Plausible multiple imputation and consolidate matched : quality of life square tables with missing in two rounds

Abstract EN

The aim of this paper is to introduce a procedure that integrates the performances of the imputer and the analyst.

Estimates produced under the existence of missing data of certain mechanism mostly generate biases and loss of precision.

This procedure is illustrated using a real life example in which we compare the effectiveness of patients' self-management (psm) of anticoagulation versus the standard care. A general framework dealing with the quality of life (qol) data of ordinal response type is presented using items collected at the base and end points of a patient's questionnaire.

Multiple imputations (mi) are conducted as A part of the editing process to deal with missing data in two rounds, i.e., before and after, that is common in such studies.

We apply a hierarchical log-linear model to describe the non-responses mechanism for the item under consideration and multiply impute 10 sets of plausible data sets.

The generated data sets can then be used by the data analyst to consolidate a matched pair's model for potential investigation of symmetry, conditional symmetry and quasi symmetry in a square table or any other statistical analyses. This paper illustrates, stage-by-stage how the procedure helps with selecting the most proper model for the non-response mechanism, and solves issues regarding.