The sample for the fMRI study consisted of 24 healthy female participants ages 18 to 19 years (M
age = 19.1 years, SD
age = .5 years). The sample size was based on prior studies using the social judgment paradigm (Somerville et al., 2006). Participants were recruited through local advertisements and through a recruitment website. All participants were screened for MRI contraindications and reported no diagnosed psychiatric disorder using a telephone interview before the scanning session. Informed consent was obtained from participants prior to the scan session. All participants’ height and weight were measured to calculate their BMI (based on weight in kilos divided by the square height in meters; cf. WHO, 2015). This resulted in an average BMI of 22.2 (SD = 3.1; range: 17.5–30). Participants received €30 for participation in a larger set of studies. This study was approved by the University’s Medical Ethical Committee and was conducted in accordance with the provisions of the World Medical Association Declaration of Helsinki.
Before the start of the study, participants were told that they would participate in a study on the processes behind forming judgments. During the lab visit, participants were instructed about the procedure of an fMRI scan. A short explanation about the Body Image Paradigm was provided, followed by six practice trials of the Body Image Paradigm. Participants were weighed and measured before the scanning session. Directly after the scanning session, participants completed a paper-and-pencil version of the questionnaires. Finally, all participants were debriefed by explaining that the feedback provided during the task was not actually based on the opinion of a majority of their peers but merely on the opinion of some of those peers.
The Body Image Paradigm is an adapted version of the Social Judgment task, which was previously used by Gunther Moor et al. (2010) and Somerville et al. (2006), in combination with ideal-body imagery format, as previously described by Veldhuis et al. (2012, 2014a). In the paradigm, 60 pretested media models (30 categorized as too thin and 30 as normal; see below) were shown on the scanner screen, and participants rated each model by indicating whether they perceived each of the media models as too thin or as normal. Subsequently, upon their own rating, the participants received feedback indicating opinions from ostensible peers (too thin or normal), as was explained to the participants beforehand. These own perceptions versus peer feedback combinations of the models’ body sizes led to incongruent (too thin–normal and normal–too thin) and congruent (too thin–too thin and normal–normal) situations.
In a pretest for the Body Image Paradigm, a large sample of 135 pictures displaying young female media models showing swimwear in beach settings were selected from the Internet (e.g., via Google search images). With Photoshop techniques the images were made comparable in formatting (e.g., cutting areas such that the model was central in the picture and her body image features clearly visible), resulting in a set of similar media model stimuli with a blue sky background, a sandy beach, and the media models facing the camera (cf. Veldhuis et al. (2012, 2014a). An independent sample of young females (N = 124; M = 15.99; SD = 1.90; age range: 12–19) rated each media model on perceived attractiveness and the model’s body size on 10-point semantic scales, varying from very ugly to very beautiful, and extremely big to extremely thin. To select stimuli for the Body Image Paradigm, the rated media models were compared on attractiveness and body size (thin to normal weight). We selected those models that had varying body sizes but were comparable in attractiveness (M = 6.55 for models rated as thin, M = 6.63 for models rated as normal weight). This resulted in a selection of 30 media models that were rated as too thin and 30 media models that were rated as normal.
In the scanner version of the Body Image Paradigm, feedback presented to the participants was randomized, such that participants did not receive the same feedback on the same answer for more than two times in a row. To enhance the credibility of the peer feedback, for 10 models (i.e., 5 times too thin, 5 times normal) the feedback was fixed to these conditions, because these models were always rated as too thin or normal during the pretest. More specifically, this led to the following set of stimuli: For the 30 too thin models, five images were consistently prerated as too thin and therefore always received peer feedback too thin; 25 images were on average prerated as too thin and peer feedback communicated 10 times too thin and 15 times normal. For the 30 normal models, five images were consistently prerated as normal and therefore always received peer feedback normal; 25 images were on average prerated as normal and peer feedback communicated 10 times normal and 15 times too thin. Next, the final conditions included in the analysis were based on the combinations of participants’ own judgments and peer feedback (e.g., participant rating too thin, peer feedback normal), and therefore the number of trials in each condition varied between participants.
During the task, each trial was preceded by a fixation cross with jittered duration between 600 and 4450 ms. Images were presented against a black background for a maximum of 3,000 ms. Within these 3,000 ms, participants had to respond by pressing the left button (too thin) or right button (normal) on a button box with their index or middle finger. Directly after their decision, the choice of the participant (too thin or normal) appeared on the left side of the image on the screen for 2,500 ms. After this, peer feedback (too thin or normal) was presented on the right side of the image for 2,500 ms (see Fig. 1). Responses that exceeded the duration of 3,000 ms were modeled separately and not included in the analyses. Instead, a screen with Too Slow was shown 2,500 ms, immediately followed by the start of the next trial. This occurred in less than 3% of the trials.
Several outcome measures were taken from this paradigm. First, we compared the participants’ ratings of the models (too thin or normal) to our previously set precategorization (too thin or normal) to check whether their evaluations matched those of our pretest panel. Second, to investigate whether it would take longer to rate models as too thin than as normal, we calculated the reaction times for these two different response options. Finally, we used the combinations of participants’ ratings and the subsequent peer feedback as conditions in our MRI analyses.
Self-esteem was assessed through Rosenberg’s (1965) 10-item Self-Esteem Scale (e.g., “I feel I have a number of good qualities”; as applied in adolescent girls in Veldhuis et al., 2014a), followed by 5-point rating scales (1 = totally disagree; 5 = totally agree). Higher scores indicated a higher self-esteem. Participants received an average sum score of 26.83 (SD = 7.17), with a range of 4–35 (Cronbach’s α = .89).
Body dissatisfaction was measured through the 9-item Body Dissatisfaction subscale from the Eating Disorder Inventory (Garner, Olmstead, & Polivy, 1983), which was extended with four items to create a balanced set of indicative (e.g., “I think my belly is too fat”) and counterindicative (e.g., “I am happy with my figure”) answers (cf. Veldhuis et al., 2014a). The 13 items could be answered on a 5-point rating scale (1 = totally disagree; 5 = totally agree). After recoding, higher scores indicated more body dissatisfaction. Participants received a mean score of 32.9 (SD = 8.2) with a range of 21–54 (Cronbach’s α = .81).
MRI data acquisition
Scans were made with a 3 Tesla Philips scanner, using a standard whole-head coil. The functional scans were acquired using a T2*-weighted echo-planar imaging (EPI). The first two volumes were discarded to allow for equilibration of T1 saturation effects (TR = 2.2 s, TE = 30 ms, sequential acquisition, 38 slices of 2.75 mm, field of view 220 mm, 80 × 80 matrix, in-plane resolution 2.75 mm). After the functional runs, a high resolution 3D T1-weighted anatomical image was collected (TR = 9.751 ms, TE = 4.59 ms, flip angle = 8°, 140 slices, 0.875 mm × 0.875 mm × 1.2 mm, and FOV = 224.000 × 168.000 × 177.333). Visual stimuli were presented on a screen that was attached in the magnet bore. Participants could see the stimuli via a mirror attached to the head coil. Head movement was restricted by using foam inserts inside the coil.
FMRI data analysis
All data were analyzed with SPM8 (Wellcome Department of Cognitive Neurology, London). Images were corrected for differences in rigid body motion. Structural and functional volumes were spatially normalized to T1 templates. Translational movement parameters never exceeded 1 voxel (3 mm) in any direction for any participant or scan. The normalization algorithm used a 12-parameter affine transform together with a nonlinear transformation involving cosine basis functions and resampled the volumes to 3 mm cubic voxels. Templates were based on the MNI305 stereotaxic space (Cocosco, Kollokian, Kwan, & Evans, 1997). Functional volumes were spatially smoothed with a 6 mm FWHM isotropic Gaussian kernel.
The onset of the stimulus and the feedback display of each trial were modeled as zero duration events (see Gunther Moor et al., 2010). We divided the feedback displays in four conditions, focusing on combinations of participant ratings with peer feedback: congruent too thin (too thin–too thin), incongruent too thin (too thin–normal), congruent normal (normal–normal) and incongruent normal (normal–too thin).
All events were time locked to the moment of the start of the feedback screen. The trial functions were used as covariates in a general linear model; along with a basic set of cosine functions that high-pass filtered the data. The least-squares parameter estimates of height of the best fitting canonical HRF for each condition were used in pair-wise contrasts. The resulting contrast images, computed on a subject-by-subject basis, were submitted to group analyses. We tested the neural response to incongruent feedback with two contrasts: incongruent thin > congruent thin (and the reversed contrast) and incongruent normal > congruent normal (and the reversed contrast). Task-related responses were considered significant if they exceeded a FWE voxel level threshold of p < .05, or a FDR cluster-corrected threshold of p < .05, with an initial threshold of p < .005 (Woo, Krishnan, & Wager, 2014).
Region of interest analysis
We used the MarsBaR toolbox (Brett, Anton, Valabregue, & Poline, 2002) for SPM8 to perform region of interest (ROI) analyses. The general contrast incongruent feedback > congruent feedback, with a threshold of FWE corrected at p < .05 at the voxel level, was used to determine suitable ROIs. This contrast was chosen because it is not biased toward normal or too thin peer feedback, but collapsed across conditions (see Fig. 3). The contrast resulted in three clusters that were extracted with the MarsBaR toolbox: left insula (x = −33, y = 2-, z = −14), right insula (x = 36, y = 20, z = −14) and dorsal medial prefrontal cortex (dmPFC)/ACC (x = 0, y = 23, z = 52). Pearson correlations were computed between the contrast values for dmPFC/ACC, left insula and right insula and self-esteem, BMI and body dissatisfaction.
First, we tested participants’ ratings of media models as being too thin or normal compared to our initial precategorization of the models as being too thin or normal (i.e., precategorized: too thin vs. participants’ rating: too thin; precategorized: too thin vs. participants’ rating: normal; precategorized: normal vs. participants’ rating: too thin; precategorized: normal vs. participants’ rating: normal). Participants’ ratings of the media models’ body sizes were mostly in accordance with those in the pretest. More specifically, 70.85% of the media models that were precategorized as too thin were also rated as too thin in the main study, while 91.7% of the media models that were precategorized as normal models were rated as normal. It therefore seems that participants were biased towards giving a normal rating to precategorized normal models.
Next, we tested reaction times (RT) for rating the portrayed media models’ body shapes. For this analysis, we only included trials where (1) the model was precategorized as (, 2) the model was precategorized as too thin but participants rated the model as normal, and (3) the model was precategorized as normal and participants rated the model as normal. There were too few trials on which the models were precategorized as normal and participants rated the model as too thin, and therefore these trials were removed from the RT analyses. An ANOVA with the aforementioned three conditions revealed that, regardless of their own ratings of a media model’s body size, it took participants significantly longer to rate the prerated too thin media models’ body sizes (too thin model with rating too thin: M = 1,572 ms, SD = 331 ms; too thin model with rating normal: M = 1,626 ms, SD = 360 ms) compared to rating precategorized normal models (normal model with rating normal: M = 1,304 ms, SD = 229), F(2, 46) = 17.31, p < .001. Thus, even when the participant rated the too thin models as normal, they were slower to react compared to rating models that were precategorized as normal as normal (p < .001).
Prior to performing the analyses we checked for outliers in the data. One significant outlier (Z value <−3.29 or >3.29; Tabachnick & Fidell, 2013) in self-esteem scores (score 4) was removed from the analyses. This resulted in analyses including 23 participants. There was a significant negative correlation between self-esteem and body dissatisfaction (r = −.61, p < .005), indicating that a lower self-esteem relates to experiencing more body dissatisfaction. No significant correlations were found between BMI and self-esteem (r = −.24, p = .26), or BMI and body dissatisfaction (r = .36, p = .09).
Whole brain analysis
To test processing of peer feedback that deviated from the participant’s response, we tested neural activation for congruent and incongruent feedback. For the first set of analyses, we compared the effects of congruent and incongruent feedback following the too thin rating from the participant, resulting in the contrast participant rating: too thin vs. peer feedback: normal > participant rating: too thin vs. peer feedback: too thin. This contrast resulted in increased activity in the left and right insula and dmPFC, extending into the ACC (dmPFC/ACC; Fig. 2a). An overview of all activated clusters for this contrast is presented in Table 1.
For the second set of analyses, we compared the effects of congruent and incongruent feedback following the normal ratings from the participant. This resulted in the contrast participant rating: normal vs. peer feedback: too thin > participant rating: normal vs. peer feedback: normal. This contrast again showed increased activity in the left and right insula and dmPFC/ACC (Fig. 2b). An overview of all activated clusters for this contrast is presented in Table 2.
Hence, results showed increased activity in the dmPFC/ACC and insula in both incongruent situations, that is, when participants’ ratings differed from the peer feedback ratings of the media models’ body shapes.
To test whether these regions were more strongly engaged for incongruent feedback after a too thin rating or after a normal rating, these two conditions were contrasted directly to each other (participant rating: too thin vs. peer feedback: normal compared to participant rating: normal vs. peer feedback: too thin). This analysis resulted in a single cluster in the ACC that was more active for participant rating: normal vs. peer feedback: too thin compared to participant rating: too thin vs. peer feedback: normal (see Fig. 2c; Table 3). Thus, in incongruent situations participants’ ACC was more strongly activated when the peer feedback signaled that the model was too thin while participants rated the model as normal. No significant activity for the reversed contrast was shown (in the incongruent situation where normal peer feedback follows participants’ rating of models being too thin).
Relation with self-esteem
To test for relations between neural activity to peer feedback and self-esteem, ROIs from the left and right insula and dmPFC/ACC were extracted from the general contrast incongruent–congruent feedback across conditions (see Method sections). For these three regions, contrast values were computed for participant rating: too thin vs. peer feedback: normal minus participant rating: normal vs. peer feedback: normal and for participant rating: normal vs. peer feedback: too thin minus participant rating: too thin vs. peer feedback: too thin. Significant negative correlations were found between self-esteem and activity in both left insula (r = −.45, p = .03), right insula (r = −.49, p =.02) and the dmPFC/ACC (r = −.48, p = .02), but only for the contrast participant rating: normal vs. peer feedback: too thin (vs, participant rating: normal vs. peer feedback: normal). These relations were not found for the other incongruent contrast (participant rating: too thin vs. peer feedback: normal compared to participant rating: too thin vs. peer feedback: too thin, all ps > .17). Thus, neural activity associated with receiving feedback that peers considered the model too thin whereas their own rating was normal, was stronger for those with lower self-esteem (illustrated in Fig. 3).
Finally, to test whether other constructs related to body image were also related to neural activation in the contrast participant rating: normal vs. peer feedback: too thin (vs. participant rating: normal vs. peer feedback: normal), the same analyses were also performed with BMI and body dissatisfaction. Neither BMI nor body dissatisfaction was correlated with any of the contrast values. For an overview of all correlations, see Table 4.