Multi-level Bayesian Analysis of Policy Preference and Political Behavior in Taiwan

It is widely believed that cross-cut social cleavages benefit democracy because the mass public has diversified opinions. Therefore, it is necessary to examine public opinions and drew possible consensus on the sub-national level, such as county or election districts. We aim to develop a multilevel Bayesian model that combines both polling and census data, thereby we can evaluate sub-level public opinions, such as county and township.

The previous project that predicts the 2004 presidential results is extended to the legislative election in 2007, which will use single-member district. Our Bayesian analysis can simulate enough samples from each of the two-wave national polling data included in this project, predicting public opinions of every county and legislative district. After estimating the influence of each parameter, we use poststratification technique to predict behavior of sub-levels, such as party vote share of each county.

The Ministry of Interior collected census data in 2000, which allows us to calculate the proportion of every category, such as gender, age, and education. Moreover, we can fit varying models with different dependent variables, such as multi-nomial logistic regression model.We also manage to collect the budgets of local governments and calculate the association between the budget and public opinions. In doing so, we can show the degree to which the local governments are responsive to public opinions.

Additionally, we will code candidates』 issue positions printed on the election bulletin. With the simulated public opinion data, we can examine the extent to which the point of view of candidates is congruent public opinion. This project will contributes to finding public opinions at the local level, measuring representation, and predicting voting shares of the small districts.