Use only when every outcome in the sample space is equally likely.

Every business decision is made under uncertainty: whether a new product will sell, whether a supplier will deliver on time, whether a stock will rise. Probability is the language we use to measure that uncertainty, and it marks a turning point in the book. Until now we have described data we already collected. From here we build the mathematics of uncertainty, which will later let us reason from an imperfect sample back to the population that produced it. We start with the basic ideas.
Before we calculate probabilities, it helps to agree on what a probability means. There are two main ways to understand it.
Frequentist view. A probability is the long-run proportion of times an event occurs when an experiment is repeated many times under the same conditions. Saying a product has a \(2\%\) return rate means that, over many sales, about \(2\) in every \(100\) items come back. In this view, probabilities only make sense for events that can be repeated.
Bayesian view. A probability is a degree of belief about how likely an event is. It is subjective in the sense that it reflects a reasoned judgment based on the evidence and experience available, not a random guess, and it is updated as new data arrive. This lets us put a probability on one-time events, such as the chance that a particular product launch succeeds. A frequentist would not assign such a probability, because the launch cannot be repeated.
The two views disagree about what a probability is, but they follow the same rules (the Addition Rule, the Multiplication Rule, and Bayes’ Theorem), so the calculations in this book work the same way under either one.
Example: A company estimates the return rate of an existing product from the share of past sales that were returned. That is a frequentist probability. For a brand-new product with no sales history, managers instead assign a probability of success from market research and experience, and revise it as early sales come in. That is a Bayesian probability.
The following concepts form the foundation of probability theory. They provide a structure that is useful when analyzing uncertainty and randomness.
An experiment is a process that leads to one of several outcomes. Ex: Tossing a Die, Tossing a Coin, Drawing a Card, etc.
An outcome is the result of an experiment. Ex: A coin landing on heads, drawing the ace of spades.
The sample space \((S)\) of an experiment contains all possible outcomes of the experiment. Ex: \(S=\{1,2,3,4,5,6\}\) is the sample space for tossing a die.
An event is a subset of the sample space. Ex: \(A=\{2,4,6\}\) is the event of tossing an even number when rolling a die. \(B=\{1A,1D,1S,1H\}\) is the event of drawing an ace from a deck of cards.
A probability is a numerical value that measures the likelihood that an event occurs. There are three common ways to assign one.
Classical method. When every outcome in the sample space is equally likely (a fair die, a well-shuffled deck of cards), the probability of an event is the share of outcomes that belong to it:
CLASSICAL PROBABILITY \[P(A)=\frac{Number\ of\ outcomes\ in\ A}{Number\ of\ outcomes\ in\ Sample \ Space}\] This formula only applies when all outcomes are equally likely.
Example: Draw one card at random from a standard deck of \(52\) cards. Every card is equally likely, so we can use the classical method. Let \(A\) be the event of drawing a red card. There are \(26\) red cards, so
\[P(A) = \frac{\text{Number of red cards}}{\text{Total number of cards}} = \frac{26}{52} = 0.5\]
Relative frequency (empirical) method. When outcomes are not equally likely, we estimate the probability from data: the proportion of times the event happened over many repetitions of the experiment. This is the frequentist view put into practice, and the estimate improves as the number of trials grows.
RELATIVE FREQUENCY PROBABILITY \[P(A)\approx\frac{Number\ of\ times\ A\ occurred}{Number\ of\ trials}\]
Example: A distributor wants to know how likely it is that an order arrives late. Deliveries are not equally likely to be on time, late, or damaged, so the classical method does not apply. Instead, the distributor looks at its records for the last \(600\) orders:
| Delivery outcome | Number of orders | Relative frequency |
|---|---|---|
| On time | 552 | \(552/600 = 0.92\) |
| Late | 18 | \(18/600 = 0.03\) |
| Damaged | 30 | \(30/600 = 0.05\) |
| Total | 600 | \(1\) |
The estimated probability that the next order arrives late is \(P(\text{late}) \approx 18/600 = 0.03\). It is an estimate: with records from \(6{,}000\) orders instead of \(600\), the distributor could pin it down more precisely. In Excel, with the outcomes in column A, =COUNTIF(A2:A601, "Late")/COUNTA(A2:A601) returns \(0.03\).
Subjective method. When there is no repeatable experiment and little or no data, a probability is assigned by reasoned judgment: an analyst or expert weighs the available evidence and experience. Subjective probabilities match the Bayesian view; they are not frequentist probabilities, because the event cannot be repeated to check them.
Example: A company is about to launch a new kind of product, so there is no sales history to count. The product manager still needs a probability that the launch succeeds. She weighs the evidence she has: a customer survey in which \(40\%\) of respondents said they would buy, the fact that two similar products from competitors succeeded while one failed, and her own experience with past launches. Putting this together, she assigns
\[P(\text{success}) = 0.35 \qquad P(\text{failure}) = 0.65\]
Another manager, weighing the same evidence differently, might say \(0.30\) or \(0.40\), and that is acceptable: subjective probabilities can differ from person to person, but they must still follow the basic requirements below (here, \(0.35 + 0.65 = 1\)). They should also change with new evidence. If a test launch in two cities sells better than expected, she might revise \(P(\text{success})\) up to \(0.55\).
Whichever method we use to assign probabilities (classical, relative frequency, or subjective), the result must satisfy two basic requirements. Any set of numbers that breaks either one is not a valid set of probabilities.
BASIC REQUIREMENTS FOR ASSIGNING PROBABILITIES
The two ends of the scale represent certainty. A probability of \(0\) means the event is impossible, and a probability of \(1\) means it is certain to happen. In both cases there is nothing left to guess.
Everything in between is uncertainty. The closer a probability is to \(0.5\), the less we know: at \(0.5\) the event is as likely to happen as not, like a coin flip. A probability of \(0.95\) or \(0.05\) still leaves some doubt, but we can act with a lot of confidence.
Information moves probabilities toward the extremes. As we learn more, we can say with greater confidence whether something will or will not happen. This is why businesses collect data: good information reduces uncertainty.
Example: A bank is deciding whether to approve a loan. Before it knows anything about the applicant, it uses the default rate for all borrowers, so \(P(\text{default}) = 0.10\). If the applicant has an excellent credit history, the estimate drops to about \(0.02\), close to certain the loan will be repaid. If instead the applicant has missed three payments on other loans, it rises to about \(0.70\), much closer to certain default. In each case, new information pushed the probability away from the uncertain middle and toward one of the extremes. Once the loan is fully repaid, the uncertainty is gone: \(P(\text{default}) = 0\).
When we run an experiment, one of its outcomes must occur. The probability that something in the sample space happens is therefore \(1\), and that total has to be shared among the outcomes. If the probabilities added up to less than \(1\), we would be leaving room for nothing to happen; if they added up to more, we would be counting some possibility more than once.
The outcomes of an experiment have two properties that make this work:
The same logic applies to any list of events, not just individual outcomes. Whenever a list of events is both mutually exclusive and exhaustive, exactly one of them will occur, so their probabilities add up to \(1\): something must happen, and only one thing can.
Example: Consider the four suits in a deck of cards (hearts, diamonds, clubs, spades). They are mutually exclusive, because a card cannot belong to two suits, and exhaustive, because every card belongs to one of them. Their probabilities add up to \(1\):
\[P(\text{hearts}) + P(\text{diamonds}) + P(\text{clubs}) + P(\text{spades}) = 4 \times \frac{13}{52} = 1\]
Both conditions matter. Hearts and spades alone are mutually exclusive but not exhaustive, so \(P(\text{hearts}) + P(\text{spades}) = 0.5\) falls short of \(1\). Hearts and red cards are not mutually exclusive, since every heart is also red, so adding \(P(\text{hearts}) + P(\text{red}) = 0.25 + 0.5 = 0.75\) counts the hearts twice.
A Venn diagram is a visual representation of sets and their relationships. It consists of overlapping circles, each representing a set. Below is an example:

The rectangle enclosing the circles represents the sample set (i.e., the 52 playing cards). Each circle represents a set or event The heart circle contains all 13 hearts; the same holds for the clubs circle. The area outside the circle that has hearts represents its complement, meaning elements that are clubs, spades, or diamonds. Since the two sets do not overlap, they are mutually exclusive (i.e., if a card is a heart, it can’t be a spade). The total area covered by both sets represents the union denoted by \(\cup\). This is the set that contains all elements that belong to at least one of the sets. Formally, we would represent this set as \((Hearts \cup Spades)\).

The overlapping region of two sets contains elements that belong to both sets. This region is called the intersection. The intersection contains elements that are Hearts and Jacks (i.e., the jack of hearts). Formally, we denote the intersection by \((Hearts \cap Jacks)\).
Probability rules help us quantify uncertainty, avoid cognitive biases, and make rational predictions based on available data. These rules allow us to make reliable decisions. Below we introduce some useful rules.
The Complement Rule states that:
THE COMPLEMENT RULE \[P(A^c)=1-P(A)\] where \(A^c\) is the complement of \(A\).
This rule is useful when calculating probabilities for complex scenarios, where finding the probability of the complement is easier than computing the event itself.
Example: The probability of drawing a spade, diamond or clubs can be estimated by just calculating the probability of drawing a heart. Since the probability of drawing a heart is \(P(Heart)=\frac{13}{52}\), then using the complement rule, \(P(Not \text{Heart})=1-\frac{13}{52}=\frac{39}{52}\).
De Morgan’s Laws relate the complement of a union to the complement of an intersection. They are handy for rewriting a complicated event as an easier-to-compute one.
DE MORGAN’S LAWS \[(A \cup B)^c = A^c \cap B^c \qquad\qquad (A \cap B)^c = A^c \cup B^c\] The complement of a union is the intersection of the complements, and the complement of an intersection is the union of the complements.
Example: Roll a die with \(A=\{1,2,3\}\) and \(B=\{2,4,6\}\). Their union is \(A \cup B=\{1,2,3,4,6\}\), so the complement is \((A \cup B)^c=\{5\}\). De Morgan’s Law gives the same answer from the complements: \(A^c \cap B^c=\{4,5,6\} \cap \{1,3,5\}=\{5\}\).
The Addition Rule states:
THE ADDITION RULE \[P(A \cup B)=P(A)+P(B)-P(A \cap B)\] where \(\cap\) is intersection and \(\cup\) is union.
This formula prevents double-counting when events overlap by including the intersection of events.
Example: Let’s calculate the probability of drawing either a heart or a face card. If we let event \(A\) be drawing a heart, and event \(B\) be drawing a face card, the probability is given by: \[P(A \cup B)=\frac{13}{52}+\frac{12}{52}-\frac{3}{52}=\frac{22}{52}\]
The Multiplication Rule states:
THE MULTIPLICATION RULE \[P(A \cap B)= P(A|B)P(B)\] where \(P(A|B)\) is the conditional probability.
Conditional probability measures the likelihood of an event occurring given that another event has already happened. If we reorganize the formula above we can get:
CONDITIONAL PROBABILITY \[P(A|B)=\frac{P(A \cap B)}{P(B)}\]
Note that this formula focuses on cases where Event B has already occurred. We can then find the fraction of the cases where Event A also happens.
Example: Suppose we randomly draw two cards from a standard deck of 52 cards, without replacement. The probability that both cards are aces is given by: \[P(Ace_1 \cap Ace_2)=P(Ace_1) \times P(Ace_2|Ace_1)= \frac {4}{52} \times \frac{3}{51}\]
Lastly, if events \(A\) and \(B\) are independent, we can simplify the multiplication rule to:
INDEPENDENCE \[P(A \cap B)= P(A)P(B)\]
Hence, we can define independence as: events \(A\) and \(B\) are independent if the occurrence (or non occurrence) of one does not affect the probability of the other occurring.
Example: Suppose we randomly draw a card from a standard deck, replace it, and then draw another card. The probability of drawing a heart in the first draw is \(13/52\), and the probability of drawing a spade in the second draw is also \(13/52\).
Example: Suppose that the weather is sunny; this has no effect on the outcome of a lottery. Hence, the weather and lottery outcomes are independent.
The Law of Total Probability states:
LAW OF TOTAL PROBABILITY \[P(A)=P(A \cap B)+P(A\cap B^c)\] which, using the multiplication rule, can also be written as \[P(A)=P(A|B)P(B)+P(A|B^c)P(B^c)\]
The formula mainly states that if an event \(A\) can occur in multiple mutually exclusive scenarios (or conditions), then the total probability of \(A\) is the sum of its probabilities under each scenario, weighted by how often those scenarios occur. The Venn diagram helps us visualize the law of total probability.

The image shows how event \(A\) can occur. Essentially, when \(B\) occurs \(A\) is possible in the left semicircle. When \(B^c\), then \(A\) is possible in the right semicircle. The addition of both semicircles, complete set \(A\).
Example: Consider the sets face cards and hearts. The entire set of hearts can be completed by taking face cards that are hearts, and non face cards (the complement of face cards) that are hearts.
The same idea extends to any number of scenarios. If \(B_1, B_2, \dots, B_k\) are mutually exclusive and together cover the whole sample space, the total probability of \(A\) is found by summing across all of them:
LAW OF TOTAL PROBABILITY (Several Events) \[P(A)=\sum_{i=1}^{k} P(A \cap B_i)=\sum_{i=1}^{k} P(A|B_i)\,P(B_i)\] where the events \(B_1, B_2, \dots, B_k\) are mutually exclusive and exhaustive (they partition the sample space).
Example: A factory makes a part on three machines. Machine 1 produces \(50\%\) of the parts with a \(2\%\) defect rate, Machine 2 produces \(30\%\) with a \(3\%\) defect rate, and Machine 3 produces \(20\%\) with a \(5\%\) defect rate. The overall probability that a randomly chosen part is defective is \[P(D)=(0.50)(0.02)+(0.30)(0.03)+(0.20)(0.05)=0.029\] so about \(2.9\%\) of all parts are defective.
Bayes’ Theorem states:
BAYES’ THEOREM \[P(A|B)= \frac{P(B|A)P(A)}{P(B)}\]
In essence, Bayes’ theorem is a way of updating probabilities in light of new evidence. The theorem tells us that the probability of an event A given that we have observed some evidence is equal to the prior probability (\(P(A)\)) times a factor that measures the informativeness of the observed evidence \(P(B|A)/P(B)\).
Example: Assume that a rare form of cancer affects 1% of the population. That is, 1 out of 100 people will likely be affected by it. There is a test for the cancer, but it is imperfect. In particular, if you don’t have cancer, the test says that you do 10% of the time (false positive rate). Also, if you do have cancer, it will tell you that you don’t have it 5% of the time. Let’s say you get tested, and your doctor calls you and says you have tested positive. How likely is it that you have this type of cancer?
We can use the Law of Total Probability and Bayes’ theorem to answer this question. Let’s first identify what we want to estimate. Since we need to figure out the likelihood of having cancer, given that we tested positive, we are looking for \(P(C|+)\). Now, we can set up Bayes’ rule to see what additional information we need to find our desired value. Bayes’ rule states:
\[P(C|+)= \frac{P(+|C)P(C)}{P(+)}\] Since the problem does not provide the \(P(+)\) we must estimate it with the Law of Total Probability, mainly:
\[P(+)=P(+|C)P(C)+P(+|NC)P(NC)=(0.95 \times 0.01)+(0.1 \times 0.99)=0.1085\] We can now use this probability to find our new belief that we have cancer:
\[P(C|+)= \frac{(0.95 \times 0.01) }{0.1085}=0.0876\] In conclusion, if we before thought we had a \(1\%\) chance of having the rare form of cancer, once we were told that we tested positive, we should update our belief to about \(8.76\%\).
A contingency table (or cross-tabulation) summarizes the relationship between two or more events (or variables). It is particularly helpful when analyzing frequencies of events and their relationship to other events.
The contingency table below shows the frequency of event A and B occurring/not occurring.
| \(B\) | \(B^c\) | Total | |
|---|---|---|---|
| \(A\) | 26 | 34 | 60 |
| \(A^c\) | 14 | 26 | 40 |
| Total | 40 | 60 | 100 |
We can see that event A occurs 60 times and that event B occurs 40 times. Most importantly, we can also see how the two events are not mutually exclusive, since A and B occur 26 times. The table can be easily translated to probabilities, by dividing each entry by the total amount of outcomes (i.e., 100). We call the table that includes probabilities the joint probability table.
| \(B\) | \(B^c\) | Total | |
|---|---|---|---|
| \(A\) | 0.26 | 0.34 | 0.60 |
| \(A^c\) | 0.14 | 0.26 | 0.40 |
| Total | 0.40 | 0.60 | 1 |
In the middle of the table we have the joint probabilities or intersections of events (i.e., \(A \cap B\), \(A \cap B^c\), etc.). The values in the totals columns or rows are the probabilities of each event happening (i.e., \(P(A)\), \(P(B^c)\), etc.). We call these probabilities marginal probabilities since they reside at the margin of the table.
Counting rules are useful in probability as they allow us to determine the number of ways events can occur without listing out every possibility. Essentially, they facilitate the enumeration of possible outcomes for large sets.
The most basic counting rule, and the one all the others build on, is the Fundamental Counting Principle (also called the Multiplication Principle). It states that if a task is completed in separate stages, the total number of ways to complete it is the product of the number of choices available at each stage.
THE FUNDAMENTAL COUNTING PRINCIPLE \[m_1 \times m_2 \times \dots \times m_k\] where \(m_i\) is the number of choices at stage \(i\), for a task made up of \(k\) stages.
Example: A restaurant offers \(3\) appetizers, \(4\) entrées, and \(2\) desserts. The number of different three-course meals is \(3 \times 4 \times 2 = 24\).
To illustrate the factorial, consider the number of ways that we can organize a group of three people (Ann, Beth, Charlie). The enumeration of all the possible orderings in this case is not that tedious and it can be shown in a table:
| Ordering | Order |
|---|---|
| 1 | (Ann, Beth, Charlie) |
| 2 | (Ann, Charlie, Beth) |
| 3 | (Beth, Ann, Charlie) |
| 4 | (Beth, Charlie, Ann) |
| 5 | (Charlie, Ann, Beth) |
| 6 | (Charlie, Beth, Ann) |
The table shows that there are six possible ways to organize a group of three people. This is really just the Fundamental Counting Principle at work: there are \(3\) choices for the first spot, then \(2\) remaining choices for the second, and only \(1\) left for the third, giving \(3 \times 2 \times 1 = 6\). The Factorial captures exactly this pattern: it counts how many ways a group of \(N\) objects can be organized. Formally, we write:
THE FACTORIAL \[Factorial=N!\] where the operator \(!\) means to take the product of all positive integers from \(N\) down to \(1\). Note that \(0!\) is equal to one.
If we then try to find in how many ways we can organize a group of three people, we can avoid enumeration by calculating \(3!=3 \times 2 \times 1=6\).
Example: Let’s assume that a manager has locked down the first five batters in a given game, but is undecided on the batting order. The number of ways that he can organize the five batters is given by \(5!=5 \times 4 \times 3 \times 2 \times 1=120\).
The Combination function counts the number of ways to choose \(x\) objects from a total of \(n\) objects. The order in which the \(x\) objects are listed does not matter. If repetition is not allowed use:
COMBINATIONS (No Repetition) \[C(n, x)= \frac{n!}{(n-x)!x!}\]
Example: Let’s imagine visiting the Mexican restaurant Chipotle. We are choosing a bowl for lunch and we can choose from the following items: beans, chicken, rice, pico, guac, and beef. If we are only allowed to select four items out of the six, in how many ways can we construct a bowl? Below we use the combination formula to show that there are \(15\) possible bowls. \[C(6, 4)= \frac{6!}{(6-4)!4!}=\frac{6 \times 5}{2}=15\]
There are situations when we wish to count all possible combinations inclusive of repetition of elements. If repetition is allowed we use the following formula:
COMBINATIONS (With Repetition) \[C(n+x-1,x)=\frac{(x+n-1)!}{(n-1)!x!}\]
Example: Imagine you visit the ice cream store. You want a three scoop dish of ice cream. In how many ways can you choose your dish if the flavors are vanilla, chocolate, smores, brownie, and chips? Below we use the combination formula with repetition to show that the answer is \(35\). \[C(7,3)=\frac{(5+3-1)!}{(5-1)!3!}= 7 \times 5=35\]
The Permutation function also counts the number of ways to choose \(x\) objects from a total of \(n\) objects. However, the order in which the \(x\) objects are listed does matter. Note that since order matters, the permutation will always count at least as many outcomes as the combination. When repetition is not allowed, permutations can be counted by:
PERMUTATIONS (No Repetition) \[P(n,x)= \frac{n!}{(n-x)!}\]
Example: The Olympic medal finals for the 100 meters sprint has eight runners competing for three medals. How many medal podiums are possible? There are a total of \(336\). Below you can see the calculation. \[P(8,3)= \frac{8!}{5!}= 8 \times 7 \times 6=336\]
As with combinations, there might be situations where we wish to count when repetition is allowed. If this is the case, use:
PERMUTATIONS (With Repetition) \[P(n,x)=n^x\]
Example: You are trying to guess the code of a lock with three dials. Since each dial has 10 numbers to choose from, there are \(10^3=1000\) possible permutations.
Excel has a collection of functions that make counting and probability calculations straightforward.
Contingency tables: Use a PivotTable (Insert → PivotTable) to build a contingency table from raw data. Place one variable in the Rows area, the other in the Columns area, and any field in the Values area as a count. To turn the counts into a joint probability table, use Value Field Settings → Show Values As → % of Grand Total. You can also count a specific combination of conditions directly with =COUNTIFS(range1, criteria1, range2, criteria2).
Factorial: The =FACT(number) function returns the factorial of a number. For example, =FACT(5) returns \(120\).
Combinations: The =COMBIN(number, number_chosen) function counts the number of combinations without repetition. Use =COMBINA(number, number_chosen) when repetition is allowed.
Permutations: The =PERMUT(number, number_chosen) function counts the number of permutations without repetition. Use =PERMUTATIONA(number, number_chosen) when repetition is allowed.
Below is a list of the Excel functions used in this section:
=COUNTIFS(range1, criteria1, ...) counts the observations that satisfy several conditions at once, useful for building contingency tables.
=FACT(number) returns the factorial of a number.
=COMBIN(number, number_chosen) returns the number of combinations without repetition; =COMBINA(number, number_chosen) allows repetition.
=PERMUT(number, number_chosen) returns the number of permutations without repetition; =PERMUTATIONA(number, number_chosen) allows repetition.
Use only when every outcome in the sample space is equally likely.
Estimates the probability from past data. The estimate improves as the number of trials grows.
Every assignment must satisfy both, including subjective ones. 0 and 1 mean certainty; values in between mean uncertainty.
The chance an event does not occur. Often easier to compute than the event itself.
Subtract the overlap so it is not counted twice. If A and B are mutually exclusive, P(A ∩ B) = 0.
The chance both occur. If A and B are independent, P(A ∩ B) = P(A) · P(B).
Adds the chance of A across scenarios Bᵢ that are mutually exclusive and exhaustive.
Turns a prior P(A) into a posterior after observing evidence B. Find P(B) with the law of total probability.
Count outcomes without listing them. Combinations ignore order; permutations respect it.
Test the main ideas of the chapter before moving on to the exercises. Pick an answer to see whether you are right and why.
A distributor reviews its last \(600\) orders and finds that \(18\) arrived late. It uses \(18/600=0.03\) as the probability that the next order is late. Which method is this?
The probability is estimated from how often the event happened in past trials. The classical method needs equally likely outcomes, which deliveries are not.
An experiment has three outcomes. Which of these is not a valid assignment of probabilities?
The probabilities of all outcomes must add up to \(1\), and \(0.5+0.4+0.2=1.1\). A probability of \(0\) is allowed; it means the outcome is impossible.
\(P(A)=0.3\), \(P(B)=0.5\), and \(P(A \cap B)=0.1\). What is \(P(A \cup B)\)?
By the addition rule, \(P(A \cup B)=0.3+0.5-0.1=0.7\). Subtracting the overlap keeps it from being counted twice.
Events \(A\) and \(B\) are independent, with \(P(A)=0.4\) and \(P(B)=0.5\). What is \(P(A \cap B)\)?
For independent events, \(P(A \cap B)=P(A)P(B)=0.4 \times 0.5=0.20\). A value of \(0\) would describe mutually exclusive events, which cannot both happen.
Events \(A\) and \(B\) are mutually exclusive, with \(P(A)=0.3\) and \(P(B)=0.4\). What is \(P(A \cap B)\)?
Mutually exclusive events cannot happen together, so their intersection has probability \(0\). The value \(0.12\) would apply only if they were independent.
Use the chapter’s joint probability table: \(P(A \cap B)=0.26\) and \(P(B)=0.40\). What is \(P(A|B)\)?
\(P(A|B)=\frac{P(A \cap B)}{P(B)}=\frac{0.26}{0.40}=0.65\). Knowing that \(B\) occurred restricts us to the \(B\) column of the table.
Of a store’s transactions, \(2\%\) are fraudulent. A screening system flags \(90\%\) of fraudulent transactions and \(5\%\) of legitimate ones. If a transaction is flagged, about how likely is it to be fraudulent?
\(P(\text{flag})=0.90(0.02)+0.05(0.98)=0.067\), by the law of total probability. Then \(P(\text{fraud}|\text{flag})=\frac{0.90(0.02)}{0.067}\approx0.27\). Most flags are false alarms because fraud is rare.
A manager picks \(3\) of her \(8\) employees to form a committee. The order of selection does not matter. How many different committees are possible?
Order does not matter, so use combinations: \(C(8,3)=\frac{8!}{5!\,3!}=56\). The value \(336\) is \(P(8,3)\), which counts ordered selections.
The following exercises will help you practice some probability concepts and formulas. In particular, the exercises work on:
Calculating simple probabilities.
Applying probability rules.
Using counting rules.
Answers are provided below. Try not to peak until you have a formulated your own answer and double checked your work for any mistakes. You can also find every answer on the Answers to Exercises page.
For the following exercises, make your calculations by hand and verify results with a calculator or Excel.
For the following exercises, make your calculations by hand and verify results with a calculator or Excel.
From the multiplication rule, \[P(A|B)*P(B)=P(A \cap B)\]
substituting values yields: \[P(A \cap B)=0.45*0.3=0.135\] From the addition rule, \[P(A \cup B)=P(A)+P(B)-P(A \cap B)\] substituting yields: \[P(A \cup B)=0.65+0.3-0.135=0.815\] From the multiplication rule once again, \[P(B|A)=\frac{P(A \cap B)}{P(A)}\] substituting values yields, \[P(B|A)=0.135/0.65=0.2076923\]
From the complement rule we have that \(P(A^c)=0.6\) and \(P(B^c)=0.5\).
Using the multiplication rule, \[P(A^c|B^c)=\frac{P(A^c \cap B^c)}{P(B^c)}\] substituting yields: \[P(A^c|B^c)=0.24/0.5=0.48\] From the addition rule, \[P(A^c \cup B^c)=P(A^c)+P(B^c)-P(A^c \cap B^c)\] substituting yields: \[P(A^c \cup B^c)=0.6+0.5-0.24=0.86\] The event that has no elements of \(A\) or \(B\) is given by \(P(A^c \cap B^c)\). Therefore \(P(A \cup B)=1-0.24=0.76\) has all the elements of A and B.
In short, the problem states \(P(A)=0.4\), \(P(B)=0.6\), and \(P(A|B)=0.5\). Where \(A\) and \(B\) are events of stocks rising in price. The question asks for: \[P(A \cup B)=P(A)+P(B)-P(A \cap B)\] Using the multiplication rule, \[P(A \cap B)=0.5*0.6=0.3\] Hence, \[P(A \cup B)=0.4+0.6-0.3=0.7\] The events are not mutually exclusive since \(P(A \cap B)=0.3 \neq 0\). The events are also not independent since \(P(A|B)=0.5 \neq 0.4=P(A)\).
For the following exercises, make your calculations by hand and verify results with a calculator or Excel. This time, each problem gives you a different set of starting information. Decide which rule connects what you know to what you need.
Rearranging the addition rule, \[P(A \cap B)=P(A)+P(B)-P(A \cup B)=0.5+0.4-0.7=0.2\] Using the conditional probability formula, \[P(A|B)=\frac{P(A \cap B)}{P(B)}=\frac{0.2}{0.4}=0.5 \qquad P(B|A)=\frac{P(A \cap B)}{P(A)}=\frac{0.2}{0.5}=0.4\] The events are not mutually exclusive since \(P(A \cap B)=0.2 \neq 0\). They are independent since \(P(A|B)=0.5=P(A)\) (equivalently, \(P(A)P(B)=0.5 \times 0.4=0.2=P(A \cap B)\)).
Since the events are independent, \[P(A \cap B)=P(A)P(B)=0.3 \times 0.6=0.18\] From the addition rule, \[P(A \cup B)=0.3+0.6-0.18=0.72\] Independence also means that knowing \(B\) happened does not change the probability of \(A\), so \(P(A|B)=P(A)=0.3\). Finally, by De Morgan’s law, \(A^c \cap B^c=(A \cup B)^c\), so \[P(A^c \cap B^c)=1-P(A \cup B)=1-0.72=0.28\] As a check, the complements of independent events are also independent: \(P(A^c)P(B^c)=0.7 \times 0.4=0.28\).
Mutually exclusive events cannot happen together, so \(P(A \cap B)=0\). The addition rule then simplifies to \[P(A \cup B)=P(A)+P(B)=0.25+0.45=0.70\] Since \(A\) and \(B\) never occur together, \(P(A|B)=0/0.45=0\). The probability that neither occurs is the complement of the union: \(1-0.70=0.30\).
The events are not independent. If \(B\) happens, the probability of \(A\) drops from \(0.25\) to \(0\), so knowing \(B\) changes the probability of \(A\). Mutually exclusive events (with positive probabilities) are always dependent.
In short, \(P(M)=0.60\), \(P(P)=0.05\), and \(P(M \cap P)=0.02\). From the addition rule, \[P(M \cup P)=0.60+0.05-0.02=0.63\] The probability that a mobile visitor buys is \[P(P|M)=\frac{P(M \cap P)}{P(M)}=\frac{0.02}{0.60}=0.033\] Visitors who do not use a mobile phone make up \(P(M^c)=0.40\) of the traffic, and they account for the remaining \(0.05-0.02=0.03\) of purchases, so \[P(P|M^c)=\frac{0.03}{0.40}=0.075\] Purchasing is not independent of the device: \(P(P|M)=0.033 \neq 0.05=P(P)\). Desktop visitors are more than twice as likely to buy as mobile visitors.
Since the failures are independent, \[P(\text{both fail})=0.02 \times 0.02=0.0004\] “At least one is working” is the complement of “both fail”: \[P(\text{at least one works})=1-0.0004=0.9996\] “Exactly one fails” can happen in two mutually exclusive ways: the first fails and the second works, or the first works and the second fails. Adding them, \[P(\text{exactly one fails})=(0.02)(0.98)+(0.98)(0.02)=0.0392\]
| \(B\) | \(B^c\) | |
| \(A\) | 26 | 34 |
| \(A^c\) | 14 | 26 |
Below is the joint probability table. The \(P(A)=0.26+0.34=0.6\), \(P(A \cap B)=0.26\), \(P(A|B)=0.26/0.4=0.65\), and \(P(B|A^c)=0.14/0.4=0.35\). Events \(A\) and \(B\) are not independent since \(P(A) \neq P(A|B)\). The events are not mutually exclusive since \(P(A \cap B)=0.26 \neq 0\).
| \(B\) | \(B^c\) | Total | |
| \(A\) | 0.26 | 0.34 | 0.6 |
| \(A^c\) | 0.14 | 0.26 | 0.4 |
| Total | 0.4 | 0.6 | 1 |
You will need the Crash data set to answer this question. The data shows information on several car crashes. Specifically, whether the crash was Head-On or Not Head-On and whether there was Daylight or No Daylight. You can download it here:
The contingency table below shows the counts for each combination of Crash Type and Light Condition:
| Daylight | Not Daylight | Total | |
|---|---|---|---|
| Head-on | 166 | 108 | 274 |
| Not Head-On | 3258 | 1326 | 4584 |
| Total | 3424 | 1434 | 4858 |
In Excel, build a PivotTable: drag Crash Type to the Rows area, Light Condition to the Columns area, and either field to the Values area as a Count. To turn the counts into a joint probability table, use Value Field Settings → Show Values As → % of Grand Total. You can also count any single cell directly. Assuming Crash Type is in column B and Light Condition in column C:
=COUNTIFS(B:B, "Head-on", C:C, "Daylight") → \(166\)a) The probability of a Head-On crash is:
\[P(\text{Head-on})=\frac{166+108}{4858}=\frac{274}{4858}=0.056\]
b) The probability that the crash is in daylight is:
\[P(\text{Daylight})=\frac{166+3258}{4858}=\frac{3424}{4858}=0.70\]
c) The probability that a crash is Head-On given daylight is:
\[P(\text{Head-on}\mid\text{Daylight})=\frac{P(\text{Head-on}\cap\text{Daylight})}{P(\text{Daylight})}=\frac{166}{3424}=0.048\]
In Excel these follow from the PivotTable counts. For example, part a) is =COUNTIF(B:B,"Head-on") / COUNTA(B2:B4859).
The two variables are dependent since \(P(\text{Head-on}\mid\text{Daylight}) \neq P(\text{Head-on})\), that is \(0.048 \neq 0.056\).
\(P(A^c)=1-P(A)=1-0.7=0.3\), \(P(A \cap B)=𝑃(𝐵|𝐴)𝑃(𝐴) = 0.55(0.70) = 0.385\), \(P(A^c \cap B)=𝑃(B|A^c)𝑃(A^c) = 0.10(0.30) = 0.03\), \(P(B)= 𝑃(A \cap B) + 𝑃(𝐴^c \cap 𝐵) = 0.385 + 0.03 = 0.415\), and \(P(A|B)= \frac{𝑃(A \cap B)}{P(B)}=0.385/0.415=0.9277\).
Let the event of failing be \(F\), the event of not failing be \(NF\), the event of having a tutor be \(T\), and the event of not having a tutor be \(NT\). The probability of failing the course is \(0.25\). \((𝐹) = 𝑃(𝐹 \cap 𝑇) + 𝑃(𝐹 \cap 𝑇^c) = 𝑃(𝐹|𝑇)𝑃(𝑇) + 𝑃(𝐹|𝑇^c)𝑃(𝑇^c) = 0.10(0.50) + 0.40(0.50) = 0.05 + 0.20 = 0.25\) The probability of not having a tutor, given that she failed the course is \(0.2\). \(P(𝑇|𝐹) = \frac{𝑃(𝐹\cap𝑇)}{𝑃(𝐹\cap𝑇)+𝑃(𝐹 \cap𝑇^c)}= 0.05/0.25 = 0.20\).
\(3!=3 \times 2 \times 1=6\), \(4!=4 \times 3 \times 2 \times 1=24\), \(C_6^8=28\), and \(P_6^8=20{,}160\).
In Excel:
=FACT(3) → \(6\)=FACT(4) → \(24\)=COMBIN(8, 6) → \(28\)=PERMUT(8, 6) → \(20{,}160\)If order doesn’t matter, there are \(252\) ways. If order matters, then there are \(30{,}240\) ways.
In Excel:
=COMBIN(10, 5) → \(252\)=PERMUT(10, 5) → \(30{,}240\)The order in which the two people are picked does not matter, so we count combinations. The number of possible pairs is \[C_2^6=\frac{6!}{2!\,4!}=15\] Every pair is equally likely, so we can use the classical method. The number of pairs made up of two analysts is \(C_2^4=6\), so \[P(\text{both analysts})=\frac{6}{15}=0.4\] “At least one intern” is the complement of “both analysts”: \[P(\text{at least one intern})=1-0.4=0.6\] We can check this by counting directly: there are \(C_1^4 \times C_1^2=8\) pairs with one analyst and one intern, and \(C_2^2=1\) pair with both interns, so \((8+1)/15=0.6\).
In Excel:
=COMBIN(6, 2) → \(15\)=COMBIN(4, 2)/COMBIN(6, 2) → \(0.4\)=1 - COMBIN(4, 2)/COMBIN(6, 2) → \(0.6\)