Using Minitab’s Assistant to Run a Johnson‑Transformed Capability Analysis on Non‑Normal Data

Why Non‑Normal Data Breaks Traditional Capability Metrics

Process engineers often rely on Cpk and Ppk to judge whether a part meets specification limits. These indices assume the underlying data are normally distributed. When the data are skewed or heavy‑tailed, the normality assumption fails and the capability metrics become misleading. The question is: how can we still use capability analysis when the data are non‑normal?

Enter Minitab’s Assistant – A Guided, Johnson‑Based Workflow

Minitab 16 introduced the Assistant menu, and from version 17 onward it includes a dedicated path for Non‑Normal Capability Analysis. The Assistant automatically detects non‑normality, recommends a Johnson transformation, and then calculates transformed capability indices with confidence intervals. This removes the guesswork of picking a distribution and ensures the user gets a statistically sound report.

Step‑by‑Step: Running a Johnson Transformation in the Assistant

1. Open the Assistant – Assistant ▸ Capability Analysis ▸ Non‑Normal ▸ Johnson Transformation.
2. Load your data – either by selecting a worksheet column or by importing a file. For this example we’ll use the built‑in sample Capability Nonnormal.MTW.
3. Choose the variable – the column that contains the quality characteristic.
4. Review the diagnostics – the Assistant shows a histogram of the raw data, the Anderson‑Darling (AD) test p‑value, and a fit‑to‑distribution plot. If the AD p‑value is 0.05 or lower, the data are considered non‑normal.
5. Accept the suggested Johnson family – the Assistant will propose one of the SB, SL, or SU families. If you have domain knowledge (e.g., lifetime data often follow a Weibull, which is a special case of the SL family), you can override the choice in the Distribution ID sub‑dialog.
6. Run the analysis – the Assistant outputs the transformation function (e.g., y = ln((x - a)/b)), the transformed data histogram, the fitted normal curve, and the capability indices Ppk and Cpk with 95% confidence intervals.
7. Compare raw vs. transformed results – the report includes side‑by‑side tables so you can see how the indices change after the transformation.

All of this happens in a single, interactive session. If you prefer automation, you can script the same steps using a Minitab macro or the mtbpy Python package, but the Assistant itself is not fully batchable.

Concrete Example: Measuring a Skewed Tensile Strength

Suppose you have a dataset of tensile strengths for a polymer that is heavily right‑skewed. The raw data histogram shows a long tail. You load the data into Minitab, run the Assistant, and receive the following key outputs:

• Raw data AD p‑value: 0.001 – confirms non‑normality.
• Suggested Johnson family: SL (log‑normal).
• Transformation function: y = ln((x - 0.5)/1.2).
• Transformed data AD p‑value: 0.42 – indicates the transformed data are now approximately normal.
• Transformed Ppk: 1.45 (95% CI: 1.30 – 1.60).
• Transformed Cpk: 1.40 (95% CI: 1.25 – 1.55).
• Raw Ppk: 0.92 (95% CI: 0.80 – 1.04).

The transformation turns a process that appears incapable (Ppk < 1.33) into a capable one once normality is achieved. The confidence intervals give you a sense of statistical uncertainty, which is crucial for decision making.

Trade‑Offs and Limitations

• Interpretability: Capability indices derived from transformed data are not directly comparable to those from raw normal data. Stakeholders must agree on reporting conventions and explain the transformation to non‑technical audiences.
• Sample Size: Johnson transformations are unreliable with fewer than 30 observations. The AD test may still reject normality after transformation, signalling that the process is inherently non‑normal.
• License Requirements: The Assistant and advanced capability features are part of the ‘Quality Companion’ or ‘Statistical’ modules in the subscription licensing model. Users on the free or basic license cannot access this workflow.
• Univariate Only: The Assistant does not handle multivariate capability analysis. If you have correlated characteristics, you must revert to the standard Capability Sixpack or write custom macros.

How to Check Your Results

After running the Assistant, verify the following:

• The transformed data histogram should resemble a normal bell curve.
• The AD test p‑value for the transformed data should be > 0.05.
• The confidence intervals for Ppk and Cpk should be reasonable (i.e., not extremely wide).
• Cross‑check the transformation function by applying it manually to a few data points and ensuring the output matches the Assistant’s transformed values.

If any of these checks fail, reconsider the transformation family or collect more data.

Actionable Take‑Away

1. Use the Assistant for quick, guided capability analysis on non‑normal data. It handles detection, transformation, and reporting in one place.
2. Document the transformation function and confidence intervals in your quality reports. This ensures transparency and facilitates stakeholder understanding.
3. Validate the results with a manual check or a macro if you need batch processing. The Assistant can serve as a prototype, but automation requires scripting.
4. Remember the license and sample‑size caveats. If you’re on a limited license or have <30 data points, consider alternative statistical approaches (e.g., non‑parametric capability metrics).

By following this workflow, process engineers can confidently assess specification conformance even when the data defy normality assumptions, turning a potential roadblock into a data‑driven decision.