The data of empirical studies often are incomplete: selected persons do not participate, variables are skipped, and items are omitted. The missingness can be Missing Completely at Random (MCAR), Missing at Random (MAR), or Missing Not at Random (MNAR). The type of missingness is studied by constructing missingness variables. These variables also may be of interest of their own. Persons, variables, and items are accidentally or nonaccidentally missing. It is plausible that accidental missingness is MCAR or MAR, but nonaccidental missingness is MNAR. Missingness is counteracted by guidelines to increase the participation rate, re-approaching participants who omitted variables or items, and using the randomized response method to ask sensitive questions. The sample is maintained at its planned size by completing the sample with new participants or oversampling of persons, but these procedures can bias parameter estimates. Naive methods to handle missing variables are listwise and pairwise deletion of participants, and carrying forward a participant’s last observation. These methods can cause bias of parameter estimates. Another naive method is mean imputation of variable and item scores. This method can bias parameters and reduces the variance of these estimates. If participants or variables are MCAR or MAR, it is adequately handled under statistical models. The preferred methods are maximum likelihood missing data and Bayesian multiple imputation methods. If participants or variables are MNAR, a worst-case strategy is recommended. This strategy imputes values that are least favorable to the research hypothesis. Usually, a worst-case strategy is appropriate for missing maximum performance items. The Two-Way with Error (TW-E) method is suitable to impute typical response items of different missingness types.
KeywordsBayesian multiple imputation Maximum likelihood missing data methods Missingness variable Naive missing data methods Randomized response method Two-way with error method Worst-case imputation
- Cronbach, L. J. (1990). Essentials of psychological testing (5th ed.). New York, NY: Harper & Row.Google Scholar
- Huisman, M., & Molenaar, I. W. (2001). Imputation of missing scale data with item response models. In A. Boomsma, M. A. J. van Duijn, & T. A. B. Snijders (Eds.), Essays on item response theory (pp. 221–244). New York, NY: Springer.Google Scholar
- Mellenbergh, G. J. (2011). A conceptual introduction to psychometrics: Development, analysis, and application of psychological and educational tests. The Hague, The Netherlands: Eleven International Publishing.Google Scholar
- Mokken, R. J. (1997). Nonparametric models for dichotomous responses. In W. J. van der Linden & R. K. Hambleton (Eds.), Handbook of modern item response theory (pp. 351–367). New York, NY: Springer.Google Scholar
- Verhelst, N. D., & Glas, C. A. W. (1995). The one-parameter logistic model. In G. H. Fischer & I. W. Molenaar (Eds.), Rasch models: Foundations, recent developments, and applications (pp. 215–237). New York, NY: Springer.Google Scholar