ABSTRACT
Consider, however, a set of paired data of the sort that might be produced if one were to weigh each of a sample of one hundred men before and after they had taken a fitness course. Previously, our concern would have been with the comparison of the men's average weight before the course with their average weight afterwards. We would certainly expect these averages to be different. A bivariate data set such as this, however, is likely to have another striking characteristic. The person who was heaviest before the course is likely to be among the heaviest in the group afterwards; the lightest person before the course should be among the lightest afterwards; and one with an intermediate score before the course is likely to be in the middle of the group afterwards. In other words, there should be a statistical association or correlation between people's weights before and after taking the course.
