State-Space Poisson Models for Event-Count Time Series： Evidence from the Legislative Productivity of the U.S. Congress

For event-count time series data, the Poisson distribution is limited because it assumes independent events. In this study, we horserace two state-space Poisson methods in analyzing the legislative productivity of the U.S. Congress：the Poisson exponentially weighted moving average (PEWMA) and Poisson autoregressive (PAR) models. They distinguish from the conventional Poisson and negative binomial event-count models by explicitly specifying the time-series process in a Bayesian parameter-feedback context. We find that (1) a divided government does not matter for the legislative productivity of the U.S. Congress; but, (2) electoral consideration of lawmakers urges them to make more laws in the election year, and (3) wider-seat margin Senatorial majority legislate more.