Strategies and Guidelines for Handling Missing Data in Social Work Research

Background & Purpose. Quantitative analyses are often afflicted by a worrying fact: the presence of missing data (MD) on some observations. Statistically, the presence of MD threatens assumptions about the completeness of data, potentially jeopardizing statistical power and validity. Dealing with MD is complex and frequently one of the most challenging components of an analysis for many investigators. Moreover, beyond statistical implications, the presence of such data is of particular substantive importance given that (a) missing observations are frequently participants in a study who are marginalized and (b) analyses that do not consider the implications of missing data can produce to misleading results that suggest implications which may reinforce dominant systems that fail to study accurately underserved populations. This is of obvious and particular importance to social work research. Overall, despite the widespread awareness of the problems of MD, there remains a great need in social work clear and concise overview of strategies to detect and handle MD.

Content. This workshop aims to address these needs via an extensive overview of MD in three main parts. First, MD will be placed within broader contexts of theory development, research question formulation, hypothesis testing, and causal inference. Particular attention will be given to (a) Rubin's seminal classifications of MD types and (b) the grave substantive and methodological consequences of unconsidered MD. Participants will be able to see first hand with real data examples the changes to parameter estimates and standard errors that can occur when researchers are not sensitive to MD. Second, strategies for handling MD will be discussed in significant detail. Participants will first be shown strategies to detect both the presence and significance of MD in their analyses. Then, traditional strategies such as item deletion, listwise deletion, and single imputation will be covered with careful consideration being given to their strengths and limitations. Following this, newer and likely superior strategies will be covered in depth. Strategies covered will include multiple imputation and maximum likelihood estimation. This section will conclude with the presentation of an evidence-based, delineated, and organized set of guidelines for handling MD. Lessons from the work of key scholars (e.g., Allison, Graham, Little) will be highlighted. Third, the workshop will consider statistical software, resources, and key references with participants being provided numerous materials for future use. Throughout, the workshop will allow space for comments and interactions, and the pedagogical approach will rely heavily on figures and illustrations for ease of interpretation.

Conclusion & Implications. Overall, this type of overview of MD can be invaluable for SSWR scholars and students who are at the beginning stages of their understanding of statistical models. There is need for an overview of MD that pairs methodological rigor with a focus on practicality. By becoming familiar with MD best practices, attendees can gain the expertise necessary to complete analyses that compete with the highest-quality research in other fields. As such, it fills a need in SSWR's Research Design and Measurement cluster, especially given that recent few recent SSWR workshops have focused on MD.