What a Binomial Count Measures
Many questions concern how many times an event occurs in a fixed set of opportunities: how many inspected devices fail, how many applications are approved, or how many days a sensor issues an alert. The relevant random variable is a count of successes, not the sequence in which they occur.
The binomial model formalizes this count when the trials satisfy specific conditions. A “success” is simply the outcome being counted; it need not be a favorable outcome. For example, if the question is how many devices fail inspection, define failure as success for the purpose of the model.
Each \(I_j\) is an indicator: it equals 1 when trial \(j\) is a success and 0 otherwise. Consequently, the count \(X\) is the number of indicators equal to 1. The model assumes a fixed number of trials, independence across trials, and the same success probability on every trial. These conditions are part of the definition, not optional details.
Parameters, Support, and Probability Function
For \(0<p<1\), the support—the set of values with positive probability—is every integer from 0 through \(n\):
Zero is possible because every trial might fail; \(n\) is possible because every trial might succeed. No value below 0 or above \(n\) is possible. At the boundary values of \(p\), the count is degenerate: if \(p=0\), then \(X=0\) with probability 1; if \(p=1\), then \(X=n\) with probability 1. Thus the full set of possible values is \(\{0\}\) or \(\{n\}\), respectively, at those boundaries.
A standard probability mass function for the binomial distribution is
The symbol \(\binom{n}{k}\) denotes the number of ways to choose \(k\) positions from \(n\) positions. The formula applies for \(0<p<1\); at \(p=0\) or \(p=1\), use the degenerate distributions just stated. The parameters \(n\) and \(p\), together with this probability function, specify the distribution. This tutorial’s main concern is recognizing what the count represents and identifying its support; a count and a particular ordered sequence of outcomes are not the same event.
Worked Example: Counting Device Failures
A hypothetical quality-control team tests six devices. For each device, the probability of failure is 0.30, and the test outcomes are independent with the same failure probability. Let success mean “fails the test.” In the observed order, the results are failure, pass, failure, failure, pass, pass. Define the count model, find the observed count and its support, and calculate the probability of that particular ordered sequence. No interest rate or mortality assumption applies; this is a hypothetical device-testing model.
Model and parameters. There are \(n=6\) fixed trials. Each has failure probability \(p=0.30\), and the trials are independent and identically distributed. Thus the number \(X\) of failures has distribution \(X\sim\operatorname{Bin}(6,0.30)\).
Observed count and support. The sequence contains failures in positions 1, 3, and 4, so the observed count is \(X=3\). Since \(0<p<1\), the support is \(\{0,1,2,3,4,5,6\}\).
Probability of this ordered sequence. The sequence has three failures and three passes. Independence gives \(0.30\times0.70\times0.30\times0.30\times0.70\times0.70 =0.30^3(0.70)^3 =0.027\times0.343 =0.009261\). To four decimal places, its probability is 0.0093.
Check and method. The count is within 0 to 6, as required, and the sequence probability is between 0 and 1. The method multiplies the probabilities for the specified outcomes in their given positions. The common mistake is to treat the probability of this one sequence as the probability that the total count is 3; other ordered sequences can also have three failures.
Why the Support Runs from Zero to \(n\)
The support follows directly from representing the count as a sum of indicators. Each indicator contributes either 0 or 1, so the smallest possible sum is 0 and the largest is \(n\). When \(0<p<1\), both outcomes have positive probability on every trial. Any chosen number \(k\) from 0 to \(n\) can therefore occur: one possible arrangement has \(k\) successes and \(n-k\) failures, and that arrangement has positive probability.
Write \(I_j=1\) for success on trial \(j\), and \(I_j=0\) for failure.
The total \(X=I_1+\cdots+I_n\) increases by one for each success and not at all for a failure.
There are \(n\) terms, each either 0 or 1, so \(0\le X\le n\). Since \(X\) is an integer, its possible values lie among \(0,1,\ldots,n\).
This argument also explains why the trial count is an upper bound on the count. It does not establish that every value in that range has positive probability when \(p=0\) or \(p=1\); in those boundary cases, only one outcome is possible.
Worked Example: Counting Alerts Across Several Days
A hypothetical monitoring system is observed on nine days. On each day it independently issues an alert with probability 0.12. Let \(X\) be the number of alert days. In one possible record, alerts occur on days 2 and 8 only. State the distribution, parameters, support, observed count, and probability of that particular record. No interest rate or mortality assumption applies; the specified model concerns daily alert outcomes.
Model and parameters. Each day is one trial, so \(n=9\). An alert is the defined success, with common probability \(p=0.12\). The outcomes are independent and identically distributed; hence \(X\sim\operatorname{Bin}(9,0.12)\).
Observed count and support. There are alerts on exactly two days, so the record gives \(X=2\). Since \(0<0.12<1\), the support is \(\{0,1,2,3,4,5,6,7,8,9\}\).
Probability of the specified record. This record has two alerts and seven non-alerts. Its probability is \(0.12^2(0.88)^7 =0.0144\times0.4086755968 =0.00588492859392\). Rounded to four decimal places, this is 0.0059.
Check and method. There are nine day-level outcomes, so a count of two lies in the stated support. The computation is for the specified record, in which alerts occur on days 2 and 8. It is not the probability of every possible record with two alerts. The common mistake is to ignore the seven non-alert days and use only \(0.12^2\); the probability of this complete record must include both successes and failures.
Setting Up the Count Carefully
Before writing \(X\sim\operatorname{Bin}(n,p)\), specify what one trial is and which outcome counts as success. Then identify the fixed trial count and the common probability of success. A statement such as “there are eight cases” is not enough by itself: the model also needs a defined event for each case and the independence and identical-probability assumptions.
The label “success” does not alter the underlying event. If a hypothetical insurer is counting claims that require additional review, “requires additional review” is success for this count even though it may be an undesirable result. Once defined, each trial contributes either one to \(X\) or zero.
Worked Example: Counting Escalated Applications
A hypothetical benefits office examines eight applications. Assume each application independently has probability 0.65 of requiring escalation, with the same probability for all eight applications. In one possible set of outcomes, applications 1, 2, 5, 6, and 8 are escalated; the others are not. Define the count, state its parameters and support, and find the observed value. No interest rate or mortality assumption applies; this is a hypothetical application-review model.
Define success and the random variable. Let success mean “the application requires escalation.” Let \(I_j=1\) if application \(j\) is escalated and \(I_j=0\) otherwise. Then \(X=I_1+\cdots+I_8\) is the number of escalated applications.
State the distribution and parameters. There are \(n=8\) fixed trials, and each has success probability \(p=0.65\). The trials are stated to be independent and to have the same probability, so \(X\sim\operatorname{Bin}(8,0.65)\).
Find the observed value and support. The listed outcomes contain five escalations, so the observed count is \(X=5\). Since \(0<0.65<1\), the support is \(\{0,1,2,3,4,5,6,7,8\}\).
Check and method. The maximum count is eight because there are eight applications, and the observed value five is between zero and eight. The model counts escalations, not applications that avoid escalation. The common mistake is to reverse the success definition and report three; three is the number not escalated, whereas the requested count is five.
Common Mistakes and Exam Traps
- Confusing the count with an indicator. An indicator for one trial takes values 0 and 1. The binomial count adds \(n\) such indicators and can take integer values from 0 to \(n\).
- Using the wrong value as the trial count. In \(X\sim\operatorname{Bin}(n,p)\), \(n\) is the number of trials, not the success probability or the observed number of successes.
- Leaving success undefined. State the event being counted. The model works whether success means passing, failing, approval, or any other specified outcome.
- Giving a support that starts at 1. The count can be zero when every trial fails. For \(0<p<1\), the support includes both 0 and \(n\).
- Ignoring boundary probabilities. If \(p=0\) or \(p=1\), the count does not have positive probability at every integer from 0 to \(n\); it is fixed at 0 or \(n\), respectively.
- Mixing a sequence with a count. A specific ordered record is one way to obtain a count. The count alone does not specify which trials succeeded.
Check Your Understanding
For each problem, identify the count model and the requested value or support. Treat each scenario as hypothetical.
- A technician independently tests five sensors, each with probability 0.08 of producing a fault signal. Let success mean a fault signal. State the distribution and support. (Answer: \(X\sim\operatorname{Bin}(5,0.08)\); support \(\{0,1,2,3,4,5\}\).)
- A reviewer independently processes seven files, each with the same probability 0.4 of needing correction. What does \(X\) count, and what is its maximum possible value? (Answer: \(X\) counts files needing correction; \(X\sim\operatorname{Bin}(7,0.4)\), with maximum 7.)
- In six independent trials with common success probability 0.25, the outcomes are failure, success, failure, success, success, failure. Find the observed count and the support. (Answer: observed count \(X=3\); support \(\{0,1,2,3,4,5,6\}\).)
- A process has four independent trials, each with success probability 1. State the distribution and the actual support. (Answer: \(X\sim\operatorname{Bin}(4,1)\); \(X=4\) with probability 1, so the actual support is \(\{4\}\).)
- A hypothetical audit has ten independent cases, each with probability 0.15 of requiring follow-up. If exactly two cases require follow-up in one record, distinguish the count from a specified ordered record. (Answer: \(X\sim\operatorname{Bin}(10,0.15)\), support \(\{0,1,\ldots,10\}\), observed count 2; the count does not identify which two cases required follow-up.)