Holistic integration of gaze cues in visual face and body perception: Evidence from the composite design

Abstract

A considerable amount of research on identity recognition and emotion identification with the composite design points to the holistic processing of these aspects in faces and bodies. In this paradigm, the interference from a nonattended face half on the perception of the attended half is taken as evidence for holistic processing (i.e., a composite effect). Far less research, however, has been dedicated to the concept of gaze. Nonetheless, gaze perception is a substantial component of face and body perception, and holds critical information for everyday communicative interactions. Furthermore, the ability of human observers to detect direct versus averted eye gaze is effortless, perhaps similar to identity perception and emotion recognition. However, the hypothesis of holistic perception of eye gaze has never been tested directly. Research on gaze perception with the composite design could facilitate further systematic comparison with other aspects of face and body perception that have been investigated using the composite design (i.e., identity and emotion). In the present research, a composite design was administered to assess holistic processing of gaze cues in faces (Experiment 1) and bodies (Experiment 2). Results confirmed that eye and head orientation (Experiment 1A) and head and body orientation (Experiment 2A) are integrated in a holistic manner. However, the composite effect was not completely disrupted by inversion (Experiments 1B and 2B), a finding that will be discussed together with implications for future research.

Introduction

Although the configuration of a face is relatively stable (i.e., two eyes above the nose, a nose above the mouth), it nonetheless inherently holds complex information, such as identity, attractiveness, gender, and emotion. Being able to perceive all of this information in a single glance is said to be crucial for socially adept behavior, and part of what makes face perception special. More specifically, it is generally assumed that faces are processed as a holistic gestalt (e.g., Farah, Wilson, Drain, & Tanaka, 1998; Maurer, Le Grand, & Mondloch, 2002; Van Belle, De Smet, De Graef, Van Gool, & Verfaillie, 2009), indicating that human observers are able to rapidly perceive faces as integrated wholes. Recently, research on holistic perception has been extended to another class of highly socially relevant stimuli, namely bodies. Evidence seems to reveal overlapping areas for faces and bodies in the macaque and human brains (Bartels & Zeki, 2004; Peelen & Downing, 2007; Pinsk, DeSimone, Moore, Gross, & Kastner, 2005; Tsao, Freiwald, Knutsen, Mandeville, & Tootell, 2003), indicative of a similar processing style, which seems to be confirmed by behavioral studies in humans (Reed, Stone, Bozova, & Tanaka, 2003; Slaughter, Stone, & Reed, 2004). 

Following another person's gaze (i.e., joint attention) is an ability that is acquired early in life, and a lack thereof is assumed to be an important marker for autism spectrum disorder (ASD; Ashwin, Hietanen, & Baron-Cohen, 2015; Baron-Cohen, Campbell, Karmiloff-Smith, Grant, & Walker, 1995; Leekam, Hunnisett, & Moore, 1998). Adequate gaze perception enables observers to infer the motivations, point of interest, and intentions of another person. Although gaze information contains important social implications for everyday life, it remains unstudied in the context of holistic processing. Nonetheless, as with identity and emotion perception, people have a lifelong experience in gaze detection, enabling them to effortlessly estimate gaze direction and differentiate between direct and averted gaze (Bock, Dicke, & Thier, 2008; Gamer & Hecht, 2007; Symons, Lee, Cedrone, & Nishimura, 2004). However, the placement of the iris relative to the sclera is not the only factor that plays a role in gaze perception, and research in the field of social attention reveals that an interaction between head orientation and eye-gaze perception takes place in estimating gaze direction (Anstis, Mayhew, & Morley, 1969; Cline, 1967; Frischen, Bayliss, & Tipper, 2007; Gibson & Pick, 1963; Hietanen, 1999; Moors, Verfaillie, Daems, Pomianowska, & Germeys, 2016; Wollaston, 1824). More specifically, perception of eye gaze has been shown to be biased toward the direction opposite the head orientation, but only when eye and head orientation are not aligned. This bias is termed the overshoot effect and has recently also been found for perception of head orientation relative to body orientation (Hietanen, 2002; Moors, Germeys, Pomianowska, & Verfaillie, 2015; Pomianowska, Germeys, Verfaillie, & Newell, 2011). 

In the present study, the question of holistic gaze integration in face and body perception is addressed using the composite design, a highly controllable method to systematically investigate holistic processing (e.g., Hole, 1994; Rossion & Boremanse, 2008; Young, Hellawell, & Hay, 1987; for a review, see Rossion, 2013). In this task, originally proposed by Young et al. (1987), new composite faces are constructed by pairing the upper face half (i.e., from the nose upwards) from one familiar person with the lower face half from another familiar person. While identifying the person from the task-relevant half (i.e., upper or lower face half), participants experience a perceptual interference from the nonattended half, an effect that disappears when the face halves are horizontally misaligned and thus break the configuration. The results indicate the importance of configurational processing on the recognition of the separate parts (i.e., the upper and lower face halves). The largest limitation of this naming task (i.e., only familiar faces could be used) was later resolved by introducing a matching-to-sample variant of the composite design (Hole, 1994), in which participants had to decide on the similarity of the relevant halves of two sequentially presented composite faces (i.e., a target and a test face). The matching-to-sample task is now the more frequently used variant of the composite design in research on holistic processing. The composite design has been widely employed in research on holistic face and body perception in general populations (e.g., Robbins & Coltheart, 2012; Robbins & McKone, 2007; Rossion & Boremanse, 2008; Willems, Vrancken, Germeys, & Verfaillie, 2014) as well as in populations with hypothesized face-perception difficulties (Busigny, Joubert, Felician, Ceccaldi, & Rossion, 2010; Gauthier, Klaiman, & Schultz, 2009; Nishimura, Rutherford, & Maurer, 2008; Teunisse & de Gelder, 2003). However, whereas research with the composite design has mostly focused on holistic processing of identity (e.g., Rossion & Boremanse, 2008), emotion (e.g., Calder & Jansen, 2005; Calder, Young, Keane, & Dean, 2000), and even gender (e.g., Baudouin & Humphreys, 2006), to our knowledge no research with the composite design exists for gaze perception. 

Present study

We created two composite designs, a face one and a body one, based on gaze perception. For both designs, gaze perception is defined as the perception of where a person or the stimulus is looking. For the face design (Experiment 1), estimation of gaze direction was assessed through the eyes. For the body design (Experiment 2), the head indicated gaze orientation, since the eyes were barely visible and were held constant at 0° relative to the head orientation. In Experiment 1, eye and head orientation were manipulated independently, with the eyes as the relevant or to-be-attended stimulus half and the head as the nonrelevant or to-be-ignored stimulus half. Manipulation of eye orientation consisted exclusively of changes in the position of the pupil relative to the sclera; all other features of the relevant halves remained stable throughout the experiment. In Experiment 2, head and body orientation were manipulated independently, with the head as the relevant or to-be-attended stimulus half and the body as the nonrelevant or to-be-ignored stimulus half. See Figure 1 for an example of the stimuli from Experiments 1 and 2. 

Figure 1

 

Stimuli from (A) Experiment 1 and (B) Experiment 2. Both are examples from “same incongruent” trials, illustrated in an “aligned” and a “misaligned” version. More specifically, relevant upper halves are identical, whereas nonrelevant lower halves are different. Face stimuli (Experiment 1) were depicted on a white background, and body stimuli (Experiment 2) were depicted on a gray background.

The face and body stimuli were presented upright in Experiments 1A and 2A and inverted in Experiments 1B and 2B. Rotating faces by 180° is hypothesized to disrupt holistic processing (e.g., Goffaux & Rossion, 2006; McKone et al., 2013; Mondloch & Maurer, 2008; Van Belle, Verfaillie, Rossion, & Lefèvre, 2010; Young et al., 1987). Presumably, the holistic representation of upright faces that is innate and further developed through visual experience is absent for inverted faces. This limits the perceptual field with which observers perceive inverted faces, resulting in a diminished influence from the whole facial context (McKone et al., 2013; Rossion, 2008, 2009). Hence, the inversion experiments were conducted to make sure that the obtained effects were not the result of generic mechanisms other than holistic processing. 

Two versions of the composite design have been proposed: standard (e.g., Robbins & McKone, 2007; Rossion, 2013) and alternative (e.g., Gauthier & Bukach, 2007; Richler & Gauthier, 2013). The main difference between the two versions is the number of manipulated factors (see Figure 2), but the standard design can easily be extracted from the alternative design. For this reason, and because research on reliability and response bias seems to point to the alternative variant as the preferable design (Meinhardt, Meinhardt-Injac, & Persike, 2014; Richler, Cheung, & Gauthier, 2011), we opt for this variant in the present study. As can be seen in Figure 2, in the alternative design three factors are manipulated independently: similarity, congruency, and alignment. 

Figure 2

 

Illustration of the alternative composite design. Relevant upper halves (depicted in white) determine the similarity factor and are either “same” or “different.” Nonrelevant lower halves (depicted in gray) can also be “same” or “different” and, together with the upper halves, determine the congruency factor. If upper and lower halves within one trial elicit the same response (i.e., both “same” or both “different”), the trial is “congruent”; if not, it is “incongruent.” Red squares indicate conditions from the standard composite design which are embedded within the alternative design.

Whereas the similarity factor is based on task-relevant halves only (i.e., same or different relevant halves in target and test stimulus), congruency is based on the physical sameness from both relevant and nonrelevant halves. Congruent trials are trials where target- and test-nonrelevant halves would elicit the same response as the target- and test-relevant halves, whereas incongruent trials are trials in which relevant and nonrelevant halves would elicit different responses (e.g., relevant halves are the same, but nonrelevant halves are different). Holistic processing is then measured as a significant Congruency × Alignment interaction, with poor performance on incongruent compared to congruent trials, but only when stimuli are aligned. We predict that this interaction will be significant for the upright stimuli (Experiments 1A and 2A) but not the inverted stimuli (Experiments 1B and 2B). Furthermore, conditions from the standard design can easily be derived from the alternative design (see Figure 2), and we thus also include a separate analysis on these conditions. According to the standard design, effects of holistic processing should be apparent specifically for the “same incongruent” conditions, with a significant difference between the “aligned” and “misaligned” trials (e.g., Jiang, Blanz, & Rossion, 2011; Robbins & McKone, 2007; Rossion, 2013). This hypothesis will also be addressed, and we predict that the difference between “same incongruent aligned” and “same incongruent misaligned” trials will be significant for the upright stimuli (Experiments 1A and 2A) but not the inverted stimuli (Experiments 1B and 2B). 

Experiment 1A: Upright face composite design

Methods

Participants

A total of 31 participants completed the first experiment (27 female, four male; age: M = 19.5 years, SD = 6.8). All had normal or corrected-to-normal vision, and received course credit after participation. Furthermore, none of the participants reported ever having experienced face-recognition problems. Prior to this and all following experiments, participants received a verbal explanation of the design and were asked to sign written informed consent forms in conformation to the Medical Ethics Committee of Katholieke Universiteit Leuven and the ethical standards laid down in the 1964 Declaration of Helsinki. 

Stimuli and design

The stimulus-set consisted of 49 male 3-D heads, created in Poser 8 (Smith Micro Software, Aliso Viejo, CA). Head and eye orientation were manipulated independently within a range from −30° to 30° in steps of 10° (with 0° as the straight-ahead view), but identity remained stable. Aligned composites were created by applying a horizontal 2-pixel line at the same height for every figure (i.e., between the eyes and nose). Next, misaligned stimuli were obtained by shifting the lower halves to the left and right relative to the vertical midline between the eyes. Confounding factors other than head orientation that could influence the perception of eye orientation were eliminated (i.e., featural changes due to shifting profile view, such as covered ears and shadows) by pasting the eye area of a standard head orientation (0°) on other head orientations. As a result, the perception of different eye orientations in “same” trials (i.e., eye orientations are identical) could be attributed only to experimentally manipulated factors (i.e., head orientation, alignment, and congruency). For “different” trials (i.e., eye orientations of test and target stimuli are different within one trial), eye orientation was manipulated by shifting the pupil relative to the sclera. All other features from the relevant eye area remained identical throughout the experiment. 

To maximize a potential perceptual interference from nonrelevant lower halves on the perception of relevant upper halves within trials, the nonrelevant head orientations had to differ at least 30°, whereas the relevant eye orientations could shift only 10° between target and test stimuli. The latter restriction was also imposed to create a certain level of difficulty and to avoid ceiling effects. Note, however, that these restrictions were imposed only where applicable (i.e., in trials where relevant and/or nonrelevant halves differed within one trial). Taking these restrictions into account, the resulting trial list consisted of 1,539 possible test–target pairs, of which 240 (30 per condition) were randomly chosen for every participant by Affect, a program for experimental design (Spruyt, Clarysse, Vansteenwegen, Baeyens, & Hermans, 2010). 

Procedure

Each participant completed a total of 240 trials, divided into five blocks of 48 trials. Each block consisted of six trials per condition, but besides this restriction, all trials were randomly presented to the participant. Before the experimental procedure started, a practice run of eight randomly chosen trials (one per condition) was presented to the participants, for which feedback was provided. On each trial, participants had to decide on the similarity of eye orientation of the two sequentially presented target and test faces, irrespective of head orientation. A trial started with a fixation cross (1° × 1°) for 1500 ms, continued with the target face (8.5° × 8°) for 250 ms and a blank screen for 1000 ms, and ended with the test face (8.5° × 8°), which again was presented for 250 ms (see Figure 3). Participants were instructed to respond as fast as possible but to beware of unnecessary errors. The next trial started immediately after a response was given, which was possible by pressing the Y or B key on a keyboard. The corresponding response-possibilities (i.e., “same” and “different”) were counterbalanced between participants. 

Figure 3

 

Experimental procedure for (top) Experiment 1 and (bottom) Experiment 2. For both sequences, a “same congruent” trial is illustrated, with identical relevant and nonrelevant halves.

Results

Accuracy, defined as the percentage of correct responses, and reaction times were recorded for each participant. Outliers in reaction times were removed for every participant separately. Due to a skewed distribution, different criteria for upper and lower bounds were defined. For the upper bound, all reaction times within a condition were removed if they exceeded the average reaction time plus three times the standard deviation for that condition. For the lower bound, all reaction times faster than 250 ms were removed. This resulted in the removal of 1.3% of the total amount of data. 

Since results from the standard design are nested within the alternative design (see Figure 2), analyses from the standard conditions alone will also be reported to compare results from both designs. In accordance with the alternative design, we were interested in a significant Congruency × Alignment interaction. More specifically, a perceptual interference from nonrelevant face halves is apparent for “incongruent” trials measured by higher reaction times and more errors compared to “congruent” trials, but only for “aligned” conditions, in which holistic processing is hypothesized to take place. This effect is termed the congruency effect and is said to be a crucial measure for holistic processing in the alternative composite design. In “misaligned” trials, however, holistic processing is disrupted by shifting nonrelevant halves horizontally, resulting in a diminished difference between “incongruent” and “congruent” trials (i.e., the congruency effect disappears in misaligned trials). To test this hypothesis, a repeated-measures analysis of variance (rmANOVA) was conducted, with similarity (same/different relevant halves), congruency (congruent/incongruent), and alignment (aligned/misaligned) as within-subject factors. Furthermore, an rmANOVA was also conducted on “same” and “different” trials separately, since the hypothesis from the standard composite design states that holistic processing should be apparent only for “same” trials, not for “different” trials (e.g., Rossion, 2013). In a last step, to test the hypothesis of the standard design specifically, two paired-samples t tests are reported. Under the assumption of holistic processing, the difference between “aligned” and “misaligned” trials should be significant for “same incongruent” trials but not for “different congruent” trials. Holistic processing in the standard design is thus measured by means of an alignment effect, whereas the alternative design uses a congruency effect. Both will be discussed. Partial eta squared ( Display Formula ) is reported as an estimation of effect size. Results of Experiment 1A are shown in Figure 4A.  

Figure 4

 

Summary results for (top) the upright design (Experiment 1A) and (bottom) the inverted design (Experiment 1B) for the face stimuli. Left (accuracy) and right (correct reaction times) panels depict means for the eight conditions. Error bars represent the standard errors.

Accuracy

Due to the multiplicity of main and interaction effects, only those relevant for the current design are discussed. Note that effects for congruency and alignment (i.e., the factors) mentioned here are not to be confused with the congruency and alignment effects mentioned earlier. For accuracy, results revealed significant main effects of similarity: F(1, 30) = 19.076, p < 0.001, Display Formula = 0.389; congruency: F(1, 30) = 40.399, p < 0.001, Display Formula = 0.574; and alignment: F(1, 30) = 15.282, p < 0.001, Display Formula = 0.337. More specifically, participants made fewer errors on “same” trials (83% vs. 71%), “congruent” trials (83% vs. 70%), and “misaligned” trials (79% vs. 75%). More relevant are the significant Congruency × Alignment interaction: F(1, 30) = 10.794, p = 0.003, Display Formula = 0.265; and Similarity × Congruency × Alignment interaction: F(1, 30) = 15.871, p < 0.001, Display Formula = 0.346. Separate analyses on “same” and “different” trials revealed significant effects for “same” trials—congruency: F(1, 30) = 58.216, p < 0.001, Display Formula = 0.660; alignment: F(1, 30) = 28.970, p < 0.001, Display Formula = 0.491; Congruency × Alignment: F(1, 30) = 20.244, p < 0.001, Display Formula = 0.403)—and a significant effect of congruency in “different” trials, F(1, 30) = 6.194, p = 0.019, Display Formula = 0.171, due to more errors on “different incongruent” trials. None of the other effects for “different” trials reached significance, however.  

Finally, analyses of accuracy in the conditions of the standard design matched those of the alternative design. As predicted, the difference between “aligned” and “misaligned” trials was significant for “same incongruent” trials, t(30) = 5.193, p < 0.001, Display Formula = 0.473, but not for “different congruent” trials. More specifically, participants made fewer errors on “same incongruent misaligned” trials compared to “same incongruent aligned” trials (82% vs. 64%). These results are indicative for holistic processing of the face stimuli, but only for “same” trials.  

Correct reaction times

Main effects of congruency, F(1, 30) = 29.351, p < 0.001, Display Formula = 0.495, and alignment, F(1, 30) = 20.987, p < 0.001, Display Formula = 0.412, were significant, but not of similarity. Specifically, subjects responded faster on “congruent” trials (914 vs. 968 ms) and “misaligned” trials (915 vs. 967 ms). The Congruency × Alignment interaction was again significant, F(1, 30) = 5.266, p = 0.029, Display Formula = 0.149, as was the three-way Similarity × Congruency × Alignment interaction, F(1, 30) = 29.753, p < 0.001, Display Formula = 0.498. All effects for the “same” trials were significant—congruency: F(1, 30) = 64.259, p < 0.001, Display Formula = 0.682; alignment: F(1, 30) = 19.488, p < 0.001, Display Formula = 0.394; Congruency × Alignment: F(1, 30) = 22.354, p < 0.001, Display Formula = 0.427—and in contrast to the accuracy data, so were all the effects for the “different” trials—congruency: F(1, 30) = 10.924, p = 0.002, Display Formula = 0.267; alignment: F(1, 30) = 4.850, p = 0.035, Display Formula = 0.139; Congruency × Alignment: F(1, 30) = 9.862, p = 0.004, Display Formula = 0.247. A closer inspection of the data revealed that the significant Congruency × Alignment interaction of “different” trials was due to longer reaction times on “different congruent aligned” trials, compared to other conditions. Contrary to the prediction, the correct reaction times revealed an inverse congruency effect for “different” trials (i.e., reaction times are slower for “congruent” compared to “incongruent” trials). For “same” trials, however, results confirm the hypothesis, and a classical congruency effect was found.  

Again, the same conclusions can be drawn from the standard conditions. The alignment effect from “same” trials was significant, t(30) = 4.810, p < 0.001, Display Formula = 0.435, with participants responding faster on “same incongruent misaligned” trials compared to “same incongruent aligned” trials (925 vs. 1080 ms). Furthermore, the effect for “different” trials was also significant, t(30) = 2.995, p < 0.01, Display Formula = 0.230, due to longer reaction times for the “different congruent aligned” trials compared to the “different congruent misaligned” trials (1020 vs. 954 ms).  

Discussion

Overall, the results according to the alternative and standard designs both support the hypothesis of holistic integration of eye and head orientation. The results proved to be straightforward for “same” trials, but not for “different” trials. Holistic processing in the alternative composite design is measured by a significant Congruency × Alignment interaction, which was confirmed for the “same” trials for both accuracy and correct reaction times. For “different” trials, however, results for accuracy revealed a main effect of congruency, but no other effects were significant. Furthermore, although the Congruency × Alignment interaction was significant for correct reaction times, a closer inspection of the results revealed that an inverse congruency effect caused this significant interaction. To conclude, for the face composite design, the overall results seem to confirm the holistic integration of eye and head orientation, as measured by a congruency effect (alternative design) and an alignment effect (standard design). However, although this effect was apparent for “same” trials in accuracy and correct reaction times, it was not straightforward for “different” trials. For a further elaboration on the topic of “same” and “different” trials, see the General discussion. 

In a next step, the inverted face stimuli were tested on a new group of participants, to rule out the possibility that the obtained results were due to effects other than holistic processing. If our hypothesis of holistic processing is correct, rotating the stimuli by 180° should eliminate the perceptual interference from the “same incongruent aligned” condition, since it is hypothesized that observers perceive inverted faces with a smaller perceptual field compared to upright faces (Rossion, 2009). 

Experiment 1B: An inverted face composite design

Methods

Participants

A total of 27 participants, none of whom participated in Experiment 1A, were recruited to take part in Experiment 1B (21 female, six male; age: M = 23 years, SD = 2.9). All had normal or corrected-to-normal vision and received eight euros for participation. 

Stimuli, design, and procedure

Aside from rotation of the original stimuli by 180°, the stimuli, design, and procedure for Experiment 1B were identical to those for Experiment 1A. 

Results

The procedure for removing outliers was identical to that in Experiment 1A, which resulted in the exclusion of 1.7% of the data for further analyses. 

The upright face design (Experiment 1A) revealed different results for “same” and “different” trials, with evidence for holistic processing coming almost exclusively from “same” trials. According to advocates for the standard composite design, this finding logically follows the theoretical assumptions of holistic processing (e.g., Robbins & McKone, 2007; Rossion, 2013; see also the General discussion for a brief discussion on “same” and “different” trials). For the present inverted design, no meaningful main or interaction effects were found for “different” trials, and hence only “same” trials are discussed. Results of Experiment 1B are shown in Figure 4B. 

Accuracy

Since we are mainly interested in the effect of inversion compared to the upright design, only significant effects of orientation will be discussed. The rmANOVA according to the alternative design with congruency, alignment, and orientation as factors revealed no significant main effect of orientation. However, the following interactions were significant: Congruency × Orientation, F(1, 56) = 17.577, p < 0.001, Display Formula = 0.239; Alignment × Orientation, F(1, 56) = 4.595, p = 0.036, Display Formula = 0.076; and Congruency × Alignment × Orientation, F(1, 56) = 4.160, p = 0.046, Display Formula = 0.069. Inspection of the data shows that the Congruency × Alignment interaction is reduced for the inverted stimuli, mainly as a result of better performance on the “same incongruent aligned” trials. However, further analyses on the inverted data alone revealed that that interaction remained significant, F(1, 26) = 31.368, p < 0.001, Display Formula = 0.347, which is indicative of holistic processing of the inverted face stimuli.  

The rmANOVA to address the hypothesis from the standard design (i.e., a significant alignment effect for “same incongruent” trials), with alignment and orientation as factors, revealed a significant main effect of orientation, F(1, 56) = 7.661, p = 0.008, Display Formula = 0.120. More specifically, accuracy on the “same incongruent” trials was better overall for the inverted compared to the upright design (83% vs. 73%). Furthermore, the Alignment × Orientation interaction was significant, F(1, 56) = 5.013, p = 0.029, Display Formula = 0.082. The difference between “aligned” and “misaligned” trials for the “same incongruent” condition showed a reduction from 18% (upright design) to 9% (inverted design). However, a separate paired t test reveals that the latter difference remained significant, t(26) = 4.785, p < 0.001, Display Formula = 0.486. The accuracy data seem to indicate that participants still experienced an interference from the nonrelevant halves when rating the similarity of the relevant halves even when the faces were inverted, although less than when faces were upright.  

Correct reaction times

The rmANOVA for congruency, alignment, and orientation revealed a significant main effect of orientation, F(1, 56) = 5.342, p = 0.025, Display Formula = 0.087. In general, correct reaction times in the inverted design were significantly faster than in the upright design (805 vs. 922 ms). Furthermore, all interactions with orientation proved significant—Congruency × Orientation: F(1, 56) = 22.272, p < 0.001, Display Formula = 0.285; Alignment × Orientation: F(1, 56) = 5.234, p = 0.026, Display Formula = 0.085; Congruency × Alignment × Orientation: F(1, 56) = 8.312, p = 0.006, Display Formula = 0.129. Although reaction times were overall faster in the inverted design, they were particularly reduced in the “same incongruent aligned” trials. However, analyses on the inverted data alone revealed that the Congruency × Alignment interaction remained significant, F(1, 26) = 8.361, p = 0.008, Display Formula = 0.243, again providing evidence for residual holistic processing.  

The rmANOVA with alignment and orientation revealed a significant main effect of orientation, F(1, 56) = 8.856, p = 0.004, Display Formula = 0.137. Correct reaction times on the “same incongruent” trials were significantly faster for the inverted design (831 vs. 1002 ms). Furthermore, the Alignment × Orientation interaction was significant, F(1, 56) = 7.897, p = 0.007, Display Formula = 0.124, indicative of a significant reduction of the difference between “same incongruent aligned” and “misaligned” trials in the inverted design (51 ms) compared to the upright design (155 ms). However, despite this reduction, the alignment effect in the inversion design remained significant, t(26) = 3.623, p = 0.001, Display Formula = 0.335, thus reflecting the observation from the accuracy data that subjects still experienced an influence of the nonrelevant face halves, indicative of holistic processing of the inverted faces.  

Discussion

As predicted, inverting the face stimuli significantly reduced holistic processing, a finding that was confirmed by analyses of the standard and alternative designs. More specifically, the holistic integration of eye and head orientation, as measured in Experiment 1A, decreased when faces were inverted by 180°. Participants experienced less interference from the nonrelevant head orientation when perceiving eye orientation, as was hypothesized. The present study is the first to report holistic integration of gaze cues in faces by means of the composite design. The obtained results in the present study are comparable to composite effects of identity and emotion (e.g., Calder et al., 2000; Goffaux & Rossion, 2006; Soria Bauser, Suchan, & Daum, 2011). However, a separate analysis of the inverted results revealed that rotating the faces by 180° did not completely eliminate the composite effect. It seems that even when the faces were inverted, participants experienced some influence from the nonrelevant head orientation when perceiving the relevant eye orientation. Although this seems to contrast with the prediction (i.e., holistic processing is disrupted through inversion), significant composite effects for inverted faces have been reported before in research on identity perception (e.g., Robbins & Coltheart, 2012; Susilo, Rezlescu, & Duchaine, 2013), as will be discussed further in the General discussion. 

Experiment 2A: Upright body composite design

Methods

Participants

A total of 29 undergraduate students were recruited to participate in Experiment 2A (23 female, six male; age: M = 19 years, SD = 1.6). All participants had normal or corrected-to-normal vision and reported never having experienced face-recognition problems. 

Stimuli and design

The stimulus set consisted of 49 male 3-D figures, created in Poser 8 (Smith Micro Software). Head and body orientation were manipulated independently within a range from −30° to 30° with steps of 10°, but identity remained stable. Aligned composites were then created using Adobe Photoshop by applying a horizontal 2-pixel line at the same height for every figure (i.e., above the shoulders). Misaligned stimuli were obtained by shifting the body halves 75 pixels to the left and right. Similar to Experiment 1, variations in the relevant half (i.e., the head) that were not part of the experimental manipulation (i.e., orientation) were eliminated (e.g., creases in the neck due to rotating the body) by pasting the head of a standard body orientation (10°) on the remaining body orientations. As a result, the perception of different head orientations in “same” trials (i.e., head orientations are identical) could only be attributed to experimentally manipulated factors, such as body orientation, congruency, and alignment. To maximize a potential perceptual interference from nonrelevant halves on the judgment of relevant halves, the same restrictions were imposed as in Experiment 1 on trials where relevant and/or nonrelevant halves differed. 

Procedure

The procedure matched that of Experiment 1. Each participant completed 400 randomized trials (50 per condition), divided into five blocks of 80 trials. On each trial, participants had to decide on the similarity of head orientation of the two sequentially presented target and test stimuli (14.5° × 4°), irrespective of body orientation. 

Results

As in Experiment 1, outliers for reaction times were removed for every participant, and criteria for the upper and lower bounds again differed due to a skewed distribution. This resulted in the removal of 1.4% of the total amount of data. 

Parallel to Experiment 1A, analyses from the alternative and standard designs will be discussed sequentially. Furthermore, “same” and “different” trials will also be discussed separately, because the hypothesis from the standard design explicitly states that no effect should be apparent for the “different” trials. Results for Experiment 2A are visually illustrated in Figure 5A. 

Figure 5

 

Summary results for (top) the upright design (Experiment 2A) and (bottom) the inverted design (Experiment 2B) for the body stimuli. Left (accuracy) and right (correct reaction times) panels depict means for the eight conditions. Error bars represent the standard errors.

Accuracy

For accuracy, main effects of similarity, congruency, and alignment were significant—similarity: F(1, 28) = 11.818, p = 0.002, Display Formula = 0.297; congruency: F(1, 28) = 26.651, p < 0.001, Display Formula = 0.488; alignment: F(1, 28) = 27.161, p < 0.001, Display Formula = 0.492—with significantly fewer errors on “same” trials (91% vs. 82%), “congruent” trials (89% vs. 84%), and “misaligned” trials (88% vs. 85%). Furthermore, the Congruency × Alignment interaction was significant, F(1, 28) = 22.249, p < 0.001, Display Formula = 0.443, as was the three-way Similarity × Congruency × Alignment interaction, F(1, 28) = 23.020, p < 0.001, Display Formula = 0.451. “Same” and “different” trials were again analyzed separately. For the “same” trials, the main effects of congruency and alignment were significant—congruency: F(1, 28) = 29.805, p < 0.001, Display Formula = 0.516; alignment: F(1, 28) = 30.321, p < 0.001, Display Formula = 0.520—as was the Congruency × Alignment interaction, F(1, 28) = 30.470, p < 0.001, Display Formula = 0.521. In sharp contrast to the “same” trials, main and interaction effects for “different” trials were not significant.  

Analyses based on conditions from the standard composite design matched the conclusions from the alternative design. For “same incongruent” trials, the difference between “aligned” and “misaligned” trials was significant, F(1, 28) = 34.322, p < 0.001, Display Formula = 0.551, but this was not the case for “different congruent” trials. More specifically, for “same” trials participants made fewer errors on “misaligned” trials compared to “aligned” trials (94% vs. 81%). The accuracy results point to the holistic integration of gaze cues in bodies, but only for “same” trials.  

Correct reaction times

Results for correct reaction times roughly reflect those for accuracy, with significant main effects of similarity, F(1, 28) = 8.049, p = 0.008, Display Formula = 0.223; congruency, F(1, 28) = 19.309, p < 0.001, Display Formula = 0.408; alignment, F(1, 28) = 10.912, p = 0.003, Display Formula = 0.280; and the Congruency × Alignment interaction, F(1, 28) = 12.124, p = 0.002, Display Formula = 0.302. More specifically, participants were faster on “same” trials (699 vs. 747 ms), “congruent” trials (703 vs. 742 ms), and “misaligned” trials (706 vs. 740 ms). The three-way Similarity × Congruency × Alignment interaction, however, was not significant. Nonetheless, a closer look at “same” and “different” trials independently reveals a significant Congruency × Alignment interaction, F(1, 28) = 13.831, p = 0.001, Display Formula = 0.331, and significant effects for both factors separately—congruency: F(1, 28) = 33.822, p < 0.001, Display Formula = 0.547; alignment: F(1, 28) = 18.012, p < 0.001, Display Formula = 0.391—for “same” trials, but again not for “different” trials.  

Finally, in accordance with the standard design, a significant alignment effect was apparent for “same incongruent” trials, F(1, 28) = 18.722, p < 0.001, Display Formula = 0.401, with faster reaction times for “misaligned” compared to “aligned” trials (693 vs. 789 ms), but not for “different congruent” trials. This is indicative of holistic processing of the body stimuli for “same” trials.  

Discussion

Experiment 2 revealed significant composite effects in the body task with both the alternative and the standard design, indicative of holistic integration of body and head orientation. This adds evidence to the line of research on holistic body processing, and to research focusing on the integration between head and body in particular (Aviezer, Hassin, & Bentin, 2012; Aviezer, Trope, & Todorov, 2012). In behavioral research, the inclusion or exclusion of the head on body stimuli has been a great point of debate, and is even hypothesized to influence the way observers perceive bodies (Brandman & Yovel, 2012; Yovel, Pelc, & Lubetzky, 2010; for fMRI studies, see Cox, Meyers, & Sinha, 2004; Morris, Pelphrey, & McCarthy, 2006; for an fMRI study on congenital prosopagnosia, see Van den Stock, van de Riet, Righart, & de Gelder, 2008). For this reason, many researchers decide to remove or blur the face when studying body perception. In Experiment 2A, we chose to regard the face not as a mere confounding or facilitating feature for body processing but rather as an intrinsic part of full body perception. Our results indeed revealed that body and head are strongly integrated, as has previously been shown in research on emotion perception (Aviezer, Hassin, & Bentin, 2012; Aviezer, Trope, & Todorov, 2012). However, in line with the results of Experiment 1A, although the overall results revealed a significant congruency effect, this was true only for “same” trials, not “different” trials. This result will be further elaborated in the General discussion. 

As for Experiment 1A, the upright body design was complemented with an inverted design. If the significant composite effects from Experiment 2A are the result of holistic body processing, the effects should be eliminated, or at least strongly reduced, when the stimuli are rotated 180°. 

Experiment 2B: Inverted body composite design

Methods

Participants

A total of 25 participants, none of whom participated in Experiment 2A, were recruited to take part in Experiment 2B (18 female, seven male; age: M = 24 years, SD = 5.4). All had normal or corrected-to-normal vision and received eight euros for participation. 

Stimuli, design, and procedure

Aside from rotation of the original stimuli by 180°, the stimuli, design, and procedure for Experiment 2B were identical to those of Experiment 2A. 

Results

The procedure to remove outliers resulted in exclusion of 1.1% of the data from further analyses. Analogous to Experiment 1B, only relevant effects for “same” trials will be discussed, as results of the upright and inverted designs again revealed no significant effects for “different” trials. Results of Experiment 2B are visually illustrated in Figure 5B. 

Accuracy

The rmANOVA with congruency, alignment, and orientation as factors revealed that all interaction effects with orientation were significant—Congruency × Orientation: F(1, 52) = 7.749, p = 0.007, Display Formula = 0.130; Alignment × Orientation: F(1, 52) = 5.291, p = 0.025, Display Formula = 0.092; Congruency × Alignment × Orientation: F(1, 52) = 9.558, p = 0.003, Display Formula = 0.155. The main effect of orientation, however, was not significant. Overall, accuracy in the inverted design was slightly worse compared to the upright design, except for the “same incongruent aligned” condition, for which accuracy was somewhat higher, resulting in the significant three-way Congruency × Alignment × Orientation interaction. However, a separate analysis on the inverted data shows that the Congruency × Alignment interaction remained significant, F(1, 24) = 5.386, p = 0.029, Display Formula = 0.183.  

Analyses in accordance with the standard design confirm these results. The main effect of orientation was not significant, but the Orientation × Alignment interaction was, F(1, 52) = 9.311, p = 0.004, Display Formula = 0.152, indicative of a significant reduction in the difference between “same incongruent aligned” and “misaligned” trials for the inverted design (5%) compared to the upright design (13%). A paired-samples t test, however, revealed that this difference remained significant for the inverted stimuli, t(24) = 3.781, p = 0.001, Display Formula = 0.373, which is evidence for residual holistic integration of the body stimuli when inverted.  

Correct reaction times

The results of the reaction-time data reflect the accuracy data. The main effect of orientation was not significant, but the interaction effects with congruency and alignment were—Congruency × Orientation: F(1, 52) = 16.592, p < 0.001, Display Formula = 0.242; Alignment × Orientation: F(1, 52) = 4.613, p = 0.036, Display Formula = 0.081—as was the three-way Congruency × Alignment × Orientation interaction, F(1, 52) = 5.228, p = 0.026, Display Formula = 0.091. As for accuracy, the overall reaction times were slightly slower in the inverted design, except in the “same incongruent aligned” condition, which was faster compared to the upright design. In contrast to the accuracy data, however, the Congruency × Alignment interaction was not significant for the inverted data.  

The rmANOVA with alignment and orientation as factors for the “same incongruent” condition show a significant Alignment × Orientation interaction, F(1, 52) = 6.181, p = 0.016, Display Formula = 0.106, but no main effect of orientation. This significant interaction indicates that the difference between “aligned” and “misaligned” trials for the “same incongruent” condition is significantly reduced in the inverted design (31 ms) compared to the upright design (96 ms). However, a separate paired-samples t test shows that the difference remains significant, t(24) = 2.882, p = 0.008, Display Formula = 0.257, reflecting the observation in the accuracy data that participants still experienced an influence from the nonrelevant body halves when perceiving the relevant halves.  

Discussion

Results for the inverted body design show the same trend as for the inverted face design. Significant interactions with orientation point to a reduction in the composite effect when stimuli were rotated by 180°, but separate analyses on the inversion data seem to indicate that some amount of influence from the nonrelevant body orientation on the perception of head orientation remains. However, unlike all previous experiments in the present study, the correct reaction times for the inverted body design point to contrasting conclusions for the alternative and standard designs. According to the alternative design, the nonsignificant Congruency × Alignment interaction for the inverted data is indicative of a complete disruption of holistic processing. The analysis of the standard design, however, would conclude otherwise, since the alignment effect for the “same incongruent” trials was still significant for the inverted stimuli, pointing to a residual interference from the nonrelevant halves on the perception of the relevant halves. This specific contradiction is addressed in the General discussion, together with a broader comparison between the two designs. 

General discussion

To summarize, we can conclude that gaze cues in faces and bodies are integrated holistically, as evidenced by significant composite effects for upright stimuli. Furthermore, this finding is confirmed by the significant reduction in composite effects when the stimuli are inverted, a manipulation that is presumed to disrupt holistic processing. However, despite a significant reduction in composite effects for faces and bodies, the effects still proved significant for the inverted stimuli. In the remainder of the General discussion, the obtained composite effects for the upright and inverted designs will be discussed and compared to previous research on identity and emotion perception. Next, we elaborate further on the similarities and differences between results from the alternative and standard designs. Lastly, implications for future research with the gaze-based composite design will be highlighted. 

The inverted composite design

Although priming and attention studies have already pointed toward the importance of cues other than eye direction in estimating eye gaze (e.g., Hietanen, 1999, 2002; Pomianowska et al., 2011), the composite design offers a highly controllable method to systematically investigate the effect of head and body orientation on gaze estimation. Furthermore, gaze perception is one aspect within a more elaborate model of face perception (Haxby, Hoffman, & Gobbini, 2000), of which many other aspects have proven to yield significant composite effects (e.g., identity, emotion). With our results, we provide evidence for holistic processing of another piece in the model. Furthermore, it seems that the obtained sizes of the composite effects in the upright designs are comparable to previous research on holistic processing of identity and emotion. Taking the difference between “same incongruent aligned” and “misaligned” trials as a marker for holistic processing (within the framework of the standard design), our results with the face stimuli (18% accuracy and 155-ms reaction-time difference) fall well within the range of the sizes found for identity and emotion composites in other studies (Calder et al., 2000; Durand, Gallay, Seigneuric, Robichon, & Baudouin, 2007; Goffaux & Rossion, 2006; Mondloch & Maurer, 2008; Robbins & Coltheart, 2012; Robbins & McKone, 2007; Rossion & Boremanse, 2008; Soria Bauser et al., 2011; Susilo et al., 2013; White, 2000). Studies have reported composite effects that range from 5% to 22% for accuracy, and from 60 to 120 ms for reaction time. Although studies with bodies are still scarce, the obtained effects in our body design (13% accuracy and 96-ms reaction-time difference) seem to be relatively large compared to those in other studies (Aviezer, Trope, & Todorov, 2012; Robbins & Coltheart, 2012; Soria Bauser et al., 2011; Willems et al., 2014), which have reported differences that range from 1% to 10% for accuracy and 34 ms for reaction time (only one study reported reaction-time data; see Willems et al., 2014). 

It thus seems that, for the upright design, the results from our gaze stimuli are of a comparable size to those in previous research on identity and emotion. However, the question emerges whether the same conclusion can be drawn for the inverted results. It is hypothesized that rotating faces by 180° disrupts holistic processing (e.g., Rossion & Boremanse, 2008). More specifically, the upright composite effect is expected to be eliminated for inverted stimuli. In the present experiments, however, this was not the case. For faces, the composite effect for the inverted stimuli was significantly reduced compared to the upright stimuli but remained significant (9% and 51 ms). This finding points to some residual holistic processing for inverted faces, which stands in sharp contrast to the theoretical assumptions of holistic processing but is nonetheless not a new finding in research with the inverted composite design. More specifically, whereas some researchers have found complete elimination of the identity composite effect for inverted faces (Carey & Diamond, 1994; McKone, 2008; McKone et al., 2013; Mondloch & Maurer, 2008; Robbins & McKone, 2007; Soria Bauser et al., 2011; Susilo et al., 2010; Young et al., 1987), this is not the case in all studies (Goffaux & Rossion, 2006; McKone et al., 2013; Robbins & Coltheart, 2012; Rossion & Boremanse, 2008; Susilo et al., 2013). Furthermore, to our knowledge only two studies have conducted the inverted composite design for bodies (Robbins & Coltheart, 2012; Soria Bauser et al., 2011). Robbins and Coltheart (2012) reported significant upright and inverted composite effects for vertically divided bodies (i.e., participants compared left and right body halves) but a significant composite effect for upright bodies only when they were horizontally divided (i.e., participants compared top and bottom halves). The results of Soria Bauser et al. (2011) are difficult to compare to the present study, since those authors did not find a significant composite effect for upright or inverted bodies. In the next section, we consider several mechanisms that could explain our significant composite effects for inverted faces and bodies. 

In an article by McKone et al. (2013), several possible methodological explanations for a significant inverted composite effect were investigated, from participant race and gender to stimulus gap effects and the chosen relevant stimulus half. The researchers concluded that no single mechanism or straightforward interaction could explain the significant inversion effects, and we likewise believe that none of the aforementioned participant and stimulus factors could completely account for the obtained results in the current experiments. A second possible mechanism to explain the results can be found in a generic response conflict, hypothesized to be inherent to the congruency design but unrelated to perceptual processing (Rossion, 2013), with the well-known Stroop effect as primary example (Stroop, 1935). Although the “different” trials did not reveal any relevant effects for the hypothesis on holistic processing, they did show a robust congruency effect for accuracy (i.e., performance on congruent trials is better than performance on incongruent trials). This effect did not disappear with misaligning or inverting the stimuli, which is reason to believe that the effect did not depend on holistic processing. This effect was small, however, and only significant for the face design; we thus propose a third mechanism as the more likely explanation. 

Research on the perception of direct versus averted gaze reports that inversion affects sensitivity to gaze discrimination and that this effect is unrelated to the presence or absence of the whole facial context (Jenkins & Langton, 2003; Schwaninger, Lobmaier, & Fischer, 2005; Vida et al., 2013). Furthermore, the effect of head rotation on the computation of gaze direction is shown to be similar for upright and inverted faces (Langton, Honeyman, & Tessler, 2004; Otsuka, Mareschal, Calder, & Clifford, 2014; Otsuka, Mareschal, & Clifford, 2015). More specifically, Langton et al. (2004) have shown that the internal (i.e., nose angle) and external (i.e., head shape) face features that are hypothesized to be important for the computation of upright gaze direction elicit a comparable effect for inverted faces. Both features can be defined as a deviation from vertical or bilateral symmetry, an aspect that previous research has shown can be extracted based on low-level features and is orientation independent (Wilson, Wilkinson, Lin, & Castillo, 2000). Thus, it is possible that the interaction of these features—and head shape in particular, since nose angle was barely visible in our stimuli—with eye orientation could account for the significant inverted composite effects. Since bodies are also highly bilaterally symmetric, we believe that the symmetry hypothesis can be extended to this stimulus class as well. To summarize, the composite effects for the inverted design remain significant, which in our opinion could be the result of orientation-independent mechanisms, as already outlined. Nonetheless, the significant composite effects for the upright stimuli and the significant reduction in composite effects for the upright compared to the inverted stimuli point to holistic processing of gaze cues in faces and bodies. 

Alternative versus standard design

In this section, we aim to compare the results according to the alternative design with results obtained from standard-design analyses. Our discussion on this particular point of comparison can be short, since conclusions were generally similar from both designs, except for the correct reaction times for the inverted bodies. Whereas advocates of the alternative design would conclude that no holistic processing for inverted bodies is apparent, as evidenced by a nonsignificant Congruency × Alignment interaction, results from the standard design suggest otherwise, since the difference between “same incongruent aligned” and “misaligned” trials remained significant for the inverted bodies. However, in the rmANOVA from the alternative design, the main effect of alignment was also significant, F(1, 24) = 8.927, p = 0.006, Display Formula = 0.271, but it did not depend on congruency (i.e., participants were generally slower on “aligned” trials compared to “misaligned” trials for both “incongruent” and “congruent” trials, although the effect was not significant for “congruent” trials), thus resulting in the absence of an interaction effect.  

In the current study, the analyses from the standard and alternative designs led to only minor differences. Although we believe that the alternative variant offers a more balanced design by including a congruency factor, we suggest that researchers should beware of interpretations based on the Congruency × Alignment interaction. It is common in research with the alternative design to calculate sensitivity measurements to discriminate between true discriminability (d′) and response bias (criterion c; e.g., Cheung, Richler, Palmeri, & Gauthier, 2008; DeGutis, Wilmer, Mercado, & Cohan, 2013; Gauthier & Bukach, 2007; Horry, Cheong, & Brewer, 2015; Meinhardt et al., 2014; Richler, Cheung, & Gauthier, 2011; Richler, Mack, Palmeri, & Gauthier, 2011). However, we did not follow this tradition in the current study. In our opinion, d′ gives a distorted image of the Congruency × Alignment interaction, since we cannot interpret it similarly for “same” and “different” trials, which are merged in d′ and c. More specifically, we hypothesize that the Congruency × Alignment interaction is defined as an interference effect for “same” trials but a facilitation effect for “different” trials. Whereas a significant interaction for “same” trials is caused by worse performance for “same incongruent” trials compared to “same congruent” and “same misaligned” trials (i.e., different nonrelevant halves interfere with the perception of identical relevant halves), a significant interaction for “different” trials is hypothesized to be caused by better performance on “different congruent” trials compared to “different incongruent” and “different misaligned” trials (i.e., different nonrelevant halves amplify the perception of different relevant halves). For d′, the interference effect from “same” trials and the facilitation effect from “different” trials are merged, which would not be a problem if both effects are equally important for holistic processing. However, the relative importance of both effects to assess holistic processing is not equal. In line with the standard design, we argue that the presence of perceptual interference for “same” trials is crucial to assess holistic processing, whereas a facilitation for “different” trials is not. The definition of holistic processing in terms of the standard composite design states that holistic processing interferes with adequate perception of identical relevant halves when they are paired with different nonrelevant halves. Different trials, however, do not give rise to such an interference effect, since identical nonrelevant halves cannot make different relevant halves appear more similar: “One small discordance overweighs a multitude of similarities and suggests a general unlikeness” (Galton, 1883, p.3). This is also the reason why a mere facilitation effect from different nonrelevant halves on the perception of different relevant halves is not sufficient for holistic processing—only one small difference between faces is enough to create a different configuration and, hence, a qualitatively different face. 

This could potentially be a problem, since a significant Congruency × Alignment interaction in d′ could be solely due to a significant facilitation effect, without the interference effect from “same” trials. This has proven to be the case in some studies (Horry et al., 2015; Liu & Behrmann, 2014; Liu, Hayward, Oxner, & Behrmann, 2014) but unfortunately hampers an adequate interpretation of the results. Although we will not elaborate on the reason we think some researchers find only a facilitation effect, we believe that in the future, interpretation on the basis of d′ should be made with caution; preferably, “same” and “different” trials should be analyzed separately, in accordance with the standard design. 

Implications for future research

By creating a gaze-based composite design, we hope to further facilitate the systematic study of comparisons between different face or body aspects on the one hand and between clinical populations on the other. More specifically, as mentioned in the Introduction, inadequate eye-gaze following (i.e., joint attention) during the first months after birth has proven to be an important predictor for ASD (e.g., Ashwin et al., 2015; Baron-Cohen et al., 1995). Furthermore, although it is not a primary symptom, people with ASD are said to experience face-processing difficulties (e.g., Gauthier et al., 2009). The underlying mechanism for a face-processing impairment in ASD has not yet been fully uncovered, but a lack of holistic processing in favor of a more piecemeal analysis is hypothesized to be a good candidate (for a review, see Watson, 2013). Research on ASD using the composite design has provided evidence for this hypothesis (Gauthier et al., 2009; Teunisse & de Gelder, 2003; but see Nishimura et al., 2008), but until now only identity-composite designs have been conducted. Administering the current composite design, based on gaze perception, in a population with ASD could hence be interesting for at least two reasons. First, it explores an aspect of face processing regarding which researchers share the consensus that it is impaired in people with ASD (i.e., gaze perception); second, this exploration is operationalized using the composite design, which measures a hypothetical impaired processing style in ASD (i.e., holistic processing). 

Whereas a deficit in eye-gaze following is a core symptom for ASD and face-processing impairment is not (or is even secondary to deficits in social interaction), the scenario is reversed for people with prosopagnosia, a problem with face recognition. Prosopagnosia indicates a significant impairment in recognizing familiar faces on a daily basis, either acquired (e.g., after brain injury) or congenital. The main hypothesis states that the face-recognition impairment in prosopagnosia is due to a lack of holistic processing, and many behavioral and neurological studies seem to confirm this hypothesis (Avidan, Tanzer, & Behrmann, 2011; Busigny et al., 2010, 2014; Liu & Behrmann, 2014; Palermo et al., 2011; Ramon, Busigny, & Rossion, 2010; Van Belle, De Graef, Verfaillie, Busigny, & Rossion, 2010; Verfaillie, Huysegems, De Graef, & Van Belle, 2014; but see Le Grand et al., 2006; Schmalzl, Palermo, & Coltheart, 2008; Susilo et al., 2010). However, while research with prosopagnosia mainly focuses on the processing of identity, investigating other face (and even body) aspects in prosopagnosia could shed more light on the condition as a whole. More specifically, studies indicate a different processing style for the eye region in people with prosopagnosia compared to control groups (Caldara et al., 2005; DeGutis, Cohan, Mercado, Wilmer, & Nakayama, 2012; Orban de Xivry, Ramon, Lefèvre, & Rossion, 2008). Furthermore, observations in our lab reveal compensatory strategies in a population with congenital prosopagnosia that revolve around eye gaze. For example, many subjects with congenital prosopagnosia indicate that they constantly scan the environment to detect direct eye gaze and subsequently anticipate signs of recognition in other people, such as a smile. On the other hand, some subjects indicate deliberately avoiding eye contact, to prevent embarrassing situations (Yardley, McDermott, Pisarski, Duchaine, & Nakayama, 2008). The holistic processing of gaze perception by means of the composite design thus seems an interesting candidate to assess for a population with prosopagnosia. 

Conclusion

The significant composite effects for the face and body design support the hypothesis of holistic integration of gaze cues in the perception of faces and bodies. For faces, we provide evidence for holistic processing of another piece in the face-perception model besides identity, emotion, and gender. For bodies, we add evidence to the line of research stating that gaze cues in full bodies are also processed in a holistic manner, alongside identity, posture, and emotion. In future research, this orientation-based composite design may facilitate systematic study on the comparison between healthy and clinical populations, such as people with ASD or prosopagnosia. 

Acknowledgments

We are grateful to Sam Willems for his comments on this research project. LV was supported by Grant FWO G.0810.13 from the Fund for Scientific Research Flanders (awarded to KV). 

Commercial relationships: none. 

Corresponding author: Karl Verfaillie. 

Email: . 

Address: Laboratory of Experimental Psychology, Katholieke Universiteit Leuven, Leuven, Belgium.