Multilevel Multivariate Modelling of Legislative Count Data, with a Hidden Markov Chain

Abstract

SummaryThe production of legislative acts is affected by multiple sources of latent heterogeneity, due to multilevel and multivariate unobserved factors that operate in conjunction with observed covariates at all the levels of the data hierarchy. We account for these factors by estimating a multilevel Poisson regression model for repeated measurements of bivariate counts of executive and ordinary legislative acts, enacted under multiple Italian governments, nested within legislatures. The model integrates discrete bivariate random effects at the legislature level and Markovian sequences of discrete bivariate random effects at the government level. It can be estimated by a computationally feasible expectation–maximization algorithm. It naturally extends a traditional Poisson regression model to allow for multiple outcomes, longitudinal dependence and multilevel data hierarchy. The model is exploited to detect multiple cycles of legislative supply that arise at multiple timescales in a case-study of Italian legislative production.