A Note on Log-Linear Regression
Dale Heien · Journal of the American Statistical Association · 1968
This paper presents a reparameterization of the conventional loglinear regression model. This reparameterization is shown to be more reasonable than the conventional specification. The stochastic assumptions of this reparameterized model are derived from the expected value properties of the dependent variable. Comparison with the conventional method of fitting log-linear models shows the regression intercept and the variance of the regression intercept to be different. A minimum variance unbiased estimate and another consistent estimate of the intercept are derived.