Covariates that are indicated in the “Include meta-data” and the “Blocking” inputs are both included in the linear model and are “adjusted” for when calculating p-values for the primary variable of interest. The difference is that the former are modeled as fixed effects while the latter are random effects. Fixed effects are typically variables with a known set of values that will have the same range for all future studies, such as sex or age, while random effects can be thought of as a small sub-set of all possible realizations of that variable, with future studies likely to have different sets of values, for example patient ID or sample processing batch. When variables are properly modeled as random effects, that model can be better for making future predictions, however it is more computationally intensive and reduces the statistical power. Since the majority of differential analyses in MetaboAnalyst will not be used for future predictions, we recommend modeling most variables as fixed effects.