8.1 Measures of Central Tendency & Dispersion

8.1 Measures of Central Tendency & Dispersion

Core Principle: Descriptive statistics summarize the location and dispersion of process data without drawing broader inferential conclusions. Central tendency measures (mean, median, mode) pinpoint the center of a distribution, while dispersion measures (range, variance, standard deviation, IQR) quantify process spread and variation. Six Sigma practitioners must select metrics based on distribution shape: the mean and standard deviation for symmetrical, bell-shaped data, and the median and interquartile range (IQR) for skewed or outlier-heavy distributions.

---

The Role of Descriptive Statistics in Six Sigma

In the Measure and Analyze phases of DMAIC, continuous improvement teams collect large volumes of baseline data representing process cycle times, product dimensions, transaction defect counts, or chemical concentrations. Raw data in its unaggregated state is cognitively overwhelming and cannot inform decision-making.

Descriptive statistics provides the mathematical toolkit to condense numerical datasets into concise summary values that describe two fundamental dimensions of process performance:

1. Location (Central Tendency): Where is the process centered? What is the typical or expected value?
2. Spread (Dispersion): How much variation exists around that center? Are outputs tightly clustered or widely scattered?

Understanding these two properties allows Green Belts to establish baseline process capability, evaluate alignment against customer specifications (Voice of the Customer), and assess whether observed differences before and after improvements represent true systemic shifts or random noise.

---

Population Parameters vs. Sample Statistics

A cornerstone concept tested on the CSSC Green Belt examination is the distinction between a population and a sample, along with their corresponding mathematical notations.

• Population: The complete, exhaustive collection of all items, units, transactions, or observations of interest in a given study (e.g., every invoice processed this fiscal year, or every machined shaft produced across the lifetime of a production run). A numerical descriptive measurement calculated from an entire population is called a parameter.
• Sample: A representative subset or fraction of observations drawn from the population (e.g., 50 randomly selected invoices, or 30 consecutive shafts measured during a shift). A numerical descriptive measurement calculated from sample data is called a statistic.

In industrial Six Sigma projects, measuring the entire population is virtually impossible due to prohibitive economic costs, destructive testing constraints (such as tensile strength testing), or infinite process streams. Consequently, practitioners calculate sample statistics to estimate unknown population parameters.

[!WARNING] Exam Warning: On CSSC exams, mixing Greek and Roman notation is an immediate failure point in quantitative questions. When a question provides sample data gathered from a test subgroup, you must compute the sample standard deviation $s$ with degrees of freedom $n - 1$, rather than the population standard deviation $\sigma$.

---

Measures of Central Tendency

Central tendency captures the single value that best typifies or summarizes the central location of a dataset. The three primary metrics are the mean, median, and mode.

                     Measures of Central Tendency
                                  │
         ┌────────────────────────┼────────────────────────┐
         ▼                        ▼                        ▼
     Arithmetic                Median                     Mode
        Mean               (50th Percentile)       (Peak Frequency)
  Mathematical Center       Positional Center        Nominal Center
 Highly Outlier-Sensitive    Robust to Outliers     Identifies Multimodality

1. The Arithmetic Mean

The mean is the sum of all individual data values divided by the total number of observations:

$\mu = \frac{\sum_{i=1}^N X_i}{N} \quad \text{(Population)} \qquad \bar{x} = \frac{\sum_{i=1}^n x_i}{n} \quad \text{(Sample)}$

• Mathematical Property: The arithmetic mean serves as the physical "center of gravity" or balance point of a dataset. The sum of signed deviations of all observations from their arithmetic mean is mathematically identically zero: $\sum_{i=1}^n (x_i - \bar{x}) = 0$
• Vulnerability: The mean incorporates every numerical value into its calculation. Therefore, a single extreme outlier pulls the mean substantially toward the tail, distorting the perceived center of the process.

2. The Median

The median (denoted as $\tilde{x}$ or $M$) is the physical middle value when observations are sorted in ascending numerical order. It divides the distribution into two equal halves, representing the 50th percentile:

• Odd Sample Size ($n$ is odd): The median is the unique middle observation located at rank position $\frac{n + 1}{2}$.
• Even Sample Size ($n$ is even): The median is the arithmetic average of the two central observations located at rank positions $\frac{n}{2}$ and $\frac{n}{2} + 1$.
• Robustness: The median is resistant (robust) to extreme outliers and heavy skewness because it depends solely on rank order, not numerical magnitudes. If a hospital emergency department records five patient wait times of 15, 20, 22, 25, and 360 minutes, the median is 22 minutes (accurately reflecting typical experience), whereas the mean is inflated to 88.4 minutes.

3. The Mode

The mode is the most frequently occurring value in a dataset. Unlike the mean and median, the mode can be identified for both quantitative (numerical) and qualitative (attribute/categorical) data:

• Unimodal: Exactly one value appears with peak frequency.
• Bimodal: Two non-adjacent values share peak frequency.
• Multimodal: Three or more values share peak frequency.
• No Mode: All observations occur with equal frequency.

[!NOTE] DMAIC Diagnostic Value: In Six Sigma root-cause analysis, encountering a bimodal distribution in process cycle times or product dimensions is a critical red flag. It indicates that the dataset is not representing a single homogeneous process, but rather two distinct sub-processes combined (e.g., two different machine operators, two distinct raw material suppliers, day shift vs. night shift, or two parallel manufacturing cavities).

Comparative Selection Guide: Choosing the Right Measure of Center

---

Measures of Dispersion (Process Variation)

Central tendency describes where a process sits, but Six Sigma is fundamentally about reducing variation. Two processes can possess identical mean cycle times of 30 minutes, yet Process A ranges from 28 to 32 minutes while Process B ranges from 5 to 55 minutes. Process B will generate severe customer dissatisfaction and high defect rates. Measures of dispersion quantify this spread.

1. Range ($R$)

The range is the simplest measure of dispersion, representing the arithmetic distance between the maximum and minimum observed values:

$R = X_{\text{max}} - X_{\text{min}}$

• Strengths: Intuitive, instant to compute. Heavily utilized in traditional Shewhart control charts (such as $\bar{X}-R$ charts) for small subgroups ($n \le 8$).
• Weaknesses: Completely ignores all intermediate data values and relies exclusively on the two most extreme observations. Range systematically increases as sample size $n$ grows, making it unreliable for comparing datasets of unequal size.

2. Variance ($\sigma^2$ and $s^2$)

Variance measures the average squared deviation of individual data points from their arithmetic mean. Squaring deviations ensures that positive and negative differences do not cancel each other out and penalizes larger deviations more heavily.

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

$s^2 = \frac{\sum_{i=1}^n (x_i - \bar{x})^2}{n - 1} \quad \text{(Sample Variance with Bessel's Correction)}$

Bessel's Correction and Degrees of Freedom ($n - 1$)

Notice the crucial denominator difference between population variance ($N$) and sample variance ($n - 1$):

• When a sample is drawn, data points tend to cluster closer to their own sample mean $\bar{x}$ than to the unknown true population mean $\mu$.
• Dividing the sample sum of squares by $n$ creates a downward-biased estimator that systematically underestimates the true population variance.
• Dividing by $n - 1$ (known as Bessel's correction) mathematically corrects this bias, rendering $s^2$ an unbiased estimator of $\sigma^2$.
• The term $n - 1$ represents the degrees of freedom ($df$). Once the sample mean $\bar{x}$ is calculated, only $n - 1$ deviations are free to vary; the $n$-th deviation is mathematically fixed because $\sum (x_i - \bar{x}) = 0$.

3. Standard Deviation ($\sigma$ and $s$)

While variance is mathematically essential for statistical modeling (e.g., ANOVA), its unit of measurement is squared (e.g., $\text{minutes}^2$, $\text{mm}^2$, or $\text{dollars}^2$), making it difficult to interpret practically. The standard deviation resolves this by taking the positive square root of the variance, returning dispersion to the original measurement units:

$\sigma = \sqrt{\sigma^2} = \sqrt{\frac{\sum_{i=1}^N (X_i - \mu)^2}{N}} \quad \text{(Population Standard Deviation)}$

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

In Six Sigma methodology, the standard deviation is the universal benchmark metric for evaluating process variation, calculating capability indices ($C_p, C_{pk}$), and establishing statistical process control limits.

4. Interquartile Range (IQR)

The Interquartile Range (IQR) measures the spread of the middle 50% of ordered observations. It is computed as the difference between the third quartile ($Q_3$, 75th percentile) and the first quartile ($Q_1$, 25th percentile):

$IQR = Q_3 - Q_1$

• Robustness: Because the IQR discards the lowest 25% and highest 25% of data, it is impervious to extreme outliers and heavy tails. It serves as the dispersion counterpart to the median.

---

Box-and-Whisker Plots & Five-Number Summary

Developed by statistician John Tukey, the Box-and-Whisker Plot (boxplot) provides an exploratory data visualization of process location, dispersion, skewness, and outliers based on the Five-Number Summary:

1. Minimum: The smallest observed value within the non-outlier fence boundaries.
2. First Quartile ($Q_1$): 25th percentile (splits lower 25% from upper 75%).
3. Median ($Q_2$): 50th percentile (central process line).
4. Third Quartile ($Q_3$): 75th percentile (splits lower 75% from upper 25%).
5. Maximum: The largest observed value within the non-outlier fence boundaries.

       Lower Outlier                                               Upper Outlier
          *                                                           *   *
   ───────┼─────────────┬──────────────┬──────────────┬───────────────┼──────────
         Fence         Min            Q1            Median           Q3         Fence
     [Q1 - 1.5*IQR]                    ┌──────────────┬──────────────┐      [Q3 + 1.5*IQR]
                       ├───────────────┤              │              ├──────┤
                                       └──────────────┴──────────────┘
                                       ◄──────────── IQR ────────────►
                                       (Middle 50% of Process Data)

Tukey's 1.5 * IQR Rule for Outlier Detection

Boxplots define objective mathematical boundaries (called "inner fences") beyond which data points are categorized as suspected outliers:

• Lower Inner Fence: $\text{LIF} = Q_1 - 1.5 \times IQR$
• Upper Inner Fence: $\text{UIF} = Q_3 + 1.5 \times IQR$

Any data point falling below the Lower Inner Fence ($x_i < LIF$) or above the Upper Inner Fence ($x_i > UIF$) is plotted as an individual asterisk or circle and flagged for investigation. In Six Sigma, outliers are never arbitrarily erased; they must be investigated to determine whether they represent measurement error, data entry mistakes, or genuine special-cause process disturbances.

---

Step-by-Step Worked Example: Calculating Dispersion & Outlier Fences

Scenario: A Green Belt monitors a CNC grinding operation producing precision stainless-steel pins. A random sample of $n = 6$ pins yields the following diameter deviations from nominal (in microns): $\{12, 14, 15, 16, 18, 21\}$

Step 1: Calculate the Sample Mean ($\bar{x}$)

$\bar{x} = \frac{\sum x_i}{n} = \frac{12 + 14 + 15 + 16 + 18 + 21}{6} = \frac{96}{6} = 16.0 \text{ microns}$

Step 2: Compute Deviations and Sum of Squared Deviations ($SS$)

Notice that the sum of signed deviations equals exactly zero, verifying the mathematical definition of the arithmetic mean.

Step 3: Compute Sample Variance ($s^2$)

Divide the sum of squared deviations by degrees of freedom $n - 1 = 6 - 1 = 5$: $s^2 = \frac{50}{6 - 1} = \frac{50}{5} = 10.0 \text{ microns}^2$

(Note: If population variance had been mistakenly calculated by dividing by 6, the result would be $50 / 6 = 8.333\text{ microns}^2$, underestimating true variance by 16.7%).

Step 4: Compute Sample Standard Deviation ($s$)

$s = \sqrt{s^2} = \sqrt{10.0} \approx 3.162 \text{ microns}$

Step 5: Compute Five-Number Summary and IQR Outlier Fences

• Ordered dataset: ${12, 14, 15, 16, 18, 21}$
• $\text{Minimum} = 12$
• Lower half: ${12, 14, 15} \rightarrow Q_1 = 14$
• $\text{Median} = \frac{15 + 16}{2} = 15.5$
• Upper half: ${16, 18, 21} \rightarrow Q_3 = 18$
• $\text{Maximum} = 21$
• $IQR = Q_3 - Q_1 = 18 - 14 = 4.0 \text{ microns}$

Evaluate outlier fences: $\text{LIF} = Q_1 - 1.5 \times IQR = 14 - 1.5(4.0) = 14 - 6.0 = 8.0 \text{ microns}$ $\text{UIF} = Q_3 + 1.5 \times IQR = 18 + 1.5(4.0) = 18 + 6.0 = 24.0 \text{ microns}$

All six observed values ($12 \dots 21$) fall safely within $[8.0, 24.0]$. If a pin measured $26.0\text{ microns}$, it would exceed the UIF of $24.0$ and be flagged as an outlier.

---

Critical CSSC Exam Traps

• Trap 1: Omitting Bessel's Correction ($n$ vs. $n - 1$) — When given a sample dataset on the exam, candidates frequently divide by $n$ instead of $n - 1$. Dividing by $n$ calculates population variance $\sigma^2$, which yields an artificially small variance and causes capability indices to appear falsely inflated.
• Trap 2: Using the Mean for Skewed Operational Metrics — Exam scenarios presenting heavily skewed data (such as invoice processing turnaround times or IT incident resolution hours) will ask for the most representative metric of process performance. Choosing the mean is incorrect; the median and IQR must be chosen when data is skewed.
• Trap 3: Averaging Bimodal Processes — If a process exhibits two distinct peaks (e.g., modes at 10 psi and 30 psi), computing an overall mean of 20 psi is mathematically valid but operationally catastrophic. The process never actually produces parts at 20 psi; it produces two distinct streams that must be segregated.
• Trap 4: Confusing Range with IQR — Range spans 100% of observations from minimum to maximum and is highly sensitive to extreme points. IQR spans the middle 50% of observations ($Q_3 - Q_1$) and is entirely resistant to tail extremes.