Multiple imputation by scale-wise principal component analysis

Nonresponse in modern studies is a common problem to deal with. Also as studies tend to become larger and larger in scope, missing data analysis techniques need to be able to manage the large scope of data that will be demanded of them. The present study focuses on a technique meant to aid in data reduction for large datasets, while maintaining as much theoretically useful variance as possible. The method involves using a principled approach to Principal Component Analysis, applying it to scales of variables which are theoretically meant to correlate together. These Principal Components were used as the predictors in the imputation to create an efficient and effective imputation model. Results indicate strong support for the method, introducing minimal bias, and less bias than the inclusive imputation. Recommendations for further improvements to the method, as well as limitations are also discussed.