Skip to main content

Archived Comments for: Hand washing with soap and water together with behavioural recommendations prevents infections in common work environment: an open cluster-randomized trial

Back to article

  1. Concern over lack of statistical adjustment for clustering

    Gordon Doig, University of Sydney

    15 December 2014

    I am very concerned that lack of statistical adjustment for clustering has led to a false positive result in this trial.

    The unit of randomization was the work environment. Given the analysis was based on the number of weekly reports, there are at least two important levels of clustering to consider: Each cluster (work environment) contains a number of participants (N) and each participant generates a number reports (n). Reports cannot be considered to be the appropriate unit of analysis. They are not independent of the participant. The simplest example of lack of independence can be demonstrated by an infection that lasts three weeks: If a participant reports an outcome (infection) in week 1, week 2 and week 3 this infectious episode cannot be considered to be three independent episodes. Treatment of reports as the unit of analysis would inflate the number of unique infectious episodes.

    If randomization occurred at the level of the participant, a repeated measures analysis of variance would clearly be required.

    Analysis based on appropriate adjustment for clustering should be presented. The reader will then be able to judge for themselves whether the results presented herein are overzealous.


    Gordon S. Doig


    Competing interests

    No competing interests.