Title: A partially pooled Bayesian hierarchical model for aggregated relational survey data Authors:  Rowland Seymour - University of Birmingham (United Kingdom) [presenting]
Abstract: Aggregated relational data can be collected in household surveys to estimate the number of people who have been affected by crimes. This method asks survey respondents questions of the form 'How many people do you know with feature X?'. Surveys of this kind can require large sample sizes when the social network structure of the population is heterogeneous. To allow us to reduce the sample size and simultaneously provide estimates for different subgroups in the population, we develop a partially pooled Bayesian hierarchical model. Through a linear predictor, we introduce correlation between the subgroup model parameters and assume that the parameters for the subgroups come from a national-level distribution, which allows us to share information between the subgroups. Inference for the model can be quickly implemented in Stan. We demonstrate this model on a new data set on child sexual abuse in the Philippines, and show how the results led to new laws in the USA and the Philippines.