How low should an ICC be to justify disaggregating (i.e. ignore clustering) a multilevel/mixed model?

by JElder   Last Updated October 20, 2019 03:19 AM - source

It is conventional wisdom that ICCs can be used to estimate how much clustering/non-independence is occurring in data consisting of two or more levels. ICCs can inform whether a multilevel/mixed model approach is necessary to account for this clustering or nonindependence. Some people suggest that if your ICC is low, you can justify analyzing your data as disaggregated rather than clustered, as it would be in a multilevel model. However, how low is low enough to justify disaggregating and ignoring clustering in such data? I have heard different numbers for what is appropriate or reasonable to ignore clustering and disaggregating. Is there a particular threshold or "magic number" for ICC to justify disaggregating or is it simply case-by-case and it depends on your data?



Related Questions


Intra Class Correlation

Updated May 31, 2017 19:19 PM