When there is some form of relationship or correlation between the various factors of which you are measuring aspects. For example, if you are comparing the speed of use of two systems, hoping that your new system is faster, a confounding variable might be user expertise: if the users of one system are more expert than those of the other, this may create a difference in response time that is not due to the system. One way to deal with such a situation would be to create a balanced design, where the expertise of the users of each system is the same. When this is not possible, one can attempt to model the effect of expertise and correct for it.
Used on page 34