Repeated Measures - Maximum number of variables

[Nives_ES]

I know that the general rule of thumb is that you need 10 observations or 10 cases per variable in your model. How about in a repeated measures model?
Do you count the clustered data as one observation or would you count each measurement as an observation in determining the maximum number of variables to use in your model?

 

[hlsmith]

Is your outcome continuous or categorical? What procedure are you planning on using for analyses.


This is a good question. I do not have an answer for you and I have thought about it before. I can see how it could seem like you now have more variable or components to control for, however, in the past it seems like I have heard people say you do need more observations. I think the more clusters you have the better in this area you are as well. Though, I am not basing any of these comments on citatble literature, just my faulty memory on things I thought I had heard.

 

[GretaGarbo]

I know that the general rule of thumb is that you need 10 observations or 10 cases per variable

That rule of thumb is simply not true!

In some cases like in a designed experiment, you can have 7 factors (i.e. 7 variables) in 8 experiments (observations). In other cases you can need thousands of observations for just a few variables (that are highly co-linear).

I believe that the incorrect rule of thumb comes from someone that did a simulation study based on one specific data set, with one specific pattern of co-linearity (i.e. correlation among the x-variables).

 

[Nives_ES]

I am dealing with both a continuous and a categorical outcome. I was planning to use proc mixed and proc genmod. Also, I will see if I can look up the simulation study! That sounds interesting.