Pairs of Random Variables and Their Joint Law
When two random variables describe the same outcome, their values must be considered together. For example, a claim may have both a reported amount and a processing time; a policyholder may have both an age and a claim count. Knowing the possible values of each variable separately does not necessarily tell us which pairs can occur or how probability is distributed among them.
Let \(X\) and \(Y\) be random variables defined on the same probability space. The ordered pair \((X,Y)\) is a two-dimensional random vector. Its joint law specifies the probability that this pair lies in any chosen set of points in the plane. The set may be a single point, several separated points, a rectangle, or a more general region.
The support describes where the pair can occur according to its law. An event, by contrast, is a selected collection of possible pairs. For instance, \(X+Y>3\) defines an event: the pairs in the plane above the line \(x+y=3\). Its probability depends on how the joint law assigns probability within that region, not just on the region's shape.
Support Is Not the Same as an Event
The support and an event answer different questions. The support answers, “Where can the pair be found under this law?” An event answers, “Which outcomes are being counted for this probability question?” A region can include points outside the support; those points do not contribute probability. A region can also intersect the support without containing all of it.
In a discrete model, the support is typically a collection of individual pairs with positive probability. In a continuous model, the support can contain infinitely many points, even though each individual point has probability zero. For example, a continuously distributed pair can have support equal to a filled rectangle. Each point in that rectangle belongs to the support because every neighborhood around it has positive probability, even though the probability of the point itself is zero.
Support is a property of the joint law, not simply a list of values that happen to appear in a sample. A finite sample may fail to include many values in the support. Conversely, a value that is mathematically possible but has probability zero does not automatically belong to the support: the neighborhood condition is the defining test.
Reading a Joint Law as a Set of Outcomes
For a finite model, one can list the possible pairs and their probabilities. To find the probability of a region, check which listed pairs satisfy its condition and add their probabilities. This is an application of the definition of probability for a union of mutually exclusive outcomes. The region may be large geometrically, but only its intersection with the support matters.
Worked Example: A Region Applied to a Finite Support
A hypothetical monitoring system records \(X\), the number of alerts on a low-load channel, and \(Y\), the number on a high-load channel, during a one-hour test. Its joint law assigns probability \(0.20\) to \((0,0)\), \(0.30\) to \((0,1)\), \(0.10\) to \((1,0)\), and \(0.40\) to \((1,2)\), with no other pairs possible. Find the support and the probability of the event \(X+Y\geq 2\).
Model and method. This is a finite joint distribution with four possible pairs and the stated probabilities. The support consists of those pairs; the event probability is the sum of probabilities for the pairs satisfying the inequality. First, the probabilities total \(0.20+0.30+0.10+0.40=1.00\).
Support. The four possible pairs are \((0,0)\), \((0,1)\), \((1,0)\), and \((1,2)\). Each has positive probability, so all four belong to the support.
Apply the event condition. The sums at the four support points are \(0+0=0\), \(0+1=1\), \(1+0=1\), and \(1+2=3\). Only \((1,2)\) satisfies \(X+Y\geq2\). Therefore, $$ P(X+Y\geq2)=P((X,Y)=(1,2))=0.40. $$
Check and trap avoided. The result is between zero and one, and the selected event contains exactly one support point. A common error is to treat every point in the half-plane \(x+y\geq2\) as an outcome. The model allows only the four listed pairs, so all other points in that region contribute zero probability.
Continuous Supports and Regions
In a continuous model, the support often contains a continuum of pairs, and a region's probability is not found by counting its points. A useful geometric model is a pair that is uniformly distributed over a rectangle: the probability of a region inside that rectangle is its area divided by the rectangle's total area. This follows from the model's equal probability per unit area.
The support of a uniform rectangle is the closed rectangle, including its edges and corners. These boundary points belong to the support because every neighborhood around one of them intersects the rectangle in a region of positive area. Yet an edge or a single point has area zero, so it has probability zero under this model. This illustrates why support membership does not mean that the point itself has positive probability.
Worked Example: A Half-Plane Intersecting a Rectangle
Suppose \((X,Y)\) is uniformly distributed over the rectangle \(0\leq X\leq2\), \(0\leq Y\leq1\). Find the probability of \(X>Y\). The joint model is uniform over the stated rectangle, so probability is proportional to area.
Model and support. The support is the rectangle \([0,2]\times[0,1]\). Its total area is \(2\cdot1=2\). The event \(X>Y\) is the half-plane to the right of the line \(x=y\); only its intersection with the support matters.
Find the event area. For \(0\leq y\leq1\), the event requires \(x>y\), while the support requires \(x\leq2\). At a fixed \(y\), the permitted horizontal length is \(2-y\). Thus the event area is $$ \int_0^1(2-y)\,dy =\left[2y-\frac{y^2}{2}\right]_0^1 =2-\frac12 =\frac32. $$
Convert area to probability. Divide the event area by the total support area: $$ P(X>Y)=\frac{3/2}{2}=\frac34=0.7500. $$
Check and trap avoided. The event occupies three quarters of the rectangle, so a probability of \(0.7500\) is geometrically plausible. The boundary \(X=Y\) has area zero and does not change the answer if the event is written \(X\geq Y\). A common mistake is to use the area of the whole half-plane; the probability calculation must be restricted to the support.
Supports Constrained to a Curve
A joint support need not fill a two-dimensional region. If one variable is determined by the other, the pair may be restricted to a line or curve. The support then records that constraint, even though a region used to define an event may cover a much larger part of the plane.
To evaluate an event in such a model, substitute the relationship between the variables into the event condition. This identifies which values of the underlying variable produce pairs in the event. The probability is then found using the distribution of that underlying variable.
Worked Example: An Event Intersecting a Line-Shaped Support
Let \(X\) have a uniform distribution on \([0,1]\), and define \(Y=2X+1\). Find the probability of the event \(1.5\leq Y\leq2.5\). This is a hypothetical model in which the second measurement is exactly determined by the first.
Model and support. The possible pairs lie on the line segment $$ \{(x,2x+1):0\leq x\leq1\}. $$ The support is this segment, not the whole rectangle \(0\leq x\leq1,\ 1\leq y\leq3\).
Translate the event to \(X\). Substitute \(Y=2X+1\): $$ 1.5\leq2X+1\leq2.5. $$ Subtracting \(1\) gives \(0.5\leq2X\leq1.5\), and dividing by \(2\) gives \(0.25\leq X\leq0.75\).
Calculate the probability. Since \(X\) is uniform on an interval of length \(1\), the probability of a subinterval is its length divided by \(1\): $$ P(1.5\leq Y\leq2.5) =P(0.25\leq X\leq0.75) =\frac{0.75-0.25}{1} =0.5000. $$
Check and trap avoided. The interval for \(X\) lies within its support, and its length is \(0.50\), giving probability \(0.5000\). A common error is to imagine that the pair is spread uniformly through the surrounding rectangle. It is concentrated on the line segment, because \(Y\) is fixed once \(X\) is specified.
Why the Joint-Law View Matters
A joint model must be read as a model for pairs, not as two unrelated lists of values. Separate information about the possible values of \(X\) and \(Y\) may describe the ranges of each coordinate, but it does not by itself determine which combinations occur. A support can impose a relationship such as \(Y=2X+1\), exclude parts of a rectangle, or consist of isolated points.
The set-based definition also gives a reliable procedure for probability questions. First write the event as a condition on \((X,Y)\). Next identify the support and intersect the event with it. Finally use the specified joint law to calculate the probability of that intersection. The calculation differs by model—adding point probabilities for a finite law, or using the model's area probabilities for a uniform region—but the event remains a set of pairs in either case.
Identify the two random variables and the rule or distribution governing their values together.
Determine which pairs belong to the joint law's support, rather than assuming every pair in the plane is possible.
Translate the question into a condition on \(X\) and \(Y\), such as \(X+Y\geq2\) or \(X>Y\).
Calculate probability only for outcomes that satisfy the event and are supported by the model.
Boundary details require care. If a boundary contains positive probability under a discrete or other noncontinuous model, changing a strict inequality to a non-strict one can change the answer. Under the uniform rectangle model, a line has area zero, so including or excluding that line does not change the probability. The model, rather than the notation alone, determines whether the boundary matters.
This distinction is foundational for later work with joint distributions: probability questions are posed about regions, while the support constrains which points can contribute. Keeping those ideas separate prevents a geometric description of an event from being mistaken for a description of the distribution itself.
Check Your Understanding
For each problem, identify the relevant support or region before calculating the requested probability.
- A finite joint law assigns probabilities \(0.15, 0.25, 0.35,\) and \(0.25\) to \((0,0),(0,2),(1,1),\) and \((2,0)\), respectively, with no other possible pairs. State the support and find \(P(X+Y\geq2)\). (Answer: Support \(\{(0,0),(0,2),(1,1),(2,0)\}\); probability \(0.8500\).)
- \((X,Y)\) is uniform over \(0\leq X\leq3,\ 0\leq Y\leq2\). Find \(P(X>2Y)\). (Answer: \(0.3750\).)
- \(X\) is uniform on \([0,2]\) and \(Y=3-X\). State the support and find \(P(Y>2)\). (Answer: Support \(\{(x,3-x):0\leq x\leq2\}\); probability \(0.5000\).)
- A finite joint law assigns probabilities \(0.20,0.15,0.40,\) and \(0.25\) to \((0,1),(1,0),(1,2),\) and \((2,2)\), respectively. Find the probability of the region \(X\leq Y\). (Answer: \(0.8500\).)
- \(X\) is uniform on \([0,1]\) and \(Y=X^2\). The support is \(\{(x,x^2):0\leq x\leq1\}\). Find \(P(0.25\leq Y\leq0.64)\). (Answer: \(0.3000\).)