Techniques of data weighting and missing value imputation

Lee ML, Rubenstein LV, Yano EM. Techniques of data weighting and missing value imputation. Paper presented at: VA HSR&D National Meeting; 2001 Feb 15; Washington, DC.

Related HSR&D Project(s)

• CPG 97-002 – Implementing Guidelines for Smoking Cessation: A Randomized Trial of Evidence-Based Quality Improvement

Workshop Objectives: Many health services research studies are reliant on primary data collection through the use of self- or interviewer-administered surveys of patients or providers. HSRandD investigators have become accustomed to having to describe the reliability and validity of measures used in health and health care surveys, but are frequently less prepared to address the equally important issues of bias emanating from patterns of data loss through refusals, losses to follow-up and incomplete responses (i.e., missing data). Such losses are particularly problematic for randomized trials, where power to detect differences, especially among subgroups, relies on adequacy of the ultimate sample size. The purpose of this workshop is therefore to: (a) Review data weighting techniques that adjust for patterns of refusal and losses to follow-up, and (b) Examine the strengths and weaknesses of alternate methods of data imputation for missing values among subjects that have remained in the study sample. Workshop Activities: A multidisciplinary panel of methodologists (statistician, epidemiologist, and clinician health services researcher) will present the basic ideas behind weighting and imputation techniques, presenting relatively straightforward solutions, and describing available software and illustrations of how to implement these solutions. Case studies will then be presented, further illustrating the situations that lead to the problems of loss of subjects and/or parts of their data, along with the approaches used to deal with these problems. The strengths and limitations of the various techniques will also be considered, enabling participants to intelligently apply these techniques on survey datasets at their home institutions with knowledge of the potentially serious implications of ignoring data losses. Target Audience: Health services researchers interested in adding data weighting and imputation techniques to their methodological toolkits and those generally interested in the issues involved in missing data patterns and the implications of such patterns on the ultimate utility of their survey and intervention data. Audience Familiarity: The workshop is designed for the broad health services research audience, with low to moderate levels of statistical understanding and familiarity needed.