Determining Suitable Options for Random Coefficient Variables

• Activity Number: 76 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Novie Younger-Coleman* and Jennifer Knight-Madded
• Companies: Caribbean Institute for Health Research (formerly TMRI), UWI, Jamaica and CAIHR, University of the West Indies, Jamaica
• Keywords: Random effects; Random Coefficients; Intra-cluster correlation coefficient; clustering; model adequacy

Abstract

In a regression model, the random coefficient has various definitions; definitions include the variable that investigators cannot control; whose values are a random sample from a population; and that can be a fixed effect in one model and a random effect in another. Data from sickle cell disease patients gathered between 1991 and 2001 were analysed to assess age of patient, year of measurement and patient ID, each with a different random effect variable definition, as candidate random coefficient variables in the relationship between oxygen saturation (SaO2), three haematology measurements and history of asthma. Intra-cluster correlation (ICC) coefficients and log-likelihood statistics for null and single explanatory variable models quantified the roles of the clustering variables in determining model adequacy and evidence of clustering in SaO2. The candidate variables each give evidence of clustering, with ICC being highest for ID and lowest for age. The log-likelihood statistics also suggest that the ID variable yields the best-fitting model. The other variables, however, do contribute to clustering in the outcome that should not be ignored in the model building process.