Figure 3

Scatterplot matrix between all pairs of variables: age (years), CVLT free recall long delay (number of correct items), intracranial volume (ICV) in mm ³ , hippocampus volume (ICV-normalized), total volume of cerbral cortex (ICV-normalized), and total volume of lateral ventricles (ICV-normalized). On the diagonal panels estimated probability density are given for each variable, together with measurement value for each observation, the latter in the form of vertical lines along the x-axis. The numerical range for each variable is given along both the horizontal and the vertical borders of the matrix plot. The off-diagnal panels allow for each variable to be compared to any other variable, interchanging the ordinate and the abscissa. For each of these bivariate scatterplots a least square linear regression line is fitted. Elliptic data-concentration contours for the fitted bivariate normal distribution are also plotted. For highly correlated data the elliptic shape is elongated, and for uncorrelated data the shape is circular. The contours are plotted at levels 0.5 and 0.9, i.e. 50% of the data is within the inner ellipse, and 90% within the outer one. (The figure was produced by the car package in R.)