Minitab Guide for Lean Six Sigma

Where Minitab Fits in a Project

Minitab is the statistical package most widely used in quality and Six Sigma work. It was built for that job: the menus follow the DMAIC toolkit, the output is in the language of the Body of Knowledge, and every analysis comes with the graphs and checks a Black Belt expects. This guide is a task-oriented companion. It shows where each tool lives, how to set up the data, how to read the results, and the mistakes that cost the most time.

Minitab is a registered trademark of Minitab, LLC. This guide is independent and is not affiliated with or endorsed by Minitab. Menu names follow recent releases of Minitab Statistical Software for the desktop; older releases and the web edition can differ in small ways, so check each path against your version.

Know the Workspace

Minitab keeps everything for a piece of work in one project: the data, the text results, the graphs, and a history of what you did. Knowing the parts makes the rest much easier.

Files. Save the whole project to keep data, output, and graphs together (recent releases use .mpx; older ones use .mpj). You can also save a worksheet on its own (.mwx in recent releases, .mtw in older ones), and Minitab can export a worksheet to Excel or a text file. Keep the original data in its own file and never overwrite it with edited data.

The dialog pattern. Almost every analysis follows the same steps: choose the columns, click Options to set the confidence level, hypothesized value, or assumptions, click Graphs to choose the plots (always ask for the residual plots or the probability plot), click Results or Storage for extra output, then click OK. If the result is not what you expected, press Ctrl+E and look at the settings.

Get the Data Right First

Most Minitab problems are data problems. Five minutes spent getting the data into the right shape saves an hour of confusing output.

Getting data in

1. From Excel or a text file: use File > Open and choose the file, or copy a range from Excel and paste it into the worksheet. Make sure the first row of your range is the column names.
2. Layout: one variable per column, one observation per row, no merged cells, no blank rows in the middle, no totals or notes in the data area.
3. Check the column type. A column of numbers is numeric. If a column header shows -T after the name, Minitab thinks the column is text, often because one cell holds a stray letter or a space. Fix the cell, or use Data > Change Data Type > Text to Numeric. A -D marks a date or time column.
4. Missing values appear as an asterisk (*). Most analyses skip those rows. Check how many rows each analysis used (the session window says so).

Stacked and unstacked data

The same data can be laid out two ways, and many analyses need a specific one.

Stacked is the safer default. Regression, DOE, Gage R&R, ANOVA with several factors, and nearly all graphs by group need the values in one column and the groups in another. Some tests, such as the two-sample t-test and one-way ANOVA, accept either layout and ask you which you used in the dialog. Convert with Data > Stack > Columns and Data > Unstack Columns.

Preparing and checking

Graph Before You Test

Always graph the data first. Minitab’s graphs are interactive and every one can be edited.

Editing and sharing graphs

• Double-click any element (axis, bar, title, symbols) to change it. Right-click for more options.
• Brushing (Editor > Brush) lets you select points on a graph and see which rows they are in the worksheet. It is the quickest way to find an outlier.
• Panels and groups: in most graph dialogs, Multiple Graphs makes one panel per group and lets you share the same scale across panels, so groups can be compared fairly.
• Layout Tool (Editor > Layout Tool) combines several graphs into one page for a report.
• Copy to a document: Edit > Copy Graph and paste into Word or PowerPoint. Graphs can also be saved as image files.
• Keep graphs current: right-click a graph and choose to update it automatically, so it redraws when the data change.

Measure: Trust the Gauge, Then the Process

Measure phase work in Minitab is about two questions: can I trust the measurement? and what does the process do now? Answer the first before the second, because a poor gauge makes every later number unreliable.

Gage R&R (crossed)

A crossed Gage R&R study has several operators measure the same parts, several times each. It splits the total variation into repeatability (the gauge itself), reproducibility (differences between operators), and part-to-part variation (the real differences you want to see). For the full method see the Measurement System Analysis guide.

1. Plan the study: 10 parts that span the range of the process, 3 operators, 2 or 3 repeats. Randomize the order, and do not let operators see earlier results.
2. Set up the worksheet in stacked form: one column for Part, one for Operator, one for the Measurement, one row per measurement:

3. Part	Operator	Measurement (mm)
   1	A	50.55
   1	A	50.64
   1	B	50.75
   1	B	50.53
   1	C	50.57
   …	…	…

4. Run it: Stat > Quality Tools > Gage Study > Gage R&R Study (Crossed). Enter the Part, Operator, and Measurement columns. Choose the ANOVA method. Under Options, set the study variation multiplier (6 is the default; some standards use 5.15), enter the process tolerance if you have one, and keep the default alpha of 0.25 for removing the interaction term.
5. Read the output using the thresholds below, and look at the six graphs: components of variation, range chart by operator, mean chart by operator, measurements by part, measurements by operator, and the operator-by-part interaction.

These are the commonly used guidelines from the AIAG Measurement Systems Analysis manual and Minitab’s help. The right limit depends on the risk and cost of a wrong decision, so agree it with your customer.

A worked example

Ten parts were measured by three operators, twice each (60 measurements), with a process tolerance of 3 mm. The session window shows:

Gage R&R Study - ANOVA Method

Two-Way ANOVA Table Without Interaction
(the Operator*Part interaction was removed: its p-value, 0.518, is above 0.25)

Source          DF        SS        MS        F      P
Part             9   17.0508    1.8945   247.63  0.000
Operator         2    0.1426    0.0713     9.32  0.000
Repeatability   48    0.3672    0.0077
Total            59   17.5607

Gage R&R

                          %Contribution
Source              VarComp  (of VarComp)
Total Gage R&R      0.01083        3.33
  Repeatability     0.00765        2.35
  Reproducibility   0.00318        0.98
    Operator        0.00318        0.98
Part-To-Part        0.31448       96.67
Total Variation     0.32531      100.00

Process tolerance = 3

                           Study Var  %Study Var  %Tolerance
Source              StdDev (SD)     (6 * SD)       (%SV)  (SV/Toler)
Total Gage R&R        0.10408      0.62451       18.25       20.82
  Repeatability       0.08747      0.52480       15.34       17.49
  Reproducibility     0.05642      0.33851        9.89       11.28
    Operator          0.05642      0.33851        9.89       11.28
Part-To-Part          0.56079      3.36472       98.32      112.16
Total Variation       0.57036      3.42218      100.00      114.07

Number of Distinct Categories = 7

• The interaction is not significant (p = 0.52), so Minitab pooled it into the error term. Operators measure all parts consistently.
• Total Gage R&R is 3.3% of the variance, 18.2% of the study variation, and 20.8% of the tolerance. That is marginal on every measure: usable for some purposes, but not good.
• The operator effect is significant (p = 0.000). Reproducibility is 9.9% of study variation, so the operators have different average readings. Repeatability is 15.3%.
• Number of distinct categories = 7, above the minimum of 5. The gauge can tell the parts apart.

Other measurement studies

Attribute agreement. Have several appraisers rate the same samples more than once, ideally against a known standard. Minitab reports the percent agreement within each appraiser, between appraisers, and against the standard, plus kappa. As a common guide, kappa above 0.9 is excellent, 0.7 to 0.9 is acceptable, and below 0.7 means the definition or the training must be fixed.

Capability and sample size

• Capability: Stat > Quality Tools > Capability Analysis > Normal, or Capability Sixpack to see the control chart, histogram, probability plot, and capability together. Enter the specification limits and the subgroup size, and remember the chart must show the process is in control first. See Capability Statistics.
• Non-normal data: Stat > Quality Tools > Capability Analysis > Nonnormal fits another distribution or transforms the data. Pass/fail data use the Binomial option and defect counts the Poisson option.
• How many measurements? Stat > Power and Sample Size has a menu for each test (1-sample t, 2-sample t, one-way ANOVA, proportions, factorial designs). Enter the difference you care about and the standard deviation. See Sample Size and Power.

Analyze: Find the Right Test

Analyze is where the menus branch most. The table below maps the question to the menu and to the Stat Dojo page that teaches the test with a worked example and the Excel route. If you are not sure which test applies, the method selector on Stat Dojo, or the Minitab Assistant, will point you to it.

How to read any Minitab output

1. Check the row count and the settings at the top: which columns, how many rows were used, which confidence level.
2. Look at the graph first. Is the pattern the one the test assumes?
3. Find the p-value and the interval. The p-value says whether an effect is likely real; the interval says how big it might be.
4. Check the assumptions. Normality, equal variance, independence, and the residual plots. Minitab flags problems in the session window (unusual observations, a warning about small samples) and the Assistant’s report card does it in plain language.
5. Write the sentence: the finding, the size of the effect, the interval, and the p-value, in words a sponsor understands.

The Assistant

The Assistant menu covers Measurement Systems Analysis, Capability Analysis, Graphical Analysis, Hypothesis Tests, Regression, Control Charts, and Design of Experiments. For each it presents a decision tree, runs the analysis, and produces three outputs: a Summary Report with the conclusion, a Diagnostic Report showing the assumption checks, and a Report Card listing each check as passed, borderline, or failed with an explanation. It also reports statistical power for hypothesis tests. It is an excellent learning aid and a good second opinion, though it does not offer every option the menus do.

Improve: Plan, Run, and Analyze an Experiment

Minitab’s design of experiments tools plan the experiment, record it in a worksheet in a standard form, analyze it, and find the best settings. The analysis is shown with a full worked example in Analyzing Designed Experiments; this section covers the workflow.

1. Plan. Stat > DOE > Factorial > Create Factorial Design. Choose the number of factors and the design (use Display Available Designs to see the options). Choose replicates, center points, and blocks. In Factors, name each factor and give its low and high levels. Leave randomization on.
2. Run the experiment in the RunOrder, not the StdOrder. The worksheet records both. Type the measured response into a new column as you go.
3. Analyze. Stat > DOE > Factorial > Analyze Factorial Design. Choose the response, include the interactions you can estimate, and under Graphs choose the Pareto chart of effects and the four-in-one residual plots.
4. Look. Stat > DOE > Factorial > Factorial Plots for main effects and interaction plots, and Cube Plots for the corner means.
5. Reduce the model to the significant terms (keep a main effect if its interaction stays), check the residuals again, then use Stat > DOE > Factorial > Response Optimizer to find the settings that hit your goal, and Predict for intervals at chosen settings.
6. Confirm. Run the recommended settings several times before changing the process.

The design worksheet looks like this for a three-factor, eight-run design. The StdOrder column is the textbook order, and RunOrder is the random order in which to run it:

Which design?

• Randomize the run order and replicate so you can estimate error. Add center points to detect curvature.
• Block when you cannot run everything under the same conditions (days, batches).
• Set the base for randomization (Calc > Set Base) if you want to reproduce the same random order.

Control: Charts That Hold the Gain

Control charts keep the improvement in place. Minitab has a chart for every kind of data and checks each point against a set of tests for special causes.

Settings that matter

• Tests (in the chart’s Options): choose which special-cause tests apply, such as one point beyond 3 sigma (Test 1), nine points in a row on one side of the center line (Test 2), six points in a row trending up or down (Test 3), and others up to Test 8. Minitab flags failing points in red and lists them in the session window. You can change how many points a test needs.
• Estimate: omit points with a known special cause so they do not distort the limits, and choose how sigma is estimated.
• Parameters: enter historical limits or a known mean and sigma to hold limits fixed, instead of recalculating them as new data arrive.
• Stages: use a stage column to calculate separate limits before and after an improvement, so the chart shows the change.
• Always check the range or moving range chart first. If variation is out of control the averages chart cannot be interpreted. See Control Chart Theory.

Work Faster and Leave an Audit Trail

When the Output Looks Wrong

Minitab or Excel?

Other statistical packages (JMP, R, Python, and others) do the same jobs, and the ideas here carry over, but the menus and the output differ. Pick the one your organization and your customers expect, and learn the statistics, which stay the same.

Menu Map

A one-page reference. Print it or save it as a PDF.

Minitab Glossary

Assistant

A guided Minitab menu that selects the method, runs it, and reports on the assumptions.

Brushing

Selecting points on a graph to see which worksheet rows they are.

Column properties

Settings for a column, such as the value order for text levels.

Dialog box

The window in which you choose columns and options for an analysis.

Exec file

A saved list of session commands (.mtb) that Minitab can run.

Layout Tool

A tool for combining several graphs into one page.

Project

A file holding worksheets, session output, graphs, and history.

Project Manager

The tree view of everything in the project.

Report Pad

A place to collect output and graphs into a report as you work.

Session window

The window that shows the text output of analyses.

Stacked data

Values in one column and the group labels in another.

Stage

A section of a control chart with its own limits, such as before and after a change.

Subgroup

A small set of measurements collected together for a control chart.

Worksheet

The data grid of columns and rows.

Minitab Guide for Lean Six Sigma: Frequently Asked Questions

Related Pages

Related tools: Process Capability Helper, Gauge R&R Calculator, Control Chart Selector, One-Way ANOVA Calculator, Regression Calculator, DOE Quick Planner.

Sources and Further Reading

• Minitab Support documentation, including the Methods and Formulas topics for Gage R&R, capability analysis, control charts, and designed experiments (support.minitab.com).
• Automotive Industry Action Group, Measurement Systems Analysis Reference Manual, 4th ed. (Gage R&R and attribute agreement guidelines).
• Montgomery, D. C., Introduction to Statistical Quality Control, Wiley (control charts and capability).
• Montgomery, D. C., Design and Analysis of Experiments, Wiley (factorial and fractional factorial designs, resolution).
• NIST/SEMATECH, e-Handbook of Statistical Methods (itl.nist.gov/div898/handbook).
• SixSigmaKaizen.com Stat Dojo for the statistics behind each test, with worked examples and Excel steps.