Mixed-effect models - too many random factor categories - how to solve? I'm attempting to specify several random effects in a lme/lmer. In specific, I have conducted an experiment where I have stimulated several test animals with other animals that belong to two groups (treated/untreated stimulus animals - the variable specifying stimulus treatment is called Stimulusgroup). Due to a lack of animals, I have had to re-use some of the stimulus animals. Test animals were used only once with two stimulus animals each - hence I have two values per test animal (Response to Stimulus animal 1 and response to Stimulus animal two). Hence I need to include both TestanimalID and StimulusanimalID.
Therefore my model looks like this:
lme(Response ~ Stimulusgroup,
    random=~1|StimulusanimalID/TestanimalID,
    data=x, method="ML")

However, when running this model I receive the following error code:

number of levels of each grouping factor must be < number of observations

I would like to know if there is a way to still conduct the analysis as the reviewers of my paper have explicitly asked for inclusion of these two random effects in my models. I have heard that one may use cbind in this case but I did not find anything online.  
 A: I think if I got your question right.. you have two sets of animals one used just once and the other used twice . I think you can just use animal as a factor and fit response =(1|animal) in the lmer syntax. You will have to switch  to the lme4 package since, if my memory serves me right, me requires that you have more than one observation per group. Lmer lets you fit observation level random effects. Perhaps @benbolker might jump in and give you some hints here. 
EDIT: After reading the comments below.. this is my understanding of the experimental design ( will edit if it needs refinement). For each test animal there are two observations associated with different stimulus animals. Some of these stimulus animals are re-used.. that is some test animals had encounters with the same stimulus animals. OP originally wanted to fit a nested model with testAnimalID nested within stimulusID. Nested effects can also be considered as maineffect+interaction , so in this case it would be something like (1|stimulusID)+ (1|stimulusID*testAnimalID). The first part will account for correlated observations, however the second part will be equivalent to observation level random effects not covariance among observations from the same testAnimalID. I agree with @rolands comment above that it would be simpler to just fit a random effect for each id i.e (1|stimulusID)+ (1|testAnimalID). 
