{"paper_id":"1da8a8da-e239-489a-b291-fe8271b208ff","body_text":"1 \n \n \n \n \n \n \n \n \n \nTitle: Social Decision Preferences for Close Others are Embedded in Neural and Linguistic \nRepresentations \n \nAbbreviated Title: Social Decision-Making and Neural Representations \n \n \n \nJoão F. Guassi Moreira*, L. Concepción Esparza, Jennifer A. Silvers & Carolyn Parkinson \n \nDepartment of Psychology, University of California, Los Angeles \n \n \n*Corresponding Author \n4572 Pritzker Tower \n502 Portola Plaza \nLos Angeles CA, 90095 \njguassimoreira@ucla.edu \n \n \n \n \nAuthors Contributions \nJFGM and CP developed research questions and designed the study concept. Data were \ncollected by JFGM (both studies) and LCE (Study 1) under the supervision of CP (Study 1) and \nJAS (Study 2). JFGM cleaned the data, verified by LCE. JFGM analyzed the data under \nsupervision of CP with feedback from JAS. JFGM drafted the manuscript with substantial \nfeedback and critical edits from CP. LCE and JAS also provided feedback and edits on a later \ndraft. All authors approved the final manuscript for submission. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 2 \nAbstract \nHumans frequently make decisions that impact close others. Prior research has shown that \npeople have stable preferences regarding such decisions and maintain rich, nuanced mental \nrepresentations of their close social partners. Yet, if and how such mental representations \nshape social decisions preferences remains to be seen. Using a combination of functional \nmagnetic resonance imaging (fMRI) and natural language processing (NLP), this study \ninvestigated how neural and linguistic representations of close others relate to social decision-\nmaking. After nominating a parent and friend, male and female participants (N = 63) rated their \ncharacteristics and made hypothetical social decisions while undergoing fMRI. Neural \nrepresentations of parents and friends, relative to the self, were linked to social decision \npreferences. Specifically, greater neural similarity between self and parent in the \ntemporoparietal junction (TPJ) and nucleus accumbens (NAcc) was associated with a \npreference for parents, while greater self-friend similarity in the medial prefrontal cortex (mPFC) \nwas associated with friend-preference. Additionally, linguistic analysis of written descriptions of \nclose others from a separate sample of males and females (N = 1,641) revealed that social \ndecision preferences could be reliably predicted from semantic features of the text. High \ncorrespondence between neural and linguistic data in the imaging sample further strengthened \nthe association with social decision preferences. These findings help elucidate the neural and \nlinguistic underpinnings of social decision-making, emphasizing the critical role of mental \nrepresentations in guiding choices involving familiar others. \n \n \n \n \n \n \n \nSignificance Statement \nThis study provides novel insights into how mental representations of close others relate to \nsocial decision-making. By combining brain imaging and natural language processing, we show \nthat both neural and linguistic representations of familiar individuals (parents and friends), can \npredict social preferences. We found that neural representations of these close others are linked \nto the choices people make about these individuals. Additionally, the way people describe their \nclose others in writing are reliably associated with their decision preferences. Our approach, \nintegrating neuroscience and language analysis, significantly advances our understanding of the \ncognitive mechanisms behind social decision-making, posing implications for fields ranging from \npsychology to artificial intelligence. These findings highlight the complexity of human \nrelationships and their impact on everyday decisions. \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 3 \nIntroduction \nHumans are frequently faced with decisions that could impact multiple other people. For \ninstance, one may need to choose between having dinner with a close friend or attending happy \nhour with a new co-worker, or whether to buy a plane ticket home for the holidays to see family \nor use that money to take a trip abroad with a romantic partner. Decision points such as these \nare pervasive and often highly consequential for one’s relationships with others. Prior work has \nshown that individuals have specific, stable preferences that guide their choices regarding \nfamiliar others (Guassi Moreira et al., 2020). Given that no two people we know are exactly \nalike, one might expect subjective representations of familiar others (i.e., one’s mental model of \nsomeone) play a key role in shaping social decision behavior. However, surprisingly little work \nhas examined this. In the current study, we used functional magnetic resonance imaging (fMRI) \nand natural language processing (NLP) to examine the link between mental representations of \nfamiliar others and social decision behavior towards those individuals.  \nBy and large, most social decision-making studies have not examined how mental \nrepresentations of others shape behavior. Instead, studies typically investigate how cognitive or \naffective heuristics are applied over characteristics of choice alternatives to guide decision \nbehavior. These are heuristics that help process a variety of decision-level features that convey \ninformation about risk, mood, ambiguity, inequity, harm, others’ perspectives, and so on \n(Austerweil et al., 2016; Cole & Bruno Teboul, 2004; Crockett et al., 2017; FeldmanHall & Chang, \n2018; Hackel et al., 2017; Kao et al., 2023; Sampaio et al., 2023; Yu et al., 2019). Because so \nmuch attention has been devoted to such heuristics and their role in parsing decision-level \nfeatures of choice alternatives, studies tend to disproportionately focus on contexts in which \none’s choices impact strangers, anonymous others, and social agents with experimentally \nmanipulated roles (Huettel & Kranton, 2012), despite the fact that everyday social decisions \nfrequently implicate familiar others. Thus, the relationship between our mental representations \nof other people—especially known, familiar others—and the decisions we make regarding them \nremains largely unexplored. \nAvailable evidence suggests that mental representations of other people should play a \nrole in the decisions we make regarding those people. Several studies have delineated the \nneural architecture for thinking about other people (Broom et al., 2021; Hassabis et al., 2014; \nMitchell et al., 2005; Yin Wang et al., 2017), and recent work has further extended this by \ninvestigating how distributed neural response patterns encode the identities and attributes of \nspecific others (Chavez & Wagner, 2020; Hassabis et al., 2014; Yin Wang et al., 2017). This \nencoding constitutes the basis of mental representations of familiar others – i.e., models that \nallow us to simulate, forecast, and extrapolate the opinions, thoughts, feelings, and actions of \nothers. Mental representations of others are thought to be evoked when predicting an \nindividual’s thoughts or behaviors after having learned about their personality traits (Fleming & \nSlank, 2015; Hassabis et al., 2014; Mitchell et al., 2006), witnessing demonstrations of reward or \npunishment(Ho et al., 2021; Yu et al., 2019), and adherence to norms and conventions (Hawkins \net al., 2021; Sarathy et al., 2017), all suggesting that mental representations of other people \nought to at least partly drive social decisions. Despite this work, however, it remains unclear \nhow such mental representations influence social choice preferences involving familiar others.  \nHere we probed if and how mental representations of familiar, close others are \nassociated with behavior towards those people. We deliberately chose to investigate mental \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 4 \nrepresentations of familiar close others for several reasons. First, close others are most \ncommonly implicated in social decision scenarios. Results obtained by investigating close \nothers are more applicable to everyday human life – more so than a confederate, fictive \nindividual, or celebrity. Second, relationships with close others can provide much richer \nrepresentational content to map onto behavior. Third, examining choice preferences among \nindividuals within the category of close others helps map behavioral tendencies in a finer-\ngrained manner (as opposed to coarsely examining choice preferences between close versus \nnon-close others) that is likely to capture the kinds of distinctions that pervade everyday life, \nwhere people must often consider the consequences of their choices for multiple familiar others \n(rather than deciding, for example, whether to favor a friend or a stranger). We specifically \nstudied choice preferences between a parent and close friend because group-level choice \npreferences between parents and friends have been well documented(Guassi Moreira et al., \n2018, 2020, 2021), and because these two particular relationship categories are motivationally \nsalient across individuals and for much of the lifespan (Crone & Fuligni, 2020; Crosnoe & \nJohnson, 2011; Fuligni, 2019; Greenberg et al., 1983; Syed & Mitchell, 2013).  \nWe assessed mental representations using multiple modalities of measurement by \nleveraging neuroimaging (specifically, fMRI) and applying NLP techniques to written text data. \nWith fMRI, we based our analyses on key theories in social neuroscience that stipulate that \none’s representation of the self serves as the template for all other mental models, and that self-\nrelevance (i.e., high self-other representational overlap) is a key driver of behavior (Mitchell et \nal., 2006; Tamir & Mitchell, 2013). Thus, we examined the degree to which self-other \nrepresentational overlap differed between parent and friends in brain regions related to social \ncognition and value-based processes—key psychological processes implicated in social \ndecision-making—and then tested whether this self-other representational overlap tracked with \nsocial choice preferences. In particular, following extensive evidence from the established \nliterature, the bilateral temporoparietal junction (TPJ) and dorsomedial prefrontal cortex \n(dmPFC) were probed as regions sensitive to social cognitive processing (Pfeifer & Peake, \n2012; Saxe & Powell, 2006) and the medial prefrontal cortex (mPFC) and bilateral nucleus \naccumbens (NAcc) were specified as value-based processing regions (Delgado et al., 2016).  \nUsing the linguistic data, we employed machine learning to test whether we could \nengineer a linguistic signature of social decision preferences from written text about one’s \nparent and friend (Fatima et al., 2021), essentially decoding choice preferences from one’s \nrepresentation of others that is available in written text data. Finally, we combined the \nneuroimaging and linguistic data to test whether high correlation between representational \nsimilarities in the brain and language were associated with stronger choice preferences. We did \nso for two related reasons. First, information from one modality can help disambiguate findings \nfrom another. Second, generally, information captured with different methods may capture \ndifferent aspects of one’s mental representations of other people, and using multiple modalities \nto measure mental representations of others can confer greater sensitivity to links between such \nrepresentations and behavior.  \nA schematic overview of the entire study’s workflow is depicted in Figure 1. We first used \nneuroimaging data to test for differences in how the brain constructs representations for one’s \nparent and friend using the self as a reference (i.e., are friends or parents represented more \nsimilarly to the self?). We anchored representations in relation to the self because of \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 5 \naforementioned prior work in psychology and neuroscience showing that humans often use \nmental representations of the self as the basis for representations of others (Cadinu & Rothbart, \n1996; Krienen et al., 2010; Mitchell et al., 2006; Todd & Tamir, 2024), and that perceived \nsimilarity to oneself is often linked to preferences regarding others (Fischer & Savranevski, \n2023; Froehlich et al., 2021; Hackel et al., 2017). We then examined whether individual \ndifferences in the degree of representational similarity between oneself and particular close \nothers were statistically associated with individuals’ social decision preferences between \nparents and friends. Next, we pivoted to linguistic data to examine whether social decision \npreferences regarding close others could be predicted from written descriptions of them. To do \nso, we validated a linguistic signature of choice preferences across several datasets with \ndifferent types of social decision tasks and with additional characterizations of relationships \n(e.g., measures of relationship quality). Afterwards, we tested for high correspondence between \ninformation from both modalities (neural, linguistic) by correlating neural representations with \nlinguistic signature scores. Finally, we finished by probing whether high correspondence \nbetween the two types of representations was related to stronger social decision preferences for \na given other (i.e., if neural and linguistic representations both show a strong parent ‘bias’, is \none even more likely to evince a parent-over-friend social choice preference?). In all, this study \naims to reveal how mental representations of specific, close others influence social decision \nbehavior towards those individuals.  \n \nMaterials & Methods \nStudy 1 \n \nParticipants. A total of 63 (Mean age = 18.81 years, SD = 1.05 years, 73% female) \nindividuals participated in the current study. Participants were recruited from the University of \nCalifornia, Los Angeles (UCLA) student body by tapping existing internal recruitment pools, \nclassroom advertisements, and an email from the registrar’s office. Our a priori sample size goal \nwas to recruit and run as many participants as possible during the winter and spring academic \nquarters (January – June, 2023). In order to be eligible for the study, participants must have \nbeen between 18-22 years at the time of enrollment and exhibit no MRI contraindications (e.g., \nclaustrophobia, non-MR safe metallic implants, etc.). All participants were fluent in English. Four \nparticipants did not have imaging data available (3 due to software issues preventing data \ncollection/reconstruction and 1 due to participant failure to fully disclose permanent jewelry \nduring safety screening), rendering a sample size of N = 59 for imaging analyses. All \nparticipants provided consent in accordance with the policies of the UCLA Institutional Review \nBoard. Participants were compensated $40 (USD) upon completion of the study. All data and \nmaterials are publicly available on the Open Science Framework (OSF; osf.io/fzds6). \nDemographically, the full sample was comprised of 46 individuals assigned the female \nsex at birth (73%). In terms of gender, 43 individuals identified as women (68.3%), 17 as men \n(27.0%), 2 as non-binary (3.2%), and 1 (1.6%) declined to respond. Ethnically, 12 identified as \nHispanic/Latinx (19.0%). Racially, 24 identified as Asian/American (38.1%), 23 identified as \nCaucasian (36.5%), 7 identified as mixed race/multi-ethnic (11.1%), 5 identified as another \ncategory (‘Other’, 7.9%), 1 declined to respond (1.6%), and 0 identified as Native \nHawaiian/Pacific Islander and American Indian/Alaska Native (0%). In terms of sexual \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 6 \norientation, 46 individuals identified as straight/heterosexual (73%), 8 identified as bisexual \n(12.7%), 4 identified as homosexual (gay/lesbian, 6.3%), 3 identified as another category \n(‘Other’, 4.8%), and 2 identified as asexual (3.2%). Finally, 57 participants reported relying on \ntheir parents for financial assistance (90.5%). The self-reported median household income \namong participants was $100,000 (range = $18,000 - $3,000,000). \n \nProtocol Overview. Participants completed the study in a one-hour session at UCLA’s \nStaglin One Mind Center for Cognitive Neuroscience. Participants were greeted by an \nexperimenter upon arrival, consented, nominated a parent and close friend, received training for \nthe two fMRI tasks, underwent task-based fMRI scanning, and then completed a post-scan \nsurvey. Major elements of this procedure are described in greater detail below.  \n \nParent-Close Friend Nomination. Prior to receiving training for the fMRI tasks, \nparticipants were informed the purpose of the study was to understand how the brain encodes \ninformation about close others and how it uses such information when making choices about \nothers. Participants were then asked to nominate a parent and their closest friend. The closest \nfriend was not allowed to be a current romantic partner or a family member. For the parent, the \nmajority of participants chose a mother (84%) whereas the remainder chose a father (mean age \nof nominated parent: 51.4 years, SD = 5.6). The majority (69%) of participants nominated a \nclose friend assigned the female sex at birth (mean age of nominated parent: 19.1 years, SD = \n1.2 years). \n \nfMRI Acquisition Parameters. fMRI data were collected using a research-dedicated 3T \nSiemens Prisma scanner and a 32-channel head coil. Functional scan sequences were \ndesigned to be consistent with the sequences used by the Human Connectome Project \n(humanconnectome.org), which uses the University of Minnesota’s Center for Magnetic \nResonance Research (CMRR) multi-band accelerated EPI pulse sequences \n(github.com/CMRR-C2P/MB). A high resolution T1 magnetization-prepared rapid-acquisition \ngradient echo (MPRAGE) structural image was acquired for registration purposes (TR = 1900 \nms, TE = 2.48 ms, Flip Angle = 9°, FOV = 256 mm2, 1 mm3 isotropic voxels, 208 slices, A >> P \nphase encoding). Functional runs were comprised of T2*-weighted multiband echoplanar \nimages (TR = 1240 ms, TE = 36 ms, Flip Angle = 78°, FOV = 208 mm2, 2 mm3 isotropic voxels, \n60 slices, A >> P phase encoding, multiband acceleration factor = 4). Single-band reference \nimages were acquired for each functional run. \n \nBehaviors and Preferences Judgment Task. Mental representations of the participant \n(self), close friend, and parent were elicited during fMRI scanning with a computerized \nbehaviors and preferences judgment task (Tamir & Mitchell, 2013; Thornton & Mitchell, 2018). \nThe task entails rating how well a series of preferences describe a given target. On each trial, a \nlabel for a given target was displayed at top of the screen (‘SELF’, ‘PARENT’, or ‘FRIEND’), a \npreference was displayed in the center of the screen, and a 4-point Likert scale was displayed \nat the bottom (1 = “Strongly disagree”, 4 = “Strongly agree”). Examples of preferences include \n“really enjoys living in big cities”, “thinks modern art is uninteresting”, and “feels comfortable \ntalking about personal issues”. Participants used a button box to indicate their response. Each \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 7 \ntrial was presented for 4500 ms, and was followed by a jittered fixation screen consisting of a \nblack ‘+’ in the center of the screen (mean = ~2000 ms, possible range = [200–10165 ms], \ndistributed exponentially). Trials were ‘chunked’ by target such that participants would rate at \nleast seven consecutive trials about a given target before moving on to another target to help \nreduce the cognitive load of rapidly switching between targets. To help ensure participants did \nnot miss a transition between chunks of targets, the fixation cue at the center of the screen was \ndisplayed in red prior to the start of a new chunk.  \nParticipants completed 225 trials total (75/target), spread across four runs. The number \nof trials per target varied across runs. Importantly, the exact same set of stimulus items \n(preferences) were used for each target (self, parent, friend), helping ensure that \nrepresentations of oneself, a parent, and friend were comparably elicited and that different \npatterns of effects between self-parent and self-friend similarities would not be driven by \nstimulus differences. Each run typically lasted between 345–375 s, given the random jitter. \nPreferences for this task were drawn from an open, online pool used in prior studies \n(https://jasonmitchell.fas.harvard.edu/links/; accessible on our OSF repository at osf.io/sfc6b). \nUsing methods consistent with prior studies(Tamir et al., 2016), we narrowed the initial 800+ \nitem pool into a final set of 75 items based on how strongly they were associated with Big Five \npersonality traits (De Raad, 2000; John & Srivastava, 1999). We chose to use the Big Five as an \norganizing framework to structure our measurement of participant representations because \npersonality traits are a key component of individual identities, are known to be broadly \ngeneralizable, and encompass other key types of information (e.g., they imply information about \nmental state-scenario associations). The optimization procedure involved applying a genetic \nalgorithm to stimulus item ratings solicited in an independent, crowd-sourced sample. Further \nstimulus optimization details are hosted on our OSF page (osf.io/zcu7n). We checked whether \nreaction time data from the task differed between condition pairs of interest (self, parent; self, \nfriend) while accounting for random effects of subject and stimulus by using hierarchical \nmodeling. No such differences emerged. The task was programmed in jsPsych (de Leeuw et al., \n2023). \n \nSocial Decision-Making Task. We used a pair of computerized delay discounting tasks to \nassess social decision preferences between parents and close friends. Our rationale for \nchoosing a discounting task is threefold. First, discounting decisions are both pervasive in \neveryday life and are thought to be important for shaping life adjustment outcomes (Golsteyn et \nal., 2014; Lebeau et al., 2016). Second, computerized discounting tasks deployed in research \nsettings are flexible in their configuration, allowing researchers to study social decision behavior \nin the face of different reward outcomes (Seaman et al., 2016). Last, we have previously shown \nthat discounting tasks with social choice manipulations reliably tap decision preferences \nbetween specific social partners and are related to socioemotional facets of interpersonal \nrelationships such as quality and subjective relationship value; we also previously observed that \nchoice preferences from similar tasks tend to be correlated (Guassi Moreira et al., 2021). \nParticipants completed two separate runs of a delay discounting task. On each trial, \nparticipants were presented with two hypothetical scenarios that pitted outcomes for their \nnominated parent and friend against each other. One scenario involved a relatively immediate, \nsmaller reward and the other involved a relatively delayed, larger reward. The delays could take \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 8 \nthe value of zero (‘Today’), 2 weeks, 4 weeks, and 6 weeks. Values of zero and 6 weeks were \nnever presented in the delayed or immediate scenarios, respectively. The two runs differed in \nthe type of rewards offered to participants. One of the runs offered participants hypothetical \nmonetary rewards (in USD) to be earned on behalf of either nominee (e.g., $16 for a given \ntarget); the other run was comprised of social rewards, offering participants hypothetical time \nspent with either target (e.g., 16 minutes of time spent with a given close other). Participants \nwere instructed to think of the latter condition as additional time they could spend with their \nparent or friend, combined with the time they already spend with either close other.  \nNumeric outcome values for each type of reward ranged from 2 – 30. There were 49 \nunique combinations of numeric values and time pairings (e.g., $2 now versus $18 two weeks \nfrom now) for each condition, resulting in 294 unique possible trial types (Guassi Moreira et al., \n2021). In the interest of reducing in-scan task demands for participants, each run was \ncomprised of 30 trial types randomly selected from the total 294. It was heavily stressed that \nparticipants were to complete the task as if rewards were real, even though they were \nhypothetical. Participants had 4500 ms to indicate their decision via button box. The trial \nterminated once they completed their response. Each trial was interspersed with a fixation \nstimulus ‘+’ and a random jitter (similar parameters as the previous task). Participants \ncompleted two runs of the task (1 monetary, 1 social). Each run lasted approximately 90–135 s, \nwith variability owing to the self-paced nature of the task. The visual configuration of task stimuli \nwere generally consistent with prior implementations (Guassi Moreira et al., 2021; Seaman et \nal., 2016). Both runs were programmed in PsychoPy (Peirce et al., 2019). This task was always \npresented to participants after the behaviors and preferences judgment task. Outcome types for \nthis task were counterbalanced between participants.  \n \nPost-Scan Survey. Participants completed a post-scan survey immediately following the \nscan. During the survey, they provided basic demographic information (sex and age) about their \nnominated parent and friend, were asked to list approximately 10 adjectives or short phrases \ndescribing each close other, and answered four written prompts asking them to recall specific \nmemories with each close other. The prompts asked participants to write about their (i) favorite \nmemory with each close other, (ii) a time when their close other made them feel supported, (iii) \na time when their close other let them down, and (iv) and a time when they let their close other \ndown. Participants were asked to provide 3-5 sentences per each memory with as much detail \nas possible. Text about parents and friends were collected in separate blocks. Block orders \nwere randomized between participants. The survey also contained measures of their \nrelationship quality with both close others (see Study 2 methods), questions about how often \nthey see each close other (actual and ideal), a measure of social loss aversion towards each \nclose other (i.e., how upset one would be if they could no longer spend time with a given social \npartner (Guassi Moreira & Parkinson, 2024)), and two one-shot social decision-making items (a \ndictator game and a forced choice question about whom they would rather spend time with). \nMore information on these latter measures is fully disclosed on our OSF page (osf.io/zcu7n). \nParticipants finally finished the survey by answering demographic items about themselves.  \n \nPreprocessing and Statistical Analysis \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 9 \n MRI data preprocessing. fMRIPrep (version 22.0.2)—an open source, freely available \ntool based on functional neuroimaging software Nipype (Gorgolewski et al., 2011)—was used to \nperform preprocessing MRI preprocessing. The T1-weighted (T1w) image was corrected for \nintensity non-uniformity (INU) with N4BiasFieldCorrection (Tustison et al. 2010), distributed with \nANTs 2.3.3 (Avants et al., 2008) (RRID:SCR_004757), and used as T1w-reference throughout \nthe workflow. The T1w-reference was then skull-stripped with a Nipype implementation of the \nantsBrainExtraction.sh workflow (from ANTs), using OASIS30ANTs as target template. Brain \ntissue segmentation of cerebrospinal fluid (CSF), white-matter (WM) and gray-matter (GM) was \nperformed on the brain-extracted T1w using fast (FSL 6.0.5.1) (Zhang et al., 2001). Brain \nsurfaces were reconstructed using recon-all (FreeSurfer 7.2.0) (Dale et al., 1999), and the brain \nmask estimated previously was refined with a custom variation of the method to reconcile ANTs-\nderived and FreeSurfer-derived segmentations of the cortical gray-matter of Mindboggle (Klein \net al., 2017). Volume-based spatial normalization to one standard space \n(MNI152NLin2009cAsym) was performed through nonlinear registration with antsRegistration \n(ANTs 2.3.3), using brain-extracted versions of both the T1w reference and the T1w template. \nThe following template was selected for spatial normalization: ICBM 152 Nonlinear \nAsymmetrical template version 2009c (TemplateFlow ID: MNI152NLin2009cAsym) (Fonov et al., \n2009). \nFor each of the BOLD runs found per subject, the following preprocessing was \nperformed. First, a single-band reference was used as a reference volume. Head-motion \nparameters with respect to the BOLD reference (transformation matrices, and six corresponding \nrotation and translation parameters) were estimated before any spatiotemporal filtering using \nmcflirt (FSL 6.0.5.1) (Jenkinson et al., 2002). The BOLD time-series were resampled onto their \noriginal, native space by applying the transforms to correct for head-motion. These resampled \nBOLD time-series will be referred to as preprocessed BOLD in original space, or just \npreprocessed BOLD. The BOLD reference was then co-registered to the T1w reference using \nbbregister (FreeSurfer) which implements boundary-based registration (Greve & Fischl, 2009). \nCo-registration was configured with six degrees of freedom. First, a reference volume and its \nskull-stripped version were generated using a custom methodology of fMRIPrep. Several \nconfounding time-series were calculated based on the preprocessed BOLD: framewise \ndisplacement (FD), DVARS and three region-wise global signals. FD was computed using two \nformulations following Power (absolute sum of relative motions) (Power et al., 2014) and \nJenkinson (relative root mean square displacement between affines)(Jenkinson et al., 2002). FD \nand DVARS are calculated for each functional run, both using their implementations in Nipype \n(following the definitions by Power et al. 2014). The three global signals are extracted within the \nCSF, the WM, and the whole-brain masks. Additionally, a set of physiological regressors were \nextracted to allow for component-based noise correction (CompCor) (Behzadi et al., 2007). \nPrincipal components were estimated after high-pass filtering the preprocessed BOLD time-\nseries (using a discrete cosine filter with 128s cut-off) for the two CompCor variants: temporal \n(tCompCor) and anatomical (aCompCor). tCompCor components were then calculated from the \ntop 2% variable voxels within the brain mask. For aCompCor, three probabilistic masks (CSF, \nWM and combined CSF+WM) are generated in anatomical space. The implementation differs \nfrom that of (Behzadi et al., 2007). in that instead of eroding the masks by 2 pixels on BOLD \nspace, a mask of pixels that likely contain a volume fraction of GM is subtracted from the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 10 \naCompCor masks. This mask was obtained by dilating a GM mask extracted from the \nFreeSurfer’s aseg segmentation, and it ensures components are not extracted from voxels \ncontaining a minimal fraction of GM. Finally, these masks are resampled into BOLD space and \nbinarized by thresholding at 0.99 (as in the original implementation). Components are also \ncalculated separately within the WM and CSF masks. For each CompCor decomposition, the k \ncomponents with the largest singular values are retained, such that the retained components’ \ntime series are sufficient to explain 50 percent of variance across the nuisance mask (CSF, \nWM, combined, or temporal). The remaining components were dropped from consideration. The \nhead-motion estimates calculated in the correction step were also placed within the \ncorresponding confounds file. The confound time series derived from head motion estimates \nand global signals were expanded with the inclusion of temporal derivatives and quadratic terms \nfor each (Satterthwaite et al., 2013). Frames that exceeded a threshold of 0.5 mm FD or 1.5 \nstandardized DVARS were annotated as motion outliers. Additional nuisance timeseries were \ncalculated by means of principal components analysis of the signal found within a thin band \n(crown) of voxels around the edge of the brain, as proposed by (Patriat et al., 2017). The BOLD \ntime-series were resampled into standard space, generating a preprocessed BOLD run in \nMNI152NLin2009cAsym space. First, a reference volume and its skull-stripped version were \ngenerated using a custom methodology of fMRIPrep. All resamplings can be performed with a \nsingle interpolation step by composing all the pertinent transformations (i.e. head-motion \ntransform matrices, susceptibility distortion correction when available, and co-registrations to \nanatomical and output spaces). Gridded (volumetric) resamplings were performed using \nantsApplyTransforms (ANTs), configured with Lanczos interpolation to minimize the smoothing \neffects of other kernels. Non-gridded (surface) resamplings were performed using mri_vol2surf \n(FreeSurfer). \n \nMultivariate Pattern Estimation. Multivariate patterns were estimated by modeling \nimaging data from the preferences task using a General Linear Model (GLM). Each run of the \ntask was modeled using a fixed effects GLM in FSL. Every GLM consisted of the following \nexplanatory variables: A regressor for each of the self, parent, and friend trials where \nparticipants rated the extent to which a given item described said target. The duration of each \nregressor corresponded to the length of the trial (4500 ms). These regressors were convolved \nwith the hemodynamic response function (double gamma). The temporal derivatives of each \nwere added to account for slice timing effects. The following confound regressors were included \nto statistically control for noise and head motion. Timeseries from the CSF+WM mask \nautomatically extracted by fMRIPrep was added into the model, in addition to the tCompCor \ncomponent. 24 head-motion regressors and individually censored volumes exceeding the FD \nand DVARS thresholds, both computed by fMRIPrep, were also included. Three linear contrasts \nwere computed: [self - baseline], [parent - baseline], and [friend - baseline].  \n \nRepresentational Similarity Analysis. We used representational similarity analysis \n(Kriegeskorte et al., 2008) to summarize neural representations of others. RSA is useful for \ndetermining how various forms of psychological or behavioral information are encoded in the \nbrain, and broadly, for comparing representational structure across modalities. Here we use \nRSA to examine how information about parents and friends is encoded in the brain (specifically \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 11 \nself, other overlap), as well as how this information is relevant for social decision-making. The \nformer set of analyses involved conducting various searchlight and ROI analyses; the latter \nused individual RDM cells to estimate statistical relationships between individual differences in \nneural self, other overlap and social decision preferences. These analyses are described in the \nfollowing three sections.  \n \nA Priori ROI Analyses. We implemented RSA on region-of-interest (ROI) data to \ndetermine whether there was a ‘bias’ in the encoding of parent or friend representations in six \nbrain regions (mPFC, dmPFC, L NAcc, R NAcc, L TPJ, R TPJ), defined a priori via meta-\nanalytic maps for cortical regions and probabilistic atlases for subcortical regions (NAcc). \nCortical regions were identified by their peak meta-analytic coordinates. We did this by \nconstructing an RDM based on between-run distances in multivariate patterns of brain activity. \nBetween-run distances were used to avoid the potential confound of temporal “pattern drift”, \nwhich has been shown to persist even after voxel-wise detrending has been performed, but \nwhich can be alleviated by focusing on comparisons between patterns collected in different runs \n(Alink et al., 2015). Focusing on between-run multivoxel pattern comparisons is also \nrecommended in designs such as that used in the current study, where particular trials (e.g., \nthose corresponding to a particular target) are “chunked” together (Mumford et al., 2014). \nConcretely, this meant that multivoxel patterns for self, parent, and friend were extracted for a \ngiven ROI (using t-statistic images) and distances were computed between patterns from each \nrun for all possible target pairings. Because we were using between run distances, this also \nmeant comparing between run patterns for the same target (e.g., the self condition for run 1 was \ncompare to same condition in run 2). We ran analyses with both (1 – Pearson’s r) and Euclidean \ndistances as the RDM-defining distance metric to check for robustness—analyses were \nconsistent between the two metrics. Euclidean distances are reported in the main document. \nThe mPFC, dmPFC, and TPJ ROIs were defined by locating the respective brain regions on the \nNeurosynth meta-analytic platform (Yarkoni et al., 2011), noting coordinates for the global \nmaximum (or within each hemisphere for the TPJ), and drawing a 5 mm radius sphere around \nthe coordinates. The NAcc masks were defined using the Harvard-Oxford subcortical atlas as \navailable in FSL. ROI masks are visualized in the Extended Data (Figure 1-1). We then \ncomputed pairwise differences between the [self and parent] and [self and friend] RDM cells for \neach of the six ROIs. Paired differences were modeled as being drawn from a Gaussian \ndistribution such that Diffi ~ N(δ*σ, σ2). The mean was parameterized as δ*σ so that draws from \nthe distribution are in “native” units but summary statistics of paired differences are in the \nCohen’s d metric (i.e., mean/standard deviation). Prior distributions were assigned to both δ \n(Cauchy(0, \n!\n√#)) and σ (Jeffreys prior, proportional to \n!\n$!). The Bayesian modeling software stan \n(Stan Development Team, 2020) was used to run these analyses via the rstan R package (no \nthinning, 4 chains, 2,000 samples/chain, 1,000 discarded burn-in samples).  \n \nExploratory Searchlight Analyses. We conducted a set of exploratory searchlight \nanalyses (Kriegeskorte et al., 2006) to identify which brain regions showed the greatest \ndifferentiation among the three representations (self, parent, friend) in a data-driven manner. \nThe searchlight was meant to complement the ROI approach by ensuring we did not miss any \nadditional brain regions that differentiated between targets. A 125 mm3 sphere (2 mm radius, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 12 \nexcluding center voxel [(2 * 2) + 1]3) was moved throughout each participant’s brain and \nmultivoxel patterns to each target were compared at each searchlight center. The first \ncomparison we performed was a direct comparison between parent and friend representations. \nThis was accomplished by extracting multivariate patterns for the parent and friend t-statistics at \neach center voxel, vectorizing the voxels within the sphere, and taking the (Pearson) correlation \ndistance (1 – r) between the two patterns. As before, we only used between-run comparisons. \nWe used between-run distances within-target as a baseline. All between-run distances were \nplaced into a vector and standardized. The mean of within-target distances (e.g., (parent run 1, \nparent run 2), (friend run 2, friend run 4)) was then subtracted from the mean of (parent, friend) \ndistances. We conducted a second series of analyses comparing representational overlap \nbetween the self and each target. Specifically, distances between neural representations of \none’s self and one’s parent were compared to distances between representations of one’s self \nand one’s friend in a procedure similar to that described above. \n Modeling Social Decision Preferences. We used Bayesian statistics for all regression \nmodels involving choice behavior on the social decision-making tasks. Given that there exists \nnotable intra- and inter-individual variability in decision-making behavior, we chose to model the \ndata at the trial-level, necessitating the use of a hierarchical logistic regression model. While a \nhierarchical model is not strictly needed to handle nested data (McNeish et al., 2017), we use it \nhere because it allows us to flexibly model both within- and between-participant sources of \nvariance while regularizing the model coefficients to enhance out-of-sample generalizability. The \nform of the within-person component of the model is notated below. \n \n𝐿𝑜𝑔𝑖𝑡&𝐷𝑒𝑐%&* = \t 𝑏'& + 𝑏!&𝐶𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛%& + \t 𝑏#&𝑅𝑒𝑤𝑎𝑟𝑑\t𝑅𝑎𝑡𝑖𝑜%& \n \n This is the core component of the model because it encodes social decision preferences \nbetween parents and friends. Here, the i-th decision (Dec, 1 = discount, 0 = non-discount) from \nthe j-th participant was modeled as a function of a Condition dummy code and a Reward Ratio \nvariable. Condition is coded such that 0 = friend outcome associated with discounting option, \nparent outcome associated with non-discounting option and 1 = parent outcome associated with \ndiscounting option, friend outcome associated with non-discounting option. The coefficient \nassociated with Condition (b1j) encodes social decision preferences between parents and \nfriends because it reflects the mean difference in choice tendencies between the two conditions \n(after statistically adjusting for other terms in the model). A b1j coefficient > 0 indicates a \npreference towards parents: there is greater likelihood of choosing to discount when the parent \noutcome is associated with the smaller immediate reward, and a greater likelihood of not \ndiscounting when parents are affected by the larger delayed reward. A b1j coefficient < 0 \nindicates a friend-oriented preference following the same logic: a reduced likelihood of choosing \nto discount when the friend outcome is associated with a larger delayed reward and an \nincreased tendency of discounting when associated with the smaller immediate reward. Reward \nRatio is the quotient of the non-discounted outcome over the discounted outcome and is a key \ncovariate for several reasons. Primarily, it serves as a ‘yardstick’ to gauge whether \nrepresentational information is meaningfully related to decision preferences beyond trial-level \nfeatures. This is notable because various cognitive heuristics used in decision-making are \ntypically applied over such features; adjusting for this here helps ensure that representational \ninformation is still related to choice behavior over and above other cognitive processes. In \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 13 \naddition, controlling for Reward Ratio helps to make choices more comparable between trials, \nand provides a ‘sanity check’ for participant engagement (greater reward ratio should generally \nbe associated with a reduced likelihood of discounting regardless of condition). Additionally, a \npost hoc examination revealed that trial-level task features (e.g., reward ratio, delay ratio, \nmagnitude of rewards) were not significantly oversampled in either choice condition, ruling out \nthe possibility that social decision preferences were not conflated with trial-level choice option \nfeatures. All three terms in the within-person component of the model were allowed to vary \nrandomly across participants.  \nAllowing the b1j coefficient to vary randomly across participants meant that we could \nmodel variability in individual decision preferences. We did so across various models by \nentering different between-participant variables. These between-participant variables are \nstatistically allowed to interact with the Condition dummy code (i.e., a cross-level interaction in a \none-level model), but are substantively conceptualized as the association between participant-\nlevel individual differences and social choice preferences. The following sets of between-person \npredictors were tested (each group in a separate model): (i) dissimilarity values for self and \nparent, as well as self and friend calculated from behavioral ratings, (ii-v) neural dissimilarities \nfrom each ROI (bilateral NAcc and TPJ ROIs were entered in the same model). Here we notate \nthe between-participants portion of the model with neural dissimilarity values from mPFC as an \nillustrative example (interested readers can consult our publicly available code for further model \ndetails). \n \n𝑏'& = \t 𝛾'' + 𝛾'!𝑆𝑒𝑙𝑓, 𝑃𝑎𝑟𝑒𝑛𝑡\t𝐷𝑖𝑠𝑠𝑖𝑚& \t + 𝛾'#𝑆𝑒𝑙𝑓, 𝐹𝑟𝑖𝑒𝑛𝑑\t𝐷𝑖𝑠𝑠𝑖𝑚& + 𝑢'& \n𝑏!& = \t 𝛾!' + 𝛾!!𝑆𝑒𝑙𝑓, 𝑃𝑎𝑟𝑒𝑛𝑡\t𝐷𝑖𝑠𝑠𝑖𝑚& \t + 𝛾!#𝑆𝑒𝑙𝑓, 𝐹𝑟𝑖𝑒𝑛𝑑\t𝐷𝑖𝑠𝑠𝑖𝑚& + 𝑢!& \n𝑏#& = \t 𝛾#' + \t 𝑢#& \n \nAnalyses involving linguistic preference scores (described in the next section) were \nentered as part of a three-way interaction between neural dissimilarity and Condition in \nsubsequent analyses. All continuous predictors were grand mean centered prior to running the \nmodel. All models were implemented using the brms package in the R statistical software library \n(no thinning, 8 chains, 2500 samples/chain, 1250 discarded burn-in samples). A weakly \ninformative standard normal prior was placed on all slope coefficients and brms default priors \nwere used for all other model parameters.  \nBecause the output of Bayesian analyses is a distribution, we performed statistical \ninference in a graded manner and thus do not report nor use p-values to conduct frequentist \ninference. Instead, we evaluated the weight of statistical evidence using a combination the \nRegion of Practical Equivalence (ROPE) method (Kruschke, 2011, 2013) and noninferiority \ntesting (Spiegelhalter et al., 1994; Wiens, 2002). Traditionally, the ROPE method entails \nspecifying a symmetric region in parameter space around the null value, calculating a credible \ninterval (CI), and then evaluating whether the CI intersects at all with ROPE. CIs that do not \noverlap with ROPE result in a rejection of the null hypothesis, CIs that are entirely contained \nwithin ROPE result in acceptance of the null, and any other configuration falls in undecided. \nOriginating in clinical trials research, noninferiority testing is similar: if the CI partly overlaps with \nthe upper limit of ROPE but does not pass through below the lower limit (regardless of its \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 14 \nrelation to the null value), the parameter value is declared to be noninferior to the null value (the \nsame method can be used in the other direction, e.g., for ‘nonsuperior’ inference).  \nHere we combine these methods in the following manner. If the CI does not overlap with \nROPE, we deem the strength of evidence to be robust. If the CI does not overlap with the null \nvalue (usually zero), but partly overlaps with ROPE, we deem the strength of evidence to be \nmoderate. If the CI overlaps with ROPE and the null value (in the manner usually described in \nnoninferiority testing), we deem the strength of evidence to be modest. If the CI completely \nspans ROPE (i.e., overlaps and exceeds both ends of ROPE), deem the evidence to be \ninconclusive. If the CI is contained within rope, we interpret the finding to support the null result. \nWe deliberately follow these inferential criteria because it displaces the inferential focus from a \nproblematic binary ‘significant vs not-significant’ label (Kruschke, 2018) to a more nuanced \nconclusion describing the degree of evidence in relation to meaningful benchmarks. We \nconducted a permutation analysis on several study variables to determine whether zero was an \nempirically reasonable null value to use. Analyses showed the expected value of a correlation \ncoefficient under a permuted distribution was indeed zero. Posterior distributions were \nsummarized with the mean of the relevant statistic (e.g., Cohen’s d, regression coefficient) and \nan 89% highest density credible interval (HDI; all samples within the interval have a higher \ndensity than those outside it). We used 89% credible intervals upon the recommendation that \nwider intervals (e.g., 95%) are more sensitive to Monte Carlo sampling error (Makowski et al., \n2019; McElreath, 2015).  \n  \nStudy 2. \n \nDatasets.  \nStudy 2 was comprised of secondary analysis of three large existing datasets (N = 1,641 \nunique participants total) containing written text about parents and friends (nominated using the \nsame procedure as described in Study 1) as well as social decisions involving these two others. \nWe used one dataset (Dataset 1) to engineer a linguistic signature of social decision \npreferences. The written text data and social decision measures in this dataset were collected \nas part of a broader study aiming to answer an orthogonal research question (Guassi Moreira & \nParkinson, 2024). The other two datasets (Dataset 2, Dataset 3), and additional social decision \nmeasures from Dataset 1, were used to validate the linguistic signature by ensuring that it \npredicted social decision preferences in additional contexts. These datasets come from a series \nof studies on social decision-making conducted by the research team (Guassi Moreira et al., \n2018, 2020, 2021). These prior studies have yielded published results, but the written text data \nhave not been analyzed before. A detailed description of each dataset follows.  \n Dataset 1 was comprised of five independently collected subsamples (S1.1 – S1.5). \nParticipants in each subsample were asked to nominate a parent and close friend of their choice \nand provide approximately ten adjective or short phrases (e.g., ‘happy go lucky’, ‘pain in the \nbutt’) describing each close other. Participants in S1.1 were also prompted to recall four types of \nmemories involving each social partner – instances in which they felt supported by the close \nother, in which they supported their close other, in which their close other let them down, and \ntheir favorite memory with their close other. Social decision preferences were assessed with a \none-shot dictator game question in which individuals allocated a hypothetical $100 pot between \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 15 \nparents and friends in $5 increments, and another in which they indicated who they would rather \nspend time with given a day with no other obligations or commitments. Subsamples S1.1 – S1.5 \ntotaled N = 870 participants (range of subsample size n = 45 – 454). Subsamples were recruited \nfrom the UCLA undergraduate psychology subject pool and online crowdsourcing platforms \n(Mturk, Prolific) in 2022. All prior analyses involving the social decision-making data from this \ndataset were recently published (Guassi Moreira & Parkinson, 2024). \nDataset 2 was comprised of five independently collected subsamples (S2.1 – S2.5) as a \npart of a series of studies on social decision-making. Participants in each subsample were \nasked to nominate a parent and close friend of their choice, write a memory they share with \neach target, and list several words or phrases describing the target. They were then \nadministered a social decision-making task that forced them to make choices between the \nparent and friend. Social decision preferences were assessed with a version of the Columbia \nCard Task (Figner et al., 2009)—a multi-trial risk-taking paradigm—that was modified to capture \nsocial decision preferences when making choices with conflicting outcomes between a pair of \nothers (Guassi Moreira et al., 2018). Subsamples S2.1 – S2.5 totaled N = 574 participants \n(range of subsample size n = 46 – 225). Subsamples were recruited from the UCLA \nundergraduate psychology subject pool and the broader West Los Angeles community between \n2016 - 2020. \nDataset 3 was comprised of four subsamples (S3.1 – S3.4). Some of the participants in \nthese subsamples also provided data for subsamples in Datasets 1 and 2. The distinctive \ncontribution of this dataset relative to the others is that it uses a different social decision-making \nparadigm. As with the other datasets, participants were prompted to nominate a parent and \nclose friend. Some subsamples followed the text generation procedure described for Dataset 1, \nothers follow the procedure for Dataset 2, meaning that all subsamples provided adjectives and \nshort phrases describing each other, and some also provided a memory of each other. Social \ndecision preferences were assessed with the same task as in Study 1: a multi-trial delay \ndiscounting task that was modified to capture social decision preferences when making choices \nwith conflicting outcomes between a pair of others (Guassi Moreira et al., 2021). Subsamples \nS3.1 – S3.4 totaled N = 920 participants (range of subsample size n = 61 – 454). Subsamples \nwere recruited from the UCLA undergraduate psychology subject pool (2018-2019) and from an \nonline crowdsourcing platform (2022). \nFurther information about each dataset, such as participant demographics and \nmethodological details, can be accessed in the original publications cited above. Self-reported \nrelationship quality between participants and their nominated parents and friends was collected \nin all three datasets. All participants provided consent in accordance with the policies of the \nUCLA Institutional Review Board. Depending on their method of recruitment, participants were \neither compensated in course credit or monetary remuneration (USD). All study materials and \ncode are publicly available on the Open Science Framework (OSF; osf.io/fzds6). For privacy \nreasons, we are unable to share the raw or preprocessed text data.  \nThe following two sections detail (i) how we used Dataset 1 to engineer a linguistic \nsignature of social decision preferences based on one social decision measure and then (ii) how \nwe used data from all three datasets to further validate the signature to verify that scores \nderived from the signature were related to social decision preferences in novel data and novel \nsocial decision measures.  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 16 \n \nEngineering a Linguistic Signature via Machine Learning.  \nSince our goal was to examine whether linguistic representations of close others encode \ninformation about social decision preferences, we trained a model to predict social decisions \nfrom semantic features of written text data. We sought to produce a linguistic signature of social \ndecision preferences that followed the logic of neural signatures and pattern expression analysis \nin the cognitive and affective neuroscience literature (Chang et al., 2015; Wager et al., 2013), \nboth in the process used to engineer the signature and its application. The first step in this \nprocess was to test candidate models with cross-validation on the subsamples from Dataset 1 \nto settle on model specifications (type of model, hyperparameters). The pipeline for this task \nwas comprised of the following steps: (i) social decision measure selection, (ii) text \npreprocessing, (iii) representation extraction, (iv) leave-one-out cross-validation using each \nsubsample, and (v) estimating model weights on the entire dataset using the model with the \nbest cross-validation accuracy.  \n \nSelecting a Social Decision Measure for Signature Construction. We used the forced \nchoice question about which close other the participant would rather spend a free day with \n(coded such that parent = 1, friend = 0) as our measure of social decision preferences. We \nchose this metric for assessing social decision preferences for four reasons. First, the binary \nnature of the item makes for an intuitive classification metric with which to evaluate model \nperformance (percent accuracy). Second, using a binary forced-choice measure prevents \nparticipants from being impacted by a neutral scale response bias and requires them to disclose \nan unambiguous preference. Third, framing social decision preferences in terms of spending \ntime (a finite resource subject to opportunity cost) with others simulates a real-world scenario \nthat could generalize better to behaviors outside of the lab. Last, the simplicity of the item may \nhelp it better translate to other, more complex measures of social decision-making. Preferences \nfor parents over friends for each sample follows: S1.1 = 51% chose parents over friends; S1.2 = \n42%; S1.3 = 42%; S1.4 = 46%; S1.5 = 42%. \n \nText Preprocessing. Written text data was preprocessed in a manner consistent with \nother similar studies (Fatima et al., 2021), implemented with the TextBlob python library. The \nfollowing preprocessing pipeline was applied to each participant’s data. We first removed \nstopwords, as defined by TextBlob, in addition to the words ‘favorite’ and ‘memory’ (as many \nparticipant responses to memory prompts began with this phrase). Afterwards, each word was \nforced into its lowercase form and lemmatized. Words were then tagged as parts of speech and \nfiltered. Only nouns, adjectives, verbs and adverbs were in the preprocessed word list, \ncorresponding to the following TextBlob tags: 'JJ', 'JJR', 'JJS', 'NN', 'NNS', 'NNP', 'NNPS', 'RB', \n'RBR', 'RBS', 'VB', 'VBD', 'VBG', 'VBP', 'VBZ'.  Parent and friend memories were preprocessed \nseparately, meaning that each participant had a set of a preprocessed words from written text \ndata about their nominated parent, and another preprocessed word set about their friend.  \n \nRepresentation Extraction. We used word embeddings to extract linguistic \nrepresentations of parents and friends from participants’ written text data. This approach \nquantified semantic meaning from the preprocessed word set in a bottom-up fashion using \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 17 \npretrained Word2Vec word embeddings. We extracted word embedding vectors for each word \nin the parent and friend word sets separately, and then averaged the vectors of all set words \ntogether. We used the pretrained embeddings from Google’s 300-dimension Word2Vec model \n(Mikolov et al., 2013), as implemented in the genism python package. Words in the word set \nthat did not have a corresponding word embedding vector were excluded from the \nrepresentation. In this approach, the semantic content of each participant’s text data for a given \nclose other (parent, friend) was summarized as a composite 300-dimensional vector. We used \nprincipal components analysis (PCA) to reduce dimensionality in order to aid modeling. Both \nparent and friend composite vectors for each subject were reduced to 30 dimensions, consistent \nwith similar work (Yingying Wang et al., 2023). On average, 30 dimensions accounted for \napproximately 83% of the variance in written text data, collapsing across parent and friend text \nin all datasets.  \n \nLeave-One-Subsample-Out Cross-Validation. We used leave-one-subsample-out cross-\nvalidation to determine the best specifications for defining a linguistic signature of social \ndecision preferences. We specifically sought to identify the best type of machine learning model \nand set of hyperparameters. We identified three possible types of machine learning models: a \nsupport vector classifier (SVC), a ridge regression classifier (RRC), and regularized logistic \nregression (RLR). Hyperparameters for each model were selected using a grid search within \neach iteration of cross-validation. The candidate values for all possible hyperparameters (C, γ, \nα) were {1, 5, 10, 50, 100}.  \nOur cross-validation approach took advantage of the independent subsamples in \nDataset 1—we chose to leave-one-subsample out for each iteration of cross-validation. Since \nthe goal of cross-validation is to estimate out-of-sample modeling performance, we reasoned \nthe natural partitioning of each subsample as independent datasets best served this goal. This \nmeant that each iteration of cross-validation proceeded such that one subsample was scaled \nand set aside, the other four were aggregated and scaled, a grid search was performed to find \nthe best hyperparameter(s), the best hyperparameter was then used to estimate model weights \non the four aggregated subsamples, and the model weights were finally applied to the left-out \nsubsample to estimate the predictive accuracy of the model. This process was repeated five \ntimes, leaving one of the subsamples out each time, for all of the three models described above. \nThis workflow was implemented using functions from the sci-kit learn python library.  \n \nEstimating Model Weights. As previously noted, the binary forced choice question about \nspending time with either a parent (coded as 1) or friend (coded as 0) served as our criterion. \nThe principal components derived from both the parent and friend text data (60 total) were \nentered into the winning model from the prior step as predictors. The ensuing 60 coefficients \n(weights) from the model indicate how a ‘preference’ towards one close other is embedded in \nthe linguistic representation and thus comprise the linguistic signature. \n \nValidating the Linguistic Signature.  \n  Once we engineered in our linguistic signature in Dataset 1, we set out to further validate \nit by correlating linguistic preference scores derived from the signature with other measures of \nsocial decision-making across the three datasets. The logic for doing so is that if our linguistic \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 18 \nsignature is truly capturing social decision preferences between parents and friends, then the \nsignature should generalize to novel measures and predict preferences across different \nmeasures in the three datasets. To that end, we used the weights from our signature to \ncompute preference scores and test whether those preference scores could predict social \ndecision preferences on previously untested social decision tasks. Linguistic preference scores \nwere calculated by taking the dot product between the model weights that comprised the \nlinguistic signature and the feature-reduced text data from a given subject. The feature-reduced \ntext data were the same as described above: PCA-reduced composite word2vec embeddings \nfor parents and friends. When validating the linguistic signature, we used decisions from a one-\nshot dictator game where participants had to directly allocate money between parents and friend \n(Dataset 1), a multi-trial decision-making task that pitted gains and losses for parents against \nthose for friends in a risky context (Dataset 2), and a multi-trial decision-making task where \nparticipants had to choose to allocate money for, or time spent with, their parents and friends in \na delay discounting context (Dataset 3; the same task used in the Study 1). Finally, because \nsocial decision preferences are often predicted by attitudes towards social partners, we also \ntested whether the linguistic signature could predict relationship quality scores for parents and \nfriends (available in all datasets). These validation tests for each outcome are described in \ngreater detail below. Bayesian statistics were used for all analyses. The inferential criteria from \nStudy 1 were applied to Study 2 analyses.  \n \nSocial Preferences in the One-Shot Dictator Game. The one-shot dictator game, \ncollected in Dataset 1, asked participants to allocate a hypothetical pot of $100 between their \nnominated parent and friend. Values varied by increments of $5, ranging from [$100 parent, $0 \nfriend] to [$0 parent, $100 friend], resulting in a 21-point ordinal variable. Responses were \nrecoded to be the percentage of money allocated towards the parent (0.0, 0.05, 0.10 … 0.90, \n0.95, 1.00). A robust Pearson correlation between preference scores and dictator game \nallocations was computed using the Bayesian software package stan. We note that while the \ntext data for this validation step is not novel (i.e., were used in model engineering), the outcome \nvariable (dictator game allocations) is, and it is valuable to know whether preference scores can \npredict decision preferences in different contexts within-participants.  \n \nSocial Preferences in a Risky Context. Linguistic preference scores were next validated \nby determining whether they could predict parent vs friend preferences in a risky decision \ncontext. Participants in this dataset (Dataset 2) completed a multi-trial decision task that \nrequired them to make choices between a risky and a safe outcome. The risky outcome carried \nthe potential for either a hypothetical monetary gain or a hypothetical monetary loss according \nto a known probability; the safe outcome was neutral (no gain or loss). The interests of parents \nand friends were pitted against each other such that any potential gain from a risky choice on a \ngiven trial would be credited to one partner, whereas any potential loss would be credited to the \nother partner. Participants completed multiple trials across two runs of the task, one in which \nrisky gains benefitted one partner (e.g., parent) while the losses were incurred by the other \npartner (e.g., friend), and another run in which the opposite was true. Hierarchical Bayesian \nlogistic regression was used to model the trial-level likelihood of making a risky (vs safe) \ndecision as a function of a parameter that encoded choice preferences on the task and other \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 19 \ntrial-level covariates. We refer readers to (Guassi Moreira et al., 2018) for details about the form \nof the within-person component of the model. We entered preference scores derived from the \nlinguistic signature as a between-person predictor to the model, allowing it to interact with the \nchoice preference parameter from the model. Because the choice preference parameter was \nallowed to vary randomly across subjects, the ensuing interaction term codes for how choice \npreferences on the task vary as a function of preference scores derived from the linguistic \nsignature. The brms package was used to fit the models (4 chains, 1000 discarded burn-in \nsamples, 2000 samples/chain, no thinning).  \n \nSocial Preferences in a Delay Discounting Context. Linguistic preference scores were \nfurther validated by testing whether they could predict parent vs friend preferences in a delay \ndiscounting context. This is the same task used in Study 1, with the only difference being that \nmore trials were administered to participants. Hierarchical Bayesian logistic regression was \nused to model the trial-level likelihood of making a discounting decision (i.e., choosing the \nimmediate over delayed reward) as a function a parameter that encoded choice preferences on \nthe task and another trial-level covariate (same within-person model as Study 1). We ran \nseparate models for choices with monetary and social outcomes. Preference scores derived \nfrom the linguistic signature were entered into this model in the same manner described above. \nThe brms package was also used to fit the models (using the same specifications that are listed \nabove). \n \nPredicting Relationship Quality Using the Linguistic Signature. Last, we opted to relate \npreference scores derived from the linguistic signature to self-reported relationship quality with \nparents and friends. In doing so, our goal was to subject the linguistic signature to another \nvalidatory step that involved predicting attitudes that may also influence social decision \nbehavior. A meaningful result for this test would indicate that motivationally relevant information \nabout one’s attitude towards a close other is encoded in their linguistic representation. \nInformation about the self-report measure used here (Inventory of Parent and Peer Attachment) \nand its reliability is included in the Extended Data.  \n \nResults \n \nInformation about social decision preferences is embedded in neural \nrepresentations of close others.  \nDifferent Categories of Close Others are Differentially Represented Across Brain \nRegions. We compared the degree to which individuals construct mental representations of \nparents and friends in terms of self-relevance (i.e., representational overlap between self-parent \ncompared to self-friend) in six a priori brain regions of interest (Table 1; Figure 2) that are \ninvolved in social processing and social decision-making (ROIs visualized in Figure 1-1). We \nfound modest evidence to suggest greater self-other overlap for parents than for friends in \nbilaterally in the TPJ and NAcc. In contrast, there appeared to be greater self-other overlap for \nfriends than for parents in mPFC.  \nWe ran two exploratory searchlight analyses to determine if brain regions other than the \na priori ROIs also distinguish between familiar others. The first of these analyses tested for \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 20 \ndistinct representations of parents vs. friends and the second tested for differences in \nrepresentational overlap between self and parent, and self and friend. Results of the first \nsearchlight suggested that the precuneus contains distinct representations of close others \n(Figure 2-1). Exploratory examination of uncorrected results of this analysis revealed more \nextensive results in the precuneus and several smaller clusters in medial prefrontal cortex \n(mPFC; see Figure 2-2). The second searchlight yielded no significant results. Technical details \nof the searchlight are hosted on our OSF page (osf.io/zcu7n). \nA traditional univariate analysis comparing the magnitude of brain activity evoked when \nanswering items for each target ([self – parent], [self – friend], [parent – friend]) is also included \nin the Extended Data (Figure 2-3). This analysis revealed robust activity along the cortical \nmidline and TPJ for both the [self – parent] and [self – friend] contrasts. For thoroughness, we \nanalyzed behavior-based representations using the task’s Likert scale ratings and found greater \nself-other overlap for friends than for parents (posterior Cohen’s d = 0.50, [0.28, 0.71] 89% \nhighest density credible intervals).  \nSocial Decision Preferences are Encoded in Neural Representations of Close Others. \nAfter examining differences in parent and friend representations, we next tested whether social \ndecision preferences involving parents and friends were embedded in their neural \nrepresentations. As noted in the Methods, we estimated trial-level choice preferences between \nparents and friends using hierarchical Bayesian modeling. The model term that captures choice \npreferences was allowed to vary randomly across participants, allowing us to model the ensuing \nvariability in preferences with participant-level data about neural representations. Thus, we \nentered representational similarities between self and other (parent or friend) for each of the six \nROIs into four separate models (bilateral ROIs were included in a single model). \nFor monetary outcomes, we found that greater self-parent similarity in neural \nrepresentations in the dorsomedial prefrontal cortex (dmPFC) was robustly associated with a \ngreater tendency to favor parents over friends (Table 2). Greater self-friend similarity in the right \nNAcc was robustly associated with a greater likelihood of favoring friends (Table 3). Curiously, \ngreater self-friend dissimilarity in the left NAcc was modestly associated with a greater likelihood \nof favoring friends. For social outcomes, we observed modest evidence that greater self-friend \ndissimilarity in mPFC was associated with a greater propensity to favor friends over parents on \nthe task (Table 2). Greater self-parent similarity in the right NAcc was also modestly associated \nwith an increased propensity to favor parents over friends (Table 3). To further ensure that \nresults were not driven by decision-level task features, the models in Tables 2-3 were re-run \nwith all task features replacing reward ratio (discounted reward value, non-discounted reward \nvalue, discounted reward time, non-discounted reward time). The results did not appreciably \ndiffer when controlling for these other decision-level features.  \n While these results descriptively indicate that social decision preferences are embedded \nin mental representations of known others, they cannot speak to the out-of-sample predictive \npower of neural activity measured when thinking about oneself and close others. To address \nthis concern, we conducted a basic predictive modeling analysis, post hoc, to verify whether a \npredictive model using neural RDM cells could predict significantly more variance than a \nbaseline model that only included task-related features. Using leave-one-subject-out cross \nvalidation, we employed the glmnet package in R to train a regularized logistic regression model \nto predict trial-level choices as a function of the condition and reward ratio terms, the neural \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 21 \nRDM cells (one model per brain region, averaged across hemispheres for parsimony where \nrelevant), and the interaction between the neural RDM cells and condition.  \nImbalanced cases (discount, no discount choices) were handled by assigning weights to \neach case proportional to the imbalance. Model-predicted choices and observed choices were \nused to compute a class-weighted composite F1 score for each model (baseline model, \nbaseline model with neural data added). This F1 score assesses joint performance on both the \npositive (discount) and negative (no discount) cases while taking class imbalance into account. \nDifferences in F1 scores were taken between the baseline model and each of the neural models \n(ΔF1). To perform a significance test of these ΔF1 values, we bootstrapped the model output \n(observed cases, predicted cases) and re-computed the F1 scores for all models, as well as the \ndifference between each neural model and the baseline model, for each of 5,000 iterations, \nthereby creating an empirical sampling distribution of the ΔF1 scores. The results of these \nanalyses are presented in Table 4. \nAs shown in Table 4, every model that included interactions between the condition \nvariable and data from neural RDMs evinced a better composite F1 than the baseline (non-\nneural) model. The added predictive value of models containing neural data was modest but \nconsistent, and amounted to increases in F1 scores of approximately 10% in some cases. In the \nmajority (75%) of cases, the bootstrapped CIs did not overlap with 0, indicating a statistically \nsignificant improvement in model performance when incorporating the brain data. In 25% of \ncases, we did not find evidence to reject the null hypothesis that the F1 difference with vs. \nwithout including the neural data is 0; this could be due to several factors, including differences \nbetween the Bayesian approach used in our main analyses vs. this bootstrapping approach, \ndifferences stemming from collapsing across hemispheres in the bootstrapping analyses but not \nthe main (Bayesian) analyses, and/or the fact that this predictive modeling approach is \npenalizing model predictions to ensure generalizability, which could come at the expense of \nunderstanding small but meaningful associations between variables. That said, in the majority of \ncases, incorporating the neural data significantly improved model predictions of participants’ \nchoices, over and above what was achieved when basing predictions only on task features. \n \n Social decision preferences are encoded in linguistic representations of close \nothers.  \nSocial Decision Preferences are Encoded in Linguistic Representations of Close Others. \nThe average leave-one-subsample-out classification accuracies for each of the three candidate \nmodels when engineering the linguistic were roughly comparable (SVC: Mean = 0.554, SD = \n.04, Range = [0.489, 0.585], RRC: Mean = 0.566, SD = .06, Range = [0.471, 0.622], RLR: Mean \n= 0.562, SD = .06, Range = [0.480, 0.622]). We decided to select weights from the SVC model \nbecause it had relatively high accuracy and the lowest performance variance. Notably, when \ntaking into account the baseline rate of parent-over-friend preferences in each of these samples, \nthe classification accuracies for each sample were statistically better than chance when \nconsidering sample size. That social decision behavior could be successfully decoded from \nwritten text was the first piece of evidence that linguistic representations of close others contain \nmotivationally relevant information about participants’ preferences.  \n To follow this up, we tested whether social decision preferences in novel data with \ndifferent assessments of social choice preferences could be predicted as function of linguistic \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 22 \npreference scores – or, one’s social decision preference implied by the model. Model weights \nfrom the linguistic signature were thus applied to novel data to estimate associations with social \ndecision preferences in a one-shot dictator task in which participants allocated a finite sum of \nhypothetical money between their parent and friend (Figure 3). Linguistic preference scores \nwere robustly associated with preferences on this task (r = 0.26, 89% HDI = [0.21, 0.31]). We \nthen observed modest evidence that linguistic preference scores exhibited a small association \nwith social decision preferences between parents and friends in the multi-trial risky decision task \n(b = 0.02 [-0.06, 0.11]). Preference scores derived from our linguistic signature were robustly \nassociated with choice preferences between parents and friends in a multi-trial delay \ndiscounting task, when both monetary (b = 0.94, [0.72, 1.20]) and social (b = 1.14, [0.91, 1.39]) \noutcomes were at stake. Finally, we were able to show that linguistic preference generalized \nsuch that they were robustly associated with attitudes about one’s relationship with their parents \n(rparent = 0.23, [0.19, 0.27]). Evidence for null effects was observed for the relationship between \npreference scores and friend relationship quality, however (rfriend = 0.00, [-0.04, 0.04]). These \nfindings are visualized in Figure 3-1.  \n \nLinguistic and Neural Representations are Correlated, and Jointly Predict Social \nDecision Preferences. The preceding findings raise the question of whether having both a \nstrong linguistic preference score and extensive overlap between neural representations of \noneself and a given other is associated with a stronger social decision preference for that \nindividual. In other words, if someone’s representations of others were ‘biased’ towards a \nparticular other (e.g., one’s friend rather than one’s parents) in both neural and linguistic data, \nwould we also observe even stronger social decision preferences? Such a pattern of results \nwould have several implications for what kinds of information are stored in neural and linguistic \nrepresentations, and how they are linked with each other and behavior.  \n The degree of correspondence between linguistic preference scores and neural \nsimilarity is shown in Figure 4 and Table 5. We found that preference scores implied by the \nlinguistic signature were correlated with similarity between self and parent neural \nrepresentations in the anticipated direction for three of the six brain regions (mPFC, R NAcc, L \nTPJ). Neural similarity between self and friend was correlated with linguistic preferences in the \nanticipated direction for five of the six brain regions (dmPFC, mPFC, L NAcc, L TPJ and R TPJ).  \nWe then examined whether linguistic preference scores moderated the link between \nneural representational similarity and choice preferences on the discounting task (Figure 5, \nTables 6-1, 6-2). The logic behind this analysis is that linguistic representations and neural \nrepresentations obtained with the methods used here may capture partially distinct aspects of \none’s mental representations of others. Thus, combining these different modalities could afford \ngreater sensitivity for detecting associations between one’s mental representations of others \nand behavior. In the monetary outcome domain, a relationship in the non-anticipated direction \nfor the L NAcc was observed with linguistic preference scores and neural dissimilarity between \nself and friend representations. No other effects met our evidentiary criteria or appeared close to \nmeeting them in the monetary outcome domain. In the social outcome domain, we found that \nlinguistic preference scores interacted with self and parent neural dissimilarity in the mPFC and \nR NAcc to promote stronger social decision preferences for parents if neural similarity was high \nand linguistic preference scores were high. A similar finding with self and friend neural \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 23 \ndissimilarity in the left TPJ was indicated by examining the posterior distribution but did not quite \nmeet our evidentiary criteria. \nWhile these results indicate some degree of informational correspondence between \nneural and linguistic representations and subsequently suggest neural representations driving \nbehavior are comprised of semantic information, it remained an open question about what kinds \nof semantic content are indexed in these representations.  \nTo unpack the statistical relationship between neural representations and linguistic \npreference scores we conducted an exploratory, descriptive analysis using the most recent \nversion of the Linguistic Inquiry and Word Count (LIWC-22) software (Boyd et al., 2022) to \nidentify the semantic themes that drive linguistic preference scores and then correlated them \nwith neural representations. We did this in two steps.  \nFirst, we passed the written text data of Study 2 participants through LIWC-22 and \nextracted the prevalence of the following thematic categories (individual facets in parentheses): \ncore summary themes (clout, analytical thinking, authentic, emotional tone), drives (affiliation, \nachievement, power), cognition (all-or-none, cognitive processes, insight, causation, \ndiscrepancy, tentative, certitude, differentiation), affect (positive tone, negative tone, positive \nemotion, negative emotion, anxiety, anger, sadness), social processes (prosocial behavior, \npoliteness, interpersonal conflict, moralization, communication), culture (politics, ethnicity, \ntechnology), lifestyle (leisure, home, work, money, religion), states (mental health, need, want, \nacquire, lack, fulfilled, fatigue), and motives (reward, risk, curiosity, allure). The prevalence of \neach of these themes in parent and friend text was then separately correlated with linguistic \npreference scores. We kept themes that showed a correlation of r > |.05|. This threshold was \nchosen because it roughly corresponded to the threshold for statistical significance in this \nsubsample and because it minimized the risk of falsely excluding an interesting or meaningful \ntheme. Results indicate that affective themes (e.g., risk, want, fatigue) in parent text are the \nprimary drivers of linguistic preference scores; cognitive and state-based themes (e.g., \ncausation, mental health) are ostensibly the primary drivers of linguistic preference scores for \nfriend text (Table 6). Notably, themes in parent text tended to covary more with linguistic \npreference scores than those in friend text insofar that more parent text themes evinced \ncorrelations that met our threshold. \nSecond, we took the themes identified in Table 6 and extracted them from Study 1 \nparticipants’ text data, and then correlated their prevalence in that data with self-other neural \nsimilarity from ROIs that showed the strongest relationships with linguistic preference scores \n(Parent: R NAcc, mPFC, dmPFC, L TPJ; Friend: L NAcc, mPFC, dmPFC, L TPJ; see Figures 4-\n5).  These correlations are depicted as Cleveland plots in Figures 6-7. The most consistently \nstrong themes for parents and friends were identified by taking the magnitude (absolute value) \nof each theme’s correlation with neural data and averaging it across the ROIs. These are listed \nin the top right corner of Figures 6-7. Because the purpose of this analysis was meant to \ndescriptively unpack what is happening in our sample and help generate future research, we did \nnot conduct inferential statistics on the estimates. Affective and state-based themes were most \nconsistently associated with self-parent representational overlap across ROIs, whereas \ncognitive themes were most consistently associated with self-friend representational overlap \nacross ROIs. Because of potential negations in the text data (e.g., ‘I do not feel happy’) we \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 24 \nencourage readers to primarily interpret the magnitude, rather than the sign, of these \ncorrelations.  \n \n \nDiscussion \n \nIn the current study, we found evidence that neural and linguistic representations of \nclose others contain information about one’s social decision preferences involving these \nindividuals. We first observed that neural representations of parents and friends are differentially \nencoded in the brain with respect to the self, such that representations of one’s parents are \nmore similar to representations of oneself in bilateral NAcc and TPJ, whereas representations of \none’s friends are more similar to representations of oneself in mPFC. Individual differences in \nthe encoding ‘bias’ of these brain regions were related to and possibly consequential for social \ndecision preferences involving close others, as self-other representational similarities in the \nNAcc, TPJ, and dmPFC tracked with between-individual social decision preferences. \nLinguistically, written memories and descriptors of parents and friends were used to train a \nlinguistic signature of social decision preferences. When applied to new data, the signature was \nable to predict choice preferences between parents and friends on novel decision tasks and was \neven related to other relationship characteristics (e.g., relationship quality). Critically, the scores \nfrom the linguistic signature and neural data were related, showing correspondence of \nrepresentations measured with linguistic and neural data. Finally, we found that individual \ndifferences in the degree of linguistic-neural correspondence were also predictive of social \ndecision preferences. These findings collectively enrich our understanding of how the mental \nmodels that humans create for specific others relate to choices regarding those others.  \n Existing studies on social decision-making predominantly focus on understanding \ndecision behavior in terms of the cognitive heuristics that individuals apply over situation-level \nfeatures or how individuals weigh the mental states of others (Austerweil et al., 2016; Cole & \nBruno Teboul, 2004; Crockett et al., 2017; FeldmanHall & Chang, 2018; Hackel et al., 2017; Kao \net al., 2023; Sampaio et al., 2023; Yu et al., 2019). Our findings suggest that an additional driver \nof social decision preferences stems from how representations of oneself and others are aligned \nin the brain. We found that self-other similarity in mPFC and dmPFC was associated with choice \npreferences between a parent and friend. Specifically, greater neural representational overlap \nbetween self and parent was associated with a greater likelihood of favoring one’s parent-over-\nfriend, whereas less self and friend representational overlap (i.e., greater dissimilarity) was \nassociated with stronger friend-over-parent preferences. The NAcc was also predictive of choice \npreferences, but the direction of the effect varied based on several factors. These varied \npatterns, paired with existing psychology and neuroscience literature showing that \nrepresentations of others are linked to a wide range of cognitive, affective, and semantic \nprocesses (Guassi Moreira et al., 2023; Lin & Thornton, 2023; Tamir & Thornton, 2018; Todd & \nTamir, 2024), suggest these brain regions could be storing and accessing different types of \nrepresentational information and/or associated mental processes. Indeed, descriptive patterns \nin the written text data appear to provide some evidence for this possibility. Variation in the \nspecific brain regions implicated as a function of outcome type (monetary or social) is consistent \nwith the notion that brain regions store and access different kinds of representational \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 25 \ninformation that may be selectively accessed in different decision contexts. That said, we note \nthat similarities in representations of oneself and others in the NAcc were associated with \ndecision preferences across both outcome types. This suggests that representations of familiar \nothers in the NAcc region may shape social decisions regarding those individuals across a wide \nrange of contexts and outcomes, which may be related to the motivational and reward-related \nprocess that this region supports. Our pattern of findings overall hint at the idea that ‘encoding \nbiases’ towards one’s parent or friend in the brain regions studied are related to choice \npreferences implies the possibility of affordances in these representations (de Wit et al., 2017; \nKourtis et al., 2018; McMahon & Isik, 2023), such that the way in which information (here, \nrepresentations of close others) is encoded might facilitate particular behavioral tendencies \nduring decision-making. \n Relatedly, our linguistic findings—the prediction of choice preferences from text data and \nmodel weights generalizing to other facets of relationships—point towards a similar concept of \n‘psychological affordances’. The most readily available mental information when prompted to \ndescribe a close other or write down a memory about them (the prompts used to elicit the text \ndata that was used to train the linguistic signature) may also be the most readily available or \nsalient information when making choices that impact said close other. That our decision tasks \ndid not allow for prolonged deliberation between trials meant there was a degree of spontaneity \nrequired when responding. Having one’s representations of other people contain behavioral \naffordances is a potentially efficient facet of the psychological architecture that links mental \nmodels with behaviors.  \nThe joint associations of neural and linguistic mental representations with decision \npreferences here are a particularly interesting facet of investigation with respect to the ability of \nlinguistic information to disambiguate brain-behavior associations. fMRI data cannot always be \nused to infer the exact operations performed by the brain. However, close correspondence \nbetween linguistic and neural data can refine hypotheses about the types of information the \nbrain is using. Here we observed mixed evidence: while linguistic preference scores were \nbroadly correlated with imaging data, high correspondence between linguistic and neural \ninformation (e.g., strong correlation between neural similarity and linguistic preference scores) \ndid not always augment social choice preferences. The results of these analyses can be \ninterpreted in different ways. On the one hand, if indices of neural and linguistic representations \nshow high correspondence, it could indicate convergence in the types of information carried by \neach and make one of the pair redundant when modeling social decision preferences (such as \nin the case when monetary outcomes were at stake). On the other hand, if a high correlation \nbetween linguistic and neural representational information was redundant, then why would it be \nassociated with augmented decision preferences (such as the case with social outcomes)? One \nexplanation could be that having a high linguistic preference score could indicate a shift in the \ntype of content encoded in mental representations in such a way that is possibly more \nconsequential for shaping decision preferences. That this occurs for social but not monetary \noutcomes could mean that decisions involving spending time with others are particularly \nsensitive to representational content that is most readily accessed in autobiographical content, \nthough this would need to be tested directly. Linguistically, neural representational overlap with \nboth parent and friend relationships was associated with motivationally relevant themes. Parent \nneural representations were more uniquely linked with state-based themes, whereas friend \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 26 \nneural representations evinced stronger relationships with social cognitive and mental health \nrelated themes. While the results of these analyses are meant to be descriptive and to inform \nhypotheses for future research, they broadly suggest that representations of specific familiar \nothers are comprised of distinct psychological content.  \nWhile our findings highlight an important association between mental representations \nand social decision-making, and we hypothesize that mental representations are a driving force \nin shaping behavior, we must also acknowledge that one’s social choice preferences involving \nclose others may very well sculpt the architecture of mental representations. Contextual factors \nunrelated to mental representations may also constitute potent forces that shape initial social \ndecision preferences, which then in turn may influence the construction of mental \nrepresentations. Another possibility is that a bidirectional relationship may exist such that \nindividuals who prioritize their relationships with particular familiar others may develop mental \nrepresentations of those people that are more similar to their mental representations of \nthemselves, which could subsequently influence their downstream decision-making. Future \nresearch should explore this interplay more deeply, potentially using longitudinal designs to \nexamine how relationship dynamics, decision preferences, and mental representations evolve \nover time. Additionally, qualitative methodologies could provide insights into how individuals \nnavigate these complexities in real-world decision-making scenarios. \nFuture studies could improve and extend on ours in several ways. First, researchers \nmay consider testing other types of familiar relationship categories, not just close others. \nUnderstanding how mental representations influence behavior is likely best accomplished by \nexamining such associations across the entire spectrum of psychological and social distance \nbetween an individual and the people in their social environment. Such efforts may also \nconsider exhaustively characterizing social decision preferences on a wider range of tasks with \nvaried experimental conditions (e.g., including self versus other conditions). Keeping with this \nspirit, it would also be helpful to conduct similar studies across different populations to assess \nthe generalizability of these findings. Second, longitudinal work is critical for quantifying the \ntemporal dynamics of how representations are formed and updated with experience, as this is \ncritical for making causal claims. Such avenues would benefit from complementary use of \nintensive longitudinal methods such as ecological momentary assessment, or incorporating \ndevelopmental populations that contain individuals who are frequently introduced to completely \nnovel social networks (e.g., incoming college students, middle or high schoolers) and thus may \nserve as ideal populations for pursuing this line of work.  \nOn a related note, given that mental representations of others are not static entities, \nanother benefit of longitudinal work would be to examine how mental representations of others \nfluctuate over time and how such fluctuations correspond to social decision-making behavior. \nAdditionally, some of the benefits of longitudinal work can also perhaps be attained without \nmultiple sessions, by varying the temporal ordering between social decision and representation-\neliciting tasks within a session to measure whether order effects exist and whether the decisions \nthat people make regarding familiar others impact their mental representations of those \nindividuals. Third and last, we believe that future work should incorporate more extensive written \ntext data about close others so as to bolster sensitivity by more extensively sampling one’s \nmental representations of those individuals. Ideally, future studies utilizing intensive longitudinal \nmethods would incorporate written diaries or essay-like prompts at baseline to gather even \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 27 \nmore text data about close others, or such studies could ethically obtain written records (e.g., \ntext messages) between participants and close others. Such approaches could also be used to \nhelp disentangle the role of mental representations in shaping decision behavior versus how \ncognitive heuristics such as aversion to inequity, risk, or ambiguity may be differentially \ndeployed when making choices that affect others compared to oneself. Longitudinal approaches \nmay be particularly helpful for this issue because they allow for the use of lagged analyses \ncould help better delineate the relative influence of mental models of others versus differential \ndeployment of cognitive heuristics.  \nFinally, our study speaks to the importance of studying known close others. This study is \nanother entry in a growing body of work that shows individuals have highly granular behavioral \nchoice tendencies involving other individuals (Brandner et al., 2021; Delton et al., 2023; Karan et \nal., 2022; Sznycer, 2022). Understanding human behavior requires examining the most relevant \nand meaningful phenomena in our daily experiences. Knowing how individuals represent \nspecific others and in turn use these representational distinctions to guide behavior is critical for \nunderstanding much of human behavior as it unfolds outside the laboratory.  \n \nAcknowledgments \nPreparation of this research was supported by a National Science Foundation SBE Postdoctoral \nResearch Fellowship to JFGM (award number 2104629) and the UCLA Department of \nPsychology (CP). We thank Dr. Rick Dale for comments on the linguistic analyses. We thank \nthe members of the Computational Social Neuroscience Laboratory at UCLA for their feedback \non study concept and manuscript.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 28 \nFigure 1. Study Overview \n \n \nNote. (A) Participants in both studies nominated a parent and same-age friend of their choice \nwho was not a current romantic partner or family member. (B) Participants in Study 1 underwent \nfMRI scanning while completing a task designed to elicit mental representations of the self, \nparent, and friend, as well as a social decision-making task that involved allocating monetary \nand social rewards between parent and friend. (C) RSA was used to compare parent and friend \nrepresentations (directly, and relative to the self) in six a priori ROIs and in exploratory \nsearchlight analyses. ROI mask visualizations can be accessed in Figure 1-1 of the Extended \nData. (D) Hierarchical modeling was used to test for associations between individual differences \nin self - other (parent and friend) representational similarity in the ROIs and decision \npreferences between parents and friends. (E) Study 2 participants provided written memories \nabout, and/or adjectives describing their nominated parent and friend and completed a one-shot \nsocial decision task. Semantic features for parent and friend text were extracted (Word2Vec \nembeddings) and used to create and validate a linguistic signature of participants’ decision \npreferences between parents and friends on the decision task. The linguistic signature yielded \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 29 \n‘linguistic preference scores’ that quantified implicit social choice preferences embedding in \nlinguistic representations. These scores generalized to additional data (other social decision \ntasks, questionnaire-based measures of relationship characteristics, such as quality and social \nloss aversion). (F) The linguistic signature was used to compute linguistic preference scores on \nwritten text data obtained from Study 1 participants. Scores were entered as moderators in \nmodels predicting decision preferences from neural representational similarity. The study \nworkflow is summarized in (G).  \n \n  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 30 \nFigure 2. Brain regions involved in social cognition and value-based processes differentially \nconstruct representations of parents and friends in relation to the self.  \n \n \n \nNote. ‘dmPFC’ refers to dorsomedial prefrontal cortex. ‘mPFC’ refers to medial prefrontal cortex. \n‘TPJ’ refers to temporoparietal junction. ‘NAcc’ refers to nucleus accumbens. ‘L’ refers to left \nand ‘R’ refers to right. Cohen’s d values reflect paired differences between (self, parent) - (self, \nfriend) dissimilarities of multivoxel response patterns extracted from the six brain regions \ndisplayed here. Error bars correspond to posterior standard deviations. Exploratory searchlight \nresults to complement this analysis are depicted in Figures 2-1 and 2-2 in the Extended Data. \nFigure 2-3 of Extended Data shows the results of mass univariate analyses for comparison. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 31 \nFigure 3. Linguistic preference scores directly predict social decision preferences on a novel \ndecision task (dictator game).  \n \nNote. Linguistic preference scores refer to model implied preferences between a parent and \nfriend based on written text data; relatively greater values indicate a parent-over-friend \npreference, relatively lower values indicate a friend-over-parent preference. The Y-axis depicts \nthe percentage of a hypothetical $100 endowment allocated to one’s parent (rather than one’s \nfriend); greater values thus indicate stronger parent-over-friend preferences in this task, \nwhereas lower values indicate the opposite. Linguistic preference scores were calculated with \nweights trained on the Word2Vec-based linguistic representations. Figure 3-1 of the Extended \nData shows the relationship of the linguistic preference scores with social loss aversion and \nrelationship quality. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 32 \nFigure 4. Neural and linguistic representations of parents and friends are correlated in the TPJ, \nNAcc, and dmPFC.   \n \n \nNote. Linguistic preference scores refer to model implied social decision preferences between a \nparent and friend based on written text data; relatively greater values indicate a parent-over-\nfriend preference, relatively lower values indicate a friend-over-parent preference. The Y-axis \ndepicts neural dissimilarity between self and other (parent, friend) representations across six \nneural ROIs. Linguistic preference scores were calculated with weights trained on Word2Vec-\nbased linguistic representations. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 33 \nFigure 5. Correspondence between linguistic and neural representations in three ROIs is \nassociated with stronger decision preferences \n \n \nNote. Linguistic preference scores refer to model implied social decision preferences between a \nparent and friend based on written text data; relatively greater values indicate a parent-over-\nfriend preference, relatively lower values indicate a friend-over-parent preference. The Y-axis \nrepresents choice preferences between parent and friend. The X-axis depicts neural \nrepresentational dissimilarity between the self and a target other (parent, friend) for a given ROI. \nThe relevant ROI is listed at the top of the top of plot, along with the outcome domain of the \ndecision task (monetary or social). Linguistic preference scores were calculated with weights \ntrained on the Word2Vec-based linguistic representations. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 34 \nFigure 6. Unpacking the correspondence between linguistic and neural representations of \nparents: Unique variance explained in self-parent neural dissimilarity by each linguistic theme.   \n \n \nNote. Linguistic themes analyzed here are those from the ‘parent’ column of Table 6. ROIs \nanalyzed here are those showing the strongest relationship with linguistic preference scores \n(Figures 4-5). Datapoints represent the squared semi-partial correlation between the prevalence \nof a linguistic theme in written text data (about parents) and self-parent neural dissimilarity in an \nROI (i.e., the unique variance each theme accounts for in the neural data given all other \nthemes). This is an exploratory descriptive analysis meant to understand what semantic content \nis encoded in self-other neural representations. Pearson’s r coefficient was used. Each term in \nthe word cloud corresponds to a linguistic theme, and their sizes are scaled by the unique \nvariance explained. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 35 \nFigure 7. Unpacking the correspondence between linguistic and neural representations of \nfriends: Unique variance explained in self-friend neural dissimilarity by each linguistic theme.   \n \n \nNote. Linguistic themes analyzed here are those from the ‘friend’ column of Table 6. ROIs \nanalyzed here are those showing the strongest relationship with linguistic preference scores \n(Figures 4-5). Datapoints represent the squared semi-partial correlation between the prevalence \nof a linguistic theme in written text data (about friends) and self-friend neural dissimilarity in an \nROI (i.e., the unique variance each theme accounts for in the neural data given all other \nthemes). This is an exploratory descriptive analysis meant to understand what semantic content \nis encoded in self-other neural representations. Pearson’s r coefficient was used. Each term in \nthe word cloud corresponds to a linguistic theme, and their sizes are scaled by the unique \nvariance explained. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 36 \nTable 1. Friend representations are more similar to self representations in cortical midline \nregions, whereas parent representations are more similar to self representations in the TPJ and \nNAcc \n \nROI Paired Difference [89% HDI] \ndmPFC 0.04 [-0.15, 0.25] \nmPFC 0.11 [-0.10, 0.30] \nL TPJ -0.11 [-0.31, 0.09] \nR TPJ -0.18 [-0.38, 0.03] \nL NAcc -0.14 [-0.35, 0.08] \nR NAcc -0.11 [-0.30, 0.10] \n \nNote. ‘dmPFC’ refers to dorsomedial prefrontal cortex. ‘mPFC’ refers to medial prefrontal cortex. \n‘TPJ’ refers to temporoparietal junction. ‘NAcc’ refers to nucleus accumbens. ‘L’ rand ‘R’ refer to \nleft and right hemisphere, respectively. Values in the table reflect Cohen’s d statistics (posterior \nmean) for paired comparisons between dissimilarities of multivoxel representations of self and \nparent, and self and friend, in each brain region. Neural dissimilarities were calculated using \nEuclidean distance. Values in brackets reflect 89% highest density credible intervals drawn \naround the posterior distribution. ROPE was determined as [-0.1, 0.1]. Differences where the \nHDI partially overlaps with ROPE and includes 0 (indicating modest evidence for a difference \nbetween self-other overlap for friends vs parents) are italicized. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 37 \nTable 2. Modeling social decision preferences as a function of the overlap of neural \nrepresentations of self and parent, and self and friend, in dmPFC and mPFC \n Monetary Outcomes Social Outcomes \n dmPFC mPFC dmPFC mPFC \nIntercept -2.49 [-3.05, -1.93] -2.52 [-3.07, -1.95] -0.92 [-1.33, -0.50] -0.92 [-1.32, -0.50] \nCondition -0.08 [-0.57, 0.44] -0.06 [-0.63, 0.48] -0.27 [-0.81, 0.21] -0.28 [-0.78, 0.19] \nReward ratio -2.96 [-3.56, -2.39] -2.94 [-3.52, -2.34] -1.29 [-1.60, -0.93] -1.28 [-1.62, -0.96] \nSelf-parent \nneural \ndissimilarity \n0.45 [0.01, 0.90] 0.31 [-0.14, 0.80] 0.02 [-0.34, 0.38] 0.09 [-0.28, 0.45] \nSelf-friend \nneural \ndissimilarity \n-0.22 [-0.65, 0.23] 0.17 [-0.31, 0.62] 0.01 [-0.36, 0.37] 0.33 [-0.04, 0.66] \nCondition x \nSelf-parent \nneural \ndissimilarity \n-0.98 [-1.51, -0.44] -0.26 [-0.81, 0.34] -0.23 [-0.78, 0.28] -0.17 [-0.70, 0.36] \nCondition x \nSelf-friend \nneural \ndissimilarity \n-0.07 [-0.58, 0.45] -0.24 [-0.82, 0.35] -0.28 [-0.79, 0.26] -0.58 [-1.11, -0.07] \nLOOIC (SE) 1322.4 (46.0) 1325.0 (46.1) 1703.5 (43.6) 1704.2 (43.5) \nPercent Pareto \nK < 0.7 95.5% 95.5% 99.8% 99.7% \n \nNote. Coefficients are on a logit scale and reflect means from a posterior distribution. ‘Condition’ \nindicates when parent or friend outcomes were associated with the discounting or delay options \nand encodes social decision preferences (positive values indicate a parent-over-friend \npreference, negative values indicate a friend-over-parent preference). Reward ratio reflects the \ndivision of the non-discounting option over the discounting option. Self-parent and self-friend \nneural dissimilarities refer to the distance between multivoxel neural patterns for oneself and a \nparent or friend, extracted from the brain region listed in the column heading. Neural \ndissimilarities were calculated via Euclidean distance. Values in brackets reflect 89% highest \ndensity credible intervals drawn around the posterior distribution. Interactions reflect \nassociations between neural dissimilarities and social decision preferences. Coefficients where \nthe HDI partially overlaps with ROPE (modest or moderate evidence depending on whether the \ninterval contained 0) are italicized; coefficients where the HDI does not overlap with ROPE \n(strong evidence) are bolded and italicized. Leave One Out Information Criterion (LOOIC) is \npresented as a value of fit for each model, along with its standard errors in parentheses. Lower \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 38 \nLOOIC values reflect better fit. The percentage of Pareto K distribution values exceeding a 0.7 \ncut-off is provided as a diagnostic for gauging the reliability of LOOIC values.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 39 \nTable 3. Modeling social decision preferences as a function of the overlap of neural \nrepresentations of self and parent, and self and friend, bilaterally in the TPJ and NAcc \n Monetary Outcomes Social Outcomes \n TPJ NAcc TPJ NAcc \nIntercept -2.51 [-3.01, -1.99] -2.56 [-3.07, -1.95] -0.91 [-1.32, -0.51] -0.94 [-1.35, -0.51] \nCondition -0.06 [-0.59, 0.50] -0.04 [-0.61, 0.49] -0.28 [-0.80, 0.20] -0.28 [-0.78, 0.22] \nReward ratio -2.90 [-3.46, -2.30] -3.02 [-3.62, -2.43] -1.28 [-1.59, -0.94] -1.31 [-1.65, -0.99] \nSelf-parent \nneural \ndissimilarity (L) \n0.19 [-0.23, 0.62] 0.07 [-0.44, 0.57] -0.28 [-0.63, 0.09] -0.07 [-0.47, 0.29] \nSelf-friend \nneural \ndissimilarity (L) \n-0.49 [-0.93, -0.05] 0.09 [-0.40, 0.60] -0.42 [-0.79, -0.03] -0.30 [-0.68, 0.08] \nSelf-parent \nneural \ndissimilarity (R) \n0.62 [0.20, 1.07] -0.13 [-0.60, 0.35] 0.15 [-0.24, 0.53] 0.19 [-0.19, 0.54] \nSelf-friend \nneural \ndissimilarity (R) \n0.70 [0.25, 1.12] -0.08 [-0.52, 0.38] 0.44 [0.07, 0.81] -0.02 [-0.37, 0.32] \nCondition x \nSelf-parent \nneural \ndissimilarity (L) \n-0.25 [-0.84, 0.36] -0.39 [-0.97, 0.21] -0.14 [-0.68, 0.42] 0.05 [-0.48, 0.62] \nCondition x \nSelf-friend \nneural \ndissimilarity (L) \n0.03 [-0.56, 0.66] -0.48 [-1.11, 0.08] -0.23 [-0.80, 0.36] -0.01 [-0.59, 0.54] \nCondition x \nSelf-parent \nneural \ndissimilarity (R) \n-0.46 [-1.05, 0.13] 0.20 [-0.38, 0.74] 0.41 [-0.17, 0.99] -0.50 [-1.08, 0.03] \nCondition x \nSelf-friend \nneural \ndissimilarity (R) \n-0.42 [-1.01, 0.20] 0.81 [0.23, 1.35] 0.03 [-0.57, 0.56] 0.02 [-0.51, 0.54] \nLOOIC 1325.1 (46.4) 1317.9 (46.1) 1700.6 (43.9) 1700.8 (43.8) \nPercent Pareto \nK > 0.7 95.7% 94.9% 99.5% 95.5% \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 40 \n \nNote. Coefficients are on a logit scale and reflect means from a posterior distribution. ‘Condition’ \nindicates when parent or friend outcomes were associated with the discounting or delay options \nand thus encodes social decision preferences (positive values indicate a parent-over-friend \npreference, negative values indicate a friend-over-parent preference). Reward ratio reflects the \ndivision of the non-discounting option over the discounting option. Self-parent and self-friend \nneural dissimilarities refers to the distance between multivoxel neural patterns for oneself and a \nparent or friend, extracted from the brain region listed in the column heading. Neural \ndissimilarities were calculated via Euclidean distance. Values in brackets reflect 89% highest \ndensity credible intervals drawn around the posterior distribution ‘L’ rand ‘R’ refer to left and right \nhemisphere, respectively. Interactions reflect associations between neural dissimilarities and \nsocial decision preferences. Interaction slopes where the HDI partially overlaps with ROPE \n(modest or moderate evidence, depending on whether the interval includes zero) are italicized; \ninteraction slopes where the HDI does not overlap with ROPE (strong evidence) are bolded and \nitalicized. Leave One Out Information Criterion (LOOIC) is presented as a value of fit for each \nmodel, along with its standard errors in parentheses. Lower LOOIC values reflect better fit. The \npercentage of Pareto K distribution values exceeding a 0.7 cut-off is provided as a diagnostic for \ngauging the reliability of LOOIC values. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 41 \nTable 4. Predictive models using data from neural RDM cells outperform baseline models that \nonly rely on task data.  \nModel Monetary Outcome Social Outcome \nBaseline (condition, reward \nratio) 0.589 0.553 \ndmPFC (condition, reward \nratio, self-other dmPFC \nRDM cells, RDM cells x \ncondition interactions) \n0.641  \nΔF1 = .052 [0.032, 0.071] \n0.562  \nΔF1 = .009 [-0.003, 0.023] \nmPFC (condition, reward \nratio, self-other mPFC \nRDM cells, RDM cells x \ncondition interactions) \n0.614 \nΔF1 = .025 [0.007, 0.035] \n0.584  \nΔF1 = .031 [0.013, 0.048] \nNAcc (condition, reward \nratio, self-other NAcc RDM \ncells, RDM cells x \ncondition interactions) \n0.607  \nΔF1 = .018 [-0.001, 0.032] \n0.578  \nΔF1 = .025 [0.001, 0.032] \nTPJ (condition, reward \nratio, self-other TPJ RDM \ncells, RDM cells x \ncondition interactions) \n0.647  \nΔF1 = .058 [0.036, 0.079] \n0.600  \nΔF1 = .047 [0.023, 0.068] \n \nNote. The first row in the table lists F1 scores for baseline models predicting choices regarding \nmonetary and social outcomes with task features only (condition, reward ratio). From the \nsecond row onward, each row lists the F1 score and a ΔF1 value for a model that was trained \nwith data from the neural RDM cells for a given ROI. Values in brackets refer to 95% \nbootstrapped confidence intervals for the difference between the F1 score for the model in \nquestion and the baseline model. ‘dmPFC’ refers to dorsomedial prefrontal cortex. ‘mPFC’ \nrefers to medial prefrontal cortex. ‘TPJ’ refers to temporoparietal junction. ‘NAcc’ refers to \nnucleus accumbens. \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 42 \nTable 5. Correlations between linguistic preference scores and neural representational \ndissimilarities between self and other across two close others (targets) and six ROIs \n \nROI Target Robust Correlation [89% HDI] \ndmPFC Parent -0.03 [-0.24, 0.18] \ndmPFC Friend 0.15 [-0.07, 0.35] \nmPFC Parent -0.19 [-0.41, 0.00] \nmPFC Friend 0.15 [-0.05, 0.36] \nL NAcc Parent -0.07 [-0.27, 0.14] \nL NAcc Friend 0.30 [0.11, 0.51] \nR NAcc Parent -0.19 [-0.38, 0.04] \nR NAcc Friend -0.02 [-0.24, 0.20] \nL TPJ Parent -0.11 [-0.32, 0.10] \nL TPJ Friend 0.15 [-0.04, 0.37] \nR TPJ Parent 0.01 [-0.21, 0.22] \nR TPJ Friend 0.12 [-0.08, 0.33] \n \nNote. ‘dmPFC’ refers to dorsomedial prefrontal cortex. ‘mPFC’ refers to medial prefrontal cortex. \n‘TPJ’ refers to temporoparietal junction. ‘NAcc’ refers to nucleus accumbens. ‘L’ and ‘R’ refer to \nleft and right hemisphere, respectively. Values in brackets reflect 89% highest density credible \nintervals drawn around the posterior distribution. Correlation values reflect a robust version of \nPearson’s r obtained by using t-distributed likelihood function. Neural dissimilarities were \nobtained with Euclidean distance. Linguistic preference scores were derived from Word2Vec-\nbased linguistic representations. Correlations where the HDI partially overlaps with ROPE \n(modest or moderate evidence depending on whether the interval contained 0) are italicized; \ncorrelations where the HDI does not overlap with ROPE are bolded and italicized (robust \nevidence). Negative correlation values are expected for parent, positive values are expected for \nfriend.  \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 43 \nTable 6. Linguistic themes from parent and friend linguistic representations most strongly \ncorrelated with linguistic preference scores \nThematic Category Parent text and linguistic \npreference score correlation \nFriend text and linguistic \npreference score correlation \nCore \nClout 0.05 - \nTone 0.15 - \nDrives \nAchievement - -0.06 \nCognition \nAll-or-None 0.06 - \nCognitive Processes - -0.11 \nInsight - -0.09 \nCausation - -0.09 \nDiscrepancy 0.07 - \nTentative 0.05 - \nCertitude 0.09 - \nDifferentiation 0.05 - \nAffect \nPositive Tone 0.14 0.07 \nNegative Tone -0.24 - \nPositive Emotion 0.09 0.05 \nNegative Emotion -0.24 - \nAnxiety -0.21 -0.05 \nAnger -0.12 - \nSadness -0.11 - \nSocial Processes \nPoliteness 0.11 - \nInterpersonal Conflict -0.05 - \nMoralization 0.05 0.08 \nLifestyle \nReligion 0.07 - \nStates \nMental Health -0.13 0.05 \nNeed -0.05 - \nWant 0.05 - \nFatigue -0.09 - \nMotives \nRisk -0.05 - \nCuriosity -0.06 -0.12 \nAllure 0.17 0.08 \n \nNote. Linguistic themes were obtained by independently passing parent and friend written \nrepresentations through LIWC-22 and then correlating prevalence for each theme with linguistic \npreference scores. Positive correlation values in the ‘parent’ column indicate that greater \nprevalence of a theme is associated with a higher linguistic preference score and thus a \nstronger parent-over-friend social decision preference. Negative correlation values in the ‘friend’ \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 44 \ncolumn indicate that greater prevalence of a theme is associated with a lower linguistic \npreference score and thus a stronger friend-over-parent social decision preference. This is an \nexploratory descriptive analysis meant to inform hypotheses regarding the semantic content that \ndrives linguistic preference scores. Correlations below r = |0.05| were excluded. This threshold \nwas chosen to be inclusive given the exploratory nature of the analysis. For results of analyses \ninvolving the interaction between linguistic preference scores and neural dissimilarity, see \nTables 6-1 and 6-2 in Extended Data. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 45 \nReferences \nAlink, A., Walther, A., Krugliak, A., van den Bosch, J. J. F., & Kriegeskorte, N. (2015). Mind \nthe drift - improving sensitivity to fMRI pattern information by accounting for temporal \npattern drift. BioRxiv. https://doi.org/10.1101/032391 \nAusterweil, J. L., Brawner, S., Greenwald, A., Hilliard, E., Ho, M. K., Littman, M., MacGlashan, \nJ., & Trimbach, C. (2016). The Impact of Other-Regarding Preferences in a Collection of \nNon-Zero-Sum Grid Games. AAAI spring symposium 2016 on challenges and opportunities in \nmultiagent learning for the real world, Palo Alto, CA. \nAvants, B. B., Epstein, C. L., Grossman, M., & Gee, J. C. (2008). Symmetric diffeomorphic \nimage registration with cross-correlation: evaluating automated labeling of elderly and \nneurodegenerative brain. Medical Image Analysis, 12(1), 26–41. \nhttps://doi.org/10.1016/j.media.2007.06.004 \nBehzadi, Y., Restom, K., Liau, J., & Liu, T. T. (2007). A component based noise correction \nmethod (CompCor) for BOLD and perfusion based fMRI. Neuroimage, 37(1), 90–101. \nhttps://doi.org/10.1016/j.neuroimage.2007.04.042 \nBoyd, R. L., Ashokkumar, A., Seraj, S., & Pennebaker, J. W. (2022). The Development and \nPsychometric Properties of LIWC-22. Unpublished. \nhttps://doi.org/10.13140/rg.2.2.23890.43205 \nBrandner, P., Güroğlu, B., van de Groep, S., Spaans, J. P., & Crone, E. A. (2021). Happy for Us \nnot Them: Differences in neural activation in a vicarious reward task between family and \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 46 \nstrangers during adolescent development. Developmental Cognitive Neuroscience, 51, 100985. \nhttps://doi.org/10.1016/j.dcn.2021.100985 \nBroom, T. W., Chavez, R. S., & Wagner, D. D. (2021). Becoming the King in the North: \nidentification with fictional characters is associated with greater self-other neural overlap. \nSocial Cognitive and Affective Neuroscience, 16(6), 541–551. \nhttps://doi.org/10.1093/scan/nsab021 \nCadinu, M. R., & Rothbart, M. (1996). Self-anchoring and differentiation processes in the \nminimal group setting. Journal of Personality and Social Psychology, 70(4), 661–677. \nhttps://doi.org/10.1037/0022-3514.70.4.661 \nChang, L. J., Gianaros, P. J., Manuck, S. B., Krishnan, A., & Wager, T. D. (2015). A Sensitive \nand Specific Neural Signature for Picture-Induced Negative Affect. PLoS Biology, 13(6), \ne1002180. https://doi.org/10.1371/journal.pbio.1002180 \nChavez, R. S., & Wagner, D. D. (2020). The neural representation of self is recapitulated in the \nbrains of friends: A round-robin fMRI study. Journal of Personality and Social Psychology, \n118(3), 407–416. https://doi.org/10.1037/pspa0000178 \nCole, T., & Bruno Teboul, Jc. (2004). Non-zero-sum collaboration, reciprocity, and the \npreference for similarity: Developing an adaptive model of close relational functioning. \nPersonal Relationships, 11(2), 135–160. https://doi.org/10.1111/j.1475-6811.2004.00075.x \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 47 \nCrockett, M. J., Siegel, J. Z., Kurth-Nelson, Z., Dayan, P., & Dolan, R. J. (2017). Moral \ntransgressions corrupt neural representations of value. Nature Neuroscience, 20(6), 879–885. \nhttps://doi.org/10.1038/nn.4557 \nCrone, E. A., & Fuligni, A. J. (2020). Self and others in adolescence. Annual Review of \nPsychology, 71, 447–469. https://doi.org/10.1146/annurev-psych-010419-050937 \nCrosnoe, R., & Johnson, M. K. (2011). Research on Adolescence in the Twenty-First Century. \nAnnual Review of Sociology, 37, 439–460. https://doi.org/10.1146/annurev-soc-081309-\n150008 \nDale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical surface-based analysis. I. Segmentation \nand surface reconstruction. Neuroimage, 9(2), 179–194. \nhttps://doi.org/10.1006/nimg.1998.0395 \nDelgado, M. R., Beer, J. S., Fellows, L. K., Huettel, S. A., Platt, M. L., Quirk, G. J., & Schiller, \nD. (2016). Viewpoints: Dialogues on the functional role of the ventromedial prefrontal cortex. \nNature Neuroscience, 19(12), 1545–1552. https://doi.org/10.1038/nn.4438 \nDelton, A. W., Jaeggi, A. V., Lim, J., Sznycer, D., Gurven, M., Robertson, T. E., Sugiyama, L. \nS., Cosmides, L., & Tooby, J. (2023). Cognitive foundations for helping and harming others: \nMaking welfare tradeoffs in industrialized and small-scale societies. Evolution and Human \nBehavior, 44(5), 485–501. https://doi.org/10.1016/j.evolhumbehav.2023.01.013 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 48 \nde Leeuw, J. R., Gilbert, R. A., & Luchterhandt, B. (2023). jsPsych: Enabling an Open-Source \nCollaborative Ecosystem of Behavioral Experiments. The Journal of Open Source Software, \n8(85), 5351. https://doi.org/10.21105/joss.05351 \nDe Raad, B. (2000). The Big Five Personality Factors: The psycholexical approach to \npersonality. Hogrefe & Huber Publishers. \nde Wit, M. M., de Vries, S., van der Kamp, J., & Withagen, R. (2017). Affordances and \nneuroscience: Steps towards a successful marriage. Neuroscience and Biobehavioral Reviews, \n80, 622–629. https://doi.org/10.1016/j.neubiorev.2017.07.008 \nFatima, A., Li, Y., Hills, T. T., & Stella, M. (2021). Dasentimental: detecting depression, \nanxiety, and stress in texts via emotional recall, cognitive networks, and machine learning. Big \nData and Cognitive Computing, 5(4), 77. https://doi.org/10.3390/bdcc5040077 \nFeldmanHall, O., & Chang, L. J. (2018). Social Learning. In Goal-Directed Decision Making \n(pp. 309–330). Elsevier. https://doi.org/10.1016/B978-0-12-812098-9.00014-0 \nFigner, B., Mackinlay, R. J., Wilkening, F., & Weber, E. U. (2009). Affective and deliberative \nprocesses in risky choice: Age differences in risk taking in the Columbia Card Task. Journal \nof Experimental Psychology: Learning, Memory, and Cognition, 35(3), 709–730. \nFischer, I., & Savranevski, L. (2023). The effect of similarity perceptions on human cooperation \nand confrontation. Scientific Reports, 13(1), 19849. https://doi.org/10.1038/s41598-023-\n46609-8 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 49 \nFleming, A. C., & Slank, K. L. (2015). Making a choice: Self-other differences in decision \nmaking in risky situations. North American Journal of Psychology, 17(3), 633–648. \nFonov, V. S., Evans, A. C., McKinstry, R. C., Almli, C. R., & Collins, D. L. (2009). Unbiased \nnonlinear average age-appropriate brain templates from birth to adulthood. Neuroimage, 47, \nS102. https://doi.org/10.1016/S1053-8119(09)70884-5 \nFroehlich, L., Dorrough, A. R., Glöckner, A., & Stürmer, S. (2021). Similarity Predicts Cross-\nNational Social Preferences. Social Psychological and Personality Science, 12(8), 1486–1498. \nhttps://doi.org/10.1177/1948550620982704 \nFuligni, A. J. (2019). The need to contribute during adolescence. Perspectives on Psychological \nScience, 14(3), 331–343. https://doi.org/10.1177/1745691618805437 \nGolsteyn, B. H. H., Grönqvist, H., & Lindahl, L. (2014). Adolescent time preferences predict \nlifetime outcomes. The Economic Journal, 124(580), F739–F761. \nhttps://doi.org/10.1111/ecoj.12095 \nGorgolewski, K. J., Burns, C. D., Madison, C., Clark, D., Halchenko, Y. O., Waskom, M. L., & \nGhosh, S. S. (2011). Nipype: a flexible, lightweight and extensible neuroimaging data \nprocessing framework in python. Frontiers in Neuroinformatics, 5, 13. \nhttps://doi.org/10.3389/fninf.2011.00013 \nGreenberg, M. T., Siegel, J. M., & Leitch, C. J. (1983). The nature and importance of attachment \nrelationships to parents and peers during adolescence. Journal of Youth and Adolescence, \n12(5), 373–386. https://doi.org/10.1007/BF02088721 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 50 \nGreve, D. N., & Fischl, B. (2009). Accurate and robust brain image alignment using boundary-\nbased registration. Neuroimage, 48(1), 63–72. \nhttps://doi.org/10.1016/j.neuroimage.2009.06.060 \nGuassi Moreira, J. F., Méndez Leal, A. S., Waizman, Y. H., Tashjian, S. M., Galván, A., & \nSilvers, J. A. (2023). Value-based neural representations predict social decision preferences. \nCerebral Cortex, 33(13), 8605–8619. https://doi.org/10.1093/cercor/bhad144 \nGuassi Moreira, J. F., & Parkinson, C. (2024). A Behavioral Signature for Quantifying the Social \nValue of Interpersonal Relationships with Specific Others. Communications Psychology, \n2(84). \nGuassi Moreira, J. F., Tashjian, S. M., Galván, A., & Silvers, J. A. (2018). Parents versus peers: \nassessing the impact of social agents on decision making in young adults. Psychological \nScience, 29(9), 1526–1539. https://doi.org/10.1177/0956797618778497 \nGuassi Moreira, J. F., Tashjian, S. M., Galván, A., & Silvers, J. A. (2020). Is social decision \nmaking for close others consistent across domains and within individuals? Journal of \nExperimental Psychology: General, 149(8), 1509–1526. https://doi.org/10.1037/xge0000719 \nGuassi Moreira, J. F., Tashjian, S. M., Galván, A., & Silvers, J. A. (2021). Computational and \nmotivational mechanisms of human social decision making involving close others. Journal of \nExperimental Social Psychology, 93. https://doi.org/10.1016/j.jesp.2020.104086 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 51 \nHackel, L. M., Zaki, J., & Van Bavel, J. J. (2017). Social identity shapes social valuation: \nevidence from prosocial behavior and vicarious reward. Social Cognitive and Affective \nNeuroscience, 12(8), 1219–1228. https://doi.org/10.1093/scan/nsx045 \nHassabis, D., Spreng, R. N., Rusu, A. A., Robbins, C. A., Mar, R. A., & Schacter, D. L. (2014). \nImagine all the people: how the brain creates and uses personality models to predict behavior. \nCerebral Cortex, 24(8), 1979–1987. https://doi.org/10.1093/cercor/bht042 \nHawkins, R., Liu, I., Goldberg, A., & Griffiths, T. (2021). Respect the code: Speakers expect \nnovel conventions to generalize within but not across social group boundaries. Proceedings of \nthe Annual Meeting of the Cognitive Science Society. \nHo, M. K., Cushman, F., Littman, M. L., & Austerweil, J. L. (2021). Communication in action: \nPlanning and interpreting communicative demonstrations. Journal of Experimental \nPsychology: General, 150(11), 2246–2272. https://doi.org/10.1037/xge0001035 \nHuettel, S. A., & Kranton, R. E. (2012). Identity economics and the brain: uncovering the \nmechanisms of social conflict. Philosophical Transactions of the Royal Society of London. \nSeries B, Biological Sciences, 367(1589), 680–691. https://doi.org/10.1098/rstb.2011.0264 \nJenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002). Improved optimization for the \nrobust and accurate linear registration and motion correction of brain images. Neuroimage, \n17(2), 825–841. https://doi.org/10.1006/nimg.2002.1132 \nJohn, O. P., & Srivastava, S. (1999). The big-five trait taxonomy: History, measurement, and \ntheoretical perspectives (Vol. 2, pp. 102–138). \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 52 \nKao, C.-H., Feng, G. W., Hur, J. K., Jarvis, H., & Rutledge, R. B. (2023). Computational models \nof subjective feelings in psychiatry. Neuroscience and Biobehavioral Reviews, 145, 105008. \nhttps://doi.org/10.1016/j.neubiorev.2022.105008 \nKaran, M., Lazar, L., Leschak, C. J., Galván, A., Eisenberger, N. I., Uy, J. P., Dieffenbach, M. \nC., Crone, E. A., Telzer, E. H., & Fuligni, A. J. (2022). Giving to others and neural processing \nduring adolescence. Developmental Cognitive Neuroscience, 56, 101128. \nhttps://doi.org/10.1016/j.dcn.2022.101128 \nKlein, A., Ghosh, S. S., Bao, F. S., Giard, J., Häme, Y., Stavsky, E., Lee, N., Rossa, B., Reuter, \nM., Chaibub Neto, E., & Keshavan, A. (2017). Mindboggling morphometry of human brains. \nPLoS Computational Biology, 13(2), e1005350. https://doi.org/10.1371/journal.pcbi.1005350 \nKourtis, D., Vandemaele, P., & Vingerhoets, G. (2018). Concurrent Cortical Representations of \nFunction- and Size-Related Object Affordances: An fMRI Study. Cognitive, Affective & \nBehavioral Neuroscience, 18(6), 1221–1232. https://doi.org/10.3758/s13415-018-0633-1 \nKriegeskorte, N., Goebel, R., & Bandettini, P. (2006). Information-based functional brain \nmapping. Proceedings of the National Academy of Sciences of the United States of America, \n103(10), 3863–3868. https://doi.org/10.1073/pnas.0600244103 \nKriegeskorte, N., Mur, M., & Bandettini, P. (2008). Representational similarity analysis - \nconnecting the branches of systems neuroscience. Frontiers in Systems Neuroscience, 2, 4. \nhttps://doi.org/10.3389/neuro.06.004.2008 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 53 \nKrienen, F. M., Tu, P.-C., & Buckner, R. L. (2010). Clan mentality: evidence that the medial \nprefrontal cortex responds to close others. The Journal of Neuroscience, 30(41), 13906–\n13915. https://doi.org/10.1523/JNEUROSCI.2180-10.2010 \nKruschke, J. K. (2011). Bayesian Assessment of Null Values Via Parameter Estimation and \nModel Comparison. Perspectives on Psychological Science, 6(3), 299–312. \nhttps://doi.org/10.1177/1745691611406925 \nKruschke, J. K. (2013). Bayesian estimation supersedes the t test. Journal of Experimental \nPsychology: General, 142(2), 573–603. https://doi.org/10.1037/a0029146 \nKruschke, J. K. (2018). Rejecting or accepting parameter values in bayesian estimation. \nAdvances in Methods and Practices in Psychological Science, 1(2), 251524591877130. \nhttps://doi.org/10.1177/2515245918771304 \nLebeau, G., Consoli, S. M., Le Bouc, R., Sola-Gazagnes, A., Hartemann, A., Simon, D., Reach, \nG., Altman, J.-J., Pessiglione, M., Limosin, F., & Lemogne, C. (2016). Delay discounting of \ngains and losses, glycemic control and therapeutic adherence in type 2 diabetes. Behavioural \nProcesses, 132, 42–48. https://doi.org/10.1016/j.beproc.2016.09.006 \nLin, C., & Thornton, M. (2023). Evidence for bidirectional causation between trait and mental \nstate inferences. Journal of Experimental Social Psychology, 108, 104495. \nhttps://doi.org/10.1016/j.jesp.2023.104495 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 54 \nMakowski, D., Ben-Shachar, M., & Lüdecke, D. (2019). bayestestR: Describing Effects and their \nUncertainty, Existence and Significance within the Bayesian Framework. The Journal of Open \nSource Software, 4(40), 1541. https://doi.org/10.21105/joss.01541 \nMcElreath, R. (2015). Statistical Rethinking: A Bayesian Course with Examples in R and Stan. \nChapman and Hall/CRC. https://doi.org/10.1201/9781315372495 \nMcMahon, E., & Isik, L. (2023). Seeing social interactions. Trends in Cognitive Sciences, \n27(12), 1165–1179. https://doi.org/10.1016/j.tics.2023.09.001 \nMcNeish, D., Stapleton, L. M., & Silverman, R. D. (2017). On the unnecessary ubiquity of \nhierarchical linear modeling. Psychological Methods, 22(1), 114–140. \nhttps://doi.org/10.1037/met0000078 \nMikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word \nrepresentations in vector space. ArXiv. https://doi.org/10.48550/arxiv.1301.3781 \nMitchell, J. P., Banaji, M. R., & Macrae, C. N. (2005). The link between social cognition and \nself-referential thought in the medial prefrontal cortex. Journal of Cognitive Neuroscience, \n17(8), 1306–1315. https://doi.org/10.1162/0898929055002418 \nMitchell, J. P., Macrae, C. N., & Banaji, M. R. (2006). Dissociable medial prefrontal \ncontributions to judgments of similar and dissimilar others. Neuron, 50(4), 655–663. \nhttps://doi.org/10.1016/j.neuron.2006.03.040 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 55 \nMumford, J. A., Davis, T., & Poldrack, R. A. (2014). The impact of study design on pattern \nestimation for single-trial multivariate pattern analysis. Neuroimage, 103, 130–138. \nhttps://doi.org/10.1016/j.neuroimage.2014.09.026 \nPatriat, R., Reynolds, R. C., & Birn, R. M. (2017). An improved model of motion-related signal \nchanges in fMRI. Neuroimage, 144(Pt A), 74–82. \nhttps://doi.org/10.1016/j.neuroimage.2016.08.051 \nPeirce, J., Gray, J. R., Simpson, S., MacAskill, M., Höchenberger, R., Sogo, H., Kastman, E., & \nLindeløv, J. K. (2019). PsychoPy2: Experiments in behavior made easy. Behavior Research \nMethods, 51(1), 195–203. https://doi.org/10.3758/s13428-018-01193-y \nPfeifer, J. H., & Peake, S. J. (2012). Self-development: integrating cognitive, socioemotional, \nand neuroimaging perspectives. Developmental Cognitive Neuroscience, 2(1), 55–69. \nhttps://doi.org/10.1016/j.dcn.2011.07.012 \nPower, J. D., Mitra, A., Laumann, T. O., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. \n(2014). Methods to detect, characterize, and remove motion artifact in resting state fMRI. \nNeuroimage, 84, 320–341. https://doi.org/10.1016/j.neuroimage.2013.08.048 \nSampaio, W. M., Freitas, A. L., Rêgo, G. G., Morello, L. Y. N., & Boggio, P. S. (2023). Effects \nof co-players’ identity and reputation in the public goods game. Scientific Reports, 13(1), \n13520. https://doi.org/10.1038/s41598-023-40730-4 \nSarathy, V., Scheutz, M., Kenett, Y., Allaham, M. M., & Malle, B. F. (2017). Mental \nRepresentations and Computational Modeling of Context-Specific Human Norm Systems. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 56 \nSatterthwaite, T. D., Elliott, M. A., Gerraty, R. T., Ruparel, K., Loughead, J., Calkins, M. E., \nEickhoff, S. B., Hakonarson, H., Gur, R. C., Gur, R. E., & Wolf, D. H. (2013). An improved \nframework for confound regression and filtering for control of motion artifact in the \npreprocessing of resting-state functional connectivity data. Neuroimage, 64, 240–256. \nhttps://doi.org/10.1016/j.neuroimage.2012.08.052 \nSaxe, R., & Powell, L. J. (2006). It’s the thought that counts: specific brain regions for one \ncomponent of theory of mind. Psychological Science, 17(8), 692–699. \nhttps://doi.org/10.1111/j.1467-9280.2006.01768.x \nSeaman, K. L., Gorlick, M. A., Vekaria, K. M., Hsu, M., Zald, D. H., & Samanez-Larkin, G. R. \n(2016). Adult age differences in decision making across domains: Increased discounting of \nsocial and health-related rewards. Psychology and Aging, 31(7), 737–746. \nhttps://doi.org/10.1037/pag0000131 \nSpiegelhalter, D. J., Freedman, L. S., & Parmar, M. K. B. (1994). Bayesian approaches to \nrandomized trials. Journal of the Royal Statistical Society. Series A (Statistics in Society), \n157(3), 357. https://doi.org/10.2307/2983527 \nSyed, M., & Mitchell, L. L. (2013). Race, ethnicity, and emerging adulthood. Emerging \nAdulthood, 1(2), 83–95. https://doi.org/10.1177/2167696813480503 \nSznycer, D. (2022). Value computation in humans. Evolution and Human Behavior, 43(5), 367–\n380. https://doi.org/10.1016/j.evolhumbehav.2022.06.002 \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 57 \nTamir, D. I., & Mitchell, J. P. (2013). Anchoring and adjustment during social inferences. \nJournal of Experimental Psychology: General, 142(1), 151–162. \nhttps://doi.org/10.1037/a0028232 \nTamir, D. I., Thornton, M. A., Contreras, J. M., & Mitchell, J. P. (2016). Neural evidence that \nthree dimensions organize mental state representation: Rationality, social impact, and valence. \nProceedings of the National Academy of Sciences of the United States of America, 113(1), \n194–199. https://doi.org/10.1073/pnas.1511905112 \nTamir, D. I., & Thornton, M. A. (2018). Modeling the predictive social mind. Trends in \nCognitive Sciences, 22(3), 201–212. https://doi.org/10.1016/j.tics.2017.12.005 \nThornton, M. A., & Mitchell, J. P. (2018). Theories of person perception predict patterns of \nneural activity during mentalizing. Cerebral Cortex, 28(10), 3505–3520. \nhttps://doi.org/10.1093/cercor/bhx216 \nTodd, A. R., & Tamir, D. I. (2024). Factors that amplify and attenuate egocentric mentalizing. \nNature Reviews Psychology, 3(3), 164–180. https://doi.org/10.1038/s44159-024-00277-1 \nWager, T. D., Atlas, L. Y., Lindquist, M. A., Roy, M., Woo, C.-W., & Kross, E. (2013). An \nfMRI-based neurologic signature of physical pain. The New England Journal of Medicine, \n368(15), 1388–1397. https://doi.org/10.1056/NEJMoa1204471 \nWang, Yin, Collins, J. A., Koski, J., Nugiel, T., Metoki, A., & Olson, I. R. (2017). Dynamic \nneural architecture for social knowledge retrieval. Proceedings of the National Academy of \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 58 \nSciences of the United States of America, 114(16), E3305–E3314. \nhttps://doi.org/10.1073/pnas.1621234114 \nWang, Yingying, Lee, H., & Kuhl, B. A. (2023). Mapping multidimensional content \nrepresentations to neural and behavioral expressions of episodic memory. Neuroimage, 277, \n120222. https://doi.org/10.1016/j.neuroimage.2023.120222 \nWiens, B. L. (2002). Choosing an equivalence limit for noninferiority or equivalence studies. \nControlled Clinical Trials, 23(1), 2–14. https://doi.org/10.1016/S0197-2456(01)00196-9 \nYarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-\nscale automated synthesis of human functional neuroimaging data. Nature Methods, 8(8), \n665–670. https://doi.org/10.1038/nmeth.1635 \nYu, H., Siegel, J. Z., & Crockett, M. J. (2019). Modeling Morality in 3-D: Decision-Making, \nJudgment, and Inference. Topics in Cognitive Science, 11(2), 409–432. \nhttps://doi.org/10.1111/tops.12382 \nZhang, Y., Brady, M., & Smith, S. (2001). Segmentation of brain MR images through a hidden \nMarkov random field model and the expectation-maximization algorithm. IEEE Transactions \non Medical Imaging, 20(1), 45–57. https://doi.org/10.1109/42.906424    \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 59 \n \nExtended Data – Figure Captions \n \nFigure 1-1. The six masks used for ROI analyses. \n \nNote. Masks for the mPFC, dmPFC, and bilateral TPJ were created using Neurosynth. Bilateral \nNAcc masks were created using the Harvard-Oxford probabilistic atlas.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 60 \nFigure 2-1. Exploratory searchlight results showing distinct representations of parents and \nfriends  \n \nNote. Group level analyses were conducted via permutation test (10,000 iterations) using \nNilearn’s non_parametric_inference function and TFCE (FWE, p < .05). Negative log10 p-values \nare displayed as cluster intensity values. P-values are corrected for multiple comparisons (FWE, \np < .05). The color bar ranges correspond to the threshold needed to achieve FWE error control \n(1.3) and the lowest possible negative log10 p-value given the number of permutations (4).  \n \n \n \n \n \n \n \n \n \n \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 61 \nFigure 2-2. Uncorrected results of the exploratory searchlight analysis probing for distinct \nrepresentations of parent and friend representations. \n \nNote. Group level analyses were conducted via permutation test (10,000 iterations) using \nNilearn’s non_parametric_inference function. t-statistics are displayed for cluster intensity \nvalues (thresholded at 3.1). No corrections for multiple comparisons have been made to this \nimage; it is plotted purely for exploratory purposes.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 62 \nFigure 2-3. Univariate contrasts for the social preference and judgment task. \n \nNote. Group-level analyses were conducted via permutation test (10,000 iterations) using \nNilearn’s non_parametric_inference function and TFCE. Negative log10 p-values are displayed \nas cluster intensity values. Clusters depict FWE-corrected p-values (p < .05).  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 63 \nFigure 3-1. Linguistic preference scores strongly generalize to other relationship characteristics \n(relationship quality, social loss aversion) for parents and weakly generalize to social loss \naversion for friends. \n \nNote. Linguistic preference scores refer to model implied social decision preferences between a \nparent and friend based on written text data; relatively greater values indicate a parent-over-\nfriend preference, relatively lower values indicate a friend-over-parent preference. The Y-axis \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 64 \ndepicts relationship quality between participants and their nominated parent (left) and friend \n(right). Relationship quality was obtained with the self-administered IPPA. Social loss aversion \nis a one-shot item where participants are asked how upset they would be if they could no longer \nspend time with a given other. Linguistic preference scores were calculated with weights trained \non the Word2Vec-based linguistic representations.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 65 \nTable 6-1. Predicting social decision preferences with interactions between linguistic preference \nscores and neural dissimilarity scores (dmPFC, mPFC) \nTerm dmPFC – \nMonetary \nmPFC – Monetary dmPFC – Social mPFC – Social \nIntercept -2.50 [-3.07, -1.93] -2.44 [-3.03, -1.88] -0.94 [-1.40, -0.52] -0.89 [-1.33, -0.47] \nCondition -0.03 [-0.55, 0.51] -0.15 [-0.74, 0.38] -0.28 [-0.78, 0.25] -0.34 [-0.84, 0.16] \nReward ratio -2.96 [-3.56, -2.40] -2.95 [-3.52, -2.35] -1.30 [-1.64, -0.97] -1.29 [-1.61, -0.96] \nSelf-parent \nneural \ndissimilarity \n0.47 [0.00, 0.96] 0.28 [-0.22, 0.76] 0.03 [-0.35, 0.41] 0.04 [-0.33, 0.43] \nSelf-friend \nneural \ndissimilarity \n-0.20 [-0.66, 0.28] 0.09 [-0.39, 0.56] 0.05 [-0.35, 0.44] 0.31 [-0.07, 0.68] \nLinguistic \npreference \nscore \n0.15 [-0.27, 0.61] 0.21 [-0.25, 0.67] -0.05 [-0.40, 0.30] -0.07 [-0.42, 0.29] \nCondition x \nSelf-parent \nneural \ndissimilarity \n-1.07 [-1.62, -0.51] -0.30 [-0.87, 0.31] -0.20 [-0.74, 0.37] -0.05 [-0.59, 0.49] \nCondition x \nSelf-friend \nneural \ndissimilarity \n-0.05 [-0.60, 0.48] -0.11 [-0.70, 0.47] -0.38 [-0.95, 0.20] -0.60 [-1.17, -0.09] \nCondition x \nLinguistic \npreference \nscore \n-0.50 [-1.01, 0.01] -0.53 [-1.06, 0.05] 0.19 [-0.33, 0.69] 0.12 [-0.42, 0.62] \nSelf-parent \nneural \ndissimilarity x \nLinguistic \npreference \nscore \n0.11 [-0.40, 0.62] 0.18 [-0.35, 0.73] 0.16 [-0.23, 0.55] 0.02 [-0.39, 0.43] \nSelf-friend \nneural \ndissimilarity x \nLinguistic \npreference \nscore \n0.08 [-0.47, 0.63] -0.30 [-0.70, 0.47] 0.12 [-0.33, 0.58] -0.23 [-0.65, 0.21] \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 66 \nCondition x \nSelf-parent \nneural \ndissimilarity x \nLinguistic \npreference \nscore \n-0.24 [-0.82, 0.35] -0.11 [-0.77, 0.53] -0.24 [-0.79, 0.32] -0.50 [-1.10, 0.08] \nCondition x \nSelf-friend \nneural \ndissimilarity x \nLinguistic \npreference \nscore \n-0.64 [-1.07, 0.20] 0.45 [-0.26, 1.12] 0.08 [-0.56, 0.73] -0.21 [-0.85, 0.39] \n \nNote. Coefficients are on a logit scale and reflect means from a posterior distribution. ‘Condition’ \nindicates when parent or friend outcomes were associated with the discounting or delay options \nand encodes social decision preferences (positive values indicate a parent-over-friend \npreference, negative values indicate a friend-over-parent preference). Reward ratio reflects the \ndivision of the non-discounting option over the discounting option. Self-parent and self-friend \nneural dissimilarities refer to the distance between multivoxel neural patterns for oneself and a \nparent or friend, extracted from the brain region listed in the column heading. Values in brackets \nreflect 89% highest density credible intervals drawn around the posterior distribution. \nInteractions reflect associations between neural dissimilarities, linguistic preferences, and/or \nsocial decision preferences. Coefficients where the HDI partially overlaps with ROPE (modest or \nmoderate evidence depending on whether the interval contained 0) are italicized; coefficients \nwhere the HDI does not overlap with ROPE (strong evidence) are bolded and italicized. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 67 \nTable 6-2. Predicting social decision preferences with interactions between linguistic preference \nscores and neural dissimilarity scores (TPJ, NAcc) \nTerm TPJ – Monetary NAcc – Monetary  TPJ – Social NAcc – Social \nIntercept -2.46 [-3.00, -1.95] -2.69 [-3.31, -2.07] -0.85 [-1.27, -0.42] -0.99 [-1.44, -0.54] \nCondition -0.03 [-0.62, 0.51] 0.13 [-0.47, 0.73] -0.38 [-0.86, 0.17] -0.27 [-0.82, 0.28] \nReward ratio -2.91 [-3.52, -2.37] -3.01 [-3.62, -2.42] -1.29 [-1.62, -0.96] -1.32 [-1.64, -0.98] \nSelf-parent neural \ndissimilarity (L) 0.14 [-0.31, 0.56] 0.06 [-0.48, 0.58] -0.33 [-0.71, 0.04] -0.11 [-0.47, 0.32] \nSelf-friend neural \ndissimilarity (L) -0.63 [-1.08, -0.16] 0.10 [-0.49, 0.65] -0.52 [-0.91, -0.13] -0.37 [-0.82, 0.05] \nSelf-parent neural \ndissimilarity (R) 0.63 [0.17, 1.08] 0.00 [-0.54, 0.53] 0.17 [-0.22, 0.56] 0.27 [-0.14, 0.67] \nSelf-friend neural \ndissimilarity (R) 0.74 [0.29, 1.18] -0.06 [-0.56, 0.46] 0.49 [0.11, 0.86] -0.02 [-0.39, 0.36] \nLinguistic \npreference score 0.10 [-0.35, 0.50] 0.15 [-0.38, 0.68] -0.11 [-0.45, 0.23] 0.21 [-0.19, 0.63] \nCondition x Self-\nparent neural \ndissimilarity (L) \n-0.25 [-0.84, 0.35] -0.30 [-0.92, 0.30] -0.12 [-0.70, 0.45] 0.06 [-0.49, 0.66] \nCondition x Self-\nfriend neural \ndissimilarity (L) \n-0.00 [-0.64, 0.66] -0.26 [-0.93, 0.41] -0.15 [-0.77, 0.45] -0.11 [-0.76, 0.50] \nCondition x Self-\nparent neural \ndissimilarity (R) \n-0.32 [-0.94, 0.31] -0.06 [-0.71, 0.53] 0.37 [-0.21, 1.00] -0.54 [-1.15, 0.04] \nCondition x Self-\nfriend neural \ndissimilarity (R) \n-0.35 [-0.96, 0.27] 0.70 [0.10, 1.33] 0.02 [-0.54, 0.61] 0.07 [-0.49, 0.66] \nCondition x \nLinguistic \npreference score \n-0.57 [-1.16, 0.00] -0.47 [-1.08, 0.16] 0.30 [-0.23, 0.84] -0.01 [-0.60, 0.60] \nSelf-parent neural \ndissimilarity (L) x \nLinguistic \npreference score \n0.36 [-0.12, 0.87] 0.08 [-0.55, 0.68] 0.25 [-0.19, 0.66] 0.28 [-0.18, 0.75] \nSelf-friend neural \ndissimilarity (L) x \nLinguistic \npreference score \n-0.47 [-1.02, 0.12] 0.51 [-0.04, 1.09] -0.40 [-0.87, 0.07] 0.31 [-0.07, 0.71] \nSelf-parent neural -0.12 [-0.60, 0.40] -0.02 [-0.52, 0.51] -0.11 [-0.54, 0.31] 0.07 [-0.33, 0.48] \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint \n\n 68 \ndissimilarity (R) x \nLinguistic \npreference score \nSelf-friend neural \ndissimilarity (R) x \nLinguistic \npreference score \n0.34 [-0.18, 0.94] 0.21 [-0.34, 0.76] 0.15 [-0.32, 0.63] 0.20 [-0.23, 0.61] \nCondition x Self-\nparent neural \ndissimilarity (L) x \nLinguistic \npreference score \n0.06 [-1.00, 0.35] -0.16 [-0.88, 0.53] -0.04 [-0.67, 0.61] 0.15 [-0.54, 0.84] \nCondition x Self-\nfriend neural \ndissimilarity (L) x \nLinguistic \npreference score \n-0.02 [-1.29, 0.24] -0.62 [-1.27, 0.00] 0.53 [-0.20, 1.24] -0.35 [-0.92, 0.23] \nCondition x Self-\nparent neural \ndissimilarity (R) x \nLinguistic \npreference score \n-0.34 [-0.63, 0.72] 0.25 [-0.34, 0.88] -0.14 [-0.77, 0.53] -0.48 [-1.08, 0.09] \nCondition x Self-\nfriend neural \ndissimilarity (R) x \nLinguistic \npreference score \n-0.53 [-0.79, 0.73] -0.33 [-0.98, 0.33] 0.03 [-0.68, 0.76] -0.16 [-0.79, 0.46] \n \nNote. Coefficients are on a logit scale and reflect means from a posterior distribution. ‘Condition’ \nindicates when parent or friend outcomes were associated with the discounting or delay options \nand encodes social decision preferences (positive values indicate a parent-over-friend \npreference, negative values indicate a friend-over-parent preference). Reward ratio reflects the \ndivision of the non-discounting option over the discounting option. Self-parent and self-friend \nneural dissimilarities refer to the distance between multivoxel neural patterns for oneself and a \nparent or friend, extracted from the brain region listed in the column heading. Values in brackets \nreflect 89% highest density credible intervals drawn around the posterior distribution. \nInteractions reflect associations between neural dissimilarities, linguistic preferences, and/or \nsocial decision preferences. Coefficients where the HDI partially overlaps with ROPE (modest or \nmoderate evidence depending on whether the interval contained 0) are italicized; coefficients \nwhere the HDI does not overlap with ROPE (strong evidence) are bolded and italicized. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted August 16, 2025. ; https://doi.org/10.1101/2024.07.16.603808doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}