To briefly describe our experiment - a handful of individuals had blood taken, their blood was exposed to different infection combinations and then we measured the immune response at two different time points for each combination. As you would expect, there is large variation between individuals. Also, the time effect is significant since immune response is not instant. However, we have no interest in either the individual effect or the time effect, we are only interested in comparing the immune response to different infection conditions. Additionally, some of the samples for a couple of the individuals were lost so it is not balanced across individuals.
Is it ok to model Individual/Time as the error term – as a sort of nested ‘random block’ for making comparisons within? Code in R that I am using is:
aov(y ~ Infection1*Infection2*Infection3 + Error(Individual/Time), data=response.data)