Lean Six Sigma Green Belt

Certification Course
Hypothesis Testing
 Learning Objectives

By the end of this lesson, you will be able to:

Explain the basics of hypothesis testing

Perform tests for Means, Variances, and Proportions

Conduct hypothesis testing with non-normal data

 Scenario

Java Coffee House implemented an improved process at facility B, which would help reduce defects. The
desire is to compare the impact of the improved process at facility B to the standard process being
performed at facility A.

How could you confidently say that the new process is significantly better than the standard process?

Yield A Yield B
80 85
80 81
80 95
93 95
93 89
87 87
82 92
81 81
93 82
Basics
 Hypothesis Testing

Hypothesis test is used to prove or disprove a theory or claim by comparing two

or more samples or comparing a single sample to a defined value.

A Six Sigma project produced significant improvements in average

performance

Product quality is independent of supplier

The average cycle time for processing similar products is the same
between two different facilities
 Hypothesis Testing

Handle uncertainty in data

Minimize subjectivity

⯑ Question assumptions

Manage decision error risk

 Hypothesis Testing

• Determine what to compare and assume that there is no difference

• Decide how much risk of being wrong is acceptable
• Use data to calculate a test statistic
• Compare the test statistic to a critical test statistic
• Make a decision by comparing p-value to accepted risk value
 Null Hypothesis vs. Alternative Hypothesis

Null Hypothesis Alternative Hypothesis

● Represented as H0 ● Represented as Ha

● Statement of no change or difference ● Statement of change or difference

● Cannot be proved, only rejected ● Challenges the null hypothesis

● Example: Movie is worth watching. ● Example: Movie is not worth watching.

🖝 H0 : µa = µb
Ha : µa ≠ µb
Null is stating there is no difference in means
Alterative is stating there is a difference in means
 Hypothesis Testing

Type I Error Type II Error

● Accepting a null hypothesis when it is false

● Rejecting a null hypothesis when it is true
● False negative
● False positive ● Also known as Consumer’s Risk
● Also known as Producer’s Risk ● ‘β’ is the chance of committing a Type II
● Significance level or ‘α’ is the chance of Error
committing a Type 1 error ● The value of ‘β’ is 0.2 or 20%
● The value of ‘α’ is 0.05 or 5% ● Any experiment should have as less β value
as possible
Example: When a movie is worth watching, it is
reviewed as ‘not worth watching.’ Example: When a movie is not worth watching,
it is reviewed as ‘worth watching.’
 Type I and Type II Errors

Court’s Decision

Accept Null: Reject Null:

Defendant is NOT Guilty Defendant is Guilty

Null is True: Type I Error

Actually Innocent Correct
α-Risk

The
Truth

Null is False: Type II Error

Correct
Actually Guilty β-Risk
 Type I and Type II Errors

Probability of making one type of error can be reduced

A false null hypothesis may be accepted (type II error)

‘Α’ is set at 0.05, which means the risk of committing a type I error will be 1
out of 20 experiments

It is important to decide what type of error should be less and set ‘α’ and ‘β’
accordingly
 Type I and Type II Errors

Power of a Test
• The probability of correctly rejecting the null hypothesis when it is false

• The complement to Type II error and represented as 1-β

• The probability of not committing a type II error

• Helps in improving the advantage of hypothesis testing

• With highest value should be preferred when given a choice of tests

Confidence Level
• The probability of correctly failing to reject the null hypothesis when it is true
• The complement to Type I error and represented as 1-α
• The probability of not committing a Type I error
 Sample Size

The sample size for continuous data can be

determined by the formula:
• How much variation is present in the population?
(σ)

• At what interval or tolerance does the true

population mean need to be estimated? ( ±Δ) The sample size for discrete data can be
determined by the formula:

• How much representation error is allowed in the

sample? ( α)
 Standard Sample Size Formula

To calculate the standard sample size for continuous data, the value of α is taken
as 5%.

According to Z table, the Z97.5 = 1.96.

The standardized sample size formula

 Standard Sample Size Formula

Q The population standard deviation for the time, to resolve customer problems, is 30
hours. What should be the size of a sample that can estimate the average problem
resolution time within ± 5 hours tolerance with 99% confidence?

A Δ = 5, σ = 30, α =0.01, and Z99.5 = 2.58

Sample size = [(2.575*30)/5]2 = 238.70 = 239

 Standard Sample Size Formula

To calculate the standard sample size for discrete data, the average
population proportion of non-defective is ‘p’ and value of α is taken as 5%.

DISCRETE DATA

Where = Tolerance allowed on either side of the population proportion average in percentage.
 Standard Sample Size Formula

Q The non-defective population proportion for pen manufacturing is 80%.

What should be the sample size to draw a sample that can estimate the
proportion of compliant pens within ± 5% with an alpha of 5%?

A Δ = 0.05, σ 2= 0.8 (1-0.8), α = 0.05, and Z97.5 = 1.96

Sample size = (1.96/0.05)2 *0.8*0.2 = 245.86 = 246

 P-Values

The P-value is the probability that any differences observed are due to random
chance or common cause variation.

A small p-value indicates a small probability that the observed results

could happen under the assumption that the null hypothesis is true.

Low p-values lead us to reject the null hypothesis.

If lower than alpha risk, the null is rejected in favor of the alternative.

High p-values indicate that we haven’t gathered sufficient data or evidence

to not reject the null hypothesis.
 Test Statistics

A test statistic is used to determine the validity of a

hypothesis test.

The test statistic is calculated based on the

observations.
Directional, Left-tail Test
If the null hypothesis is true the test statistic will be a
random variable with a known distribution.

If we can conclude that the test statistic came from the

population, then the null is not rejected.

If it is not likely for the test statistic to come from Directional, Right-tail Test
population, then the null is rejected.

This is determined by comparing the test statistic to a

critical value.

The calculation of the test statistic and critical value will

depend on the particular hypothesis test.
Non-Directional, Two-tail Test
 2 Tailed Probability and 1 Tailed Probability

Use of 2-tailed probability and 1-tailed probability depends

on the direction of the alternative hypothesis.

2-Tailed Probability 1-Tailed Probability

If the alternate hypothesis tests more than one If the alternate hypothesis tests one direction,
direction, either less or more, use a 2-tailed use a 1-tailed probability value from the test.
probability value from the test.
Example: Example: If Mean of A is greater than Mean of B,
If Mean of A is not equal to Mean of B, then it is then it is 1-tailed probability.
2-tailed probability.
 Hypothesis Test Conclusions

Compare a calculated • If test statistic > critical value, Reject H0

test statistic to a • If test statistic < critical value, Fail to reject H0
critical value

Compare the p-value • If p-value < alpha risk, Reject H0

to the alpha risk • If p-value > alpha risk, Fail to reject H0
 Statistical and Practical Significance of Hypothesis Test

Significant but not Practical

The Java Coffee House distribution center is comparing two methods for fulfilling
orders in which the new method is significantly faster than the standard method by
30 minutes; however to justify the cost of changing to the new method, a reduction
of an hour is required.

Practical but not Significant

A call center is piloting new technology to process payments which has increased
transactions by 500 per day. Although the new technology has a higher average
that will have a practical impact to the business, there is no statistically significant
difference since p ≥ 0.05
Tests for Means, Variances, and Proportions
 Hypothesis Testing Roadmap

Hypothesis
Test

Continuous, Discrete
Normal Data Data

1 group 2 groups >2 groups 1 Factor 2 Factor

1-Way
Mean Variation Variation Mean 1 Group 2 Groups >2 Groups
ANOVA

n>30; n<30 &/or 1- 2–

Dependent Independent Contingency
F test Proportion Proportion
σ known σ unknown Data Data Table or χ2
Test Test

n1,n2<30 &/or
1 Sample Z 1 Sample t n1, n2 ≥30 ;
Paired t Test σ1,σ2
Test Test σ1,σ2 known
unknown

2 Sample Z 2 Sample t
test Test
 Hypothesis Testing Formulas

Test Test For Test Distribution Conditions

Statistic

1 Sample Z test

1 Sample t test

1 Proportion
test
CALCULATE TEST STATISTICS
2 Sample Z test

2 Sample t test

Paired t Test

2 Proportion
test
 1 Sample Test

H0: Average height of North American men is165 cm

Ha: Average height of North American men is not 165 cm

Z-Test t-test

Sample Size (n) = 117 Sample Size (n) = 25

Sample Average = 164.5 cm Sample Average = 164.5 cm

Z Test Statistic = 1.04 t Test Statistic = 0.5

 1 Sample Test

In hypothesis test for variance, Chi-square test is used.

H0: Proportion of wins in Australia or abroad is independent of the country played against

Ha: Proportion of wins in Australia or abroad is dependent on the country played against

χ2 Critical = 6.251

χ2 Calculated = 1.36

Result:
Since calculated value is less than the critical value, the proportion of wins of Australia hockey team
is independent of the country played or place.
 1 Sample Test

H0: Proportion of smokers among men in a place named R is 0.10 (p0)

Ha: Proportion of smokers among men is R is different that 0.10

H0: p = p0 against Ha: p p0

Among n = 150 adult men interviewed, 23 were found smokers.

Sample proportion p = 23/150 = 0.153

Compute test statistic: Zcalc = = 1.80

Reject H0 at level of significance α if zcalc > z1-a/2

Since, z.975 = 1.96, the null hypothesis is not rejected at 5% level of significance.

Result: It can be concluded based on the sample that the proportion of smokers in R is 0.10.
 Means of Two Groups

Understand the significant difference in the outcome of the two processes

Understand whether a new process is better than an old process

Understand whether the two samples belong to the same population or a

different population

Benchmark the existing process with another process

 2 Sample Test

H0: μ1 = μ2 against Ha: μ1 ≠ μ2

Two samples of sizes n1 = 125 and n2 = 110 are taken from the two populations
= 167.3, = 165.8, s1 = 4.2, s2 = 5.0 are the sample means and SDs respectively
Compute test statistic

Reject H0 at level of significance α if I Computed t I > tDF,a/2; DF = (n1 – 1) + (n2 – 1) = 124 + 109 = 233
Since t233,0.025 = 1.97[“=T.INV.2T(.05,233)”], the null hypothesis is rejected at 5% level of significance
 2 Sample Test

Q Obtained earnings data for Company A = 31 years

Obtained earnings data for Company B = 41 years
Sample standard deviation of Company A’s earnings = $4.40
Sample standard deviation of Company B’s earnings = $3.90.

Determine whether the earnings of Company A have a greater

standard deviation than those of Company B at 5% level of significance.

A
H0 : σA2= σB2 ; the variance of Company A’s earnings is equal to the
variance of Company B’s earnings.
Ha : σA2 ≠ σB2 ; the variance of Company A’s earnings is different.
σA2= variance of Company A’s earnings.
σB2= variance of Company B’s earnings.
 F-Test Example

dfA (degrees of freedom of A) = 31 – 1 = 30

Results
dfB (degrees of freedom of B) = 41 – 1 = 40
The F-test statistic (1.273) is not greater than the

critical value (1.74). Therefore, at 5% significance

Critical value from F-table = 1.74
level, the null hypothesis cannot be rejected.
[“=F.INV.RT(0.50,30,40)”].

Calculation of F-test statistic:

F= (SA2/SB2) = 4.402/3.902 = 1.273

σA > σB. In calculating the F-test statistic, always put the greater
variance in the numerator.
 F-Test Example

A restaurant wants to explore the overuse of avocados. It suspects that there is a

difference in the way the two chefs use avocados and the number of avocados
used, in pounds.

Group A (Chef 1) Group B (Chef 2)

4.2 4

4.5 4.5

7.2 5

6.1 5.2

8.9 5.3

5.2 6.1
Conducting F-Test In MS Excel
 F-Test Example

Null Hypothesis
• There is no significant statistical difference between the variances of the two groups

• This is Common Cause of Variation

Alternate Hypothesis
• There is a significant statistical difference between the variances of the two groups

• This is Special Cause of Variation

 F-Test Example

F-Test Two-Sample for Variances

• The p-value is 0.03. Variable

Variable 2
1

• If p-value is < 0.05, null must be rejected. Mean 6.016666667 5.016666667

Variance 3.197666667 0.517666667

• Null hypothesis with 97% confidence is rejected.
Observations 6
6
• The fact that variation could only occur due to
df 5 5
Common Cause of Variation is rejected.

F 6.177076626
• There could be Assignable Causes of Variation or
P(F<=f) one-tail 0.033652302
Special Causes of Variation.

F Critical one-tail 5.050329058

 t-Test

Select 2-Sample
Open MS Excel,
Independent t-
click Data, and
test assuming
click Data
unequal
Analysis.
variances.

In Variable 1
range, select the
data set for
Group A.

In Variable 2 Keep the

range, select the “Hypothesized
data set for Mean Difference”
Group B. as 0.

Click OK.
 t-Test

Group A (Chef 1) Group B (Chef 2)

4.2 4

4.5 4.5

7.2 5

6.1 5.2

8.9 5.3

5.2 6.1
 t-Test

Null Hypothesis
• There is no significant statistical difference between the means of the two groups

• This is Common Cause of Variation

Alternate Hypothesis
• There is a significant statistical difference between the means of the two groups

• This is Special Cause of Variation

H0 : Mean of Group A = Mean of Group B

Ha : Mean of Group A ≠ Mean of Group B

 t-Test

t-Test: Two-Sample Assuming Unequal

Variances
Variable 1 Variable 2

Mean 6.016666667 5.016666667

The p-value of 2-tailed probability testing is 0.24
Variance 3.197666667 0.517666667
Observations 6 6 This value is greater than 0.05
Hypothesized
0
Mean
The null hypothesis is not rejected
df 7
T Stat 1.270798616
Both the groups are statistically same
P(T<=t) one-tail 0.122200546
T Critical one-tail 1.894578605
P(T<=t) two-tail 0.244401092
T Critical two-tail 2.364624252
 t-Test

Jan Feb
360 365
The paired t-test is conducted before and after the
324 325
process to measure:
377 359
336 352
Customer satisfaction before and after improvements 383 397
361 351
369 367
Employee performance before and after training 349 397
301 335
354 338
344 349
329 393
337 370
387 400
378 411
 ANOVA

Used to compare the means of more than two samples

Stands for analysis of variance

Helps in understanding that all sample means are not

equal

Shortlisted samples can further be tested

Generalizes the t-test to include more than two

samples.
 ANOVA

Q
Outlet 1 Outlet 2 Outlet 3
The table shows the takeaway food delivery time of
48 50 49
three different outlets.
Is there any evidence that the averages for the three 49 48 48
outlets are not the same?
48 36 39

53 50 49

A
58 50 34
The null hypothesis will assume that the three
means are equal. If the null hypothesis is rejected, it
50 62 33
would mean that there are at least two outlets that 46 45 57
are different in their average delivery time.
50 47 48

49 51 47

47 44 39
 ANOVA

Select the
Enter the data in ANOVA – single
Select all the
an Excel factor test from
cells for analysis
spreadsheet the Data Analysis
“Toolpack”
 Chi-Square Distribution

Chi-square distribution (χ²-distribution) or Chi-squared

● Is a widely used probability distribution in inferential statistics;
● Needs one sample for the test to be conducted; and
● Has k-1 degrees of freedom and is the distribution of a sum of the squares of k independent
standard normal random variables.
 Chi-Square Distribution

Estimated Population Sample Statistics

Parameters
Chi-Square Distribution
 Chi-Square Distribution

There is a different Chi-square distribution for each of the different numbers of degrees of freedom.
For Chi-square distribution, degrees of freedom are calculated according to the number of rows and
columns in the contingency table.
Hypothesis Testing with Non-normal Data
 Non-Parametric Test

Non-parametric tests do not make any assumptions about a distribution model where the
data could fit.

Non-parametric tests compare groups of medians using the relative ranks of the data
within the groups.
 Non-Parametric Test

Corresponding Parametric
Non-parametric Test Main Characteristics
Tests

1 sample Sign test Test on the median, for non symmetric distribution 1 sample t or Z test

Test on the median for symmetric distribution

1 sample Wilcoxon test 1 sample t or Z test
(pairwise averages)

Mann and Whitney test Test on ranks to compare center of 2 groups 2 samples t or Z test

Test on ranks to compare center of 2 or more groups;

Kruskal-Wallis test based on the Chi Square Distribution
One Way ANOVA
More powerful than Mood’s median test but less
robust to outliers
Test on the overall median; based on the Chi Square
test
Mood’s median test One Way ANOVA
More robust to outliers than the Kruskal-Wallis test,
but less powerful

Freidman Test Test on ranks, based on the Chi Squared distribution Two-way randomized ANOVA

Same rules apply: If p < α reject the Null Hypothesis

 Non-Parametric Test

Mann-Whitney test is a non-parametric test used to compare the center between two
unpaired groups

MANN-WHITNEY TEST—DEFINITION
 Non-Parametric Test

The obtained U has to be equal to or less than this critical value.

The calculated value is not equal to or less than 2. Therefore, there is no statistical difference between
the means of the two groups.
 Non-Parametric Test

The Kruskal-Wallis test is used for testing the source of origin of the samples.

Only way to analyze the variance by ranks

Medians of two or more samples are compared to find

the source of origin

Does not assume the normal distribution of the

residuals

Testing the ratings of a product from three different groups to see if the ratings are the
same or different
 Non-Parametric Test

The Mood’s Median is a non-parametric test used to test the equality of medians from two
or more different populations. The test works if:
• The output (Y) variable is continuous, discrete-ordinal, or discrete count.
• The input (X) variable is discrete with two or more attributes.

Find the number of

Find median of the values in each sample
combined data set that are greater than
the median

Form a contingency Find expected value

table for each cell

Compare Chi-square
Find Chi-square value value to critical Chi-
Square value

Example:
To determine whether temperature changes in the ocean water near a nuclear power plant will have a
significant effect on the animal life in the region, an environmental group places groups of fish in four bowls that
are identical in except for water temperature. Six months later, they measure the weights of the fish.
 Non-Parametric Test

The Friedman Test is a form of non-parametric test that does not make any
assumptions on the origin of the sample.

A marketing company wants to compare the relative effectiveness of three different modes of advertising:

Direct Mail Newspaper Magazine Advertisements.

The company conducts a randomized block design experiment. For 14 customers, the marketing company
used all three modes during a 1-year period and recorded the percentage response to each type of
advertising.
 Non-Parametric Test

1 Sample Sign test is the simplest of all non-parametric tests

that can be used instead of one sample t test

Count the Count the

number of number of Test the
positive negative values
values values

HR of a large company analyzes its payroll to determine whether the company's median
salary differs from the industry average.
 Non-Parametric Test

1 Sample Wilcoxon Test is a non-parametric test.

It is equivalent to 1 Sample t-Test and is more powerful than 1 Sample sign test.

It is used to estimate the population median and compare it to a target or reference value.

It assumes the existing sample is randomly taken from a population.

The median customer satisfaction score of an organization has always been 3.7.
Management wants to see if it has changed. They conducted a survey and got the results
grouped by the customer type

Conclusion:
⮚ If median = 3.7 then Fail to Reject Null
⮚ If median ≠ 3.7 then Reject Null
 Key Takeaways

Hypothesis test is used to prove or disprove a theory or claim by testing

two or more samples or comparing a sample to a defined value.

In deciding to reject or not reject the null hypothesis, we can make two
possible decision errors—Type I and Type II errors.

To calculate the standard sample size for discrete data, the average
population proportion non-defective is ‘p’ and value of α is taken as 5%.

The P-value is the probability that any differences observed are due
to random chance or common cause variation.

A test statistic is used to determine the validity of a hypothesis test.

Non-parametric tests include Mann-Whitney Test, Kruskal-Wallis

Test, Mood’s Median Test , Freidman Test, 1 Sample Sign Test, and 1
Sample Wilcoxon Test.
Knowledge Check
Knowledge
Check If p value is 0.05 and the confidence level is 90%, what would be the hypothesis test
1 conclusion?

A. Fail to reject the null hypothesis

B. Reject the null hypothesis

C. Reject the alternative hypothesis

D. Not enough information

 Knowledge
Check If p value is 0.05 and the confidence level is 90%, what would be the hypothesis test
1 conclusion?

A. Fail to reject the null hypothesis

B. Reject the null hypothesis

C. Reject the alternative hypothesis

D. Not enough information

The correct answer is B

Alpha value is 10%. Since confidence level is 90% and p is smaller than the alpha value, the null hypothesis is rejected
Knowledge
Check An Assembly team desired to see if there was any performance improvement after
2 completing a Six Sigma project. What hypothesis could be used?

A. F-test

B. Two Sample t-test

C. One Sample t-test

D. Paired t-test
 Knowledge
Check An Assembly team desired to see if there was any performance improvement after
2 completing a Six Sigma project. What hypothesis could be used?

A. F-test

B. Two Sample t-test

C. One Sample t-test

D. Paired t-test

The correct answer is D

To see if a process has improved, a paired t-test should be used to compare the before and after improvement state.
Knowledge
Check
Which non-parametric test is similar to a single factor ANOVA?
3

A. Sample Sign

B. Wilcoxon Sign test

C. Mood’s Median

D. Freidman’s Test
 Knowledge
Check
Which non-parametric test is similar to a single factor ANOVA?
3

A. Sample Sign

B. Wilcoxon Sign test

C. Mood’s Median

D. Freidman’s Test

The correct answer is C

Mood’s median is similar to a single factor ANOVA . It is able to test for the difference in medians for more than 2
groups.
Knowledge
Check A team wants to test if a new drug reduced pain in the patients. What would be the
4 Type II error?

A. The new drug really works and team concludes it works

B. The new drug does not work and team concludes it works

C. The new drug really works and team concludes it does not work

D. The new drug does not work and the team concludes it does not work
 Knowledge
Check A team wants to test if a new drug reduced pain in the patients. What would be the
4 Type II error?

A. The new drug really works and team concludes it works

B. The new drug does not work and team concludes it works

C. The new drug really works and team concludes it does not work

D. The new drug does not work and the team concludes it does not work

The correct answer is C

Type II error fails to reject the null hypothesis when it is false. Therefore, if the null hypothesis is the drug, it does not
cause a difference in pain levels.
Knowledge
Check Which hypothesis test is used to compare the variance for two groups with normal
5 data?

A. Z Test

B. F Test

C. t-Test

D. χ2 Test
 Knowledge
Check Which hypothesis test is used to compare the variance for two groups with normal
5 data?

A. Z Test

B. F Test

C. t-Test

D. χ2 Test

The correct answer is B

The F test is used to compare variance for two or more groups with normal data.
Knowledge
Check The population standard deviation for the time, to resolve customer problems, is 20 hours.
What should be the size of a sample that can estimate the average problem resolution time
6 within ± 2 hours tolerance with 95% confidence?

A. 385

B. 384

C. 386

D. 400
 Knowledge
Check The population standard deviation for the time, to resolve customer problems, is 20 hours.
What should be the size of a sample that can estimate the average problem resolution time
6 within ± 2 hours tolerance with 95% confidence?

A. 385

B. 384

C. 386

D. 400

The correct answer is A

Knowledge
Check After conducting a hypothesis test at 5% significance level the test statistic was 30, the
7 critical value was 25, and the p-value was .10. What can we conclude?

A. Reject the null hypothesis

B. Fail to reject the null hypothesis

C. Not enough information to decide

D. Something went wrong with the test

 Knowledge
Check After conducting a hypothesis test at 5% significance level the test statistic was 30, the
7 critical value was 25, and the p-value was .10. What can we conclude?

A. Reject the null hypothesis

B. Fail to reject the null hypothesis

C. Not enough information to decide

D. Something went wrong with the test

The correct answer is D

The p-value indicates failure to reject the null and the test statistic indicates rejection of the null. Therefore, there is a
discrepancy because both methods should always lead to the same conclusion.
Knowledge
Check
Which non-parametric test is similar to a 1 sample t test?
8

A. Freidman

B. Kruskal-Wallis

C. Wilcoxon Signed Rank

D. Mood’s Median
 Knowledge
Check
Which non-parametric test is similar to a 1 sample t test?
8

A. Freidman

B. Kruskal-Wallis

C. Wilcoxon Signed Rank

D. Mood’s Median

The correct answer is C

Wilcoxon Signed Rank test is is similar to a 1 sample t test. It is also known as the 1 Sample Wilcoxon Test.