|Friday, February 22|
|PS2 Poster Session 2 (with refreshments)||
Fri, Feb 22, 4:45 PM - 6:15 PM
Estimating the Correlation Coefficient with Censored Data*Yanming Li, University of Michigan
Keywords: bivariate normal, sensitivity analysis, heavy tails,
Estimating the correlation coefficient, rx,y, with left-, right-, or interval-censored data is not available in standard software. We illustrate a new R package that will produce an innovative plot to visualize correlation for bivariate censored data. It also gives estimates of the correlation coefficient, p-values and confidence intervals. Estimation is based on profile likelihood, assuming a bivariate normal distribution. To assess the sensitivity of the estimation to heavier tails than the normal distribution, we can estimate rx,y assuming a bivariate t-distribution. Left-censored data often arise in the context of toxicology and environmental data where some measurements are below a limit of detection. Right-censored data arise in measuring times to events such as death or disease outcome, which may be censored at last follow-up. Interval-censored data arise from events occurring within a fixed interval, such as HIV conversion between office visits. The R package can accommodate combinations of these censored data cases, as well as missing data for either X or Y.