Strategies for Handling Missing Data in Social Work Research

Abstract: The challenges associated with missing are pervasive throughout social work research and can present even the most seasoned researcher with difficult choices about how to proceed. Handling missing data is not a question of “if”, but of “when” and “how bad”. There are a variety of statistical methods for addressing problems associated with missing data, each with its own set of benefits and challenges. Advancements in software applications, accessibility, and ease of use now provide all researchers with rigorous and efficient methods for maximizing the information available in their data and estimating unbiased results. The purpose of this workshop is to familiarize social work researchers with different strategies available for handling missing data. The workshop will include a review of the different types of missing data, “missing completely at random”, “missing at random”, and “not missing at random” (Allison, 2001; Little, 1988; Little & Rubin, 1987), and techniques for assessing type of missingness. The workshop will include discussion of the most common strategies for handling missing data, including conventional methods such as deletion and replacement, as well as more contemporary and preferable stochastic methods such as maximum likelihood estimation and multiple imputations. Guidelines for describing missing data and missing data analysis for peer-reviewed publication will be reviewed (Schlomer, et al., 2010). A demonstration of techniques for handling missing data will be given using missing data analysis options available in PASW/SPSS 18 applied to real data. Participants are encouraged to bring their own data sets for hands-on application of these techniques with support from the presenter. At the conclusion of the workshop, participants will be able to (1) distinguish between different types of missing data, (2) determine type of missingness, (3) choose appropriate strategies based on the characteristics of the data and the desired outcomes of the analysis, (4) implement a variety of missing data strategies, and (5) interpret/report results involving data sets with missing data.