Hi everyone,
I am currently working on a project with the following sample:
26458 individuals (level-1) nested within 16458 households (level-2) within 12 countries (level-3). Thus, many of my households (around 40%) are singleton (just one individual per household). I need to fit a multilevel logistic model, with roughly 5 predictors at level-1, 2 at level-2, and 4 at level-3.
I know that in these types of situation the first and second order quasi-likelihood procedures perform poorly whereas the second order penalised quasi-likelihood method would be a better one to use. To my understanding, I should also compare the PQL2 results with findings obtained from alternative estimation procedures (such as MCMC).
My problem is that –when i consider the household level –the model does not converge, even when i try to build the null model. I also tried to use the parameter estimates from the MQL 1model as starting values for MQL2; still no convergence. I get either 1.#QO(000) or I get a window saying that there were numeric errors replaced with missing values.
If i ignore that level (and only consider individuals and countries), i have no problems/errors.
Does anyone know what is going on? Am i doing something wrong? Any help with this issue would be highly appreciated!
Kind regards,
Giorgio