Introduction

“When I was a boy of 14, my father was so ignorant I could hardly stand to have the old man around. But when I got to be 21, I was astonished at how much the old man had learned in seven years. (Mark Twain)”

Individuals’ perceptual and interpretative frameworks for understanding the world can vary substantially. Moreover, these frameworks can shift within a single individual over time, as aptly described by Mark Twain’s observation above. This study investigates the neural correlates underlying the dynamic interpretation of real-life events within the context of political identity.

Political identity encompasses multiple characteristics, with political ideology and beliefs widely recognized as its primary components1. Consequently, research examining political identity typically categorizes participants along ideological dimensions, such as “Left”/“Right”2,3,4,5 or “Liberal”/“Conservative”6,7. Such studies have revealed that political identity influences various psychological processes, including in-group identification8, disgust sensitivity9, the formation of false memories regarding politicians10, and the interpretation of politically-charged articles11. Neuroimaging research has further demonstrated that political identity differences manifest in brain activity, mainly within regions in the Default Mode Network (DMN) and the limbic system (e.g.12,13,14,15,16,17,). Studies using naturalistic stimuli have shown that political affiliation modulates neural responses within the DMN during exposure to content about immigration18, abortion19, and political debates20,21. Moreover, a recent study found that different parts of partisan discourse (policy- and identity-based political stimuli) generated partisan-dependent neural responses of regions within the DMN and affective network22. Our recent work has extended these findings, demonstrating that political identity-dependent differences in neural synchronization emerge even in early sensory and motor cortices23.

While previous research has explored how individual differences in political affiliation affect brain responses to political content, the neural correlates of interpretative changes within individuals over time remain unexplored. This gap stems from the typically resistant nature of worldviews and opinions to external influences. However, recent events presented a unique opportunity to study the neural basis of political identity-dependent interpretation changes at the individual level. In 2019, Israel entered a political crisis when no faction could form a government following national elections. This crisis persisted for several years, marked by political deadlock, repeated elections, and significant shifts in party alignments and political leadership. The situation precipitated a departure from traditional “left-right” divisions, giving rise to a new political landscape defined by support for or opposition to specific political figures24,25,26. For example, right-wing leaders Naftali Bennett and Ayelet Shaked initially failed to meet the voters’ threshold but were given another opportunity in the subsequent elections, successfully returned to the Knesset, broke away from Netanyahu’s political bloc, and established a national unity government with Bennett being the prime minister, and with the support of political allies from parties with vastly different ideologies. Many right-wing voters considered Bennett act as a betray, while many left-wing voters from the “Anti-Netanyahu” bloc supported this, although having deep disagreements with Bennett ideological views (see ref. 27). This upheaval in Israel’s political landscape—marked by shifting alliances, ideological realignments, and breaches of voter expectations—offers a rare opportunity to examine how neural responses evolve in tandem with a rapidly transforming political environment. This transformation aligned with two global political phenomena: increasing polarization and personalization28,29,30. In this era of heightened political polarization and personalization, political identity has become increasingly defined by in-group affiliation rather than ideological principles31. The central question has shifted from ideological positioning to group loyalty—essentially, “Are you with us or against us?”.

This study investigated how changes in individual interpretation and emotional reactions stemming from the political crisis manifested in brain responses. We conducted a longitudinal investigation by re-recruiting participants from our April 2019 study23 for a second scanning session in August 2021, during which they viewed the same political videos. We hypothesized that alterations in interpretation would correlate with changes in neural activity, particularly within regions of the Default Mode Network (DMN), based on previous research examining partisan-dependent interpretative differences between individuals18,19,23. Furthermore, given the intensified polarization and personalization that characterized the political crisis, we predicted that changes in neural response would correlate more strongly with shifts in in-group/out-group identification than with changes in abstract ideological positions.

Methods

Participants

This fMRI study focused on two rounds of scanning sessions, two and a half years apart. In the first session took part 41 right-handed participants (24 males and 17 females, gender was determined by self-identification, mean age 26.5 ± 5.75). All participants were healthy and had normal or corrected to normal vision. Additionally, all participants were fluent in Hebrew at a mother tongue level. Data on race/ethnicity were not collected. Recruitment was conducted through social networks. Before their participation, participants completed a preliminary questionnaire that included items regarding conditions under which a participant should not undergo scanning (e.g., an irremovable piercing, implanted metal devices, etc.). Furthermore, participants completed a questionnaire that assessed their level of political involvement and their political orientation. We specifically recruited individuals who reported high levels of political involvement and were markedly left-wing or right-wing. This sample size was chosen based on previous studies exploring similarities and differences in neural responses to naturalistic stimuli32,33, as well as based on power analysis of the fMRI analysis method we used34. Out of those 41 participants, four participants were discarded from the analysis: Two due to incidental clinical findings, and two due to excessive head motion (>2 mm).

More than two years later, out of those initial 37 participants, a total of 21 agreed to take part in another scanning session (13 left-wing, 8 right-wing, 12 males, 9 females, mean age 28.05 ± 5.18). The experimental procedures were approved by Tel Aviv University’s Ethics Committee and the Institutional Review Board at the Sheba Medical Center. All participants provided written informed consent and received payment for their time. This study was not preregistered.

Stimuli and experimental design

April 2019 Scan

The first scanning session took place three weeks prior to the April 2019 national elections in Israel, between the dates 03/18/2019 and 04/08/2019 (Fig. 1). Participants watched eight video clips while inside the MRI scanner. In this study, we analyzed the brain response data for seven video clips, which were shown to participants in both April 2019 and August 2021 scans: (i) Short documentary about an individual who converted an old bus into his house (Neutral, 151 s); (ii) Campaign ad by the major right-wing party (MajorRWC, 198 s); (iii) Campaign ad by the minor right-wing party (MinorRWC, 201 s); (iv) Campaign ad by a center party (CWC, 188 s); (v) Campaign ad by left-wing party (LWC, 206 s); and speeches by (vi) Right-wing politician (RWS, 235 s) and (vii) Left-wing politician (LWS, 312 seconds). Each clip was preceded and followed by a gray screen: 8 s before, and 10 s after (which were discarded from the analysis). After each clip, participants were asked three questions: (i) “How much did you agree with the main message of this clip?”, (ii) “How much did this clip interest you?”, and (iii) “How emotionally engaged were you?”. They answered these questions by indicating their ratings (using a magnet-compatible mouse) on a VAS that ranged from “Not at all” (0) to “Very” (100). The order of the clips was randomized into four versions, all of which began with the neutral clip. All political stimuli had similar (and relatively high) amounts of ideological content, in-group\out-group relevance, and polarizing language (see Supplementary Fig. 1).

Fig. 1: Experimental design.
Fig. 1: Experimental design.
Full size image

a Prior to the 2019 first elections, 37 participants were recruited for the experiment. Out of them, 21 returned to the August 2021 scan. b In both scans, participants watched the same varied political videos, and following the scan, they answered a detailed questionnaire related to the videos they watched, and sentiments evaluations towards the characters displayed in the videos. Artistic rendering of the stimulus material generated for display purposes by the authors.

Immediately after the scan, participants completed a behavioral assessment session that was held in a separate, adjacent room. The questionnaire included three types of questions: (i) questions designed to ensure participants were attentive to the videos’ content (27 questions, e.g., “Did Benjamin Netanyahu wore a green tie in the video?”); (ii) questions which directly addressed the agendas or parties displayed in the videos (43 questions, e.g., “Does the Labor party portrayal of the right-wing public as ‘abnormal’ suggest that the left is condescending and out of touch?”); and (iii) questions aimed to assess general sentiments (trustworthiness, pride, hope, love, anger, contempt, hate, anxiety, despair,

apathy, and disgust) towards the seven politicians appearing in the video clips (62 questions).

August 2021 Scan

This scanning session took place approximately two months after the formation of the government, between the dates 08/16/2021 and 09/12/2021. The procedure and stimuli were exactly the same as in the first scan, with one change in the stimuli (pre-election political survey clip was replaced with a video clip containing a news article).

As in the first session, after the scan, participants completed a behavioral assessment session, which was held in a separate, adjacent room. The questionnaire included 109 questions that were the same as in the first session, as well as 28 new questions that directly asked participants to estimate shifts in their perspectives and opinions towards the videos content between the 2019 scan and 2021 scan (e.g., “How much has your opinion about the Labor Party video changed since you first saw it in March 2019?”). Overall, in the August 2021 session, there were 137 questions in the behavioral analysis: 27 questions designed to ensure participants were attentive to the videos’ content, 43 questions which directly addressed the agendas or parties displayed in the videos and 62 questions aimed to assess general sentiments.

MRI data acquisition

Participants underwent two scanning sessions, one in March-April 2019 and the other in August-September 2021, using a 3 T Siemens Prisma scanner with a 64-channel head coil. T1-weighted structural images were acquired using a magnetization prepared rapid gradient echo pulse sequence (MPRAGE), as follows: TR = 2530 ms, TE = 2.88 ms, TI = 1100 ms, flip angle= 7° and 250 Hz/px, isotropic voxel size of 1mm3. For functional scans, images were acquired by means of a T2*-weighted multiband echo planar imaging protocol. Repetition time (TR) = 1000 ms, echo time (TE) = 34 ms, flip angle (FA) = 60°, multiband acceleration factor of six without parallel imaging. Isotropic resolution was 2 mm*2 mm*2 mm (no gaps) with full brain coverage; slice-acquisition order was interleaved.

Preprocessing

Raw DICOM format imaging data were converted to NIfTI with the dcm2nii tool. The NIfTI files were organized according to the BIDS format v1.0.151. fMRI data preprocessing was conducted using the FMRIB’s Software Library’s (FSL v6.0.2) fMRI Expert Analysis Tool (FEAT v6.00) 52. All data were subjected to the following preprocessing procedures: brain-extraction for skull-stripping anatomy image; slice-time correction; high-pass filtering (2 cycles per stimulus’s length); motion-correction to the middle time-point of each run; and smoothing with a 4-mm FWHM kernel. All images were registered to the high-resolution anatomical data using boundary-based reconstruction and normalized to the Montreal Neurological Institute (MNI) template, using nonlinear registration. Blood-oxygen-level-dependent (BOLD) response was normalized (z-scored) within subjects for every voxel for each video clip. Hemodynamic response delay was calculated for each participant according to the peak start time of the BOLD response in early auditory areas (A1+), using the pre-election political survey video clip (which was not part of the stimuli analyzed in this paper). The shift was then calculated as the duration from the stimulus onset to the first peak of the hemodynamic response in A1+ (mean shift = 3.18 s, SD = 0.72). We further analyzed the data only in voxels that were part of the gray matter and that had a reliable BOLD signal (<3000 AU) in at least 90% of the participants across all analyzed videos and sessions. This procedure resulted in 97,332 voxels for all videos and sessions.

Behavioral analysis

To evaluate differences in responses between the two timepoints, for each Interpretation question we performed a Wilcoxon Signed-Rank test, as the data were not normally distributed. Specifically, we tested whether the absolute differences in responses to recurring questions deviated from zero. Additionally, we tested the hindsight question responses from 2021 against zero. We further corrected for multiple comparisons using False Discovery Rate (FDR) correction35.

Interpretation change coefficient (ICC)

To measure the interpretation of the video clips and emotions elicited by it, immediately after each scanning session, participants completed a behavioral assessment questionnaire. The questionnaire included questions that were asked on both April 2019 and August 2021, enabling us to directly evaluate changes in interpretation over time, as well as questions that were asked only on August 2021 in which participants retrospectively assessed changes in their interpretation over time. Those questions pertained both to the evaluation of certain worldviews and to politicians featured in the video clips.

In our analysis, we aimed to develop an integrated coefficient that would capture participants’ change in attitudes and interpretation of the political video clips over time. To do so, we developed an integrated score—Interpretation Change Coefficient (ICC). For each participant and each video, we calculated this score based on their responses to the interpretation-related questions administered after both the 2019 and 2021 scans. Specifically, we (i) calculated the absolute change in responses to each of the 47 questions about video content that appeared in both sessions; (ii) calculated the absolute change in responses to the questions about political figures that were repeated across sessions. For videos featuring the same politician in both sessions, we first averaged participants’ responses to the 11 sentiment-related questions (e.g., trust, pride, anger), and then measured the change in this average score; (iii) Included participants’ raw responses to 26 “retrospective” questions about video content that were asked only in the 2021 session, which directly asked about perceived shifts in interpretation since 2019; (iv) Finally, we averaged the values from all three components to produce a single ICC value for each participant and each video.

This process resulted in a matrix of 147 ICC scores (21 participants × 7 videos), each representing the overall change in participant’s interpretation of a specific political video over time.

Ideology vs. In-group/Out-group ICC

To further characterize interpretation changes that depend on political identity, the questions were divided into two categories: “Ideology” questions that focused on evaluative judgment and opinion reflection on the content of the videos, and “In-group/out-group” questions that focused on character perception and emotional response to specific political figures. Although all videos appealed to both ideological and group-related aspects of politicization (see Supplementary Fig. 1), as real-world political discourse typically intertwines these facets, the inclusion of both question types in the post-scan questionnaire enabled us to separately assess changes in ideological attitudes and in-group/out-group perceptions despite their co-occurrence within the stimuli.

“Ideology” questions encompassed inquiries related to attitudes towards policy, worldview, or agenda, without involving elements associated with political in-groups, political out-groups, or specific political figures representing those groups. These questions were intended to gauge participants’ ideological stances in a broader sense, independent of their affiliations with particular political entities or personalities (e.g., “Do you believe that filing an indictment against elected official before the elections, with no possibility of concluding the procedure until the end of the elections, may have negative potential implications for democracy?”). Note that the “ideology” questions also included general questions such as “How much has your opinion about the video changes since you first saw it in March 2019?”. In contrast, the “in-group/out-group” questions constituted those inquiries that specifically addressed aspects related to political in-groups, political out-groups, or individual political figures representing those groups. Those questions were distinct from the “ideology” questions as they focused on participants’ perceptions and attitudes concerning specific political entities, affiliations, or personalities, rather than examining abstract agendas or worldviews (e.g., “To what extent do you consider Benjamin Netanyahu to be trustworthy?”) (see Supplementary materials for a full list of questions).

The categorization of the questions was done by two independent judges, both majors in political science. After completing their categorization, the judges compared their categorization to identify any discrepancies. They agreed on the vast majority of the questions (52 questions), and in the rare case of discrepancies (2 questions) they discussed and reached a shared decision. Consequently, the categorization resulted in 23 questions classified under the “ideology” category and 31 questions classified under the “in-group/out-group” category per video (see Supplemental Materials for more details).

Upon completing this categorization process, we calculated the ideology ICC as well as the in-group/out-group ICC, using the same methodology as the general ICC calculation. As a result, we obtained ideology ICC vector with 147 values (21 participants x 7 videos), and in-group/out-group ICC vector with 147 values (21 participants x 7 videos).

Neuroimaging analysis

Difference in the neural response

To evaluate differences in neural responses between the April 2019 and the August 2021 sessions for each participant, we calculated the Euclidean distance voxel-by-voxel across the entire brain32. The Euclidean distance values were computed using the following equation:

$${{\rm{D}}}=\sqrt{\Sigma {{\rm{t}}}{\left(\right.{{\rm{S}}}2021({{\rm{t}}})-({{\rm{S}}}2019({{\rm{t}}}))}^{2}}$$

Here, for each voxel and each participant, S(2021)(t) represents the BOLD time course measured during the August 2021 session, and S(2019)(t) represents the BOLD time course measured during the April 2019 session. By applying this procedure, we obtained a squared distance value for each voxel, for each participant, and for each video clip. To account for differences in the number of timepoints across video clips, the obtained distance was divided by the number of TRs. Consequently, we obtained seven matrices (one per video clip) with dimensions of distance values across voxels (97,332) by participants (N = 21).

To characterize the overall difference in the neural response between the two sessions, we first averaged the distance across the 6 political movie clips in each voxel. Next, voxels were ranked based on their Euclidean distance value and divided into three equal-sized bins (the first bin included 0–33.3 percentage, the second bin 33.3–66.6 percentage, and the third bin included 66.6–100 percentage). These bin categories (small to large neural difference) were then projected onto the cortical surface for visualization.

Voxels with difference in neural response larger than “baseline”

We wanted to characterize voxels in which the change in their response over the two and a half years that passed was larger than the changes that occurred only due to watching the same stimulus again two and a half years apart. To estimate these “baseline” changes in the neural responses we used the neutral video clip. In each voxel we compared the neutral Euclidean distance to the average political videos Euclidean distance using a two-tailed paired t-test, to test for voxels with larger change in the political clips. We corrected for multiple comparisons using FDR correction35. We used these voxels for further analyses.

Neural correlates of changes in interpretation

To test whether a change in interpretation was reflected in the neural response, we correlated the neural response difference with the interpretation difference within the voxels of interest. This was done by computing Pearson correlation between the Euclidean Distance vector and (i) general ICC vector; (ii) ideology ICC vector and (iii) in-group/out-group ICC vector. This analysis resulted in 3 correlation values for each voxel (one for change in general interpretation, one for change in ideology interpretation and one for change in in-group/out-group interpretation). Note that Intra-class correlation coefficients were calculated for each evaluation type across participants and stimuli. The consistently low intra-class correlation values (Supplementary Fig. 2) indicated minimal constant variance within these grouping factors, supporting the use of non-hierarchical correlation analyses in the study.

Statistical significance testing of the Euclidean distance-ICC correlation

To assess whether the correlation values were significantly larger than what would be expected by chance, we simulated a null distribution of correlations using a permutation method. Initially, we created null distributions of the Euclidean distance values. For this procedure, we phase randomized each voxel’s values from the April 2019 scan for 1000 iterations, generating a different time course each iteration while maintaining the same autocorrelation36. Subsequently, we calculated the Euclidean distance between the randomized voxel values from the April 2019 scan and the real voxel values from the August 2021 scan. We performed this procedure for each video, resulting in six distinct pseudo-Euclidean distance matrices (24,848 × 21 × 1000). Finally, we calculated the correlation between these pseudo-Euclidean distances and the (i) ICC vector; (ii) ideology ICC vector and (iii) in-group/out-group ICC vector. This process produced three null matrices of size 24,848×1000 (voxels x number of permutations), one for each ICC vector.

For each analysis (general, ideology, in-group/out-group), we conducted cluster-size correction analysis. As permutation testing allows for the empirical estimation of a null distribution without assuming an underlying statistical model37, we used the aforementioned null distribution and iteratively evaluated the formation of clusters in each iteration, identifying clusters of voxels that exceeded a predefined correlation threshold. This threshold was set at the 95th percentile of the original correlation values to ensure that only values within the top 5% were considered. Specifically, these correlation values were derived from the three distinct categories of questions (general, ideology, in-group/out-group). In each iteration, we calculated the size of these null clusters. The final significance threshold for cluster size was determined by taking the 95th percentile of cluster sizes from all iterations, establishing a corrected criterion for statistical significance. Both threshold parameters (the correlation cutoff and cluster-size significance threshold) were selected based on a hyperparameter tuning procedure designed to optimize sensitivity and specificity (for more details, see Supplementary Fig. 3). This process allowed us to control for multiple comparisons and retain only clusters of voxels that were robustly significant.

Using Brainnetome atlas to provide a quantitative and objective description of regions

To describe voxel distribution across neural regions, we summarized the results of these analyses as associated with different brain gyri as defined by the Brainnetome atlas38. For each gyrus, we computed the total voxel count in the data, as well as voxel counts specific to the left and right hemispheres. Additionally, we calculated the percentage of voxels in the data relative to those in the atlas for the entire gyrus and separately for each hemisphere.

Results

Behavioral analysis: interpretation change

The behavioral results revealed substantial interpretative changes across the two-and-a-half-year interval, as captured by the Interpretation Change Coefficient (ICC). We used the Wilcoxon Signed-Rank test, as the data were not normally distributed, to test whether the interpretation of each of the political video clips significantly changed. We found that the change in interpretation of all 54 questions was significant between April 2019 and August 2021 (see Fig.2 and Supplementary Table 1).

Fig. 2: Interpretation change Between 2019 and 2021.
Fig. 2: Interpretation change Between 2019 and 2021.
Full size image

a Differences in responses between 2019 and 2021, along with retrospective opinion changes reported in 2021. Data are categorized into In-Group/Out-Group questions (dark red) and Ideology questions (dark blue), and further subdivided by the corresponding stimulus. Horizontal lines represent the interquartile range (IQR); circles denote the median; rotated light-blue squares indicate the mean; and gray triangles represent outliers, defined as values more than 1.5 IQRs below Q1 or above Q3. Solid lines indicate Recurring Questions, while dashed lines indicate 2021-Only Questions. b Pearson and partial correlations between ICC vectors: General ICC, In-Group / Out-Group ICC, and Ideology ICC.

For the general ICC, the most significant changes were observed in the Minor Right-Wing Campaign (MinorRWC; M= 44.48, SD = 19.60), the Center-Party Campaign (CWC; M = 36.58, SD = 12.79), and the Left-Wing Campaign (LWC; M = 35.62, SD = 15.70). These three videos—representing the New Right, Blue and White, and Labor-Meretz, respectively—correspond to political entities that underwent dramatic transformations between 2019 and 2021.

Notably, the MinorRWC campaign showed the highest ICC across all categories, which may reflect its exceptional trajectory: originally a close ally of Netanyahu and the Likud, the New Right later joined forces with center and left-wing parties to form a broad coalition government—an extremely rare political realignment in Israeli politics (see 27). While this shift might not entail deep ideological change, it represents a sharp in-group/out-group boundary transition, aligning with the exceptionally high in-group/out-group ICC for that video (M = 49.37, SD = 20.63).

Similarly, the Center-Party Campaign (Blue and White) presented substantial general and ideological ICCs (M = 36.58, SD = 12.79; M = 37.55, SD = 18.18), potentially reflecting disappointment or cognitive reevaluation: while initially positioned as the primary opposition to Netanyahu, the party ultimately chose to join his government, violating the expectations of many supporters and reshaping perceptions of its political identity (see 39).

The Left-Wing Campaign (Labor–Meretz) also showed elevated ICCs (general: M = 35.62, SD = 15.70; in-group: M = 30.74, SD = 13.71), likely tied to major shifts in leadership and party size. Beginning the 2019 campaign as the second-largest party under Avi Gabbay, it reemerged in 2021 as a much smaller left-wing faction under Merav Michaeli. These internal transformations likely altered voters’ emotional and social alignment with the party more than their core ideological positions (see 40).

By contrast, videos associated with stable or irrelevant political actors, such as Netanyahu’s speech (RWS) and Shelly Yachimovich’s speech (LWS) (Once a prominent political figure, in 2021 was at least two years outside of the political field), exhibited low ICC values across all categories (e.g., general ICCs of M = 20.80, SD = 13.20 and M = 19.95, SD = 10.13, respectively). These results suggest that participants’ attitudes toward these individuals were relatively stable and already well-formed in 2019. Taken together, the data show that interpretative change was greatest when participants were exposed to parties or figures whose political positioning, alliances, or leadership dramatically changed during the study interval.

To examine the inter-relationships among the three ICC measures (general, in-group\out-group and ideology) we examined both Pearson and partial correlations between each pair of vectors. General ICC showed strong positive correlations with both In-Group/Out-Group ICC (r = 0.88, p <0.001) and Ideology ICC (r = 0.78, p <0.001) (Fig.2B), which is expected given that it serves as the overarching construct from which the other two measures were derived. In contrast, In-Group/Out-Group and Ideology ICC share only a moderate zero-order correlation (r = 0.48, p < 0.001), and—once the variance common to General ICC is removed—their relationship reverses direction (partial r = –0.69, p < 0.001), indicating that they capture distinct facets of change. Finally, variability was comparable across the three measures (General ICC SD = 17.16; In-Group/Out-Group ICC SD = 19.53; Ideology ICC SD = 22.74).

Difference in the neural response between the two sessions

To evaluate differences in neural responses between the April 2019 and the August 2021 sessions for each participant, we calculated the Euclidean distance voxel-by-voxel across the entire brain (see methods for details). We ranked these 97,332 voxels based on their neural Euclidean distance, and then divided them into three equal-sized bins. This division is presented in Fig. 3, showing bins with relatively small differences between April 2019 and August 2021 (M = 0.194, SD = 0.014, red), bins with intermediate differences between sessions (M = 0.216, SD = 0.004, orange), and bins with relatively large differences between the sessions (M = 0.23, SD = 0.004, yellow). We found that voxels with the least difference between sessions were predominantly located in posterior regions such as the occipital lobe and part of the parietal lobe (Fig. 3, marked in red), and regions with the largest Euclidean distances between sessions were located in sub-cortical regions such as the Cingulate gyrus, Hippocampus, Basal Ganglia, Amygdala as well as the Insular gyrus and mPFC (Fig. 3, marked in yellow, Supplementary Table 2).

Fig. 3: Differences in neural response between the two scans.
Fig. 3: Differences in neural response between the two scans.
Full size image

a Surface Space and (b) Volume Space (Horizontal View) of the Euclidean distance between April 2019 and August 2021 while watching all political videos, divided into three bins: 0–33.3 percentile (red), 33.3–66.6 percentile (orange), and 66.6–100 percentile (yellow). The black line marks voxels where the Euclidean distance was significantly greater during political videos compared to the neutral video (FDR corrected).

Voxels with larger than baseline difference in the neural responses

We performed a two-tailed paired t-test to identify voxels where the change in response over the two and a half years was greater for political movie clips than for the neutral movie clip. This analysis revealed 24,848 voxels, mainly sub-cortical regions such as the Thalamus, Basal Ganglia and Amygdala, as well as the Cingulate gyrus (see Fig.3, marked with a black contour line, supplementary Table 3, and supplementary Fig. 4). We then defined these voxels as voxels of interest for the correlation between change in neural response and change in interpretation.

Change in (in-group/out-group) interpretation is associated with change in the neural response of memory, limbic and reward regions

To test for brain regions reflecting changes in interpretation due to evolving political environments, we correlated participants’ ICC values, with the differences (Euclidean distance) in their neural activity between the April 2019 and August 2021 scans. This analysis revealed 8584 voxels with a significant correlation (p < 0.05) between difference in neural response and difference in interpretation. A cluster size correction analysis was performed to account for multiple comparisons in voxel-wise correlation values. The correlation threshold, calculated as the 95th percentile of the original correlation values from the three distinct categories of questions (general, ideology, in-group/out-group), was 0.2965. The threshold cluster size, determined as the 95th percentile of cluster sizes across all iterations, was 14 voxels. After correction, 14 significant clusters consisting of 703 voxels were identified. The sizes of these significant clusters ranged from 18 to 105 voxels. Within the significant clusters, correlation values ranged from a minimum of 0.2967 to a maximum of 0.447 (M = 0.34, SD = 0.03, see supplementary Fig. 5). These voxels were mainly located in the bilateral, caudate, striatum and right amygdala and Hippocampus (see Fig. 4 and supplementary Tables 4, 5).

Fig. 4: Correlation between neural difference and interpretation difference.
Fig. 4: Correlation between neural difference and interpretation difference.
Full size image

Colored voxels represent correlation values with p-value smaller than 0.05. The black line outlines clusters that remained significant after cluster-size correction. The left column indicates correlation with general interpretation; middle column indicates correlation with in-group/out-group interpretation, and right column indicates correlation with ideology interpretation.

To specifically test the correlation to in-group/out-group interpretation, the same procedure was conducted for the Euclidean distance and the in-group/out-group ICC vector. This analysis revealed 10,061 voxels with a significant correlation (p < 0.05) between the difference in neural response and the difference in in-group/out-group interpretation. A cluster size correction analysis was performed to account for multiple comparisons in voxel-wise correlation values. As before, the correlation threshold was 0.2965. The threshold cluster size, determined as the 95th percentile of cluster sizes across all iterations, was 19 voxels. After correction, 15 significant clusters consisting of 1,133 voxels were identified. The sizes of these significant clusters ranged from 20 to 295 voxels. Within the significant clusters, correlation values ranged from a minimum of 0.297 to a maximum of 0.488 (M = 0.35, SD = 0.04, see Supplementary Fig. 3). Similar to the overall change, these voxels were located mainly within the bilateral Caudate Nucleus, Striatum, Amygdala and right Hippocampus (see Fig. 4 and supplementary Tables 4, 5).

To specifically test the correlation to ideology interpretation, the same procedure was conducted for the Euclidean distance and the ideology ICC vector. This analysis revealed 2466 voxels with a significant correlation (p < 0.05) between the difference in neural response and the difference in ideology interpretation. A cluster size correction analysis was performed to account for multiple comparisons in voxel-wise correlation values. Again, the correlation threshold was 0.2965. The threshold cluster size, determined as the 95th percentile of cluster sizes across all iterations, was 2 voxels. After correction, 4 significant clusters consisting of 9 voxels were identified. The sizes of these significant clusters ranged from 2 to 3 voxels. Within the significant clusters, correlation values ranged from a minimum of 0.302 to a maximum of 0.349 (M = 0.323, SD = 0.02, see supplementary Fig. 3), located in the inferior frontal gyrus and striatum (see Fig. 4 and supplementary Tables 4, 5).

Discussion

This study examined the neural correlates of interpretative changes within individuals following a significant political crisis. Using functional MRI, we conducted two scanning sessions separated by two and a half years, during which 21 participants viewed identical political video clips. Analysis revealed that the most pronounced differences between timepoints occurred in regions associated with limbic, reward, and memory networks. Specifically, the amygdala, hippocampus, and caudate exhibited activity patterns that tracked temporal changes in interpretation. Notably, neural changes in these regions were associated with more voxels correlating with shifts in political in-group affiliations than with changes in ideological positions.

Our analysis revealed a hierarchical pattern of neural response differences between the 2019 and 2021 scans. Primary sensory regions showed minimal changes, while the medial prefrontal cortex (mPFC), inferior frontal gyrus (IFG), cingulate and insula gyri, and subcortical regions (including the amygdala, hippocampus, and basal ganglia) exhibited the most substantial differences. A key finding was that while sensory and frontal regions showed similar magnitudes of change for both political and neutral content, the cingulate, insula, IFG, and subcortical regions demonstrated significantly larger differences specifically in response to political content. There is very sparse research examining longitudinal changes in neural responses to naturalistic stimuli within individuals. Our observation of minimal changes in sensory cortices aligns with the few studies that did compare intra-subject correlation and found relatively stable responses in primary visual and auditory cortices when individuals view the same movie multiple times41,42. Our findings extend this work by demonstrating a hierarchical pattern of neural plasticity, with increasingly pronounced differences observed along a gradient from sensory regions to memory, emotion, and reward networks.

Our findings demonstrated that regions within the striatum and amygdala tracked longitudinal changes in individual interpretation. The striatum has been established as crucial for value and reward processing43 and plays a role in narrative interpretation44,45, while the amygdala serves as a central hub for emotional processing46. In political contexts, ventral striatal activation correlates positively with politician preference rankings47, and engagement with attitude-incongruent information activates the striatum13, suggesting that resolving cognitive inconsistencies may be inherently rewarding15. Both implicit48 and explicit16 attitudes toward Democratic and Republican politicians modulate amygdala activation during politician image viewing. Additionally, heightened political engagement correlates with increased amygdala and ventral striatal activation during the evaluation of agreeable opinions14. Recent research has shown that the striatum exhibits similar activation patterns among politically aligned individuals during political word processing tasks, explaining enhanced striatal and amygdala synchronization during political video viewing49. Our study extends this literature by demonstrating that the same neural substrates that distinguish between individuals with opposing political views also undergo modulation within individuals over time, as evidenced by changes observed across our two-and-a-half-year study period.

The substantial changes we observed in hippocampal activity align with this structure’s established role in memory encoding, consolidation, and retrieval50,51. The finding that hippocampal activity tracked interpretative changes complements previous research demonstrating differential hippocampal responses among individuals who experienced identical narratives following exposure to different contextual information, whether in auditory32 or visual52 modalities. Our observation that hippocampal responses tracked interpretative changes within the same individual over time further reinforces the hippocampus’s role in narrative construction53 and social navigation54.

Based on previous studies examining neural correlates of political and non-political stimulus interpretation (e.g.18,19,20,21,55), we hypothesized that interpretative changes occurring over the two-and-a-half-year interval between scans would be reflected in Default Mode Network (DMN) activity. However, our data did not support this hypothesis. We propose that this unexpected finding may be explained by the DMN’s role in integrating information into abstract narrative representations56. Previous research has shown that DMN activity remains aligned between individuals who share similar narrative interpretations, regardless of variations in language57, modality58, or specific event details59,60,61. When applied to the findings in this study, one possibility is that DMN responses were less sensitive to within-individual changes as these were driven not by shifts in abstract narrative construction (as is the case in between-individual differences), but rather by evolving emotional responses to the narrative content. However, because null findings do not constitute evidence against the hypothesis, further research is needed.

Interestingly, we found that changes in neural responses over the two-and-a-half-year period correlated strongly with shifts in in-group/out-group ratings (particularly in the caudate, amygdala, and hippocampus) and less so with changes in ideological positions. This aligns with extensive literature demonstrating the involvement of the amygdala and striatum, along with regions within the Default Mode Network, in in-group/out-group evaluation62,63,64,65,66,67,68,69,70,71,72. Previous research has shown enhanced amygdala response to out-group faces73, with response magnitude correlating with implicit in-group bias measures74,75. These differential amygdala and striatal responses to in-group/out-group members persist even in artificially created groups using the minimal group paradigm76. Additionally, striatal activity has been linked to outgroup prejudice77 and predicts reduced helping behavior toward outgroup members78. Our study contributes to this literature by demonstrating that these in-group/out-group-dependent neural responses can be observed longitudinally within the same individual.

A recent study tested the neural response to different types of political discourse: policy-based (which aligns more with ideological evaluation) and identity-based (which aligns more with in-group/out-group evaluation)22. Partisan-dependent response was found in regions involved in socio-political reasoning (e.g., mPFC and right TPJ) for policy-based discourse, in regions involved in mentalizing (e.g., posterior STG) for identity-based discourse, and in regions involved in affective processing (e.g., insula and ventral striatum) for both kinds of stimuli. Our study complements this study by using a greater range of stimuli such as campaign ads and political speeches that contained similar amounts of policy and identity-based content (Supplementary Fig. 1) and testing the neural correlates of changes occurring in ideology and in-group/out-group ratings of the perceivers (and not the stimuli). We found that within regions such as the insula and ventral striatum, not only are different forms of stimulus partisanship reflected in the response, but also different forms of individual partisanship.

The substantial overlap between our general and in-group/out-group analyses highlights how political identity perception was predominantly shaped by in-group evaluation. This shift reflects the diminishing role of ideology in political assessment, where the central question in evaluating political entities has evolved from “What is your suggested policy?” to “Are you with us or against us?”28. A compelling review of neural correlates in artificial group dynamics suggested that “the fluid nature of functional relations between groups, however, is also reason to be optimistic about interventions in intergroup conflicts: If perceptions or the salience of intergroup relations can be changed, individuals may overcome hostility toward previous enemies, even in the context of socially meaningful identities that are linked to deeply held political beliefs”70. Our study, capitalizing on unexpected political developments in Israel, tracked such transitions in intergroup relations and demonstrated their neural representation within limbic, memory, and reward networks.

Limitations

Several limitations should be considered when interpreting these findings. Our relatively small sample size (N = 21) may have limited our statistical power to detect subtle effects and potentially affect the generalizability of our results to the broader population. Specifically, our inability to stratify participants based on their political orientation due to insufficient sample sizes of distinctly left-wing (N = 13) and right-wing (N = 8) individuals prevents us from establishing the directionality of the observed neural response patterns and their relationship to political identity. While our results demonstrate significant alterations in neural activity over the two and a half years that past between the two scans, we cannot determine whether these changes reflect a progressive shift from conservative to liberal perspectives, or vice versa. Future studies with larger, politically diverse samples will be essential to elucidate the directional nature of these neural mechanisms and their relationship to shifts in political identity. A second limitation is that changes in ideological processing may reflect both shifts in participants’ own ideology and changes in politicians’ perceived ideology over time. Although the stimuli were identical across sessions, the broader political context and evolving public perceptions could have contributed to the observed differences, complicating the interpretation of shifts in Ideology. Such constraints are inevitable in ecological designs; however, future studies may employ more controlled laboratory paradigms to minimize this contextual variability.

Conclusion

A central challenge in political neuroscience concerns the bidirectional relationship between neural activity and political beliefs—whether brain structure and function determine political views or political views shape neural responses79. Jost and colleagues propose that “it seems more likely to us that ideological differences in neurocognitive structure and functioning reflect a constellation of social and psychological processes that unfold over time and give rise to the expression of beliefs, opinions, and values.” As the first study to examine longitudinal changes in individual brain responses to political stimuli, our findings provide empirical support for this directional hypothesis, suggesting that social and psychological processes indeed shape neural responses. Furthermore, our dissection of political identity into ideological and in-group/out-group components reveals that changes in neural responses to political situations are driven more by shifts in group allegiance than by alterations in fundamental worldviews.

Global politics is experiencing intensifying polarization across multiple dimensions—ideological, religious, and socio-political. Our investigation of the neural correlates of political identity represents a significant advancement in understanding these divisions at a neurobiological level. We hope these insights may inform novel approaches to understanding societal divisions and mitigating the effects of political polarization.