Designing a Reliable Capability Analysis in Minitab: Requirements, Checks, and When to Adjust

Requirements for a Valid Capability Analysis

Before running Minitab’s Stat > Quality Tools > Capability Analysis > Normal, the process must meet three core conditions:

• Statistical control – no special‑cause variation visible on a control chart.
• Approximate normality – the data distribution should not be heavily skewed or kurtotic.
• Sufficient sample size – at least ~30 observations to keep sampling error low for σ̂ estimates.

If any of these is violated, the Cp and Cpk indices can be misleading.

Smallest Suitable Design

The minimal analysis that satisfies the requirements consists of:

1. Load a data column containing the process measurements.
2. Specify the lower specification limit (LSL) and upper specification limit (USL).
3. Choose the normal capability method (no transformation).
4. Run the analysis and review the histogram, normal probability plot, and summary table.

This design uses only the built‑in capability routine; no extra modules or add‑ons are required.

Trust and Data Boundaries

Minitab treats the worksheet column as the sole input. The trust boundary lies at:

• Data entry – ensure values are numeric and represent the process of interest.
• Specification limits – LSL and USL must reflect engineering requirements, not arbitrary numbers.
• Assumptions – normality and stability are assumed by the tool; violating them moves the analysis outside its trusted domain.

All calculations (σ̂, Cp, Cpk, Pp, Ppk) are performed internally; the user cannot alter the formulas without using a different method (e.g., non‑normal capability).

Operational Checks

After the analysis runs, perform these verification steps:

1. Examine the histogram with normal overlay – look for gross deviations.
2. Check the normal probability plot – points should fall roughly along the straight line.
3. Compare the session‑window Cp and Cpk values to a quick manual check using Cp = (USL‑LSL)/(6·σ̂) and Cpk = min[(USL‑μ̂)/(3σ̂), (μ̂‑LSL)/(3σ̂)] where μ̂ and σ̂ are the sample mean and standard deviation (available via Stat > Basic Statistics > Display Descriptive Statistics).
4. If the histogram or probability plot shows systematic curvature, repeat the analysis with a Box‑Cox transformation (Stat > Control Charts > Box‑Cox Transformation) or switch to the non‑normal capability option.

These checks give confidence that the indices are not an artifact of assumption violation.

Failure Modes

Common ways the capability analysis can produce misleading results:

• Out‑of‑control points – special cause variation inflates or deflates σ̂, biasing Cp/Cpk.
• Non‑normal data – skewed distributions cause the normal method to overestimate capability; the histogram will show a clear tail.
• Too few observations – sampling error leads to wide confidence intervals; the indices may appear capable by chance.
• Incorrect specification limits** – using engineering targets instead of true limits yields meaningless indices.

When any of these occurs, the capability numbers should not be used for decision‑making until the underlying issue is addressed.

Conditions That Would Change the Design

The analysis design must be updated if:

• The process fails a control‑chart test – stabilize the process first, then re‑run capability.
• Normality tests (e.g., Anderson‑Darling) reject the normal assumption – apply a Box‑Cox or Johnson transformation, or select Stat > Quality Tools > Capability Analysis > Nonnormal.
• The available sample size is below 30 – collect more data or acknowledge increased uncertainty and consider confidence‑interval reporting.
• Specification limits change – simply edit the LSL/USL fields and re‑run; no methodological change is needed.

Concrete Example (Using Minitab’s Sample Data)

To see the workflow in action, use the built‑in worksheet Capability.MTW:

1. Open Minitab and choose File > Open Worksheet, then load Capability.MTW.
2. Select Stat > Quality Tools > Capability Analysis > Normal.
3. In the dialog, set Variable to the column Diameter (or any measurement column).
4. Enter LSL = 10 and USL = 20.
5. Click OK.
6. Minitab will display a histogram with a normal curve, a normal probability plot, and a session‑window table showing Cp, Cpk, Pp, Ppk.
7. For a quick sanity check, run Stat > Basic Statistics > Display Descriptive Statistics on the same column, note the mean (μ̂) and standard deviation (σ̂), and compute Cp and Cpk manually as described above.

If the histogram shows a noticeable skew, repeat the analysis after applying a Box‑Cox transformation (Stat > Control Charts > Box‑Cox Transformation) and observe how the capability indices shift. This demonstrates the importance of verifying assumptions before trusting the numbers.

Limitations and Practical Verification

The normal capability method assumes a stable, approximately normal process. When data are heavily tailed or multimodal, the method may overstate capability. A practical way to detect this is to:

• Run a normality test (Stat > Basic Statistics > Normality Test).
• If the p‑value is below the chosen alpha (e.g., 0.05), consider a transformation or the non‑normal capability option.

Additionally, capability indices are point estimates; reporting confidence intervals (available via the Options button in the capability dialog) gives a sense of uncertainty, especially for small samples.

Summary

A dependable capability analysis in Minitab starts with confirming process stability, normality, and adequate sample size. The smallest viable design uses the normal capability routine with user‑defined specification limits. Operational checks—histogram, probability plot, and a manual Cp/Cpk cross‑check—provide immediate feedback on assumption validity. When those checks fail, the design must shift to transformations, non‑normal methods, or additional data collection before the capability numbers can be trusted for engineering decisions.