It is now time to put everything we have learned together into a kind of hybrid type of experimental study. The analysis of covariance is not a study design but an analysis model. It is assumed that you have used one of the standard experimental designs discussed in the previous section to examine treatments effects. The only differences are that while measuring the response of interest you also measured some factor (preexisting condition say) that you suspect has a relationship with the response. This factor, called a covariate ( from: to co-vary, it varies with the response), is not typically available to be used as a blocking factor, or it is simply easier to measure it than to block with it. Regardless, at the end of the study we have for our general linear analysis model a set of dummy variables describing the blocking that was done (if any) , another set of dummy variables for the treatment factors, and this covariate. The trick is to fit this covariate into the analysis model and test hypotheses about it.

This method of analysis is important in a number of situations where blocking is only part of the solution to controlling variability. This model is used quite often in almost all areas of science.