Missing Data? Fear Not! Best Practices for Handling, Reporting, and Embracing Missing Data

This article is an accessible resource about missing data, handling and reporting missing data, plus introduces planned missing data designs. The first section provides a straightforward introduction to missing data mechanisms: missing completely at random, missing at random, and missing not at random. By understanding the different characteristics behind these three mechanisms, researchers can plan how to turn missing not at random data into missing at random data by the inclusion of items related to potential reasons for missingness or nonresponse. The second section reviews how handling missingness has evolved with an emphasis on the different modern options for handling missing data-multiple imputation and full-information maximum likelihood (FIML)-and how the approach used to handle missingness influences parameter bias, power, and generalizability of the results. Software programs have started to use FIML estimation by default. How this default may unintentionally reduce consideration of what could affect the handling of the missing data with FIML is reviewed. The second section also describes how to go beyond reporting percent missing; specific statistics are discussed that can be reported after multiple imputation or FIML regarding how the handling of the missingness influenced the results. The third section presents different planned missing data designs: multiform surveys, two-method design, and wave-missingness. Each capitalizes on missing completely at random data by randomly assigning participants to their data collection experience and reduces participant burden by reducing the total data collected. The final section illustrates how handling missing data impacts results with an empirical data example including Supplemental Syntax Files.