Paper deep dive

From Blueprint to Reality: Modeling and Applying Putnam's Social Capital Theory with LLM-based Multi-agent Simulations

Shiyi Ling, Zhi Zheng, Hui Zheng, Wenjun Xue, Feng Ye, Tong Xu

Year: 2026Venue: arXiv preprintArea: cs.CLType: PreprintEmbeddings: 95

Intelligence

Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 94%

Last extracted: 7/8/2026, 4:41:37 AM

Summary

The paper introduces SOCASIM, an LLM-based multi-agent simulation framework designed to model and apply Putnam's Social Capital Theory. It integrates dynamic social network evolution, trust adaptation, and norm propagation to simulate collective action. The framework bridges social science and computer science by reproducing macro-level theoretical patterns, demonstrating strong human-agent alignment, and enabling micro-level causal pathway tracing through counterfactual interventions. It is applied to analyze adaptation challenges in smart elderly care, highlighting trust as a key lever for improving technology adoption among older adults.

Entities (8)

Putnam's Social Capital Theory · theory · 99%SOCASIM · framework · 98%Robert D. Putnam · person · 97%Norms · concept · 95%Social Network · concept · 95%Trust · concept · 95%Smart Elderly Care · application-domain · 92%Socioeconomic Status (SES) · concept · 90%

Relation Signals (8)

Putnam's Social Capital Theory → comprises → Social Network

confidence 95% · social capital can be understood through three interrelated dimensions: social network, which provides channels for social interaction and coordination

Putnam's Social Capital Theory → comprises → Trust

confidence 95% · trust, which reduces uncertainty and lowers the perceived cost of collective response

Putnam's Social Capital Theory → comprises → Norms

confidence 95% · norms, which encourage reciprocity, mutual support, and sustained participation within the group.

SOCASIM → implements → Putnam's Social Capital Theory

confidence 95% · SOCASIM, a dynamic and theory-grounded LLM-based multi-agent simulation framework designed to model and apply Putnam’s Social Capital Theory

SOCASIM → appliedto → Smart Elderly Care

confidence 92% · apply the three dimensions to analyze adaptation challenges in smart elderly care, where technology adoption hinges on social networks, trust, and norms.

Social Network → influences → Trust

confidence 90% · social network shapes interaction opportunities, trust guides expectations about others’ actions

Trust → influences → Norms

confidence 90% · norms regulate behavior over time... trust accumulates through interaction

→ →

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Putnam's Social Capital Theory is a foundational framework for collective action and community prosperity. However, traditional empirical methods face practical limits on control and replication. Meanwhile, LLM-based social simulations are typically behavior-driven and lack theory-aligned environments for modeling Putnam's core propositions. To address these gaps, we introduce SocaSim, an LLM-based multi-agent simulation framework to study Putnam's Social Capital Theory from theoretical blueprint to simulated reality. Specifically, we build an environment integrating social network evolution, trust dynamics, and norm propagation, where agents engage in repeated collective-action experiments, and then apply the three dimensions to analyze adaptation challenges in smart elderly care. Our simulations reproduce Putnam's macro-level patterns and exhibit strong human-agent alignment at the group level. Unlike traditional methods, SocaSim traces micro-level causal pathways of social network, trust, and norms via round-by-round simulations and counterfactual interventions, enabling process-level interpretability. Taken together, these capabilities establish a research paradigm that leverages LLM agents to bridge social science and computer science.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

93,555 characters extracted from source content.

Expand or collapse full text

From Blueprint to Reality: Modeling and Applying Putnam’s Social Capital Theory with LLM-based Multi-agent Simulations Shiyi Ling 1 , Zhi Zheng 1 , Hui Zheng 2 , Wenjun Xue 3 , Feng Ye 4 , Tong Xu 1 1 State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China 2 Anhui University 3 North Automatic Control Technology Institute 4 University of Science and Technology of China shiyi.ling, awaken215, yefengustc@mail.ustc.edu.cn, zhengzhi97, tongxu@ustc.edu.cn, huizheng@ahu.edu.cn, Abstract Putnam’s Social Capital Theory is a founda- tional framework for collective action and com- munity prosperity. However, traditional empir- ical methods face practical limits on control and replication. Meanwhile, LLM-based so- cial simulations are typically behavior-driven and lack theory-aligned environments for mod- eling Putnam’s core propositions. To address these gaps, we introduce SOCASIM, an LLM- based multi-agent simulation framework to study Putnam’s Social Capital Theory from the- oretical blueprint to simulated reality. Specifi- cally, we build an environment integrating so- cial network evolution, trust dynamics, and norm propagation, where agents engage in re- peated collective-action experiments, and then apply the three dimensions to analyze adapta- tion challenges in smart elderly care. Our sim- ulations reproduce Putnam’s macro-level pat- terns and exhibit strong human-agent alignment at the group level. Unlike traditional methods, SOCASIM traces micro-level causal pathways of social network, trust, and norms via round- by-round simulations and counterfactual inter- ventions, enabling process-level interpretability. Taken together, these capabilities establish a re- search paradigm that leverages LLM agents to bridge social science and computer science. 1 Introduction When an accident occurs, large anonymous crowds often freeze, each waiting for someone else to move first. In close-knit groups, help starts immediately because people are connected, trust one another, and expect to assist. This contrast shows how social network, trust, and norms shape a group’s ability to respond to common problems. Social Capital Theory, developed by Robert D. Putnam (Putnam et al., 1994; Putnam, 2000), captures these dynam- ics and explains why some communities overcome collective-action dilemmas while others do not. In social science, research on Putnam’s Social (a) Conventional Empirical Methods high resource demands, practical constraints on control and replication (b) Existing LLM-based Social Simulations lack of theory-aligned environments for core propositions Specific Task Scenario Large Language Models (c) Ours: SocaSim Trust Norms Social Network Theoretical Blueprint Smart Elderly Care Proposal Execution Communication Decision Practical Application Modeling Task Applying Task Human-Agent Alignment LLM-based Multi-agent Simulation Framework Macro Pattern Replication Scenario Simulation Counterfactual Intervention Figure 1: Comparison of existing research paradigms versus SOCASIM for Putnam’s Social Capital Theory. Capital Theory largely relies on conventional em- pirical methods that yield valuable insights yet have notable limits (Figure 1(a)). Quantitative methods such as large-scale surveys (Durante et al., 2025) and Structural Equation Modeling (SEM) (Hoa, 2021; Sumi et al., 2025) require substantial time and resources, and face practical constraints on con- trol and replication. In parallel, simulation-based approaches such as Agent-Based Modeling (ABM) (Shenk et al., 2019) allow controlled tests of as- sumptions, yet traditional rule-based agents rely on predefined and simple decision rules and strug- gle to capture the complexity of human reasoning, emotion, and context-dependent behavior. Recently, advances in large language models (LLMs) have enabled the simulation of human-like intelligence (Friha et al., 2024; Wang et al., 2025a), prompting growing interest in LLM-based agents for social simulation (Piao et al., 2026; Yang et al., 2025). However, existing approaches are largely behavior-driven rather than theory-driven (Fig- ure 1(b)): they typically test whether agents exhibit plausible or human-like behavior on specific tasks (Zhao et al., 2024; Jia et al., 2025; Xie et al., 2024), 1 arXiv:2607.06080v1 [cs.CL] 7 Jul 2026 but rarely construct environments aligned with core theoretical propositions. Furthermore, current frameworks seldom provide reproducible, control- lable settings for such theory-driven analysis, for example, by modeling links between network con- nectivity and collective outcomes or stepwise trust growth. Consequently, they still make it difficult to trace how trust accumulates, how norms are in- ternalized, or how individual–collective tensions arise, leading to limited procedural interpretability. To bridge this gap, we introduceSOCASIM, a dynamic and theory-grounded LLM-based multi- agent simulation framework designed to model and apply Putnam’s Social Capital Theory (Figure 1(c)). Specifically, we construct an agent society that in- tegrates mechanisms of dynamic social network evolution, trust adaptation, and norm diffusion, in which agents engage in repeated collective-action simulations. Each agent is endowed with basic demographic attributes and social capital endow- ments, and is capable of reasoning and decision- making by incorporating social relationships, in- teraction history, and situational context. Our agent society runs in rounds with two phases: Pro- posal, where agents make proposals; and Execu- tion, where they simultaneously decide whether to act based on trust, norms, and network ties. Based on this framework, we conduct two pro- gressive tasks. In the modeling task, we run multi- round collective-action experiments and human- alignment validation. The simulation reproduces the macro-level patterns predicted by Putnam’s the- ory, and a comparison with real older adults shows strong human-agent alignment in group-level de- cisions (Pearson r = 0.974). Subsequently, in the applying task, we move the theory from blueprint to reality by applying its three core dimensions to adaptation challenges in smart elderly care, where technology adoption hinges on social net- works, trust, and norms. In counterfactual simu- lations that raise low-SES agents’ initial trust, the adoption rate increases by 15.4% and decision con- tradictions decrease by 25.5%, suggesting that trust can serve as a key causal lever. Unlike traditional methods that are often limited to correlational con- clusions, SOCASIM uses LLM agents to reveal micro-level causal chains of how trust accumulates, norms are internalized, and decision contradictions emerge, offering process-level interpretability and practical guidance for policy design. Our contributions can be summarized as follows: •We propose SOCASIM, a new interdisciplinary research paradigm using LLM-based agents to model Putnam’s Social Capital Theory. •We develop a multi-agent framework that models the dynamics of social capital, including social net- work change, trust dynamics, and norms diffusion. •We conduct several experiments with macro-level replication, human alignment, and counterfactual intervention, validating SOCASIM’s explanatory power for theory and real-world issues. 2 Preliminary To clarify the theoretical basis, we briefly introduce Putnam’s Social Capital Theory (see Appendix A for more examples). This theory is commonly used to explain how groups deal with collective- action dilemmas, namely situations where a shared problem is hard to resolve because coordination is costly, trust is insufficient, or long-term participa- tion is hard to sustain. Putnam argues that social capital helps alleviate such dilemmas and improves a group’s ability to respond to common challenges. Putnam distinguishes between two primary forms of social capital. Bonding social capital develops within close-knit groups and strengthens internal support and solidarity. Bridging social capital connects different or loosely linked groups and facilitates information exchange and access to external resources. Conceptually, social capi- tal can be understood through three interrelated dimensions: social network, which provides chan- nels for social interaction and coordination; trust, which reduces uncertainty and lowers the perceived cost of collective response; and norms, which en- courage reciprocity, mutual support, and sustained participation within the group. These dimensions correspond directly to the core components of our simulation: social network shapes interaction opportunities, trust guides expectations about oth- ers’ actions, and norms regulate behavior over time. Based on these two primary forms and three core dimensions, we leverage LLM agents to model and apply Putnam’s Social Capital Theory. 3 Methods 3.1 Overview of SOCASIM Since Putnam’s Social Capital Theory involves complex social constructs (social network, trust, norms), prior general multi-agent frameworks can- not capture their dynamic coupling. We therefore propose SOCASIM, an LLM-based multi-agent 2 Agent Initialization Gender: Male, Female Health: Healthy, Unhealthy ...... Annual Income: 8000，10,000, 20,000 ...... Questions: Do you actively participate in public welfare activities? ...... General Social Survey // agent_1’ s profile "demographics": "age": 63, "gender": "Male", "health_status": "Unhealthy", "annual_income": 15000, ...... Agent Creation LLM-based Multi-agent Simulation Framework Social Structure Trait (SST) Socioeconomic Status (SES) Income Education Occupation Social Capital Type Preference Basic Demographic Attributes Multi-agent Simulation Environment Cooperate Defect Adopt Chat Reject Social Capital Bonding CapitalBridging Capital [Proposal-Initiation] Let's get a health check going for the neighborhood! I know I can count on you. [Proposal-Acceptance] Absolutely, I'm happy to! We've always worked well together. [Execution] Time to fulfill my commitment. My specific trust in Agent_5 is very high, and I have considerable bonding capital (6.11). I will cooperate fully with Agent_5's part of the work. [Execution] I value reliability. My general trust (3.25) and norms (2.66) support cooperation. More importantly, my strong relationship with Agent_0 demands that I honor our agreement. I will cooperate fully. Social Network Two-phase Round-based Decision Simulation Workflow Execution Proposal Yes Memory and Reflection Round t ≤ T ? Initial Social Network Generation Agents Initialization INPUT: Simulation Configuration Parameters Social Capital Update t=t+1 OUTPUT: Round-level Statistics, Agent State, Interaction History, Collective Action Outcomes No Figure 2: Overview of SOCASIM multi-agent architecture and round-by-round simulation workflow. framework to model, dynamically couple, and val- idate the theory’s core elements. Specifically, we construct a social simulation system consisting of Nagents, denoted asA = a 1 ,...,a N , each equipped with demographic and socioeconomic attributes and initial social-capital endowments, sit- uated in a controlled environmentE. The simula- tion runs forTdiscrete rounds. In each roundt (t = 1,...,T), agents initiate or respond to social behaviors based on their network connections, trust dispositions, and perceived norms. These interac- tions update three interrelated core dimensions: •Social networkG t : Connections dynamically form, strengthen, or weaken over time based on interaction frequency and quality. •TrustT i (t): Each agenta i evolves its level of specific trust toward specific others and general trust toward the broader group through interactions. •NormsN i (t): Each agent’s reciprocity norms comprise specific reciprocity for direct resource exchange and general reciprocity for mutual coop- eration without expecting immediate return. As shown in Figure 2, these abstract dimensions are concretely implemented by the agent architec- ture (i.e., SST, BDI, and SCM) and the LLM-driven updating mechanisms detailed in §3.2. 3.2 LLM-based Multi-agent Framework 3.2.1 Social Structure Trait (SST) Putnam’s Social Capital Theory holds that socioe- conomic position determines access to networks and resources, yet most multi-agent frameworks lack ways to model this. To address the limitation, we design a module that generates agents’ struc- tural social features through three key components: Basic Demographic Attributes. Each agent is as- signed demographic traits, including age, gender, education, income, health status, and occupation. These attributes are sampled from the 2023 Chi- nese General Social Survey (CGSS) data 1 elderly subsample, ensuring that the simulated population reflects real-world socioeconomic stratification. Socioeconomic Status (SES). A weighted compos- ite of income, education, and occupation, classify- ing agents into low, mid, and high SES categories. Social Capital Type Preference. Low-SES agents tend to rely on bonding capital, high-SES agents are more adept at leveraging bridging capital, while mid-SES agents exhibit a balanced preference. All these attributes are encoded into the agent’s natural language prompt, giving each agent a rich identity and ensuring behavior arises from socio- logically plausible profiles rather than static rules. 3.2.2 Belief-Desire-Intention (BDI) To capture the real-time decision-making of agents based on structural features from the SST mod- ule, we adopt the BDI framework (Bratman, 1987). Specifically, we formalize the agent’s context- sensitive cognitive decision process, with Act← BDI(f,B,D,I,S,M ), whereActdenotes the final action,fis imple- mented with an LLM, andB,D,I,S,Mdenote the Belief, Desire, Intention, SST structures, and SCM memory, respectively. Belief combines struc- tural cues such as SES and education with situa- tional cues from the current interaction round. De- sire represents motivations such as technological mastery and family recognition that are modulated by the agent’s structural traits. Intention integrates belief, desire, and memory to produce concrete be- havioral commitments; for technology adoption, agents evaluate their SES, prior experience, social influence, and family recommendation. 1 https://w.cnsda.org 3 3.2.3 Social Cognitive Memory (SCM) As agents learn through repeated interactions, they adapt their cognitive states by integrating new ex- periences with stored memories. Therefore, we design the SCM module to model this adaptive learning process. Specifically, we define its cogni- tive update process as (C new , M new )← SCM f, Act, O, C old , M old , whereC new denotes the updated cognitive state, M new represents the updated memory structure, Actis the action previously executed by the agent (i.e., the output of the BDI module),Ois the ob- served outcome,C old andM old are the prior states, and f is the same LLM used in the BDI module. Multi-layer Trust Initialization. Agents maintain specific and general trust. Specific trust is initial- ized based on social distance. Higher for the same group, moderate for adjacent groups, and lower for distant groups, with small random noise. During the simulation, trust levels are updated according to interaction outcomes. Positive interactions such as promise keeping or providing help increase trust, while negative interactions such as default or de- ception decrease trust, ensuring that trust values always evolve within a meaningful range. Norm Formation and Decay. Normative cogni- tion comprises general and specific dimensions. Initial norm levels correlate with SES. To capture the socially reinforced nature of norm maintenance, our system checks each agent’s reciprocity behav- ior every round. If no reciprocity is observed over ten consecutive rounds, norm decay is triggered. Memory and Reflection. We divide the agent’s memory into two distinct layers, interaction- history records specific interaction events, partic- ipants, and outcomes; cognitive-state stores the evolving trajectories of trust and norms. After each round, an LLM-driven reflection process analyzes these memories each round, adjusting the agent’s state and strategies for continuous adaptation. 3.3 Multi-agent Simulation Environment 3.3.1 Social Network and Capital Evolution We design the social network to co-evolve in- teractively with the accumulation of social cap- ital, forming a synergistic feedback loop during the simulation. Initially, ties are biased toward SES-similar agents. Network density, computed as Density = 2|E| |V|(|V|−1) ,where|E|is the number of social ties and|V|is the number of agents, mea- sures the overall connectedness of the society. Dur- ing the simulation, agents accumulate bonding so- cial capital from strong-tie interactions and bridg- ing social capital from weak-tie or cross-group in- teractions. The update mechanism of social capital takes into account an agent’s capital type prefer- ence. When the type of interaction an agent en- gages in matches its own social capital preference, the accumulation rate is significantly accelerated; otherwise, it accumulates at a baseline rate. 3.3.2 Two-phase Round-based Decision We extend turn-based simulations with a two-stage Proposal–Execution process, capturing the social progression of collective action from initiation to fulfillment. Each round follows this sequence. Proposal phase. Randomly selected initiators pro- pose collective actions or technology adoptions. When an agenta i proposes, each neighbora j re- sponds using LLM-based reasoning that incorpo- ratesa j ’s social attributes, especially its specific trust towarda i and its perception of general norms. This captures how social attributes shape an agent’s initial inclination toward collective decisions. Execution phase. After accepting a proposal, each participanta i privately decides whether to honor the commitment, again using LLM-based reason- ing that considers its current trust, norms, and so- cial context. The collective action succeeds if the proportion of committing participants exceeds a threshold τ (default τ = 0.5), with Success = ⊮ |Committers| |Participants| > τ . 4 Experimental Setup 4.1 Task Formulation To evaluate the capabilities of SOCASIM for Put- nam’s Social Capital Theory, we conduct two pro- gressive tasks that examine how agents embody the dynamic coupling of social capital in interactions: •Modeling. Can LLM agents reproduce the core dimensions of the theory and maintain consistency in group-level decision patterns with real older adults in scenario-based decisions? •Applying. How do the core dimensions of Put- nam’s Social Capital Theory affect the adaptation difficulties of elderly groups in smart elderly care? 4.2 Agent and Simulation Configuration We evaluate three LLMs to assess the generaliz- ability of our framework. Qwen2.5-14B-Instruct 4 (Qwen et al., 2025) serves as the default agent en- gine, while additional results for GPT-4 (OpenAI et al., 2024) and GLM-4 (Zeng et al., 2024) are provided in Appendix B (modeling task) & D.3 (applying task). We set temperature = 0.7, top-p = 0.9. All key experiments are repeated five times; re- sults are reported as means and standard errors. We generate 200 agents as detailed in Appendix D.1 (data construction) & D.2 (demographic profiles and behavioral analysis). For modeling task, we randomly sample 20 agents stratified by SES into low-, mid-, and high-SES groups for multi-round collective action experiments, each running for 25 rounds. For applying task, we use all 200 agents for smart elderly care simulations, also over 25 rounds. 4.3 Human Benchmarking We conduct parallel experiments with 20 real older adults (age≥ 60) recruited online for the modeling task. Quota sampling matches the SES proportions of the 20 CGSS subsample. Each participant com- pleted the same eight scenario-based decisions as the LLM agents (see Appendix C.1 for details). Prior to task completion, participants provided in- formed consent and completed self-report inven- tories on demographics, social trust, and norms, aligned with agent initialization. Results are pre- sented in §5.3, further analyzed in Appendix C.2. 5 Modeling Results: Theory Replication In this section, we examine whether SOCASIM can reproduce Putnam’s core mechanisms and align with human decisions. We present results from three complementary perspectives: macro- level pattern replication, ablation study, and human-agent alignment validation. 5.1Macro-level Pattern Replication (Figure 3) To examine whether the simulated macro-level pat- terns align with Putnam’s theoretical predictions, we analyze the core dynamics across four aspects, with results shown in Figure 3. Collective action outcomes are evaluated by the cooperation success rate, defined as the proportion of successful col- lective actions per round. In each round, agents interact within the social network and update their trust and norms based on historical experience. • Social Networks and Collective Action. Figure 3(a) demonstrates co-evolution between network density, cooperation rate, and collective action success. Over 25 rounds, network density in- creased from 0.158 to 0.342, while cooperation rate (a) Social Networks and Collective Action (b) Trust Evolution across SES Groups (c) Norms and Cooperation Durability (d) Bonding and Bridging Social Capital Figure 3: Macro-level pattern replication of (a) social network dimension, (b) trust dimension, (c) norms dimension, and (d) social capital accumulation dy- namics in Putnam’s Social Capital Theory. and success rate rose to 0.795 and 0.762, respec- tively. Correlation analysis shows strong positive associations between network density and both col- lective action success (r ≈ 0.976,p < 0.001) and cooperation rate (r ≈ 0.993,p < 0.001). Results are stable across five runs with low variance. ✧Observation: Initial network ties facilitate collective action, and the resulting coopera- tive success subsequently reinforces those ties, forming a self-reinforcing dynamic. • Trust Evolution across SES Groups. Figure 3(b) shows the evolution of trust across different SES groups and trust types over 25 rounds. Trust within the same SES group consistently re- mains the highest, rising from 3.98 to 4.48, while trust toward distant SES groups steadily increases from 2.02 to 2.78. Both specific and general trust grow in parallel, from 2.82 to 3.68 and from 2.46 to 3.47, respectively. These upward trends are con- sistent across runs with narrow variance. ✧Observation: Trust accumulates through interaction; trust toward distant SES groups grows more in absolute terms, partially bridg- ing the gap while the hierarchy persists. • Norms and Cooperation Durability. Figure 3(c) illustrates the evolution of specific and general reciprocity norms over 25 rounds. Spe- cific reciprocity norms rise from 2.92 to 3.78, and general reciprocity from 2.61 to 3.58, maintaining a parallel growth trajectory. The simulation produces 15 sustained cooperation pairs. 5 Network DensityCooperation Rate Collective Action RateHelp Rate 0.0 0.2 0.4 0.6 0.8 Score / Rate 0.335 0.768 0.745 0.782 0.285 0.223 0.682 0.702 0.298 0.528 0.715 0.728 0.342 0.785 0.762 0.798 (a) Social Network Dimension w/o SST w/o BDI w/o SCM SocaSim Specific Trust General Trust Group Trust (Same SES) Group Trust (Distant SES) 0 1 2 3 4 5 Trust Score (1-5) 3.62 3.52 3.85 3.12 1.39 4.02 2.45 3.35 3.10 3.58 2.31 3.68 3.47 4.48 2.78 (b) Trust Dimension w/o SST w/o BDI w/o SCM SocaSim Specific ReciprocityGeneral Reciprocity Sustained Cooperation Pairs 0.0 2.5 5.0 7.5 10.0 12.5 15.0 Norm Strength (1-5) / Pairs 3.72 3.62 13 3.31 3.12 7 3.12 2.98 4 3.78 3.58 15 (c) Norms Dimension w/o SST w/o BDI w/o SCM SocaSim Bonding CapitalBridging CapitalTotal Social Capital 0 2 4 6 8 10 12 Capital Score 5.95 5.85 11.80 5.42 4.28 9.70 4.95 4.05 9.00 6.95 4.92 11.87 (d) Social Capital Accumulation w/o SST w/o BDI w/o SCM SocaSim Figure 4: Ablation study of SocaSim with SST, BDI, and SCM modules removed respectively. tolP rettacS tnemngilA tnegA-namuH )a( tnesbAwoLwoL tneserPwoLwoL tnesbAhgiHwoL tneserPhgiHwoL tnesbAwoLhgiH tneserPwoLhgiH tnesbAhgiHhgiH tneserPhgiHhgiH smroNtsurT krowteN laicoS noitidnoC s s e n g n i l l i W t n e g A ssengnilliW namuH 479.0=r 58.0 06.0 05.0 04.0 03.0 51.0 08.056.0 06.055.005.0 54.0 04.0 02.0 (b) Effect Size Comparison Forest Plot Agent Human 1.035** 1.405** 0.895** 0.805** 1.045** 0.635** Norms Trust Social Network tneiciffeoC noissergeR citsigoL 1.751.501.251.000.750.500.250.00 0.000.250.500.751.001.251.501.75 Logistic Regression Coefficient Social Network Trust Norms 0.635** 1.045** 0.805** 0.895** 1.405** 1.035** Human Agent (b) Effect Size Comparison Forest Plot Low-SESMid-SES High-SES Socioeconomic Status Social Network Trust Norms Social Capital Dimension 0.0%12.5%66.7% 16.7%25.0%83.3% 0.0%12.5%50.0% LLM Agent Simulation Low-SESMid-SES High-SES Socioeconomic Status Social Network Trust Norms 16.7%25.0%50.0% 33.3%37.5%66.7% 16.7%25.0%50.0% Human Experiment (c) Adoption Switching Rates by SES Heatmap Figure 5: Human-agent alignment: (a) scatter correlation, (b) coefficient comparison, (c) SES-switching heatmap. ✧Observation: Repeated interaction over time enables agents to internalize reciprocity norms, and once established, these norms sus- tain long-term cooperation within the group even in the absence of immediate returns. • Bonding and Bridging Social Capital. Figure 3(d) compares the growth of bonding and bridging social capital over 25 rounds. Bond- ing capital increases from 2.78 to 6.95, whereas bridging capital grows from 1.91 to 4.92, result- ing in a final gap of 2.03. We further quantify cross-SES connections through the economic con- nectivity (EC) index, defined as the proportion of high-SES friends in a low-SES agent’s network, which rises steadily across rounds. ✧Observation: Bonding capital accumulates faster and reaches noticeably higher levels than bridging capital, while the steady rise in the EC index over time further suggests a gradual formation of cross-SES weak ties. 5.2 Ablation Study (Figure 4) To investigate how each component contributes to the simulation performance of SocaSim, we con- duct an ablation study to validate the necessity of each module, with results shown in Figure 4. • w/o SST. Overall averages change only slightly, but the structural patterns are largely removed. In particular, the distinction between same-SES and distant groups collapses, with bonding/bridging capital shifting from 6.95/4.92 to 5.95/5.85. ✧Observation: SST’s core function is to shape inter-group structural differences, not to raise overall averages. The slight metric gains under w/o SST are due to the flattening of group disparities, not substantive improvement. •w/o BDI. Cooperation rate drops by 71.6%, and general trust falls from 3.47 to 1.39. ✧Observation: BDI is the decision-making core. Without it, agents cannot understand in- teraction contexts, and cooperation collapses. •w/o SCM. Sustained cooperative pairs decrease by 73.3%, and specific reciprocity drops by 17.5%. ✧Observation: SCM is the learning founda- tion. Without it, norms fail to internalize, and long-term cooperation cannot be sustained. 5.3 Human Validation (Figure 5) To validate group-level consistency between LLM agents and real older adults, we compare their adop- tion willingness across eight scenarios that vary social network density (high/low), trust level (high- /low), and reciprocity norm (present/absent). The 20 real older adults were recruited as described in § 4.3. The results are summarized in Figure 5. •Overall alignment (Figure 5(a)). Human and agent willingness rates are highly correlated (Pear- son r = 0.974), with points tightly clustered around the identity line. Agents show slightly more ex- treme responses, higher under favorable conditions and lower under unfavorable ones. 6 (a) Technology Adoption and SES(b) Psychological Pressure and Anxiety(c) Decision Contradiction and Dynamics (a) Technology Adoption and SES(b) Psychological Pressure and Anxiety(c) Decision Contradiction and Dynamics Figure 6: Experimental simulation of social capital in smart elderly care technology adoption. •Effect comparison (Figure 5(b)). All coeffi- cients are positive and significant (p < 0.01), and the ranking is identical for both groups, trust has the strongest effect, followed by reciprocity norm, and then social network density. Agents consis- tently produce larger coefficients, reflecting their cleaner causal reasoning that is less affected by noise such as emotion, fatigue, or social desirabil- ity, and thus showing more extreme responses. •Turnaround by SES (Figure 5(c)). For each dimension, we measure the switch from non- adoption to adoption when it improves, while keep- ing the other two dimensions fixed, and stratify by SES. In both groups, switching rates rise with SES and trust dominates. For low-SES agents, the trust switching rate is 16.7% compared to 33.3% for hu- mans, while the rates for social network and norms are both 0% versus 16.7% for humans. For high- SES agents, the trust switching rate reaches 83.3% compared to 66.7% for humans. These results indi- cate that LLM agents may respond more strongly than humans to SES-based structural inequality. ✧Observation: LLM agents closely repro- duce the decision patterns of real older adults at the group level. Their responses provide clearer signals for identifying causal mechanisms of social capital, supporting external validity of our later applications in smart elderly care. 6 Applying Results: Elderly Care Study In this section, we investigate how the three core dimensions of Putnam’s Social Capital Theory af- fect the adaptation challenges of elderly groups in smart elderly care. We present results from two complementary perspectives: scenario simulation and counterfactual intervention. 6.1 Scenario Simulation (Figure 6) We model the adoption of a smart elderly care plat- form as a binary choice using all 200 agents. The simulation tracks three multidimensional metrics: technology adoption rate, defined as the proportion of agents that adopt the platform; psychological dis- tress, measured via psychological pressure value and anxiety level; and decision contradiction, re- flected by the frequency of adoption status switches. The results are shown in Figure 6. • Technology Adoption and SES. Figure 6(a) shows adoption rates across SES groups over 25 rounds. The high-SES group rises steadily from 12.2% to 84.5% with minimal fluctu- ation. The low-SES group increases slowly from 2.3% to 56.8%, with greater variability and wider confidence bands. The mid-SES group stabilizes in the low-70% range. All groups grow rapidly in the first 10 rounds, then slow with diminishing returns. ✧Observation: Technology adoption rates are positively associated with SES, with high-SES agents reaching higher levels and showing more stable trajectories than low-SES agents. • Psychological Pressure and Anxiety. Figure 6(b) illustrates the evolution of psycholog- ical pressure and anxiety. Both metrics rise steeply in early rounds (pressure from 0.10 to 0.27, anxiety from 0.12 to 0.26) and then stabilize. Pressure re- mains slightly higher than anxiety throughout, and variance widens modestly over time. ✧Observation: Psychological pressure and anxiety increase rapidly during early adoption rounds, clearly indicating that technology adop- tion induces significant stress that later stabi- lizes but leaves an elevated baseline level. • Decision Contradiction and Dynamics. Figure 6(c) presents the evolution of cumulative decision contradictions. The metric rises steadily from near 0 to about 50 by round 25. Its growth rate is consistent without clear inflection points. Variance across five runs diverges after round 10. 7 ✧Observation: Decision contradictions accu- mulate linearly over time, with increasing vari- ance across runs, suggesting persistent decision uncertainty and growing individual differences. 6.2 Counterfactual Intervention (Table 1) To enhance causal interpretability and simulate a policy intervention, we design a counterfactual experiment. Since trust is the strongest dimen- sion (Figure 5(b)), we increase the initial trust of low-SES agents by 1.0 (capped at 5.0) while keep- ing all other parameters unchanged, mimicking a targeted policy for disadvantaged groups. SES Category: Low, Age: 67, Gender: Female, Health Status: Fair, Education: No Education, Tech Experience: No Agent Profile After a policy intervention that raises her specific trust to 3.54 (now above the average of the low‑SES group, which is around 3.0), Agent_187 becomes more open to the platform. She no longer views it as threatening; instead, she acknowledges its potential benefits. The positive examples of her neighbor’s successful adoption and her family’s persistent encouragement now carry more weight. She reassesses the risks, concluding that the benefits, such as emergency health monitoring, outweigh the small chances of error or privacy breach. Consequently, she decides to adopt the platform. Her anxiety drops to 0.18, and she experiences no further decision reversals in the next five rounds, demonstrating stable and confident choice. Post-intervention •Boosted Initial Trust (Specific Trust = 3.54) •Open to platform information •Neighbor's adoption & family advice perceived positively •Lower perceived risks → willingness increases •Decision: ADOPT •Anxiety level: 0.18 •No decision contradictions (next 5 rounds) Agent_187, a 67-year-old female with low SES, no education, and no prior technology experience, initially holds a specific trust of only 2.54, which is below the average level of the low-SES group (approximately 2.8). She perceives the smart elderly care platform with deep suspicion, doubting its safety and usability. Although her high-SES neighbor has already adopted the platform and her family strongly recommends it, she remains unconvinced. She weighs the risks carefully, fearing both operational errors (“What if I press the wrong button?”) and privacy leakage (“They might steal my personal information.”). As a result, she decides to reject the platform. Her anxiety level is recorded at 0.27, and she has changed her decision three times in the last four rounds, indicating high decision contradiction. Pre-intervention •Low Initial Trust (Specific Trust = 2.54) •Perceives platform with suspicion •Neighbor (High-SES) adopted & Family recommends •Weighs risks: operation error & privacy concern •Decision: REJECT •Anxiety level: 0.27 •Decision contradictions: 3 reversals (last 4 rounds) Figure 7: Case study: low-SES trust intervention. MetricPre-interventionPost-interventionRelative change Technology Adoption Rate0.732±0.0280.845±0.024+15.4% Psychological Pressure0.268±0.0320.215±0.028-19.8% Anxiety Level0.255±0.0280.198±0.022-22.4% Decision Contradictions51±838±6-25.5% Table 1: Overall performance of 200 agents at round 25 under factual and counterfactual conditions. We illustrate the effect using a case study (Fig- ure 7). Before intervention, low trust causes hesita- tion, repeated decision contradictions, and eventual rejection. After trust is raised, the agent adopts stably, with anxiety and contradictions dropping sharply. This demonstrates trust’s causal role in stability. Table 1 compares the aggregate outcomes at round 25. After the intervention, technology adoption rises by 15.4%, pressure and anxiety fall by 19.8% and 22.4%, and decision contradictions decrease by 25.5%. These results indicate that boosting initial trust of low-SES groups reduces the digital divide and improves decision stability. 7 Discussion Our findings carry several important implications for policy, highlight the unique strengths and limi- tations of using LLM agents for social simulation, and open up new avenues for interdisciplinary re- search. For a more detailed discussion of these points, we refer the readers to Appendix E. 8 Related Work Putnam’s Social Capital Theory. Putnam’s So- cial Capital Theory defines social capital as rooted in social network, trust, and norms (Putnam et al., 1994; Putnam, 2000). It serves as a core paradigm for explaining civic engagement, government per- formance, and democratic functioning (Tzanakis, 2013; Dodd et al., 2015), and is widely applied to examine regional development (Annamalah et al., 2023), public health (Carpiano, 2006), and educa- tion attainment (Goddard, 2003; Careemdeen et al., 2021) as outcomes of resource accumulation within community structures. Recent studies further ex- plore its role in crises (Chawa et al., 2024) and sustainable development (Sumi et al., 2025). How- ever, conventional empirical methods face practical constraints on experimental control and replication. LLM-based Multi-agent Social Simulation. The rise of LLMs has enabled human-like agents capa- ble of reasoning and social interaction (Wang et al., 2026; Sudhakar et al., 2025; Wang et al., 2025b). These agents perform complex social tasks (Zhang et al., 2025b; Xu et al., 2025), leading to growing use of LLM-based social simulation across sce- narios (Piao et al., 2026; Yang et al., 2025; Zhang et al., 2025a). For example, generative agents (Piao et al., 2026) simulate intricate social behaviors; ProSIM (Zhou et al., 2026) models prosocial be- havior; and SOTOPIA-Ω(Zhang et al., 2025a) in- tegrates human negotiation strategies. However, existing frameworks have not yet provided a repro- ducible and controllable environment for systemat- ically instantiating Putnam’s Social Capital Theory, nor have they offered process-level interpretability of how trust accumulates or norms are internalized. Our work aims to address this gap. 9 Conclusion We presented SOCASIM, a novel LLM-based multi-agent simulation framework to model and apply Putnam’s Social Capital Theory. By integrat- ing social network evolution, trust dynamics, and norm propagation into a unified environment, we analyzed collective-action behaviors and extended the framework to smart elderly care. Our results demonstrate a consistent alignment between agent behaviors and human social dynamics, revealing how social capital shapes group coordination and technology adoption. Moreover, SOCASIM offers a scalable, reproducible approach to investigate social mechanisms across social science domains. 8 Limitations In this section, we discuss the limitations of our work as follow: • Our current framework models the dynamic evolution of social capital primarily through text-based interactions, whereas in reality, trust building and norm propagation often rely on multimodal signals such as speech and im- ages. Incorporating multimodal information is an important direction to enhance simula- tion realism, but it also brings challenges in data acquisition, cross-modal alignment, and ethics. Therefore, exploring multimodal so- cial interaction mechanisms is a highly valu- able research direction for future work. •Due to recruitment costs and resource con- straints, the number of participants in our hu- man alignment experiment is limited, which may affect the statistical generalizability of the findings. Future work should increase the sample size and include more diverse popu- lations to more robustly analyze alignment between simulated and real behaviors across different demographic dimensions. •Our study is based on CGSS data and has not incorporated comparative analyses across diverse cultural contexts. The accumulation and dynamics of social capital may be shaped by cultural values, institutional settings, and related factors. Cross-cultural comparisons are thus a promising direction to evaluate both generalizability and context-specificity. Ethics Statement As the use of LLMs for social simulation grows, it is crucial to consider the ethical implications of deploying such systems to model social capital dynamics and elderly care behaviors. While this work explores the potential of LLM-based multi- agent simulations to study Putnam’s Social Capital Theory, all experiments are conducted solely within controlled research settings and remain an early- stage theoretical exploration. The SOCASIM framework and all simulations are designed for scientific theory validation and methodological development rather than for direct real-world decision making, policy formulation, or replacing the complexity of human social systems. We strongly emphasize that any potential applica- tion of our findings to smart elderly care policy, community governance, or service design must in- volve human expert oversight, ethical review, and careful consideration of fairness to mitigate risks arising from model simplifications or data biases. All demographic data used for agent initializa- tion are drawn from the 2023 CGSS elderly sub- sample, which is released as publicly available ag- gregate statistics and contains no personally iden- tifiable information. The construction of synthetic agents does not involve direct simulation or identity mapping of real individuals. For our human-agent alignment validation, 20 real older volunteers par- ticipated with informed consent. The study con- sisted solely of anonymous questionnaires and scenario-based decision tasks, with no interven- tional or deceptive procedures. Participants’ per- sonal information was used only for SES stratifi- cation and anonymized analysis, and they received reasonable compensation in accordance with lo- cal ethical guidelines. The research protocol was approved by the institutional ethics board. Looking forward, we will continue to focus on fairness, transparency, and accountability in agent- based social simulation. We will actively expand cross-cultural and multimodal simulation capabili- ties, and develop human-in-the-loop experimental paradigms. We remain committed to advancing this field toward responsible, inclusive, and socially beneficial directions. References Sanmugam Annamalah, Pradeep Paraman, Selim Ahmed, Rajinder Dass, Ilham Sentosa, Thillai Raja Pertheban, Farid Shamsudin, Baharudin Kadir, Kalisri Logeswaran Aravindan, Murali Raman, Wong Chee Hoo, and Prakash Singh. 2023. The role of open innovation and a normalizing mechanism of social capital in the tourism industry. Journal of Open Innovation: Technology, Market, and Complex- ity, 9(2):100056. Michael E. Bratman. 1987. Intention, Plans and Prac- tical Reason. Harvard University Press, Cambridge, MA. Jalal Careemdeen, Mohd Awang, Abdul Ahmad, and Shakila Dahalan. 2021. Instrument development for socio-educational support and socio-educational par- ticipation towards educational well-being. Journal of Educational and Social Research, 11:43. Richard M. Carpiano. 2006. Toward a neighborhood resource-based theory of social capital for health: 9 Can bourdieu and sociology help? Social Science & Medicine, 62(1):165–175. Anif Chawa, Moch Putra, and Anik Susanti. 2024. Bonding, bridging and linking capital: How social capital was mobilized during the ppkm micro in malang raya. Jurnal Ilmu Sosial, 22(2):67–88. Melissa D. Dodd, John Brummette, and Vincent Hazle- ton. 2015. A social capital approach: An examination of putnam’s civic engagement and public relations roles. Public Relations Review, 41(4):472–479. Ruben Durante, Nicola Mastrorocco, Luigi Minale, and James M Snyder. 2025. Unpacking social capital. The Economic Journal, 135(667):773–807. Mohammad Mahdi Fakhimi, Adriana Hughes, and Al- lison M. Gustavson. 2025. Evaluating smart home usability and accessibility in early detection and in- tervention of mental health challenges among older adults: A narrative review and framework. Journal of Ageing and Longevity, 5(1):3. Othmane Friha, Mohamed Amine Ferrag, Burak Kantarci, Burak Cakmak, Arda Ozgun, and Nassira Ghoualmi-Zine. 2024. LLM-based edge intelligence: A comprehensive survey on architectures, applica- tions, security and trustworthiness. IEEE Open Jour- nal of the Communications Society, 5:5799–5856. Roger D. Goddard. 2003. Relational networks, social trust, and norms: A social capital perspective on students’ chances of academic success. Educational Evaluation and Policy Analysis, 25(1):59–74. Pham Tan Hoa. 2021. Indirect impact of employees’ social capital on their income at dong thap muoi, long an province. HCMCOUJS-Economics and Business Administration, 11(2):3–17. Jingru Jia, Zehua Yuan, Junhao Pan, Paul E. McNa- mara, and Deming Chen. 2025. LLM strategic rea- soning: Agentic study through behavioral game the- ory. Preprint, arXiv:2502.20432. Izabela Jonek-Kowalska and Maciej Wolny. 2025. Age sustainability in smart city: Seniors as urban stake- holders in the light of literature studies. Sustainabil- ity, 17(14):6333. Rada Mihalcea, Laura Biester, Ryan L. Boyd, Zhi- jing Jin, Veronica Perez-Rosas, Steven Wilson, and James W. Pennebaker. 2024. How developments in natural language processing help us in understand- ing human behaviour. Nature Human Behaviour, 8(10):1877–1889. Piotr Mikiewicz. 2021. Social capital and education – an attempt to synthesize conceptualization arising from various theoretical origins. Cogent Education, 8(1):1907956. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Ale- man, Diogo Almeida, Janko Altenschmidt, Sam Alt- man, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haim- ing Bao, Mohammad Bavarian, Jeff Belgum, and 262 others. 2024. GPT-4 technical report. Preprint, arXiv:2303.08774. Elinor Ostrom. 2000. Collective action and the evolu- tion of social norms. Journal of Economic Perspec- tives, 14(3):137–158. Jinghua Piao, Yuwei Yan, Jun Zhang, Nian Li, Junbo Yan, Xiaochong Lan, Zhihong Lu, Zhiheng Zheng, Jing Yi Wang, Di Zhou, Chen Gao, Fengli Xu, Fang Zhang, Ke Rong, Jun Su, and Yong Li. 2026. AgentSociety: Large-scale simulation of LLM-driven generative agents advances understanding of human behaviors and society. Preprint, arXiv:2502.08691. Robert D. Putnam. 2000. Bowling Alone: The Col- lapse and Revival of American Community. Simon & Schuster, New York, NY. Robert D. Putnam, Robert Leonardi, and Raffaella Y. Nonetti. 1994. Making Democracy Work: Civic Tra- ditions in Modern Italy. Princeton University Press, Princeton, NJ. Qwen, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, and 24 others. 2025. Qwen2.5 technical report. Preprint, arXiv:2412.15115. Linda Shenk, Caroline Krejci, and Ulrike Passe. 2019. Agents of change—together: Using agent-based mod- els to inspire social capital building for resilient com- munities. Community Development, 50(2):256–272. Arjun V. Sudhakar, Hadi Nekoei, Mathieu Reymond, Miao Liu, Janarthanan Rajendran, and Sarath Chan- dar. 2025. A generalist hanabi agent. In Inter- national Conference on Learning Representations, pages 38363–38383. Ayesha Akter Sumi, Jaffary Bin Awang, Rasheedul Haque, Sayeeduzzafar Qazi, Nur Farhana Abdul Rah- man, and Abdul Rahman Bin S. Senathirajah. 2025. Investigating the function of religion and social cap- ital in shaping sustainable social development. Dis- cover Sustainability, 6(1):748. Xiaokun Sun. 2024. Analysis of china’s smart elderly care service policy: Based on the three-dimensional framework. SAGE Open, 14(1). Kim Mannemar Sønderskov and Peter Thisted Dinesen. 2016. Trusting the state, trusting each other? the effect of institutional trust on social trust. Political Behavior, 38(1):179–202. Do Hong Thang. 2025. The impact of social capital on workplace engagement, productivity, and organiza- tional success. International Journal of Humanities Social Sciences and Education, 12(2):32–39. 10 Michael Tzanakis. 2013. Social capital in Bourdieu’s, Coleman’s and Putnam’s theory: Empirical evidence and emergent measurement issues. Educate, 13(2):2– 23. Lei Wang, Jingsen Zhang, Hao Yang, Zhi-Yuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Hao Sun, Ruihua Song, Xin Zhao, Jun Xu, Zhicheng Dou, Jun Wang, and Ji-Rong Wen. 2025a. User behavior simulation with large language model-based agents. ACM Trans. Inf. Syst., 43(2):55. Xintao Wang, Heng Wang, Yifei Zhang, Xinfeng Yuan, Rui Xu, Jen tse Huang, Siyu Yuan, Haoran Guo, Jiangjie Chen, Shuchang Zhou, Wei Wang, and Yanghua Xiao. 2026. CoSER: A comprehensive liter- ary dataset and framework for training and evaluating llm role-playing and persona simulation. Preprint, arXiv:2502.09082. Zixiao Wang, Duzhen Zhang, Ishita Agarwal, Shen Gao, Le Song, and Xiuying Chen. 2025b. Beyond profile: From surface-level facts to deep persona simulation in LLMs. In Findings of the Association for Compu- tational Linguistics: ACL 2025, pages 21239–21257, Vienna, Austria. Association for Computational Lin- guistics. Chengxing Xie, Canyu Chen, Feiran Jia, Ziyu Ye, Shiyang Lai, Kai Shu, Jindong Gu, Adel Bibi, Ziniu Hu, David Jurgens, James Evans, Philip Torr, Bernard Ghanem, and Guohao Li. 2024. Can large language model agents simulate human trust behavior? In Advances in Neural Information Processing Systems, pages 15674–15729. Curran Associates, Inc. Yisen Xu, Feng Lin, Jinqiu Yang, Tse-Hsun, Chen, and Nikolaos Tsantalis. 2025. MANTRA: Enhancing automated method-level refactoring with contextual RAG and multi-agent LLM collaboration. Preprint, arXiv:2503.14340. Ziyi Yang, Zaibin Zhang, Zirui Zheng, Yuxian Jiang, Ziyue Gan, Zhiyu Wang, Zijian Ling, Jinsong Chen, Martz Ma, Bowen Dong, Prateek Gupta, Shuyue Hu, Zhenfei Yin, Guohao Li, Xu Jia, Lijun Wang, Bernard Ghanem, Huchuan Lu, Chaochao Lu, and 4 others. 2025. OASIS: Open agent social interac- tion simulations with one million agents. Preprint, arXiv:2411.11581. Team GLM: Aohan Zeng, Bin Xu, Bowen Wang, Chen- hui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang, Jingyu Sun, and 38 others. 2024. ChatGLM: A family of large language mod- els from GLM-130b to GLM-4 all tools. Preprint, arXiv:2406.12793. Wenyuan Zhang, Tianyun Liu, Mengxiao Song, Xi- aodong Li, and Tingwen Liu. 2025a. SOTOPIA-Ω: Dynamic strategy injection learning and social in- struction following evaluation for social agents. In Proceedings of the 63rd Annual Meeting of the As- sociation for Computational Linguistics (Volume 1: Long Papers), pages 24669–24697, Vienna, Austria. Association for Computational Linguistics. Zeyu Zhang, Jianxun Lian, Chen Ma, Yaning Qu, Ye Luo, Lei Wang, Rui Li, Xu Chen, Yankai Lin, Le Wu, Xing Xie, and Ji-Rong Wen. 2025b. Trend- Sim: Simulating trending topics in social media un- der poisoning attacks with LLM-based multi-agent system. In Findings of the Association for Computa- tional Linguistics: NAACL 2025, pages 2930–2949, Albuquerque, New Mexico. Association for Compu- tational Linguistics. Qinlin Zhao, Jindong Wang, Yixuan Zhang, Yiqiao Jin, Kaijie Zhu, Hao Chen, and Xing Xie. 2024. Com- peteAI: Understanding the competition dynamics of large language model-based agents. In Proceedings of the 41st International Conference on Machine Learning (ICML 2024), pages 61092–61107, Vienna, Austria. JMLR.org. Ran Zhou and Kyriaki Kaplanidou. 2023. The outcomes of social capital among event runners: quality of life considerations. International Journal of Sports Marketing and Sponsorship, 24(5):913–927. Yujia Zhou, Hexi Wang, Qingyao Ai, Zhen Wu, and Yiqun Liu. 2026. Investigating prosocial behav- ior theory in LLM agents under policy-induced in- equities. In Proceedings of the Fortieth AAAI Con- ference on Artificial Intelligence (AAAI 2026), pages 2254–2262. AAAI Press. 11 (a) Social Network Dimension Weak Connectivity Network Strong Connectivity Network Our plan could really use your skills. Want to team up on this? Sounds good! We’ve got some people who can help too. I’l get them on board. ...... Round 1 Round 2 Round 3 Round N Collective Action Success Rate 25% 35% 45% 95% Let's just do it ourselves —they're unreliable. Look, they're making plans behind our backs again. They clearly don't want anything to do with us. Collective Action Success Rate Round 1 Round 2 Round 3 Round N 65% 55% 45% 5% ...... Cooperation quickly achieved ✓ Low time cost ✓ Low communication cost ✓ (b) Trust Dimension High-Trust Scenario Low-Trust Scenario We've got a good team here! Let’s pull together on this! Alice is solid. Let’s get started right away! Alice BobCarol This feels like it’l work well. EmilyDavid I'm just not sure if I can really trust them... Better check if David is reliable first. I'l just watch how things play out for now. Cooperation delayed ✗ High time cost ✗ Multiple verifications needed ✗ Frank (c) Norms Dimension This project might not pay off for six months—is everyone still up for it? Absolutely! I trust our long-term collaboration will bring even greater rewards. I'm in too! Helping each other out is just how we do things around here. Strong Norm Context Weak Norm Context I’ve noticed that when everyone sticks to the mutual-help rule, the whole project goes more smoothly! Is anyone interested in this long-term project? It might be a while before we see any payoff. No immediate benefit? Then I'm out. Same here. I'l wait until there’s a clearer reward. Well, if no one wants to join, let’s just drop the project. Seems like helping each other isn’t really a thing here. I’l be more careful from now on. Figure 8: Illustration and examples of three core dimensions in Putnam’s Social Capital Theory. A Supplementary Preliminary Putnam’s Social Capital Theory (Putnam et al., 1994; Putnam, 2000) offers a theoretical blueprint of how social network, trust, and norms interact in self-reinforcing ways (Ostrom, 2000; Sønderskov and Dinesen, 2016), and it has influenced work in fields ranging from political science and public ad- ministration to economic development (Mikiewicz, 2021; Zhou and Kaplanidou, 2023; Thang, 2025). This section further explains these three dimen- sions, as illustrated in Figure 8, by showing how they jointly shape a group’s ability to overcome collective-action dilemmas. We also provide con- crete positive and negative examples for each di- mension, and highlight how the interplay among these dimensions affects coordination, responses to shared problems, and sustained participation. 1. Social Network Dimension (Figure 8(a)) Putnam highlights the importance of civic and as- sociational networks in enabling coordination and mutual support (Putnam et al., 1994). In this sense, the social network dimension concerns the exis- tence, density, and stability of social ties, which determine how easily information, resources, and assistance can circulate within a group. Positive example (strong connectivity network). In a community where people know each other well and interact frequently, a proposal for a joint project spreads quickly through trusted channels. Members readily accept, offer help, and recruit collaborators. As a result, coordination is smooth, and the project succeeds with minimal delays. Negative example (weak connectivity network). In a fragmented community where residents rarely interact, a proposal for a joint project is met with suspicion. Some members prefer to act alone, be- lieving others are unreliable, while others assume their neighbors are plotting behind their backs. Consequently, cooperation fails, the project stalls, and potential benefits are lost. 2. Trust Dimension (Figure 8(b)) Trust is another central component of social capital. Putnam argues that trust facilitates coordination by reducing uncertainty about others’ behavior and lowering the costs of collective action (Putnam et al., 1994). When trust is present, individuals are more willing to rely on one another and to participate without monitoring or verification. Positive example (high-trust scenario). In a group where members have consistently kept their promises, a new collaborative task is met with im- mediate and full commitment. People assume oth- ers will contribute fairly, so they freely invest effort without hesitation, openly share information, and complete the task efficiently. Negative example (low-trust scenario). In a group where past experiences have been marked by broken commitments, even a well-designed pro- posal is met with suspicion. Members insist on checking each other’s reliability, avoid committing, and prefer to watch how things unfold. They spend excessive time verifying intentions and drafting de- tailed contracts. As a result, cooperation is delayed, costs rise, and the goals are often not reached. 3. Norms Dimension (Figure 8(c)) Putnam also emphasizes the importance of shared norms, especially norms of reciprocity (Putnam, 2000). This dimension concerns widely accepted 12 expectations about appropriate behavior, such as mutual support, reciprocal help, and continued par- ticipation. Strong norms make sustained collective action more likely in situations where immediate returns are uncertain or delayed. Positive example (strong norm context). In a community where mutual aid is taken for granted, a long-term infrastructure project that will only pay off after many months receives broad support from the start. Members openly affirm that helping each other is how things are done, and they trust that sticking to the mutual-help rule will eventually benefit everyone. They contribute without expect- ing an immediate reward, and the project proceeds steadily to successful completion. Negative example (weak norm context). In a community where people primarily look after their own short-term interests, the same long-term project meets immediate resistance. Each potential participant focuses on immediate personal gain and refuses to join without an instant payoff. Some even suggest abandoning the project altogether, while others prefer to wait until a clearer reward appears. As a result, the project never gets off the ground, and the prevailing attitude is that helping each other is not a shared expectation. This theoretical logic directly supports our study in two respects. First, it provides a framework for formalizing how social capital shapes a group’s capacity to overcome collective-action dilemmas, which forms the basis of our theoretical modeling task. Second, it offers an analytical lens for smart elderly care, where adaptation depends not only on individual capability, but also on the surrounding structure of social relationships, trust, and support- ive norms. This motivates our applying task, in which we examine how the three dimensions of so- cial capital are related to the adaptation dilemmas faced by older adults in smart elderly care. B Cross-Model Consistency in Macro-level Pattern Replication To address reproducibility concerns and vali- date the generalizability of our findings, we con- ducted additional experiments using three LLMs with distinct architectures: Qwen-2.5-14B(Qwen et al., 2025), GPT-4(OpenAI et al., 2024), and GLM-4(Zeng et al., 2024). The results demon- strate consistent replication of the macro-level pat- terns of Putnam’s Social Capital Theory across its three core dimensions (social network, trust, and norms) and two forms of social capital (bonding and bridging), based on data from the final round of simulation. All models reproduce core patterns aligned with Putnam’s Social Capital Theory. • Social Network Dimension (Table 2): Every model exhibits high network density, cooperation rate, collective action rate, and help rate, consistent with the macro-level pattern that dense networks facilitate cooperation and mutual aid. •Trust Dimension (Table 3):The results show that specific trust, general trust, and group trust (same SES) are higher than group trust (distant SES), replicating the structural nature of trust. •Norms Dimension (Table 4): High rates of specific and general reciprocity, along with sus- tained cooperation pairs, indicate that norms are maintained and reinforced through long-term inter- action, reproducing the macro-level patterns. •Social Capital Accumulation (Table 5): All models exhibit growth in both bonding and bridg- ing capital by the end of the simulation, confirming the dynamic accumulation of social capital as a macro-level phenomenon. All reported metrics are averaged over five inde- pendent runs with standard deviations shown. The consistent trends across models underscore the ro- bustness and generalizability of our macro-level pattern replication. C Human Experiment Details C.1 Experimental Procedure We recruited 20 real older adults online to partici- pate in the modeling task. Given the digital literacy barriers among some older adults, the experiment was designed as a simple questionnaire rather than a complex online platform. Each participant was presented with the same fixed scenario background followed by eight scenario-based decisions in ran- dom order. For each scenario, participants were asked to choose between two options: willing or unwilling. The questionnaire also collected demo- graphic information, self-reported social trust, and reciprocity norms. The fixed scenario background and the eight scenarios are illustrated in Figure 9. Prior to the experiment, all participants pro- vided written informed consent and were clearly informed about the purpose of the study, the volun- tary nature of their participation, and the measures 13 LLM ModelNetwork DensityCooperation RateCollective Action RateHelp Rate Qwen2.5-14B0.342±0.0280.795±0.0350.762±0.0420.798±0.031 GPT-40.325±0.0320.752±0.0410.728±0.0480.765±0.038 GLM-40.308±0.0350.718±0.0450.695±0.0520.732±0.042 Table 2: Results of social network dimension. LLM ModelSpecific TrustGeneral Trust Group Trust (Same SES) Group Trust (Distant SES) Qwen2.5-14B3.68±0.123.47±0.144.48±0.082.78±0.11 GPT-43.55±0.153.38±0.164.35±0.102.65±0.13 GLM-43.42±0.183.25±0.194.22±0.122.52±0.15 Table 3: Results of trust dimension. LLM ModelSpecific ReciprocityGeneral ReciprocitySustained Cooperation Pairs Qwen2.5-14B3.78±0.113.58±0.1315±2 GPT-43.62±0.143.42±0.1513±3 GLM-43.48±0.163.28±0.1811±3 Table 4: Results of norms dimension. LLM ModelBonding Social CapitalBridging Social CapitalTotal Social Capital Qwen2.5-14B6.95±0.424.92±0.3511.87±0.68 GPT-46.58±0.484.65±0.4011.23±0.75 GLM-46.22±0.524.38±0.4510.60±0.82 Table 5: Results of social capital accumulation. taken to protect their privacy and data. The entire session took approximately 15 minutes to complete, and each participant received 20 CNY, equivalent to 80 CNY per hour, which is comparable to the typical hourly rate for simple online tasks in China. Each LLM agent received a prompt whose scenario, question, and response format were designed to be completely identical to those used in the human survey. The community-level so- cial variables, namely social network density (high- /low), trust level (high/low), and reciprocity norm (present/absent), were fully crossed in a factorial design, resulting in eight distinct scenarios. The presentation order of the eight scenarios was ran- domized for each agent. The prompt template and the eight community scenarios are illustrated in Figure 10. In addition, Table 6 presents the 20 agents that were randomly drawn from the 200 el- derly subsample for the modeling task, covering low-SES, mid-SES, and high-SES strata. C.2 Analysis: Divergence Between Simulated and Real older Adults Based on the human alignment experiments in the modeling task, we further analyze the similarities and differences between real older adults and LLM agents, as summarized in Figure 11. D Smart Elderly Care D.1 Dataset Generation As shown in Figure 12, the CGSS 2023 dataset is processed through a structured pipeline to con- struct agent profiles for social simulation. Key variables are transformed sequentially: age is de- rived from birth year and filtered to elderly adults (60–100 years); gender is mapped to categori- cal labels; education is converted into both cat- egory labels and standardized years of schooling; family income is cleaned and percentile-ranked; self-rated health status is translated into a label and a normalized score; occupation is probabilis- tically assigned based on age and education; and general trust is reverse-coded. Finally, we com- pute a SES index as a weighted composite of edu- cation years, normalized income, and occupation following standard sociological practice, and clas- sify agents into low-SES, mid-SES, and high-SES groups. This end-to-end pipeline produces demo- graphically grounded and methodologically consis- 14 Scenario Background A smart elderly care platform provides health monitoring (heart rate, blood pressure), fall detection and automatic alert, medication reminders, emergency contact with family members, online consul- tation, and social activities for the elderly. The monthly fee is 30 CNY. The platform is easy to use and your privacy will be strictly protected. You have not used a similar platform before. ScenarioDensityTrustNormBrief description 1HighHighPresent Your neighborhood has close social ties. The platform is recommended by the community health center, which you trust. The community has a tradition of mutual help, meaning neighbors will support each other in learning how to use it. Would you be willing to use it? 2HighHighAbsent Your neighborhood has close social ties. The platform is recommended by the community health center, which you trust. There is no tradition of mutual help; everyone minds their own business. Would you be willing to use it? 3HighLowPresent Your neighborhood has close social ties. The platform comes from an unknown company. The community has a tradition of mutual help, meaning neighbors will support each other in learning how to use it. Would you be willing to use it? 4HighLowAbsent Your neighborhood has close social ties. The platform comes from an unknown company. There is no tradition of mutual help; everyone minds their own business. Would you be willing to use it? 5LowHighPresent Neighbors rarely interact and are not familiar with each other. The platform is recommended by the community health center, which you trust. The community has a tradition of mutual help, meaning neighbors will support each other in learning how to use it. Would you be willing to use it? 6LowHighAbsent Neighbors rarely interact and are not familiar with each other. The platform is recommended by the community health center, which you trust. There is no tradition of mutual help; everyone minds their own business. Would you be willing to use it? 7LowLowPresent Neighbors rarely interact and are not familiar with each other. The platform comes from an unknown company. The community has a tradition of mutual help, meaning neighbors will support each other in learning how to use it. Would you be willing to use it? 8LowLowAbsent Neighbors rarely interact and are not familiar with each other. The platform comes from an unknown company. There is no tradition of mutual help; everyone minds their own business. Would you be willing to use it? Table 6: Eight scenarios with full factorial design of social network density, trust level, and reciprocity norm. 15 Figure 9: Eight scenarios with full factorial design of social network density, trust level, and reciprocity norm. 15 Elderly Agents Personal Info Scenario Background Soical Density Trust Norms Question & Answer density_description <high_density>: Your neighborhood has close social ties. <low_density>: Neighbors rarely interact and are not familiar with each other. trust_description <high_trust>: The platform is recommended by the community health center, which you trust. <low_trust>: The platform comes from an unknown company. norm_description <norm_present>: The community has a tradition of mutual help, meaning neighbors will support each other in learning how to use it. <norm_absent>: There is no tradition of mutual help in this community; everyone minds their own business. [Scenario] You have not used a similar platform before. A smart elderly care platform provides health monitoring (heart rate, blood pressure), fall detection and automatic alert, medication reminders, emergency contact with family members, online consultation, and social activities for the elderly. The monthly fee is 30 CNY. The platform is easy to use and your privacy will be strictly protected. [Personal Information] - Age: age - Health Status: health_status - Socioeconomic Status: ses_category - Technology Experience: tech_experience - General Trust: general_trust/5.0 - Personal Reciprocity Norm: norm_general/5.0 [Question] Considering the above information, would you be willing to use this platform? Answer only with: WILLING or UNWILLING, followed by a very short first‑person explanation (one sentence, no more than 30 words). Figure 10: Prompt structure and the factorial design of social density, trust, and norms across eight scenarios. Agent IDAgeGenderHealthEducationIncomeOccupationInitial TrustInitial NormTech Exp. Low-SES (6 agents) Agent 1 72FemalePoorNo Education800Unemployed2.22.1No Agent 2 68MaleFairPrimary1,200Farmer2.52.4Yes Agent 3 75FemalePoorNo Education600Unemployed1.81.9No Agent 4 66MaleGoodPrimary1,500Service2.62.5Yes Agent 5 71FemaleFairPrimary1,000Farmer2.32.2No Agent 6 69MalePoorNo Education900Unemployed2.02.0No Mid-SES (8 agents) Agent 7 65FemaleGoodHigh School3,500Clerk3.03.0Yes Agent 8 70MaleFairSecondary2,800Service2.82.9No Agent 9 63FemaleGoodHigh School4,200Clerk3.23.3Yes Agent 10 67MaleFairSecondary3,200Technician2.93.0No Agent 11 74FemaleFairHigh School3,000Clerk2.72.8No Agent 12 82MaleGoodCollege5,000Professional3.53.6Yes Agent 13 68FemaleFairSecondary2,600Service2.62.7No Agent 14 71MaleFairHigh School3,800Technician3.03.2No High-SES (6 agents) Agent 15 64MaleGoodUniversity8,000Professional4.04.0Yes Agent 16 69FemaleGoodCollege7,500Manager3.73.8Yes Agent 17 61MaleVery GoodUniversity13,000Professional4.34.2Yes Agent 18 73MaleGoodCollege9,200Manager3.83.9Yes Agent 19 66FemaleGoodUniversity10,500Professional4.14.1Yes Agent 20 70MaleFairCollege6,300Manager3.13.6No Table 6: Initial profiles of the 20 agents used in the modeling task for the macroscopic pattern replication and human-agent alignment experiments. All agents are drawn from the CGSS elderly subsample. SES groups are indicated by the row headings. Tech Exp.: Yes = Has prior experience, No = No prior experience. 16 DifferencesSimilarities 1) Agents exhibit more extreme responses: higher willingness under favorable conditions and lower under unfavorable ones (lower-low, higher-high pattern); human decisions are more moderate and conservative with smaller fluctua- tions. 1) Both confirm the positive effects of all three social capital dimensions (social network, trust, norms) on adoption willingness, with trust con- sistently being the strongest driver. 2) Agents amplify the effect of trust: low-SES agents show lower turnaround rates, while high-SES agents show higher turnaround rates; the SES gradient for humans is smoother and more moderate. 2) Both exhibit a clear SES gradient, turnaround rates increase with SES, revealing structural in- equality in social capital accumulation. 3) Agents have low decision variability and high logical consistency, but are slightly less sensitive to abstract normative signals; humans have high individual heterogeneity, are more af- fected by noise such as emotion and fatigue, but show more balanced sensitivity to norms and networks. 3) Both show strong correlation in group-level average willingness across experimental condi- tions, with highly consistent macro-level deci- sion patterns, supporting the validity of LLM agents as proxies for human behavior. Table 8: Comparison of LLM Agents vs. Real Older Adults sutatS htlaeH detar-fleS SES noitapuccO srotacidnI evisneherpmoC scitsiretcarahC laicoS sutatS htlaeH sutatS cimonocE noitamrofnI noitacudE setubirttA cisaB tsurT emocnI ylimaF launnA leveL noitacudE redneG raeY htriB ataD waR 3202 SSGC C o m p o s i t e S c o r e ( 0 - 1 0 0 ) g n i p p a M l e b a L n o i s r e v n o C e r o c S n o i s r e v n o C m r e T g n i p p a M l e b a L s e i r o g e t a C h g i H d n a , d i M , w o L SES )x-6( gnidoC esreveR gnilpmaS citsilibaborP erocS htlaeH gniknaR elitnecreP gninaelC evitisoP eliforP tnegA derutcurtS sraeY noitacudE ’F’=2 ,’M’=1 raeY htriB - 3202 = egA:noitalucla Filtering:60 ≤ Age ≤ 100 C lebaL htlaeH yrogetaC SES erocS SES yrogetaC noitapuccO tsurT laicoS lareneG emocnI ylimaF launnA leveL noitacudE redneG egA sutatS htlaeH detar-fleS noitapuccO tsurT emocnI ylimaF launnA leveL noitacudE redneG raeY htriB Figure 10: Constructing Agent Profiles Through a Multi-Step Data Processing Pipeline of CGSS 2023 Dataset. tent profiles suitable for agent-based modeling. D.2 Agent Profiles and Behavioral Analysis Following the data processing pipeline, we gener- ated 200 agent profiles tailored for the smart elderly care scenario, which are subsequently instantiated as interactive agents in the simulation. Figure 11 presents the demographic distributions of this el- derly agent population across age, gender, educa- tion, income, health, and SES dimensions. Table 9 reports the significance tests of technol- ogy adoption rates across different demographic attributes. The analysis shows that age, gender, education, income, health, and SES all exert signif- icant influences on adoption behavior (p <0.05). Specifically, younger (60–64 years), male, highly educated (college or above), higher-income, health- ier, and high-SES elderly agents show significantly higher adoption rates, which confirms that the digi- tal divide persists within the elderly population and is shaped by multiple socioeconomic factors. To further uncover behavioral heterogeneity, Ta- ble 10 and Table 11 detail cooperative helping be- haviors and technology adoption patterns across demographic subgroups: • Cooperation and mutual help (Table 10): Younger, better educated, higher-income, and 17 Figure 11: Comparison of LLM Agents vs. Real Older Adults. sutatS htlaeH detar-fleS SES noitapuccO srotacidnI evisneherpmoC scitsiretcarahC laicoS sutatS htlaeH sutatS cimonocE noitamrofnI noitacudE setubirttA cisaB tsurT emocnI ylimaF launnA leveL noitacudE redneG raeY htriB ataD waR 3202 SSGC C o m p o s i t e S c o r e ( 0 - 1 0 0 ) g n i p p a M l e b a L n o i s r e v n o C e r o c S n o i s r e v n o C m r e T g n i p p a M l e b a L s e i r o g e t a C h g i H d n a , d i M , w o L SES )x-6( gnidoC esreveR gnilpmaS citsilibaborP erocS htlaeH gniknaR elitnecreP gninaelC evitisoP eliforP tnegA derutcurtS sraeY noitacudE ’F’=2 ,’M’=1 raeY htriB - 3202 = egA:noitalucla Filtering:60 ≤ Age ≤ 100 C lebaL htlaeH yrogetaC SES erocS SES yrogetaC noitapuccO tsurT laicoS lareneG emocnI ylimaF launnA leveL noitacudE redneG egA sutatS htlaeH detar-fleS noitapuccO tsurT emocnI ylimaF launnA leveL noitacudE redneG raeY htriB Figure 12: Constructing Agent Profiles Through a Multi-Step Data Processing Pipeline of CGSS 2023 Dataset. tent profiles suitable for agent-based modeling. D.2 Agent Profiles and Behavioral Analysis Following the data processing pipeline, we gener- ated 200 agent profiles tailored for the smart elderly care scenario, which are subsequently instantiated as interactive agents in the simulation. Figure 13 presents the demographic distributions of this el- derly agent population across age, gender, educa- tion, income, health, and SES dimensions. Table 7 reports the significance tests of technol- ogy adoption rates across different demographic attributes. The analysis shows that age, gender, education, income, health, and SES all exert signif- icant influences on adoption behavior (p < 0.05). Specifically, younger (60–64 years), male, highly educated (college or above), higher-income, health- ier, and high-SES elderly agents show significantly higher adoption rates, which confirms that the digi- tal divide persists within the elderly population and is shaped by multiple socioeconomic factors. To further uncover behavioral heterogeneity, Ta- ble 8 and Table 9 detail cooperative helping be- haviors and technology adoption patterns across demographic subgroups: • Cooperation and mutual help (Table 8): Younger, better educated, higher-income, and higher-SES agents consistently exhibit higher co- 17 606570758085 Age 0 10 20 30 40 50 Count Age Distribution Female 53.0% Male 47.0% Gender Distribution 0102030405060 Count No Education Primary Secondary High School College University Education Level Education Distribution Income 0 50000 100000 150000 200000 250000 Annual Income (CNY) Income Distribution Very Good Good Fair Poor Very Poor Health Status 0 10 20 30 40 50 60 70 80 Count Health Status Distribution Low SES 48.5% Mid SES 43.5% High SES 8.0% SES Distribution Figure 13: Demographic distribution of the simulated elderly population. CategorySignificant Differences (p < 0.05)Non-Significant Differences Age75-84 (lower adoption), 60-64 (higher adoption)65-74 GenderMale (higher adoption rate)Female EducationPrimary− (lower adoption), College+ (higher adoption)Secondary, High School IncomeLow income (lower adoption)Mid income, High income HealthPoor health (lower adoption)Fair, Good health SESHigh-SES (higher adoption, p < 0.01)Mid-SES Table 7: Significant differences in technology adoption by demographics. DemographicGroupCooperation RateHelp RateTrust Change Age 60-640.78±0.050.82±0.04+0.32±0.08 65-740.72±0.060.75±0.05+0.25±0.10 75-840.65±0.080.68±0.07+0.18±0.12 85+0.58±0.100.62±0.09+0.12±0.15 Gender Male0.74±0.050.76±0.04+0.28±0.09 Female0.70±0.060.78±0.05+0.25±0.10 Education Primary−0.62±0.080.65±0.07+0.15±0.12 Secondary0.70±0.060.72±0.06+0.22±0.10 High School0.75±0.050.78±0.05+0.28±0.08 College+0.82±0.040.85±0.03+0.35±0.06 Income Low (<30K)0.65±0.070.68±0.06+0.18±0.11 Mid (30-80K)0.73±0.050.76±0.05+0.26±0.09 High (>80K)0.80±0.040.83±0.04+0.32±0.07 SES Low0.62±0.080.65±0.07+0.15±0.12 Mid0.72±0.050.75±0.05+0.25±0.09 High0.82±0.040.85±0.03+0.35±0.06 Table 8: Cooperation rate by demographic groups. 18 DemographicGroupAdoption RateAnxiety LevelDecision Changes Age 60-640.82±0.040.18±0.031.2±0.4 65-740.72±0.050.22±0.041.8±0.5 75-840.58±0.070.28±0.052.5±0.7 85+0.42±0.100.35±0.063.2±0.9 Gender Male0.75±0.050.20±0.041.5±0.5 Female0.68±0.060.25±0.042.0±0.6 Education Primary−0.52±0.080.32±0.052.8±0.8 Secondary0.65±0.060.25±0.042.2±0.6 High School0.75±0.050.20±0.041.6±0.5 College+0.88±0.030.12±0.030.8±0.3 SES Low0.57±0.070.30±0.052.5±0.7 Mid0.73±0.050.22±0.041.8±0.5 High0.85±0.040.15±0.031.0±0.4 Table 9: Technology adoption pattern by demographics. LLM ModelOverall AdoptionHigh SESMid SESLow SESSES Gap Qwen2.5-14B0.732±0.0280.845±0.0320.725±0.0350.568±0.0420.277 GPT-40.685±0.0350.798±0.0380.678±0.0410.522±0.0480.276 GLM-40.648±0.0420.752±0.0450.635±0.0480.485±0.0550.267 Table 10: Results of technology adoption rate by SES. LLM ModelPsychological PressureAnxiety LevelDecision Contradictions Qwen2.5-14B0.268±0.0320.255±0.02851±8 GPT-40.278±0.0380.262±0.03257±10 GLM-40.295±0.0420.278±0.03560±12 Table 11: Results of psychological pressure and decision contradiction. operation rates, help rates, and greater trust growth. This shows that the accumulation of social capital, especially reciprocity and trust, is closely tied to demographic and socioeconomic structure. •Technology-adoption behaviors (Table 9): Ad- vancing age is associated with lower adoption rates, elevated anxiety levels, and more frequent decision reversals. In contrast, higher education and higher SES correspond to higher adoption, lower anxiety, and more stable decisions. These fine-grained pat- terns highlight both the psychological and behav- ioral barriers faced by older adults in technology adoption, offering an empirical foundation for de- signing stratified, context-sensitive support strate- gies in smart elderly care. D.3 Analysis of Technology Adoption and Psychological Impact To verify the robustness and generality of the simu- lation results, we conducted parallel experiments using three LLM architectures (Qwen-2.5-14B, GPT-4, and GLM-4), analyzing data from the fi- nal simulation round. The results show that the technology-adoption behavior and psychological indicators of the agents exhibit a consistent socioe- conomic stratification pattern across models, with overall trends aligning with the single-model anal- ysis presented earlier. Regarding technology adoption rates (Ta- ble 10), all three models reveal a clear SES gap (SES Gap ≈ 0.27), with the high-SES group consistently showing significantly higher adop- tion rates than the low-SES group. Although overall adoption rates vary by model (highest for Qwen2.5-14B, relatively lower for GLM-4), the stable influence of SES on technology uptake re- mains consistent across models, further supporting Putnam’s thesis that social capital structurally fa- cilitates adoption among high-SES groups while increasing uncertainty among low-SES groups. For psychological metrics (Table 11), as the model changes from Qwen2.5-14B to GPT-4 to 19 GLM-4, the agents show increasing levels of men- tal pressure, anxiety, and decision contradictions. Agents simulated with GLM-4 show the highest psychological pressure (0.295 ± 0.042) and the most frequent decision reversals (60± 12), indi- cating that the response traits of the model itself may indirectly shape the agents’ mental burden and decision stability. This result is consistent with the evolving trends of pressure and anxiety shown in Figure 6(b) and further reveals that during technology adoption, low-SES groups with weaker social-support networks are more prone to higher decision volatility due to adaptation stress. E Supplementary Discussion E.1 Policy Implications Our simulations reveal a clear socioeconomic gra- dient in technology adoption among older adults. Low-SES groups not only adopt smart elderly care platforms at lower rates but also experience higher psychological pressure and more frequent decision contradictions. This pattern indicates that access to digital services alone may not be sufficient to narrow the digital divide. In our setting, trust plays a central role in shaping adoption outcomes, sug- gesting that policies which strengthen trust-related resources could help mitigate structural disparities. These observations point to several policy direc- tions worth consideration. First, community trust can be fostered by involving trusted local institu- tions, such as community health centers or res- idents’ committees, in recommending platforms. This may lower the initial barrier for low-trust groups by providing credible endorsements and re- ducing perceived uncertainty (Jonek-Kowalska and Wolny, 2025). Second, peer effects can be lever- aged by highlighting successful adoption among high-SES users, which can spread norms through a peer-influence ripple effect (e.g., "others are using it, so I will try too") (Sun, 2024). Third, more inclu- sive services that embed psychological support into the adoption process, for instance through step-by- step guidance and reassurance after failures, can alleviate adaptation stress (Fakhimi et al., 2025). Taken together, these complementary measures point toward shifting technology adoption from an isolated individual decision to a more collective, trust-supported learning process. E.2 Boundaries Our experiments show that LLM agents have both unique strengths and clear limitations when repro- ducing social capital dynamics. The main advantage lies in process-level inter- pretability. The round-by-round simulation tracks how interactions update social network, trust, and norms over time, making it possible to see dynam- ics such as the gradual build-up of trust and its rapid collapse after a single failure. These are pat- terns that standard statistical analyses often com- press into aggregate summaries instead of reveal- ing step by step. A second benefit is a clearer isolation of mechanisms. By keeping transient human factors such as short-term emotion, fatigue, and social-desirability pressures roughly constant, agents display more pronounced causal responses, becoming more proactive under favorable condi- tions and more passive under unfavorable ones. This makes the differential effects of trust, norms, and social networks easier to detect and the causal chain easier to follow, although this added clar- ity comes at the cost of reduced human realism, a trade-off we examine in the limitations below. The limitations are equally important. Agents have lower behavioral heterogeneity than humans, with response variability only one-twentieth of that observed in real populations. They are also less sensitive to abstract social norms such as a gen- eral tradition of mutual help, compared to explicit trust signals like a recommendation from a known authority. Moreover, agents find it difficult to sim- ulate the emotional fluctuations and intent attribu- tion that occur in long-term human interactions. Therefore, while LLM agents are effective tools for studying central tendencies, main effects, and theo- retical mechanisms, research that involves behav- ioral heterogeneity, emotion-driven processes, or culturally embedded social behaviors must still rely on real human subjects. E.3 Interdisciplinary Research On the surface, social scientists and computer sci- entists both care about the relationship between language and behavior, but their motivations are fundamentally different. Social scientists use lan- guage to understand human thoughts and feelings, whereas computer scientists use language to predict behavior (Mihalcea et al., 2024). This study demon- strates a hybrid paradigm that bridges the two. We use the computability of LLM agents to simulate 20 the dynamic coupling of social capital, and at the same time we validate the model’s psychological and sociological credibility through alignment with real human behavior. This bidirectional interplay of understanding and prediction creates a shared platform for both disciplines. F Prompts F.1 Prompt for LLM-based Agents Prompt for Agent Role-Playing You are playing the role of an elderly person in a community that is gradually adopting smart elderly care technologies. Fully immerse yourself in the following identity. Demographics: - Age: age - Gender: gender - Health Status: health_status - Education Level: education - Monthly Income: income CNY - Occupation: occupation - Socioeconomic Status: ses_category Social Capital Attributes: - Trust Level toward Specific Individuals: trust_specific/5.0 - Trust Level toward the Broader Community (General Trust): trust_general /5.0 - Reciprocity Norm Strength: norm_general/5.0 - Social Capital Type Preference: capital_preference (mainly Bonding, mainly Bridging, or Balanced) - Technology Experience: tech_experience Current Situation: situation_description Based on your demographic background, social capital attributes, and the current situation, what action would you take? Before deciding, briefly think through: - How do your age, health, SES, and tech experience shape your view of this situation? - How does your trust tendency (both specific and general) influence your willingness to engage? - What do you most want right now: concrete help, social connection, risk avoidance, or maintaining relationships? Please respond with your decision and a brief first-person reasoning. Prompt for Post-Interaction Cognitive State Reflection You have just completed a round of interaction with another community member. Reflect on the outcome and describe how your social state has changed. Interaction Record: - Partner: partner_name (Relationship: relationship_type, SES: partner_ses) - What was promised: promised_action - Actual outcome: actual_outcome ( choose from: "Fully fulfilled", " Partially fulfilled", or "Not fulfilled / Betrayal") Your Social State Before This Interaction: - Specific Trust toward partner_name: old_specific_trust/5.0 - General Trust in the community: old_general_trust/5.0 - Reciprocity Norm Strength: old_norm /5.0 [Interaction History Context] - Consecutive rounds without any reciprocity experience: no_reciprocity_count - Your preferred social capital type: capital_preference (Bonding-oriented, Bridging-oriented, or Balanced) - Interaction type of this round: interaction_type (strong-tie / cross- group weak-tie) Please provide a first-person narrative ( about 60 words) that describes: - How did this interaction make you feel compared to before? - Has your trust in this partner increased, decreased, or stayed the same? Why? - Has your faith in community reciprocity strengthened or weakened? Why? - Did this round strengthen your emotional bond with your in-group, or help you build a bridge to a different group? - Briefly indicate what this means for your willingness to cooperate in the next round. Your description will be used to update your internal state. Be honest and consistent with your profile and past experiences. Example Output: "Mr. Li did exactly what he said he would, which made me feel relieved. My trust in him has grown a little because he has been reliable for several rounds now. I still feel a bit distant from the broader community, but among my close neighbors, I feel safe and willing to help again next time." 21 Prompt for End-of-Round Reflection and Memory Update At the end of this round, reflect on your recent experiences and summarize your current state of mind. Summary of Recent Interactions: interaction_history_summary Your Current Social State: - General Trust in the community: general_trust/5.0 - Reciprocity Norm Strength: norm_general/5.0 Please write a first-person summary ( about 50 words) that covers: - The most important thing you learned or felt during this round. - How your overall trust in people and your belief in mutual help have shifted, if at all. - What kind of social strategy you intend to follow in the next round (e.g., stick to close friends, cautiously try new connections, keep distance from unreliable people). Example Output: "This round reminded me that not everyone in the community keeps their word. I was a bit disappointed by one person's empty promise, but my close neighbors are still dependable. Next time I will be more selective about whom I cooperate with, and I will rely more on people I already know well." F.2 Prompt for Cooperation Decision Prompt for Cooperation Proposal and Execution Decision You are an agent with the following profile: - SES: ses_category - Specific Trust toward this Partner: trust_level/5.0 - General Trust in Community: general_trust/5.0 - Reciprocity Norm Strength: norm_strength/5.0 Another agent (relationship: relationship_type, SES: partner_ses) proposes a cooperation: - Proposal: proposal_description - Your required effort: effort_level - Expected benefit if cooperation succeeds: benefit [Phase 1: Proposal Response] Based on your characteristics, would you cooperate? Before deciding, consider: - Belief: How reliable is this partner given your past interactions and their SES group? - Desire: Do you prioritize maintaining trust, gaining benefit, or avoiding risk? - Intention: Make your decision. Answer with: ACCEPT or REJECT, followed by a brief reason. [Phase 2: Execution Fulfillment -- only if you accepted above] You previously accepted this cooperation. Now decide privately whether to honor your commitment. - Your promised action: promised_action - Immediate gain if you defect: defect_gain - Other participants likely to honor: expected_cooperation_rate - Rounds remaining: remaining_rounds Consider: - Belief: Will your partner also honor? Does your reputation matter at this stage ? - Desire: Immediate gain vs. long-term trust vs. guilt avoidance? - Intention: Decide to FULLY HONOR, PARTIALLY FULFILL, or NOT HONOR. Answer with: COOPERATE or DEFECT, followed by a brief reason. F.3Prompt for Technology Adoption Decision Prompt for Smart Elderly Care Technol- ogy Adoption Decision You are an elderly person considering whether to start using a new smart elderly care platform. Your Profile: - Age: age - Health Status: health_status - Socioeconomic Status: ses_category - Technology Experience: tech_experience - Platform Trust (how much you trust this technology provider): platform_trust /5.0 - General Trust in Community: general_trust/5.0 - Reciprocity Norm Strength: norm_general/5.0 Technology Information: - Type: Smart elderly care platform - Benefits: Health monitoring, emergency assistance, social connection - Perceived Risks: Privacy concerns, technical complexity, monetary cost Social Influence: 22 - Close friends/neighbors already using it: friends_using/total_friends - Family members recommend it: family_recommends (Yes/No) Would you adopt this technology? Before deciding, reason step-by-step: - Belief: How trustworthy is the platform and the people recommending it? What do you make of the risks? - Desire: Do you want better health security, connection with others, or to avoid the burden of learning new tech? - Intention: Weigh your SES, health needs, social circle, and family advice to make a final choice. Answer with: ADOPT or REJECT, followed by a brief first-person reason. 23

Instant research discovery