Judgment Post-Stratification Estimation of Population Proportion with High Missing Data Rate

• Activity Number: 141
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Tian Chen*+ and Elizabeth Stasny and Tao Shi
• Companies: The Ohio State University and The Ohio State University and The Ohio State University
• Keywords: missing data ; judgment post-stratification ; random forests adapted JPS

Abstract

Abstract: In many data mining problems where the goal is to estimate a population proportion, the percentage of missing data can be quite high. The usual practice of ignoring missing data assumes a missing completely at random (MCAR) mechanism, which might be seriously violated in some applications. Judgment post-stratification (JPS) estimation of a population proportion has been shown to increase the precision over the commonly used simple random sample proportion estimator. We compare the JPS estimator with an estimator based on random forests (RF) outcomes assuming that missingness is related to either the response or explanatory variable, referred as Missing Not at Random (MNAR) or Missing at Random (MAR) respectively. In particular, we develop and analyze a random forests adapted JPS estimation method. We use a dataset collected by NASA's satellite remote sensing instruments MODIS and CloudSat/CALIPSO as a test bed to demonstrate the benefits of JPS, RF and RF adapted JPS for estimating a population proportion when missingness is not MCAR.