ABSTRACT
Linear regression is the simplest statistical procedure that gains from geometric visualization. In turn, bivariate regression is the simplest version of this procedure. Although all visualization in bivariate regression is easily done with a scatterplot in variable space, the subject-space picture is worth examining, particularly the aspects of it that do not change when one introduces multiple predictors. As a bonus, the geometry gives a natural way to develop the regression formulae. Linear regression can be used to solve a problem seemingly quite different from prediction: the assessment of the difference between the means of two groups of scores. Suppose that one wishes to see whether the average value of a variable Y is different in two groups. Although this test is completely satisfactory, an approach to the problem through regression is also valuable, for it introduces some techniques that are needed to analyze more complex designs.
