Missing data analysis and multiple imputation


Few missing values are universal accompanier of data in every research. However, missing value load exceeding 5% of total data burden of that variable will signal alarm. There are two key types of missing data: a) Missing is unpredictable and random: Non-response in data outcomes is randomly distributed and cannot be attributed to any unobserved or observable factor in the study & b) Missing is systematic and predictable: Here certain cause (observed or unobserved) can be assigned to missing values.

Approaches to deal with missing data involve complete case analysis (deleting subjects with missing values), imputing missing values with series mean or median value, last observation carried forward when follow up values are missing in a repeated measure design study and multiple imputation method using regression modeling approaches.