Democratic Confidence Intervals for Adjusted Means and Incidence Rates
Uraiwan Sampantarak, Noodchanath Kongchouy, Metta Kuning

Abstract
This paper presents confidence intervals for adjusted means using linear regression with weighted sum contrasts and applies the method to a study comparing blood lead concentrations among Pattani school children. These intervals are “democratic” in the sense that they compare the mean for each factor level with the overall mean, rather than selecting a referent. Confidence intervals based on negative binomial GLM models with sum contrasts are also recommended for comparing incidence rates for levels of a study factor after adjusting for biases due to associations with covariate factors. This method is applied to the terrorism events in regions of Southern Thailand that occurred over the period from 2004 to 2009.

Full Text: PDF