Reading the Binomial Parameters
A binomial distribution describes a count: how many successes occur in a fixed collection of trials. Its two parameters answer different questions. The parameter \(n\) is the number of trials, while \(p\) is the probability of success on each trial. Keeping those roles separate is essential when interpreting the model, identifying its support, and evaluating probabilities at the boundaries.
The binomial trial scheme has a fixed positive integer number \(n\) of trials. Each trial has two outcomes, designated success and failure; the success probability is the same value \(p\) on every trial; and the trials are independent. The random variable \(X\) counts the successes. Thus \(X\) has a binomial distribution with parameters \(n\) and \(p\), written \(X\sim\operatorname{Bin}(n,p)\), where \(n\geq 1\) and \(0\leq p\leq 1\).
The value of \(n\) is not the number of successes observed, and it is not automatically the size of any data set. It counts the trials in the model. The value of \(p\) is not the fraction of successes in a particular realized sample; it is the probability assigned to success on each trial. A realized count \(x\) is one possible outcome of \(X\), not a parameter.
For \(0<p<1\), the probability mass function is \(P(X=k)=\binom{n}{k}p^k(1-p)^{n-k}\) for integer \(k\) from 0 to \(n\). This is the formula obtained by counting the possible arrangements of \(k\) successes among \(n\) trials. The probability at each count depends on both parameters: \(n\) determines which counts can occur, while \(p\) controls the probabilities assigned to them.
Worked Example: Interpreting a Trial Count
A laboratory tests 12 specimens. Suppose each test, independently of the others, has probability \(0.03\) of producing a positive result. Let \(X\) be the number of positive results. Identify the binomial parameters and state what values \(X\) can take.
There are 12 tests, so \(n=12\). A positive result is designated success, and its probability on each test is \(p=0.03\). The independence and common-probability assumptions are part of the stated model, so \(X\sim\operatorname{Bin}(12,0.03)\).
The count can be zero if all tests are negative, one if exactly one is positive, and so on, up to 12 if all are positive. Therefore its possible-count range is \(\{0,1,2,\ldots,12\}\). Since \(0<0.03<1\), every one of these counts has positive probability, so this is also the actual support.
For instance, the probability of zero positive results is \(P(X=0)=\binom{12}{0}(0.03)^0(0.97)^{12}=(0.97)^{12}\). The trial count remains 12 regardless of which count is observed.
Possible Counts and Actual Support
The possible-count range is the set \(\{0,1,\ldots,n\}\): no count of successes can be negative or exceed the number of trials. The support of a discrete random variable is the set of values that have strictly positive probability. These sets agree for an interior binomial probability, but need not agree at the boundary values \(p=0\) and \(p=1\).
Proof. A count outside \(\{0,1,\ldots,n\}\) is impossible: there cannot be fewer than zero successes or more successes than trials. Now suppose \(0<p<1\), and fix any integer \(k\) with \(0\leq k\leq n\). There are \(\binom{n}{k}\) arrangements with exactly \(k\) successes. Each specified arrangement has probability \(p^k(1-p)^{n-k}\), which is strictly positive because both \(p\) and \(1-p\) are positive. Also, \(\binom{n}{k}>0\). Hence
Thus every count from 0 through \(n\) belongs to the support. If \(p=0\), each trial is a failure with probability one, so the only possible count is \(X=0\). If \(p=1\), each trial is a success with probability one, so the only possible count is \(X=n\). In each case that sole value has probability one, proving the support statements. \(\square\)
This distinction matters in the boundary cases. The count range \(\{0,\ldots,n\}\) describes values that are structurally within the limits of \(n\) trials. But when \(p=0\), counts above zero have probability zero; when \(p=1\), counts below \(n\) have probability zero. Those zero-probability values are not in the support.
Boundary Values of the Success Probability
The endpoints \(p=0\) and \(p=1\) are valid probability values, not exceptions to the binomial model. They make the outcome of every trial certain. Consequently, the count is no longer random: it is zero at \(p=0\) and \(n\) at \(p=1\).
For \(n\geq1\), the probabilities at the two endpoints of the count range can also be written, for every \(0\leq p\leq1\), as
At an interior value of \(p\), zero successes requires every trial to be a failure, and \(n\) successes requires every trial to be a success. Independence gives the displayed powers. At \(p=0\), the first expression equals 1 and the second equals 0; at \(p=1\), the first equals 0 and the second equals 1. Since \(n\geq1\), these endpoint expressions do not involve a zero raised to the zeroth power.
Worked Example: A Certain Failure Probability
A model assigns success probability \(p=0\) to each of \(n=6\) independent trials. Let \(X\) count successes. Find its support and calculate the probabilities of zero, one, and six successes.
Since success has probability zero on every trial, all six trials are failures with probability one. Therefore \(X=0\) with probability one and its support is \(\{0\}\). In particular, \(P(X=0)=1\), \(P(X=1)=0\), and \(P(X=6)=0\).
The endpoint expression gives \(P(X=0)=(1-0)^6=1\), while \(P(X=6)=0^6=0\). The count \(1\) is also outside the support, so its probability is zero. The values 1 and 6 lie in the possible-count range from 0 through 6, but only 0 is in the actual support.
Worked Example: A Certain Success Probability
Now take \(n=6\) trials with success probability \(p=1\). Find the support and the probabilities of zero and six successes.
Every trial is a success with probability one. Thus \(X=6\) with probability one and the support is \(\{6\}\). The endpoint probabilities are \(P(X=0)=(1-1)^6=0\) and \(P(X=6)=1^6=1\).
The result follows from the event described by each count: zero successes would require six failures, which cannot occur when each trial succeeds surely; six successes is the certain outcome. This example also shows why a support endpoint need not have positive probability: zero is the lower end of the possible-count range, but it is not in the support here.
Probabilities at the Count Endpoints
The endpoints of the count range have particularly simple interpretations. The event \(X=0\) means that every trial is a failure; the event \(X=n\) means that every trial is a success. The endpoint probabilities follow directly from these events, rather than from any approximation or tail calculation.
Proof. Write the trial outcomes as \(Y_1,\ldots,Y_n\), where \(Y_i=1\) for success and \(Y_i=0\) for failure. Then \(X=\sum_{i=1}^{n}Y_i\). The event \(X=0\) occurs exactly when every trial is a failure. Each failure has probability \(1-p\), and independence gives
Similarly, \(X=n\) occurs exactly when every trial is a success. Each success has probability \(p\), so independence gives
Both products are valid when \(p=0\) or \(p=1\) as well as when \(0<p<1\). \(\square\)
Worked Example: Calculating Both Endpoint Probabilities
Suppose \(X\sim\operatorname{Bin}(5,0.4)\). Calculate the probabilities of zero successes and five successes, and check that both are positive.
Here \(n=5\), \(p=0.4\), and \(1-p=0.6\). For zero successes, all five trials must fail: \(P(X=0)=(0.6)^5=0.6\times0.6\times0.6\times0.6\times0.6=0.07776\).
For five successes, all five trials must succeed: \(P(X=5)=(0.4)^5=0.4\times0.4\times0.4\times0.4\times0.4=0.01024\).
Both values are strictly positive, as expected because \(0<0.4<1\). The support is therefore \(\{0,1,2,3,4,5\}\), including both endpoints. The two endpoint probabilities do not need to be equal: they refer to different all-failure and all-success outcomes, whose probabilities depend on \(1-p\) and \(p\), respectively.
Common Mistakes and Why the Details Matter
One tempting shortcut is to substitute \(p=0\) into the general probability formula without checking the boundary. At \(k=0\), this produces a factor \(p^0\), and the expression \(0^0\) is convention-dependent if considered in isolation. The probability itself is not ambiguous: when \(p=0\), all trials fail, so \(P(X=0)=1\). The safest approach is to use the deterministic boundary description or the endpoint formula \((1-p)^n\), with \(n\geq1\).
- Confusing a parameter with an outcome. \(n\) is the fixed number of trials; \(X\) is the random number of successes. A realized value \(x\) does not replace \(n\).
- Assuming every count in the possible-count range is in the support. That is true only when \(0<p<1\). At \(p=0\) or \(p=1\), the support contains just one value.
- Reversing the endpoint probabilities. Zero successes means \(n\) failures, so \(P(X=0)=(1-p)^n\). All \(n\) successes gives \(P(X=n)=p^n\).
- Treating \(p\) as an observed sample proportion. The parameter is the per-trial success probability specified by the model; an observed fraction is a data summary and can differ from it.
These distinctions are useful whenever a binomial model is translated into a probability statement. Naming \(n\) and \(p\) correctly determines which counts are possible, while checking whether \(p\) is at a boundary determines whether all those counts actually have positive probability. Endpoint calculations also provide quick checks on a probability model: if \(p\) is interior, both all-failure and all-success outcomes must have positive probability; if \(p\) is a boundary value, exactly one count is certain.
Check Your Understanding
Answer each question using the definitions and results above.
- A study models 20 independent trials, each with success probability \(0.15\). Identify \(n\) and \(p\), and state the support of the success count.
- For \(X\sim\operatorname{Bin}(8,0.25)\), write the probabilities of zero successes and eight successes as powers.
- For \(X\sim\operatorname{Bin}(4,0)\), distinguish the possible-count range from the support.
- Explain why \(P(X=0)\) involves \(1-p\), rather than \(p\), in a binomial model.
- What is the support of \(X\sim\operatorname{Bin}(7,1)\), and what is \(P(X=7)\)?