9.3 Basic Distribution Types: Normal, Binomial & Poisson

Basic Distribution Types: Normal, Binomial & Poisson

Quick Answer: Statistical analysis in Six Sigma relies on probability distributions to model process behavior. The continuous Normal (Gaussian) distribution is symmetrical (mean = median = mode) and governed by the Empirical Rule (68.27% within $\pm 1\sigma$, 95.45% within $\pm 2\sigma$, 99.73% within $\pm 3\sigma$), utilizing $Z$-scores to calculate process yields. Discrete distributions model count data: the Binomial distribution tracks binary pass/fail outcomes across independent trials, whereas the Poisson distribution tracks defect frequencies across an area of opportunity. Deviations from normality, such as skewness or bimodal clustering, provide immediate visual clues of underlying operational mixing or boundary effects. Independent CSSYB study guide by OpenExamPrep.

Continuous vs. Discrete Data Distributions in Six Sigma

In the Analyze phase of DMAIC, Six Sigma practitioners evaluate process performance by fitting collected baseline data to theoretical probability distributions. A probability distribution provides a mathematical model describing the likelihood of observing various values of a process output metric ($Y$).

The fundamental starting point for selecting an appropriate distribution model is distinguishing between continuous and discrete data:

• Continuous Data (Variables Data): Measurements taken on a continuous, uninterrupted numerical scale (such as dimensions in millimeters, cycle times in seconds, temperatures in Celsius, or tensile strength in megapascals). Continuous data is modeled using continuous distributions, predominantly the Normal (Gaussian) distribution.
• Discrete Data (Attribute Data): Categorical counts, classifications, or integer events (such as conforming versus nonconforming parts, customer billing complaints per month, or solder voids on a circuit board). Discrete data is modeled using discrete probability distributions, primarily the Binomial and Poisson distributions. The CSSYB BoK names the normal and binomial distributions; the Poisson distribution is included here because it is the usual model for counts of defects, and exam answer choices often contrast it with the binomial.

---

The Continuous Normal (Gaussian) Distribution

The Normal distribution is the mathematical cornerstone of statistical quality control and process capability modeling. Described by Carl Friedrich Gauss, it depicts the behavior of processes influenced by many independent, random common causes acting together.

Key Mathematical & Geometric Properties

• Symmetrical Bell Shape: The curve is perfectly symmetrical around the central arithmetic mean ($\mu$).
• Coincidence of Central Tendencies: The mean, median, and mode are identical and located exactly at the center of the distribution.
• Asymptotic Tails: The tails of the curve extend infinitely in both positive and negative directions without ever touching the horizontal axis ($y = 0$).
• Total Probability: The total area under the probability density curve equals exactly $1.0$ (or $100\%$ probability).

The Empirical Rule (68-95-99.7 Rule)

For any data set that follows a normal distribution, the dispersion of observations relative to the population mean ($\mu$) and standard deviation ($\sigma$) conforms to fixed, universal percentages:

• $\mu \pm 1\sigma$: Approximately 68.27% of all data points fall within one standard deviation of the mean (roughly 2 out of every 3 observations).
• $\mu \pm 2\sigma$: Approximately 95.45% of all data points fall within two standard deviations of the mean (roughly 19 out of every 20 observations).
• $\mu \pm 3\sigma$: Approximately 99.73% of all data points fall within three standard deviations of the mean (roughly 369 out of every 370 observations).

Only 0.27% of observations fall outside the $\pm 3\sigma$ control boundaries (0.135% in each tail), which forms the theoretical foundation for Shewhart three-sigma control limits.

The Standard Normal Distribution and the $Z$-Score

Because real-world processes operate with different units and scales, practitioners convert raw data into the Standard Normal Distribution ($Z$-distribution), which has a standardized mean of $\mu = 0$ and standard deviation of $\sigma = 1$.

The $Z$-score (standard score) measures the exact number of standard deviations an individual observation ($X$) lies above or below the population mean:

$$Z = \frac{X - \mu}{\sigma}$$

Where:

• $X$ = Specific value or customer specification limit (USL or LSL)
• $\mu$ = Process mean
• $\sigma$ = Process standard deviation

Yellow Belt Calculation Example: A precision shaft grinding process has a mean diameter $\mu = 25.00\text{ mm}$ with $\sigma = 0.05\text{ mm}$. The customer's Upper Specification Limit (USL) is $25.10\text{ mm}$. The $Z$-score for the upper limit is:

$$Z = \frac{25.10 - 25.00}{0.05} = \frac{0.10}{0.05} = 2.0$$

Consulting a standard normal table reveals that $Z = 2.0$ leaves approximately $2.28\%$ of the distribution in the upper tail, indicating that about $2.28\%$ of shafts will exceed the customer's maximum allowable diameter.

---

Discrete Probability Distributions

When evaluating counts, classifications, and attribute pass/fail occurrences, Yellow Belts apply two core discrete distributions.

1. The Binomial Distribution

The Binomial distribution models discrete counts of successes or failures across a sequence of independent trials.

• Governing Conditions:
  1. The process consists of a fixed number of identical trials ($n$).
  2. Each trial has only two mutually exclusive outcomes: success or failure (conforming or nonconforming).
  3. The probability of success ($p$) remains constant from trial to trial.
  4. All trials are statistically independent.
• Six Sigma Application: The Binomial distribution models the number of defective (nonconforming) units in an inspection lot. It provides the mathematical foundation for attribute $p$-charts (proportion nonconforming) and $np$-charts (number nonconforming).

2. The Poisson Distribution

The Poisson distribution models the frequency of discrete events or defects occurring within a specified, continuous area of opportunity (such as an interval of time, physical area, volume, or length).

• Governing Conditions:
  1. Events occur independently at a constant average rate ($\lambda$).
  2. The occurrence of an event in one interval does not influence occurrences in another.
  3. Multiple occurrences in an infinitesimally small interval are practically zero.
  4. Key Statistical Property: The mathematical variance of a Poisson distribution equals its mean ($\sigma^2 = \mu = \lambda$).
• Six Sigma Application: The Poisson distribution models the count of defects per unit—such as paint blemishes on an automobile hood, typing errors per standard contract page, or customer complaints logged per hour. It serves as the statistical engine for $c$-charts (constant area of opportunity) and $u$-charts (variable area of opportunity).

Critical Difference: Defective Units vs. Defects

Yellow Belts must never confuse defective units with defects:

• Defective Unit (Binomial): An entire product or transaction that fails to meet criteria (a cracked windshield). A unit is either good or bad.
• Defect (Poisson): An individual flaw or blemish on a unit (three small chips on a single windshield). A single defective unit may contain multiple distinct defects.

---

Diagnostic Interpretation of Distribution Shapes

When data is plotted on a histogram, the resulting geometric shape provides immediate diagnostic clues regarding underlying process dynamics.

When a Yellow Belt encounters a bimodal distribution during the Analyze phase, the mandatory investigative action is stratification—disaggregating the dataset by shift, machine, operator, or vendor to isolate and analyze the two underlying distributions separately.