Random effects clustering in multilevel modeling: choosing a proper partition