 |
 |
> Thinking of community based interventions always raises the question > of statistical power for the findings (or no findings) of the study. > These sorts of interventions are naturally situated within community > 'clusters' (hospital, neighbourhood, town etc). The best available > evidence on the analysis of clustered RCTs - that I am aware of - > recommends that randomisation and analysis both should be conducted at > the cluster level. This isn't particularly statistically efficient though, and there are techniques (multilevel modelling or one of its synonyms) for using both group and individual level data. This may (or may not) help with the power problems of there not being enough 'clusters' in the country/world to randomise, although I guess some/many might criticise it as allowing too much (subjective?) researcher intervention/involvement to construct the models. A relatively non technical introduction is Campbell MK et al. Analysis of cluster randomised trials in primary care: a practical approach. Family Practice 2000, 17(2), 192-196. I have no expertise in, or direct experience of cluster randomised RCTs (although I have used multilevel modelling with observational data) so I don't know how generally accepted these ideas are by people who do this kind of RCTs. Arguably, this kind of technique is one way to put back in the social context that RCTs usually strip out or don't report. Bruce Bruce Guthrie, Department of General Practice, University of Edinburgh, 20 West Richmond Street, Edinburgh EH8 9DX Tel 0131 650 9237 Fax 0131 650 9119 e-mail [log in to unmask]