Mary Edmond posted on Saturday, March 13, 2010 - 1:57 pm
In terms of validity, how small is too small for a cluster? I am proforming a LPA on 3 different samples, which have an N of about 500, 300, and 250 respectivley. Some of by clusters are very small, less than 10 percent of the total sample. Is that an acceptable size? The Bics indicate a well fitting model, but does anyone know of a rule of thumb or a test of validity concerning this issue? Thank you!!
I think you are asking about the size of classes found in an LPA. I don't know of any minimum class size. There are several things to consider. Is the class meaningful? Is the class well-defined from other classes? Does it have different predictors and relate differently to distal outcome(s)? Does the class have more observations than parameters being estimated for that class?
Thank you so much for the response! I had also received two articles that were helpful to me: (Marsh, Lüdtke, Trautwein & Morin 2009) and (Wade, Crosby, & Martin 2006). These articles seem to give similar advice as you have given. I appreciate your input very much!