9.2 Descriptive Statistics & Measures of Variation

9.2 Descriptive Statistics & Measures of Variation

Quality engineering relies on descriptive statistics to transform large volumes of raw operational data into actionable process intelligence. In manufacturing and service environments, processes never produce identical outputs; natural and assignable causes inevitably introduce variation. Understanding the central location, dispersion, and distributional shape of process data allows quality professionals to evaluate stability, compute capability indices, and identify operational degradation before customer defects occur.

---

1. Measures of Central Tendency (Location)

Measures of central tendency identify the center point or typical value around which numerical data clusters. The three primary metrics tested on the ASQ CQIA examination are the Mean, Median, and Mode.

+-------------------------------------------------------------------------+
|                     MEASURES OF CENTRAL TENDENCY                        |
+-------------------------------------------------------------------------+
|                                                                         |
|  1. MEAN (Arithmetic Average)                                           |
|     * Sum of all values divided by total count n                        |
|     * Highly sensitive to extreme outliers                              |
|     * Formula: x̄ = (Σ xi) / n                                            |
|                                                                         |
|  2. MEDIAN (50th Percentile)                                            |
|     * Exact middle value of an ordered dataset                          |
|     * Robust against outliers and heavy tails                           |
|     * Ideal for skewed cycle times, repair durations, and salaries      |
|                                                                         |
|  3. MODE (Peak Frequency)                                               |
|     * Most frequently occurring data value                              |
|     * The only central tendency metric applicable to Nominal data       |
|     * Identifies multi-stream process mixing (bimodal distributions)    |
|                                                                         |
+-------------------------------------------------------------------------+

Mathematical Definitions and Formulas

1. The Arithmetic Mean ($\bar{x}$ or $\mu$)

• Sample Mean ($\bar{x}$): The sum of all observed sample values divided by the sample size $n$.

$\bar{x} = \frac{\sum_{i=1}^{n} x_i}{n} = \frac{x_1 + x_2 + \dots + x_n}{n}$

• Population Mean ($\mu$): The arithmetic average of all $N$ elements in the entire population.

$\mu = \frac{\sum_{i=1}^{N} X_i}{N}$

• Operational Characteristics: The mean is mathematically tractable and serves as the baseline for hypothesis testing and control charts (e.g., $\bar{X}$ charts). However, it is non-robust because a single extreme outlier pulls the mean heavily in the direction of the anomaly.

2. The Median ($\tilde{x}$ or $M$)

• Definition: The physical midpoint of a dataset when the values are arranged in ascending order ($50\text{th}$ percentile). It divides the distribution into two equal halves.
• Calculation Protocol:
  • If $n$ is odd, the median is the value at position $(n + 1) / 2$.
  • If $n$ is even, the median is the arithmetic average of the two middle values at positions $n / 2$ and $(n / 2) + 1$.
• Operational Characteristics: The median is resistant (robust) to extreme outliers. It is the preferred measure of central tendency for skewed datasets, such as customer wait times, equipment repair turnaround times (MTTR), and income distributions.

3. The Mode

• Definition: The value that appears with the highest frequency in a dataset.
• Distributional Types:
  • Unimodal: Exactly one peak value.
  • Bimodal: Two distinct values sharing peak frequency, often indicating the mixing of two distinct machine streams, tool cavities, or operator shifts.
  • Multimodal: Three or more peak values.
  • No Mode: Occurs when all values appear with equal frequency.
• Operational Characteristics: The mode is the only central tendency metric valid for Nominal attribute data (e.g., identifying the most common defect code).

Comparison of Central Tendency Measures

---

2. Measures of Dispersion (Variation / Spread)

In quality management, knowing the central location of a process tells only half the story. A famous engineering adage warns: "A person with one foot in boiling water and one foot in ice water is, on average, comfortable." Variation is the primary enemy of quality. Measures of dispersion quantify the extent of spread, scattering, or variability around the central tendency.

+-------------------------------------------------------------------------+
|                        MEASURES OF DISPERSION                           |
+-------------------------------------------------------------------------+
|                                                                         |
|  1. RANGE (R)                ──► R = Max - Min (Simplest, small n <= 10)|
|                                                                         |
|  2. SAMPLE VARIANCE (s²)     ──► Mean squared deviation using (n - 1)   |
|                                  degrees of freedom (Bessel's correction)|
|                                                                         |
|  3. SAMPLE STANDARD DEV. (s) ──► Square root of variance; expressed in  |
|                                  original measurement engineering units |
|                                                                         |
|  4. COEFF. OF VARIATION (CV) ──► CV = (s / x̄) * 100% (Relative variation)|
|                                                                         |
+-------------------------------------------------------------------------+

Mathematical Formulations

1. Range ($R$)

• Definition: The arithmetic difference between the largest value ($x_{\max}$) and the smallest value ($x_{\min}$) in a sample.

$R = x_{\max} - x_{\min}$

• Application: Standard measure of dispersion used in sub-grouped Statistical Process Control ($\bar{X}-R$ charts) when subgroup size is small ($n \le 10$). For larger sample sizes, the range becomes inefficient and unreliable because it ignores all intermediate data points.

2. Population Variance ($\sigma^2$) vs. Sample Variance ($s^2$)

• Population Variance ($\sigma^2$): The average of the squared deviations from the population mean $\mu$, evaluated over all $N$ units.

$\sigma^2 = \frac{\sum_{i=1}^{N} (X_i - \mu)^2}{N}$

• Sample Variance ($s^2$): The sum of squared deviations from the sample mean $\bar{x}$, divided by the degrees of freedom ($n - 1$).

$s^2 = \frac{\sum_{i=1}^{n} (x_i - \bar{x})^2}{n - 1}$

Bessel's Correction ($n - 1$ Degrees of Freedom): When calculating sample variance, dividing by $n$ produces a biased estimate that systematically underestimates the true population variance. This occurs because sample observations naturally cluster closer to their own sample mean $\bar{x}$ than to the true population mean $\mu$. Dividing by $n - 1$ removes this downward bias, making $s^2$ an unbiased estimator of $\sigma^2$.

3. Sample Standard Deviation ($s$)

• Definition: The positive square root of the sample variance. Unlike variance (which is expressed in squared units, such as $\text{mm}^2$), standard deviation is expressed in the original linear engineering units of the measurement (e.g., $\text{mm}$, $\text{psi}$, $\text{seconds}$).

$s = \sqrt{s^2} = \sqrt{\frac{\sum_{i=1}^{n} (x_i - \bar{x})^2}{n - 1}}$

4. Coefficient of Variation ($CV$)

• Definition: A dimensionless metric of relative dispersion expressing the standard deviation as a percentage of the mean.

$CV = \left( \frac{s}{\bar{x}} \right) \times 100\%$

• Application: Enables direct comparison of process variability across different machines or measurement characteristics having widely different units or absolute magnitudes (e.g., comparing the precision of an injection mold producing $2\text{ g}$ parts versus a press producing $200\text{ kg}$ engine blocks).

Step-by-Step Sample Standard Deviation Calculation

Consider a quality inspector recording the thickness (in mm) of $n = 5$ machined steel shims: ${10, 12, 14, 16, 18}$.

1. Step 1: Compute the Sample Mean ($\bar{x}$): $\bar{x} = \frac{10 + 12 + 14 + 16 + 18}{5} = \frac{70}{5} = 14.0\text{ mm}$
2. Step 2: Calculate Deviations and Squared Deviations:
   • $(10 - 14)^2 = (-4)^2 = 16$
   • $(12 - 14)^2 = (-2)^2 = 4$
   • $(14 - 14)^2 = (0)^2 = 0$
   • $(16 - 14)^2 = (+2)^2 = 4$
   • $(18 - 14)^2 = (+4)^2 = 16$
3. Step 3: Sum the Squared Deviations (Sum of Squares, $SS$): $SS = \sum_{i=1}^{5} (x_i - \bar{x})^2 = 16 + 4 + 0 + 4 + 16 = 40.0$
4. Step 4: Compute Sample Variance ($s^2$) using $n - 1 = 4$: $s^2 = \frac{40.0}{5 - 1} = \frac{40.0}{4} = 10.0\text{ mm}^2$
5. Step 5: Compute Sample Standard Deviation ($s$): $s = \sqrt{10.0} \approx 3.162\text{ mm}$

---

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

The Normal (Gaussian) Distribution is the most vital continuous probability distribution in statistical quality control. It is a unimodal, perfectly symmetric, bell-shaped distribution governed entirely by two parameters: its mean $\mu$ (location) and standard deviation $\sigma$ (scale).

+-------------------------------------------------------------------------+
|                    THE EMPIRICAL RULE (68-95-99.7)                      |
+-------------------------------------------------------------------------+
|                                    │                                    |
|                                 .-'"'-.                                 |
|                              .-'   │   '-.                              |
|                           .-'      │      '-.                           |
|                        .-'  34.13% │ 34.13%  '-.                        |
|                     .-'            │            '-.                     |
|                  .-' 13.59%        │        13.59% '-.                  |
|             ..-' 2.14%             │             2.14% '-..             |
|       .----'   0.135%              │              0.135%   '----.       |
|  ─────┴────────────┴───────────────┴───────────────┴────────────┴─────► |
|      μ-3σ         μ-2σ            μ-1σ            μ+1σ         μ+3σ     |
|                                    μ                                    |
|       ├────────────── 68.26% ──────────────┤                            |
|       ├────────────────────── 95.44% ──────────────────────┤            |
|       ├───────────────────────────── 99.73% ───────────────────────────┤ |
+-------------------------------------------------------------------------+

Mathematical Intervals of the Empirical Rule

For any dataset that is approximately normally distributed, the Empirical Rule provides exact probability boundaries:

1. $\mu \pm 1\sigma$ Interval:
   • Encompasses approximately $68.26%$ of all observed values (roughly 2 out of every 3 data points).
   • Exactly $34.13%$ lies between $\mu$ and $\mu + 1\sigma$, and $34.13%$ lies between $\mu - 1\sigma$ and $\mu$.
2. $\mu \pm 2\sigma$ Interval:
   • Encompasses approximately $95.44%$ of all observed values (roughly 19 out of every 20 data points).
   • An additional $13.59%$ of data lies in each slice between $1\sigma$ and $2\sigma$.
3. $\mu \pm 3\sigma$ Interval:
   • Encompasses approximately $99.73%$ of all observed values (roughly 370 out of every 371 data points).
   • Only $0.27%$ ($2,700\text{ parts per million}$) falls outside the $\pm 3\sigma$ boundaries ($0.135%$ in each tail).
   • This $\pm 3\sigma$ boundary forms the statistical foundation for Shewhart Control Chart Upper and Lower Control Limits (UCL and LCL).

Normal vs. Non-Normal Distributions: Chebyshev's Inequality

If a process distribution is not normal (or its shape is entirely unknown), the Empirical Rule cannot be applied. Instead, Chebyshev's Inequality dictates that for any arbitrary mathematical distribution, the proportion of data falling within $k$ standard deviations of the mean ($k > 1$) is at least:

$P(|X - \mu| < k\sigma) \ge 1 - \frac{1}{k^2}$

• For $k = 2$: At least $1 - 1/4 = 75.0%$ of data lies within $\mu \pm 2\sigma$ (compared to $95.44%$ for normal data).
• For $k = 3$: At least $1 - 1/9 = 88.89%$ of data lies within $\mu \pm 3\sigma$ (compared to $99.73%$ for normal data).

---

4. Distributional Shape: Skewness and Kurtosis

When assessing whether process data conforms to standard statistical assumptions, quality associates evaluate two higher-moment shape descriptors: Skewness ($3\text{rd}$ standardized moment) and Kurtosis ($4\text{th}$ standardized moment).

+-------------------------------------------------------------------------+
|                        SKEWNESS (ASYMMETRY)                             |
+-------------------------------------------------------------------------+
|                                                                         |
|  POSITIVE (RIGHT) SKEW       SYMMETRICAL            NEGATIVE (LEFT) SKEW|
|       Mode                      Mean                     Mode           |
|        |                        |                          |            |
|       / \                      / \                        / \           |
|      /   \                    /   \                      /   \          |
|     /  M  \                  /  M  \                    /  M  \         |
|    /       \____            /       \              ____/       \        |
|   /         Mean\          /  Mode   \            /Mean         \       |
|  ─────────────────►       ─────────────►         ─────────────────►     |
|  Mean > Median > Mode     Mean = Median = Mode   Mean < Median < Mode   |
|  Tail extends RIGHT       Zero Skew (Symmetric)  Tail extends LEFT      |
|                                                                         |
+-------------------------------------------------------------------------+

Skewness (Direction of Asymmetry)

Skewness measures the degree of lopsidedness or lack of symmetry in a frequency distribution.

1. Symmetric Distribution (Skewness $= 0$):
   • The left and right tails are mirror images.
   • $\text{Mean} = \text{Median} = \text{Mode}$.
2. Positive Skew (Right-Skewed, Skewness $> 0$):
   • The long tail of extreme values extends toward the right (positive infinity).
   • The mean is pulled to the right by high outliers, while the mode remains at the peak.
   • Order of Metrics: $\text{Mean} > \text{Median} > \text{Mode}$.
   • Common Quality Phenomena: Machine repair times (MTTR), customer call hold times, defect counts per lot, tool wear rates, particle contamination levels.
3. Negative Skew (Left-Skewed, Skewness $< 0$):
   • The long tail extends toward the left (negative infinity).
   • The mean is pulled to the left by low outliers.
   • Order of Metrics: $\text{Mean} < \text{Median} < \text{Mode}$.
   • Common Quality Phenomena: Product purity percentages approaching $100%$, assembly yields with an upper physical ceiling, student exam scores on an easy certification test.

Kurtosis (Peakedness and Tail Weight)

Kurtosis measures the "tailedness" and outlier-proneness of a probability distribution relative to a normal distribution.

+-------------------------------------------------------------------------+
|                         KURTOSIS (TAILEDNESS)                           |
+-------------------------------------------------------------------------+
|                                                                         |
|      Leptokurtic (Heavy tails, sharp peak, Excess Kurtosis > 0)         |
|      Mesokurtic  (Normal bell curve shape, Excess Kurtosis = 0)         |
|      Platykurtic (Thin tails, flat/broad peak, Excess Kurtosis < 0)     |
|                                                                         |
|                         /\  <-- Leptokurtic                             |
|                        /  \                                             |
|                       / /\ \ <-- Mesokurtic (Normal)                    |
|                      / /  \ \                                           |
|                    /───----───\ <-- Platykurtic                         |
|                  /              \                                       |
|  ───────────────'────────────────'──────────────►                       |
|                                                                         |
+-------------------------------------------------------------------------+

• Mesokurtic (Normal Kurtosis, Excess Kurtosis $= 0$): Baseline bell curve. Standard outlier probability.
• Leptokurtic (Excess Kurtosis $> 0$): Characterized by a sharp, slender central peak and heavy, fat tails. Indicates a high probability of extreme outliers ("black swan" process shocks).
• Platykurtic (Excess Kurtosis $< 0$): Characterized by a broad, flat plateau and thin, light tails. Data is spread out more evenly across the range with fewer extreme outliers (approaching a uniform distribution).

Practical Quality Impact: Applying standard normal-theory statistical tools (such as standard $C_p / C_{pk}$ capability indices or standard Shewhart 3-sigma control limits) to heavily skewed or leptokurtic process data leads to severe calculation errors, generating false out-of-control alarms and inaccurate scrap predictions.