Distinguishing Network Ties

Political preferences and collective action are conditioned by the density and structure of network ties, but different types of network connections -- such as social, economic, and political ties -- are too rarely distinguished from one another. The plots above illustrate the differences between the social and political network in one slum neighborhood in Jaipur, India. The social network is relatively "flat" and non-hierarchical, exhibiting a Bernoulli-type degree distribution. Moreover, the social network exhibits a high degree of homophily: Hindus (yellow nodes) tend to connect to other Hindus, and Muslims (green nodes) tend to connect with other Hindus. By contrast, the political network in the same neighborhood is highly hierarchical, exhibiting a degree distribution indicative of a power-law or preferential-attachment data-generating process. Moreover, the political network is much more integrated, with most of the clusters containing both Hindus and Muslims. Distinguishing between different types of ties allows researchers to uncover variation of this type, with implications for political preferences and collective action potential.

Dimensions of Segregation

Recent work on the effects of segregation have pointed to the importance of distinguishing between different spatial patterns of segregation. However, existing segregation metrics are of limited utility in measuring whether segregation occurs along a single dividing line -- as in the figure on the left -- or is characterized by multiple interlocking clusters, as in the figure on the right. I have developed a new segregation measure, which is based on the location of individuals, rather than the ethnic compositions of aggregate areal units. My measure provides a distribution of the outgroup exposure for an entire population; the first moment of the distribution (the mean) corresponds to the overall amount of segregation, while the second moment (the variance) corresponds to the extent to which the spatial pattern is characterized by a single dividing line (as in the figure on the left). In a current working paper, I use simulations to establish the properties of this segregation measure for binary- and cluster-type data generating processes, as shown above, and compare it to existing metrics.

Issues of Scale

Studies of ethnic heterogeneity and segregation frequently make use of administrative data at the level of census tracts or city wards; an example of this type of data is shown above at left, which gives ethnic composition at the ward level for Jaipur, India. However, such measures are subject to the modifiable areal unit problem (MAUP), whereby measures over aggregated units can be highly contingent on the scale of the measurement and the location of the boundaries between units. Moreover, aggregate measures can mask substantial variation within units. For example, in Jaipur, residents in the blue zones in the southeast have few low-caste individuals in their own ward, but are quite close to wards with very high concentrations of low castes. By disaggregating segregation and exposure measurement to the level of the individual, we can capture this fine-grained variation. This approach can be applied to geocoded survey samples, as I demonstrate in a current working paper, or it can be applied to synthetic population data, as shown at the top right for the neighborhood where I lived in NYC.