Comparison between method penalized quasi- likelihood and Marginal quasi-likelihood in estimating parameters of the multilevel binary model

Abstract

Multilevel models are among the most important models widely used in the application and analysis of data that are characterized by the fact that observations take a hierarchical form, In our research we examined the multilevel logistic regression model (intercept random and slope random model) , here the importance of the research highlights that the usual regression models calculate the total variance of the model and its inability to read variance and variations between levels ,however in the case of multi-level regression models, the calculation of  the total variance is inaccurate and therefore these models calculate the variations for each level of the model, Where the research aims to estimate the parameters of this model using approximation methods (penalized quasi-likelihood and Marginal quasi-likelihood), A simulation method was used to compare the estimation methods for different sample sizes, through Mean squared error (MSE) to get the best method to estimate the parameters, the result obtained using the simulation method showed that the estimation methods gave close result, but the method (MQL) is the best in all sizes .