Binomial Mean From Indicator Variables

Why Indicators Make the Binomial Mean Clear

A binomial random variable counts how many successes occur in a fixed number of trials. Its mean has a compact formula, but the formula is more useful when its reasoning is clear: each trial contributes either zero or one to the total, and the expected contribution from each trial is the success probability. Adding those expected contributions gives the expected count.

This argument uses linearity of expectation. It does not require the trial outcomes to be independent to calculate the expected sum. Independence is part of the usual binomial model, but it is not the reason the means add. Keeping those two facts separate helps prevent a common error: treating a mean calculation as if it required multiplying probabilities across trials.

Definition: For a specified event $A$, its indicator variable is $I_A=1$ if $A$ occurs and $I_A=0$ otherwise. Its expectation is $E[I_A]=P(A)$. For trial $i$, write $I_i=1$ when trial $i$ is a success and $I_i=0$ when it is a failure.

From Trial Indicators to the Binomial Mean

Suppose there are $n$ trials. For each trial $i$, define $I_i$ to indicate success on that trial. The number of successes is the sum of these indicators: a successful trial adds one, and a failed trial adds zero. In a binomial model, the trials are independent and each has the same success probability $p$. Thus every indicator has expectation $p$.

X=\sum_{i=1}^{n} I_i,\qquad I_i\sim\operatorname{Bernoulli}(p)

Linearity of expectation states that the expectation of a finite sum is the sum of the expectations. This statement holds whether or not the summands are independent, provided the expectations exist. Applying it to the indicator sum gives the binomial mean.

E[X] =E\!\left[\sum_{i=1}^{n} I_i\right] =\sum_{i=1}^{n}E[I_i] =\sum_{i=1}^{n}p =np

Theorem: If $X\sim\operatorname{Binomial}(n,p)$, where $n$ is a fixed nonnegative integer and each of the $n$ independent trials has success probability $p$, then $E[X]=np$. The result is the expected number of successes, not the probability of a particular outcome.

The same indicator argument gives the expected number of failures. If $F$ is the failure count, then each failure indicator has probability $1-p$, so $E[F]=n(1-p)$. Because every trial is exactly one success or one failure, $X+F=n$. The two expected counts therefore add to $n$, as they should.

E[F]=n(1-p),\qquad E[X]+E[F]=np+n(1-p)=n

Why the Derivation Works

The key step is to represent a count as a sum of zero-one contributions. Consider two trials first. If their indicators are $I_1$ and $I_2$, then the total count is $I_1+I_2$. Across repeated versions of the experiment, the average value of that total is the average value of $I_1$ plus the average value of $I_2$. More generally, for a finite sum, expectation distributes over addition.

Define one indicator per trial.
Set $I_i=1$ for a success on trial $i$ and $I_i=0$ for a failure. Then the observed success count is exactly $\sum_{i=1}^{n}I_i$.

Find each indicator’s expectation.
Since $I_i$ is one with probability $p$ and zero with probability $1-p$, $E[I_i]=1\cdot p+0\cdot(1-p)=p$.

Add the expectations.
Linearity gives $E[\sum I_i]=\sum E[I_i]=np$. No step in this expectation calculation multiplies the trial probabilities.

This derivation also clarifies what is and is not needed. To call the count binomial, the model requires a fixed number of independent trials with a common success probability. To add the expectations of indicators, linearity alone is sufficient; independence is not needed for that operation. If success probabilities differ between trials, the expected total is the sum of the individual probabilities, but the binomial mean formula $np$ does not apply with a single common $p$.

Applying the Mean

A mean can be fractional even though a count can only be a whole number in any one run. For example, a mean of $10.8$ does not mean that one run produces $10.8$ successes. It means that the average count over many repetitions of the model is $10.8$. Do not round an expected count unless a question explicitly asks for a rounded report.

Worked Example: Expected Passes and Failures

A hypothetical quality-control procedure checks 12 sensors. Assume each sensor independently passes with probability $0.90$, with the same pass probability for all sensors. Let $X$ be the number that pass. Find the expected number of passes and failures, and check the two answers together. The distribution is $X\sim\operatorname{Binomial}(12,0.90)$. No interest rate or mortality assumption applies.

Method: indicator variables. Define $I_i=1$ if sensor $i$ passes and $I_i=0$ otherwise. Then $X=I_1+\cdots+I_{12}$, and each indicator has expectation $0.90$. Therefore,

$$ E[X]=12(0.90)=10.80 $$

The failure probability for one sensor is $1-0.90=0.10$. If $F$ is the failure count, its mean is

$$ E[F]=12(0.10)=1.20 $$

Check. Every sensor either passes or fails, so the expected counts should sum to the total number of sensors:

$$ 10.80+1.20=12 $$

The check holds. The value $10.80$ is an average across repetitions, not a possible pass count in a single inspection. A common mistake is to round it to 11 before using it in later calculations; retain the exact mean unless rounding is requested.

Worked Example: Expected Timely Applications

A hypothetical service team receives 40 applications. Under a planning model, each application is independently processed on time with probability $0.75$. Let $T$ be the number processed on time. Find the expected number processed on time and the expected number not processed on time. The model is $T\sim\operatorname{Binomial}(40,0.75)$. No interest rate or mortality assumption applies.

Method: apply the binomial mean to the success count. Here a success means an application is processed on time. There are $n=40$ trials and $p=0.75$, so

$$ E[T]=np=40(0.75)=30 $$

The probability of not being processed on time is $1-0.75=0.25$. Thus the expected count in that category is

$$ E[40-T]=40(0.25)=10 $$

Check. The categories partition the 40 applications, and $30+10=40$. The answer describes expected counts; it does not guarantee that exactly 30 applications will be timely in a particular batch.

Common mistake avoided: do not calculate $0.75^{40}$ when asked for the expected number processed on time. That power is associated with all 40 applications being timely, not with the expected count. The expected count comes from adding 40 individual success probabilities.

Worked Example: Expected Total Incentive Payment

In a hypothetical trial, 25 customers are each offered the same optional service upgrade. Assume their decisions are independent and each customer accepts with probability $0.20$. The provider pays a fixed incentive of $40 for each acceptance. Find the expected total incentive payment. The acceptance count $X$ has distribution $\operatorname{Binomial}(25,0.20)$. No interest rate or mortality assumption applies.

Method: write the total as a sum of payment indicators. Let $I_i=1$ if customer $i$ accepts and $0$ otherwise. Then $X=\sum_{i=1}^{25}I_i$, and the total payment $C$ is $40X$. First find the expected number of acceptances:

$$ E[X]=25(0.20)=5 $$

Since the payment is $40 per acceptance, the total payment is $40 times the count. Therefore,

E[C]=E[40X]=40E[X]=40(5)=\$200

Equivalently, each customer contributes an expected payment of $40(0.20)=\$8$, so the 25 expected contributions sum to $25(8)=\$200$. Both routes agree. This calculation does not assert that the actual payment must be $200; actual payments occur in multiples of $40.

Common mistake avoided: multiplying $40 by the probability that all 25 customers accept would answer a different question. The total payment depends on the number of acceptances, so first calculate the expected count and then scale it by the payment per acceptance.

Common Mistakes and Scope of the Result

Confusing a mean with a probability. $np$ is an expected count. It is not $P(X=n)$, $P(X=0)$, or the probability of any single count.
Using the wrong success probability. State what counts as success. For the failure count, the relevant probability is $1-p$, not $p$.
Multiplying probabilities instead of adding expected contributions. A probability such as $p^n$ concerns all $n$ trials succeeding under independence. The mean count is $np$.
Rounding too early. Keep the exact value of $np$ for subsequent calculations. A fractional expected count is valid.
Using $np$ without a common success probability. The binomial formula assumes the same $p$ for every trial. If the probabilities differ, the expectation of the total is the sum of those probabilities, not necessarily $np$ for a single $p$.

The indicator method is useful because it connects a count to the outcomes that create it. It also gives a reliable way to check an answer: identify the contribution from one trial, find its expected contribution, and add across the fixed number of trials. This focus on the mean alone does not determine how spread out the count is; that is a separate question.

Key takeaway: For $X\sim\operatorname{Binomial}(n,p)$, express $X$ as the sum of $n$ success indicators. Each indicator has expectation $p$, so linearity of expectation gives $E[X]=np$.

Check Your Understanding

For each problem, identify the count and apply the indicator-variable argument. Assume the stated binomial conditions hold.

A hypothetical inspection checks 16 independent components, each with probability $0.25$ of failing. What is the expected number of failures? (Answer: 4)
A batch contains 35 applications, each independently complete with probability $0.80$. Find the expected number complete and the expected number incomplete. (Answer: 28 complete and 7 incomplete)
For 24 independent service requests, the probability of resolution on the first contact is $0.15$. Find the expected number resolved on first contact and the expected number not resolved on first contact. (Answer: 3.6 resolved and 20.4 not resolved)
A hypothetical program offers a $25 credit for each of 18 independent customers who enroll, with enrollment probability $0.40$ per customer. Find the expected total credit paid. (Answer: $180)
A fixed group of 50 independent trials has a common success probability $p$. The expected number of successes is 32. Find $p$, then find the expected number of failures. (Answer: $p=0.64$; 18 expected failures)

What You'll Learn