My understanding is that to carry out a multilevel (model, a rule of thumb is that around 12 level 2 units, or groups, is needed.

Is this a helpful rule of thumb, without additional information? What is this based on? What considerations, technical and otherwise, should be considered?

I searched, and this doesn't seem to have been asked as a question, at least as the focus.

Note. The data I'm working to analyze is from around 200 students grouped into one of 10 programs. The number of students in each group is about equal (around 20 per program).

  • $\begingroup$ What are your concerns/"expected safety issues" against using a multilevel model? $\endgroup$
    – IWS
    Commented Jan 24, 2017 at 12:38
  • $\begingroup$ I guess I don't have any. It's about going against the vaunted rule of thumb. $\endgroup$ Commented Jan 24, 2017 at 12:39
  • $\begingroup$ Edited the question. $\endgroup$ Commented Jan 24, 2017 at 12:59
  • 1
    $\begingroup$ You may find Maas and Hox (2005) (joophox.net/publist/methodology05.pdf) helpful. $\endgroup$
    – user36400
    Commented Jan 24, 2017 at 14:46
  • $\begingroup$ The common rule is 30 units at the highest level of analyses, but this depends on the complexity of your model see: academic.oup.com/esr/article/32/1/3/2404247 $\endgroup$
    – user264415
    Commented Nov 1, 2019 at 10:04

1 Answer 1


There are a number of reasons why having 10 units in level 2 may not be advisable, including bias. However, “rules of thumb such as only doing multilevel modeling with 15 or 30 or 50 level 2 units can be found and are often personal opinions based on personal experience and varying reasons e.g. getting a non zero variance, being able to check the normality assumption etc.” - Center for Multilevel Modelling.

William Browne’s ppt found in CMM’s site, goes over the sample size question more at length.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.