CMStatistics 2017: Start Registration
View Submission - CMStatistics
B1624
Title: Robust assessment of cross-over designs against missing values Authors:  Ed Godolphin - University of Surrey (United Kingdom) [presenting]
Peter Godolphin - University of Nottingham (United Kingdom)
Abstract: In scientific experiments where human behaviour or animal response is intrinsically involved, such as clinical trials, there is a strong possibility that some observations will not be recorded. Missing data in a clinical trial has the potential to impact severely on study quality and precision of estimates. In studies which use a cross-over design, even a small number of missing values can lead to the eventual design being disconnected. In this case, some or all of the treatment contrasts under test cannot be estimated and the experiment is compromised since little can be achieved from it. We consider experiments that use a cross-over design. Methods to limit the impact of missing data on study results are explored. It is shown that maximal robustness and perpetual connectivity of the planned design are properties which are related and which guard against the possibility of a disconnected eventual design. A procedure is proposed which assesses planned designs for robustness against missing values and the method is illustrated by assessing several designs that have been considered in the recent literature on cross-over studies.