Kruskal-Wallis Test in Six Sigma: How to Use this Non-Parametric Test

The Kruskal-Wallis test plays a critical role in Six Sigma analysis. It helps you compare multiple groups when your data does not meet normality assumptions. Many real-world processes produce skewed, ordinal, or non-normal data. Therefore, this test becomes a powerful alternative to ANOVA.

In this guide, you will learn how the Kruskal-Wallis test works, when to use it, and how to apply it in Six Sigma projects. You will also see step-by-step examples, practical tables, and real manufacturing scenarios.

What Is the Kruskal-Wallis Test?

The Kruskal-Wallis test is a non-parametric statistical test. It compares three or more independent groups. Unlike ANOVA, it does not assume normal distribution.

Instead of using raw data, it uses ranks. This approach makes it robust against outliers and skewed distributions.

Key idea

• It tests whether the medians of multiple groups differ
• It uses ranked data instead of raw values
• It works well with ordinal or non-normal data

Why the Kruskal-Wallis Test Matters in Six Sigma

Six Sigma focuses on reducing variation and improving processes. However, real-world data rarely behaves perfectly.

You often encounter:

• Skewed cycle times
• Non-normal defect counts
• Ordinal customer satisfaction scores
• Small sample sizes

Because of this, traditional ANOVA may fail. That is where the Kruskal-Wallis test adds value.

Practical benefits

Benefit	Why It Matters in Six Sigma
No normality assumption	Works with real process data
Handles outliers	Reduces distortion in analysis
Works with ordinal data	Useful for surveys and ratings
Simple interpretation	Easy to explain to stakeholders

When Should You Use the Kruskal-Wallis Test?

You should use this test under specific conditions. Otherwise, you risk incorrect conclusions.

Use it when:

• You compare 3 or more groups
• Your data is non-normal
• Your data is ordinal or ranked
• Samples are independent

Avoid it when:

• Data is normal and meets ANOVA assumptions
• Groups are dependent (use Friedman test instead)
• Sample sizes are extremely small

Kruskal-Wallis vs ANOVA

Many practitioners struggle to choose between these two tests. The decision depends on your data.

Comparison table

Feature	Kruskal-Wallis Test	One-Way ANOVA
Data type	Ordinal or continuous	Continuous
Distribution	Not required	Must be normal
Outlier sensitivity	Low	High
Uses ranks	Yes	No
Compares	Medians	Means

Quick rule

• Use ANOVA for clean, normal data
• Use Kruskal-Wallis for real-world messy data

How the Kruskal-Wallis Test Works

The test follows a structured process. It converts raw data into ranks and then compares rank sums.

Step-by-step logic

1. Combine all data from all groups
2. Rank all values from smallest to largest
3. Calculate rank sums for each group
4. Compute the test statistic (H)
5. Compare H to a chi-square distribution

The Kruskal-Wallis Formula

$H = \frac{12}{N(N+1)} \sum_{i=1}^{k} \frac{R_i^2}{n_i} – 3(N+1)$

Where:

• N = total number of observations
• k = number of groups
• Rᵢ = sum of ranks for group i
• nᵢ = sample size of group i

Hypotheses in the Kruskal-Wallis Test

Every Six Sigma analysis starts with hypotheses.

Null hypothesis (H₀)

All group medians are equal.

Alternative hypothesis (H₁)

At least one group median differs.

Step-by-Step Example (Manufacturing Case)

Let’s walk through a practical Six Sigma example.

Scenario

A process engineer wants to compare cycle times across three machines.

Data (minutes)

Machine A	Machine B	Machine C
10	14	18
12	16	20
11	15	19

Step 1: Combine and Rank Data

Value	Rank
10	1
11	2
12	3
14	4
15	5
16	6
18	7
19	8
20	9

Step 2: Sum Ranks by Group

Group	Rank Sum
Machine A	1 + 2 + 3 = 6
Machine B	4 + 5 + 6 = 15
Machine C	7 + 8 + 9 = 24

Step 3: Apply Formula

You plug values into the formula and compute H.

Step 4: Compare to Critical Value

If H exceeds the critical chi-square value, you reject H₀.

Conclusion

Machine cycle times differ significantly. Therefore, the process engineer should investigate root causes.

Interpreting Results in Six Sigma

Interpretation matters as much as calculation.

Key points

• A low p-value (< 0.05) indicates significant differences
• The test does not show which group differs
• You need post-hoc testing for deeper insights

Post-Hoc Analysis After Kruskal-Wallis

The Kruskal-Wallis test tells you that a difference exists. However, it does not tell you where.

Common post-hoc methods

Method	Purpose
Dunn’s Test	Pairwise comparisons
Bonferroni Correction	Controls error rate
Mann-Whitney Tests	Compare pairs

Example: Post-Hoc Insight

From the earlier example:

• Machine A vs B → significant
• Machine B vs C → significant
• Machine A vs C → highly significant

Action

Focus on Machine C. It shows the highest cycle times.

Real Six Sigma Applications

The Kruskal-Wallis test fits many industries.

Manufacturing

• Compare machine performance
• Analyze defect rates across shifts
• Evaluate supplier quality

Healthcare

• Compare patient wait times
• Analyze treatment effectiveness
• Evaluate satisfaction scores

Service Industry

• Compare call center performance
• Analyze customer ratings
• Evaluate response times

Example: Customer Satisfaction Analysis

Scenario

A company collects satisfaction ratings (1–10 scale) across three regions.

Data summary

Region	Median Score
East	6
West	8
Central	7

Insight

The Kruskal-Wallis test reveals significant differences. Therefore, leadership should focus on improving the East region.

Advantages of the Kruskal-Wallis Test

This test offers several advantages for Six Sigma teams.

Key strengths

• Handles non-normal data easily
• Works with small sample sizes
• Reduces impact of extreme values
• Supports ordinal data analysis

Limitations You Should Know

Despite its strengths, the test has limitations.

Key drawbacks

• Does not identify specific group differences
• Less powerful than ANOVA for normal data
• Requires post-hoc testing
• Assumes similar distribution shapes

Kruskal-Wallis in the DMAIC Framework

You can apply this test across multiple DMAIC phases.

Define

Identify process variation across groups.

Measure

Collect data from multiple categories.

Analyze

Use Kruskal-Wallis to detect differences.

Improve

Target the worst-performing group.

Control

Monitor improvements using ongoing analysis.

Example: DMAIC Application

Phase	Action
Define	Compare supplier defect rates
Measure	Collect defect data
Analyze	Run Kruskal-Wallis test
Improve	Improve worst supplier
Control	Track defect trends

Software Tools for Kruskal-Wallis Test

Most Six Sigma professionals use statistical software.

Common tools

Tool	Capability
Minitab	Built-in non-parametric tests
Excel	Requires add-ins
Python	SciPy library
R	kruskal.test() function

Example in Python

from scipy.stats import kruskalgroup1 = [10, 12, 11]
group2 = [14, 16, 15]
group3 = [18, 20, 19]stat, p = kruskal(group1, group2, group3)print(stat, p)

Best Practices for Six Sigma Professionals

You should follow best practices to ensure valid results.

Key recommendations

• Always check data distribution first
• Use boxplots to visualize variation
• Combine with graphical analysis
• Follow up with post-hoc testing
• Communicate results clearly

Common Mistakes to Avoid

Many practitioners misuse this test.

Avoid these errors

• Using it for dependent samples
• Ignoring post-hoc analysis
• Misinterpreting p-values
• Skipping data visualization

Visualizing Kruskal-Wallis Results

Visualization improves understanding.

Recommended charts

• Boxplots
• Violin plots
• Rank plots

Example insight

A boxplot quickly shows which group has higher median performance.

Advanced Insight: Effect Size

Statistical significance alone is not enough. You should also measure effect size.

Common metric

• Eta-squared (η²)

Interpretation

Value	Effect Size
0.01	Small
0.06	Medium
0.14	Large

Kruskal-Wallis vs Mann-Whitney Test

Both are non-parametric tests. However, they serve different purposes.

Comparison

Feature	Kruskal-Wallis	Mann-Whitney
Groups	3 or more	2
Purpose	Multi-group comparison	Pairwise comparison
Output	Overall difference	Specific comparison

Real-World Case Study

Problem

A factory sees variation in defect rates across three shifts.

Data

Shift	Defects
Morning	Low
Afternoon	Medium
Night	High

Analysis

The Kruskal-Wallis test shows a significant difference.

Action

• Investigate night shift processes
• Train operators
• Improve maintenance

Result

Defect rates drop by 25%.

How to Explain Results to Stakeholders

Clear communication drives action.

Simple explanation

“The test shows that at least one group performs differently. Therefore, we should focus on identifying and fixing the root cause.”

Key Takeaways

• Use Kruskal-Wallis when data is non-normal
• It compares medians across multiple groups
• It relies on ranked data
• It requires post-hoc testing for deeper insight
• It fits perfectly into the Analyze phase of DMAIC

Conclusion

The Kruskal-Wallis test gives Six Sigma professionals a powerful tool for analyzing non-normal data. It works well in real-world environments where ideal assumptions fail.

You can use it to compare multiple groups, detect variation, and guide improvement efforts. However, you should always combine it with post-hoc analysis and visualization.