What you will learn: how to decide whether a difference in sample data is strong enough to support a workplace conclusion.
StatisticsQuality EngineeringLean Six Sigma
By the end, you should be able to
Turn a workplace question into a null and alternative hypothesis.
Use the p-value and significance level to make the correct statistical decision.
Explain Type I and Type II errors in practical language.
Choose a basic hypothesis test for a common business or quality problem.
Run and interpret a 1-sample t-test in Minitab.
Why this matters at work
A sample can look different even when the underlying process has not changed. Hypothesis testing helps you separate a real signal from normal sample-to-sample variation. It gives you a disciplined way to decide whether to adjust a process, challenge a supplier, approve a change, or collect more evidence.
Important: a hypothesis test does not make the business decision for you. It measures the strength of the statistical evidence. You must still judge cost, risk, specifications, and practical importance.
Hypothesis testing in plain English
Start with a default explanation: nothing meaningful has changed. Then ask whether the data are too unusual for that explanation to remain believable.
Start with H0No difference or no effect
→
Challenge it with dataIs the result too unusual under H0?
Memory phrase:
Assume no real difference. Measure how surprising the sample would be. Reject that assumption only when the evidence is strong enough.
The two competing statements
Null hypothesis, H0
The starting position. It usually says there is no difference, no effect, or the population value equals a target.
Alternative hypothesis, H1
The claim supported when the evidence against H0 is strong enough. It states the difference or direction you are testing.
Example: average wall thickness
The nominal average wall thickness is 8.00 mm. You want to know whether the process average has moved in either direction.
H0: μ = 8.00 mm H1: μ ≠ 8.00 mm
One-sided or two-sided?
Question
Alternative hypothesis
Test direction
Has the mean changed in either direction?
μ ≠ target
Two-sided
Has the mean fallen below the minimum?
μ < target
Lower one-sided
Has the mean increased above the baseline?
μ > target
Upper one-sided
Choose the direction before looking at the result. Changing from a two-sided test to a one-sided test after seeing the data weakens the credibility of the analysis.
The six-step workflow
State the business question. Example: “Has the average wall thickness moved away from 8.00 mm?”
Write H0 and H1. Use population language, not just the values in your sample.
Choose α before analyzing. A common choice is 0.05, but the risk of a false alarm should match the decision.
Select the right test. Base the choice on the response type, number of groups, and whether observations are independent or paired.
Check assumptions and run the test. Look for poor sampling, dependence, outliers, and strong skewness.
Interpret both statistical and practical significance. Report the estimated difference, confidence interval, p-value, and business meaning.
The p-value and alpha
Alpha (α) is the false-alarm risk you choose before the test. The p-value measures how incompatible the sample result is with H0, assuming H0 and the test assumptions are correct.
Do not say “the p-value is the probability that H0 is true.” It is calculated under the assumption that H0 is true. It tells you how unusual the observed evidence would be under that assumption.
Interactive decision lab
Separate statistical significance from practical significance
Change the values below. The tool will give you the formal test decision and a separate practical interpretation.
Must be between 0 and 1.
Set this before analyzing the data.
Example: sample mean minus target.
Smallest change worth acting on.
Type I and Type II errors
Type I error: false alarm
You reject H0 even though it is true. Example: you stop and adjust a stable process because the sample looked unusual by chance. Alpha controls this risk.
Type II error: missed signal
You fail to reject H0 even though a real difference exists. Example: you miss a meaningful process shift because the sample is too small or too noisy.
Power = 1 − β. Higher power means a better chance of detecting a real difference of practical importance. Larger samples, lower variation, and larger effects usually improve power.
Interactive basic-test selector
Which test fits the question?
This is a beginner guide for common mean and proportion questions. Complex designs may require a more specialized method.
Worked example in Minitab
Is average wall thickness different from 8.00 mm?
A quality engineer randomly selects 12 pipe measurements. The question is whether the population mean differs from the 8.00 mm nominal value.
Hypotheses: H0: μ = 8.00 mm; H1: μ ≠ 8.00 mm. Use α = 0.05.
Minitab steps
1
Enter the 12 values in one worksheet column and name it Wall Thickness.
2
Choose Stat > Basic Statistics > 1-Sample t.
3
Select One or more samples, each in a column, then enter Wall Thickness.
4
Select Perform hypothesis test and enter 8.00 as the hypothesized mean.
5
Under Options, use a 95% confidence level and the alternative mean ≠ hypothesized mean. Click OK.
Simplified Minitab-style output
One-Sample T: Wall Thickness
Test of μ = 8.000 vs μ ≠ 8.000
N Mean StDev SE Mean 95% CI
12 7.919 0.098 0.028 (7.857, 7.982)
T-Value = -2.85 P-Value = 0.016
Statistical decision: 0.016 ≤ 0.05, so reject H0.
Evidence statement: the sample provides statistically significant evidence that the population mean differs from 8.00 mm.
Direction and size: the estimated mean is 0.081 mm below nominal. The 95% confidence interval for the mean is 7.857 to 7.982 mm.
Business interpretation: determine whether a shift of 0.081 mm is large enough to matter. Compare it with engineering tolerances, risk, and the preselected practical threshold.
Assumption check before trusting the result
Representative sample
The measurements should represent the population and time period covered by your conclusion.
Independent observations
One observation should not determine another. Repeated measurements on the same item may require a paired or repeated-measures method.
No unexplained extreme outliers
Investigate data-entry errors, measurement problems, and unusual special causes before running the test.
Distribution is reasonable
For a small sample, severe skewness can make a t-test less reliable. Use a histogram, boxplot, or probability plot.
Common interpretation mistakes
“Fail to reject” does not mean “prove equal.” It means the sample did not provide strong enough evidence of a difference at the chosen alpha level.
Statistical significance does not guarantee practical importance. A large sample can detect a difference that is too small to justify action.
A hypothesis test is not a specification test. Testing whether the process mean equals a target does not tell you whether every unit meets specification limits.
Do not report the p-value alone. Include the estimated effect, confidence interval, sample size, assumptions, and business meaning.
A professional reporting sentence
“A 1-sample t-test found that average wall thickness differed from the 8.00 mm nominal value (mean = 7.919 mm, 95% CI: 7.857 to 7.982 mm, p = 0.016). The estimated shift was 0.081 mm below nominal. Engineering review is required to determine whether this difference is practically important.”
Check your understanding
Lesson summary
H0 is the default claim; H1 is the difference or direction being tested.
Choose the test direction and alpha before looking at the result.
Reject H0 when p ≤ α; otherwise, fail to reject H0.
A p-value is not the probability that H0 is true.
Always separate statistical significance from practical significance.
Report the effect estimate, confidence interval, p-value, assumptions, and workplace meaning.
UpSkill Sprint Consulting
Start a 15-min lesson