|
Abstract:
|
Many analyses and imputations on missing values are based on the assumption that data are missing completely at random (MCAR) or missing at random (MAR). This study provides a review of commonly used methods in testing missingness mechanism. We summarize the methods developed for different settings, including cross-sectional data and longitudinal data. We further distinguish the methods from testing univariates and multivariate, unit nonresponse and item nonresponse, continuous and categorical variables. Eight methods are reviewed, such as Little’s MCAR test, Listing and Schilittgen’s test, Ridout's logistic regression method, false discovery rate, and so on. The theory, model assumption, testing procedure, and applicable conditions of these methods are discussed. We further applied these methods to a two-wave longitudinal dataset from Health and Retirement Study. Results indicate the eight methods consistently find the data are MAR. The advantages and limitations of these methods are discussed. Some comments of applying these methods in social survey data are provided.
|
