"Sampling zero-cell" in Generalised Estimating Equations (GEE)

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
Report Content as Inappropriate

"Sampling zero-cell" in Generalised Estimating Equations (GEE)


I am analysing family data using a generalised estimating equations (GEE) approach. Since my dependent variable is dichotomous I use a binary logistic regression as model type. My problem now is that I have one empty cell (a "sampling zero-cell") which affects/stops my analysis. However, I have the assumption that this 'observation' is actually very interesting and I would like to find a way to handle this problem, keep the category and proceed with the analysis. Agresti (1996) suggests adding a very small constant to the cell count to determine the effect. Unfortunately, I do not know how to do that in SPSS. Is there a way in SPSS to "eliminate" the zero-cell and conduct a GEE logistic regression? Or is there another procedure I could apply to analyse corrleated observations using a logistic regression when a sampling zero is present?

I would be greatful for any help!