A stochastic blockmodels framework for the analysis of treatment response with social interaction

Quem: 

Thiago Costa

Onde: 

Praia de Botafogo, 190 - sala 317

Quando: 

10 de Agosto de 2017 às 16h

We introduce the use of a stochastic blockmodels approximation framework (SBA) to develop a methodology for assessment of social interaction in treatment response, assuming that the response of each individual is influenced by the treatment given in his social neighborhood. We show how information provided by the SBA about the structure of the network can be used to improve identifiability and to optimize estimation of social effect. Identifiability of treatment response is determined by the space of realized effective treatments, so in order to design optimal assignments it is necessary to describe this space. Understanding the space of realized effective treatment associated with a given treatment strategy can be a difficult challenge, given the complex ways the individuals' reference groups usually intersect. We propose a method to design treatment assignments that uses connection patterns given by the SBA to control the formation of effective treatments, potentially increasing the identification power of the experiment and improving estimation of model parameters. We apply our ideas to develop a methodology of experimental design for optimal estimation of treatment and social effects in linear models.

*Texto informado pelo autor. 

Palestrante: 

Thiago Costa possui graduação em Engenharia de Computação pela Unicamp, mestrado em Matemática pelo IMPA e doutorado em Matemática Aplicada pela Universidade de Harvard. Antes de juntar-se à EMAp, ele foi Moore Sloan Data Science postdoc na Universidade de Washington, em Seattle. Em sua pesquisa, ele desenvolve teoria e métodos para análise de dados relacionais, com foco em modelos não-paramétricos. Thiago também é co-fundador do Atlas Político.