Kjetil Halvorsen Løland
2009-Mar-16 21:52 UTC
[R] Logistic regression and mixed effects / hierarchial structure
Hi, I’ve tried to find a general approach to my problem without any success, although this might very well be due to my inexperience with R help resources (and statistics in general). My general problem is a straightforward 2 by 2 table (“Belonging to the upper quartile” vs “not-belonging to the upper quartile”, intervention vs non-intervention), but with a random effect addressing the hierarchical structure of the individual data points. The different data points are pseudoreplicated, some hailing from the same patients (some patients have contributed with 1, while others as much as 7 – i.e. not constant). Naturally I would like to adjust for this clustering within patients, but I have failed to see which model/approach that would be correct. Is there some way to do a logistic regression with only a binary explanatory variable and a added random effect adjusting for data entry nesting within patients, preferentially giving an odds ratio with a confidence interval? I’ve tried to use the lme function in the nlme package, but I feel this would be stretching it a bit. I’m terribly sorry if this problem is to obvious, but I would appreciate any feedback and pointers in the right direction. Sincerely, Kjetil Checked by AVG. 07:04 [[alternative HTML version deleted]]