Edited by: Frank Krueger, George Mason University, USA
Reviewed by: Ruolei Gu, Chinese Academy of Sciences, China; Chunliang Feng, Beijing Normal University, China
*Correspondence: Juan F. Domínguez D
†Equal first authors.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
People often find it hard to disagree with others, but how this disposition varies across individuals or how it is influenced by social factors like other people's level of expertise remains little understood. Using functional magnetic resonance imaging (fMRI), we found that activity across a network of brain areas [comprising posterior medial frontal cortex (pMFC), anterior insula (AI), inferior frontal gyrus (IFG), lateral orbitofrontal cortex, and angular gyrus] was modulated by individual differences in the frequency with which participants actively disagreed with statements made by others. Specifically, participants who disagreed less frequently exhibited greater brain activation in these areas when they actually disagreed. Given the role of this network in cognitive dissonance, our results suggest that some participants had more trouble disagreeing due to a heightened cognitive dissonance response. Contrary to expectation, the level of expertise (high or low) had no effect on behavior or brain activity.
The freedom to make autonomous choices (within the limits of the law) without fear of harm or prosecution is a fundamental value at the core of the Universal Declaration of Human Rights, democratic societies and market economies. However, individual choice is never fully autonomous. We may be highly susceptible to the opinions of others (Cialdini and Goldstein,
The truth bias, however, represents a challenge for societies to balance. It entails a reduced inclination for individuals to disagree with their peers, with potentially adverse effects on autonomous choice. There is evidence that this reduced inclination to disagree may vary across individuals (Laird and Berglas,
Neuroscientists have recently started to investigate the neural mechanisms underlying disagreement. Studies in this area have implicated a set of posterior medial frontal cortex (pMFC) structures [comprising dorsal medial prefrontal cortex, dorsal anterior cingulate cortex (dACC), and pre-supplementary motor area] as well as anterior insula (AI; Westen et al.,
In this fMRI study, we aimed to further explore why people often find it hard to disagree with others, and how this effect is modulated by individual differences and other people's expertise. We were specifically interested in exploring whether cognitive dissonance is a factor in people's disposition to agree/disagree and, consequently, their truth bias. Previous studies have investigated the brain responses to others' subjective opinions that are the same or different from participants' opinions (Klucharev et al.,
We also wanted to ensure participants' responses reflected their disposition to disagree and the effect of expertise on it, unconfounded by other variables like knowledge of the subject area, personal opinions, valuations or preferences. Therefore, unlike previous studies, the chosen object of disagreement was not subjective opinions but objectively defined statements that could be correct or incorrect. Moreover, the statements were designed to be difficult with the purpose of inducing participants to rely more on the person making the statement and their level of expertise (high vs. low), rather than their own knowledge.
We predicted that, due to the existence of the truth bias, participants would more often agree with the statements than disagree. We expected this would be motivated, at least in part, by the experience of cognitive dissonance during disagreement. Accordingly, we anticipated increased activation in response to disagreement in pMFC and AI. Importantly, individual differences were expected to modulate the above effects as suggested by earlier studies (Campbell-Meiklejohn et al.,
Thirty-nine healthy right-handed participants (19 females; mean age = 22.1 years,
Participants were presented with a total 192 true or false statements from four fields of knowledge: biology (e.g., “Orchid flowers have the most species”), history (e.g., “The first public library was opened in England”), medicine (e.g., “Protanopia is the inability to see the color green”), and physics (e.g., “The faster you move the heavier you get”). For the full list of statements see Supplementary Table
Participants were asked to decide whether the statements were true or false. Agreement was therefore operationalized as true judgments and disagreement as false judgments. Participants were told statements were checked for their accuracy, with many of them found to be incorrect. This was done to ensure participants disagreed (
The difficulty of the statements was confirmed in a prior validation behavioral experiment with an independent cohort of 20 participants (18 females; mean age = 21.6 years,
One half of the true (T1) and one half of the false (F1) statements were randomly assigned to the professors and the other halves of true (T2) and false (F2) statements were assigned to the students. Which subsets (T1, T2, F1, F2) were assigned to professors or students was counterbalanced across participants. A one-way ANOVA revealed no significant differences between the mean truthfulness ratings of any these subsets of statements,
Each of four functional runs contained 12 randomly presented blocks, including eight task blocks (lasting 65 s each) and four rest blocks (lasting 20 s each). Task blocks grouped statements by expertise level: four blocks contained statements attributed to a professor (
Structural and functional MR images were acquired with a 3 Tesla scanner (Siemens AG, Erlangen, Germany) and a 32-channel head coil. Functional images were acquired using gradient-echo planar imaging (EPI) with the following parameters: repetition time (TR) = 2.5 s; echo time (TE) = 36 ms; flip angle = 90°; 64 × 64 matrix; voxel size = 3 × 3 × 3 mm. A total of 246 whole brain images per run were acquired, each consisting of 36 transversal slices with 10% gap between each slice. A high-resolution 3D T1-weighted image covering the entire brain was also acquired after the second run for anatomical reference (TR = 1900 s; TE = 2.32 ms; flip angle = 9°; 192 × 192 matrix; voxel size = 0.9 × 0.9 × 0.9 mm).
In this study, there were two factors—agreement and expertise—with two levels each—agree/disagree (corresponding to true/false judgments), and high/low, respectively. This resulted in four conditions: agreement with a professor (AP), disagreement with a professor (FP), agreement with a student (AS), and disagreement with a student (FS). An agreement score was then computed by subtracting disagree (false) from agree (true) scores, collapsed across the levels of expertise (
Functional MRI data were pre-processed and analyzed with SPM8 Software (Wellcome Department of Imaging Neuroscience, Institute of Neurology, London) through MATLAB (MathWorks Inc., USA). All EPI images were first realigned to the first image of each run to offset any effects of head movements. The T1-weighted anatomical scan for each participant was then coregistered to the mean functional image created during realignment. Using the segment routine, the coregistered anatomical scan was then normalized with a voxel size of 1 × 1 × 1 mm to the MNI T1 standard template (Montreal Neuropsychological Institute). Following this, the same parameters were used to normalize all of the EPI images to map onto the template using a voxel size of 3 × 3 × 3 mm. The images were then smoothed with a 9-mm isotropic Gaussian kernel.
A general linear model was used to estimate regions of significant Blood Oxygen Level Dependent (BOLD) response in each voxel for each participant and included event-related regressors for each of four conditions (AP, DP, AS, and DS); that is, each condition was modeled based on the participant's agree/disagree (true/false) responses to the statements. The onset of each event corresponded to the start of the slides in which the statements were shown and had a duration of 9 s. Contrast images for each participant across all conditions were included at the second level in a 2 (agreement: Agree vs. Disagree) × 2 (expertise: Professor vs. Student) factorial design, and the main effects and interaction effects were estimated. The above experimental design can therefore be effectively conceived as mixed, since statements were blocked by level of expertise, but could only be sorted by agreement
To investigate if participants who disagreed less often had more activation in the pMFC and AI when they disagreed, we modeled the relationship between the BOLD response exclusive to the disagreement compared to agreement condition (
The agreement score was not significantly different from 0 (
Whole-brain analysis revealed no main effect of agreement or expertise. There was also no interaction between the two factors. However, there was a significant association between BOLD response and the agreement score exclusive to the Disagree minus Agree contrast in four clusters comprising the following network: one cluster corresponded to pMFC bilaterally, specifically comprising dorsal medial prefrontal frontal cortex (dMPFC), dACC, and pre-supplementary motor area (pre-SMA); a second cluster with peak voxel on lateral orbitofrontal cortex (LOFC) also included AI on the right; another cluster with peak voxel on LOFC incorporated AI and inferior frontal gyrus (IFG); one last cluster involved the left angular gyrus (AG; see Table
pMFC bilaterally | 1182 | B | 0 | 35 | 43 | 6.58 | < 0.001 |
AI/IFG/OFC | 217 | L | −42 | 23 | −5 | 4.43 | 0.033 |
AI/OFC | 211 | R | 51 | 41 | −8 | 6.01 | 0.035 |
AG | 309 | L | −45 | −64 | 49 | 4.59 | 0.012 |
We carried out an additional analysis as a test of whether a low proportion of disagreement truly reflects a difficulty to disagree. First, we calculated the relative agreement RT (
To rule out potential confounding effects on the results, a number of additional analyses were performed. First, participants' responses could have been biased by the truth-like quality of statements (
The pattern of brain activity we found could have also been influenced by a salience effect stemming from a low proportion of instances of disagreement for participants who disagreed less often. To address this possibility, we estimated the proportion of trials where participants who disagreed less often (defined stringently as the bottom 25th percentile on the agreement score) actually disagreed. We found these participants (
In this study, we used fMRI to ascertain why people often have an aversion to disagree with others, and how this effect is modulated by individual differences and other people's expertise. We therefore asked participants, for the first time, to make the active decision to agree or disagree with others. We also ensured that this decision reflected participants' disposition to disagree unconfounded by other factors. Results confirm our prediction that individual differences modulate a brain network previously involved in disagreement and cognitive dissonance comprised of posterior medial prefrontal cortex (pMFC) structures and AI. Specifically, participants who disagreed less frequently exhibited greater activation in this network when they actually disagreed.
Both pMPC and AI have been consistently found to respond to disagreement (Klucharev et al.,
Broadly speaking, pMFC is involved in on-line monitoring and control of action, which is especially important in situations involving conflict between alternatives. In particular, this area plays a key role in reinforcement learning (in both social and non-social contexts, Ruff and Fehr,
As in previous studies involving disagreement (Klucharev et al.,
AI is thought to be involved in cognitive dissonance by virtue of its role in negative affect and autonomic arousal (van Veen et al.,
A recent model postulates AI performs a more general function as an internal hub in salience detection (Menon and Uddin,
Together with pMFC and AI, a number of other areas, including IFG and angular gyrus (AG) in the left hemisphere, as well as lateral orbitofrontal cortex (LOFC) bilaterally, were active in those who disagreed less often. IFG (Campbell-Meiklejohn et al.,
More broadly, IFG and AG have been shown to play a role in inhibition of prepotent responses (Aron et al.,
LOFC has an important function in the representation of displeasure (Kringelbach and Rolls,
Previous studies have reported individual brain activity differences in the network reported in the present study (including pMFC, AI, IFC, LOFC), depending on how much participants were influenced by others' opinions (Berns et al.,
Our agreement score can be interpreted as a measure of the disposition to disagree (or agree), effectively a proxy for the truth bias, and may be considered a personality trait encoded in this network's response. This was explicitly anticipated in the original formulation of cognitive dissonance (Festinger,
Confounding factors could have potentially affected our results. First, participant's responses could have been biased by the truth-like quality of the statements. While the proportion of agreement with true statements was statistically higher than the proportion of agreement with false statements, we found that the difference between the proportion of agreement with false vs. true statements had no effect on either the agreement score or the brain response to disagreement. This indicates that the truth-like quality of the statements had no effect on our results. Second, rather than greater difficulty to disagree for those who disagreed less, the pattern of brain activations we found could reflect higher salience associated with a low proportion of instances of disagreement for participants who disagreed less often. Contradicting this possibility, participants who disagreed less often still disagreed a sizeable ~35% of the time.
Conflicting with our predictions, only half of participants agreed more often with the statements. Therefore, we found overall no evidence of an agreement effect or truth bias (Vrij,
The failure to observe an expertise effect, on behavior or brain activity, may be related to a potentially weak priming of the expertise condition, with expertise priming presented only briefly (during 5 s) at the beginning of each (65 s) block. Alternatively, there may have been a group membership confound. In particular, a student in-group bias may have canceled out the effect of the high expertise group—on account of favoritism for one's in-group (Crocker and Luhtanen,
A small number of studies have investigated disagreement behavior and its neural basis, together with how behavior and brain responses vary across individuals and in response to social factors (Klucharev et al.,
Having a lot of trouble disagreeing due to a heightened cognitive dissonance response may be indicative of an array of emotional, attitudinal or social issues compromising an individual's capacity to make autonomous choices. This can potentially lead to poor decision-making, anxiety, or difficulties in interpersonal relationships. For example, introversion has been shown to be associated with the experience of greater dissonance discomfort (Matz et al.,
This work was supported by an Australian Research Council (ARC) Early Career Research Award (DE130100120), Heart Foundation Future Leader Fellowship (100458), and ARC Discovery Grant (DP130100559) awarded to PM.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
We would like to thank Zoie Nott for her help with data acquisition.
The Supplementary Material for this article can be found online at: