In cluster randomisation sampling, I know that one loses precision if clusters become too large. But is there a limit on how small can a cluster be?
I have previously heard that you can have as few as 12-13 per cluster. But can you drop it down to 4-5 per cluster, and have more clusters in the sample.
Appreciate if you can describe the methodology or rationale for arriving at your conclusion. Thanks
For general orientation on this - see
http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#datadesign
You can use the following software to do the power calculations
http://www.bristol.ac.uk/cmm/software/mlpowsim/
the manual is a very good primer
and
Optimal Design for Longitudinal and Multilevel Research Documentation for the “Optimal Design” Software
http://hlmsoft.net/od/od-manual-20111016-v300.pdf
The analysis should be undertaken by a methodology that respects that individuals are nested in clusters such as random effects multilevel analysis - that technology deals with imbalance and can handle the situation when there is only 2 individuals per cluster ( the same type of modelling can analyse twin data). They key then is the target of inference - with small number of subjects within a cluster you will not be able to say much about a particular cluster but if you have lots of clusters, you will be able say something about between cluster differences.
I have found this book to be very useful in terms of practical matters
Cluster Randomised Trials, Second Edition
Richard J. Hayes, Lawrence H. Moulton (2017)
An excellent article on this
https://academic.oup.com/ije/article/35/5/1292/762170
- Badges
- Science topic
- Similar topics
- Correction