A data distribution describes how values in a dataset are spread across their possible range. It shows which values appear most often, which appear rarely, and whether the pattern is symmetric, skewed, or clustered into multiple groups. In Six Sigma, understanding data distribution is not just a statistical formality.
The shape of a distribution determines which statistical tools are valid for that data. Using the wrong tool on the wrong distribution produces incorrect conclusions. Building a DMAIC project on incorrect conclusions wastes time and fails to solve the problem.
Table of contents
Meaning of Data Distribution in Six Sigma
A data distribution is the pattern showing how often each value or range of values occurs in a dataset. In Six Sigma, practitioners examine the distribution of their process data during the Measure phase to determine whether it is normal (bell-shaped), skewed, bimodal, or follows another pattern.
This matters because most Six Sigma statistical tools — including process capability indices (Cp, Cpk), control charts, and hypothesis tests — assume the data follows a normal distribution. When the data is non-normal, different tools or data transformations are needed.
The Anderson-Darling normality test is the standard Six Sigma method for formally testing whether data follows a normal distribution, with a p-value above 0.05 indicating the data is consistent with normality.
Key Takeaways
- A data distribution describes how values in a dataset are spread across their possible range, showing frequency, central tendency, and shape.
- The shape of a distribution determines which statistical tools are valid. Process capability indices, control charts, and most hypothesis tests assume normally distributed data.
- The normal distribution is the most important distribution in Six Sigma. It is defined by its mean and standard deviation. In a normal process, 68.27% of data falls within one sigma, 95.45% within two sigma, and 99.73% within three sigma of the mean.
- Normal distribution was developed by astronomer Karl Friedrich Gauss, which is why it is also called the Gaussian distribution.
- The Anderson-Darling normality test is the standard Six Sigma test for checking whether data follows a normal distribution. A p-value above 0.05 indicates the data does not significantly deviate from normality.
- When data is non-normal, practitioners can apply data transformations (logarithmic, square root, Box-Cox) or use non-parametric statistical tests that do not assume a specific distribution shape.
- Eight distributions appear most frequently in Six Sigma project data: normal, uniform, skewed, bimodal/multimodal, exponential, log-normal, Poisson, and chi-square.
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 Data Distribution?
A data distribution is a function or table that shows how often each value or range of values appears in a dataset. Plot those frequencies on a graph and the shape that emerges is the distribution.
Every dataset has a distribution. The shape of that distribution communicates three things:
Where the data centers. The mean, median, and mode locate the center of the distribution. In a symmetric distribution, all three measures coincide. In a skewed distribution, they diverge in a predictable way.
How spread out the data is. Standard deviation and variance measure the width of the distribution. A narrow distribution clusters tightly around the center. A wide distribution spreads far in both directions.
What shape the data takes. Is it bell-shaped? Flat? Skewed to one side? Does it have two peaks? The shape tells the analyst what kind of process produced the data and which statistical tools apply.
In Six Sigma, every Measure phase investigation begins with examining the distribution of the project’s Y variable. That examination drives every subsequent analytical decision.
Also Read: Unimodal Distribution
Why Data Distribution Matters in Six Sigma
Every statistical tool in Six Sigma rests on assumptions about the distribution of the data it analyzes.
Process capability indices (Cp and Cpk) assume normally distributed data. If the data is skewed or bimodal, these indices produce misleading results. The Six Sigma Institute states explicitly that when data is non-normal, practitioners must transform the data from non-normal to normal before calculating capability indices.
Control charts (X-bar, R, S charts) assume that subgroup means follow a normal distribution. When this assumption holds, the ±3 sigma control limits correctly identify out-of-control signals. When it does not hold, the false alarm rate changes.
Hypothesis tests such as the two-sample t-test, one-way ANOVA, and regression analysis assume normality of residuals or of the data itself, particularly in small samples. iSixSigma confirms that understanding the distribution of data directly determines whether standard normal-theory methods apply or whether non-parametric alternatives are needed.
The bottom line: identifying the distribution of process data is not an academic exercise. It is the prerequisite to using Six Sigma statistical tools correctly.
The Eight Key Data Distributions in Six Sigma
1. Normal Distribution (Gaussian Distribution)
The normal distribution is the most important distribution in Six Sigma practice. Karl Friedrich Gauss developed the mathematical framework for it, which is why it is also called the Gaussian distribution. It is symmetric and bell-shaped, with data clustering around the mean and tailing off symmetrically in both directions.
The Six Sigma Institute describes the normal distribution as “the most prominently used distribution in statistics,” with applicability to many situations where population characteristics must be used to predict sample behavior.
Key properties:
- 68.27% of data falls within one standard deviation (±1σ) of the mean
- 95.45% falls within two standard deviations (±2σ)
- 99.73% falls within three standard deviations (±3σ)
- 99.9937% falls within four standard deviations (±4σ)
At six sigma process performance, 99.999998% of outputs fall within the specification limits.
A normal distribution is fully described by two parameters: the mean (µ) and standard deviation (σ). Written in statistical notation as N(µ, σ).
Six Sigma context: Process capability analysis, control charts, and most hypothesis tests all assume normally distributed data. The normal distribution is the baseline against which all other distributions are compared.
2. Uniform Distribution
A uniform distribution assigns equal probability to every value within a defined range. When plotted, it appears as a flat rectangle. No value is more likely than any other.
Uniform distributions appear when measurement instruments round all readings to a coarse scale, creating an artificial pattern where all values in a range appear equally. They also appear in random number generators and some simulation scenarios.
Six Sigma context: A histogram showing a suspiciously flat distribution often signals measurement system issues, specifically resolution problems where the gauge cannot distinguish between adjacent values. This pattern prompts Measurement System Analysis investigation.
3. Skewed Distribution
A skewed distribution is asymmetric. One tail is longer than the other. Two types exist.
Positively skewed (right-skewed): The tail extends to the right. The mean is greater than the median. Process data with a natural lower boundary — such as cycle time, response time, or defect count — frequently skews right because outputs cannot fall below zero but have no upper limit.
Negatively skewed (left-skewed): The tail extends to the left. The mean is less than the median. This pattern appears when a natural upper boundary limits the data, such as a purity percentage that cannot exceed 100%.
Six Sigma context: Skewed data requires attention before capability analysis. Standard Cpk calculations on skewed data overestimate or underestimate the true defect rate. Log transformation, square root transformation, or Box-Cox transformation often normalizes skewed data.
4. Bimodal and Multimodal Distributions
A bimodal distribution has two peaks. A multimodal distribution has three or more. Each peak represents a cluster of data from a different sub-population or process condition.
Bimodal distributions almost always signal that data from two distinct groups — two shifts, two machines, two operators, two suppliers — has been combined into one dataset without stratification.
Six Sigma context: A bimodal histogram is one of the most important diagnostic signals in the Measure phase. It tells the team that process data is mixed and must be stratified before analysis. Calculating Cpk on bimodal data produces meaningless results. The team must separate the two populations and analyze each independently.
5. Exponential Distribution
The exponential distribution models the time between independent events that occur at a constant average rate. Its shape is right-skewed with a long tail. High frequencies cluster near zero and fall off rapidly.
Six Sigma context: Exponential distributions appear in reliability and maintenance contexts — the time between equipment failures, the time between customer arrivals, or the time between defects in a low-defect process. These situations call for reliability engineering tools rather than standard capability analysis.
6. Log-Normal Distribution
When the logarithm of a variable follows a normal distribution, the original variable follows a log-normal distribution. Log-normal data is right-skewed and bounded below by zero.
Log-normal distributions appear in particle size measurements, income distributions, failure times in degradation processes, and many biological measurements.
Six Sigma context: Taking the natural log of log-normal data produces a normal distribution. This transformation allows the full suite of normal-assumption statistical tools to be applied to the transformed data.
7. Poisson Distribution
The Poisson distribution models the number of independent events occurring within a fixed interval of time, space, or volume, when events happen at a constant average rate. It is a discrete distribution — it applies to counts, not continuous measurements.
Six Sigma context: Poisson distributions are used in attribute data analysis when counting the number of defects per unit (not the number of defective units). The c-chart and u-chart in Statistical Process Control are built on the Poisson distribution. When average defect count is low (typically below 5 per unit), the Poisson model applies.
8. Chi-Square Distribution
The chi-square distribution is used in hypothesis testing. It arises as the distribution of a sum of squared standard normal random variables. It is right-skewed, with the degree of skewness decreasing as degrees of freedom increase.
Six Sigma context: The chi-square distribution underpins chi-square tests of goodness-of-fit and tests of independence for categorical data. It also forms the basis for confidence intervals and hypothesis tests on variance.
How to Identify a Data Distribution: The Six Sigma Approach
Identifying the distribution of process data follows a three-step sequence in a DMAIC project.
Step 1: Plot a Histogram First
The histogram is the primary visual tool for assessing distribution shape. Plot the data before performing any formal statistical test. A histogram immediately reveals whether the data appears symmetric, skewed, bimodal, or flat.
iSixSigma confirms this directly: “Often, a simple plot of the data on either a histogram or probability plot will provide you with enough insight into how your data looks. This will keep you from having to do more complicated analysis.”
What to look for in a histogram:
- Bell-shaped and symmetric: Consistent with normal distribution
- Right or left tail: Consistent with skewed distribution
- Two separate peaks: Signals bimodal distribution — investigate stratification
- Flat, rectangular shape: Consistent with uniform distribution
- Sharp peak near zero with long right tail: Consistent with exponential or log-normal
Step 2: Perform the Anderson-Darling Normality Test
If the histogram suggests a normal or approximately normal distribution, confirm it with the Anderson-Darling normality test. This formal statistical test measures how much the observed data deviates from a theoretical normal distribution.
The Anderson-Darling test was proposed by Theodore W. Anderson and Donald A. Darling in 1952. It is the standard normality test used in Six Sigma practice because it is particularly sensitive to deviations in the tails of the distribution.
Interpreting the Anderson-Darling test result:
- p-value > 0.05: The data does not significantly deviate from normality. Proceed with normal-assumption tools (Cpk, X-bar chart, t-tests).
- p-value ≤ 0.05: The data significantly deviates from normality. Normal-assumption tools are not valid. Consider transformation or non-parametric methods.
The Lean 6 Sigma Hub confirms this interpretation: “A p-value greater than 0.05 typically suggests the data is consistent with normal distribution.”
One important caution: a p-value above 0.05 does not prove the data is normal. It means there is no strong statistical evidence against normality. A high p-value should always be combined with visual assessment of the histogram and probability plot.
Step 3: Run a Probability Plot (Normal Probability Plot)
A probability plot (also called a Q-Q plot or normal probability plot) plots each data point against the value it would have if the data were perfectly normal. If the data follows a normal distribution, the points fall approximately along a straight diagonal line. Departures from the line indicate non-normality.
The probability plot is particularly useful for identifying the nature of non-normality — whether it is in the tails (the ends of the line bend away) or in the center (the middle of the line curves).
In Minitab: Stat > Basic Statistics > Normality Test. Select the variable, choose the Anderson-Darling test, and check “Include confidence interval.” Minitab produces both the probability plot and the Anderson-Darling p-value in one output.
Also Read: Leptokurtic Distribution
What to Do When Data Is Non-Normal
When process data fails the normality test, practitioners have three options.
Option 1: Apply a data transformation. Mathematical transformations convert non-normal data into approximately normal data. The Six Sigma Institute recommends:
- Log transformation: Effective for right-skewed data, particularly when values span several orders of magnitude.
- Square root transformation: Effective for count data following a Poisson distribution.
- Box-Cox transformation: A family of power transformations that Minitab can optimize automatically to best normalize a given dataset.
After transformation, re-test for normality and calculate capability on the transformed scale.
Option 2: Use non-parametric statistical tests. Non-parametric tests make no assumption about the distribution shape. The Mann-Whitney test replaces the two-sample t-test. The Kruskal-Wallis test replaces one-way ANOVA. These tests analyze medians rather than means and work on any distribution.
Option 3: Use a distribution-specific model. If the data clearly follows a known non-normal distribution — log-normal, Weibull, exponential — the appropriate distribution-specific capability analysis can be run. Minitab’s capability analysis allows practitioners to specify the distribution that best fits the data.
Data Distributions Across DMAIC Phases
Understanding and correctly handling data distributions is relevant across multiple DMAIC phases.
| DMAIC Phase | Role of Data Distributions |
| Define | The project Y variable is identified. Its data type (continuous or attribute) and likely distribution shape are considered when planning data collection. |
| Measure | Baseline data is plotted in a histogram. The distribution shape is identified. The Anderson-Darling test confirms or rejects normality. Transformations are applied if needed before capability analysis. |
| Analyze | Hypothesis tests are selected based on the confirmed distribution type. Normal data uses parametric tests. Non-normal data uses non-parametric tests or distribution-specific methods. |
| Improve | Pilot data from the improved process is compared against the baseline distribution. The team tests whether the distribution has shifted or narrowed as expected from the improvement. |
| Control | Control chart type is selected based on distribution. X-bar and R charts assume normality. Attribute charts (p, np, c, u) use binomial or Poisson models. SPC monitoring assumes the correct distributional model. |
Frequently Asked Questions: Data Distributions
Q: What is a data distribution in simple terms?
A: A data distribution is the pattern showing how often each value or range of values appears in a dataset. When you plot data in a histogram, the shape of that histogram is the distribution. It tells you where values cluster, how spread out they are, and whether the pattern is symmetric, skewed, or has multiple peaks. In Six Sigma, the distribution shape determines which statistical tools can be applied to the data.
Q: Why does the distribution type matter in Six Sigma?
A: The distribution type matters because most Six Sigma statistical tools assume the data follows a normal (bell-shaped) distribution. Process capability indices, X-bar control charts, and most hypothesis tests all make this assumption. If the data is non-normal — skewed, bimodal, or following an exponential pattern — applying normal-assumption tools produces incorrect results. Identifying the correct distribution is the prerequisite to choosing the right tool.
Q: What is the Anderson-Darling test and when is it used?
A: The Anderson-Darling test is a statistical normality test proposed by Theodore W. Anderson and Donald A. Darling in 1952. It measures how much a dataset deviates from a theoretical normal distribution and produces a p-value. In Six Sigma, it is used during the Measure phase to formally confirm whether process data follows a normal distribution. A p-value above 0.05 indicates the data does not significantly deviate from normality. A p-value of 0.05 or below indicates significant non-normality.
Q: What are the most common data distributions in Six Sigma?
A: Eight distributions appear most frequently in Six Sigma project data: normal (the standard bell curve used in most Six Sigma tools), uniform (equal probability across all values), skewed (asymmetric with a longer tail in one direction), bimodal or multimodal (two or more peaks indicating mixed sub-populations), exponential (models time between independent events), log-normal (skewed data whose logarithm is normal), Poisson (counts of events in a fixed interval), and chi-square (used in hypothesis tests on categorical data and variance).
Q: What should I do if my data fails the normality test?
A: When data fails the Anderson-Darling normality test, practitioners have three options. First, apply a data transformation such as a logarithmic, square root, or Box-Cox transformation to convert the data to approximate normality, then re-test.
Second, use non-parametric statistical tests that do not assume a specific distribution shape, such as Mann-Whitney instead of the two-sample t-test.
Third, identify the actual distribution the data follows (log-normal, Weibull, exponential) and use distribution-specific capability analysis tools available in Minitab.
Q: What does a bimodal distribution mean in a Six Sigma project?
A: A bimodal distribution — a histogram with two distinct peaks — almost always means that data from two different sub-populations or process conditions has been combined into one dataset without stratification. Common causes include two shifts, two machines, two operators, or two suppliers producing output at different centers.
The correct response is to identify the stratification variable causing the two peaks, separate the data into two groups, and analyze each group independently. Calculating process capability on bimodal data produces meaningless results.
Q: How do I identify the distribution of my data in Minitab?
A: In Minitab, go to Stat > Basic Statistics > Normality Test to run the Anderson-Darling test and view a normal probability plot. For identifying other distributions, use Stat > Quality Tools > Individual Distribution Identification, which fits the data to 16 different distribution families and displays goodness-of-fit statistics for each, allowing you to identify which distribution best describes your data.
Data Distributions Training in Six Sigma
Identifying, testing, and correctly handling data distributions is a core competency across all Six Sigma belt levels. Yellow Belts learn to recognize basic distribution shapes in histograms. Green Belts conduct Anderson-Darling normality tests, select appropriate statistical tools based on distribution type, and apply basic data transformations.
Black Belts apply distribution-specific capability analysis, use non-parametric tests, and design data collection plans that account for expected distribution shapes.
At Six Sigma Development Solutions Inc, data distributions and normality testing are covered as applied Measure and Analyze phase skills in our Green Belt and Black Belt training programs. Practitioners learn to identify distribution types visually, run and interpret the Anderson-Darling test in Minitab, and make the correct tool selection decision based on what the data actually shows.
We offer training in three formats:
- Onsite training — delivered at your facility, using your actual process data in histogram and normality testing exercises.
- Live virtual training — instructor-led sessions with real-time Minitab exercises covering distribution identification and normality testing.
- Online training — self-paced Green Belt and Black Belt certification programs covering all IASSC-testable distribution and statistical content.
Explore our Six Sigma training programs or contact our team to find the right program for your 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.
