Recorded presentation

G. Gordon Brown, PhD and Elvessa Aragon-Logan, MS. Statistics Research Division, RTI International, 3040 Cornwallis Road, Research Triangle Park, NC 27709-2194, 919-485-5647, ggbrown@rti.org

Our presentation will focus on the application of methods for calculating robust variance estimates for cluster-correlated survival data from surveys, observational studies or experiments. Failure to account for cluster-correlation can result in poor estimates of variance, which lead to poor interval estimates and hypothesis tests. Typically, variances are underestimated and hypothesis tests tend to be too liberal (high false-positive rate) in the presence of cluster-correlation. Robust variance estimates are obtained using one of three procedures: Taylorized deviations, Jackknife, or Balanced Repeated Replication (BRR). These methods are applied to the Kaplan-Meier estimates for different grouping variables in the data, and to the regression coefficients and adjusted treatment means obtained using Cox’s proportional hazards model. Our presentation will focus on the interpretation of results from analysis of cluster-correlated survival data and on comparison of our robust variance estimates to non-robust variance estimates. We will illustrate the advantages of correctly adjusting variances for cluster-correlation, and we will utilize procedures available in the latest release of the statistical software package SUDAAN®.

Learning Objectives:

• Our primary objective is to inform participants about robust variance estimation methods for cluster-correlated survival data. Participants will gain an appreciation for the impact of cluster-correlation on variance estimates and the associated confidence intervals and hypothesis tests. Participants will learn to recognize situations where failure to adjust for cluster-correlation would result in erroneous inference, and be able to apply our methodology to correct for cluster-correlation. Finally, participants will be able to assess the use of our methods versus traditional methods of variance estimation, and be able to use SUDAAN’s SURVIVAL procedure to perform the analyses. The use of our methodologies will improve the participant’s ability to accurately interpret the results derived from survey data.

Keywords: Survey, Statistics

Related Web page: www.rti.org/sudaan/

Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: We will discuss and illustrate the use SUDAAN software, a licensed product of RTI International.
I have a significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
Relationship: SUDAAN® is a product of my employer, RTI International.

Recorded presentation