I was using lmer(REML) from lme4 package to perform variance analysis. The data set consists of 96 observations collected from 16 subjects (ID), 3 treatments (Treatment ) and 2 time points (before/after treatment). There were few data points missing, in total, I have 80 observations. I fit a model as following to estimate the fixed effects of the condition, time and their interaction, plus a random effect for each specific subject. I got the error saying observation is less than the number of random effects, is this normal? I thought with REML the model should have handled the missing values? Do I have to mark the missing data as NA?
fit1 <- lmer(Mean ~ Treatment * Time + (1 + Treatment * Time | ID), data = dat)
Error: number of observations (=80) <= number of random effects (=96) for term (1 + Treatment * Time | ID); the random-effects parameters and the residual variance (or scale parameter) are probably unidentifiable
Another question is do you think this model is correct? I went to lmer with RMEL because I learned it could deal with missing values and I was about to perform a 2-way rmANOVA analysis to show if there are any significant treatment effects.