Joint Densities and Their Probability Meaning

Why a Joint Density Is Not a Point Probability

For two continuous random variables, probabilities usually concern regions of possible pairs of values, such as \(X+Y>1\) or \(X\) lying in one interval while \(Y\) lies in another. A joint density describes how probability is distributed across the plane. It is not, by itself, the probability of observing any particular pair.

The distinction matters in calculations: probabilities are obtained by integrating the density over the event region. A density can be greater than 1, and a point can have positive density while its probability remains zero. The relevant quantity is the density accumulated over an area, not its value at one coordinate.

Definition: A pair of random variables \((X,Y)\) has a joint probability density function \(f_{X,Y}(x,y)\), or joint density, if \(f_{X,Y}(x,y)\geq 0\), its integral over the plane is 1, and probabilities are calculated by integrating it over the corresponding regions. The support is the set of points where the density may be positive. The density is zero outside the support.

If \(R\) is a measurable region in the plane, the probability that \((X,Y)\) falls in that region is

P\bigl((X,Y)\in R\bigr)=\iint_R f_{X,Y}(x,y)\,dA

Here \(dA\) denotes an area element, commonly written \(dx\,dy\) when the region is described using vertical or horizontal slices. The limits of integration must cover the part of the support that lies in \(R\). Any part of the region outside the support contributes zero.

Probability Accumulates Across Area

The integral formula follows directly from the definition of a density. The indicator \(1_R(X,Y)\) equals 1 when the random pair lies in \(R\) and 0 otherwise. Its expected value is the probability of that event. For a pair with joint density, the expectation is the integral of the indicator times the density over the plane:

\begin{aligned} P\bigl((X,Y)\in R\bigr) &=E[1_R(X,Y)]\\ &=\iint_{\mathbb{R}^2}1_R(x,y)f_{X,Y}(x,y)\,dx\,dy\\ &=\iint_R f_{X,Y}(x,y)\,dx\,dy. \end{aligned}

The final equality holds because the indicator is zero outside \(R\) and one inside it. This also explains why a density value alone is not a probability: integration combines the density with the area over which it applies. If the density is constant at \(c\) on a region of area \(A\), the probability of that region is \(cA\), not \(c\).

For a continuous pair with a joint density, an individual point has area zero. Therefore,

P(X=x_0,\ Y=y_0)=\iint_{\{(x_0,y_0)\}}f_{X,Y}(x,y)\,dx\,dy=0.

More generally, a measurable set with area zero has probability zero under a joint density. This includes ordinary curves and line segments. A density can be positive on a curve, but integrating over the curve still gives probability zero. This statement relies on the joint distribution having a density; it need not hold for a distribution that assigns probability directly to a line or point.

Worked Example: A Corner Region in a Uniform Square

A hypothetical pair of measurements \((X,Y)\) has joint density \(f_{X,Y}(x,y)=1\) for \(0<x<1\) and \(0<y<1\), and zero elsewhere. Find \(P(X+Y>3/2)\), and find the probability that \(X+Y=3/2\). The model is a uniform joint distribution on the unit square.

Method. Integrate the density over the event region, then treat the equality event as a line. The stated density is valid: the support has area 1, so its total integral is \(1\cdot1=1\).

Within the square, \(X+Y>3/2\) is the small triangle above the line \(y=3/2-x\). It is present for \(1/2<x<1\), and at each such \(x\), \(3/2-x<y<1\). Thus

\begin{aligned} P(X+Y>3/2) &=\int_{1/2}^{1}\int_{3/2-x}^{1}1\,dy\,dx\\ &=\int_{1/2}^{1}\bigl(1-(3/2-x)\bigr)\,dx\\ &=\int_{1/2}^{1}(x-1/2)\,dx\\ &=\left[\frac{(x-1/2)^2}{2}\right]_{1/2}^{1}\\ &=\frac{(1/2)^2}{2}=\frac18=0.1250. \end{aligned}

The equality \(X+Y=3/2\) describes a line segment, which has zero area. Therefore \(P(X+Y=3/2)=0\), even though the density is 1 along that segment. Check: the triangle has legs of length \(1/2\), so its area is \(\tfrac12(1/2)(1/2)=1/8\), matching the integral.

Common mistake avoided: using the density value 1 as the event probability. The event probability is the density integrated over the triangle, not the density at any one point.

Reading the Density and the Region Together

A joint density and its support must be read together. The support tells which pairs are possible under the model; the event determines which of those pairs to include. When the density is constant, probability is the constant density multiplied by the area of the included region. When the density varies, the integral weights different parts of the region differently.

The density is probability per unit area. If \(X\) and \(Y\) are measured in different units, such as hours and dollars, the density has units of inverse hours times inverse dollars. A numerical density greater than 1 is therefore not inherently problematic. For example, a high density on a sufficiently small support can still integrate to 1.

Worked Example: A Narrow Band Inside a Triangular Support

A hypothetical pair \((X,Y)\) has density \(f_{X,Y}(x,y)=2\) when \(0<x<y<1\), and zero elsewhere. Find the probability that the gap satisfies \(0<Y-X<1/4\), and find \(P(X=0.4,\ Y=0.6)\).

Method. Integrate over the part of the triangular support where the gap is less than \(1/4\). First verify that the given model is a probability density:

\int_0^1\int_0^y 2\,dx\,dy =\int_0^1 2y\,dy =\left[y^2\right]_0^1 =1.

For \(0<y<1/4\), every \(x\) between 0 and \(y\) gives a gap \(y-x<1/4\). For \(1/4\leq y<1\), the condition requires \(y-1/4<x<y\). The event probability is therefore

\begin{aligned} P(0<Y-X<1/4) &=\int_0^{1/4}\int_0^y 2\,dx\,dy +\int_{1/4}^{1}\int_{y-1/4}^{y}2\,dx\,dy\\ &=\int_0^{1/4}2y\,dy +\int_{1/4}^{1}2(1/4)\,dy\\ &=\left[y^2\right]_0^{1/4} +\frac12\left(1-\frac14\right)\\ &=\frac1{16}+\frac38 =\frac7{16} =0.4375. \end{aligned}

The point \((0.4,0.6)\) is inside the support, and its density value is \(f_{X,Y}(0.4,0.6)=2\). But the probability of that exact point is zero because a point has zero area: \(P(X=0.4,\ Y=0.6)=0\). Check: the probability \(7/16\) is between 0 and 1, and the band occupies less than the entire triangular support.

Common mistake avoided: reporting \(2\) as the probability of the point or of the band. The value 2 is a density; only its integral over an event gives a probability.

Variable Density: Integrate the Weight, Not Just the Area

When the density changes over the support, a region's area alone does not determine its probability. Locations where the density is larger contribute more probability per unit area. The next example uses a density that increases across the unit square, so a geometric area calculation by itself would not be enough.

Worked Example: A Varying Density Over a Square

A hypothetical pair \((X,Y)\) has density \(f_{X,Y}(x,y)=x+y\) for \(0<x<1\) and \(0<y<1\), and zero elsewhere. Find \(P(X+Y<1)\) and the probability of the point \((1/4,1/2)\).

Method. Integrate the varying density over the triangular event region. First confirm the model integrates to 1:

\begin{aligned} \int_0^1\int_0^1(x+y)\,dy\,dx &=\int_0^1\left(x+\frac12\right)\,dx\\ &=\frac12+\frac12=1. \end{aligned}

The event \(X+Y<1\) has \(0<x<1\) and \(0<y<1-x\). Hence

\begin{aligned} P(X+Y<1) &=\int_0^1\int_0^{1-x}(x+y)\,dy\,dx\\ &=\int_0^1\left(x(1-x)+\frac{(1-x)^2}{2}\right)\,dx\\ &=\int_0^1\left(\frac12-\frac{x^2}{2}\right)\,dx\\ &=\left[\frac{x}{2}-\frac{x^3}{6}\right]_0^1\\ &=\frac12-\frac16=\frac13=0.3333. \end{aligned}

At the specified point the density is \(f_{X,Y}(1/4,1/2)=1/4+1/2=3/4\). That is a density value, not a point probability: \(P(X=1/4,\ Y=1/2)=0\). Check: the complementary region \(X+Y>1\) has probability \(2/3\), so the two regions together have total probability 1; their shared boundary has probability zero.

Common mistake avoided: using the triangle's area, \(1/2\), as the probability. The density is not constant on the triangle, so the correct calculation integrates \(x+y\); the result is \(1/3\).

Common Exam Traps

Confusing density with probability. \(f_{X,Y}(x_0,y_0)\) is not \(P(X=x_0,Y=y_0)\). For a pair with a joint density, every singleton point has probability zero.
Forgetting the area element. A region's probability is a double integral. A one-dimensional length calculation does not give its probability unless the problem has first been reduced correctly to a one-dimensional distribution.
Integrating outside the support. Even if an event extends beyond the support, only the intersection of the event with the support contributes probability.
Assuming a density must be at most 1. The requirement is nonnegativity and total integral 1. A density can exceed 1 when its support is small enough.
Assigning probability to a boundary line. For a distribution with a joint density, a line has area zero and therefore probability zero. This conclusion depends on the continuous-density model.

Key takeaway: A joint density describes probability per unit area. For any event represented by a planar region \(R\), calculate its probability as \(\iint_R f_{X,Y}(x,y)\,dx\,dy\). A density value at one point is not the probability of that point.

Check Your Understanding

For each problem, identify the event region and use its density to calculate the requested probability.

A pair has density 1 on \(0<x<2,\ 0<y<1/2\), and zero elsewhere. Find \(P(X>3/2,\ Y<1/5)\). (Answer: 0.1000)
A pair is uniform on the unit square, so its density is 1 for \(0<x<1,\ 0<y<1\). Find \(P(Y>X+1/3)\). (Answer: \(2/9=0.2222\))
A pair has density 2 on \(0<x<y<1\), and zero elsewhere. Find \(P(Y-X>1/2)\). (Answer: 0.2500)
A pair has density \(x+y\) on the unit square, and zero elsewhere. Find \(P(X+Y>1)\). (Answer: \(2/3=0.6667\))
A pair has density 2 on \(0<x<1,\ 0<y<1-x\), and zero elsewhere. Find \(P(X<1/2)\). (Answer: 0.7500)

What You'll Learn