A Data-Driven Agent-Based Model of Generalized and Particularized Trust

Empirical studies about trust strongly support that our experiences with known others influence our trust towards specific individuals (or particularized trust) and experiences with strangers our trust in society (or generalized trust). However, the degree to which they influence each other is still up for debate. Originally proposed by Putnam, it is suggested that generalized trust is grown form specific social circles in which we have built particularized trust with specific individuals, such as our family, friends, neighbors, colleagues, and acquaintances. This is believed to be specifically effective in dense, diverse and horizontally structured clusters in which individuals learn cooperative values. Testing this mechanism empirically has not been done before as it has been difficult to approximate the structure of social networks at a larger community scale. Whether this structural element in the social network is a prerequisite in the generalizing effect of trust, and how different network structures could facilitate different ways of development of generalized trust, is not clear.

To solve this puzzle, we propose a data-driven agent-based model that simulates the development of generalized trust and particularized trust through experiences embedded in a social network. Agents play investment games as a means of simulating experiences in which trust is given, reciprocated or betrayed, and altered. We use ERGMs on personal network data collected through a survey on social cohesion (N = 6,000) to estimate society-scale social network characteristics and trust distributions, in order to reflect on both realistic macro-structures (in terms of e.g., clustering or density) as well as realistic dyadic relationships (in terms of e.g., social class or origin) in our simulations. This allows us to explore the effects of the network structure (and the properties and distribution of its members) on the development in trust in realistic societal configurations, as well as stylized configurations.