Mixed Bernoulli and Binomial Probability Workshop

Choosing the Model and Defining Success

Bernoulli and binomial questions begin with a modeling decision: is the variable one binary outcome, or is it a count of successes across several trials? After choosing the model, define “success” precisely. It is simply the outcome being counted, not necessarily a favorable outcome. A good definition prevents verbal phrases such as “at least,” “no more than,” and “between” from being translated into the wrong probability.

The Bernoulli and binomial models are aligned with the SOA VEE Mathematical Statistics syllabus. The earlier material on Bernoulli and binomial models establishes their standard assumptions and probability rules; here the emphasis is on applying them together to contextual problems.

Definition: A Bernoulli random variable $I$ records one trial with two possible outcomes. If the designated success occurs with probability $p$, then $I=1$ for success and $I=0$ for failure, where $0\le p\le 1$. Thus $P(I=1)=p$ and $P(I=0)=1-p$. A binomial random variable $X\sim\operatorname{Bin}(n,p)$ counts successes in $n$ independent Bernoulli trials, each having the same success probability $p$, where $n$ is a fixed nonnegative integer.

A single Bernoulli trial is also a binomial model with $n=1$. For $n>1$, the binomial model counts the number of successes, which can take any integer value from $0$ through $n$. All trials must use the same definition of success. For a binomial model, the number of trials is fixed in advance, the outcomes are independent, and the probability of success is constant from trial to trial. If any of these assumptions fail, the binomial probability formula may not apply.

The probability mass function is

P(X=k)=\binom{n}{k}p^k(1-p)^{n-k},\qquad k=0,1,\ldots,n.

It is important to state the support $k=0,\ldots,n$. The formula gives probabilities only for those values; a count outside that range is impossible. For integer $r$, the cumulative and upper-tail probabilities can be written with explicit boundary cases:

P(X\le r)= \begin{cases} 0,&r<0,\\ \displaystyle\sum_{k=0}^{r}\binom{n}{k}p^k(1-p)^{n-k},&0\le r<n,\\ 1,&r\ge n, \end{cases} \qquad P(X\ge r)= \begin{cases} 1,&r\le 0,\\ \displaystyle\sum_{k=r}^{n}\binom{n}{k}p^k(1-p)^{n-k},&1\le r\le n,\\ 0,&r>n. \end{cases}

For $0\le r\le n$, the complementary form $P(X\ge r)=1-P(X\le r-1)$ is often quicker. For an inclusive interval within the support, $P(a\le X\le b)=\sum_{k=a}^{b}P(X=k)$ when $0\le a\le b\le n$. If the requested interval extends beyond the support, only the values from $0$ to $n$ can contribute.

Why the Binomial Formula Counts the Right Outcomes

For a specified pattern with exactly $k$ successes and $n-k$ failures, independence gives probability $p^k(1-p)^{n-k}$. There are $\binom{n}{k}$ different patterns with $k$ successes: one chooses which $k$ of the $n$ trial positions are successes. These patterns are disjoint, so adding their probabilities gives the binomial formula.

P(X=k)=\underbrace{\binom{n}{k}}_{\text{patterns with }k\text{ successes}} \underbrace{p^k(1-p)^{n-k}}_{\text{probability of each pattern}}.

The same reasoning explains why an event such as “at least two successes” includes multiple values of $k$, rather than just the probability of exactly two. In a worked problem, first write the event using $X$, then identify the exact values or complementary values required.

Worked Applications

Worked Example: One Application and a Bernoulli Outcome

In a hypothetical review program, one application is selected. The probability it requires a manual document check is $0.08$. Let $I=1$ if the selected application requires the check and $I=0$ otherwise. A check costs a hypothetical \$250. Find the distribution of $I$, the probability that no check is needed, and the probability that the check cost is at most \$0.

Method and model. There is one application and one binary event, so use a Bernoulli model with success defined as “requires a manual document check.” Thus $I\sim\operatorname{Bernoulli}(p)$ with $p=0.08$; the failure probability is $1-p=0.92$.

Distribution. By the definition of $I$, $P(I=1)=0.08$ and $P(I=0)=0.92$. If $C$ denotes the check cost, then $C=250I$, so $C=250$ when a check is needed and $C=0$ otherwise.

No check. The event is $I=0$, so $P(\text{no check})=P(I=0)=0.92$.

Cost at most \$0. The only possible costs are \$0 and \$250. Therefore $C\le 0$ means $C=0$, which is the no-check event, and $P(C\le 0)=0.92$. The answers, to four decimal places, are $0.9200$ and $0.9200$.

Check and common mistake avoided. The two probabilities sum to $0.08+0.92=1$, as required. A common error is to use $0.08$ for “no check” because the check event was called success. Success is only a label: here “no check” is the failure outcome, with probability $1-p=0.92$.

Worked Example: Exactly Two or At Least Two Flags

In a hypothetical quality audit, five independently selected records each have probability $0.20$ of containing a specified coding flag. Let $X$ be the number of flagged records. Find (a) the probability of exactly two flags and (b) the probability of at least two flags.

Method and model. Define success as “the record contains the specified flag.” There are a fixed $n=5$ records, the flag probability is $p=0.20$ for every record, and the records are assumed independent. Thus $X\sim\operatorname{Bin}(5,0.20)$.

(a) Exactly two. Apply the binomial mass function at $k=2$:

\begin{aligned} P(X=2) &=\binom{5}{2}(0.20)^2(0.80)^3\\ &=10(0.04)(0.512)\\ &=0.2048. \end{aligned}

So the probability of exactly two flags is $0.2048$.

(b) At least two. The event $X\ge 2$ contains counts $2,3,4,5$. Its complement is $X\le 1$, which contains counts $0,1$. It is shorter to calculate that complement:

\begin{aligned} P(X\ge 2) &=1-P(X=0)-P(X=1)\\ &=1-(0.80)^5-\binom{5}{1}(0.20)(0.80)^4\\ &=1-0.32768-0.40960\\ &=0.26272. \end{aligned}

Rounded to four decimal places, the answer is $0.2627$. As a check, the event $X\ge2$ is contained in the event $X\ge1$; the calculation also gives $P(X\ge1)=1-0.32768=0.67232$, which is larger than $0.26272$, as it should be.

Common mistake avoided. Using $1-P(X=0)=0.67232$ for “at least two” would include the case $X=1$. The correct complement of $X\ge2$ is $X\le1$, not merely $X=0$.

Worked Example: A Count Between Two Bounds

A hypothetical monitoring team checks six independently operating sensors. Each sensor has probability $0.25$ of triggering a specified alert during the check. Let $X$ be the number of alerts. Find (a) the probability of exactly two alerts and (b) the probability of between two and four alerts, inclusive.

Method and model. Define success as “the sensor triggers the specified alert.” The trial count is fixed at $n=6$, each success probability is $p=0.25$, and the six outcomes are assumed independent. Therefore $X\sim\operatorname{Bin}(6,0.25)$, with failure probability $0.75$.

(a) Exactly two. Substitute $k=2$ in the binomial mass function:

\begin{aligned} P(X=2) &=\binom{6}{2}(0.25)^2(0.75)^4\\ &=15(0.0625)(0.31640625)\\ &=0.296630859375. \end{aligned}

Rounded to four decimal places, this is $0.2966$.

(b) Two through four, inclusive. The requested event is $X=2$, $X=3$, or $X=4$. Its complement consists of $X=0,1,5,6$. Compute the complement using the mass function:

\begin{aligned} P(X=0)&=(0.75)^6=0.177978515625,\\ P(X=1)&=\binom{6}{1}(0.25)(0.75)^5=0.35595703125,\\ P(X=5)&=\binom{6}{5}(0.25)^5(0.75)=0.00439453125,\\ P(X=6)&=(0.25)^6=0.000244140625. \end{aligned}

Add the four excluded probabilities and subtract from one:

\begin{aligned} P(2\le X\le4) &=1-(0.177978515625+0.35595703125\\ &\qquad\quad+0.00439453125+0.000244140625)\\ &=1-0.53857421875\\ &=0.46142578125. \end{aligned}

Rounded to four decimal places, the probability is $0.4614$. As a check, the probability is positive and less than one, and the included values $2,3,4$ are precisely the values left after excluding $0,1,5,6$ from the support $0,\ldots,6$.

Common mistake avoided. “Between two and four” is inclusive here, so both endpoints must be included. Taking only $P(X=3)$, or omitting $P(X=2)$ or $P(X=4)$, would answer a different event.

Translating Words Into Events

For a binomial count $X$, write the event before choosing a formula. The wording and its mathematical translation should agree exactly:

Wording	Event	Useful calculation
Exactly $r$	$X=r$	One probability mass
At most $r$, no more than $r$	$X\le r$	Add masses from $0$ through $r$
Fewer than $r$	$X<r$, or $X\le r-1$ for integer counts	Add masses through $r-1$
At least $r$	$X\ge r$	Often use $1-P(X\le r-1)$
More than $r$	$X>r$, or $X\ge r+1$ for integer counts	Often use $1-P(X\le r)$
From $a$ through $b$, inclusive	$a\le X\le b$	Add the masses for $a,\ldots,b$

The bounds must also be interpreted relative to the support. For instance, if $r\ge n$, then $P(X\le r)=1$; the count cannot exceed $n$. If $r>n$, then $P(X\ge r)=0$. These boundary cases matter when using a formula or cumulative-probability table: do not evaluate a binomial mass at an impossible count.

Common Errors and a Reliable Check

A particularly common error is confusing “at least one” with “at least two.” For the five-record audit above, the wrong calculation $1-P(X=0)=1-0.32768=0.67232$ answers $P(X\ge1)$, not $P(X\ge2)$. The correction is to remove both excluded counts, zero and one:

P(X\ge2)=1-P(X=0)-P(X=1) =1-0.32768-0.40960 =0.26272.

Before finalizing a result, check the model, event, and arithmetic separately:

Model: Is the variable a single binary outcome or a count over a fixed number of trials? For the binomial case, are independence and a common success probability justified?
Event: Have you defined success and translated the requested wording into the correct equality or inequality?
Support: Are all summed values between $0$ and $n$? Are the boundary cases handled if a bound lies outside that range?
Probability: Is the answer between zero and one, and does the complement describe exactly the outcomes not requested?

The arithmetic can be correct while the model or event is wrong. In particular, a binomial calculation is not appropriate just because a problem mentions a number of people or objects; the trials must satisfy the model’s assumptions. Likewise, the word “success” does not tell you which outcome the question asks for. Define it in the context, write the event, and only then calculate.

Key takeaway: Use a Bernoulli model for one binary trial and a binomial model for the count of successes in a fixed number of independent trials with a common success probability. Define success explicitly, translate the verbal event into notation, and respect the support $0,\ldots,n$ when computing exact, cumulative, or complementary probabilities.

Check Your Understanding

For each problem, state the relevant event before calculating. Each scenario and set of values is hypothetical.

One sensor has probability $0.13$ of issuing a warning during a test. Let $I=1$ mean it issues a warning. What is the distribution of $I$, and what is the probability it issues no warning? (Answer: $I\sim\operatorname{Bernoulli}(0.13)$; $0.8700$.)
Four independently selected files each have probability $0.30$ of being incomplete. Let $X$ count incomplete files. Find the probability exactly one is incomplete. (Answer: $X\sim\operatorname{Bin}(4,0.30)$; $0.4116$.)
Five independent equipment checks each have probability $0.40$ of detecting a defect. Find the probability of at most one detection. (Answer: $0.3370$.)
Eight independent applicants each have probability $0.10$ of needing an additional document. Find the probability at least two need one. (Answer: $0.1869$.)
Seven independently inspected packages each have probability $0.20$ of having a seal issue. Find the probability that one through three packages, inclusive, have an issue. (Answer: $0.7569$.)

Wording	Event	Useful calculation
Exactly \(r\)	\(X=r\)	One probability mass
At most \(r\), no more than \(r\)	\(X\le r\)	Add masses from \(0\) through \(r\)
Fewer than \(r\)	\(X<r\), or \(X\le r-1\) for integer counts	Add masses through \(r-1\)
At least \(r\)	\(X\ge r\)	Often use \(1-P(X\le r-1)\)
More than \(r\)	\(X>r\), or \(X\ge r+1\) for integer counts	Often use \(1-P(X\le r)\)
From \(a\) through \(b\), inclusive	\(a\le X\le b\)	Add the masses for \(a,\ldots,b\)

What You'll Learn