Covariates are continuous and natural, a block is discrete. Blocks need one of each treatment and must be decided before experiment is started. Blocking affects how replicates are assigned to treatments. Thus, once a decision has been made, it cannot later be decided to remove ‘block’ as a term in the analysis (some people would today if it does not explain significant variation). The decision to measure covariates can be made at any point. Many potential covariates can be measured. If the analysis shows that they do not explain significant variation in your response, they can be dropped from the model.
Recent comments