A0216
Title: The impact of (un)congenial multiple imputation approaches on GPAbin biplots
Authors: Johane Nienkemper-Swanepoel - Stellenbosch University (South Africa) [presenting]
Niel Le Roux - Stellenbosch University (South Africa)
Sugnet Lubbe - Stellenbosch University (South Africa)
Abstract: Multiple imputation is considered a superior technique for handling missing data. This approach results in multiple completed data sets, which are analysed separately by means of standard complete data techniques. Estimates from the separate analyses are combined using suitable combining rules, referred to as Rubin's rules. In the context of exploratory analysis, GPAbin biplots enable the unified visualisation of the individual plots constructed from multiple imputed data sets. This visualisation approach combines configurations by means of generalised orthogonal Procrustes analysis (GPA) followed by the application of Rubin's rules (-bin) on the aligned configurations. In the context of multivariate categorical data, the configurations are multiple correspondence analysis (MCA) biplots. Multiple imputation with multiple correspondence analysis (MIMCA) is therefore regarded as a suitable benchmark approach for other imputation methods, due to the congeniality between the imputation and analysis models. MIMCA is a joint modelling multiple imputation approach, since the same imputation model is used for all variables. In a previous study, the performance of the GPAbin biplots after MIMCA has been evaluated in an extensive simulation study. Now, the performance of GPAbin under joint modelling and fully conditional specification imputation approaches is compared and discussed. Through simulation, the choice of multiple imputation approach is investigated.
