A number of recent articles have claimed strong relationships--i.e., very high "proportions of shared variance"--between pairs of teaching and learning questionnaires. These claims have been the subject of debate and it has emerged that the proportion of shared variance is defined as the complement of Wilks' lambda. The present article argues that this is not an appropriate measure for the data being analyzed. First some intuitive understandings of the situation are developed. A case study finds that the (Wilks-based) proportion of shared variance between two questionnaires is 79%, but that the first questionnaire explains only 27% of the variance in the second and the second explains only 40% of the variance in the first. The technical literature is then considered and alternatives to the complement of Wilks' lambda are recommended as measures of the relationship between two sets of variables.