4.2 Measures of Central Tendency and Dispersion

4.2 Measures of Central Tendency and Dispersion

Introduction to Summary Statistics in Quality Engineering

When inspecting machined features, test coupons, or assembly cycle times, quality technicians rarely present raw data tables containing dozens or hundreds of individual numbers to engineering or operations teams. Instead, raw data must be condensed into two complementary statistical dimensions:

1. Measures of Central Tendency (Location): Numbers describing the central point, typical value, or gravitational center around which the data cluster.
2. Measures of Dispersion (Spread or Variability): Numbers quantifying the extent to which individual measurements scatter, spread out, or deviate from that central point.

Reporting a measure of central tendency without a measure of dispersion is dangerous in manufacturing metrology. A CNC turning process producing pins with a mean diameter of exactly 25.000 mm might appear perfectly centered on blueprint nominal; however, if the process dispersion is excessively wide, half the pins may exceed the Upper Specification Limit while the other half fall below the Lower Specification Limit.

---

Measures of Central Tendency

1. The Arithmetic Mean

The arithmetic mean (commonly called the average) is the sum of all observed values divided by the total number of observations. It represents the physical balance point of the data distribution.

• Sample Mean ($\bar{x}$): $\bar{x} = \frac{\sum_{i=1}^n x_i}{n} = \frac{x_1 + x_2 + \dots + x_n}{n}$
• Population Mean ($\mu$ wash): $\mu = \frac{\sum_{i=1}^N x_i}{N}$

Key Characteristics:

• Mathematical Efficiency: Uses every single data value in its calculation, making it mathematically stable for symmetrical distributions.
• Extreme Sensitivity to Outliers: A single rogue measurement (e.g., an unseated workpiece or an operator transcription typo) pulls the mean strongly in the direction of the extreme value.

2. The Median

The median (denoted $\tilde{x}$ or $M$) is the physical middle value of an ordered dataset arranged in ascending or descending sequence. It divides the distribution into two equal halves: exactly 50% of the observations lie at or below the median, and 50% lie at or above it.

• Odd Sample Size ($n$): The median is the single middle value located at position $\frac{n + 1}{2}$.
• Even Sample Size ($n$): The median is the arithmetic average of the two central values located at positions $\frac{n}{2}$ and $\frac{n}{2} + 1$.

Key Characteristics:

• Robustness (Resistance): The median is largely unaffected by extreme outliers or heavy distribution tails. For this reason, environmental particle counts, customer complaint response times, and cycle times with long delays are frequently summarized using the median.

3. The Mode

The mode is the specific value or category that appears with the highest frequency in a dataset.

• Unimodal: A dataset with a single distinct peak.
• Bimodal: A dataset exhibiting two distinct frequency peaks. In manufacturing, a bimodal distribution is a critical diagnostic red flag, typically signaling that parts from two different machine spindles, different shifts, or two raw material lots have been blended together into a single tote.
• Multimodal: A dataset with three or more distinct peaks.
• No Mode: Occurs when all recorded values appear with equal frequency.

Key Characteristics:

• The mode is the only measure of central tendency that can be used for nominal data (e.g., determining the most common defect type on an assembly line: "Burr").

---

Distribution Skewness and Central Tendency Relationships

The relative positions of the mean, median, and mode reveal the symmetry or skewness of a manufacturing process distribution.

  Left-Skewed (Negative)         Symmetrical (Normal)          Right-Skewed (Positive)
      Long Left Tail                 Bell Curve                    Long Right Tail
           /\                            /\                             /\
          /  \                          /  \                           /  \
      ---/    \                     ---/    \---                      /    \---
   Mean < Median < Mode           Mean = Median = Mode           Mode < Median < Mean

1. Symmetrical (Normal) Distribution

In a perfectly balanced, unimodal, symmetrical distribution: $\text{Mean} = \text{Median} = \text{Mode}$ The center of gravity, the 50th percentile, and the highest point of frequency coincide exactly.

2. Right-Skewed (Positively Skewed) Distribution

A distribution with a long tail extending toward higher positive values. Extreme high values pull the arithmetic mean upward, while the median is pulled only moderately, and the mode remains at the cluster peak: $\text{Mode} < \text{Median} < \text{Mean}$ Common Manufacturing Examples: Surface roughness ($R_a$), total indicator reading (TIR) runout, hole perpendicularity, flatness deviations, cycle times, and defect counts per unit. These parameters are physically bounded by zero on the left but have unbounded tails on the right.

3. Left-Skewed (Negatively Skewed) Distribution

A distribution with a long tail extending toward lower values. Extreme low values pull the arithmetic mean downward: $\text{Mean} < \text{Median} < \text{Mode}$ Common Manufacturing Examples: Tensile strength of components subjected to early embrittlement, plating chemical purity percentages clustering near 100%, and tool life hours experiencing premature chipping.

[!TIP] Memory Trick for Skewness: The arithmetic mean is the most vulnerable metric and always follows the tail! If the tail stretches to the right (positive), the mean is pulled to the right, making Mean > Median. If the tail stretches to the left (negative), the mean is pulled to the left, making Mean < Median.

---

Measures of Dispersion (Variability)

Two production processes can share the exact same mean dimension of 50.00 mm while exhibiting drastically different quality performance. Dispersion metrics quantify this internal variation.

1. The Range ($R$)

The range is the simplest measure of dispersion, defined as the difference between the largest observed value ($X_{\max}$) and the smallest observed value ($X_{\min}$): $R = X_{\max} - X_{\min}$

• Advantages: Fast and intuitive to compute on the shop floor without complex calculators.
• Disadvantages: Extremely sensitive to outliers and ignores all intermediate data. Range loses statistical efficiency rapidly for sample sizes $n > 10$. In Statistical Process Control, range is standard for rational subgroups of size $n = 2$ to $5$ on $\bar{X}-R$ control charts.

2. Sample Variance ($s^2$) and Bessel's Correction

The sample variance is the sum of squared deviations of each observation from the sample mean, divided by the degrees of freedom ($n - 1$): $s^2 = \frac{\sum_{i=1}^n (x_i - \bar{x})^2}{n - 1}$ An algebraically equivalent computing formula that minimizes rounding error on calculators is: $s^2 = \frac{\sum x_i^2 - \frac{(\sum x_i)^2}{n}}{n - 1}$

Why Divide by $(n - 1)$? Bessel's Correction and Degrees of Freedom

One of the most frequently tested theoretical concepts on the ASQ CQT exam is Bessel's correction:

• When we calculate sample variance, we measure deviations from the sample mean $\bar{x}$, not from the true population mean $\mu$.
• Because sample points naturally cluster more closely around their own sample mean $\bar{x}$ than around the true population mean $\mu$, dividing by $n$ produces a sample variance that systematically underestimates the true population variance $\sigma^2$ (it is a biased estimator).
• Dividing by degrees of freedom ($n - 1$) mathematically compensates for this downward bias, making $s^2$ an unbiased estimator of $\sigma^2$.

3. Sample Standard Deviation ($s$)

The sample standard deviation is the positive square root of the sample variance: $s = \sqrt{s^2} = \sqrt{\frac{\sum_{i=1}^n (x_i - \bar{x})^2}{n - 1}}$

• Key Advantage: Standard deviation is expressed in the exact same engineering units as the original physical measurements (e.g., millimeters, pounds, or degrees Celsius), unlike variance which is expressed in squared units ($mm^2$).

4. Population Variance ($\sigma^2$) and Population Standard Deviation ($\sigma$)

When all $N$ units of an entire population are measured, the true population mean $\mu$ is known, and there is no estimation bias. Therefore, the divisor is the full population size $N$: $\sigma^2 = \frac{\sum_{i=1}^N (x_i - \mu)^2}{N}, \quad \sigma = \sqrt{\frac{\sum_{i=1}^N (x_i - \mu)^2}{N}}$

---

Relative Dispersion: The Coefficient of Variation ($CV$)

The standard deviation measures absolute dispersion. However, comparing absolute dispersion across processes with vastly different means or completely different units of measurement is misleading. The Coefficient of Variation ($CV$), also known as relative standard deviation (RSD), expresses the standard deviation as a percentage of the mean: $CV = \left(\frac{s}{\bar{x}}\right) \times 100\%$

Engineering Application of $CV$

Consider two CNC operations:

• Operation A (Micro-Pin Diameter): $\bar{x}_A = 2.000\text{ mm}$, $s_A = 0.050\text{ mm}$
• Operation B (Heavy Shaft Diameter): $\bar{x}_B = 200.000\text{ mm}$, $s_B = 0.500\text{ mm}$

Absolute comparison shows $s_B$ is 10 times larger than $s_A$. However, evaluating their relative dispersion: $CV_A = \left(\frac{0.050}{2.000}\right) \times 100\% = 2.50\%$ $CV_B = \left(\frac{0.500}{200.000}\right) \times 100\% = 0.25\%$ Operation B is ten times more consistent relative to its nominal dimension than Operation A. $CV$ is dimensionless and can also compare tensile strength in MPa against elongation in percent.

---

Standard Error of the Mean ($\sigma_{\bar{x}}$ or $s_{\bar{x}}$)

The Standard Error of the Mean quantifies the variability of sample averages across repeated samples of size $n$ drawn from the same population: $\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}} \quad \text{or} \quad s_{\bar{x}} = \frac{s}{\sqrt{n}}$ As sample size $n$ increases, the standard error shrinks by a factor of $\sqrt{n}$. This explains why averages of measurements exhibit far less scatter than individual measurements, forming the theoretical bedrock of $\bar{X}$ control charts.

---

Five-Number Summary, Box Plots, and Outlier Identification

Exploratory Data Analysis (EDA), developed by John Tukey, relies on the Five-Number Summary to describe data without assuming a normal distribution:

1. Minimum ($X_{\min}$): The smallest observed value.
2. First Quartile ($Q_1$): The 25th percentile (25% of data lies below $Q_1$).
3. Median ($Q_2$): The 50th percentile.
4. Third Quartile ($Q_3$): The 75th percentile (75% of data lies below $Q_3$).
5. Maximum ($X_{\max}$): The largest observed value.

The Interquartile Range ($IQR$)

The Interquartile Range ($IQR$) measures the spread of the central 50% of the distribution: $IQR = Q_3 - Q_1$ Like the median, the $IQR$ is highly resistant to extreme outliers.

Outlier Identification ($1.5 \times IQR$ Rule)

In a Box-and-Whisker plot, potential statistical outliers are identified using inner fences: $\text{Lower Inner Fence} = Q_1 - (1.5 \times IQR)$ $\text{Upper Inner Fence} = Q_3 + (1.5 \times IQR)$

• Any observed value falling below the lower fence or above the upper fence is flagged as an outlier.
• Whiskers on the box plot extend to the most extreme data points that still fall within the fences. Points outside the fences are plotted individually with asterisks or dots for engineering root-cause investigation.

---

Complete Step-by-Step Worked Metrology Calculation

Scenario

A quality technician measures the outer diameter of $n = 10$ precision dowel pins machined on a CNC lathe. The raw measurements (in millimeters) are: $25.04, \quad 25.01, \quad 25.08, \quad 25.03, \quad 25.05, \quad 25.02, \quad 25.04, \quad 25.06, \quad 25.04, \quad 25.13$

Step 1: Arrange the Data in Ascending Order

$25.01, \quad 25.02, \quad 25.03, \quad 25.04, \quad 25.04, \quad 25.04, \quad 25.05, \quad 25.06, \quad 25.08, \quad 25.13$

Step 2: Calculate Central Tendency Measures

• Sample Mean ($\bar{x}$): $\sum_{i=1}^{10} x_i = 250.50\text{ mm}$ $\bar{x} = \frac{250.50}{10} = 25.050\text{ mm}$
• Median ($Q_2$): Since $n = 10$ (even), average the 5th and 6th ordered values: $\text{Position 5} = 25.04, \quad \text{Position 6} = 25.04$ $\text{Median} = \frac{25.04 + 25.04}{2} = 25.040\text{ mm}$
• Mode: The value $25.04\text{ mm}$ appears three times (more than any other value). Thus, the distribution is unimodal with $\text{Mode} = 25.040\text{ mm}$.
• Skewness Check: Since $\text{Mode } (25.040) \le \text{Median } (25.040) < \text{Mean } (25.050)$, the distribution exhibits slight positive (right) skewness caused by the high measurement ($25.13\text{ mm}$).

Step 3: Calculate Dispersion Measures

• Range ($R$): $R = X_{\max} - X_{\min} = 25.13 - 25.01 = 0.120\text{ mm}$

• Deviations and Sum of Squared Deviations ($\sum(x_i - \bar{x})^2$):

• Sample Variance ($s^2$): $s^2 = \frac{\sum (x_i - \bar{x})^2}{n - 1} = \frac{0.010600}{10 - 1} = \frac{0.010600}{9} \approx 0.001178\text{ mm}^2$

• Sample Standard Deviation ($s$): $s = \sqrt{0.001178} \approx 0.03432\text{ mm}$

• Coefficient of Variation ($CV$): $CV = \left(\frac{0.03432}{25.050}\right) \times 100\% \approx 0.137\%$

• Standard Error of the Mean ($s_{\bar{x}}$): $s_{\bar{x}} = \frac{0.03432}{\sqrt{10}} = \frac{0.03432}{3.1623} \approx 0.01085\text{ mm}$

Step 4: Five-Number Summary and Outlier Analysis

• $X_{\min} = 25.01\text{ mm}$
• $Q_1$ (median of lower half: $25.01, 25.02, 25.03, 25.04, 25.04$) $= 25.03\text{ mm}$
• $\text{Median } (Q_2) = 25.04\text{ mm}$
• $Q_3$ (median of upper half: $25.04, 25.05, 25.06, 25.08, 25.13$) $= 25.06\text{ mm}$
• $X_{\max} = 25.13\text{ mm}$
• Interquartile Range ($IQR$): $IQR = Q_3 - Q_1 = 25.06 - 25.03 = 0.030\text{ mm}$
• Check for Outliers: $\text{Lower Fence} = Q_1 - 1.5 \times IQR = 25.03 - (1.5 \times 0.030) = 25.03 - 0.045 = 24.985\text{ mm}$ $\text{Upper Fence} = Q_3 + 1.5 \times IQR = 25.06 + (1.5 \times 0.030) = 25.06 + 0.045 = 25.105\text{ mm}$
• Outlier Conclusion: The maximum observation, $25.13\text{ mm}$, is greater than the upper fence ($25.105\text{ mm}$). It is an outlier that warrants checking for an unseated part, chip pack in the collet, or tool thermal expansion.