SOURCE:

Faculty: Physical Sciences
Department: Statistics

CONTRIBUTORS:

Okoli, C. N.
Onyeagu, S.I.

ABSTRACT:

Log linear models are statistical tools for the analysis of categorical data. The problem of estimating the parameters and best fits of log linear models for q-dimensional contingency table as well as procedure for testing the significance of interaction of variables which involves many parameter models has been identified and considered in this research work. We proposed new method of estimation of parameters of log linear models for q-dimensional contingency table. More so, we developed algorithms on the new method for estimation of the parameters and best model fits of log linear model for q-dimensional contingency table. In estimating these parameters and best model fits, a computer program in R was written for the implementation of the algorithms. Furthermore, the Iterative proportional fitting procedure algorithm is used to estimate the parameters and best fits of models of the log linear model. The proposed method compared favorably with the existing methods, and yielded higher p-values and least likelihood ratio estimates than the existing methods. The proposed method was also implemented on a computer program in R using real life data and the results showed the best model fits for three, four and five dimensional contingency tables as with their p-values 0.84, 0.96,0.86 and likelihood ratio estimates 0.83, 3.79 , 13.40 respectively. The models were subjected to the test for the significance of interaction of the variables using proposed approach. The proposed method involves fewer parameter models. This consequently makes the proposed method faster than the conventional Brown’s method. Hence, the proposed method requires lesser parameter models, saves computer memory and time than Brown’s method.