Abstract
An attempt is given to the problem of analyzing the two-way binary attribute data using the logistic regression model in order to find a sound statistical methodology. It is demonstrated that the analysis of variance (ANOVA) may not be good enough, especially for the case that the proportion is very low or high. The logistic transformation of proportion data could be a help, but not sound in the statistical sense. Meanwhile, the adoption of generalized least squares (GLS) method entails much to estimate the variance-covariance matrix. On the other hand, the logistic regression methodology provides sound statistical means in estimating related confidence intervals and testing the significance of model parameters. Based on simulated data, the efficiencies of estimates are ensured with a view to demonstrate the usefulness of the methodology.