Conditional Probability Order And Denominator Traps

Why Conditional Probability Order Matters

In a multi-stage probability problem, the events may be described in the order they occur, but the question may ask about them in the reverse direction. For example, a problem may give the chance of a positive result for someone who has a condition, then ask for the chance that someone has the condition given a positive result. These are different probabilities. The denominator changes with the condition, so using the full population—or simply reversing the words in a known probability—can produce a plausible but incorrect answer.

The reliable method is to name the events, identify the event after “given,” and use that event as the conditioned sample space. In multi-stage problems, also keep track of which conditions are already in force before applying another probability.

Definition: For events $A$ and $B$ with $P(B)>0$, the conditional probability of $A$ given $B$ is $$ P(A\mid B)=\frac{P(A\cap B)}{P(B)}. $$ The numerator is the probability that both events occur; the denominator is the probability of the condition $B$. In particular, $P(A\mid B)$ and $P(B\mid A)$ generally differ.

Read the Condition Before Choosing a Denominator

The expression $P(A\mid B)$ asks what fraction of the $B$ cases are also $A$ cases. Its denominator is $P(B)$, not $1$, and not $P(A)$. When data are given as counts, the same rule says to divide the number in both groups by the number in the conditioned group.

If events $B_1,\ldots,B_k$ form a partition of the population—exactly one of them occurs, and their probabilities sum to 1—then the probability of $A$ can be found by adding the contributions from each group:

P(A)=\sum_{j=1}^{k}P(A\mid B_j)P(B_j).

This is the law of total probability. Combining it with the definition of conditional probability gives Bayes’ theorem:

P(B_i\mid A)= \frac{P(A\mid B_i)P(B_i)} {\sum_{j=1}^{k}P(A\mid B_j)P(B_j)}, \qquad P(A)>0.

The denominator is the total probability of the observed event $A$, collected across all the possible groups. It is not generally the probability of $A$ within the one group named in the numerator.

Why Bayes’ Theorem Has This Form

Bayes’ theorem follows directly from the definition, rather than from a rule for informally swapping the order of events. The numerator of $P(B_i\mid A)$ is the joint probability of $B_i$ and $A$. The denominator $P(A)$ can be expanded across the partition.

Write the conditional probability.
By definition, $P(B_i\mid A)=P(B_i\cap A)/P(A)$, provided $P(A)>0$.

Express the joint event.
The definition also gives $P(B_i\cap A)=P(A\mid B_i)P(B_i)$.

Expand the denominator.
Because the $B_j$ partition the population, $P(A)=\sum_j P(A\mid B_j)P(B_j)$.

Substitute.
Putting the joint probability and expanded denominator into Step 1 yields Bayes’ theorem.

This derivation explains the key exam check: the denominator must include every way the observed event $A$ could have occurred. Omitting a group changes the sample space and usually overstates the posterior probability.

Worked Examples

Worked Example: Passing by Two Different Routes

A hypothetical qualification program has 80 applicants. Of 50 applicants who completed an assessment, 35 passed. Of the 30 who did not complete it, 6 passed through an alternate route. Given that a randomly selected applicant passed, find the probability that the applicant completed the assessment. Also compare it with the probability of passing among those who completed the assessment.

Method. Let $C$ denote completing the assessment and $P$ denote passing. The question asks for $P(C\mid P)$, so the denominator must be all applicants who passed, regardless of route.

The number who passed is $35+6=41$. Of those 41 applicants, 35 completed the assessment. Therefore,

P(C\mid P)=\frac{35}{41}=0.8536585\ldots\approx 0.8537.

For comparison, the probability of passing given that the applicant completed the assessment is

P(P\mid C)=\frac{35}{50}=0.7000.

The two values answer different questions: among passers, the fraction who completed the assessment is $0.8537$; among those who completed it, the fraction who passed is $0.7000$. Check: the 41 passers split into 35 who completed the assessment and 6 who did not, so the denominator and numerator match the requested condition. The common mistake is to use $35/50$ for the first question, thereby calculating $P(P\mid C)$ instead of $P(C\mid P)$.

Worked Example: Updating a Defect Probability After an Inspection

A hypothetical supplier receives 60% of its components from Line A and 40% from Line B. The defect probabilities are 0.02 for Line A and 0.05 for Line B. An inspection flags 90% of defective components and incorrectly flags 4% of nondefective components. Given that a component is flagged, find the probability that it is defective.

Method. Let $D$ mean defective and $F$ mean flagged. First calculate the joint probability of being defective and flagged. Then calculate the full probability of being flagged, including flags on nondefective components.

The probability of a defect is found by considering both source lines:

$$ P(D)=0.60(0.02)+0.40(0.05)=0.012+0.020=0.032. $$

A defective component is flagged with probability 0.90, so

P(D\cap F)=P(D)P(F\mid D)=0.032(0.90)=0.0288.

The probability of a nondefective component is $1-0.032=0.968$. A nondefective component is flagged with probability 0.04. Thus the total flag probability is

P(F)=P(D\cap F)+P(D^c\cap F) =0.0288+0.968(0.04) =0.0288+0.03872 =0.06752.

Bayes’ theorem now gives

P(D\mid F)=\frac{0.0288}{0.06752} =\frac{90}{211} \approx 0.4265.

Thus, the probability that a flagged component is defective is 0.4265, or about 42.65%. Check: the numerator, 0.0288, is one part of the total flag probability, 0.06752, so the conditional probability is between 0 and 1. The common mistake is to report $P(F\mid D)=0.9000$ as the requested answer. That is the chance of a flag given a defect, not the chance of a defect given a flag.

Worked Example: Applying Two Conditions to a Claim Cohort

A hypothetical review contains 1,000 claims. Six hundred are from Region R and 400 are from Region S. In Region R, 30% of claims have complete documentation; in Region S, 50% do. Among claims with complete documentation, 20% from Region R and 10% from Region S are approved. Given that a claim is both fully documented and approved, find the probability it came from Region R.

Method. Let $R$ and $S$ denote the region, $D$ denote complete documentation, and $A$ denote approval. The condition is $D\cap A$, so the denominator counts claims satisfying both conditions, not all documented claims and not all claims.

In Region R, $600(0.30)=180$ claims are fully documented, and $180(0.20)=36$ of these are approved. In Region S, $400(0.50)=200$ claims are fully documented, and $200(0.10)=20$ of these are approved. There are $36+20=56$ claims that meet both conditions. Therefore,

P(R\mid D\cap A)=\frac{36}{56} =\frac{9}{14} \approx 0.6429.

So about 64.29% of the claims that are both documented and approved came from Region R. Check: the 56 claims in the conditioned group divide into 36 from R and 20 from S. Using $36/600$ would instead calculate the fraction of all Region R claims that are both documented and approved, not the fraction of qualifying claims that are from R. The common mistake is to stop conditioning after $D$ and overlook the additional approval condition.

Worked Example: Reversing the Order in a Without-Replacement Draw

A container holds 5 red and 3 blue tokens. Two tokens are drawn uniformly at random without replacement. Given that the second token is blue, find the probability that the first token was red. Compare this with the probability that the second token is blue given that the first token was red.

Method. Let $R_1$ mean the first token is red and $B_2$ mean the second is blue. For the requested reverse conditional probability, calculate the joint probability $P(R_1\cap B_2)$ and divide by $P(B_2)$.

The joint probability is

P(R_1\cap B_2)=\frac{5}{8}\frac{3}{7}=\frac{15}{56}.

The second token is blue either when the first is red and the second is blue, or when both are blue. The second path has probability

P(B_1\cap B_2)=\frac{3}{8}\frac{2}{7}=\frac{6}{56}.

These are the only two ways the second token can be blue, so $P(B_2)=15/56+6/56=21/56$. Hence,

P(R_1\mid B_2)=\frac{15/56}{21/56}=\frac{5}{7}\approx 0.7143.

In the other direction, once the first token is known to be red, 7 tokens remain, of which 3 are blue. Therefore,

P(B_2\mid R_1)=\frac{3}{7}\approx 0.4286.

Check: both answers use the correct conditioned sample space and are unequal, as expected. The common mistake is to use $3/7$ for the first question; that is the probability of a blue second draw given a red first draw, not the probability of a red first draw given a blue second draw.

A Reliable Denominator Check

Before calculating a conditional probability, write the condition in words and translate it into a denominator. If the question asks “given a flagged result,” the denominator includes every flagged result. If it asks “given a documented and approved claim,” only claims satisfying both conditions belong in the denominator. For a multi-stage path, multiply the probabilities along that path, then add other paths that also satisfy the condition.

A useful way to prevent order errors is to keep the joint event separate from the conditional probability. For any $A$ and $B$,

P(A\cap B)=P(A\mid B)P(B)=P(B\mid A)P(A).

Both products describe the same joint event, but the conditional factors and their denominators differ. The equality does not mean the conditional probabilities are equal. They are equal only in special cases, such as when the events are independent and the relevant probabilities are positive.

Exam check: For a requested $P(A\mid B)$, mark $B$ as the conditioned group. Verify that the denominator counts or sums all cases in $B$, and that the numerator contains only cases in both $A$ and $B$. This check is aligned with the SOA Exam P (Probability) syllabus.

Check Your Understanding

For each problem, identify the conditioned group before calculating.

A hypothetical program has 45 applicants in Group M and 55 in Group N. The probability of passing is 0.80 for M and 0.50 for N. Given that an applicant passed, find the probability the applicant was in M. (Answer: 0.5669)
Among 300 hypothetical applicants, 180 attended a workshop and 126 of those qualified. Of the 120 who did not attend, 18 qualified. Given that an applicant qualified, find the probability they attended. (Answer: 0.8750)
A hypothetical screening population has 70% from Group C and 30% from Group D. The event of interest occurs with probability 0.01 in C and 0.04 in D. A positive result occurs for 80% of those with the event and 5% of those without it. Given a positive result, find the probability the event is present. (Answer: 0.0867)
A container has 4 green and 6 yellow counters. Two are drawn uniformly without replacement. Given that the second counter is green, find the probability the first was yellow. (Answer: 0.6667)
A hypothetical delivery service uses Carrier X for 40% of parcels and Carrier Y for 60%. The probability of late delivery is 0.10 for X and 0.05 for Y. Given a late delivery, find the probability that Carrier X handled the parcel. (Answer: 0.5714)

What You'll Learn