Why Joint Probability Tables Are Useful
When two discrete random variables are recorded together, a probability table gives a compact description of which pairs of values can occur and how likely each pair is. In actuarial work, a hypothetical model might pair the number of flagged items in a review with the review outcome. A question about both variables must be answered using probabilities of pairs, not by looking at just one variable in isolation.
This tutorial concerns finite discrete joint laws: each variable has a finite set of possible values, and the table records the probability assigned to every possible pair. There is no interest-rate or mortality model in these probability calculations. The model is the finite joint distribution specified by the table; its entries are the parameters needed for the questions below.
Reading and Checking the Table
The row and column labels are part of the information in the table: they identify the pair associated with each cell. For example, the entry where row \(X=1\) meets column \(Y=B\) is \(P(X=1,Y=B)\). The order is not interchangeable with a different pair, although the same event can of course be written in either order: \(P(X=1,Y=B)=P(Y=B,X=1)\).
A valid finite joint probability table has no negative entries, and the sum of all its entries is one. These conditions express that every possible outcome pair has a nonnegative probability and that the table accounts for the entire sample space. A combination that is impossible may be shown as a zero. A blank cell should be treated as zero only when the table explicitly indicates that the combination is impossible or that blank cells mean zero; a blank caused by missing information is not enough to determine a probability.
For an event consisting of several pairs, add the probabilities of the corresponding cells. Distinct pairs cannot happen simultaneously, so these cell events are disjoint. If a description uses “or,” first determine whether it includes both conditions occurring together. If it does, an overlapping cell must be counted only once.
Why Cell Probabilities Add
The sample space for the pair \((X,Y)\) consists of ordered pairs \((x_i,y_j)\). The event \(\{X=x_i,Y=y_j\}\) is one such pair. Different pairs are mutually exclusive: if \(X=x_i\) and \(Y=y_j\), then \(X\) and \(Y\) cannot simultaneously equal a different pair of values. Therefore, for any collection \(C\) of pairs, the event that the outcome belongs to \(C\) is a disjoint union of its cell events.
This follows from the additivity of probability for disjoint events. In the special case of a union of two events \(A\) and \(B\), if their cells overlap, adding \(P(A)\) and \(P(B)\) counts every shared cell twice. Subtracting the intersection once gives
In a table, this formula is a useful check. A direct sum of the cells in the union must agree with the inclusion-exclusion calculation.
Worked Examples
Worked Example: A Union of Row and Column Events
In a hypothetical quality review, \(X\) is the number of flagged items, either 0 or 1, and \(Y\) is the review category \(R\), \(S\), or \(T\). The specified joint distribution is shown below. Find \(P(X=1\text{ or }Y=T)\).
| \(X\backslash Y\) | \(R\) | \(S\) | \(T\) |
|---|---|---|---|
| 0 | 0.16 | 0.14 | 0.10 |
| 1 | 0.12 | 0.18 | 0.30 |
Check the model. All entries are nonnegative, and their total is \(0.16+0.14+0.10+0.12+0.18+0.30=1.00\). Thus the table is a valid joint probability table.
Identify the overlap. The event \(X=1\) includes the three cells in the second row, with total \(0.12+0.18+0.30=0.60\). The event \(Y=T\) includes the two cells in column \(T\), with total \(0.10+0.30=0.40\). Their intersection is the single cell \((X=1,Y=T)\), whose probability is \(0.30\).
Apply inclusion-exclusion. \(P(X=1\text{ or }Y=T)=0.60+0.40-0.30=0.70\). As a direct check, the union consists of the cells with probabilities \(0.12, 0.18, 0.30,\) and \(0.10\), which sum to \(0.70\).
Method and mistake avoided. Identify the overlap and subtract it once, or add the union’s cells directly. Adding \(0.60+0.40\) without subtracting \(0.30\) would double-count the pair \((1,T)\) and incorrectly give \(1.00\).
Worked Example: A Collection of Disjoint Pairs
In a hypothetical inspection model, \(X\) is the number of delayed components, with possible values 0, 1, and 2. \(Y\) records whether the result is routine (\(R\)) or requires review (\(V\)). Find the probability that there are at least one delayed component and a review result.
| \(X\backslash Y\) | \(R\) | \(V\) |
|---|---|---|
| 0 | 0.09 | 0.11 |
| 1 | 0.17 | 0.23 |
| 2 | 0.08 | 0.32 |
Check the model. The entries sum to \(0.09+0.11+0.17+0.23+0.08+0.32=1.00\), and none is negative.
Translate the event. “At least one” means \(X=1\) or \(X=2\). Together with \(Y=V\), the event consists of exactly two pairs: \((1,V)\) and \((2,V)\). These pairs are disjoint.
Add their cells. \(P(X\geq 1,Y=V)=0.23+0.32=0.55\). The answer is \(0.55\), or \(55\%\).
Method and mistake avoided. Translate the wording into the qualifying pairs before adding. The cell \((0,V)\), with probability \(0.11\), is excluded because zero delayed components is not at least one. Including it would answer a different question.
Worked Example: Subtracting an Overlap From a Table
A hypothetical monitoring system records \(X\), a count taking values 0, 1, or 2, and \(Y\), a status labeled \(A\), \(B\), or \(C\). Find the probability that \(X=0\) or \(Y=B\), including outcomes where both conditions hold.
| \(X\backslash Y\) | \(A\) | \(B\) | \(C\) |
|---|---|---|---|
| 0 | 0.07 | 0.13 | 0.10 |
| 1 | 0.16 | 0.14 | 0.10 |
| 2 | 0.08 | 0.12 | 0.10 |
Check the model. The total is \(0.07+0.13+0.10+0.16+0.14+0.10+0.08+0.12+0.10=1.00\).
Compute the two event totals. The \(X=0\) row totals \(0.07+0.13+0.10=0.30\). The \(Y=B\) column totals \(0.13+0.14+0.12=0.39\). The pair satisfying both conditions is \((0,B)\), with probability \(0.13\).
Correct for the overlap. \(P(X=0\text{ or }Y=B)=0.30+0.39-0.13=0.56\). Directly, the union contains the cells \(0.07,0.13,0.10,0.14,\) and \(0.12\), which sum to \(0.56\).
Method and mistake avoided. The word “or” includes the intersection here. Adding the row and column totals without removing their shared cell gives \(0.69\), an overcount of \(0.13\). If a problem instead says “exactly one” of the conditions, the shared cell must be excluded entirely.
What the Table Does and Does Not Say
A joint table specifies probabilities for pairs. It does not, by itself, say that \(X\) and \(Y\) are independent. Independence would require a separate condition: for every pair of possible values, the joint probability must equal the product of the corresponding individual probabilities. Do not multiply probabilities merely because two conditions involve different variables. For this tutorial’s event calculations, use the cell entries and add the cells that satisfy the event.
A reliable workflow is to check the table, translate the event into pairs, and then add or correct for overlap. In particular, a complicated verbal event should not be handled by guessing which row or column to use. Marking each qualifying cell makes the intended event visible and provides a direct way to check the arithmetic.
Confirm that every cell is nonnegative and that all cells sum to one.
List the pairs that satisfy the wording, paying attention to “and,” “or,” “at least,” and “exactly.”
Add the selected cells. For a union, either sum its cells directly or use inclusion-exclusion, then check that overlap was handled once.
Check Your Understanding
For each valid table, translate the event into cells before calculating. Answers are given so you can check the result.
- A hypothetical paired measurement has \(X\in\{0,1\}\) and \(Y\in\{L,H\}\), with table rows \((0.24,0.16)\) for \(X=0\) and \((0.21,0.39)\) for \(X=1\), in column order \(L,H\). Find \(P(X=0,Y=H)\). (Answer: 0.16)
- For \(X\in\{0,1\}\) and \(Y\in\{A,B,C\}\), the rows of a joint table are \((0.08,0.12,0.10)\) and \((0.15,0.20,0.35)\). Find \(P(X=1\text{ or }Y=C)\), including the overlap. (Answer: 0.80)
- A valid table for \(X\in\{0,1,2\}\) and \(Y\in\{R,S\}\) has rows \((0.12,0.08)\), \((0.17,0.23)\), and \((0.09,0.31)\). Find \(P(X\geq1,Y=S)\). (Answer: 0.54)
- For \(X,Y\in\{0,1\}\), the rows for \(X=0\) and \(X=1\), in column order \(Y=0,1\), are \((0.30,0.20)\) and \((0.10,0.40)\). Find the probability that exactly one of \(X=0\) and \(Y=1\) occurs. (Answer: 0.70)
- A joint table has \(X\in\{1,2,4\}\), \(Y\in\{0,1,2\}\), and rows \((0.10,0.05,0)\), \((0.15,0.20,0.10)\), and \((0,0.25,0.15)\), respectively. Find \(P(X+Y\geq4)\). (Answer: 0.50)