B1266
Title: Generalized linear mixed models for binary responses: An application to resignation from the Brazilian Army
Authors: Cleber Iack - Faculdade de Ciencias - Universidade de Lisboa / Ministerio da Defesa - Brasil / Bolsista do CNPQ - Brasil (Portugal)
Helena Mourino - Faculdade de Ciencias - Universidade de Lisboa (Portugal) [presenting]
Abstract: Modeling the relationship between explanatory variables and a response variable is a crucial task in Statistics. When the response variable is binary, the generalized linear model is used. However, this model assumes that observations are independent of each other. For clustered and/or longitudinal data, the fixed effects model cannot be applied any more. In this context, there is the need to incorporate the clustered (observations are nested within larger units) and/or longitudinal (repeated observations are nested within subjects) nature of the data. The resulting model is a generalized linear mixed model, which can incorporate both fixed and random effects for the regressors. For longitudinal data, the random effects models allow the regression coefficients to vary between subjects, which describe the effect of each individual on its own repeated observations. An application of the model to resignation from the Brazilian Army is given. It aims at identifying the variables intrinsically related to early exit. From 2011 to 2014, 16540 militaries were analysed. The sampled data was clustered by military's background (IME, AMAN, EsSEx and EsAEx). Also, each military was evaluated by 5-points Likert scale on a semester basis. Models with different random effects specification were compared using the likelihood ratio tests.
Journal