslam, 2017), as well as by showing that inclusion of other in the self may extend to the visual domain. The present research also helps advance understanding of the hierarchy of face processing. These studies suggest that faces of close others are processed preferentially by demonstrating that we have more difficulty distinguishing the self from close others than from non-close others, which supports the idea that personally familiar faces may be processed higher in the hierarchy than famous familiar faces (Carbon, 2008). A self–other overlap in facial processing may serve an adaptive function as we rapidly process the faces of close others. This may allow us to be attuned to the facial expressions of close others, directing our attention to those whose emotions and actions may be of great importance. Alternatively, such overlap may free up cognitive resources for more pressing concerns in the environment, such as threat detection.
The present findings extend our understanding of how close relationships affect the self-concept. Our findings are consistent with cognitive interdependence theory, which posits that the self is characterized by a collective representation of the self in a close relationship (Agnew, Rusbult, Van Lange, & Langston, 1998). Further, results from the present research indicate that perceptual judgments of others may be affected by self–other overlap. This raises potential implications for how closeness shapes our perceptions of others in visual and cognitive domains and may extend beyond these domains with behavioral or applied effects. For example, close relationships research has linked greater self–other overlap with both individual and relational benefits, including increased satisfaction and decreased dissolution in close relationships (Aron et al., 2002; Le et al., 2010); showing greater self–other overlap in the visual domain might be one specific mechanism through which people experience these benefits.
Caveats and Conclusions
We acknowledge that the present studies are limited in their generalizability by examining specific samples—North American college women—and specific relationships—close friendships. Although both samples were relatively diverse with respect to race/ethnicity, future studies should include more diverse samples with respect to age, gender, and geographical region. For example, future studies should investigate whether men would show the same effects in processing the faces of close versus non-close others, which would extend the current studies and build on previous studies by Tong and Nakayama (1999), who used all-male samples in their investigation of self versus personally familiar faces. Future studies also should examine diverse types of close relationships, such as kin relationships (e.g., siblings, following Platek and Kemp, 2009) or romantic relationships (e.g., spouses, following Aron et al., 1991).
We also acknowledge that inter-rater agreement for similarity was good in Study 2 (MICC = .69) but only fair in Study 1 (MICC = .57), according to Cicchetti’s (1994) guidelines, which might have limited the ability to fully account for potential effects of similarity in Study 1. Furthermore, our method of accounting for similarity and familiarity represents just one approach to controlling for potential confounds; future research might, for example, test friends matched for similarity and familiarity but differing in levels of closeness, or experimentally generate closeness between randomly paired strangers. Both approaches would allow researchers to explore the moderating role of closeness by, for instance, examining whether the observed effects are stronger for people in closer friendships and/or weaker for people who feel closer to the target celebrity.
Furthermore, both studies had smaller sample sizes, so it will be important to replicate their results. Although the samples were small, they were not underpowered. In both studies, a large number of data points resulted from the within-subjects design and multiple trials (288 images in Study 1 and 1,512 images in Study 2). It is also worth noting that the sample sizes in the current studies are notably larger than or comparable to past studies. For example, similar research on face processing of self and personally familiar faces include Tong and Nakayama’s (1999) set of studies, which had 16 or fewer participants per study (i.e., 8 participants in Study 1, 16 participants in Study 2, and 10 participants in Study 3), and Rooney, Keyes, and Brady’s (2012) study, which had 24 participants (i.e., 12 pairs of friends).
Despite these potential limitations, the present research employed strong experimental designs—replicated across two independent studies with different methods—and accounted for possible confounds of familiarity and similarity. As such, these studies deepen our knowledge of the underlying dynamics of a central feature of human social life and point to new methodological and theoretical directions for future research in this area.
Supplemental material
Supplementary_material - Seeing you in me: Preliminary evidence for perceptual overlap between self and close others
Supplementary_material for Seeing you in me: Preliminary evidence for perceptual overlap between self and close others by Sarah Ketay, Lindsey A. Beck, Suzanne Riela, Cristen Bailey and Arthur Aron in Journal of Social and Personal Relationships
Footnotes
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Open research statement
This research was not pre-registered. The data and materials used in the research are available upon request by emailing
Supplemental material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
