The risk of aggregating networks when diffusion is tie-specific

Abstract

Empirical studies of the spread of something through social networks, a process often called diffusion, tend to rely on network data assembled from the measurement of multiple kinds of social ties. These can be different kinds of relationships, such as friendship and kinship, or different instances of concrete interactions, such as borrowing money and eating meals together. Aggregating multiple measures of ties into a single social network has become standard practice, typically done by taking a union of the various tie types. Although this has intuitive appeal, we show that in many realistic cases, this approach adds sufficient error to bias and mask true network effects. We further demonstrate that the problem depends on: (1) whether the diffusion occurs generically or in a tie-specific way, and (2) the extent of overlap between the measured network ties. Aggregating multiple measures of ties when diffusion is tie-specific and overlap is low will, on average, attenuate and potentially mask network effects that are in fact present.