I am running a generalized linear model on a dataset with 19 individuals and have 4 variables of interest. There are furthermore a number of interactions that might be interesting to look at. I was wondering if there is a general rule of thumb (with reference please) about the nr of terms you can use in a model based on the sample size. Thanks.
Is there a general rule about max nr of variables to use in (generalized) linear model?
There's Tukey's suggestion of a minimum of 5 observations per mean parameter (he also suggested 25 observations per variance or covariance parameter). I don't recall the exact location of that suggestion, sorry.
But it really depends on the accuracy you want. If you want to be able to get a reasonable idea of actual effects sizes, this suggests that for a logistic regression something more like 50 observations per parameter might be more in the ballpark.
This has sample-size-related rules of thumb; if you can specify the required information you might be able to figure out what you need.
3$\begingroup$ Harrell has similarish ROTs (with citations): books.google.com/… $\endgroup$– dimitriyJan 30, 2013 at 1:43