Effects of focal brain damage on political behaviour across different political ideologies

Abstract

Intense political behaviour is associated with brain regions involved in emotional and cognitive processing. However, it remains unclear if this neuroanatomy is causal, compensatory or otherwise correlated.

We employed lesion network mapping in a cross-sectional study of 124 male military veterans with penetrating head trauma. Approximately 40–45 years after the injury, participants reported current political behaviour and recollection of political behaviour pre-injury. Using a normative connectome database (n = 1000), we mapped the circuitry functionally connected to lesions associated with changes in intensity of political involvement, ideological polarity and party affiliation.

No significant neuroanatomical circuit was associated with political ideology or party affiliation, but a distinct circuit was associated with intensity of political involvement. Political involvement was more intense after lesions connected to the left dorsolateral prefrontal cortex and posterior precuneus, in the full sample and in conservative-leaning participants. Political involvement was less intense after lesions connected to the amygdala and anterior temporal lobe, in the full sample and in liberal-leaning participants. These effects survived cross-validation in the full sample (P = 0.01) and in both conservative-leaning and liberal-leaning participants.

These findings may inform cognitive mechanisms of political behaviour as well as clinical assessment after brain lesions.

Using lesion network mapping in military Veterans with a history of head trauma, Siddiqi et al. identify a distinct pattern of brain damage associated with increased intensity of political involvement across both liberals and conservatives.

Introduction

Understanding the neural processes involved in political behaviour could inform how policymakers and political candidates develop approaches to influence their constituents, political rivals and others. If the underlying neuroanatomy can be discretely localized, it may also suggest that political behaviour could be informative in patients with neuropsychiatric disorders.

Political neuroscience has lent preliminary insights into the neuroanatomy of political behaviour and views. Conservatism has been associated with lesions and hypoactivity in the dorsolateral prefrontal cortex (DLPFC),^1,2 which is believed to be responsible for cognitive control and executive function.³ Liberalism, by contrast, has been consistently associated with a smaller amygdala on volumetric MRI,^4,5 but similar localization was not found in a lesion study.² However, most lesion studies have used relatively small sample sizes and have recruited a disproportionately liberal population.⁵ This has limited the ability to disentangle variables such as political views versus political involvement. This distinction is likely important, as polarized views may be meaningless if a person does not act on them, but political involvement can still be meaningful across the spectrum of liberal, moderate or conservative ideologies.

Some political neuroscience studies have attempted to distinguish variables more precisely using functional neuroimaging.⁶ However, these techniques are designed to detect correlations between neuroanatomy and behaviour, and do not reveal the direction of causality.⁷ Lesion studies can provide causal insights by studying patterns of focal brain damage associated with a particular behaviour,⁶ but are limited in statistical power because most brain lesions do not overlap each other.⁷ This limitation is partly addressed by lesion network mapping, a technique that uses the human connectome as a wiring diagram to map the whole-brain circuitry connected to any given lesion, thus providing data for every participant at every brain voxel.⁸ The resulting increase in statistical power can be used to disentangle variables more precisely with rigorous statistical testing.

Here, we used lesion network mapping to study intensity of political involvement across different ideological poles and party affiliations. We hypothesized that lesions connected to a distinct brain network would be selectively associated with changes in political behaviour.

Materials and methods

Study population and assessment

As part of phase 4 of the Vietnam Head Injury Study (VHIS),^9,10 124 male US military veterans completed an extensive behavioural testing battery over a 5-day period between 2008 and 2012, approximately 40–45 years after sustaining penetrating head trauma. A total of 35 control participants experienced similar combat exposure, but did not sustain a brain injury. The testing battery included a political questionnaire in which participants were asked to rate their intensity of political involvement and their political views, including ideological polarity and party affiliation.

The primary outcome was intensity of political involvement, which was assessed using three items that are listed in Supplementary Table 1. These three items covered the level of interest in politics, frequency of following politics in the media and frequency of discussing politics with others. Items were reverse scored such that higher scores correspond to greater political involvement. A composite score was calculated as the average of the three items, and was inverted such that higher scores represent greater political involvement.

Ideological polarity was self-rated (as described in Supplementary Table 2) on a scale of 1 (extremely liberal) to 4 (moderate) to 7 (extremely conservative). Party affiliation was self-rated on a scale of 1 (extremely Democratic) to 4 (Independent) to 7 (extremely Republican). Participants also completed the Armed Forces Qualification Test (AFQT), a standardized cognitive test that was completed at the time of military enlistment and in subsequent evaluations, as a measure of general cognitive aptitude.¹¹ The AFQT score represents the participant’s percentile relative to other examinees who were attempting to enlist in the military. Examinees with low scores are not accepted into the military, so average scores for enlisted soldiers are higher than the 50th percentile. Scores in the 60–70th percentile approximately correspond to average intelligence relative to the general population.¹¹

We also analysed a political judgement task, a 75-item questionnaire that asks participants to rate several statements along three dimensions: conservatism, individualism and radicalism.

Lesion network mapping

Lesion localization was conducted using a CT scan that was collected during phase 3 of the study 5–10 years before the present assessment. Lesions were manually traced on 1-mm-thick slices in Analysis of Brain Lesion (ABLe) software.⁹ Each lesion trace was personally reviewed by the senior author of the present manuscript (J.H.G.).

Resting-state functional connectivity of each lesion was estimated using a normative connectome database as in our prior work.¹² Each participant’s lesion location was used as a seed to compute mean connectivity of the lesion to each brain voxel across 1000 healthy participants. At each voxel, lesion connectivity was compared with behavioural outcomes, including intensity of political involvement, using the partial Spearman correlation. Covariates included age, education and AFQT score (a proxy for intellectual aptitude), as these variables have previously been shown to be correlated with political involvement.^13-15 This yielded a map depicting the connectivity profile of lesions that are most associated with changes in political involvement. To ensure that the results were not driven by covariates, we repeated the analysis without controlling for age/education/AFQT, and also repeated the analysis after additionally controlling for pre-lesion political involvement, pre/post-lesion ideological polarity and pre/post-lesion party affiliation.

This analysis was also conducted using lesion connectivity to distinct brain networks rather than connectivity to every single brain voxel. Each lesion’s connectivity profile was compared to seven canonical resting-state brain networks, as defined by Yeo et al.,¹⁶ using spatial correlations. For each participant, this yielded seven correlation values representing how much their lesion connectivity profile overlaps with each network. Across all participants, lesion connectivity to the seven networks was Fisher transformed and compared with intensity of political involvement as above. Bonferroni correction was applied for seven comparisons.

The same analyses were also conducted for polarity of political ideology, party affiliation and the three items on the political judgement task.

Statistical validations

We used three procedures to ensure that the results were stronger than chance: voxel-level family-wise error (FWE) correction using the permutation-based Westfall–Young max t-stat procedure,¹⁷ network-level FWE correction using the Bonferroni procedure¹⁸ and out-of-sample cross-validation using a leave-one-out procedure.

Voxel-wise Westfall–Young correction was implemented as described by Winkler et al.¹⁷ The lesion network mapping procedure was repeated after randomly permuting each participant’s behavioural outcomes against a different participant’s neuroimaging results. The absolute peak voxel value of the resulting map was recorded. This procedure was repeated 10 000 times, yielding a distribution of peak voxel intensities expected by chance. The 95th percentile of this distribution was treated as a cutoff to determine which voxels are stronger than chance.

Network-level Bonferroni correction was applied based on the P-values generated from the correlation of political involvement with lesion connectivity to the Yeo networks. The resulting P-values were corrected for seven comparisons (i.e. seven Yeo networks).

Leave-one-out cross-validation was applied by re-generating the lesion network map after excluding one participant. The resulting leave-one-out map was compared with the left-out lesion connectivity map using spatial correlations. This procedure was repeated for all 124 participants, yielding 124 leave-one-out spatial correlation values. These spatial correlations were Fisher transformed and compared with the political involvement score. To test specificity, this correlation was also computed for multiple control measures.

Exploratory subgroup analyses

To explore the effect of different political ideology or party affiliation, we conducted two types of subgroup analysis. First, participants were classified according to political ideology as either liberal, moderate or conservative. Second, participants were classified according to party affiliation as either Democrat, Independent or Republican.

Of note, this exploratory analysis was underpowered relative to the primary analysis. For this reason, a more liberal multiple comparisons correction approach was used for whole-brain lesion network analyses. Instead of permutation-based FWE correction, we employed cluster extent thresholding with a detection threshold of P < 0.001 and minimum cluster size of 240 mm³, following the recommended cluster size threshold for functional MRI analysis with 2 mm isotropic voxel size and 6-mm smoothing kernel.¹⁹

Results

Sample characteristics

Participant demographics are listed in Table 1. For the three items in the political involvement survey, internal consistency was in the acceptable range (Cronbach’s α = 0.738). Thirty-three participants identified as liberal, 30 as moderate and 61 as conservative. Age and income were similar between the three groups. Education was higher in the liberal group (P = 0.024) and aptitude test scores trended towards higher scores in the liberal group (P = 0.054). On a scale of 0 to 4, mean political involvement was 3.03 ± 0.78 after the lesion, and 1.56 ± 0.92 before the lesion. In control participants, political involvement was 3.06 ± 0.91 after the lesion, and 1.55 ± 0.79 before the lesion (P = 0.50). Item-level scores are summarized in Supplementary Table 3.

Table 1.

Participant demographics by ideological pole

	Liberal (n = 33)	Moderate (n = 30)	Conservative (n = 61)	ANOVA
Age, years	58.1 ± 2.4	58.0 ± 2.3	58.4 ± 3.33	P = 0.79
Education, years	15.7 ± 1.8	14.1 ± 2.6	14.9 ± 2.23	P = 0.024
Income, rangea	$50 000 to $75 000	$50 000 to $75 000	$50 000 to $75 000	n/a
Aptitude testb	74.0 ± 20.0	62.2 ± 23.3	67.4 ± 20.2	P = 0.054
Party affiliation	22 Democrat 9 Independent 2 Republican	14 Independent 10 Democrat 6 Republican	38 Republican 12 Democrat 11 Independent	n/a
Political involvement	3.23 ± 0.68	2.54 ± 0.90	3.16 ± 0.67	P = 0.0003***

^aParticipants were asked to report their income as a range; the median range is reported here for each group.

^bAptitude test scores measured using the Armed Forces Qualification Test, which was repeated at the time of the phase 4 assessment.

***Post hoc testing shows that this result is driven by moderate versus liberal (P = 0.0003) and moderate versus conservative (P = 0.0002), but not liberal versus conservative (P = 0.67).

Of note, ideological polarity and party affiliation were not synonymous. Of the 44 participants identifying as Democrat, 12 participants (27%) reported conservative ideology. Consequently, ideological polarity and party affiliation were analysed independently of each other.

Lesion network mapping

Lesions were warped to a common MNI152 template space (Fig. 1A). The overall lesion overlap is depicted in Supplementary Fig. 1. Each lesion’s whole-brain connectivity profile (Fig. 1B) was compared with intensity of political involvement using voxel-wise partial Spearman correlation, controlling for age, education and AFQT. This yielded a whole-brain map representing connectivity of lesions that preferentially influence intensity of political involvement (Fig. 1C).

Figure 1.

Lesion network mapping procedure. (A) Lesions were manually traced using a head CT scan. (B) Whole-brain connectivity of each lesion was estimated using a normative connectome database. (C) Across all participants, lesion connectivity to each voxel was compared with political intensity, yielding a map of connectivity of lesions that selectively modify political intensity.

This whole-brain map was corrected for multiple comparisons using the permutation-based Westfall–Young method (Fig. 2A),¹⁷ which showed that the peak voxel value was stronger than expected by chance (P_FWE = 0.002). The results remained significant when dropping all covariates (P_FWE = 0.017) or when additionally controlling for pre-lesion political involvement (P_FWE = 0.021), pre- and post-lesion ideological polarity (P_FWE = 0.019), or pre- and post-lesion party affiliation (P_FWE = 0.018).

Figure 2.

Network localization of political intensity. (A) Greater political intensity (warm colours) was associated with lesions connected to the left dorsolateral prefrontal cortex and the posterior precuneus. Less political intensity (cool colours) was associated with lesions connected to the bilateral superior temporal sulcus, amygdala and anterior temporal lobe. (B) Greater lesion connectivity to the frontoparietal control network (orange) was associated with greater political intensity. FPCN = frontoparietal control network.

More intense political involvement was associated with lesions connected to the left DLPFC, posterior precuneus and left superior parietal lobule (P_FWE < 0.01). Less intense political involvement was associated with lesions connected to the bilateral amygdala, anterior temporal lobe and superior temporal sulcus (P_FWE < 0.01). This effect was unchanged when adding self-reported pre-lesion political involvement as a covariate. Thus, a discrete brain circuit was connected to lesions that modify intensity of political involvement. For convenience, we henceforth describe this as the political intensity network. Several other regions were also significantly associated with political involvement at a more liberal statistical threshold (P < 0.05), and are listed in Supplementary Table 4.

To assess whether this effect was constrained to specific large-scale brain networks, we estimated each lesion’s connectivity to the seven consensus large-scale networks.¹⁶ After Bonferroni correction for seven comparisons, more intense political involvement was associated with lesions connected to the frontoparietal control network (FPCN) (Fig. 2B), whether controlling for age/education/AFQT (r = 0.30, P_FWE = 0.005) or not (P_FWE = 0.014). The results also remained unchanged when additionally controlling for pre-lesion political involvement (P_FWE = 0.020), pre- and post-lesion ideological polarity (P_FWE = 0.014), or pre- and post-lesion party affiliation (P_FWE = 0.004). No other resting-state network showed significant results across all these analyses. These results suggest that lesions to the FPCN may intensify political involvement.

The above analyses were also repeated for political views, including ideological polarity and party affiliation, in lieu of intensity of political involvement. Neither ideological polarity nor party affiliation were significantly associated with lesion connectivity to any region (P_FWE > 0.47) or any network (P_FWE > 0.45). Thus, unlike political involvement, ideological polarity (liberal-moderate-conservative) and party affiliation (Democrat-Independent-Republican) did not localize to a discrete brain circuitry. After false discovery rate (FDR) correction for six comparisons (whole-brain and Yeo network analyses for political involvement, polarity and affiliation), the associations with political involvement remained significant (whole-brain peak P = 0.012, FPCN network-level P = 0.015).

We also repeated the whole-brain analysis using the political judgement task instead of questionnaires. This task was not used for the primary analysis because pre-lesion status was not assessed. The task revealed no significant localization for implicit conservatism (P_FWE > 0.08), individualism (P_FWE > 0.13) or radicalism (P_FWE > 0.61).

Cross-validation and specificity

To confirm that the network-level results explain out-of-sample variance, we used a leave-one-out cross-validation.²⁰ The overall political intensity network (Fig. 2A) was re-generated after excluding each participant, one at a time (Fig. 3A). Across all participants, the similarity of the left-out lesion to the leave-one-out map was associated with post-lesion political involvement, controlling for age, education and AFQT (r = 0.31, P = 0.001) (Fig. 3B). Thus, lesion location with respect to the leave-one-out political intensity network explained r² = 10% of the variance in political involvement, which is generally considered to be a ‘moderate’ association.²¹ The results remained unchanged when additionally controlling for pre-lesion political involvement (r = 0.27, P = 0.002), pre- and post-lesion ideological polarity (r = 0.31, P < 0.001), or pre- and post-lesion party affiliation (r = 0.31, P < 0.001). These findings demonstrate that lesion overlap with the political intensity network explains significant variance in intensity of political involvement, independent of ideology.

Figure 3.

Cross-validation and specificity. (A) The network mapping procedure was repeated after excluding one participant at a time. (B) The resulting leave-one-out map was compared with the left-out map (C) using spatial correlations. This procedure was then iterated across all participants. (D) Similarity of each participant’s lesion connectivity to the leave-one-out map was correlated with intensity of political involvement, adjusting for age and education. (E) The leave-one-out spatial correlation was more associated with political involvement than with 39 control measures.

To confirm specificity, the leave-one-out spatial correlation was also compared with several control metrics. These control metrics included ideological polarity (from extremely liberal to extremely conservative), party affiliation (from extremely Democrat to extremely Republican), emotion contagion (propensity towards being influenced by others’ positive or negative emotions),²² five personality dimensions (based on the Neuroticism/Extraversion/Openness five-factor inventory)²³ and 27 other neuropsychiatric symptoms (based on clinician-rated neurobehavioural rating scale).²⁴ Lesions to the political intensity network were associated with more conservative ideology (r = 0.19, P = 0.04) and propensity towards negative emotional contagion (r = 0.19, P = 0.05), but these results did not survive multiple comparisons correction. Item-level analyses are summarized in Fig. 3E. In addition to the item-level scores on the neurobehavioural rating scale, we repeated the same analysis for conglomerate factor scores²⁴ representing metacognition, somatic/anxiety and cognition/energy, which similarly yielded non-significant results (r = 0.11, r = 0.09 and r = −0.02, respectively).

We also tested whether political involvement was correlated with any of the 38 control metrics, irrespective of lesion location. Political involvement was positively correlated with extraversion (r = 0.26, P = 0.005) and negatively correlated with emotional withdrawal (r = −0.26, P = 0.006). Neither of these correlations survived FDR correction. Of note, neither extraversion nor emotional withdrawal mapped to circuits that were similar to the political intensity circuit (spatial r = −0.003 and r = −0.09, respectively).

Influence of political views

Next, we conducted subgroup analyses to explore the effect of ideological polarity (liberal, moderate or conservative) and party affiliation (Democrat, Independent or Republican) on the political intensity network. The effect size of cross-validation was similar across all six of these subgroup analyses (Fig. 4A). This relationship was statistically significant in four of the six subgroups—the two exceptions were those with liberal ideology (P = 0.07) or Independent party affiliation (P = 0.20). Thus, the political intensity network was not specifically driven by any individual subgroup of ideological polarity or party affiliation.

Figure 4.

Subgroup analyses according to ideological polarity and party affiliation. (A and B) In the leave-one-out cross-validation, similar effect sizes were seen across all three ideological groups and all three party affiliations, although this effect did not reach significance for liberal ideology or independent party affiliation. (C) Peak regions connected to lesions affecting political involvement differed by ideology and party. In participants identifying as liberal or Democrat, political involvement decreased with lesions connected to the left superior temporal sulcus. In participants identifying as conservative or Republican, political involvement increased with lesions connected to the left dorsolateral prefrontal cortex.

We re-generated the political intensity network in each subgroup and identified peaks using cluster extent thresholding (detection P < 0.001, cluster extent > 240 mm³). In participants identifying as liberal, political involvement decreased after lesions connected to the superior temporal sulcus, amygdala and anterior temporal lobes, with a larger cluster on the right side than the left side. A similar localization was seen in those identifying as Democrat but was limited to the right superior temporal sulcus. In participants identifying as conservative, political involvement increased after lesions connected to the left DLPFC. A similar localization was seen in those identifying as Republican, but it also included the left lateral orbitofrontal cortex and the left posterior cingulate cortex. Individual cluster locations and sizes are listed in Supplementary Table 5.

Directionality of the circuit

Our results may be driven by lesions that increase political involvement, lesions that decrease involvement or both. Given that lesion location was related to political involvement but political involvement was similar between brain-injured participants and controls, this suggests that the result was driven by both increases and decreases. To explore this possibility, we conducted a post hoc analysis in which participants were split into three equal-sized groups based on magnitude of connectivity to the leave-one-out circuit. One-way ANOVA showed that these three groups significantly differed in terms of political involvement (P = 0.012, Supplementary Fig. 2). Political involvement was significantly higher in the positive connectivity group [3.29, 95% confidence interval (CI) 3.08–3.50] relative to the negative connectivity group (2.79, 95% CI 2.52–3.05) (P = 0.004). Political involvement in participants with relatively neutral connectivity (3.01, 95% CI 2.79–3.24) was similar to controls with no brain injury (3.06, 95% CI 3.35–2.78). This suggests that lesions to the positive part of the circuit can increase political involvement, lesions to the negative part of the circuit can decrease political involvement and lesions with relatively neutral connectivity were similar to controls with no lesion.

Discussion

Here, we identified a common brain circuit connected to lesions that modify the intensity of political involvement across both ends of the political spectrum, although the neuroanatomical peaks differed between those identifying as conservative/Republican versus liberal/Democrat. This localization was broadly consistent with prior political neuroscience research,^2,6,25,26 but included a larger sample of conservative participants and a larger overall sample, employed causal lesions with more sensitive network-level inference techniques and controlled for pre-lesion behaviour. The resulting increase in statistical power enables specificity analyses and subgroup analyses to disentangle political views from political involvement and other relevant variables.

This increased statistical power enabled highly rigorous inferential approaches such as permutation testing, Bonferroni correction and cross-validation.^17,20,27 First, permutation testing confirmed that the peak locations in the circuit were stronger than chance. Second, analysis of seven large-scale brain networks with Bonferroni correction confirmed that the involvement of the frontoparietal control network is unlikely to be a false positive from a multiple comparisons artefact. Third, leave-one-out cross-validation confirmed that the overall circuit explained out-of-sample variance in political involvement. This effect was specific to political involvement relative to other metrics, but was not specific to any one ideological pole or party affiliation. These statistical validations lend high confidence in our anatomical localization.

Political involvement was more intense after lesions connected to the DLPFC and the posterior precuneus, both of which are functionally heterogeneous regions.²⁸ The DLPFC and precuneus subregions identified here are part of the FPCN, which is believed to manage cognitive control and executive function.²⁹ In a post hoc exploration of this DLPFC peak, it was specifically detected in those identifying as conservative or Republican, consistent with prior studies showing that conservative ideology was associated with lesions and hypoactivity in the DLPFC.^2,6 Furthermore, conservatism appears to be expressed more intensely in alcohol intoxication and time pressure,³⁰ conditions that impede cognitive control. Of note, prior analyses of the VHIS have found that DLPFC lesions may cause depression,³¹ impaired social perception,³² reduced language coherence,³³ more positive implicit associations with violence,³⁴ increased Machiavellianism,³⁵ reduced working memory³⁶ and greater fundamentalism.³⁷

Political involvement was less intense after lesions connected to the amygdala and anterolateral temporal lobes. The amygdala is involved in fear to external stimuli, while the anterolateral temporal lobes are believed to be involved in empathy.^38,39 In a post hoc exploration of the amygdala and anterolateral temporal lobe peaks, they were particularly notable in those identifying as liberal or Democrat, consistent with prior work showing that smaller amygdala volume was associated with liberal political views⁴ and dissatisfaction with the existing social order.²⁵ Prior research also suggests that liberalism may be associated with a greater desire to feel empathy,⁴⁰ and empathic rhetoric can increase voting behaviour in Democrats.⁴¹

Importantly, even if these strategies successfully engage individuals with different political views, our results do not necessarily imply that conservatism is driven by cognitive impairment or that liberalism is driven by self-righteous posturing. Political views, including ideological polarity and party affiliation, were not associated with any detectable neuroanatomy and did not mediate the association between neuroanatomy and political involvement. While this finding may appear at odds with prior studies suggesting that lesions may affect political ideology,^2,25,42 most prior studies did not disentangle the polarity of viewpoints from the expression of the viewpoints. Our results build on this prior work by defining a lesion pattern that differentially modifies the expression of pre-existing views rather than modifying the views themselves.

There are many possible cognitive factors driving this observation. The prefrontal cortex generates doubt and incredulity, so lesions in this region may consequently increase susceptibility to misleading or authoritarian viewpoints.^43,44 Relatedly, damage to regions involved in cognitive control and empathy may increase group polarization, the process by which a group’s behaviour shifts in a more extreme direction.^42,45 This may provide a neurocognitive explanation for the longstanding observation that political viewpoints tend to be driven by sociocultural identity rather than cognitive processes.⁴⁶ The present study did not directly test cognitive mechanisms, but may help guide hypotheses for future studies. For instance, we would hypothesize that interventions focused on cognitive or rhetorical reasoning may be useful for modifying expression of pre-existing viewpoints, while interventions focused on sociocultural identity may be more useful for modifying the viewpoints themselves.

Limitations of this study include the homogeneous sample of older male US military veterans who were more likely to identify as conservative than liberal; it is unclear if all the findings are generalizable to other populations. This is particularly notable given that the assessments were conducted between 2008 and 2012, a time during which the US political climate was becoming increasingly polarized. Thus, it is plausible that our localizations reflect increased susceptibility to broader societal trends rather than cognitive mechanisms involved in political behaviour. These concerns are partly mitigated by similar localizations identified in prior studies of male and female participants who disproportionately identified as liberal and had a shorter time lag between imaging and political assessment.^2,4 Nevertheless, future studies may attempt a similar localization in other populations, particularly given that political candidates often develop distinct strategies for influencing voters in different demographic groups.⁴⁷

Of note, while lesion studies are valuable for many reasons, they also carry inherent limitations—for instance, the participant could not be assessed before the lesion. Thus, our analysis assumes that lesion location is randomly distributed with respect to pre-lesion political involvement, which may introduce noise into the analysis and limit the explainable variance in the model. Although we controlled for the participant’s recollection of their pre-lesion behaviour, this may be subject to recall bias. There are also other potential sources of noise; for instance, the lesion sample and the normative connectome sample were not matched for age and sex, which may introduce another source of variance. We observed a moderate association, explaining 10% of out-of-sample variance in political involvement, despite these limitations that attenuate the effect size. Future studies may address these inherent limitations by prospectively applying focal brain stimulation to the lesion-derived circuit. When lesions to a particular circuit cause a particular behaviour, stimulation of the same circuit tends to reduce the same behaviour.^7,48-52 Thus, we would hypothesize that stimulation of our lesion-derived circuit selectively leads to changes in political involvement. If so, this may have important ethical and regulatory implications. It may also be an unrecognized side effect of therapeutic brain stimulation. A stimulation study to test this hypothesis is now underway (NCT06376734).

This study may also have clinical implications. Currently, questions about political behaviour are not part of a standard neuropsychiatric assessment.⁵³ While this omission may be partly attributable to the personal and sensitive nature of political behaviour, the clinical assessment already includes sensitive topics such as sexual trauma, domestic violence, criminal history, substance abuse, hypersexuality, suicidality and access to weapons. If political involvement is differentially influenced by damage to circuits involved in cognition and emotion, this entails that different changes in political involvement may be seen in different neuropsychiatric disorders. These changes were not captured by any of the 38 control metrics assessed in this study—although political involvement was correlated with extraversion and emotional withdrawal, it did not map to similar circuitry. Future research may seek to better characterize how changes in a patient’s political behaviour may influence differential diagnosis and treatment, especially if such changes influence family dynamics.

Overall, the implication of different brain regions in intensity of liberal versus conservative political involvement may provide a neuroscientific explanation for the prevailing view that these two groups respond to different campaign strategies. Specific lesion patterns led to changes in intensity of political involvement, but not changes in political views. These results align with cognitive group polarization studies that may guide political strategists in deciding how to influence different groups of voters. However, such strategies may be more useful for intensifying pre-existing viewpoints rather than influencing a voter to change their viewpoint. Although further research is needed to test these hypotheses directly, our neuroanatomical localization may inform how political strategists plan their approach to influencing others and how clinicians approach patients with focal brain changes.

Data availability

The political intensity map, the leave-one-out cross-validation code can be found at http://siddiqi.bwh.harvard.edu/data-code, along with code and maps generated for our other papers. Raw lesion datasets cannot be shared publicly because each lesion is unique and thus may be identifiable, but questions about the VHIS can be directed to Dr Grafman.

Funding

The present work was supported by the National Institute of Mental Health (Grant No. K23MH121657 and R01MH136248 to S.H.S.). S.H.S. has also received funding from non-profit organizations including the Baszucki Family Foundation and the Brain & Behavior Research Foundation, as well as investigator-initiated industry-sponsored funding from BrainsWay Ltd and Neuronetics LLC. The funders were not directly involved in the conceptualization, design, data collection, analysis, decision to publish or preparation of the manuscript.

Competing interests

S.H.S.: owner of intellectual property involving the use of brain connectivity to target transcranial magnetic stimulation, scientific consultant for Magnus Medical, investigator-initiated research funding from Neuronetics and Brainsway, speaking fees from Brainsway and Otsuka (for PsychU.org), shareholder in Brainsway (publicly traded) and Magnus Medical (not publicly traded). None of these entities were directly involved in the present work. The authors report no other conflicts of interest related to the present work.

Supplementary material

Supplementary material is available at Brain online.