Multivariate Analysis: Discrete Variables (Loglinear Models)

摘要: Loglinear models are commonly used to analyze relationships involving discrete variables. They include for independence and conditional of polytomous variables, logit prediction binary multinomial response Poisson regression models. On the other hand, loglinear special cases exponential families. A unified treatment may be achieved by description data concerning variables in terms tables frequencies. In a model, each frequency is random variable with finite positive expectation, logarithms expectations frequencies assumed satisfy linear model. The either independent or one more models, exact tests confidence intervals available, although such procedures often difficult apply. Large-sample approaches based on maximum likelihood used, analyses weighted analysis encountered many important cases. Bayesian methods have also been proposed.