Using National Surveys to Estimate Local Public Opinions: An Application of Bayesian Multilevel Regression Model with Post-stratification





Published date: 

June, 2006


Eric Chen-Hua Yu
Chia-Hung Tsai


Survey research in Taiwan almost always focuses on studying public opinion in the national level. Due to the potential small n problem for less populous sub-geographic units, scholars have not utilized national survey data to estimate public opinions in any local level (e.g., county, town, or village). To overcome this sample size limitation, we use a method combining the multilevel modeling approach with the population information for post-stratification. We apply it to a set of pre-election national surveys of the 2004 Taiwan presidential election (sample size <2200) and produce county-level estimates of vote choice. Specifically, we first construct a multilevel logistic regression model to estimate the mean of the vote choice variable given demographics and county of residence. Second, we post-stratify on all the variables in the model by using the joint population distribution of the demographic variables within each county. Comparing our estimates with the actual county-level election outcomes, the average absolute error is less than 2 percentage points.