Analysis of covariance is widely used when comparing post-administration values between groups, such as drug groups, taking into account covariates such as pre-administration values. Traditionally, when applying analysis of covariance, it is assumed that there is no interaction between the covariate and drug group. It is also known that interactions between covariates and drug groups are likely to occur when there is a lower limit to the observed response. In high-precision experiments, even if there is a slight difference in slope between drugs, the interaction is likely to be statistically significant, while in low-precision experiments, even if there is a slight difference in slope, it is unlikely to be statistically significant. When the interaction between the covariate and drug group is included in the model in the "Fit Model" function of JMP, centering is automatically performed only for the covariate, and least squares means for the drug group can also be output. So, how should least squares means be interpreted when there is an interaction? Analysis of covariance can be applied when there is a confounding variable, but obvious interactions are likely to occur. We have gained new knowledge about how to utilize least squares means when centering interactions in JMP.
