5.4 X-bar & s Charts and Individual-Moving Range (I-MR) Charts

5.4 X-bar & s Charts and Individual-Moving Range (I-MR) Charts

While the classic $\bar{X}$ and $R$ chart is ideal for standard manufacturing operations with small subgroup sizes ($n = 4$ or $5$), quality technicians frequently encounter operational environments where the range chart is either statistically inefficient or physically impossible to deploy.

When automated coordinate measuring machines (CMMs) collect large subgroups of data ($n \ge 10$), or when parts arrive in variable lot sizes, the sample range discards valuable information, mandating the $\bar{X}$ and $s$ control chart. Conversely, when parts take 18 hours to machine, tests destroy expensive aerospace components, or a single 5,000-gallon chemical tank is mixed, subgrouping multiple parts is meaningless, mandating the Individual and Moving Range ($I-MR$) chart.

Mastering these two advanced variables charting techniques is essential for the ASQ CQT examination and modern quality engineering practice.

---

1. $\bar{X}$ and $s$ Control Charts: When and Why to Replace the Range

The Statistical Inefficiency of Range for Large Samples

The sample range ($R = X_{\max} - X_{\min}$) is computationally simple, which made it the standard in the pre-computer era of 1930s shop-floor math. However, the range suffers from a fatal theoretical flaw: it utilizes only two data points (the highest and the lowest), completely ignoring all intermediate observations.

As subgroup sample size ($n$) increases, the statistical efficiency of $R$ relative to the sample standard deviation ($s$) drops precipitously:

• At $n = 2$, relative efficiency is 100% ($R$ and $s$ contain identical information).
• At $n = 5$, relative efficiency is 95.5% (a tiny, acceptable loss of efficiency).
• At $n = 10$, relative efficiency drops to 85.0%.
• At $n = 15$, relative efficiency plummets to 75.5%.
• At $n = 25$, relative efficiency is under 65%.

In samples of $n \ge 10$, an extreme outlier artificially inflates the range, while significant clustering or shift among the remaining 8 or 9 parts goes completely undetected. The sample standard deviation ($s$) evaluates the distance of every single data point from the subgroup mean, making it a far more robust and efficient measure of dispersion.

When to Select an $\bar{X}$ and $s$ Chart

1. Subgroup Sample Size $n \ge 10$: Industry standards universally mandate switching from $R$ to $s$ when $n \ge 10$ (many automotive and aerospace standards mandate $s$ at $n \ge 9$).
2. Variable Subgroup Sizes ($n_i \ne n$): When subgroups contain different numbers of parts (e.g., due to missing parts, tool breakdowns, or varying production lot runs), range charts cannot be mathematically standardized without complex adjustments. The $s$ chart accommodates varying sample sizes seamlessly.
3. Automated and Computerized Data Systems: Modern digital calipers, inline optical probes, and CMMs compute standard deviation instantaneously, eliminating any computational barrier.

---

2. Mathematical Formulas and Factors for $\bar{X}$ and $s$ Charts

For a study of $k$ subgroups with sample size $n$:

Subgroup Statistics

For each subgroup $i$ ($i = 1, 2, \dots, k$):

• Subgroup Average ($\bar{X}_i$): $$\bar{X}_i = \frac{\sum_{j=1}^n X_{ij}}{n}$$
• Subgroup Sample Standard Deviation ($s_i$): $$s_i = \sqrt{\frac{\sum_{j=1}^n (X_{ij} - \bar{X}_i)^2}{n - 1}}$$

Centerlines

• Grand Mean ($\bar{\bar{X}}$): $$\bar{\bar{X}} = \frac{1}{k} \sum_{i=1}^k \bar{X}_i$$
• Average Sample Standard Deviation ($\bar{s}$): $$\bar{s} = \frac{1}{k} \sum_{i=1}^k s_i$$

Dispersion ($s$) Chart Control Limits

As with $\bar{X}-R$ charts, process dispersion must be evaluated first using factors $B_3$ and $B_4$:

$$UCL_s = B_4 \bar{s}$$ $$CL_s = \bar{s}$$ $$LCL_s = B_3 \bar{s}$$

(Note: $B_3 = 0$ for $n \le 5$. A positive lower limit exists for $n \ge 6$.)

Location ($\bar{X}$) Chart Control Limits

Control limits for process location use factor $A_3$:

$$UCL_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{s}$$ $$CL_{\bar{X}} = \bar{\bar{X}}$$ $$LCL_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{s}$$

(Derivation: $A_3 = \frac{3}{c_4 \sqrt{n}}$)

Estimating Population Standard Deviation ($\hat{\sigma}$)

The sample standard deviation $s$ is a slightly biased estimator of the true population parameter $\sigma$ for small samples. The unbiasing constant $c_4$ is used to obtain the true within-subgroup standard deviation:

$$\hat{\sigma} = \frac{\bar{s}}{c_4}$$

Standard Factor Table for $\bar{X}$ and $s$ Charts

Handling Variable Subgroup Sizes

When sample sizes vary from subgroup to subgroup ($n_i \ne n$):

1. Average Sample Size Method: If sample sizes vary by no more than $\pm 10\%$ to $15\%$ (e.g., $n$ varies between 9 and 11), technicians can calculate an average sample size $\bar{n} = \sum n_i / k$ and use fixed limits based on $\bar{n}$.
2. Individual Control Limit Method (Stair-Step Limits): If sample sizes vary widely (e.g., some subgroups have $n=5$, others $n=15$), separate control limits must be computed for each individual subgroup using that subgroup's specific $n_i$ and corresponding factors $A_3(n_i)$, $B_3(n_i)$, and $B_4(n_i)$. On the control chart, the control limits expand when $n_i$ is small and contract when $n_i$ is large, creating a characteristic 'stair-step' appearance.

---

3. Individual and Moving Range ($I-MR$ / $X-MR$) Charts

In many production environments, collecting rational subgroups of $n \ge 2$ is physically impossible, economically unfeasible, or statistically meaningless. In these situations, the subgroup sample size is exactly one ($n = 1$), and the process must be monitored using an Individual and Moving Range ($I-MR$) chart (also known as an $X-MR$ chart).

              INDIVIDUALS (X) CHART
  UCL_X  --------------------------------- X-bar + 2.660 * MR-bar
             *       *           *
  CL     --*-------*-------*---*-------*-- X-bar
                 *           *       *
  LCL_X  --------------------------------- X-bar - 2.660 * MR-bar

              MOVING RANGE (MR) CHART
  UCL_MR --------------------------------- 3.267 * MR-bar
                 *           *
  CL     --*-------*---*---*-------*------ MR-bar
             *           *       *
  LCL_MR --------------------------------- 0

Industrial Applications Mandating $n = 1$

Quality technicians must deploy an $I-MR$ chart under four primary conditions:

1. Destructive Testing: Measuring missile casing burst pressure, tensile fracture strength of aerospace fasteners, or impact resistance. Because each tested part is destroyed at high cost, pulling multiple pieces per subgroup is economically prohibitive.
2. Very Slow Production Rates: Machining massive turbine rotors, locomotive frames, or aerospace wing spars where only one part is produced every 8 to 24 hours. Subgrouping $n=5$ parts would span an entire week of production, violating rational subgrouping by mixing tool wear, ambient temperature swings, and multiple operator shifts into within-subgroup variation.
3. Homogeneous Fluids and Chemical Batches: In chemical vats, liquid pharmaceutical mixing tanks, molten alloy furnaces, or electroplating baths, the fluid is completely homogeneous. Taking five consecutive beakers from the same tank does not measure process variation—it only measures measurement error! Each individual batch represents $n=1$.
4. Automated 100% Inspection: Inline laser gaging, optical inspection, or automated electronic testing measuring every consecutive part as it flows down a transfer line.

---

4. Mathematical Formulas and Construction for $I-MR$ Charts

An $I-MR$ chart consists of two paired charts:

1. The Individuals ($X$) Chart: Monitors process central tendency across single readings.
2. The Moving Range ($MR$) Chart: Monitors process dispersion by calculating the absolute difference between successive individual readings.

Step 1: Compute Moving Ranges

For a series of $k$ individual observations ($X_1, X_2, \dots, X_k$), the moving range between consecutive observations using a standard span of $w = 2$ is:

$$MR_i = |X_i - X_{i-1}| \quad \text{for } i = 2, 3, \dots, k$$

(Note: For $k$ individual observations, there are exactly $k - 1$ moving ranges.)

Step 2: Compute Centerlines

• Centerline for Individuals Chart ($\bar{X}$): $$\bar{X} = \frac{1}{k} \sum_{i=1}^k X_i$$
• Centerline for Moving Range Chart ($\bar{MR}$): $$\bar{MR} = \frac{1}{k - 1} \sum_{i=2}^k MR_i$$

Step 3: Compute Moving Range ($MR$) Control Limits

Because the moving range is calculated between successive pairs of parts, the effective sample size is $n = 2$. Using standard Shewhart factors for $n = 2$ ($D_4 = 3.267$ and $D_3 = 0$):

$$UCL_{MR} = D_4 \bar{MR} = 3.267 \bar{MR}$$ $$CL_{MR} = \bar{MR}$$ $$LCL_{MR} = D_3 \bar{MR} = 0$$

Step 4: Compute Individuals ($X$) Control Limits

The population standard deviation is estimated from the average moving range using the $n=2$ factor $d_2 = 1.128$:

$$\hat{\sigma} = \frac{\bar{MR}}{d_2} = \frac{\bar{MR}}{1.128}$$

The $3\sigma$ control limits for individual observations are therefore:

$$UCL_X = \bar{X} + 3\hat{\sigma} = \bar{X} + 3\left(\frac{\bar{MR}}{1.128}\right) = \bar{X} + \left(\frac{3}{1.128}\right)\bar{MR} = \bar{X} + 2.660 \bar{MR}$$ $$LCL_X = \bar{X} - 3\hat{\sigma} = \bar{X} - 2.660 \bar{MR}$$ $$\mathbf{UCL_X = \bar{X} + E_2 \bar{MR}}$$ $$\mathbf{LCL_X = \bar{X} - E_2 \bar{MR}}$$

(Where factor $E_2 = 2.660$ for a span of $w = 2$.)

---

5. Critical Statistical Limitations of $I-MR$ Charts

While extraordinarily versatile, $I-MR$ charts possess three severe statistical vulnerabilities that quality technicians must understand:

A. Extreme Sensitivity to Non-Normality

On an $\bar{X}$ chart, the Central Limit Theorem guarantees that subgroup averages are approximately normally distributed even if individual parts are skewed. The Central Limit Theorem does not apply to the Individuals chart. The $X$ chart plots raw, un-averaged individual measurements.

If the underlying process distribution is skewed (e.g., flatness, roundness, surface roughness, hole true position, or chemical cycle times), plotting raw individual data against symmetric $\pm 3\sigma$ limits will generate massive false alarm rates on the skewed tail and zero sensitivity on the short tail. Individual data must be tested for normality (e.g., via Anderson-Darling test or probability plots) before deploying an $I-MR$ chart. If non-normal, data must undergo a Box-Cox transformation or be fitted to a Weibull/lognormal model.

B. Mathematically Induced Autocorrelation on the $MR$ Chart

In standard rational subgrouping, each subgroup range is completely independent. On an $I-MR$ chart, consecutive moving ranges are mathematically correlated because each individual observation $X_i$ is used to calculate two successive moving ranges:

$$MR_i = |X_i - X_{i-1}| \quad \text{and} \quad MR_{i+1} = |X_{i+1} - X_i|$$

If observation $X_i$ is an extreme outlier, it will cause two consecutive points on the $MR$ chart to spike upwards! Inexperienced technicians often misinterpret this paired spike as two independent out-of-control events. Technicians must recognize that a single bad part inevitably induces a paired alarm on the moving range chart.

C. Reduced Power to Detect Small Shifts

Because $n = 1$, the standard error of the estimate is large ($\sigma / \sqrt{1} = \sigma$). An $I-MR$ chart requires significantly more observations to detect a moderate process shift ($1.0\sigma$ to $1.5\sigma$) than an $\bar{X}$ chart of $n=5$. When rapid detection of small shifts is essential, technicians should supplement the $I-MR$ chart with a CUSUM (Cumulative Sum) or EWMA (Exponentially Weighted Moving Average) control chart.

---

6. Comprehensive Worked Numerical Examples

Worked Example 1: $\bar{X}$ and $s$ Control Chart Calculations

Scenario: An automated CMM measures the bore diameter of 20 consecutive automotive engine blocks per shift. The quality technician establishes an $\bar{X}$ and $s$ chart with subgroup size $n = 10$ across $k = 25$ baseline subgroups.

• Grand Mean: $\bar{\bar{X}} = 85.025\text{ mm}$
• Average Sample Standard Deviation: $\bar{s} = 0.0120\text{ mm}$

Step 1: Retrieve factors for $n = 10$ From the factor table: $c_4 = 0.9727, \quad A_3 = 0.975, \quad B_3 = 0.284, \quad B_4 = 1.716$.

Step 2: Calculate dispersion ($s$) chart limits

$$UCL_s = B_4 \bar{s} = 1.716 \times 0.0120 = 0.02059 \approx 0.0206\text{ mm}$$ $$CL_s = \bar{s} = 0.0120\text{ mm}$$ $$LCL_s = B_3 \bar{s} = 0.284 \times 0.0120 = 0.00341 \approx 0.0034\text{ mm}$$

Step 3: Calculate location ($\bar{X}$) chart limits

$$\text{3-Sigma Margin} = A_3 \bar{s} = 0.975 \times 0.0120 = 0.0117\text{ mm}$$ $$UCL_{\bar{X}} = \bar{\bar{X}} + A_3 \bar{s} = 85.025 + 0.0117 = 85.0367\text{ mm}$$ $$CL_{\bar{X}} = 85.0250\text{ mm}$$ $$LCL_{\bar{X}} = \bar{\bar{X}} - A_3 \bar{s} = 85.025 - 0.0117 = 85.0133\text{ mm}$$

Step 4: Estimate true population standard deviation ($\hat{\sigma}$)

$$\hat{\sigma} = \frac{\bar{s}}{c_4} = \frac{0.0120}{0.9727} = 0.01234\text{ mm}$$

---

Worked Example 2: Individual and Moving Range ($I-MR$) Chart Calculations

Scenario: A chemical processing facility manufactures specialty polymer resin in 2,000-pound batches. Because each batch is a homogeneous liquid, the technician tracks batch Melt Flow Index (MFI) using an $I-MR$ chart ($n = 1$). Baseline data from $k = 15$ consecutive batches yields:

• Mean Melt Flow Index: $\bar{X} = 22.40\text{ g/10 min}$
• Sum of 14 moving ranges: $\sum_{i=2}^{15} MR_i = 21.70\text{ g/10 min}$

Step 1: Calculate Average Moving Range ($\bar{MR}$) Remember: For 15 individual batches, there are $15 - 1 = 14$ moving ranges!

$$\bar{MR} = \frac{\sum MR_i}{k - 1} = \frac{21.70}{14} = 1.550\text{ g/10 min}$$

Step 2: Calculate Moving Range ($MR$) Chart Limits Using $D_4 = 3.267$ and $D_3 = 0$:

$$UCL_{MR} = D_4 \bar{MR} = 3.267 \times 1.550 = 5.064\text{ g/10 min}$$ $$CL_{MR} = \bar{MR} = 1.550\text{ g/10 min}$$ $$LCL_{MR} = 0$$

Step 3: Calculate Individuals ($X$) Chart Limits Using factor $E_2 = 2.660$:

$$\text{3-Sigma Margin} = E_2 \bar{MR} = 2.660 \times 1.550 = 4.123\text{ g/10 min}$$ $$UCL_X = \bar{X} + E_2 \bar{MR} = 22.40 + 4.123 = 26.523\text{ g/10 min}$$ $$CL_X = \bar{X} = 22.40\text{ g/10 min}$$ $$LCL_X = \bar{X} - E_2 \bar{MR} = 22.40 - 4.123 = 18.277\text{ g/10 min}$$

Step 4: Estimate Process Standard Deviation ($\hat{\sigma}$)

$$\hat{\sigma} = \frac{\bar{MR}}{d_2} = \frac{1.550}{1.128} = 1.374\text{ g/10 min}$$

---

7. Summary Comparison: Variables Control Chart Selection Guide

| Control Chart | Data Type | Subgroup Size ($n$) | Dispersion Measure | Factors Used | Primary Applications & Advantages | |:---|:---|:---:|:---:|:---:|:---|| | $\bar{X}$ and $R$ | Continuous Variables | $2 \le n \le 9$ (Typically 4–5) | Range ($R$) | $A_2, D_3, D_4, d_2$ | Standard shop-floor machining, stamping; computationally simple; high efficiency for small $n$. | | $\bar{X}$ and $s$ | Continuous Variables | $n \ge 10$ or variable $n$ | Sample Standard Deviation ($s$) | $A_3, B_3, B_4, c_4$ | High-speed CMM inspection, automated gages; statistically superior for large or varying subgroups. | | $I-MR$ ($X-MR$) | Continuous Variables | $n = 1$ | Moving Range ($MR$) of span 2 | $E_2 = 2.660$, $D_4 = 3.267, d_2 = 1.128$ | Destructive testing, homogeneous chemical vats, slow cycle times, automated 100% sorting lines. |