Sam Harris's new study on political beliefs (Neuroscience)

[Petros Agapetos]

Abstract

People often discount evidence that contradicts their firmly held beliefs. However, little is known about the neural mechanisms that govern this behavior. We used neuroimaging to investigate the neural systems involved in maintaining belief in the face of counterevidence, presenting 40 liberals with arguments that contradicted their strongly held political and non-political views. Challenges to political beliefs produced increased activity in the default mode network—a set of interconnected structures associated with self-representation and disengagement from the external world. Trials with greater belief resistance showed increased response in the dorsomedial prefrontal cortex and decreased activity in the orbitofrontal cortex. We also found that participants who changed their minds more showed less BOLD signal in the insula and the amygdala when evaluating counterevidence.

These results highlight the role of emotion in belief-change resistance and offer insight into the neural systems involved in belief maintenance, motivated reasoning, and related phenomena.


Introduction

Few things are as fundamental to human progress as our ability to arrive at a shared understanding of the world. The advancement of science depends on this, as does the accumulation of cultural knowledge in general. Every collaboration, whether in the solitude of a marriage or in a formal alliance between nations, requires that the beliefs of those involved remain open to mutual influence through conversation. Data on any topic—from climate science to epidemiology—must first be successfully communicated and believed before it can inform personal behavior or public policy. Viewed in this light, the inability to change another person’s mind through evidence and argument, or to have one’s own mind changed in turn, stands out as a problem of great societal importance. Both human knowledge and human cooperation depend upon such feats of cognitive and emotional flexibility.
It is well known that people often resist changing their beliefs when directly challenged, especially when these beliefs are central to their identity1 (http://www.nature.com/articles/srep39589#ref1),2 (http://www.nature.com/articles/srep39589#ref2),3 (http://www.nature.com/articles/srep39589#ref3),4 (http://www.nature.com/articles/srep39589#ref4),5 (http://www.nature.com/articles/srep39589#ref5),6 (http://www.nature.com/articles/srep39589#ref6).

In some cases, exposure to counterevidence may even increase a person’s confidence that his or her cherished beliefs are true7 (http://www.nature.com/articles/srep39589#ref7),8 (http://www.nature.com/articles/srep39589#ref8). Although neuroscientists have begun to study some of the social aspects of persuasion9 (http://www.nature.com/articles/srep39589#ref9) and motivated reasoning10 (http://www.nature.com/articles/srep39589#ref10), little research is aimed directly at understanding the neural systems involved in protecting our most strongly held beliefs against counterevidence.
One model of belief maintenance holds that when confronted with counterevidence, people experience negative emotions borne of conflict between the perceived importance of their existing beliefs and the uncertainty created by the new information11 (http://www.nature.com/articles/srep39589#ref11),12 (http://www.nature.com/articles/srep39589#ref12),13 (http://www.nature.com/articles/srep39589#ref13),14 (http://www.nature.com/articles/srep39589#ref14). In an effort to reduce these negative emotions, people may begin to think in ways that minimize the impact of the challenging evidence: discounting its source, forming counterarguments, socially validating their original attitude, or selectively avoiding the new information15 (http://www.nature.com/articles/srep39589#ref15). The degree to which such rationalization occurs depends upon several factors, but the personal significance of the challenged belief appears to be crucial. Specifically, beliefs that relate to one’s social identity are likely to be more difficult to change16 (http://www.nature.com/articles/srep39589#ref16),17 (http://www.nature.com/articles/srep39589#ref17),18 (http://www.nature.com/articles/srep39589#ref18),19 (http://www.nature.com/articles/srep39589#ref19).

Based on this model, predictions can be made about the neural systems that govern resistance to belief change. First, resistance to evidence may entail disengagement from external reality and increased inward focus. The brain’s default mode network (DMN), including posterior and anterior midline structures and the lateral inferior parietal lobes, appears to support these psychological processes20 (http://www.nature.com/articles/srep39589#ref20),21 (http://www.nature.com/articles/srep39589#ref21). Identity-related beliefs might also invoke internal models of the self, a form of cognition that is associated with increased activity within the DMN22 (http://www.nature.com/articles/srep39589#ref22),23 (http://www.nature.com/articles/srep39589#ref23). Second, if resistance to belief change is partly motivated by negative emotion, having one’s beliefs contradicted may produce activity in associated regions of the brain, such as the amygdala, the insular cortex, and other structures involved in emotion regulation.
In this study, we performed functional MRI to measure the brain activity of 40 individuals with strong political views as they encountered arguments against their beliefs. All the subjects were self-identified as political liberals of deep conviction. Inside the fMRI scanner, participants saw a series of statements they previously indicated strongly believing, followed by several challenging counterarguments. After participants read all five counterarguments, the original statement was shown again and they reported their post-challenge belief strength. The difference between pre-scan and post-challenge ratings was used as a measure of belief change. In order to compare high belief persistence to low belief persistence, in one condition we challenged strongly held political beliefs, and in another condition we challenged strongly-held non-political beliefs. While the non-political beliefs were just as strongly held according to the participants who held them, we did not expect these beliefs to be defended with the same vigor.

We predicted that the political condition would result in less belief change than the non-political condition, and that resisting challenges to political beliefs would be associated with increased activity in brain systems involved in contemplating identity and internally-focused cognition.

Furthermore, we predicted that there would be a relationship between activity in emotion-related brain structures and individual differences in persuadability. We also sought to identify brain activity that correlated with the strength with which specific beliefs were maintained in our sample.

[Petros Agapetos]

Materials and Methods

Participants

Forty healthy participants with no history of psychological or neurological disorders were recruited from the University of Southern California community and the surrounding Los Angeles Area (mean age: 24.30 ± 0.92 years, range: 18–39 years, 20 male). All participants were right-handed according to their own report. Subjects were paid $20 per hour for their participation and gave informed consent. All experimental protocols were approved by the Institutional Review Board of the University of Southern California and procedures were carried out in accordance with the approved guidelines. All participants had spent the majority of their life living in the United States and spoke fluent English, identified themselves as politically liberal, and had strongly held political and non-political beliefs. Specifically, participants answered a screening questionnaire in which they were asked about their political identification. On the question “Do you consider yourself a political person?” answers ranged on a scale from 1 (not at all) to 5 (very much). Participants were only included if they answered at least a 4 on this question. For the question “Which of the following describes your political self-identification?” answers ranged from 1 (strongly liberal) to 7 (strongly conservative) and participants were only included if they answered 1 or 2. Additionally, participants rated their agreement with several political and non-political statements and were only included in the experiment if they strongly agreed with at least 8 political and 8 non-political statements. Of 116 people who responded to our advertisements, 98 met the requirements for age, handedness, and political orientation. From those 98 people, 40 subjects met the requirements for strongly agreeing to at least 8 statements in each category.

Stimuli
In this experiment, each participant read 8 political statements and 8 non-political statements with which they had previously indicated strong agreement. Each statement was followed by 5 challenges. Each challenge was a sentence or two that provided a counter-argument or evidence against the original statement.

The 8 political statements for each participant were drawn from a pool of 9 political statements. These statements concerned policy issues on which we expected predictable, identity-consistent positions from our subjects, such as “Abortion should be legal” and “Taxes on the wealthy should generally be increased”. The statements can be found in full in Table S3 (http://www.nature.com/articles/srep39589#s1). The 8 non-political statements were drawn from a pool of 14 non-political statements. The pool of non-political statements was larger because while the inclusion criteria guaranteed the participants would hold certain political beliefs, they did not guarantee belief in any specific non-political statement. The non-political statements covered a wide range of topics including health (e.g. “Taking a daily multivitamin improves ones health”), education (e.g. “A college education generally improves a person’s economic prospects”), and history (e.g. “Thomas Edison invented the light bulb”).
Each political and non-political statement was associated with 5 challenges. In order to be as compelling as possible, the challenges often contained exaggerations or distortions of the truth. For instance, one challenge to the statement “The US should reduce its military budget” was “Russia has nearly twice as many active nuclear weapons as the United States”. In truth, according to statistics published by the Federation of American Scientists: Status of World Nuclear Forces (www.fas.org (http://www.fas.org/)) in 2013, Russia has approximately 1,740 active nuclear warheads, while the United States has approximately 2,150. Examples of the challenges are provided in Table S4 (http://www.nature.com/articles/srep39589#s1).
The political and non-political statements did not differ in number of words (political: 11.22 ± 1.51, non-political: 11.14 ± 1.33, p = 0.97), letters (political: 59.33 ± 7.71, non-political: 58.64 ± 6.04, p = 0.94), or Flesch reading ease (political: 60.7 ± 17.89, non-political: 48.3 ± 28.64, p = 0.26)24 (http://www.nature.com/articles/srep39589#ref24). The political and non-political challenges also did not differ in number of words (political: 20.44 ± 2.83, non-political: 18.92 ± 1.13, p = 0.15), letters (political: 104.18 ± 15.50, non-political: 96.24 ± 6.08, p = 0.15), or Flesch reading ease (political: 53.9 ± 21.02, non-political: 55.74 ± 19.11, p = 0.65).
Because we were interested in brain structures that are known to respond to social and mental stimuli, we used a word counting method to count the frequency of social and cognitive words within the stimuli. This technique is similar to linguistic inquiry word counting (LIWC25 (http://www.nature.com/articles/srep39589#ref25)), but used the open-source software tool TACIT26 (http://www.nature.com/articles/srep39589#ref26) version 1.0.0 in combination with the LIWC 2007 dictionary to count words in the social and cognitive process categories. We found that such words were infrequent in our stimuli, and similar across the two categories (social words occurred with a frequency of 5.07% in the political challenges and 5.35% in the non-political challenges; cognitive words occurred with a frequency of 10.14% in the political challenges and 11.9% in the non-political challenges).

Experimental Procedure
In preparation for the study, participants filled out a survey of demographic information, answered questions about their political and religious affiliations, and indicated the degree to which they agreed with political and non-political statements. Only statements for which participants chose 6 or 7 (where 1 was strongly disagree and 7 was strongly agree) were used during their scan. If a given subject strongly believed more than 8 statements in a category, the statements were chosen for that subject as follows: first, preference was given to more strongly held beliefs (7 vs. 6). Second, all else being equal, preference was given for statements that were not as commonly believed, in order to balance the frequency of statements in the experiment.

When participants arrived for their fMRI scan, they were given instructions and were given the opportunity to ask questions of the experimenter. After the instructions, they performed a practice task, which consisted of a shortened version of one trial of the experiment using the statement “Cats make better pets than dogs”. followed by three challenges to that statement. Following the practice task, participants underwent BOLD fMRI. For each participant there were 4 belief-challenging scans (420 seconds each). During the belief-challenging scans, each statement was presented for 10 seconds, followed by a variable delay of 4–6 seconds. Participants were instructed to press a response button when they had read and understood the statement. Five challenges to the original statement were then presented, each for 10 seconds. Again, participants pressed a response button when they had read and understood the challenge. After all five challenges had been presented, the original statement was presented again and participants had 12 seconds to rate their strength of belief in the statement. The participant indicated his or her response via a button press on an MRI-compatible button box held in the right hand. They pressed buttons to move a cursor left and right along a Likert scale to indicate the strength of their belief on a scale from 1 (strongly disbelieve) to 7 (strongly believe). The cursor started in the middle position of the scale. Two political and two non-political statements were presented in each of the four fMRI scans. The order of these conditions was randomized within each scan, and the statements within each condition were assigned random positions within the experiment for each subject. The temporal structure of the trials and runs is depicted in Fig. S1 (http://www.nature.com/articles/srep39589#s1).

Following the fMRI session, participants filled out a short questionnaire. They were asked to rate how credible they found the challenges in general, and how challenging they were to their beliefs. Participants did not make separate ratings per item or per category, but rather answered these questions about their reaction to the stimulus set in general. During the debriefing, subjects were given a packet of sourced information which detailed the truth of each challenge they read inside the scanner and provided resources on where to find further information.

MRI Scanning
Imaging was performed using a 3T Siemens MAGNETON Trio System with a 12-channel matrix head coil at the Dana and David Dornsife Neuroscience Institute at the University of Southern California. Functional images were acquired using a gradient-echo, echo-planar, T2*-weighted pulse sequence (TR = 2000 msec, one shot per repetition, TE = 25 msec, flip angle = 90°, 64 × 64 matrix, phase encoding direction anterior to posterior, GRAPPA acceleration factor = 2, fat-sat fat suppression). Forty slices covering the entire brain were acquired with an in-plane voxel resolution of 3.0 × 3.0 and a slice thickness of 3.4 mm with no gap. Slices were acquired in interleaved ascending order, and 210 functional volumes were acquired in each run, not including 3 volumes discarded by the scanner to account for T1 equilibrium effects. A gradient-echo field map was also acquired with the same slices and resolution as the functional images using a Siemens field map sequence (TR = 1000 ms, TE1 = 10 ms, TE2 =  12.45 ms, flip angle = 90°, 64 × 64 matrix).
A T1-weighted high-resolution (1 × 1 × 1 mm) image was acquired using a three-dimensional magnetization-prepared rapid acquisition gradient (MPRAGE) sequence (TR = 2530 msec, TE = 3.13 msec, flip angle = 10°, 256 × 256 matrix, phase encoding direction right to left, no fat suppression). Two hundred and eight coronal slices covering the entire brain were acquired in interleaved order with a voxel resolution of 1 × 1 × 1 mm. We also collected a T2-weighted anatomical scan (TR = 10,000 ms, TE = 88 ms, flip angle = 120°, 256 × 256 matrix) with 40 transverse slices with a voxel resolution of 0.82 × 0.82 × 3.5 mm that was reviewed by a radiologist to rule out incidental findings.

fMRI Data Analysis
fMRI analysis was performed using FEAT version 6.00, FSL’s fMRI analysis tool (FMRIB’s Software Library http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/) and other FSL tools from FSL version 5.0.8. Data were first corrected for magnetic field inhomogeneities using the field maps acquired for each subject and FSL’s FUGUE utility for geometrically unwarping EPIs, unwarping in the anterior-posterior (−y) direction with a 10% signal loss threshold. Data were then preprocessed using standard steps in the following order: motion correction using a rigid-body alignment to the middle volume of each run, slice-timing correction using Fourier-space time-series phase-shifting, removal of skull using FSL’s BET brain extraction tool, 5 mm FWHM spatial smoothing, and highpass temporal filtering using Gaussian-weighted least-squares straight line fitting with a sigma of 60 s (corresponding to a period of 120 s). Finally, temporal autocorrelation was removed using FSL’s prewhitening algorithm before statistical modeling27 (http://www.nature.com/articles/srep39589#ref27).

The skull was removed from the T1 images using the BET brain extraction tool with a fractional intensity thresholding setting of 0.4, and specifying the voxel that represented the approximate center of the brain. We then used FLIRT to register the functional images to the skull-stripped T1-weighted MP-RAGE using its boundary-based registration (BBR) algorithm. Next, the MP-RAGE was registered to the standard MNI atlas with a 12 degrees of freedom affine transformation, and then this transformation was refined using FNIRT nonlinear registration with a warp resolution of 10 mm28 (http://www.nature.com/articles/srep39589#ref28).

Data were then analyzed within the General Linear Model using a multi-level mixed-effects design. Each component of the task (statement, challenge, and rating) was modeled by convolving the task design with a double-gamma hemodynamic response function with a phase of 0 s. The task periods were defined from stimulus onset to stimulus offset. Political and non-political trials were modeled using separate regressors, yielding six task regressors. The temporal derivative of each task regressor and six motion correction parameters were also included in the design. At the individual subject level, statistical maps were generated for each functional scan. These were then combined into individual participant-level maps in a fixed effects analysis across each subject’s four scans. Subject-level maps were then entered into a higher-level group analysis to examine group-level effects using a “mixed effects” design with FLAME1.

To explore the relationship between the degree of belief change and brain activity in response to specific statements, we also performed a whole brain item-wise analysis. In this analysis, we first modeled each lower-level run with a design that specified a single regressor for each trial’s statements and challenges. Therefore, in this design, there were 8 task regressors (4 statements and 4 challenge periods) in addition to the six motion parameters. Task periods were modeled as in the previous analysis, using the time from stimulus onset to offset convolved with a double-gamma hemodynamic response function with phase 0 s. We then computed brain-activity maps for each specific stimulus item, combining across all subjects who read that stimulus using a second-level FLAME1 “mixed effects” design to produce item-level activity maps. These item-level activity maps were then tested for correlation with the average belief change across items in a third-level FLAME1 design that included belief change as a between-items covariate.

For all whole-brain analyses, statistical thresholding was performed using FSL’s cluster thresholding algorithm to control for multiple comparisons. This algorithm uses Gaussian Random Field Theory to estimate the probability of clusters of a given size taking into account the smoothness of the data. We used a Z threshold of 2.3, and a cluster size probability threshold of p < 0.05.

In addition to whole brain analysis, we performed a region of interest (ROI) analysis focusing on a-priori ROIs in the amygdala and insular cortex. We chose these two ROIs because of their well-known roles in emotion and feeling. For this analyis, beta values from the GLM analysis were extracted for each subject, and averaged within each ROI. The contrast used for this analysis combined activity from the period when participants were reading all political and nonpolitical challenges to their beliefs. Because there was very little belief change for political statements, we used belief change on non-political statements as our measure of individual variability. The beta values and the average belief change scores were subjected to a Shapiro-Wilk test for normality. These values were then correlated with each participant’s average belief change score in a Pearson’s correlation. The regions of interest were defined as follows: For the amygdala, we used the Harvard-Oxford Atlas amygdala mask, thresholded at 25. For the insula, we used masks of the dorsal anterior, ventral anterior, and posterior insula defined by a study that performed a cluster analysis of functional connectivity patterns29 (http://www.nature.com/articles/srep39589#ref29).

Followup Questionnaire
Following their participation in the fMRI portion of the study, participants were sent an on-line questionnaire asking them to indicate how strongly they agreed with each statement they had seen during their fMRI scan. The average time between a participant’s scan and completing the questionnaire was 48.36 ± 5.85 days.

[Petros Agapetos]

ResultsBehavioral results: Belief changeAfter reading the challenges to the statements, participants’ strength of belief in the statement decreased more for non-political statements than for political statements (Fig. 1A (http://www.nature.com/articles/srep39589?utm_source=Main+List&utm_campaign=9357a723ab-EMAIL_CAMPAIGN_2016_12_27&utm_medium=email&utm_term=0_f1c2a2c9db-9357a723ab-207309801&mc_cid=9357a723ab&mc_eid=969ecf126b#f1), political: 0.31 ± 0.06, non-political: 1.28 ± 0.11, t(39) = 9.76, p < 0.001). Additionally, the degree of belief change for political statements was correlated with the degree of belief change for non-political statements across subjects (Fig. 1B (http://www.nature.com/articles/srep39589?utm_source=Main+List&utm_campaign=9357a723ab-EMAIL_CAMPAIGN_2016_12_27&utm_medium=email&utm_term=0_f1c2a2c9db-9357a723ab-207309801&mc_cid=9357a723ab&mc_eid=969ecf126b#f1), r = 0.52, p = 0.001). In a follow-up questionnaire several weeks after the scan, participants’ strength of belief was still lower than their original strength of belief in the pre-test for both the political and non-political statements (political: 0.20 ± 0.05, t(33) = 3.62, p = 0.001; non-political: 0.75 ± 0.1, t(33) = 7.82, p < 0.001). There was no difference between belief strength in the follow-up compared to during the fMRI scan for political statements (0.12 ± 0.06, t(33) = 1.83, p = 0.076), however there was a difference between belief strength in the follow-up compared to during the fMRI scan for non-political statements (0.51 ± 0.13, t(33) = 4.07, p < 0.001). When separated by statement, average belief change across the 40 participants varied from 0.07 (abortion) to 0.32 (Thomas Edison). Generally, political statements showed the smallest degree of belief change.

[Petros Agapetos]

http://i65.tinypic.com/eke8z.jpg

[LeBrok]

Can you tell us in your own words what we learned from this experiment?