CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Investigating a time-varying degrees of freedom parameter in the robust modelling of longitudinal data Authors:  Melanie Campbell - Queen's University Belfast (United Kingdom) [presenting]
Abstract: Longitudinal data is commonly found in a medical setting, for example, monitoring patients following the progression of disease or response to treatment. Linear mixed-effects models are one of the most popular models for representing longitudinal data. The model contains both fixed and random effects allowing for the within-subject correlation to be accounted for. Regrettably, the standard model is sensitive to outliers due to the fact that both the random errors and random effects are assumed to be normally distributed. The relatively untouched area of robust mixed modelling attempts to deal with this problem to minimise the risk of biased results when outliers are not accounted for. The robust models replace the typical Gaussian assumptions with t-distributional assumptions. Most work to date has both the random error and random effects modelled with a single degrees-of-freedom parameter. Yet it has been recognised that following treatment, for example, or other contributing factors, an outlying individual or the trend of an individual's measurement may conform to the population trends over time or conversely become outlying. This is the motivation behind a time-varying approach. Recent work has utilised splines to model the variation of outliers over time. The purpose is to expand more on these concepts and investigate the need for time-varying degrees of freedom parameter within a robust mixed model context.