Introduction
Many probability errors are not arithmetic errors. They arise because a condition is read in the wrong direction, a relevant base rate is ignored, or two events are treated as independent without evidence. These mistakes can produce plausible-looking numbers that answer a different question from the one asked.
In actuarial work, the distinction matters whenever information about one event is used to assess another—for example, when interpreting a screening result or estimating whether two types of claims occur together. This tutorial focuses on diagnosing those errors and choosing the probability calculation that the stated information actually supports.
Read the Condition in the Correct Direction
The notation \(P(A\mid B)\) means “the probability of \(A\), among cases in which \(B\) is known to have occurred.” The event after the vertical bar is the condition and determines the denominator. Thus \(P(A\mid B)\) and \(P(B\mid A)\) generally have different denominators and need not be equal.
The conditional-probability definition gives \(P(A\cap B)=P(A\mid B)P(B)\). It also gives \(P(A\cap B)=P(B\mid A)P(A)\), when \(P(A)>0\). Equating these expressions yields Bayes’ identity:
The factor \(P(A)\) is the base rate of \(A\). A large value of \(P(B\mid A)\) does not, by itself, imply a large value of \(P(A\mid B)\): the result also depends on how common \(A\) is and how often \(B\) occurs when \(A\) does not.
Worked Example: Interpreting a Positive Screening Result
In a hypothetical population of 10,000 people, 2% have a condition. A screening test is positive for 90% of people who have the condition and is positive for 5% of people who do not. Find the probability that a randomly selected person has the condition given a positive result. Name the method and identify the common mistake it avoids.
Method. Use the population counts, or equivalently Bayes’ identity. The population size is a convenient way to keep the two conditional directions separate.
Step 1: Count people with and without the condition. The number with the condition is \(10{,}000(0.02)=200\). The number without it is \(10{,}000-200=9{,}800\).
Step 2: Count positive results in each group. Among the 200 people with the condition, \(200(0.90)=180\) test positive. Among the 9,800 without the condition, \(9{,}800(0.05)=490\) test positive.
Step 3: Use all positive results as the denominator. There are \(180+490=670\) positive results, of which 180 are from people with the condition. Therefore,
The requested probability is about 0.2687, or 26.87%. The 90% figure is \(P(\text{positive}\mid\text{condition})\), not \(P(\text{condition}\mid\text{positive})\). Reversing those conditional probabilities would answer the wrong question. The calculation also shows why the base rate matters: the 490 positive results among people without the condition must be included.
Use the Multiplication Rule Without Assuming Independence
The general multiplication rule is \(P(A\cap B)=P(A\mid B)P(B)\), provided \(P(B)>0\). It does not require independence. The shorter expression \(P(A\cap B)=P(A)P(B)\) is valid only when \(A\) and \(B\) are independent.
This distinction is useful because information about an intersection or a conditional probability can reveal whether independence is plausible. If \(P(A\cap B)\) differs from \(P(A)P(B)\), the events are dependent. A numerical problem that gives both marginal probabilities but no joint or conditional information does not, by itself, establish independence.
Worked Example: Two Types of Claims With Dependence
For a hypothetical portfolio in a particular year, let \(S\) be the event of at least one storm claim and \(W\) the event of at least one water-damage claim. Suppose \(P(S)=0.20\), \(P(W)=0.10\), and \(P(S\cap W)=0.06\). Find the probability of at least one of these claim types, the probability of neither, and \(P(W\mid S)\). Determine whether the events are independent.
Method. Use the inclusion-exclusion formula for the union and the definition of conditional probability. Do not infer independence from the two marginal probabilities.
Step 1: Calculate the probability of at least one claim type. The intersection is counted in both marginal probabilities, so subtract it once:
Step 2: Calculate the probability of neither claim type. “Neither” is the complement of \(S\cup W\), so
Step 3: Calculate the conditional probability and check independence.
If the events were independent, their intersection would have probability \(P(S)P(W)=(0.20)(0.10)=0.02\). The supplied intersection is 0.06, not 0.02, so the events are dependent. The conditional check agrees: \(P(W\mid S)=0.30\), whereas \(P(W)=0.10\).
Common mistake avoided. Multiplying the probabilities of no storm and no water claim, \(0.80(0.90)=0.72\), assumes independence of the events and their complements. That assumption is not given and is contradicted by the joint probability. Under the supplied model, the probability of neither is \(0.76\), as calculated from the complement of the actual union.
Separate a Valid Product Rule From an Independence Claim
The product \(P(A\cap B)=P(A\mid B)P(B)\) is valid whenever \(P(B)>0\). It may be used with a conditional probability supplied by the problem, even if the events are dependent. By contrast, replacing \(P(A\mid B)\) by \(P(A)\) is an additional step that requires independence.
A useful check is to compare the probability of the event after conditioning with its marginal probability. If \(P(A\mid B)\ne P(A)\), then \(A\) and \(B\) are not independent. When the problem gives \(P(B\mid A)\) and \(P(B\mid A^c)\), those values can be used to calculate \(P(B)\) by splitting the population into the disjoint cases \(A\) and \(A^c\).
Worked Example: Combining Conditional Information Correctly
In a hypothetical review process, 30% of applications are flagged for manual review; call this event \(A\). A payment discrepancy occurs for 40% of flagged applications and for 10% of applications not flagged; call a discrepancy \(B\). Find \(P(A\cap B)\), \(P(B)\), and \(P(A\mid B)\). Then assess whether \(A\) and \(B\) are independent.
Method. First use the conditional multiplication rule for the intersection. Then partition \(B\) according to whether \(A\) occurred. This avoids assuming independence.
Step 1: Find the intersection. Since \(P(B\mid A)=0.40\) and \(P(A)=0.30\),
Step 2: Find the marginal probability of a discrepancy. The probability of \(A^c\) is \(1-0.30=0.70\). The disjoint cases \(A\cap B\) and \(A^c\cap B\) make up \(B\), so
Step 3: Reverse the condition using the intersection.
Step 4: Check independence. The product of the marginal probabilities is \((0.30)(0.19)=0.057\), which is not the intersection probability 0.12. Equivalently, \(P(B\mid A)=0.40\ne P(B)=0.19\). Thus the events are dependent.
Common mistake avoided. Multiplying \(P(A)P(B)=0.057\) to find the intersection would incorrectly assume independence. The given conditional rate is 0.40, so the valid product for the intersection is \(P(A)P(B\mid A)=0.12\).
A Short Diagnostic for Probability Problems
Before calculating, identify what the problem asks for and what each given probability conditions on. Then check whether a product rule is being used in its general or independence form. This brief sequence can prevent the most common errors.
For \(P(A\mid B)\), \(B\) is the condition and is the denominator event. Do not interchange it with \(P(B\mid A)\).
For a reverse conditional probability, determine how common the target event is and account for observations both inside and outside that event.
Use \(P(A\cap B)=P(A\mid B)P(B)\) when the conditional probability is known. Use \(P(A)P(B)\) only after independence is established or stated.
For “neither,” take the complement of the union. Verify that the result uses the actual union probability, not one calculated under an unsupported independence assumption.
These checks are aligned with the SOA VEE Mathematical Statistics syllabus because they require precise use of conditional probability, complements, unions, and independence. In applied settings, the numerical result is only as sound as the assumptions used to obtain it.
Common Mistakes
- Reversing a conditional probability. \(P(\text{positive}\mid\text{condition})\) is not generally the probability of the condition given a positive result. Identify the requested condition and denominator before substituting numbers.
- Ignoring the base rate. A test or event may be common within a group but still be uncommon among all observed cases if that group is rare.
- Multiplying marginal probabilities automatically. \(P(A)P(B)\) gives the intersection only under independence. Otherwise, use a supplied conditional probability in the general multiplication rule.
- Assuming independence from separate marginal probabilities. Knowing \(P(A)\) and \(P(B)\) alone does not determine \(P(A\cap B)\).
- Using an independence calculation for complements without checking it. Independence of events implies independence of their complements, but a dependent pair cannot be made independent by switching to “neither.”
Check Your Understanding
For each problem, identify the condition or independence assumption required by the calculation.
- In a hypothetical group, \(P(A)=0.15\), \(P(B\mid A)=0.40\), and \(P(B\mid A^c)=0.10\). Find \(P(A\cap B)\), \(P(B)\), and \(P(A\mid B)\). (Answer: \(P(A\cap B)=0.0600\), \(P(B)=0.1450\), and \(P(A\mid B)=0.4138\).)
- A condition occurs in 4% of a hypothetical population. A test is positive for 85% of those with the condition and for 3% of those without it. Find the probability of the condition given a positive result. (Answer: \(0.5414\).)
- Suppose \(P(C)=0.25\), \(P(D)=0.40\), and \(P(C\cap D)=0.12\). Find \(P(D\mid C)\), determine whether \(C\) and \(D\) are independent, and find the probability of neither event. (Answer: \(P(D\mid C)=0.4800\); dependent; \(P(C^c\cap D^c)=0.4700\).)
- For events \(M\) and \(N\), suppose \(P(M)=0.60\), \(P(N\mid M)=0.20\), and \(P(N\mid M^c)=0.50\). Find \(P(M\cap N)\), \(P(N)\), and \(P(M\mid N)\), and assess independence. (Answer: \(P(M\cap N)=0.1200\), \(P(N)=0.3200\), \(P(M\mid N)=0.3750\); dependent.)
- Suppose \(P(E)=0.30\), \(P(F)=0.25\), and \(E\) and \(F\) are mutually exclusive. Find \(P(E\cup F)\), \(P(E\cap F)\), and \(P(E\mid F)\). Are \(E\) and \(F\) independent? (Answer: \(P(E\cup F)=0.5500\), \(P(E\cap F)=0\), \(P(E\mid F)=0\); not independent.)