Contextual knowledge provided by a movie biases implicit perception of the protagonist.

We are constantly categorizing other people as belonging to our in-group ('one of us') or out-group ('one of them'). Such grouping occurs fast and automatically and can be based on others' visible characteristics such as skin color or clothing style. Here we studied neural underpinnings of implicit social grouping not often visible on the face, male sexual orientation. A total of 14 homosexuals and 15 heterosexual males were scanned in functional magnetic resonance imaging while watching a movie about a homosexual man, whose face was also presented subliminally before (subjects did not know about the character's sexual orientation) and after the movie. We discovered significantly stronger activation to the man's face after seeing the movie in homosexual but not heterosexual subjects in medial prefrontal cortex, frontal pole, anterior cingulate cortex, right temporal parietal junction and bilateral superior frontal gyrus. In previous research, these brain areas have been connected to social perception, self-referential thinking, empathy, theory of mind and in-group perception. In line with previous studies showing biased perception of in-/out-group faces to be context dependent, our novel approach further demonstrates how complex contextual knowledge gained under naturalistic viewing can bias implicit social perception.

Introduction

Picture yourself walking in the city center and accidentally seeing a male person who was just yesterday interviewed on TV about his daily experiences as a homosexual person. Would you go to him to express your appreciation of his openness, or would you rather look away to avoid any possibility of interaction? Would your behavior depend on your own sexual orientation?

We tend to join, trust and like group members with whom we share similarities (Yuki ). Although the similarity might be only illusory and temporary, it does bias our social behavior (Hornsey, 2008). Favoring people similar to us is probably rooted in biological survival mechanisms (Allport, 1954). We are better in predicting the behavior of in-group members (Bruner, 1957), which facilitates collaboration. However, in-group favoritism is often accompanied with negative attitude toward out-group members. People are also less empathic to the pain and suffering of out-group individuals, which can support intergroup hostility (Xu ; Avenanti ).

Intergroup processes are based on a multitude of psychological mechanisms with complex neural basis (for a review, see Cikara and Van Bavel, 2014). Investigating why we favor one person over another faces many challenges, since critical neurocognitive mechanisms involved in this behavior are to a large extent automatic and implicit. Furthermore, the parallel implicit and explicit processing can lead to inconsistent findings when different paradigms and stimuli are used (for a review, see Amodio, 2014). For example, consequences of implicit prejudices can be inhibited by conscious suspension (Cunningham ; Payne, 2005) and thus tolerance at conscious explicit level does not necessarily mean that we do not have implicit prejudices (Olsson, 2005). Furthermore, low social status of oppressed individuals can cause favoring of high-status groups leading to out-group favoritism and bias against one’s own group (Allport, 1954; Jost and Hunyady, 2003; Jost ).

Pictures of faces have been extensively used as stimuli in investigating implicit neural processes during social perception. We continuously evaluate our social environment based on what kind of faces we see (for a review, see Todorov ). It has been shown that a brief 32–100 ms exposure to a face is sufficient for formation of a social judgment (Bar ; Ballew and Todorov, 2007; Porter ; Borkenau ; Rule and Ambady, 2009; Todorov , 2010). Face stimuli have been used to reveal intergroup bias among white as well as black participants (Hart ; Phelps ; Cunningham ; Avenanti ; Kubota ). However, most social groups, such as groups based on sexual orientation, religion or political views, are often characterized by factors other than salient facial features (Tajfel and Turner, 1979).

Recent studies have shown that face perception is dependent on contextual cues presented with the face. For example, verbal information given about a face (Schwarz ), visual background of the face (Righart and De Gelder, 2008) and seeing the face alone or with other faces (Mumenthaler and Sander, 2012) are all factors that can affect how we perceive a face. Such context dependency calls for stimuli that can better simulate real-life interactions involving faces (for a review, see Wieser and Brosch, 2012). Indeed, it has been shown that we are better at recognizing faces and individuals when presented with moving images approaching real-life perception in comparison to static images (O’Toole ).

A growing body of literature has demonstrated that movies can successfully be used as stimuli to study real-life-like social perception (Hasson, 2004; Malinen ; Jääskeläinen ; Wilson ; Lahnakoski ; Lähteenmäki ; Saarimäki ). Movie narratives can temporarily transport the viewers to the world of the protagonist (Hall and Bracken, 2011) and introduce them to complex emotional and contextual knowledge of the protagonist in a short time. Furthermore, watching the same movie synchronizes perception and underlying neural activity across viewers (Hasson, 2004; Hasson ; Jääskeläinen ; Bacha-Trams ) and makes them similar in interpretation of the protagonists and their behavior. Therefore, having subjects to view the same movie can be used to decrease variance in their perception, for example, in studying prejudices (Corrigan ).

Here we used a movie to reveal implicit brain activity, measured with functional magnetic resonance imaging (fMRI), triggered by the face of a movie character. Brains of homo- and heterosexual males were scanned with functional MRI while watching a movie about a homosexual man, whose face was also presented subliminally before and after the movie. Based on recent studies having shown negative attitude of heterosexuals toward homosexuals (Jellison ; Cullen and Barnes-Holmes, 2008) and in-group favoritism in both heterosexual and homosexual individuals (Banse ; Jellison ), we hypothesized that knowledge about the character’s sexual orientation will bias the implicit neural response to his face in homosexual vs heterosexual subjects. More specifically, we expected homosexuals to show increased activity in regions associated with empathy and in-group perception such as temporal parietal junction (TPJ) (Saxe, 2006; Decety and Lamm, 2007; Decety ) and medial prefrontal cortex (mPFC) (Vogeley and Fink, 2003; Amodio and Frith, 2006; Mitchell ; Krueger ; Van Overwalle, 2009; Morrison ). On the other hand, based on behavioral study showing prejudice of heterosexual subjects toward homosexuals to be associated with feelings of disgust and pity (Cottrell and Neuberg, 2005), we expect the heterosexual subjects to have increased activity in the insula. Although insula activation has been reported in social pain and empathy and is related to various types of emotional processing, it is also sensitive to disgust (for a review, see Vicario ). More specifically, we expect an activation in the insular frontal operculum that is specifically associated with the processing of disgust (Calder ; Adolphs ; Carr ; Wicker ; Jabbi , 2008). As another hypothesis, though it is not an indicator of implicit bias, we expected our homosexual subjects to self-report increased identification and feeling of closeness with the movie character after the movie reveals he is homosexual. Conversely, we expected heterosexual subjects to show decreased identification with the movie character after the movie reveals him as homosexual (Petta and Walker, 1992; Doosje ; Smith and Tyler, 1997; Spears ).

Material and methods

Participants

A total of 29 right-handed males (15 heterosexuals, mean age, 26; 14 homosexuals, mean age, 28; age range of participants, 20–46), volunteered for the experiment. All participants were Finnish speakers. The subjects reported neither history of neurological or psychiatric diseases nor medications affecting the central nervous system. The fMRI scanning was conducted between 10 am and 2 pm. The study protocol was approved by the Research Ethics Committee of the Aalto University, and each subject signed an informed consent form prior to participation.

Experimental design

During the fMRI scanning, the participants watched a 20 min movie consisting of scenes from the movie ‘Priest’ (Directed by Antonia Bird, 1994). The movie tells the story of a man struggling between his devotion to serve as a Roman Catholic Christian priest and his desire to be loved by another man, something that is forbidden for a Catholic priest. The movie was edited so that initially the priest appears to be heterosexual and after exactly 10 min the events in the movie reveal that he is actually homosexual. The structure of the movie was tested before the experiment on a pilot audience (9 male subjects, age 21–27) who unanimously indicated that the homosexuality of the character was identified only after he enters the gay bar (at 10 min).

After scanning, the subjects were asked to report how close they felt to the character in the first and in the second part of the film. Participants were not informed of the movie’s content prior to the experiment. A professional film-maker (the first author of this paper) edited the original movie to assure that the stimulus has a story that flows naturally with an engaging plot line. The subjects reported that they had not seen the original film before and were not familiar in the type of experiment they participated in.

To test if knowledge of the sexual orientation of the priest modulates implicit perception of his face, we measured brain hemodynamic responses to 40 ms images of the character’s face before and after viewing the movie. During the fMRI the subjects were asked to fixate at a mark in the center of the screen and watch a 4 min stimulus sequence that contained white noise and images of the character’s face (see details of the stimuli below). The participants were not informed of the nature of the subliminal stimuli and were instructed with the following slide (written in Finnish): ‘You will see a calibration clip. This clip is meant for calibrating the MRI scanner for your responses. The clip is only four minutes long and will look like white noise on a TV screen. Please keep your eyes fixated at the mark in the center of the screen until notified otherwise.’

The 4 min subliminal stimuli contained 16 blocks of 15 s each. Blocks were of two types: ‘face’ and ‘objects’. Only blocks with the character’s face were analyzed; the objects block served as a wash-out period. In pilot experiments, we used face blocks containing only faces. However, 4 out of 10 pilot participants could see that there were images of faces embedded in the stimulus. This could probably be due to higher sensitivity to face perception in some participants who therefore required further masking (Lähteenmäki ). We therefore decided to additionally mask the faces by inserting 40 ms images of objects between the faces that showed to be successful in our piloting. Object blocks consisted only of objects (Figure 1).

Fig. 1

Illustration of the experimental paradigm. ‘Upper panel A and C’. Pictures of the face of the priest presented subliminally. ‘Upper panel B’. A 20 min movie telling a story of a homosexual priest. ‘Lower panel’: first six pictures of face and object blocks.

The face block consisted of 10 40 ms black-and-white images of faces presented at 1460 ms inter-stimulus interval filled with white noise. Each face image was followed by an object image. The object images were randomly selected from a set of 197 images at http://natural-scenes.cps.utexas.edu/db.shtml; (Geisler and Perry, 2011). A 15 s object block followed each face block. The two block types were presented alternatingly, starting with the object block.

Experimental procedure

The stimuli were delivered during the fMRI scanning using the Presentation software (Neurobehavioral Systems Inc., Albany, CA, USA). The stimuli were back-projected on a semitransparent screen using a Panasonic PT-DZ110X projector (Panasonic Corporation, Osaka, Japan), and from there via a mirror to the subject. Auditory stimulation was delivered through Sensimetrics S14 insert earphones (Sensimetrics Corporation, Malden, MA, USA) via plastic tubes through porous EAR-tip (Etymotic Research, ER3, IL, USA) earplugs. Sound intensity was adjusted individually to be comfortable but loud enough to be heard over the scanner noise.

After the fMRI scan, we asked the participants to indicate how close they felt to the character in the first part of the film (they did not know the character is homosexual) as well as in the second part of the film (they knew the character is homosexual). This measurement was done to assure that the change in identification with the character is a factor of knowing the character’s sexual orientation (vs before) and not due to other possible factors that were present all along in the movie such as the character’s age, nationality or occupation as a Catholic priest. The measurement of the identification with the character was done on a computer using an online form with a sliding continues measurement from 0 (no identification) to 1 (100% total identification). We subtracted the score of the second part from the first part of the movie to quantify the change in identification. For example, if a subject’s identification level with the character in the first part of the movie is at 0.47 points and in the second part at 0.63 points, then the identification increased by 0.16 points. For this measurement two subjects were excluded from analysis for not providing their results for this section. Identification scores could have been obtained right after each part of the film. However, we wanted to exclude the possibility that such an inquiry would have influenced the viewing experience of the second part of the film. In our previous work, we have observed that this type of retrospective self-reporting can accurately capture at least experienced humorousness during watching of comedy clips, probably since re-seeing the clip serves as a memory cue (Jääskeläinen ). Furthermore, we conducted an additional behavioral measurement to assess the participants’ unconscious bias to hetero- or homosexuality using the Implicit Association Task (IAT) using presentation software (details below). The IAT measured differences in reaction time (RT) in associating positive and negative words to images representing homosexual or heterosexual sexual orientation, as measurement of implicit bias (Greenwald ). In our experiment, IAT scores were collected to have an estimate of implicit bias of the subjects to homo- and heterosexuality.

fMRI data acquisition and preprocessing

MR imaging was performed at the Advanced Magnetic Imaging Centre at Aalto University. Images were acquired with a 3T Siemens MAGNETOM Skyra (Siemens Healthcare, Erlangen, Germany), using a standard 20-channel receiving head–neck coil. Anatomical images were acquired using a T1-weighted magnetization prepared rapid acquisition gradient echo (MPRAGE) pulse sequence (TR 2530 ms, TE 3.3 ms, TI 1100 ms, flip angle 7°, 256 × 256 matrix, 176 sagittal slices, 1 mm3 resolution). Whole-brain functional data were acquired with T2*-weighted echo-planar imaging (EPI) sequence sensitive to the BOLD contrast. Imaging parameters for the functional images were TR 1700 ms, TE 24 ms, flip angle 70°, FOV 217.6 mm, 64 × 64 matrix, 4.0 mm slice thickness with 1 mm gap between slices, 29 oblique slices acquired in interleaved ascending order covering the whole brain, resolution 3.4 × 3.4 × 4.0 mm.

Standard fMRI preprocessing steps were applied using the FSL software (www.fmrib.ox.ac.uk, version 5.0.9) and custom MATLAB code (BRAMILA pipeline v2.0, available at https://version.aalto.fi/gitlab/BML/bramila). After slice timing correction, the functional images were realigned to the middle scan by rigid body transformations with MCFLIRT to correct for subject motion. Non-brain matter from functional and anatomical images was removed using Brain Extraction Tool (Smith, 2002). Functional images were registered to the MNI152 standard space template (Montreal Neurological Institute) with 2 mm resolution. The transformation parameters were acquired by first calculating transformations from structural to standard space (12 degrees of freedom) and from functional to structural space (9 degrees of freedom), and then concatenating these parameters. Next, these transformation parameters were used to co-register functional datasets to the standard space. Both registration steps were performed using FLIRT (Jenkinson, 2002). To remove scanner drift, a 240 s long Savitzky–Golay filter (Çukur ) was applied. To control for motion and physiological artefacts, EPI time series were cleaned with linear regression using 24 motion-related signals, signal from deep white matter, ventricles and cerebral spinal fluid locations as described in Power ). Additional spatial smoothing step with a Gaussian kernel of full width at half maximum (FWHM) 6 mm was also applied. The neural response to the face of the character was computed using a general linear model (GLM) implemented in FSL with film_gls with default parameters (i.e. prewhitening of time series).

GLM was run separately for the subliminal stimuli before and after the movie; the model contained two regressors for the face block (presentation of text and presentation of subliminal faces block and their corresponding temporal derivate) as well as two more regressors for the object block (presentation of text and presentation of subliminal object block and their corresponding temporal derivate). Regressors were convolved with the canonical hemodynamic response function. To analyze the differences between the homosexual and heterosexual group we run a two-way mixed-effect analysis of variance (ANOVA), which compares the differences in BOLD responses between the face after the film given the response before the film as a baseline. As a control analysis and to test if the results are unique for the face block we also re-ran the analysis using the object block. Finally, to correct for the multiple comparisons, we used a permutation-based non-parametric approach as implemented by FSL randomise with cluster-forming threshold at T-value = 3 and 5000 permutations. Visualization threshold were set at corrected P < 0.05. Significance threshold were set at cluster corrected P < 0.05.

The IAT

IAT was used to assess the subjects’ implicit bias toward hetero- and homosexuals, measured via changes in RT to the different IAT conditions. During the IAT, the subjects were presented with images that represent heterosexual/homosexual orientation in conjunction with positive/negative words. The IAT included two random conditions (associate positive words with homosexuality and negative words with heterosexuality and for the other condition the other way around). Differences in RT caused by a certain association condition were interpreted as an implicit bias. Positive words in IAT were the following: joyful, beautiful, marvelous, wonderful, pleasure, glorious, lovely and superb. The negative words were the following: agony, terrible, horrible, humiliate, nasty, painful, awful and tragic. The heterosexual/homosexual orientation images were downloaded from http://www.millisecond.com/download/library/IAT, which has been previously used to address similar topics (Biele and Grabowska, 2006; Rowatt ). Within each condition the order of words and images were random. The subjects were instructed to respond as fast and as accurately as possible. The IAT was preceded by a 2 min training phase that assured that the subjects fully understood the instruction and the task.

Results

Figure 2 depicts the normalized IAT scores of homo- and heterosexual subjects. Mean scores were significantly different (homosexuals = 0.3, heterosexuals = −0.26, t = 3.72, P < 0.01). Furthermore, IAT scores of each group were significantly different from zero, (heterosexual P = 0.043 and homosexual P = 0.0059). Figure 3 depicts the identification scores of the hetero- and homosexual subjects from the first and second part of the movie, a line connects the mean scores. The mean score in the heterosexual subjects is lower in the second part of the movie but higher in the homosexual subjects. As depicted in Figure 3, the pair-wise comparisons of homosexual and heterosexual identification (IDN) scores (i) at first part of the film, and (ii) at second part of the film showed significant interaction between the groups and film parts (F = 5.72, P = 0.02). Changes in IDN and IAT scores showed a significant positive correlation (r = 0.41, P = 0.033).

Fig. 2

Mean IAT scores (dotted lines, homosexuals = 0.3, heterosexuals = −0.26). Y-axis indicates the level of pro-homo/heterosexuality bias from 1 (very strong) to −1 (very weak). Red lines indicate medians (homosexuals = 0.45, heterosexuals = −0.20). Whiskers indicate the range of the scores.

Fig. 3

Identification scores of all subjects in the first and second part of the film. Lines connect the mean values.

A mixed-effect ANOVA analysis revealed significantly larger BOLD signals in the homosexual vs heterosexual subjects in response to the face after the movie viewing in mPFC, bilateral frontal pole (FP), anterior cingulate cortex (ACC), right temporal parietal junction and bilateral superior frontal gyrus (sFG), as depicted in Figure 4. Although the peak activation coordinates in mPFC is slightly dorsal, the activation in the present study did spread into the ventral medial regions. Furthermore, our control analysis using the object block (instead of the face block) failed to show any significant effects. We failed to see any significant activation in the reverse contrast heterosexuals vs homosexuals. The activation in insula did not survive cluster correction. Unthresholded result map can be found in NeuroVault at https://neurovault.org/collections/ELLZKWSV/images/61868. Other unthresholded maps of additional analysis that did not survive correction can be found at https://neurovault.org/collections/ELLZKWSV (i) a two-sample t-test (homosexual vs heterosexual) in response to the subliminal face presentation before watching the movie, (ii) a two-sample t-test (homosexual vs heterosexual) in response to the subliminal face presentation of the character after watching the movie and (iii) a two-way ANOVA using change in IDN scores as regressor on the subliminal face perception (iv) areas showing IAT-dependent differences on subliminal face perception (two-way ANOVA at different significant levels).

Fig. 4

Significantly stronger BOLD signals in homosexual vs heterosexual subjects to the subliminal presentation of the face of the character after viewing the movie (P < 0.01, cluster corrected).

Discussion

Aiming to bring our experiment close to real social life, we used a movie to induce implicit social bias. As expected, by revealing the priest’s homosexuality the film modulated how strongly the subjects identified with him. Furthermore, this knowledge modulated the implicit neural response to his face. Confirming our hypothesis, homosexual vs heterosexual subjects showed significantly stronger activation in response to the subliminally presented face of the homosexual character in mPFC and TPJ along with ACC, FP and sFG. This further demonstrates the important role of contextual knowledge on automatic and fast initial social evaluation of faces as was previously suggested (Freeman ; Stolier and Freeman, 2016).

Previous studies using subliminal presentation of faces show activity in brain areas thought to be related to social perception and empathy (for a review, see Brooks ). Our new results that accord with these findings suggest that neural mechanisms underlying intergroup empathy may occur already at the level of subliminal perception. The brain areas that were modulated in the present study have been previously reported to be involved in social perception including intergroup bias (for a review, see Amodio, 2014; Cikara and Bavel, 2014). For example, mPFC along with ACC has been reported as important regions for the processing of social information (Mitchell ; Amodio and Frith, 2006; Krueger ; Van Overwalle, 2009; Janowski ), as well as to the self-referential thinking (Vogeley and Fink, 2003; Mitchell ; Morrison ). Furthermore, mPFC has been reported to be involved in social information processing (see Grossmann, 2013 and in-group perception Volz ). In addition, TPJ has been consistently reported to support empathy, theory of mind and social perception (Saxe and Kanwisher, 2003; Saxe, 2006; Decety and Lamm, 2007; Decety ). Interestingly, the frontal gyrus and regions in the FP were also reported previously to be related to processing in-group faces as well as one’s own face (Scheepers ). Thus, activations observed in the present study in the homosexual subjects in response to the face of the homosexual character might have been due to knowledge gained about him via the movie as an in-group member, which modulated implicit social perception and elicited empathy and in-group response.

Our IAT results indicate that heterosexual subjects had a bias toward homosexuality. In addition, identification scores show that heterosexual subjects experienced significant decrease of identification with the priest after knowing he is homosexual. As a consequence, we expected to find stronger neural activations in heterosexual (vs homosexual) subjects in the insular region after viewing the movie. A tendency for such activation was detected in insular region previously reported to be related to disgust (Stark ), yet the activation did not survive cluster correction.

An alternative explanation to the differences in brain activation between homosexual vs heterosexual subjects could be a difference in empathic sensitivity levels between the heterosexuals and homosexuals as suggested previously (Perry ). Since our film protagonist is a Catholic priest, which in the middle of the film is revealed to be homosexual, this might have caused an experience of violation of social norms in some of the subjects (Burgoon and Hale, 1988). Although it is unlikely to expect that one group experienced more violation of expectancy than the other, it is still possible that some of the results might be related to differences in experiencing a violation of expectancy (Le Poire and Burgoon, 1996). Another possibility might be an increase in perceived attractiveness. It has been shown that preferred-sex faces are more rewarding than non-preferred-sex faces (Hahn ). However, this preferred-sex bias was particularly pronounced when attractive faces were presented, while in our study the differences in activation were found to the same face and only after presenting contextual information via the movie. However, this does not exclude the possibility that the attractiveness of the priest’s face increased in homosexual subjects after seeing the film and this could explain part of the observed activations such as the ones in the frontal gyrus (Turk ; Proverbio ).

Limitations of the study

The current study had some limitations. First, our participants were all Finnish men, so it is possible that choosing a Finnish movie would have evoked stronger responses among the subject groups, even though all subjects were sufficiently proficient in English to follow the English movie with ease. Our choice of an English film from the 90s was to assure that none of our subjects had seen the film before, which would have damaged our experimental design where the main protagonist is revealed to be homosexual in the middle of the film. Secondly, our recruited subjects might have varied in their given implicit bias, i.e. some might have in-group bias but not out-group bias and vice versa, which we cannot discern based on the data collected in the present study. Finally, because the number of subjects in our experimental groups was relatively small, future experiments using larger sample sizes are needed to confirm our results and shed more light on possible additional brain regions that might be involved. To facilitate meta-analyses we have uploaded unthresholded statistical parametric maps to Neurovault.

Conclusion

We demonstrate here that subliminal presentation of the face of a film character with a specific social identity can be used to measure the brain’s implicit neural responses to individuals with various social relationships to the viewer. Our approach can be a naturalistic alternative to investigate different types of intergroup biases.

Funding

This work was supported by the Academy of Finland (259952 and 276643).