Data imputation

• Main Author: ROSENTHAL, Sonny
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2017
• Subjects: imputation; linear regression; maximum likelihood; measurement error; missing values; Communication; Critical and Cultural Studies
• Online Access: https://ink.library.smu.edu.sg/cis_research/212
• Institution: Singapore Management University

Data imputation involves representing missing values in a dataset. Missing data create a number of potential challenges for statistical analysis. Missing values can increase the chances of making Type I and Type II errors, reduce statistical power, and limit the reliability of confidence intervals. There are a number of statistical procedures available for researchers to replace missing values with reasonable estimations. Basic methods, such as mean substitution, regression imputation, and hot deck imputation may bias imputed values depending on the mechanism of missingness. Advanced methods include expectation maximization, full information maximum likelihood, and multiple imputation. These methods produce more reliable estimations of missing values, particularly when missingness is not at random.