Advances and Applications in Statistics
Volume 17, Issue 1, Pages 29 - 40
(July 2010)
|
|
LIKELIHOOD INFERENCES FOR CORRELATED NOMINAL DATA WITHOUT KNOWING THE INTRA-CLUSTER CORRELATIONS
Chi-Chuan Yang and Tsung-Shan Tsou
|
Abstract: The multinomial distribution is the main statistical model for categorical data, including ordinal and nominal responses. A prerequisite assumption underlying this model is the independence between the individuals constituting the total number of the multinomial. This article shows that the multinomial model could be made asymptotically valid no matter how individuals are correlated. We focus on likelihood inference for the regression parameter of interest associated with the polytomous logistic regression model. Simulations and several examples are used to demonstrate the efficacy of proposed parametric robust method. |
Keywords and phrases: multinomial, robust likelihood, polytomous logistic regression. |
|
Number of Downloads: 339 | Number of Views: 1204 |
|