Handling Item-Level Missing Data in Linear Regression: A Tutorial

What the paper says

With advances in methodology and statistical software, modern methods for handling missing data have become more accessible and straightforward to apply. In psychological studies, researchers often use questionnaires or scales composed of multiple items to measure constructs of interest. As a result, missing values frequently occur at the item level, whereas data analyses are typically conducted at the scale (composite) level. However, properly addressing item-level missing data remains a common challenge for many applied psychologists, including researchers who are otherwise well experienced in handling missing data at the scale level. In this tutorial, we introduce six approaches for handling item-level missing data: listwise deletion, hybrid methods that include proration with listwise deletion and proration with full-information maximum likelihood, item-level full-information maximum likelihood, item-level multiple imputation, two-stage maximum likelihood, and composite score factored regression. Using a published empirical data set, we provide step-by-step guidance on applying these methods in linear regression models. We include R code for each method and corresponding Mplus syntax if applicable. Finally, we summarize the key assumptions, advantages, and limitations of each approach and offer practical recommendations for researchers.