Sociological research highlights the importance of social influence in the formation of individual preferences and the propagation of aggregate trends. Social scientists have long recognized the importance of social influence in opaque settings where outcomes are highly uncertain, such as financial markets, political elections, and cultural industries. Within the sociology of markets, cultural industries have received particular attention because they typically lack objective standards of valuation.
Social influence can give rise to complex dynamics and hard-to-predict collective outcomes. Due to the uncertainty of valuation in cultural markets, individuals look to the choices of others to determine which cultural products they will likely judge favorably or should even consider in the first place. Situations in which people react to other individuals who are reacting likewise are prone to give rise to processes of cumulative advantage. These dynamics can result in highly skewed outcomes, such as the emergence of a few “superstars,” that do not necessarily reflect inherent qualities of the product (the “bad bestsellers” phenomenon).
At IAS we study these phenomena in markets for music and literature. Our research applies models of social influence to prominent sites of social valuation like the Nobel Prize for Literature. More recently, the advent of global online music platforms has allowed the study of cultural dynamics on a very large scale. Scrutinizing cultural choice at Spotify.com, a leading online music platform, we identify a social-influence mechanism that can widen individuals’ behavioral repertoires in song selection (see figure below). Using the estimated influence parameters in an agent-based simulation, we find that the type and strength of social influence we observe on Spotify does indeed give rise to cultural change.
Two-dimensional representation of cultural taste profiles for a subset of 50,000 Spotify users. Each dot represents a user, and the color indicates the particular taste cluster each user was assigned to by a k-means algorithm.