In Section 2.4, we conducted an analysis of covariance adjusting for the baseline data to estimate the treatment effect where we observed a high positive correlation coefficient ρ = 0.782 between the baseline measurement and the measurement at week 3 (Figure 2.4). Similar positive correlation coefficients were also observed for the log-transformed serum levels of glutamate pyruvate transaminase (GPT) for each of treatment groups between the baseline measurement and the measurement at week 4 (Figure 4.1) in a randomized controlled trial with a 1:4 repeated measures design for chronic hepatitis patients to investigate whether the anti-hepatitis drug Gritiron tablets improve liver function abnormality1 (Yano et al., 1989). In the previous chapter, the Rat Data was analyzed using the SAS procedure PROC MIXED, which allowed you to choose the good-fit covariance model from many candidate covariance models. However, this type of ad hoc analysis based on the best-fit covariance without rhyme or reason will not tell us why and how these positive correlation coefficients arise while repeated measures were usually taken independently for each patient. Furthermore, to estimate the treatment effect in the 1:1 pre-post design, an analysis of covariance is usually carried out to adjust for baseline data. But why?