Helani Kottage, a PhD candidate of School of Mathematics and Applied Statistics attended the Modern Modeling Methods Conference 2017 at the University of Connecticut from 22nd -25th May, 2017. Her main research supervisor is Dr Carole Birrell and co-supervisor is Associate Professor Marijka Batterham.
She presented a poster at the conference under the topic, ‘Multiple imputation for incomplete categorical variables in multilevel data’. This study compares the performance of four imputation models in imputing missing values in mixed variable types. Four models are:
- Joint multivariate normal imputation via variance covariance model
- Joint multivariate normal imputation via multivariate linear mixed model
- Imputation via fully conditional specification
- Single level joint multivariate normal imputation via variance covariance model (fixed effects imputation)
These models are assessed through a simulation study by examining the outcomes, mean squared errors (MSE) of model coefficients, confidence interval coverage and relative bias.