A binomial experiment is a statistical experiment that consists of a fixed number of repeated trials where each trial produces exactly one of two possible outcomes and every trial is independent of all others. In Six Sigma, recognizing that a process produces binomial data — pass or fail, defective or conforming, yes or no — is a Measure phase decision with direct consequences for which control chart to use, which capability analysis to run, and how to interpret process performance against a customer specification.
Table of contents
Meaning of Binomial Experiment
A binomial experiment is a sequence of trials that satisfies four specific conditions: (1) the number of trials is fixed in advance, (2) each trial results in exactly one of two possible outcomes, conventionally labeled success and failure, (3) every trial is independent of every other trial, and (4) the probability of success stays constant across all trials.
When all four conditions hold, the count of successes across the trials follows a binomial distribution. In Six Sigma, attribute data collected in pass/fail or defective/conforming categories frequently meets these four conditions, making the binomial experiment the statistical foundation for p-charts, np-charts, and binomial capability analysis.
Key Takeaways
- A binomial experiment must satisfy four conditions: fixed number of trials, binary outcomes only, independence between trials, and constant probability of success.
- The four conditions are most commonly remembered with the acronym FBIT: Fixed trials, Binary outcomes, Independent trials, same probability (constant p) for each Trial.
- In Six Sigma, attribute data — data collected as counts of defective units, pass/fail results, or yes/no outcomes — frequently follows a binomial distribution when all four conditions hold.
- The p-chart (proportion defective) and np-chart (number of defectives) in Statistical Process Control both rely on the binomial distribution and are appropriate only when the underlying data meets the binomial experiment conditions.
- The most commonly violated condition in real Six Sigma projects is independence: inspection results are not independent when a defect in one unit is caused by a machine condition that also affects subsequent units in the same production run.
- A binomial experiment is distinct from a binomial distribution: the experiment describes the data-generating process and its four conditions, while the distribution describes the mathematical probability model for the number of successes that experiment produces.
Public, Onsite, Virtual, and Online Six Sigma Certification Training!
- We are accredited by the IASSC.
- Live Public Training at 52 Sites.
- Live Virtual Training.
- Onsite Training (at your organization).
- Interactive Online (self-paced) training,
What Is a Binomial Experiment?
A binomial experiment is a precisely defined type of statistical experiment. Not every experiment with two possible outcomes qualifies. The term has a specific technical meaning in statistics and in Six Sigma, and understanding that precise meaning determines whether a practitioner can legitimately apply binomial-based analysis to their process data.
The word “binomial” combines the Latin bi (two) and nomial (term or name), meaning two named outcomes. A binomial experiment is one structured so that only two possible outcomes exist at each trial, and those outcomes are collectively exhaustive and mutually exclusive — every trial must result in exactly one of the two outcomes, never both, never neither.
The result of a binomial experiment is a count: specifically, the number of trials out of the fixed total that produced the outcome labeled “success.” This count is a discrete number (0, 1, 2, 3… up to the total number of trials), and it follows the binomial distribution when all four conditions below hold.
The Four Conditions of a Binomial Experiment
A valid binomial experiment must satisfy all four of the following conditions simultaneously. If any condition fails, the binomial distribution does not accurately model the data, and binomial-based statistical tools — including p-charts and np-charts — will produce unreliable results.
Condition 1: Fixed Number of Trials (n)
The total number of trials must be determined and fixed before the experiment begins. The practitioner cannot decide to run additional trials based on the results observed so far. The fixed trial count is designated as n.
In Six Sigma process inspection, this condition means the sample size for each inspection period must be determined in advance — either a fixed constant (inspect 50 units per shift) or a pre-determined variable plan. Changing the sample size based on observed defect rates violates this condition.
What violates this condition: Continuing to inspect units until a predetermined number of defects is found. That experiment has a variable number of trials determined by the outcome and follows a negative binomial distribution instead.
Condition 2: Binary Outcomes Only (Success or Failure)
Each trial must produce exactly one of exactly two possible outcomes. In statistics, these are conventionally labeled success and failure, but those labels carry no inherent positive or negative meaning — “success” simply names the outcome being counted, which in a quality context is often the defect or failure event.
In Six Sigma, binary outcomes appear wherever a unit is classified as conforming or nonconforming, a transaction is categorized as error-free or erroneous, a patient is discharged on time or late, an invoice is accurate or inaccurate. The key requirement is that no third outcome is possible — every unit must be classified as one or the other.
What violates this condition: Classifying units into three or more categories (minor defect, major defect, critical defect) without collapsing those categories into two groups. Multi-category outcome data requires multinomial distribution models, not binomial.
Condition 3: Independent Trials
Each trial must be statistically independent of every other trial. The outcome of one trial must not influence the probability of any outcome on any subsequent trial. Independence is the condition that allows the probabilities to multiply across trials in the binomial probability formula.
In manufacturing, independence means the defect status of one unit does not affect whether the next unit is defective. This holds when defects occur randomly and individually. It fails when a root cause — such as a deteriorating tool, a miscalibrated machine, or a contaminated material batch — makes defects cluster together: one defective unit in a run makes subsequent units in the same run more likely to also be defective.
What violates this condition: Autocorrelated defects where process deterioration, operator fatigue, or material lot effects mean that a defective outcome at trial k raises the probability of a defective outcome at trial k+1.
Condition 4: Constant Probability of Success (p)
The probability of success must remain the same for every trial in the experiment. This constant probability is designated as p. Because the only two outcomes are success and failure, the probability of failure on each trial is always 1 − p = q.
In a stable, in-control process, the defect rate remains approximately constant over time, which satisfies this condition. In an unstable process — one experiencing drift, tool wear, shift changes, or incoming material variation — the defect probability changes from trial to trial, violating this condition.
What violates this condition: A process that starts a production run with a low defect rate and ends with a high defect rate due to tool wear. The probability of producing a defective unit is not constant across all units in that run.
How to Verify the Four Conditions in a Six Sigma Project
Before applying binomial-based statistical tools to process data, a Six Sigma practitioner should explicitly verify each of the four conditions. The following checks are practical, fast, and appropriate for the Measure phase.
Verifying fixed n: Confirm that the sampling plan specifies the number of units to inspect per sample period and that this number is determined before inspection begins, not adjusted based on incoming results.
Verifying binary outcomes: Confirm that the data classification system produces exactly two categories with no unit falling outside both, no unit classified as both, and no third category that would require a separate count.
Verifying independence: Check for autocorrelation in attribute data using a run chart of the defective proportion over time. Random scatter around the centerline supports independence. Clustering of high-defect periods followed by low-defect periods suggests dependence. Review process knowledge: if the same root cause (tool condition, material lot, operator) simultaneously affects multiple consecutive units, the independence condition is suspect.
Verifying constant p: Monitor the proportion defective over a stable, controlled period. A p-chart with most or all points within the control limits and no systematic patterns (trends, runs above or below the centerline) supports the constant-p assumption. Significant trends or out-of-control signals indicate that p is changing over time and the constant probability condition may not hold across the full dataset.
Also Read: Attribute Sampling
Binomial Experiment Examples: What Qualifies and What Does Not
Understanding which experiments are genuinely binomial and which only appear to be is a critical skill for Six Sigma practitioners selecting their data analysis approach.
Examples That Meet All Four Conditions
Incoming inspection of manufactured components. A quality team inspects a fixed sample of 100 components (n = 100) from each production lot. Each component is classified as conforming or nonconforming (binary). Components are drawn independently by random sampling. The process is stable so the defect rate is approximately constant (constant p). This is a valid binomial experiment.
Healthcare appointment no-show tracking. A clinic tracks whether each of 50 scheduled patients (n = 50) shows up or does not (binary). Appointment attendance decisions are independent across patients. The no-show rate has been stable for the past quarter (constant p). This is a valid binomial experiment.
Software regression test pass/fail. A test suite runs 200 independent automated tests (n = 200). Each test passes or fails (binary). Tests are designed to be independent. The defect rate for a given software build is fixed. This is a valid binomial experiment.
Examples That Do Not Meet All Four Conditions
Inspecting until 5 defects are found. The number of trials is not fixed — it is determined by the outcome. This violates Condition 1 and follows a negative binomial (Pascal) distribution instead.
Classifying defects as minor, major, or critical. Three possible outcomes violate Condition 2. If the practitioner collapses these into “any defect” vs. “no defect,” the experiment becomes binary and may qualify as binomial.
Tracking defects in a run with known tool wear. If a cutting tool degrades progressively across the production run, the probability of producing a defective unit increases as the run continues. Condition 4 (constant p) is violated.
Sampling without replacement from a small population. If a team inspects 20 units from a lot of 30 without replacement, removing each inspected unit changes the composition of the remaining population. Each successive trial is not independent because the pool changes. This follows a hypergeometric distribution instead. As a practical rule of thumb, the binomial approximation holds adequately when the sample size is no more than 10% of the total population size.
Binomial Experiment vs. Binomial Distribution
These two terms are closely related but describe different things, and practitioners benefit from keeping the distinction clear.
A binomial experiment describes the data-generating process: the actual sequence of trials and the structural conditions (fixed n, binary outcomes, independence, constant p) that those trials must satisfy.
A binomial distribution describes the mathematical probability model that applies to the count of successes produced by a valid binomial experiment. The distribution specifies, for any combination of n (trials) and p (success probability), the exact probability of observing each possible outcome: 0 successes, 1 success, 2 successes, and so on up to n successes.
The relationship is: if an experiment satisfies all four binomial conditions, then the count of successes it produces follows a binomial distribution with parameters n and p.
In Six Sigma training, both the experiment (which conditions structure the data) and the distribution (which probability model describes it) are tested separately on the IASSC exam, which is why the site maintains distinct glossary entries for each.
| Concept | What It Describes | Key Parameters |
| Binomial Experiment | The data-generating process and its four structural conditions | n (fixed trials), binary outcomes, independence, constant p |
| Binomial Distribution | The probability model for counting successes in a valid binomial experiment | n and p; produces probabilities for each possible count of successes |
| Binomial Random Variable | The variable X that counts the number of successes in the experiment | X can take any integer value from 0 to n |
How Binomial Experiments Connect to Six Sigma DMAIC
Recognizing that a process produces binomial data is not just a statistical classification exercise — it has direct, practical consequences for tool selection and data analysis at every phase of DMAIC.
Define Phase
The project definition identifies what is being measured. If the output being tracked is attribute data — a unit either meets specification or does not — the team is working with a binary outcome, which is the foundation of a binomial experiment. The project Y (output variable) is often stated as a proportion: “the defect rate is 4.2%.” That proportion comes from counting defective units in a fixed sample, which is a binomial experiment.
Measure Phase
The Measure phase establishes whether the process data meets the binomial experiment conditions and selects the appropriate control chart. When all four conditions hold, the practitioner can use:
- The p-chart (proportion defective chart) when sample size varies across inspection periods.
- The np-chart (number of defectives chart) when sample size is constant across inspection periods.
Both charts use the binomial distribution to calculate their control limits. If the data does not satisfy all four conditions — particularly the independence and constant-p conditions — these charts will produce incorrect control limits and misleading signals.
Analyze Phase
The Analyze phase tests whether a suspected root cause significantly affects the binary outcome. Chi-square tests of association and tests of proportions (comparing two defect rates across groups) both assume binomial experiment conditions. A practitioner who tests whether a process change reduced the defect rate is comparing two binomial proportions.
Improve Phase
A pilot implementation in the Improve phase typically produces binary outcome data: each unit produced under the new process either meets specification or does not. If the binomial experiment conditions hold for the pilot, the team can calculate a confidence interval on the new defect proportion and test whether the improvement is statistically significant.
Control Phase
The Control phase monitors ongoing attribute data using p-charts or np-charts. A stable p-chart with all points within control limits confirms that the process continues to produce a constant defect rate — the constant-p condition — and that no new special causes have emerged.
Frequently Asked Questions: Binomial Experiment
Q: What is a binomial experiment?
A: A binomial experiment is a statistical experiment consisting of a fixed number of repeated, independent trials where each trial produces exactly one of two possible outcomes and the probability of success stays constant across all trials. When all four conditions hold, the count of successes in the experiment follows a binomial distribution.
Q: What are the four conditions of a binomial experiment?
A: The four conditions are: (1) the number of trials is fixed before the experiment begins; (2) each trial results in exactly one of two mutually exclusive outcomes, typically labeled success and failure; (3) every trial is independent of every other trial, meaning the outcome of one trial does not affect the probability of any outcome on any subsequent trial; and (4) the probability of success remains constant and identical across all trials.
Q: How is a binomial experiment different from a binomial distribution?
A: A binomial experiment describes the actual data-generating process — the sequence of trials and the four structural conditions they must satisfy. A binomial distribution is the mathematical probability model that describes the expected probabilities for each possible count of successes when those four conditions hold. The experiment produces the data; the distribution models the probabilities of each possible outcome from that data.
Q: What happens if one of the four binomial conditions is violated?
A: If any of the four conditions fails, the binomial distribution does not accurately model the data. Using binomial-based tools — such as p-charts or np-charts — on data that violates these conditions produces incorrect control limits and misleading conclusions. Common violations include non-constant sample size (violating fixed n), clustered defects from a shared root cause (violating independence), or a drifting process defect rate (violating constant p).
Q: Why does sampling without replacement sometimes violate binomial conditions?
A: When units are sampled without replacement from a finite population, removing each inspected unit changes the composition of the remaining pool. This means each successive trial is not truly independent — the probability of drawing a defective unit changes slightly as each unit is removed. The hypergeometric distribution models this situation more accurately. A common practical rule allows the binomial approximation when the sample size is at most 10% of the total population.
Q: How do binomial experiments connect to p-charts in Six Sigma?
A: A p-chart monitors the proportion of defective units across inspection samples over time. Its control limits are calculated using the binomial distribution, which assumes that the underlying data-generating process is a valid binomial experiment. Specifically, a p-chart requires that each inspection sample represents a fixed or known number of trials with binary outcomes, independent classifications, and a constant underlying defect probability.
When these conditions hold, the p-chart correctly signals whether observed variation in the defect proportion is due to common cause or special cause variation.
Q: Is a coin flip a binomial experiment?
A: Yes, a series of coin flips is the classic textbook example of a binomial experiment. The number of flips can be fixed in advance (flip 10 times), each flip has exactly two outcomes (heads or tails), successive flips are independent, and the probability of heads stays constant at 0.5 for a fair coin. The count of heads in n flips follows a binomial distribution with parameters n and p = 0.5.
Q: What is the difference between a p-chart and an np-chart, and which binomial experiment condition distinguishes them?
A: Both p-charts and np-charts use the binomial distribution and require valid binomial experiment conditions. The distinguishing factor is whether the sample size varies across inspection periods. A p-chart plots the proportion defective and accommodates variable sample sizes. An np-chart plots the raw count of defectives and requires a constant sample size across all inspection periods. When sample size is constant, either chart is valid; when sample size varies, only the p-chart should be used.
Also Read: Lurking Variable
Binomial Experiment Training in Six Sigma
Understanding the four conditions of a binomial experiment and knowing how to verify them in real process data is a testable competency on both the IASSC Green Belt and Black Belt exams. The IASSC Body of Knowledge explicitly includes probability distributions — including the binomial distribution and its application to attribute data — as part of the Measure phase curriculum.
In practice, practitioners who correctly identify binomial data choose the right control charts, design the right sample sizes, and run the right hypothesis tests. Those who misidentify data type run the risk of applying normal-distribution-based tools (such as Xbar-R charts or Cpk indices) to attribute data that violates the continuous distribution assumptions those tools require.
At Six Sigma Development Solutions Inc, data type classification, the binomial experiment conditions, and attribute control chart selection are covered as applied, exam-relevant skills in our Green Belt and Black Belt training programs — not as isolated textbook concepts.
We offer Six Sigma training in three formats:
- Onsite training — Delivered at your facility, using your real process data and your own attribute measurement systems in data type classification exercises.
- Live virtual training — Instructor-led online sessions covering probability distributions, control chart selection, and binomial capability analysis with real-time instructor support.
- Online training — Self-paced Green Belt and Black Belt certification programs covering all IASSC-testable probability distribution content.
Explore our Six Sigma training programs or contact our team to find the right program for your certification goals.
About Six Sigma Development Solutions, Inc.
Six Sigma Development Solutions, Inc. offers onsite, public, and virtual Lean Six Sigma certification training. We are an Accredited Training Organization by the IASSC (International Association of Six Sigma Certification). We offer Lean Six Sigma Green Belt, Black Belt, and Yellow Belt, as well as LEAN certifications.
Book a Call and Let us know how we can help meet your training needs.
