8.2 The Normal Distribution & The Empirical Rule (68-95-99.7)

8.2 The Normal Distribution & The Empirical Rule (68-95-99.7)

Core Principle: The Normal Distribution (Gaussian curve) is the foundational probability distribution of Six Sigma quality engineering. Characterized by its symmetrical, unimodal bell shape, it is defined entirely by two parameters: the mean ($\mu$), which governs central location, and the standard deviation ($\sigma$), which governs spread. The Empirical Rule (68-95-99.7 Rule) provides instant calculation of conforming yields and defect percentages across $\pm 1\sigma$, $\pm 2\sigma$, and $\pm 3\sigma$ tolerances, while skewness and kurtosis quantify structural departures from normality.

---

Mathematical Foundations of the Normal Distribution

First conceptualized by Abraham de Moivre and later refined by Carl Friedrich Gauss, the Normal (Gaussian) Distribution is a continuous probability distribution that models phenomena where variations arise from the cumulative effect of numerous small, independent random factors. In manufacturing assembly, transactional processing, and chemical synthesis, natural variation naturally mirrors this bell-shaped profile.

Mathematically, the probability density function (PDF) of a normal random variable $X$ is expressed as:

$f(x) = \frac{1}{\sigma \sqrt{2\pi}} e^{-\frac{1}{2}\left(\frac{x - \mu}{\sigma}\right)^2} \quad \text{for } -\infty < x < \infty$

Where:

• $\mu$ is the true population mean (location parameter).
• $\sigma$ is the true population standard deviation (scale parameter).
• $\pi \approx 3.14159$ and $e \approx 2.71828$ are fundamental mathematical constants.

                                Normal Distribution
                                 Perfect Symmetry
                                 Total Area = 1.0
                                       │
                                      / \
                                     / │ \
                                    /  │  \
                                   /   │   \
                                 /     │     \
                               /       │       \
                           _.-'        │        '-._
                       _.-'            │            '-._
                   _.-'                │                '-._
             ─────┴────────────────────┼────────────────────┴─────
                  -3σ        -1σ       μ       +1σ        +3σ
                                 Mean = Median = Mode

Core Theoretical Properties

1. Unimodal and Symmetrical: The distribution has a single peak. The left and right halves are exact mirror images across the central vertical axis.
2. Coincidence of Central Measures: In a perfectly normal distribution, the Mean, Median, and Mode are mathematically identical ($\mu = \tilde{x} = \text{Mode}$).
3. Asymptotic Tails: The tails extend infinitely in both positive and negative directions without ever touching the horizontal axis ($f(x) > 0$ for all $x$).
4. Total Area Equals 1.0: The total area under the probability density curve is exactly equal to $1.0$ (representing 100% probability).
5. Points of Inflection: The curve transitions from concave downward to concave upward at exactly one standard deviation on either side of the mean ($x = \mu \pm \sigma$).
6. Governed by Two Parameters: Once $\mu$ and $\sigma$ are known, the normal distribution is completely specified. Changing $\mu$ shifts the curve along the horizontal axis without changing its shape; changing $\sigma$ widens or narrows the curve while maintaining a total area of 1.0.

---

The Empirical Rule (The 68-95-99.7 Rule)

For any dataset that is reasonably normally distributed, the Empirical Rule (also known as the Three-Sigma Rule) defines the exact proportion of observations expected to fall within integer multiples of the standard deviation from the mean.

                           The Empirical Rule Breakdown

                                       μ
                                       │
                                ┌──────┴──────┐
                                │   68.27%    │
                            ────┴─────────────┴────
                            μ - 1σ           μ + 1σ
                        ┌─────────────────────────────┐
                        │           95.45%            │
                    ────┴─────────────────────────────┴────
                    μ - 2σ                           μ + 2σ
                ┌─────────────────────────────────────────────┐
                │                   99.73%                    │
            ────┴─────────────────────────────────────────────┴────
            μ - 3σ                                           μ + 3σ

Detailed Interval and Tail Probabilities

1. Within $\mu \pm 1\sigma$:

   • Inside Interval: Exactly 68.27% (commonly rounded to 68%) of all observations.
   • Outside Interval: $100% - 68.27% = \mathbf{31.73%}$.
   • Tail Probabilities: Because the curve is symmetrical, dividing the excluded area equally gives 15.865% in the upper tail ($> \mu + 1\sigma$) and 15.865% in the lower tail ($< \mu - 1\sigma$).

2. Within $\mu \pm 2\sigma$:

   • Inside Interval: Exactly 95.45% (commonly rounded to 95.5% or 95%) of all observations.
   • Outside Interval: $100% - 95.45% = \mathbf{4.55%}$.
   • Tail Probabilities: Exactly 2.275% falls in the lower tail ($< \mu - 2\sigma$) and 2.275% falls in the upper tail ($> \mu + 2\sigma$). In industrial parts-per-million (PPM) metrics, each single tail corresponds to 22,750 PPM non-conformance.

3. Within $\mu \pm 3\sigma$:

   • Inside Interval: Exactly 99.73% (commonly rounded to 99.7%) of all observations.
   • Outside Interval: $100% - 99.73% = \mathbf{0.27%}$.
   • Tail Probabilities: Exactly 0.135% (1,350 PPM) falls in each tail. The total bilateral defect rate for a traditional 3-sigma process is 2,700 PPM ($0.0027$).

Master Reference Table: Empirical Rule Metrics

---

Step-by-Step Worked Industrial Calculation: Applying the Empirical Rule

Scenario: An automated CNC lathe manufactures precision drive shafts. The diameter of the shafts follows a normal distribution with a mean $\mu = 50.00\text{ mm}$ and a standard deviation $\sigma = 0.05\text{ mm}$. The customer's engineering tolerance specification is $50.00 \pm 0.10\text{ mm}$ (Lower Specification Limit $\text{LSL} = 49.90\text{ mm}$, Upper Specification Limit $\text{USL} = 50.10\text{ mm}$).

Step 1: Express Specification Limits in Terms of Standard Deviations

Determine how many standard deviations the specification limits lie from the process mean: $\text{Distance to USL} = \frac{50.10 - 50.00}{0.05} = \frac{0.10}{0.05} = +2.0\sigma$ $\text{Distance to LSL} = \frac{49.90 - 50.00}{0.05} = \frac{-0.10}{0.05} = -2.0\sigma$

The customer tolerance corresponds exactly to the $\mu \pm 2.0\sigma$ interval.

Step 2: Calculate Conforming Process Yield

Applying the Empirical Rule for $\pm 2\sigma$: $\text{Conforming Yield} = P(49.90 \le X \le 50.10) = 95.45\%$

Out of every 100,000 shafts produced, 95,450 will meet specification.

Step 3: Calculate Bilateral Defect Proportions

The total non-conforming proportion is: $\text{Total Defect Rate} = 100\% - 95.45\% = 4.55\% \quad (0.0455)$

Converting to total process PPM: $\text{Total Defect PPM} = 0.0455 \times 1,000,000 = 45,500 \text{ PPM}$

Step 4: Categorize Non-Conformance by Scrap vs. Rework

Because the normal curve is perfectly symmetrical around $\mu = 50.00\text{ mm}$:

• Undersized Shafts ($< 49.90\text{ mm}$): Cannot be salvaged and must be scrapped. $\text{Scrap Rate} = \frac{4.55\%}{2} = 2.275\% = 22,750 \text{ PPM}$
• Oversized Shafts ($> 50.10\text{ mm}$): Can be returned to the lathe for re-machining. $\text{Rework Rate} = \frac{4.55\%}{2} = 2.275\% = 22,750 \text{ PPM}$

This simple Empirical Rule calculation enables production managers to budget for exact material scrap costs and machine rework capacity.

---

Distribution Shape Characteristics: Skewness & Kurtosis

When process data is not perfectly normal, Green Belts evaluate two critical higher-order moments: skewness (asymmetry) and kurtosis (tail heaviness and peakedness).

1. Skewness (Third Standardized Moment)

Skewness measures the degree of directional asymmetry of a distribution relative to its mean.

   Positive (Right) Skew                Symmetrical                 Negative (Left) Skew
       Tail to Right                       Normal                       Tail to Left

           ▲                                  ▲                                  ▲
          / \                                / \                                / \
         /   \                              /   \                              /   \
        /     \                            /     \                            /     \
       /       \                          /       \                          /       \
      /         \_                       /         \                       _/         \
     /            '--.                  /           \                  .--'            \
  ──┴─────────────────┴──            ──┴─────────────┴──            ──┴─────────────────┴──
    Mode < Med < Mean                 Mean = Med = Mode               Mean < Med < Mode

[!TIP] Exam Rule of Thumb: In a skewed distribution, the Mean is always pulled furthest in the direction of the tail. The Mode remains under the peak, and the Median sits comfortably in between. Therefore, for positive skew: $\text{Mode} < \text{Median} < \text{Mean}$. For negative skew: $\text{Mean} < \text{Median} < \text{Mode}$.

2. Kurtosis (Fourth Standardized Moment)

Kurtosis measures the peakedness of a distribution and, more importantly, the thickness/heaviness of its tails relative to a normal distribution.

• Raw Kurtosis: The standard normal distribution has a raw kurtosis of exactly $3.0$.
• Excess Kurtosis: To simplify interpretation, statistical software computes Excess Kurtosis by subtracting 3: $\text{Excess Kurtosis} = \text{Kurtosis} - 3.0$.

                             Kurtosis Classifications
                                        ▲
                                      / │ \
                                     /  │  \    Leptokurtic (Excess > 0)
                                    /   │   \   Sharp peak, fat heavy tails
                                   │    │    │
                                   /    │    \   Mesokurtic (Excess = 0)
                                 /  \   │   /  \ Standard Normal curve
                               /     '─.│.─'     \
                           _.-'         │         '-._  Platykurtic (Excess < 0)
                     ─────┴─────────────┼─────────────┴───── Flat peak, thin tails

1. Mesokurtic (Excess Kurtosis $\approx 0$):
   • A standard normal distribution. Baseline tail risk where the 68-95-99.7 Empirical Rule holds true.
2. Leptokurtic (Excess Kurtosis $> 0$):
   • Characterized by a sharp, tall central peak and fat, heavy tails ("lepto" = slender).
   • Six Sigma Risk: Indicates higher probability of extreme events or "black swans" than predicted by the normal curve. Outliers occur far more frequently in the tails.
3. Platykurtic (Excess Kurtosis $< 0$):
   • Characterized by a flatter, broader central peak and thin, light tails ("platy" = broad/flat, like a plateau).
   • Observations cluster evenly across the center with few extreme outliers.

---

Critical CSSC Exam Traps

• Trap 1: Confusing 3-Sigma Quality with Six Sigma Quality — The Empirical Rule states that $\pm 3\sigma$ contains 99.73% of data, yielding a 2,700 PPM defect rate. Candidates mistakenly think 2,700 PPM represents "Six Sigma." In reality, Six Sigma quality demands a process spread where the nearest specification limit is at least $6\sigma$ away, producing only 3.4 DPMO (after accounting for the 1.5-sigma drift).
• Trap 2: Forgetting to Halve the Excluded Tail Area — When an exam question asks for the defect rate above an Upper Specification Limit set at $\mu + 2\sigma$, candidates often answer 4.55%. That is the bilateral (two-tailed) defect rate! The unilateral (one-tailed) defect rate is half of that: $4.55% / 2 = \mathbf{2.275%}$.
• Trap 3: Reversing Skewness Ordering — Remembering whether positive skew means Mean > Median or Mean < Median. Remember the mnemonic: "The Mean follows the tail." A right-skewed distribution has a right tail, so the mean is pulled to the right (greater than median).
• Trap 4: Blindly Applying the Empirical Rule to Non-Normal Data — Applying the 68-95-99.7 percentages to cycle-time data that is heavily exponential or lognormal. The Empirical Rule is mathematically valid only when data conforms to the normal distribution.