A Bayesian Two-Part Latent Class Model for Longitudinal Government Expenditure Data: Assessing the Impact of Vertical Political Alliances and Political Support

• Activity Number: 88
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Felipe Nunes*+
• Keywords: Longitudinal ; Government expenditures ; MCMC ; Brazil

Abstract

Latin American presidents have great discretion over targeted spending decisions. Such expenses, however, are rarely modeled correctly. This paper implements a bayesian model to analyze an original longitudinal government expenditure data from Brazil in the timeframe of 1990 and 2012. This data presents three key features: first, the data is semicontinuous, assuming nonnegative values with a spike at zero for municipalities that did not receive any investments from the president, followed by a continuous, right-skewed distribution for those places which received something. Secondly, each municipality contributes an observation for each of the years, introducing within-subject correlation. Finally, because municipalities tend to share characteristics related to government investments, investment data fall into categorical classes. The model proposed is a mixture of a degenerate distribution at zero and a positive continuous distribution for the nonzero values. The model was fitted in R using an MCMC algorithm provided by Neelon et. al (2011). I run MCMC chains for 200,000 iterations each. The results seem to be more appropriate for the analysis required in the social sciences.