Step 02 · Densities & Matrices

Joint Density Surfaces in Two Variables

One quantity is a curve; two quantities are a landscape. The joint density is a surface floating above a flat plane, and the volume beneath it is the probability of everything.

A surface above the plane

Place one quantity along the horizontal direction and the second along the vertical direction. Above each point of that plane, raise a height equal to the density at that pair of values. The result is a smooth hill, highest where the pair is most likely and falling toward zero far from that region. Probability is read as volume: the chance that both quantities fall inside a chosen region equals the volume of the surface above that region.

Because a volume is involved, the joint density is never read at a single point in the naive way. A tall spike at one pair of values does not mean that exact pair is likely; it means the density is high there. The reader who keeps the volume picture in mind will never make that mistake twice.

Contours as a map

Cut the surface with a horizontal plane and the outline left behind is a contour. A stack of such cuts, drawn on the flat plane, gives a contour map of the joint density. Contours that are tight circles say the two quantities vary together in a round, independent-looking way. Contours that are stretched ellipses say the two quantities move in step, and the tilt of the ellipses says in which direction. The contour map is the most useful single view of a bivariate density.

Tilted elliptical contours: the ridge points up and to the right, which is the visual signature of positive association.

Marginal and conditional cuts

Two reductions of the surface are fundamental. Collapse the surface onto one axis by summing over the other, and you obtain the marginal density of the first quantity alone. Slice the surface at a fixed value of the second quantity, then rescale the slice so its area becomes one, and you obtain a conditional density. These two operations turn one hard picture into two easy ones, and the whole of the next papers rests on performing them cleanly.

Independence as a product form

When the joint density can be written as the product of one function of the first quantity and one function of the second, the two are independent. Visually, independence appears as contours whose axes line up with the coordinate directions, with no tilt at all. Dependence appears as a tilt, a bend, or a change in the shape of the contours as you move across the plane. Reading a contour map, then, is mostly the art of noticing tilt and bend.

A lecture hall board showing axes, plotted points and a shaded region — Two axes and a shaded region: the geometry of a joint density, drawn large enough for a whole room to read.

Why volume, not height

Readers new to joint densities often look for the most likely pair of values and expect to find it at the tallest point of the surface. This habit is natural and wrong. For a continuous pair of quantities, no single pair carries positive probability; the surface gives a density, and only a region with extent can carry a chance. The tallest column is the place where a small neighbourhood holds the most probability, which is a statement about regions and not about points.

Keeping the volume picture in mind also explains why rescaling a joint density is such a delicate step. Stretch one axis and the surface changes its shape while the total volume stays one, because the density must compensate for the stretch. Every operation on the surface is really an operation on volumes, and the reader who remembers this will never misread a spike as a point of high chance.