Proportion Tests and Mean Tests

Learning Objectives

Explain the purpose of hypothesis testing in Six Sigma.
Understand the difference between a population parameter and a sample statistic.
Distinguish between proportion data and mean-based measurements.
Formulate null and alternative hypotheses.
Understand the significance level and p-value.
Apply one-sample proportion tests at a Green Belt level.
Apply two-sample proportion tests at a Green Belt level.
Apply one-sample mean tests.
Apply two-sample mean tests.
Understand the difference between paired and independent samples.
Interpret hypothesis-test results correctly.
Understand the role of confidence intervals in statistical inference.
Recognize when chi-square and ANOVA may be appropriate.
Understand the importance of selecting the correct statistical test.

1. Introduction

Six Sigma improvement decisions should be based on evidence rather than assumptions.

During a project, a Green Belt may encounter questions such as:

Is the current defect rate different from the required level?
Did the defect rate change after an improvement?
Is the average cycle time different from the target?
Is the average performance of two processes different?
Is an observed difference large enough to be considered statistically significant?

Descriptive statistics can show that two groups have different averages or proportions.

However, an observed difference in a sample does not necessarily mean that a real difference exists in the underlying population.

Sampling variation can produce differences even when the underlying population characteristics are the same.

Hypothesis testing provides a structured method for evaluating such questions.

The central idea is:

Use sample evidence to evaluate a claim about a population.

2. Concept / Theory

2.1 What Is Hypothesis Testing?

Hypothesis testing is a statistical method used to evaluate evidence concerning a population parameter or relationship.

A typical hypothesis test involves:

Defining the question.
Stating the hypotheses.
Selecting the appropriate statistical test.
Establishing the significance level.
Collecting appropriate data.
Calculating the test statistic and/or p-value.
Applying the decision rule.
Drawing a conclusion in the context of the problem.

The statistical test helps determine whether the observed sample evidence is sufficiently inconsistent with the null hypothesis.

2.2 Null Hypothesis

The null hypothesis, commonly written as H₀, represents the baseline assumption being tested.

Examples:

H₀: p = 0.05

This may represent a claim that the population defect proportion is 5%.

Or:

H₀: μ = 10 minutes

This may represent a claim that the population mean cycle time is 10 minutes.

2.3 Alternative Hypothesis

The alternative hypothesis, commonly written as H₁ or Hₐ, represents the competing claim.

Examples include:

Hₐ: p ≠ 0.05

Hₐ: μ ≠ 10 minutes

The alternative may also be directional when the improvement question specifically concerns an increase or decrease.

For example:

Hₐ: μ < 10 minutes

would represent a claim that the population mean is less than 10 minutes.

2.4 One-Sided and Two-Sided Tests

A two-sided test investigates whether a parameter is different in either direction.

Example:

H₀: μ = 10

Hₐ: μ ≠ 10

A one-sided test investigates a specified direction.

Example:

H₀: μ = 10

Hₐ: μ < 10

The direction should be established from the improvement question before examining the results.

2.5 Significance Level

The significance level, commonly represented by α, is the threshold used for the statistical decision.

A commonly used value is:

α = 0.05

This represents a 5% significance level.

The significance level should be selected before interpreting the test result.

2.6 P-Value

The p-value measures how inconsistent the observed data are with the null hypothesis under the assumptions of the statistical test.

A small p-value indicates that the observed data would be relatively unusual if the null hypothesis were true.

A commonly used decision rule is:

If p-value ≤ α → Reject H₀

If p-value > α → Fail to reject H₀

The phrase fail to reject is preferred to saying “accept the null hypothesis.”

Failing to reject H₀ does not prove that H₀ is true.

2.7 Statistical Significance

If the p-value is less than or equal to the selected significance level, the result is described as statistically significant under that testing framework.

Statistical significance does not automatically mean that the difference is practically important.

For example, a very small difference may be statistically significant when a very large sample is used.

The Green Belt should therefore consider both:

Statistical significance.
Practical or business significance.

3. Key Topics

3.1 Proportion Data

A proportion represents the fraction of observations in a population having a particular attribute.

For example:

Defective / non-defective.
Pass / fail.
Complaint / no complaint.
On-time / late.
Approved / rejected.

If 20 of 200 units are defective:

Sample proportion = 20 / 200 = 0.10

or:

10%

3.2 One-Sample Proportion Test

A one-sample proportion test evaluates whether a population proportion differs from a specified value.

Example question:

Is the population defect proportion different from 5%?

The hypotheses could be:

H₀: p = 0.05

Hₐ: p ≠ 0.05

The test uses sample data to determine whether there is sufficient evidence against the null hypothesis.

3.3 Two-Sample Proportion Test

A two-sample proportion test compares proportions from two populations or groups.

Example:

A company wants to determine whether defect proportions differ between:

Shift A.
Shift B.

The hypotheses may be:

H₀: p₁ = p₂

Hₐ: p₁ ≠ p₂

The test evaluates whether the observed difference between sample proportions provides sufficient evidence of a population difference.

3.4 Mean-Based Data

A mean describes the average value of a numerical measurement.

Examples include:

Cycle time.
Weight.
Temperature.
Processing time.
Diameter.
Cost.

Mean tests are appropriate when the question concerns a population mean under the assumptions of the selected method.

3.5 One-Sample Mean Test

A one-sample mean test evaluates whether a population mean differs from a specified value.

Example:

A service organization requires an average response time of 10 minutes.

The Green Belt wants to determine whether the population mean response time differs from 10 minutes.

Hypotheses:

H₀: μ = 10 minutes

Hₐ: μ ≠ 10 minutes

The appropriate test depends on factors such as sample size, distributional assumptions, and whether the population standard deviation is known.

3.6 Two-Sample Mean Test

A two-sample mean test compares the means of two groups.

Example:

A Green Belt wants to compare average cycle time between:

Process A.
Process B.

The hypotheses may be:

H₀: μ₁ = μ₂

Hₐ: μ₁ ≠ μ₂

The test evaluates whether the observed difference provides sufficient evidence of a difference between the population means.

3.7 Independent Samples

Two samples are considered independent when observations in one sample are not naturally paired with observations in the other sample.

Example:

Cycle times from:

Machine A

and

Machine B

may constitute independent samples if the observations are collected independently.

3.8 Paired Samples

Paired data occurs when observations are naturally linked.

Examples include:

Before and after measurements on the same unit.
Performance of the same employee before and after training.
Measurements taken from the same machine before and after an adjustment.

The analysis must account for the pairing.

A paired analysis focuses on the differences within each pair.

3.9 Confidence Interval

A confidence interval provides an interval estimate for a population parameter using a specified confidence level and statistical method.

For example, a confidence interval for a population mean may provide:

Estimated mean = 12 minutes

with a:

95% confidence interval

The interval provides information about plausible values for the population parameter under the assumptions of the method.

Confidence intervals are useful because they communicate both an estimate and its uncertainty.

4. Tools / Methodology

4.1 General Hypothesis-Testing Procedure

A Green Belt can follow this sequence:

Step 1 — Define the question

Clearly state what needs to be determined.

Step 2 — Identify the response variable

Determine whether the outcome is:

A proportion.
A mean.
A count.
Another type of measurement.

Step 3 — Identify the comparison

Determine whether the analysis involves:

One population versus a target.
Two independent populations.
Paired observations.
Multiple groups.

Step 4 — State H₀ and Hₐ

Write the hypotheses before analyzing the results.

Step 5 — Select α

Choose an appropriate significance level.

Step 6 — Select the statistical test

Use the data type, study design, and assumptions to select the method.

Step 7 — Analyze the data

Calculate the appropriate test statistic and p-value, commonly using statistical software such as Minitab.

Step 8 — Apply the decision rule

Compare the p-value with α.

Step 9 — State the conclusion in context

Explain what the statistical result means for the Six Sigma problem.

4.2 Minitab Application

Statistical software such as Minitab can support hypothesis testing.

Typical workflow:

Statistical Question → Identify Data Type → Select Test → Enter Data → Review Output → Interpret P-Value and Confidence Interval → State Conclusion

The Green Belt should understand the statistical question and assumptions before using the software.

Software does not determine whether the selected test is appropriate.

4.3 Example — One-Sample Proportion

Suppose a process historically has a target defect proportion of:

A sample of production units is collected.

The Green Belt wants to determine whether the current defect proportion differs from 5%.

Hypotheses:

H₀: p = 0.05

Hₐ: p ≠ 0.05

If the resulting p-value is:

0.02

and:

α = 0.05

then:

0.02 < 0.05

The decision is to reject H₀.

The conclusion should be stated in context:

The sample provides sufficient statistical evidence that the population defect proportion differs from 5%, under the assumptions of the test.

4.4 Example — Two-Sample Proportion

Suppose two production shifts have defect proportions:

Shift A = 4%

Shift B = 7%

The observed proportions differ.

However, the difference alone does not establish that the underlying population proportions are different.

A two-sample proportion test can be used to evaluate:

H₀: pA = pB

against an appropriate alternative.

The p-value and confidence interval provide evidence for interpreting the observed difference.

4.5 Example — One-Sample Mean

Suppose a process has a required average cycle time of:

10 minutes

A sample produces an observed mean of:

10.8 minutes

The Green Belt wants to determine whether the population mean differs from 10 minutes.

The hypotheses may be:

H₀: μ = 10

Hₐ: μ ≠ 10

The observed difference of 0.8 minutes is not, by itself, sufficient to establish statistical significance.

The appropriate hypothesis test must be performed.

4.6 Example — Two-Sample Mean

Suppose two machines produce the following average cycle times:

Machine A = 12.1 minutes

Machine B = 11.6 minutes

The observed difference is:

0.5 minutes

A two-sample mean test can be used to determine whether the evidence supports a difference between the population means.

Again, the difference between sample means alone does not determine statistical significance.

4.7 Chi-Square Test Awareness

The chi-square test is used in several statistical applications.

In Six Sigma, a Green Belt may encounter chi-square methods when analyzing categorical data, including tests of association or independence between categorical variables.

For example:

Defect Type × Production Shift

A chi-square analysis can help evaluate whether the observed relationship between categorical variables is statistically significant under the relevant assumptions.

The exact test should be selected according to the question and data structure.

4.8 ANOVA Awareness

ANOVA — Analysis of Variance — is used to compare means across multiple groups.

For example:

A company wants to compare average cycle time across:

Shift A.
Shift B.
Shift C.
Shift D.

Rather than performing multiple two-sample comparisons without an appropriate overall strategy, ANOVA can be used to evaluate whether there is evidence that at least one group mean differs.

ANOVA will be covered in greater depth where appropriate in the statistical analysis curriculum.

5. Worked Example / Case Study

Case Study — Comparing Defect Rates Between Two Shifts

A manufacturing organization wants to determine whether defect performance differs between two shifts.

The Green Belt collects independent samples from:

Shift A

and

Shift B

The response variable is:

Defective / Non-defective

Therefore, the response is categorical and can be represented by proportions.

Step 1 — Define the Question

Is the defect proportion different between Shift A and Shift B?

Step 2 — Define the Hypotheses

H₀: pA = pB

Hₐ: pA ≠ pB

Step 3 — Select the Test

Because the response is categorical and two independent groups are being compared, an appropriate two-sample proportion method can be considered, subject to its assumptions.

Step 4 — Analyze

The Green Belt enters the appropriate data into statistical software.

The output may provide:

Sample proportions.
Difference in proportions.
Test statistic.
P-value.
Confidence interval.

Step 5 — Interpret

Suppose the analysis produces:

p-value = 0.03

with:

α = 0.05

Since:

0.03 < 0.05

the decision is to reject H₀.

Step 6 — State the Conclusion

A suitable conclusion is:

The data provides sufficient statistical evidence that the defect proportions differ between the two shifts, under the assumptions of the test.

The Green Belt should then investigate the practical meaning of the difference and possible process factors contributing to it.

5.1 Important Distinction

The hypothesis test establishes evidence concerning the difference in population proportions.

It does not, by itself, establish the reason for the difference.

Determining causes belongs to the broader DMAIC Analyze activity.

6. Practical Application

Exercise 1 — Defect Rate

A process historically operates at a defect rate of 3%.

The Green Belt wants to determine whether the current defect rate differs from 3%.

Questions:

What type of data is involved?
Is this a one-sample or two-sample question?
What should H₀ represent?
What should Hₐ represent?
What significance level will be used?
How will the p-value be interpreted?

Exercise 2 — Cycle Time

A company wants to determine whether the average cycle time is different from the required target of 15 minutes.

Questions:

Is the response variable categorical or numerical?
Is the analysis concerned with a proportion or a mean?
Is this a one-sample or two-sample question?
What would H₀ represent?
What would Hₐ represent?

Exercise 3 — Before and After Improvement

A Green Belt records the processing time of the same group of transactions before and after an improvement.

Questions:

Are the observations naturally paired?
Should the analysis account for the pairing?
What is the response variable?
What population difference is being evaluated?

6.1 Practical Versus Statistical Significance

Suppose a large dataset shows that an improvement reduces average cycle time by:

0.05 minutes

and the p-value is very small.

The result may be statistically significant.

However, the organization should still ask:

Is the improvement operationally meaningful?
Does it justify implementation cost?
Does it improve customer performance?
Does it affect capacity?
Does it reduce meaningful business loss?

Six Sigma decisions should therefore consider both statistical evidence and practical impact.

6.2 Common Hypothesis-Testing Errors

Error 1 — Starting With the P-Value

The Green Belt should first understand the business question and select the appropriate test.

Error 2 — Choosing a Test Based Only on the Data’s Appearance

Test selection depends on the response type, study design, assumptions, and question.

Error 3 — Treating a Small P-Value as Proof of the Alternative

A small p-value provides evidence against H₀; it does not establish every aspect of Hₐ as proven fact.

Error 4 — Saying “Accept H₀”

The preferred wording is:

Fail to reject H₀

because the test does not prove that H₀ is true.

Error 5 — Confusing Statistical and Practical Significance

A statistically significant result may have little practical importance.

Error 6 — Ignoring Pairing

Before-and-after measurements on the same units or subjects are not generally treated as independent observations.

Error 7 — Ignoring Assumptions

The selected statistical method should be appropriate for the data and study conditions.

7. Lesson Summary

Hypothesis testing provides a structured method for evaluating whether observed sample evidence supports a difference or effect in the population.

Important concepts include:

Null hypothesis.
Alternative hypothesis.
Significance level.
P-value.
Statistical significance.
Confidence interval.
Decision rule.

For categorical outcomes, proportion tests can be used when appropriate.

For numerical measurements, mean tests can be used when appropriate.

Important applications include:

One-sample proportion tests.
Two-sample proportion tests.
One-sample mean tests.
Two-sample mean tests.
Paired-data analysis.

A Green Belt should also have awareness of:

Chi-square methods for appropriate categorical-data questions.
ANOVA for comparing means across multiple groups.

Statistical software such as Minitab can perform the calculations, but the Green Belt must understand the underlying statistical question and assumptions.

8. Lesson Learnt / Conclusion

Hypothesis testing helps the Green Belt distinguish between an observed difference in sample data and evidence of a difference in the underlying population.

The correct sequence is:

Define the Question → Identify the Data → Select the Test → State Hypotheses → Set α → Analyze → Interpret P-Value / Confidence Interval → State the Conclusion

A sound statistical conclusion should always be connected back to the original business or process question.

The Green Belt should also remember that statistical significance is not the same as practical significance.

Green Belt Takeaway:

Use the right statistical test, apply a clear decision rule, and interpret the result in the context of the process problem.