Conditional Probabilities from Two-Way Tables

Why the Denominator Changes

A two-way table organizes counts according to two categorical variables. It can answer questions such as “Among students in their first year, what proportion ride the bus?” or “Among bus riders, what proportion are first-year students?” These questions involve the same people and the same cell of the table, but they have different answers because they refer to different groups. The key is to use the total count for the group named after given as the denominator.

A conditional probability focuses on outcomes that satisfy a condition. In a two-way table, the condition selects a row or a column; the numerator is the count in the cell that also satisfies the event of interest. If the condition selects a row, use that row’s total. If it selects a column, use that column’s total.

The notation \(P(A\mid B)\) is read “the probability of \(A\) given \(B\).” The event after the vertical bar, \(B\), specifies the restricted group. It determines the denominator. The event before the bar, \(A\), specifies what we count within that group. When both events hold, the table entry is in their shared cell.

Finding the Relevant Counts

Consider this hypothetical table of 200 students, classified by year at college and usual way of commuting. Each student is counted once, in one row and one column. The totals at the ends of rows and bottoms of columns are called marginal totals: each sums the counts across the other variable.

The entry 42 counts students who are both first-year students and bus riders. It is a joint count: both descriptions apply. The number 90 is the total number of first-year students, while 70 is the total number of bus riders. These totals answer different “given” questions, so they cannot be substituted for one another indiscriminately.

A useful way to select the denominator is to read the conditional notation from right to left: first identify the event after the bar, then find its total in the table. Only after the restricted group is clear should you locate the cell that also meets the event before the bar.

Conditioning on a Column

The condition does not always select a row. If the group named after “given” corresponds to a column, the column total is the denominator. The same table can therefore support conditional probabilities with different denominators.

A Short Justification

The count formula follows directly from restricting the group under consideration. Suppose a table represents \(N\) equally likely individuals. Event \(B\) contains \(n_B\) individuals, and \(A\cap B\) contains \(n_{A\cap B}\). Once attention is limited to \(B\), there are \(n_B\) possible individuals left in the restricted group. Of these, \(n_{A\cap B}\) also satisfy \(A\). Therefore, the fraction of the restricted group that satisfies \(A\) is:

In the original group of \(N\), the corresponding probabilities are \(P(A\cap B)=n_{A\cap B}/N\) and \(P(B)=n_B/N\). Since \(n_B>0\), dividing these probabilities gives:

The grand total cancels. This explains both the conditional probability formula and why the denominator in a table is the total for the conditioning group. If the condition has count zero, then \(P(B)=0\), and the conditional probability is not defined by this formula.

Reading Carefully When Both Variables Are Named

In ordinary language, two questions can mention the same categories but condition in opposite directions. “Among people who received a reminder, what proportion attended?” asks for attendance given a reminder. “Among people who attended, what proportion received a reminder?” asks for a reminder given attendance. The relevant column or row total changes along with the condition.

Common Mistakes and Checks

The most frequent error is dividing by the grand total even though the question asks for a conditional probability. The grand total is appropriate for a proportion of the entire table, but conditioning narrows the relevant group. Another common error is choosing the correct number from the table but using the wrong marginal total.

• Wrong reasoning: “There are 42 first-year bus riders out of 200 students, so the probability a student rides the bus given that the student is in the first year is \(42/200\).”
• Correction: The condition says the student is in the first year. Use the 90 first-year students as the denominator: \(42/90\). The fraction \(42/200\) instead describes the proportion of all students who are both first-year and bus riders.
• Wrong reasoning: “The numerator is 42 in either question, so \(P(\text{bus}\mid\text{first year})\) and \(P(\text{first year}\mid\text{bus})\) must be equal.”
• Correction: The conditions are different, so the denominators are different: \(42/90\) versus \(42/70\). Always identify the group after “given” before dividing.
• Wrong reasoning: “A conditional probability can be calculated even if the condition’s row or column total is zero.”
• Correction: If the condition has no individuals, its total is zero and the required division is undefined.

Before finalizing a calculation, check that the numerator is a part of the denominator’s group, that the denominator matches the condition, and that the answer lies between 0 and 1. State the group in words so the interpretation is clear. For example, “46.7% of first-year students ride the bus” identifies the condition and the event; “46.7% ride the bus” does not.