Every morning, most people look in the mirror to check their appearance. Every time we look in the mirror, we know we are looking at ourselves. The self-face has a special meaning to humans due to its uniqueness and its importance for our identity and our sense of self. Looking at ourselves in the mirror gives us access to our own image, as well as to proprioceptive, tactile, and motor sensory cues, which enables individuals to update their mental representation of their own face (Tsakiris, 2008). Contrary to other self-related information, our own face is a unique stimulus, since we do not share it with other people (Devue & Bredart, 2008). Therefore, it has an important value for our own sense of identity and also constitutes one of the main components of our official identity, as we can observe through identity cards, passports, and driving licenses.

The interest in studying self-face processing emerged mainly from the studies conducted by Gallup and colleagues in chimpanzees. In a series of studies using the mark test, Gallup (1977) suggested that chimpanzees were able to recognize their own image. Almost concomitantly with Gallup’s studies, Amsterdam also used the mark test (or mirror self-recognition task) to investigate children’s responses to a mirror (Amsterdam, 1972). Indeed, the ability to recognize oneself is not an innate ability but develops gradually (Courage, Edisona, & Howeb, 2004), probably in a series of stages (Rochat, 2003). It is suggested that humans are able to recognize their own image in a mirror around the second year of life. Nevertheless, it is only later that infants develop the ability to recognize themselves in pictures and videos (Rochat, 2003). Studies have also suggested that being able to recognize oneself implies a symbolic representation of the self (the ability to perceive oneself as someone who has an outside appearance) and, consequently, the representation of others, allowing the self-other distinction (Rochat, Broesch, & Jayne, 2012).

Although the mirror self-recognition task is frequently mentioned as the “gold standard” for measuring visual self-recognition, more recent studies have mainly used images displayed on computer screens. When researchers explore self-face recognition, one of their main goals is to better understand the self as a whole and thus determine which aspects make the self a unique stimulus. On a more theoretical level, the question of whether the self-face can be considered as a unique stimulus with specific properties is still a matter of debate. As several lines of research have shown, self-relevant stimuli are processed faster and more accurately than other types of stimuli. For instance, self-face processing has been suggested to differ from other faces (both familiar and unfamiliar) in terms of behavioral, neural and neurophysiological responses (e.g., Devue & Bredart, 2011), although not all researchers seem to agree (Gillihan & Farah, 2005). For instance, Gillihan and Farah (2005) found no consistent evidence supporting a physical or psychological self as being unique. Regarding the self-face, they stated that evidence is mixed and yielded inconsistent results regarding anatomical localization for self-face recognition.

Previous reviews on this topic have mainly focused on the neuroanatomical aspects of the self (Devue & Bredart, 2011; Platek, Wathne, Tierney, & Thomson, 2008). Nevertheless, another criterion that can be used to determine whether a stimulus is unique is to look at its functional uniqueness, that is, whether the stimulus is processed in a unique way. For this, it would be necessary to consider the behavioral studies performed in the domain. Surprisingly, at behavioral level, no study has yet tried to summarize and statistically test whether self-face is really a unique stimulus associated with specific mechanisms. One of the criteria often employed to determine whether self-face processing is unique (or has a unique meaning for individuals) or not relates to reaction time, that is, whether or not self-face is prioritized and processed faster in comparison with other faces. Although several studies argue that the self-face is a unique stimulus and that individuals respond faster to their own face, current evidence reveals contradictory results. Crucially, several factors might be associated with these contradictory results, notably, methodological and sample related factors. In an attempt to explain heterogeneity and understand the potential effect of experimental and individual factors on self-face recognition, we considered different moderators.

Moderators

First, we considered the fact that studies included different stimuli. Self-face processing has been compared with the processing of unfamiliar (Hughes & Nicholson, 2010), famous (Tacikowski & Nowicka, 2010), and familiar faces such as friends (Sugiura et al., 2006; Sugiura et al., 2005), family members (Martini, Bufalari, Stazi, & Aglioti, 2015), colleagues (Liew, Ma, Han, & Aziz-Zadeh, 2011), or a partner (Kircher et al., 2000). There is consistent evidence suggesting different mechanisms involved in processing familiar and unfamiliar faces (Burton, Jenkins, & Schweinberger, 2011; Natu & O’Toole, 2011). The differences between familiar and unfamiliar faces include (1) the type of information extracted from the faces (identity-specific semantic codes for familiar faces only vs. visually derived semantics for both unfamiliar and familiar faces; Bruce & Young, 1986); (2) the amount of exposure we have to one particular face, which is associated with the robustness of familiar face representations and with the fact that familiar faces are more tolerant to changes in pose or expressions compared to unknown faces (Burton et al., 2011; Jenkins & Burton, 2011); and (3) the emotional aspects that are associated with the recognition of a familiar face (Gobbini & Haxby, 2007; Gobbini, Leibenluft, Santiago, & Haxby, 2004). In view of the arguments described above, three initial moderators were considered. First, the level of familiarity, that is, whether the face was familiar or not to the participants (familiarity). Second, the identity of the face, whether it belonged to someone she or he knows personally, whether it was a famous person or a stranger (identity). Third, we also considered whether the face was presented in “profile” view or “frontal” view (head angle), since a better mental representation of familiar faces may prompt individuals to recognize them faster than unfamiliar faces, especially in profile view (e.g., Tong & Nakayama, 1999). We also considered the interaction between familiarity and head angle.

The task employed and, more precisely, the cognitive function on which that task relies may also have an impact on self-face processing. Some tasks request participants to identify and/or recognize some stimulus properties such as identity or head orientation, while others request participants to detect or search for a specific stimulus property. Thus, while the first may rely more on memory or perceptual functions, the later depends mostly on attention resources. Therefore, tasks may be classified according to whether they request participants to rely more on pictorial/perceptual information (e.g., head angle or static pose) or semantic information (e.g., face identity or level of familiarity). Although the extraction of semantic information also takes place during attentional tasks or perceptual tasks, since we automatically attach meaning to what we are seeing, the task itself does not require the extraction of this information to be successful. Therefore, the fourth moderator considered in the present study is type of task.

Some authors have suggested that whether participants perform the task using the left or right hand may also impact on the self-face advantage. In fact, some studies investigating face found a right-brain hemisphere advantage for self-face processing (Keenan, Freund, Hamilton, Ganis, & Pascual-Leone, 2000; Keenan, Ganis, Freund, & Pascual-Leone, 2000; Keenan et al., 1999; Keenan, Nelson, O’Connor, & Pascual-Leone, 2001; Keenan, Wheeler, Platek, Lardi, & Lassonde, 2003; Keenan, Wheeler, Gallup, & Pascual-Leone, 2000). However, some evidence also points towards a left-brain hemispheric advantage (Turk et al., 2002; Uddin, Rayman, & Zaidel, 2005) or no hemispheric advantage (Keyes & Brady, 2010). Therefore, the fifth moderator considered is laterality.

The last moderator considered is culture. There is consistent evidence suggesting that the self-face does not have the same importance across different cultures (Broesch, Callaghan, Henrich, Murphy, & Rochat, 2010). A greater self-face advantage for British than for Chinese subjects has been described (Sui, Liu, & Han, 2009). It has been proposed that Western cultures are known to be more individualistic and characterized by an independent self-construal, while Eastern cultures are characterized by an interdependent self-construal.

Method

Inclusion criteria

Studies were eligible for inclusion if they met the following criteria:

  1. 1.

    The study included healthy individuals between 18 and 60 years of age.

  2. 2.

    The study used an experimental design measuring self-face processing during an implicit or explicit self–other processing (i.e., face-recognition task, face-familiarity task, visual-search task, video-morphing task, self–other discrimination, head-angle task).

  3. 3.

    The dependent variable consisted of behavioral measures, more specifically, reaction time or a similar measure indicating time to respond, or the amount of information needed to make a choice and decide that the information belongs to oneself or someone else (e.g., percentage in a morphing continuum). Accuracy was not included since most studies report ceiling effects for self-face and familiar-face recognition, thus making comparison impossible.

  4. 4.

    Studies included the face as stimulus. That is, participants had to implicitly or explicitly respond to the image of their own face compared with someone else’s face.

  5. 5.

    Studies included someone else’s face to contrast with the self-face processing. The other face might be a familiar, friend’s, famous, or any other known or unknown face.

  6. 6.

    The contrast between self–other faces was assessed and tested through a within-subject design.

  7. 7.

    Selected studies included sufficient data for the calculation of Hedges’s gav, or authors were able to provide additional data when required. When data were not available, authors were contacted twice in an attempt to obtain more information.

Exclusion criteria

Studies meeting the following criteria were excluded:

  1. 1.

    Studies including only participants suffering from a psychiatric disorder or any neurological disorder.

  2. 2.

    Studies reporting only accuracy, neuroimaging, neurophysiological, or physiological data or self-reported measures.

  3. 3.

    Studies employing tasks of priming effect, double task, or interference task.

  4. 4.

    Studies employing visual-search paradigms in which the self-face was a distractor (since the self-face seems to hold attention longer than others’ faces, it interferes with the response to unknown faces; Devue, Van der Stigchel, Bredart, & Theeuwes, 2009).

  5. 5.

    Studies that do not evaluate the direct comparison between self–other processing.

Literature search strategy

To identify relevant articles, we first carried out an online search of the PsychINFO, and Medline databases up to December 2016 using the two main keywords self-face OR own face OR self-recognition. We did not add other keywords to further describe the type of task, measure, or participants in order to avoid excluding eligible studies. We searched for studies that evaluated self-face processing (implicit or explicit) in contrast to other-face processing. Second, we also searched Google Scholar citations of the articles included as well as unpublished studies, dissertations, and conference papers. Third, in the reference list of studies was scrutinized for studies not included in the electronic database. Fourth, theses and dissertations and other unpublished works were also searched for in OATD (Open Access Theses and Dissertations), ProQuest, and the ERIC index. Fifth, we sent requests via newsgroups and mailing lists to find unpublished studies (i.e., Society for Experimental Psychology and Cognitive Science, Society of Experimental Social Psychology). Sixth, the first author of the present article requested relevant unpublished research as a question topic on her Research Gate webpage. Finally, researchers whose articles were retrieved for inclusion were contacted by email as well as those with incomplete data. As a result of these efforts, we retrieved one unpublished thesis (two studies; three experiences; five effect sizes) and another partially published thesis, meaning that the author published the data of fewer participants than what was available in the thesis. Therefore, the final data set included a small subset of unpublished research.

Screening for eligible studies

The search resulted in 340 studies from PsycINFO, 280 additional studies from Medline, 244 studies from ProQuest, and 91 from the ERIC index, resulting in a total of 955 studies (see Fig. 1). Researchers also searched Google Scholar citations of the articles included as well as unpublished studies, dissertations, and conference papers. Two independent researchers analyzed the articles, and any disagreements were discussed. After reviewing the articles’ abstracts (Step 1), 212 studies were initially included. Subsequently, articles’ abstracts were analyzed, and 121 met our inclusion criteria. In Step 2, researchers read the full article. In the end, 62 were eligible to for inclusion. We analyzed these studies in order to determine whether they included enough information to calculate the effect size. Of these 62 studies, 21 studies lacked sufficient information, and 1 study lacked part of the data. We obtained additional data from the authors for seven of them. This resulted in the exclusion of 14 studies from the meta-analysis. Two studies reported data from the same sample. Thus, in total, 48 studies were included involving 54 samples and 116 effect sizes. Information regarding the processes of coding the studies is present as Supplementary Materials.

Fig. 1
figure 1

Flowchart of literature search and screening

Full size image

Meta-analytic procedures

To analyze the differences between self-face and other-face processing in terms of reaction time, we conducted analyses by calculating an effect size, namely, Hedges’s gav, the standardized mean difference in reaction time between the self-face- and other face, as recommended by Lakens (2013). Negative effect sizes indicate favorable reaction time for self-face compared with other face (for more information on effect-size calculation, please refer to the Supplementary Materials). Most studies investigating self-face recognition in contrast to the recognition of other people’s faces have generated multiple effect sizes per study. Thus, it is very likely that the effect sizes from the sample studies are more alike compared with effect sizes from different studies. In other words, our meta-analysis included nonindependent effect sizes. Nevertheless, traditional univariate meta-analytic approaches require no dependency between effect sizes. Consequently, we employed a three-level meta-analytic approach (Cheung, 2014; Hox, 2010), which is the best approach to account for the dependency of effect sizes. This approach takes into consideration three sources of variance: the sampling variance of the extracted effect sizes (Level 1), the variance between effect sizes extracted from the same study (Level 2), as well as the variance between studies (Level 3). Thus, we can analyze both differences in outcomes between studies (between-study heterogeneity) and within studies (within-study heterogeneity). The analyses were conducted with the rma.mv function of the metafor package (Viechtbauer, 2010) in the R statistical software environment (R Development Core Team, 2005) by using guidelines formulated by Assink and Wibbelink (2016) for modeling a three-level random-effects model as described by Van den Noortgate, López-López, Marín-Martínez, and Sánchez-Meca (2013). For more information regarding data analysis, please refer to the Supplementary Materials (Table 1).

Table 1. Summary of the coding system
Full size table

Results

Characteristics of the studies included

Fifty-three samples from the 48 articles were included in the study, representing 116 effect sizes. In total, 1,299 adults were included in the study. Table 2 summarizes the characteristics of the studies evaluating self-face processing. More information regarding the studies included in the meta-analyses can be found in Table S1 in the Supplementary Materials.

Table 2 Characteristics of the samples included in the studies investigating self-face processing
Full size table

Meta-analysis results

Overall effect

The overall average effect size of the contrast between self-face and other-face processing (including familiar, famous, and unknown) was −0.346 (95% CI [−0.484, −0.207], p < .001) indicating faster responses to self-face than to other people’s faces, which was, however, small. There was evidence of heterogeneity among the studies, Q(114) = 1127.257, p < .001.

Heterogeneity in effect sizes

Significant heterogeneity both at Level 2 within study (σ2 = 0.094, χ2 = 95.217, p < .001) and Level 3 between study (σ2 = 0.176, χ2 = 47.042, p < .001) was observed. Variance at Level 3, Level 2, and Level 1 explained, respectively, 60.27% (moderate heterogeneity), 32.04% (low to moderate heterogeneity), and 7.68% of the overall variance.

Egger’s intercept and sensitivity analysis

Egger’s intercept (intercept = 0.989, t = 5.347, p = .0001) appeared significant, indicating that the precision and size of studies included in the data set was considered asymmetrical, and therefore, biased. As mentioned before, we searched for influential outliers in order to evaluate the sensitivity of our analyses by comparing fitted models with and without effect sizes that we defined as influential outliers. Nevertheless, our results did not detect influential outliers in any of the data sets.

Moderating analysis

The following categorical variables were considered in the moderating variables: (1) familiarity (familiar vs. nonfamiliar); (2) identity (personally familiar, famous, and unknown); (3) head angle (profile vs. front face); (4) type of task; (5) laterality in terms of hand employed or visual field presentation (left, right, or no effect tested); (6) and culture (Eastern vs. Western).

Familiarity (familiar vs. unfamiliar)

No significant effect was observed when considering the moderator familiarity, F(1, 113) = 1.368, p = .245, indicating that participants responded faster to their own face in contrast to both unfamiliar (gav = −0.406, SE = 0.087, t = −4.662, p = .001) and familiar (gav = −0.313, SE = 0.075, t = −4.149, p = .001) ones.

Identity (personally familiar, famous, and unknown)

The effect of the moderator identity was not significant, F(2, 112) = 0.800, p = .452, indicating that the participants responded faster to their own face than to unknown faces (gav = −0.409, SE = 0.088, t = −4.673, p < .001), faces belonging to personally familiar faces such as friends and relatives (gav = -0.312, SE = 0.086, t = −3.640, p = .001) as well as famous faces (gav = −0.304, SE = 0.113, t = −2.682, p = .008).

Head angle

No significant differences were found when considering the moderator effect of head angle, F(1, 113) = 0.920, p = .340. This result indicates that participants responded faster to their own face than to other people’s faces regardless of whether faces were presented only in frontal view (gav = −0.302, SE = 0.083, t = −3.620, p = .001) or in profile/frontal view (gav = −0.431, SE = 0.113, t = −3.804, p = .001).

Interaction between familiarity and head angle

The effect size of each type of task indicated that when performing a task involving attentional processes (e.g., visual search task), participants responded with similar speed to their own face and to other people’s faces (gav = −0.185, SE = 0.217, t = −0.851, p = .396). On the other hand, when requested to perform an identification task based on memory (e.g., identifying whether the face belonged to oneself or to someone else; gav = −0.339, SE = 0.081, t = −4.190, p < .001), participants were faster when responding to their own face compared with other people’s faces. Likewise, when participants performed an identification task of other features of the image rather than identity (e.g., head orientation toward left or right; gav = −0.495, SE = 0.172, t = −2.873, p < .005), participants were faster when responding to their own face than to other people’s faces. Nevertheless, no significance was observed regarding the overall moderator effect, F(2, 113) = 0.658, p = .520.

Type of task: Attentional, memory, or perception

The effect size of each type of task indicated that when performing a task involving attentional processes (e.g., visual search task), participants responded equally quickly to their own face and to other people’s faces (gav = −0.185, SE = 0.217, t = −0.851, p = .396). On the other hand, when requested to perform an identification task based on memory (e.g., identifying whether the face belonged to oneself or to someone else; gav = −0.339, SE = 0.081, t = −4.190, p < .001), participants were faster when responding to their own face than to other people’s faces. Likewise, when participants performed an identification task of other features of the image rather than identity (e.g., head orientation toward left or right; gav = −0.495, SE = 0.172, t = −2.873, p < .005), participants were faster when responding to their own face than to other people’s faces. Nevertheless, no significance was observed regarding the overall moderator effect, F(2, 113) = 0.658, p = .520.

Laterality

Overall analysis indicated that the effect of the moderator laterality was not significant, F( 2, 80) = 0.210, p = .811). Differences were observed between self–other processing in terms of reaction time regardless of whether participants used their left hand (gav = −0.478, SE = 0.174, t = −2.737, p = .008), right hand (gav = −0.391, SE = 0.117, t = −3.344, p = .001), or both hands (gav = −0.474, SE = 0.156, t = −3.043, p = .003).

Culture

Finally, a significant effect was observed regarding the moderator culture, F(1, 113) = 10.900, p = .001. Participants from Western cultures responded faster to their own face than to other people’s faces (gav = −0.566, SE = 0.097, t = −5.820, p < .001), while for participants from Eastern cultures, the effect was not significant (gav = −0.124, SE = 0.097, t = −1.275, p = .205).

Discussion

Self-face advantage has been massively investigated in the domain of cognitive psychology, neuropsychology, and social neurosciences. While attention has been devoted to further determine which cerebral regions are specifically involved in self-face processing, less effort has been made in summarizing and understanding in depth whether individuals also behaviorally respond in a specific way when seeing their own face. Thus, the main aim of this meta-analysis was to evaluate whether there is self-advantage when responding to one’s own face. The overall effect of the present meta-analysis indicated that self-face processing benefits from a self-advantage, inducing a small effect size. Thus, this meta-analysis is the first to indicate that individuals respond faster to their own face in contrast to other faces. However, this initial result must be considered with caution, especially due to a high level of heterogeneity detected and publication bias.

We also considered different moderators. Below, we discuss each one of these moderators individually. The three first moderators considered in the present meta-analysis were Familiarity, Identity of the face, and Head angle. First, we considered the level of familiarity, since previous studies have shown that we respond faster to familiar faces compared with unfamiliar faces (Burton et al., 2011), and thus we could expect that the overall effect observed could be larger when considering self-face versus unfamiliar faces. Second, we considered the effect of identity, since differences have been found in terms of behavioral performance (Liccione et al., 2014), as well as physiological (Guerra, Sanchez-Adam, Anllo-Vento, Ramirez, & Vila, 2012a; Guerra et al., 2012b; Vico, Guerra, Robles, Vila, & Anllo-Vento, 2010), and neuronal (Gobbini & Haxby, 2007; Gobbini et al., 2004) processes involved in the processing of famous (people we have seen only through the media) and familiar (people we have personally encountered in life) faces. These differences may be explained by the amount and type of experience individuals have with the familiar and famous (Burton et al., 2011) as well as other affective aspects (Guerra et al., 2012a). As suggested by Liccione et al. (2014), “affective and emotional aspects related to personal narratives with others seem to play a special role in face processing” (page 8). Thus, we may expect that the self-face may rely on different processes when compared to a famous face (and unknown faces) but may be characterized by a robust mental representation such as other familiar faces, since the self-face is a familiar face as well. Thirdly, Head angle was also considered as a possible moderator since familiar faces and, notably self-faces, benefit from a view-invariant characteristic due to the amount of experience individuals have with these faces (Johnston & Edmonds, 2009). Overall, our results showed that the effect of all three moderators (Familiarity, Identity or Head angle) were not statistically significant.

Thus, our results indicated that regardless of the face identity or level of familiarity, we tend to respond faster to our own face than to other people’s faces. When considering the role of identity levels, we can observe that the effect size associated with the self and other comparison increased from famous (gav = - 0. 304) and personally familiar faces (gav = - 0. 312) to unknown faces (gav = -0.409). In other words, there may be a progression in terms of response time across these different levels of familiarity. These results also suggested that self-face processing is faster than other face processing regardless of the head angle of the face presented, in accordance with the study by Tong and Nakayama (1999) as well as Troje and Kersten (1999), who provided evidence that the self-face includes some view-invariant representations. Our results suggest that these view-invariant representations seem to be even more important when considering one’s own face compared to other familiar faces.

Overall, these results indicated that the self-face may benefit from a stronger and more robust mental representation, which includes both a frontal and profile view and may result in a self-face advantage in terms of reaction time. We suggest that our results can be further discussed in light of different theories emphasizing the role of familiarity as well as the role of multisensory information in self-face recognition (see next section for a detailed discussion) obtained through photographs, the experience of seeing oneself in the mirror and by touching one’s own face.

The fourth factor included in our moderator analyses was the Type of task employed, more specifically, whether the task to be performed relied more on memory, perception or attention. We found a self-face advantage for memory-based identification tasks. Therefore, our results are in line with previous research showing that familiar faces benefit from a robust mental representation, which results in a faster reaction time when participants are requested to identify or recognize the face (Jenkins & Burton, 2011; Johnston & Edmonds, 2009) and suggests that the self-face may benefit from an even more robust mental representation compared to other familiar faces. The same faster reaction time was observed for the tasks relying on perception-based identification. It should be noted that most studies included in this category overlap with the studies included in the “profile/frontal view” category discussed before.

Conversely, our results suggest no self-face advantage in terms of reaction time when participants were performing a visual search task or a face detection task. In agreement, Devue et al. (2009), found that the self-face does not attract attention more than others’ faces, but instead participants may have difficulties in disengaging attention from their own face. This effect does not seem to be specific to self-face, but it is also present in other familiar faces. Indeed, 8 out of 12 effect sizes included under the factor “Attentional tasks” included familiar (personally familiar or famous) faces. Since the self-face as a distractor tends to interfere with the ongoing task, which would increase the difference in terms of reaction time between the self-face and other people’s faces, we included in the present meta-analysis only those studies that used unknown faces as a distractor and self or familiar faces as a target. Thus, our results cannot be explained by the interference effect of self-face. It seems that when the task to be performed relies mostly on attention, all faces are detected at a similar speed. Conversely, it is also possible that the analysis was underpowered due to the small number of studies. It is also important to have in mind that the moderator type of task was not significant.

Regarding the effect of Laterality, our results revealed that the effect of this moderator was not statistically significant. This result can be interpreted in the light of previous evidence showing a right hemisphere specialization for processing faces in general and, more precisely, for the recognition of familiar faces (Hole & Bourne, 2010), when the task was devoid of explicit verbal requirements (Ramon & Rossion, 2012). It is suggested that the right hemisphere is involved in the storage of semantic person information based on sensory information, while the left hemisphere stores verbally coded information (Gainotti, 2015). Moreover, two meta-analyses performed by Devue and Bredart (2011) and Platek et al. (2008) seem to indicate no specific cerebral region nor a hemispheric dominance for self-face processing. Another study also suggested that there is no evidence showing that one’s own face is associated with specific processes at neuronal level (Gillihan & Farah, 2005).

Finally, we also considered the effect of Culture on reaction time to self-face compared to other people’s faces. Our results revealed that the self-face advantage was stronger for participants recruited in western countries compared to eastern countries. This result is in agreement with previous studies by Sui et al. (2009) and Liew et al. (2011) who demonstrated a larger and more consistent self-face advantage in British participants compared to Chinese participants. Results could be understood in the light of cultural differences in terms of the place of the self in society. Western countries are normally characterized by an independent self-construal while eastern cultures are mainly characterized by an interdependent self-construal (Markus & Kitayama, 1991).

Self-face advantage

Except for the significant moderator Culture, the other factors considered (Familiarity, Identity, Head angle, Type of Task and Laterality) did not moderate the overall effect of self-face advantage. In other words, our results suggest that individuals respond faster when seeing their own face regardless of these five aspects. As mentioned before, several hypotheses might help us to better understand our results.

First, our results can be interpreted in view of previous studies showing that familiar faces are processed differently from unfamiliar faces and that this effect seems to be modulated by the intensity and variability of the experience with a person (Burton et al., 2011; Jenkins & Burton, 2011). In other words, the more we have seen a face the better and more robust will be the corresponding stored representation. In light of this theory, we might suggest that individuals are in constant contact with their face through, for example, the mirror and photographs. It may enhance the stored representation of their face leading to response times in experimental studies.

Secondly, the affective response to the face might also be an important factor to consider. A series of studies conducted by Guerra, Vico, et al. (2012b) provide further support for the existence of a stronger emotional arousal associated with loved familiar faces in contrast to famous and unknown faces. Regarding the self-face, studies have shown that self-face processing is associated with stronger skin conductance compared to familiar, famous and unfamiliar faces (Bagnato et al., 2010), which may be indicative of increased emotional arousal (Guerra, Vico, et al. 2012b).

Thirdly, the self-face advantage could also be explained by implicit positive attitudes toward the self (Ma & Han, 2010). Studies have investigated whether individuals implicitly associated positive characteristics with the self. Epley and Whitchurch (2008) showed that individuals tend to have a more attractive mental representation version of their own face when compared to the real version. In short, these studies concur with social psychology studies showing that individuals tend to implicitly associate the self with positive characteristics (Koole & DeHart, 2007), which further influence response time to self-face.

Another hypothesis suggests that self-face advantage might be related to specific face processing strategies. Contrary to other individuals’ faces, we mostly have contact with our own face through the mirror. The image provided by the mirror differs substantially from the direct perception of others’ faces. Thus, it is suggested that different strategies are employed during self-face processing. For instance, Keyes (2012) found that self-face representation relies both on featural (processing of isolated facial features) and configural processing (processing the relationship between the different features of the face).

Fifthly, contrary to familiar, famous and unknown faces, self-face processing also relies on multisensory information including a combination of visual, somatosensory, proprioceptive, and motor information. This combination of information might lead to faster processing of self-face compared to other peoples’ faces. According to Sugiura (2013), the physical self or bodily self-recognition (e.g. recognition of one’s own face or body), is “grounded by the experience of bodily action accompanied by visual, somatosensory, vestibular, and interoceptive feedback”, namely, sensorimotor schema (Sugiura, 2013; page 3). Thus, the ability to recognize one’s own face in a static image is suggested to depend on the experience with one’s face in the mirror and thus on the integration of visual, proprioceptive, motor, and tactile information. Other studies also suggest that the mental representation of one’s own face may be enriched by both the visual experience we have with our own face and the combination of visual, somatosensory, proprioceptive, and motor information (Tajadura-Jimenez, Longo, Coleman, & Tsakiris, 2012; Tsakiris, 2008).

Sixthly, the self-face advantage may be explained by the “integrative self” hypothesis (Sui & Humphreys, 2015). According to Sui and Humphreys (2015), self-representations “acts as an integrative hub for information processing, helping to bind together different types of information and even different stages of processing”. Consequently, self-related information will be assimilated and accumulated, creating “core self-representations”, which will facilitate the processing of stimuli associated with distinct aspects of the self (e.g. physical or psychological aspects of the self), such as enhancing memory of stimuli in relation to the self (Symons & Johnson, 1997) and the integration of different self-related shapes into a single representation (Sui, He, & Humphreys, 2012).

In summary, several hypotheses might explain self-face advantage observed in the present meta-analysis. Interestingly, these arguments are not mutually exclusive, since the stored representation of one’s own face might be composed of visual, proprioceptive, motor, tactile, emotional and autobiographical information, which might be what really differentiates self-face from other face processing. This stored representation of self-face might further contribute to the way individuals perceive themselves in society and in relation to others (Sugiura, 2013).

Implications for future studies on self-face processing

The results provided in the current meta-analysis contribute to the current debate on whether self-face is a unique stimulus that involves specific mechanisms. While some authors have argued and also provided evidence in favor of a self-face advantage (Keenan et al., 2001; Ma & Han, 2010), others have rather suggested that the self-face is merely a familiar face (Gillihan & Farah, 2005). Still, others have found support for specialized self-face processing brain regions, without any differences at behavioral level (Kircher et al., 2000; Kircher et al., 2001). On the one hand, by showing that the self-face indeed benefits from an advantage in terms of reaction time, our findings corroborate the hypothesis that the self-face is associated with specific processes at behavioral level that may be explained by the fact that the self-face is a unique stimulus or because we are more familiar with our own face compared to other people’s faces including other familiar faces.

Although Familiarity, Identity, Head angle, Type of task, and Laterality do not moderate the self-face advantage, the findings of our meta-analysis should incentivize researchers to consider new aspects that might help to further understand self-face processing at behavioral level. Firstly, one should consider how factors that normally affect face recognition, such as viewpoint, expression, and context, also affect self-face processing. It is well known that recognition of familiar faces is less sensitive to these factors compared to unknown faces. Most studies included in the present meta-analysis only explored changes in head angle using standard images. To our knowledge, only one study has employed natural images during self-face processing (Bortolon et al., 2017) and found a self-face advantage over famous and unknown faces, but not over friends’ faces. Secondly, another factor that should be considered is the emotional responses associated with self-face processing in contrast to other kinds of familiar faces, similar to the studies performed by Guerra and colleagues (Guerra, Sanchez-Adam, et al., 2012a; Guerra, Vico, et al., 2012b; Vico et al., 2010). Third, one should also consider the impact of the level of the bond between the individual and the familiar person. Most studies included friends’ faces and did not properly access the degree of friendship or the intensity and variability of individuals’ experience with this familiar person. One study by Wang and Zheng (2015), for instance, showed that the level of intimacy impacted on the processing of other people’s faces by reducing the response time. Fourth, more studies should consider the cultural implications for self-processing. So far, only a few studies (Liew et al., 2011; Sui et al., 2009) have directly compared Eastern and Western cultures. Other cultures are still to be explored, notably African cultures. Previous studies have shown that African toddlers do not show the same pattern of behavior as American toddlers during the self-mirror recognition test (Broesch et al., 2010). Fifth, the performance on different tasks relying more or less on memory or attention should also be further explored. Finally, although hemispheric dominance during self-face processing has often been investigated in the past, most recent studies have neglected this aspect, which still deserves attention.

Limitations of the present study

The present meta-analysis could not include some studies due to the lack of necessary data for calculating effect size. Initially, we tried to circumvent this limitation by contacting all authors at least twice by email. Nevertheless, a large number of authors did not reply to our request. Others were unable to retrieve the data. Thus, even though we tried to include the largest possible number of studies, it is possible that other studies have not been included in this meta-analysis. Moreover, although we made great efforts to retrieve unpublished data, we were able to find only two thesis dissertations and only access the data from two studies (three experiments) of one of these theses. Moreover, authors also made a great effort to select the study characteristics for which a strong theoretical case can be made in order to avoid identifying false moderator variables (Hunter & Schimidt, 2004).

The present study presented higher levels of heterogeneity and publication bias as indicated by the Egger’s test, which weakens the inferences presented here, especially considering the small effect size observed. The significant effects observed in the Egger’s test can indeed be explained by a publication/reporting bias, but also by the high levels of heterogeneity observed (Sterne et al., 2011) and the fact that most data included in the present meta-analyses were extracted from studies with small samples sizes that observed positive self-face advantage effects. Although we explored the effect of six different moderators, our moderator analysis did not allow us to explain the high levels of heterogeneity described. Therefore, the results presented here should be considered with caution. Regardless of whether the bias encountered is indeed due to a publication bias and/or due to higher levels of heterogeneity, it represents an important limitation of the present meta-analysis that may limit the validity of the results.

One previous review on the subject found no evidence of self-physical uniqueness or advantage, either at the behavioral or neurocognitive level (Gillihan & Farah, 2005). Therefore, Gillihan and Farah (2005) concluded that at the time there was no consistent evidence to suggest that self-processing was a unique stimulus dissociated from nonself processing. Similarly, Devue and Bredart (2011) concluded that even though a bilateral network seems to be associated with self-face recognition, it remains difficult to dissociate specific brain areas that are associated with self-face processing, in part because of the variability of control stimuli and experimental tasks. A similar conclusion may be applied to the present results. The variability of type of stimuli and experimental task, together with the unavailability of some data, may prevent us from reaching more definite conclusions.

Moreover, our study was limited by the fact that we could not adjust for publication bias (Duval & Tweedie, 2000), since such procedures are not available for the rma.mv function of the metafor package. On the other hand, we analyzed sensitivity by defining the influential outliers and comparing the fitted model with and without these values. No influential outliers that may have affected our results were found. Moreover, one of the main strengths of this study is the use of a three-level random-effects model, since it enables us to take into account different levels of variability and dependency between effect sizes.

Conclusions

Overall, our findings suggest that the self-face benefits from advantage at behavioral level (reaction time), which might be explained by different factors such as over-familiarization with one’s own face, unique experience with one’s own face, multisensory integration, and attribution of positive characteristics to the self. Regarding our moderator analysis, we found that culture seems to play an important role in the way we respond to our own face in behavioral terms, probably due to the opposition between independent self-construal and interdependent self-construal that have been suggested to characterize Western and Eastern cultures, respectively.