Underestimation of Standard Errors in Regression Analysis for Pollution Exposure Assessment Using Multi-Source Data

• Activity Number: 200
• Type: Contributed
• Date/Time: Monday, August 1, 2016 : 10:30 AM to 11:15 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): Tomoshige Nakamura* and Mihoko Minami
• Companies: Keio University and Keio University
• Keywords: regression imputation ; generalized linear model ; estimating equation ; particulate matter ; community health survey ; land use regression

Abstract

We are interested in investigating the effect of particulate matter exposure on human heath in Japan using community health survey data. However, in Japan, the number of the monitoring stations around the survey area is very limited and the observations on these measurements at local community area are not generally available. When particulate matter concentrations are not observed in survey area, Land Use Regression (LUR) is often used to fill the missing values.

In general, if we use regression imputation to fill the missing values, the inference based on regression imputed data might be wrong. For example, the consistency of estimator may be violated, and the variance of estimator may be underestimated. So, in our research, we try to clarify the problem using regression imputation when we estimate the effect of particulate matter.