A Comparison of Interval Estimation Methods for the Concordance Correlation Coefficient

• Activity Number: 299
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Dai Feng*+ and Richard Baumgartner and Vladimir Svetnik
• Companies: Merck and Merck and Merck
• Keywords: concordance correlation coefficient ; jackknife ; delta method ; Bayesian MCMC

Abstract

This study compared state of art methods for confidence interval estimation of the concordance correlation coefficient (CCC). They comprised of the delta method and the jackknife (both with and without z-transformation), as well as the Bayesian MCMC with the Jeffreys and conjugate priors. In a Monte Carlo simulation study, the coverage and precision of the CIs obtained by the methods investigated were examined in various settings with different sample sizes, numbers of raters, mean and variance structures. The performance of the methods was also compared for both normal and non-normal data. In addition, the CIs were calculated for endpoints from an electroencephalography (EEG) sleep study. Results of the simulations showed that the jackknifing with the Z-transformation provided the most accurate coverage probabilities. For normal data, the Bayesian method with the Jeffreys prior also performed well with minimal differences between the two approaches. The simulation results demonstrated that non-normality in the data compromised the performance of all methods.