Separating interviewer and area effects by using a cross-classified multilevel logistic model: simulation findings and implications for survey designs