From a Condition to a Probability
A joint probability model describes two random quantities together. Once the model is known, a question such as “What is the probability that the second count is less than the first?” becomes a precise task: identify all possible pairs satisfying the condition, then add the probabilities assigned to those pairs. The main work is often not the addition; it is interpreting the event correctly.
This skill is useful whenever an outcome depends on two discrete measurements at once, such as two types of claim counts or the numbers of two kinds of defects in a batch. The event may be defined by inequalities, an exact value of a sum, or a union of conditions. In each case, the joint distribution—not either marginal distribution alone—determines the answer.
In this formula, the event is a set of ordered pairs. The order matters: \((x,y)\) means \(X=x\) and \(Y=y\), not \(X=y\) and \(Y=x\). A pair with probability zero contributes nothing, so it is enough to inspect the support. When the random variables take integer values, a strict inequality such as \(Y<X\) excludes equality, while \(Y\le X\) includes it.
Why the Joint Probabilities Are Added
For a discrete model, an event made up of qualifying pairs is a union of individual outcomes. Distinct ordered pairs cannot occur simultaneously: the outcome \((X,Y)=(1,2)\) is different from \((X,Y)=(2,1)\). The corresponding singleton events are therefore mutually exclusive. Additivity of probability gives the event-sum formula.
Record the possible pairs \((x,y)\) and their joint probabilities.
For each pair, test the event’s inequalities or other requirements as written.
Add the joint probabilities of the selected points; exclude all other points.
More formally, write the event as a union of singleton events: $$ \{(X,Y)\in E\}=\bigcup_{(x,y)\in E}\{X=x,Y=y\}. $$ Because these singleton events are disjoint, finite additivity gives $$ P((X,Y)\in E)=\sum_{(x,y)\in E}P(X=x,Y=y) =\sum_{(x,y)\in E}p(x,y). $$ This justification depends on discrete outcomes. For a continuous joint distribution, probabilities are generally calculated from a density over a region rather than by adding probabilities at individual points.
Worked Examples
Worked Example: Two Inequalities on a Finite Support
A hypothetical model for two counts has the following joint probabilities: \(p(0,2)=2/15\), \(p(1,0)=3/15\), \(p(1,2)=4/15\), \(p(2,1)=1/15\), and \(p(3,0)=5/15\), with all other pairs assigned probability zero. Find the probability that \(Y<X\) and \(X+Y\le 3\).
Method. Test each listed support point against both conditions. The joint model is valid because the probabilities are nonnegative and \(2/15+3/15+4/15+1/15+5/15=15/15=1\).
At \((0,2)\), \(2<0\) is false, so this point does not qualify. At \((1,0)\), \(0<1\) and \(1+0=1\le3\), so it qualifies and contributes \(3/15\). At \((1,2)\), \(2<1\) is false, so it does not qualify. At \((2,1)\), \(1<2\) and \(2+1=3\le3\), so it qualifies and contributes \(1/15\). At \((3,0)\), \(0<3\) and \(3+0=3\le3\), so it qualifies and contributes \(5/15\).
Add the three qualifying probabilities: $$ P(Y<X,\ X+Y\le3)=\frac{3}{15}+\frac{1}{15}+\frac{5}{15} =\frac{9}{15}=0.6000. $$ The denominator is already the total probability mass, so the result is a valid probability. The common mistake to avoid is treating \(Y<X\) as though it meant \(Y\le X\); equality would not satisfy the stated strict inequality.
Worked Example: Reading a Joint Probability Table
For a hypothetical portfolio segment, \(X\) is the number of one type of incident and \(Y\) is the number of another type during a fixed period. Assume the joint probabilities are given by this table; each entry is \(p(x,y)\).
| \(X \backslash Y\) | 0 | 1 | 2 |
|---|---|---|---|
| 0 | \(1/24\) | \(2/24\) | \(3/24\) |
| 1 | \(2/24\) | \(1/24\) | \(3/24\) |
| 2 | \(1/24\) | \(2/24\) | \(1/24\) |
| 3 | \(3/24\) | \(2/24\) | \(3/24\) |
The entries sum to \((6+6+4+8)/24=24/24=1\), so they form a joint probability table. Find \(P(X\le Y,\ X+2Y\ge4)\).
Step 1: Apply \(X\le Y\). The possible pairs satisfying this condition are \((0,0),(0,1),(0,2),(1,1),(1,2),(2,2)\). No pair with \(X=3\) can satisfy it because the largest possible \(Y\) is 2.
Step 2: Apply \(X+2Y\ge4\) to those pairs. The values of \(X+2Y\), in the listed order, are \(0,2,4,3,5,6\). Thus only \((0,2),(1,2),(2,2)\) satisfy both conditions.
Step 3: Sum their joint probabilities. From the table, these entries are \(3/24,3/24,\) and \(1/24\). Therefore $$ P(X\le Y,\ X+2Y\ge4) =\frac{3}{24}+\frac{3}{24}+\frac{1}{24} =\frac{7}{24}=0.2917. $$ The fraction is exact; the decimal is rounded to four places. As a check, the answer lies between 0 and 1 and is smaller than the total probability of the points satisfying \(X\le Y\). The exam trap is to stop after applying the first condition and include points that fail the second.
Worked Example: A Sum Condition with a Second Restriction
A hypothetical quality-control model records \(X\), the number of minor flaws, and \(Y\), the number of major flaws on a selected item. The only possible pairs and their probabilities are shown below. Find the probability that there are at least two flaws in total and no more than one major flaw.
| \(X \backslash Y\) | 0 | 1 | 2 |
|---|---|---|---|
| 0 | \(4/30\) | \(3/30\) | \(1/30\) |
| 1 | \(5/30\) | \(6/30\) | \(2/30\) |
| 2 | \(2/30\) | \(4/30\) | \(3/30\) |
The table is a valid model: its row totals are \(8/30\), \(13/30\), and \(9/30\), which sum to \(30/30=1\). Translate the requested event as \(X+Y\ge2\) and \(Y\le1\). Check the cells with \(Y=0\) or \(Y=1\), since the second condition excludes the entire \(Y=2\) column.
For \(Y=0\), the condition \(X+Y\ge2\) requires \(X\ge2\). Among the table’s row values, only \(X=2\) qualifies, contributing \(2/30\). For \(Y=1\), the condition requires \(X+1\ge2\), or \(X\ge1\). The qualifying cells are \((1,1)\), with probability \(6/30\), and \((2,1)\), with probability \(4/30\). The cell \((0,1)\) does not qualify because the total is only one.
Adding the disjoint qualifying cells gives $$ P(X+Y\ge2,\ Y\le1) =\frac{2}{30}+\frac{6}{30}+\frac{4}{30} =\frac{12}{30}=\frac{2}{5}=0.4000. $$ The result is consistent with the event being a subset of \(Y\le1\), whose probability is \((4+3+5+6+2+4)/30=24/30\). A common error is to include the \(Y=2\) column because its total is at least two; those outcomes violate the separate requirement \(Y\le1\).
Useful Checks and Common Traps
A reliable calculation separates event interpretation from arithmetic. First state the event in symbols, then test candidate pairs, and only then add probabilities. In a table, keep row and column meanings visible: confusing the \(X\)-row with the \(Y\)-column reverses the coordinates and can change which outcomes qualify.
- Strict versus non-strict inequality. \(X<Y\) excludes ties; \(X\le Y\) includes them. For integer-valued variables, replacing one by the other changes the event.
- Every condition must hold. A conjunction such as \(X\le Y\) and \(X+Y\ge2\) selects the intersection of the two requirements, not all points satisfying either one.
- Use joint probabilities. Knowing the separate distributions of \(X\) and \(Y\) does not, by itself, determine the probability of an event involving both. The relevant entries are \(p(x,y)\).
- Do not count a point twice. Each qualifying ordered pair contributes once. If an event is split into cases, make the cases disjoint or explicitly correct for overlap.
- Check the model and the answer. The joint probabilities should sum to one, and the event probability must be between zero and one. It cannot exceed the probability of a broader event that contains it.
Check Your Understanding
For each problem, identify the qualifying outcomes before adding their probabilities.
- A joint model assigns probabilities \(2/12,3/12,1/12,4/12,2/12\) to \((0,1),(1,0),(1,2),(2,1),(2,2)\), respectively. Find \(P(Y<X)\). (Answer: \(5/12=0.4167\))
- A table has rows \(X=0,1,2\) and columns \(Y=0,1,2\), with row entries \((1/10,2/10,1/10)\), \((2/10,1/10,1/10)\), and \((1/10,1/10,0/10)\). Find \(P(X+Y\ge2,\ Y\le1)\). (Answer: \(5/10=0.5000\))
- A joint model assigns probability \(1/20\) to each of \((0,0),(0,2),(1,1),(1,3),(2,0),(2,2),(3,1),(3,3)\), and zero elsewhere. Find \(P(X\ge Y,\ X+Y=4)\). (Answer: \(2/20=0.1000\))
- A table has rows \(X=0,1,2\) and columns \(Y=0,1,2\), with entries \((2/18,1/18,2/18)\), \((1/18,3/18,2/18)\), and \((2/18,2/18,3/18)\). Find the probability that \(X=2\) or \(Y=0\), taking care not to count the overlap twice. (Answer: \(8/18=0.4444\))
- A joint model has support \((0,0),(0,1),(1,0),(1,2),(2,1),(2,2)\) with respective probabilities \(1/14,2/14,3/14,2/14,4/14,2/14\). Find \(P(X+Y\ge3,\ X>Y)\). (Answer: \(2/14=0.1429\))