

Inquiry about number of observations 

Message/Author 

HuiGuo Liu posted on Wednesday, April 15, 2009  6:40 am



I remember that I have read somewhere about the number of observations needed to do CFA/EFA for categorical data. It sounds like that as long as any number of observations of the 2 by 2 matrix generated by any two of the items/variables is larger than 5, categorical CFA/EFA is suitable. And in that statement, there is also an exmple following to show that even sometimes the whole number of observations is not very large, categorical EFA/CFA could be performed since the required condition is not demanding at all. I cannot remember where I have read that and whether this is a correct statement. Does anyone get any idea about this? 


In general, you need more observations for categorical data than for continuous data. The only way to know how large a sample is needed in any particular situation is to do a Monte Carlo study. See the following reference: Muthén, L.K. & Muthén, B.O. (2002). How to use a Monte Carlo study to decide on sample size and determine power. Structural Equation Modeling, 4, 599620. 

HuiGuo Liu posted on Tuesday, April 21, 2009  2:55 pm



Thanks a lot of the response. I saw the paper at a quick glance,and found that the CFA examples it used were all with continuous outcomes. But my own research is with 20 ordered variables. I just wonder are the test method and conclusion from the examples also applicable to my own research? Many thanks. 


You can do the same thing with ordered categorical variables. 

Back to top 

