Mit

Hypothesis Tests:
Continuous Data
Hypothesis Tests: Continuous Data
Objectives
• Evaluate the difference between a process (population) mean and a target value
using 1-sample and confidence intervals.
• Assess the power of a hypothesis test using power analysis.
• Test for a difference between two population means using a 2-sample t-test.
• Test for a difference between paired observations using a paired t-test.
• In this laboratory
Examples and Exercises Purpose
Hypothesis Testing and Confidence Intervals
Example 1: Filling Cereal Boxes Evaluate the difference between a mean and a
target value using a 1-sample t-test
Exercise C: Filling Cereal Boxes: Evaluate the difference between a mean and a
Minitab Assistant target
Power and Sample Size
Example 2: Evaluating Power Assess the power of a t-test

1-Sample t-Test
Example 3: Increased Power Demonstrate the effect of sample size on a hypothesis test.
Exercise D: Detecting Changes Determine t-test sample sizes and detectable differences.
in Bearing Diameter Evaluate the difference between a mean and a target.
Exercise E: Investigating Determine the power of a t-test for a given sample sizes.
Supplier Quality Evaluate the difference between a mean and a target.
Power and Sample Size for 2-
2-Sample t-
t-Test
Example 4: Sample Size for Determine the required sample size for a 2-sample t-test.
Supplier Comparison
2-Sample t-Test
Example 5: Plastic Strength Evaluate the difference between means using an independent 2-
sample t-test.
Exercise F: Anode Height Determine the required sample size for a 2-sample t-test.
Use the test to evaluate the difference between means.
Paired t-Test
Example 6: Parking Cars Evaluate the difference between paired observations using a
paired t-test.
Exercise G: Comparing Calipers Evaluate the difference between paired observations using a
paired t-test. Determine power for a paired t-test.

Example 1: Filling Cereal Boxes
Problem
• A cereal manufacturer wants to determine whether the box-filling
process is on target.
• The target fill weight for cereal boxes is 365 grams.
Data collection
• Engineers choose six boxes of cereal random, weigh them, and use
the sample data to estimate the mean of the population (the process
mean).
Tools
Variable Description
• Descriptive Statistics
BoxWeigh Cereal box weight (g)
• 1-Sample t
• Normality Test
• StatGuide
• Individual Value Plot
Data set
• L01_Dataset_CerealBx.MPJ

Displaying descriptive statistics
• Use descriptive statistics to summarize important features of the
data.
• In particular, descriptive statistics provide useful information about
the location and variability of the data.
Display Descriptive Statistics
1. Open CerealBx.MPJ.
2. Choose Stat > Basic Statistics > Display Descriptive Statistics.
3. In Variables, enter BoxWeigh.
4. Click OK.

Interpreting your results
• The statistics indicate that the sample mean is 366.70 gram.
• This is slightly above the target weight of 365 grams.
• The one sample t-test will compare this difference (1.7 grams) to the
variation in the data.
• While the standard deviation of the sample data is 2.40 grams, the
one sample t-test will use the SE Mean.
Descriptive Statistics: BoxWeigh
Variable N N* Mean SE Mean StDev Minimum Q1 Median Q3 Maximum

BoxWeigh 6 0 366.70 0.981 2.40 363.43 364.25 367.22 368.31 370.13

Using the StatGuide
• We can see that the SE Mean is simply the standard deviation divided
by the square root of the number of data points and represents the
dispersion or variation in the distribution of sample means.
• The one-sample t-test uses the distribution of the sample mean (not
the distribution of the data) for the analysis.
• Therefore, the standard error of the mean will be used as the
estimate of variation for the t-test and confidence interval.
StatGuide
1. In the Session window, place your cursor anywhere in the
Descriptive Statistics output.
2. Click on the ∑ icon.
3. Select Standard error of the mean (SE Mean) as shown below.

Hypothesis testing
What is a hypothesis test
• A hypothesis test uses sample data to test a hypothesis about the
population from which the sample was taken.
• The 1-sample t-test is one of many procedures available for
hypothesis testing in Minitab.
• Using information from a sample to make a conclusion about a

population is known as statistical inference.
Hypothesis testing
When to use a hypothesis test
• Use a hypothesis test to make inferences about one or more
populations when sample data are available.
Why use a hypothesis test
A hypothesis test answers questions such as:
• Is a process centered correctly?
• Is the product from one supplier better than the product from
another?
• Do treatment groups in an experiment differ?
Hypothesis testing
For example,
• Is the mean width of paper stock the desired 8.5 inches?
• Does fuel from one supplier burn more cleanly than fuel from another
supplier?
• Do customers prefer one formulation of soft drink over another?
1-sample t-
t-test
What is a 1-sample t-test
• Use 1-sample t-test to determine whether (the population mean) is
equal to the hypothesized mean.
• The test uses the standard deviation of the sample to estimate (the
population standard deviation).
• If the difference between the sample mean and the hypothesized
mean is large relative to the variability of the sample mean, then is
unlikely to be equal to the hypothesized mean.
1-sample t-
t-test
When to use a 1-sample t-test
• Use a 1-sample t-test with continuous data from a single random
sample.
• The test assumes the population is normally distributed.
• However, the test is robust to violations of this assumption, provided
the observations are collected randomly and the data are continuous,
unimodal, and reasonably symmetric (see[1]).
1-sample t-
t-test
Why use a 1-sample t-test
A 1-sample t-test answers questions such as:
• Is a process on target?
• Does a key characteristic of a supplier’s material have the desired
mean value?
For example,
• Is the mean width of razor blades higher or lower than target?
• Is the mean strength of the bolts from a supplier lower than the
required minimum?
Testing the null hypothesis
• The manufacturer needs to determine whether the mean weight for
the packaging process differs significantly from the target weight of
365 grams. In statistical terms, the process mean is the population
mean, or (mu).

Statistical hypotheses
Either is equal to 365 or it is not. You can state these alternatives as
two hypotheses:
• The null hypothesis ( ): is equal 365 grams.
• The alternative hypothesis ( ): is not equal to 365 grams.
• Because the engineers cannot feasibly measure every box in the
population, they can never know for sure which hypothesis is correct.
• However, an appropriate hypothesis test can help the make an
informed decision.
• For the box weight data, the appropriate test is a 1-sample t-test.

1-Sample t
1. Choose Stat > Basic Statistics >
1-Sample t.
2. Complete the dialog box as
shown below.
3. Click OK.
The logic of hypothesis testing

All hypothesis test follow the same steps:
1. Assume is true.
2. Determine how different the sample is from what you expected
under the above assumption.
3. If the sample is sufficiently unlikely under the assumption that is
true, then reject .

For example,
• The t-test results indicate that the sample mean is 366.705 grams.
• The test answers the question, “If is equal to 365 grams, how likely
is it to see a sample mean this different (or even more different)?”
• The answer is given as a probability value (P), which for this test is
equal to 0.143.
The t-statistic (1.74) is:

t = (sample mean – hypothesized mean) / SE Mean
• Where SE Mean is the standard error of the mean (a measure of
variability).
• For a given sample size, as the value of t increases, the p- value
decreases.
One-Sample T: BoxWeigh
Test of μ = 365 vs ≠ 365
Variable N Mean StDev SE Mean 95% CI T P

BoxWeigh 6 366.705 2.403 0.981 (364.183, 369.226) 1.74 0.143
Making a decision
To make a decision, choose the significance level, (alpha), before the
test:
• If P is less than or equal to , reject .
• If P is grater than , fail to reject . (Technically, you never accept
, you simply fail to reject it.)

• A typical value for is 0.05, but you can choose higher or lower
values depending on the sensitivity required for the test and the
consequences of incorrectly rejecting the null hypothesis.
• Assuming an -level of 0.05 for the box weight data, not enough
evidence is available to reject because P (0.143) is greater than .
What’s next
• Check the assumption of normality.
Test of μ = 365 vs ≠ 365

BoxWeigh 6 366.705 2.403 0.981 (364.183, 369.226) 1.74 0.143
Testing the assumption of normality

• The 1-sample t-test assumes the data are sampled from a normally
distributed population.
• Use a normality test to determine whether the assumption of
normality is valid for these data.
Note
• The t-test is robust to violations of the normality assumption,
provided that observations are collected randomly and the data are
continuous, unimodal, and reasonably symmetric.
Normality Test
Normality Test.
shown below.
3. Click OK.

• Use the normal probability plot to verify that the data do not deviate
substantially from what is expected when sampling from a normal
distribution.
• If the data come from a normal distribution, the points will roughly follow the
fitted line.
• If the data do not come from a normal distribution, the points will not follow
the line.
Anderson-Darling normality test
• The hypotheses for the Anderson-Darling normality test are:
• : Data are from a normality distributed population.
• : Data are not from a normality distributed population.
• The p-value from the Anderson-Darling test (0.599) assesses the
probability that the data are from a normally distributed population.
• Using an of 0.05, you cannot detect significant departure from
normality.

Conclusion
• Based on the plot and the test, assume that the data are from a
normally distributed population.
• When the sample size is small, the assumption of normality is often
based on experience with historical data.
Note
• When data are not normally distributed, you may be able to
transform them using a Box-Cox transformation or use a
nonparametric procedure such as the 1-sample sign test.
Probability Plot of BoxWeigh
Normal
99
Mean 366.7
StDev 2.403
95 N 6
AD 0.248
90
P-Value 0.599
80
70
Percent
60
50
40
30
20
10
5
1
360 362 364 366 368 370 372
BoxWeigh
Confidence intervals
What is a confidence interval
• A confidence interval is a range of likely values for a population
parameter (such as ) that is based on sample data.
For example:
• With a 95% confidence interval for , you can be 95% confidence that
the interval contain .
• In other word, 95 out of 100 intervals will contain upon repeated
sampling.
When to use a confidence interval
Use confidence intervals to:
• Make inferences about one or more populations from sample data.
• Quantify the precision of the estimate of a population parameter,
such as .
Why use a confidence interval
A confidence interval often answer the same questions as a hypothesis
test:
• Is on target?
• How precise is the estimate of ?
• How low or high might be?
For example:
• Is the mean length of a part longer than 5 cm?
• What is the range of likely values for the mean weight of bags of
coffee beans?
Using the confidence interval
• The Session window results for 1-Sample t include values for the
upper and lower bounds of the 95% confidence interval.
• Obtain a graphical representation of the interval by selecting
Individual value plot in the One-Sample t: Graphs subdialog box.
Using the confidence interval

1-Sample t
1. Choose Stat > Basic Statistics > 1-Sample t.
2. Click Graphs.
3. Complete the dialog cox as shown below.
4. Click OK in each dialog box.

Confidence interval
• The confidence interval is a range of likely values for .
• Minitab displays the interval graphically as a blue line on the
individual value plot.
• Under repeated sampling from the same population, the confidence
intervals from about 95% of the samples would include .
• Thus, for any one sample, you can be 99% of the data; this is a
common misconception.
Individual Value Plot of BoxWeigh

(with Ho and 95% t-confidence interval for the mean)
_
X
Ho
363 364 365 366 367 368 369 370 371

BoxWeigh
Hypothesis test
• The middle tick mark:
, represents the mean of the sample and the red circle.
• labeled X
• labeled H , represents the hypothesized population mean (365 g).
• You can be 95% confident that the process mean is at least 364.2 g,
and at most 369.2 g.

Use the confidence interval to test the null hypothesis:
• If H is outside the interval, the p-value for the hypothesis test will be
less than 0.05.
• You can reject the null hypothesis at the 0.05 -level.
• If H is inside the interval, the p-value for the hypothesis test will be
greater than 0.05.
• You cannot reject the null hypothesis at the 0.05 -level.
• Because H falls within the confidence interval, you cannot reject the
null hypothesis.
• Not enough evidence is available to conclude that is different from
the target of 365 grams at the 0.05 significance level.
Test of μ = 365 vs ≠ 365

BoxWeigh 6 366.705 2.403 0.981 (364.183, 369.226) 1.74 0.14
Individual Value Plot of BoxWeigh

_
X
Ho
363 364 365 366 367 368 369 370 371

BoxWeigh
Final considerations
Summary and conclusions
• Based on the sample data, you cannot reject the null hypothesis at
the 0.05 -level.
• Not enough evidence is available to suggest that the process mean fill
weight is different from 365 gram.
• Like the t-test, the 95% confidence interval does not provide enough
evidence to reject the null hypothesis that the population mean for
cereal box weights is 365 grams.
• However, the confidence interval provides extra information that with
95% confidence, the process mean is at least 364.2 g and at most
369.2 g.
Hypotheses
A hypothesis test begins with two opposing hypotheses.
The null hypothesis ( ):

• Usually states that some property of a population (such as the mean)
is not different from a specified value or from that of another
population.
• Is assumed to be true until sufficient evidence indicates the contrary.
• Is never proven true; you simply fail to disprove it.
Hypotheses
The alternative hypothesis ( ):
• States that the null hypothesis is wrong.
• Can also specify the direction of the difference.
Significance level
Choose the -level before conducting the test.
• Increasing increases the chance of detecting a difference, but it also
increases the chance of rejecting when it is actually true (a Type I
error).
• Decreasing decreases the chance of making a Type I error, but also
decreases the chance of detecting a difference when one actually
exists (a Type II error).
Assumptions
Each hypothesis test is based on one or more assumptions about the
data. If these assumptions are not met, the conclusions may not be
correct.
When using a 1-sample t-test:

• The sample must be random.
• Sample data must be continuous.
• Sample data are from a normally distributed population.
Assumptions
• The t-test is robust to violations of the normality assumption,
provided that observations are collected randomly and the data are:
• Continuous
• Unimodal
• Reasonably symmetric
Confidence interval
The confidence interval provides a likely range of values for (or other
population parameters). You can conduct a 2-tailed hypothesis test
(alternative hypothesis of ≠) using a confidence interval.
For example:
• If the test value is not within a 95% confidence interval, you can reject
at the 0.05 -level.
• Likewise, if you construct a 99% confidence interval and it does not
include the test mean, you can reject at the 0.01 -level.
Exercise C: Filling Cereal Boxes: Minitab Assistant
Problem
• A cereal manufacturer wants to determine whether the box-filling
process is on target.
• The target fill weight for cereal boxes is 365 grams.
Data Collection
• Engineering choose six boxes of cereal at random, weigh them, and
use the sample data to estimate the mean of the population (the
process mean).
Exercise C: Filling Cereal Boxes: Minitab Assistant

Instructions
1. Determine whether the process is on target by conducting a one-
sample t-test ( : ≠ 365) using the Minitab Assistant by choosing
Assistant > Hypothesis Tests > 1-Sample t.
Data Set
• L01_Dataset_CerealBx.MPJ
BoxWeigh Cereal box weight (g)
Example 2: Evaluating Power
Problem
• The engineers are concerned about the results of the fill weight
analysis because of its small sample size
• They decide to conduct a power analysis to determine whether they
collected enough sample data to detect a difference.
• They want to be sure the process mean fill weight does not differ
from the target weight of 365 grams by more than 2.5 grams.

Example 2: Evaluating Power
Data Collection
• The engineering base the power analysis on the results of the t-test
from the Filling Cereal Boxes example.
Tools
• Power and Sample Size > 1-Sample t.
Power analysis
What is power analysis
• Power is the ability of a test to detect a difference when one exists.
• A hypothesis test has the following possible outcomes:
Null hypothesis
Decision True False
Fail to reject Correct decision probability = 1- Type II error probability =
Reject Type I error probability = Correct decision probability = 1-
(Power)
Power analysis
• The power of the test is the probability that you will correctly reject
the null hypothesis, given that the null hypothesis is false.
• Use a power analysis to determine how much power a test has or to
design a new test with adequate power.
Power analysis
When to use power analysis
Use a power analysis:
• Before collecting data, to determine the sample size.
• After collecting data, to evaluate the power to detect a difference.
Why use power analysis
A power analysis answer question such as:
• Is the sample size large enough?
• How large a difference can the test detect?
• Is the test powerful enough to give credibility to its conclusion?
Power analysis
For example,
• How many samples are needed to determine whether the mean
thickness of paper is greater than 0.04 mm?
• How large a difference will be detectable between the mean strength
of new steel beams and a historic mean, using eight samples?
• Can you trust the conclusion from a t-test that indicates the strengths
of two glue formulations are not different?
Determining the difference parameter
Values
To estimate power, you must specify values for any two of the following
parameters of the test; Minitab calculates the remaining parameter.
• Sample sizes – the number of observations in the sample.
• Differences – a meaningful shift away from the target that you are
interested in detecting with high probability.
• Power values – the power (probability of rejecting when it is false)
that you would like the test to have.
Determining the difference parameter

Differences
To choose a meaningful difference amount, you must determine a shift
that will lead to unacceptable increase in the defect rate.
In this example, assume:
• The data are from a normally distributed population.
• The specification limits are 360 and 370.
• The graph at below a shift of 2.5 grams to the high side of the target.
• If such a shift occurred, the number of defects above the upper
specification limit of 370 would be unacceptable.
Determining power
• Using power analysis, evaluate how much you can rely on the results
of the cereal-fill-weight analysis.
Values
• If you enter more than one value for a parameter, Minitab performs
the calculations separately for each value.
Determining power
Standard deviation
• Because variability in the data affects the power of a test, you must
provide an estimate of .
• Use a historical estimate or the sample standard deviation.
• For the cereal data, the standard deviation (2.403) comes from the t-
test results.
Determining power
Power and Sample Size for 1-Sample t

1. Choose Stat > Power and
Sample Size > 1-Sample t.
shown below.
3. Click OK.

• With 6 observations, a standard deviation of 2.403, and an of 0.05,
the power is only 0.537662.
• Hence, if is off target by 2.5 grams, the chance of detecting it with a
sample of size 6 is 53.77%
• Therefore, if the process mean is off target by 2.5 grams, there is a
46% chance that you will not detect it with a sample size of 6.
1-Sample t Test
Testing mean = null (versus ≠ null)

Calculating power for mean = null + difference
α = 0.05 Assumed standard deviation = 2.403
Sample
Difference Size Power
-2.5 6 0.537662
2.5 6 0.537662

• The power curve allows you to see the probability of detecting a
difference (power) for a variety of differences and optionally for a
variety of sample sizes.
• The specific differences entered in the dialog box are indicated on the
plot with black points.
• In this example, the difference represents a shift in the process mean
from target, in grams.
• A negative difference corresponds to a below target process mean, a
positive difference corresponds to an above target process mean
• When the process mean is one gram above or below the target, the
power to detect the difference with a sample size of 6 is very low.

• When the process mean is two grams above or below the target, the
power to detect a difference with a sample size of 6 is close to 0.40,
still fairly low.
• Even the power at plus or minus three grams is unacceptable by most
standards.
• One way to increase power is to increase sample size.
Power Curve for 1-Sample t Test
1.0
Sample
Size
6
0.8 Assumptions
α 0.05
StDev 2.403
Alternative ≠
0.6
Power
0.4
0.2
0.0
-5.0 -2.5 0.0 2.5 5.0
Difference

What’s next
• Determine the sample size required to achieve adequate power.
• How many observations do you need to have an 80 % chance of
detecting a shift from the target of at least 2.5 grams?
• How many observations do you need for an 85%, 90% or 95% chance
of detecting this difference?
What’s next
• Determine the sample size required to achieve adequate power.
• How many observations do you need to have an 80 % chance of
detecting a shift from the target of at least 2.5 grams?
• How many observations do you need for an 85%, 90% or 95% chance
of detecting this difference?
Determining power
• With 6 observations, the power of the test was only 0.5377.
• To have a better chance of detecting a difference, increase the power
of the test to at least 0.80 by increasing the sample size.
• Calculate the sample sizes that are required to achieve power levels
of 0.80, 0.85, 0.90, and 0.95.
Determining power
Power and Sample Size for 1-Sample t

shown below.
3. Click OK.

• To have a power of at least 0.80 – to detect a difference of 2.5 grams
at the 0.05 -level – requires a sample size of 10.
• Because sample size must be an integer, the Actual Power of the test
with 10 observations (0.832695) is slightly greater than the Target
Power.
Additional observations give you more power:

• With 11 observations, the power is 0.873928.
• By doubling the sample size from 6 to 12 boxes, the chance of
detecting a 2.5-gram difference increases from 53.77% to 90.58%.
• As you increase the number of observations, the power of the test
increase and the test is able to detect smaller shifts in the mean.
• If the power is too high (say, above 99%), the test may detect smaller
shifts that do not have any practical importance.

1-Sample t Test
Testing mean = null (versus ≠ null)

Calculating power for mean = null + difference
α = 0.05 Assumed standard deviation = 2.403
Sample Target
Difference Size Power Actual Power
-2.5 10 0.80 0.832695
-2.5 11 0.85 0.873928
-2.5 12 0.90 0.905836
-2.5 15 0.95 0.962487
2.5 10 0.80 0.832695
2.5 11 0.85 0.873928
2.5 12 0.90 0.905836
2.5 15 0.95 0.962487
• The power curves for the larger sample sizes increase more rapidly as
the difference moves away from zero.
• The power is reasonably good for detecting a difference of 2.5 grams
with sample sizes between 10 and 15.

1.0
Sample
Size
10
0.8 11
12
15
Assumptions
0.6 α 0.05
Power
StDev 2.403
Alternative ≠
0.4
0.2
0.0
-3 -2 -1 0 1 2 3
Difference
• The results for the first test of cereal fill weights may have been
insignificant simply because the power was too low.
• Based on the number of observations (6), the difference to detect
(2.5), and the variability in the process, the test had a power of only
0.537662.
• Using a larger sample will give more power, enabling you to detect a
difference if one exists.
Additional considerations
• To ensure that the test will have sufficient power, conduct a power
analysis prior to collecting data.
• To increase the power of a test:
• Increase the sample size.
• Decrease the variability that is not attributed to the effect of interest.
• Increase (although this increases the chance of Type I error).
• Higher power means a greater probability of detecting a difference.
• However, it also increases the chance of detecting small effects that
may not be of practical interest.
• Use process knowledge to determine the smallest difference worth
detecting and the optimal level of power for a test.
1-Sample t-
t-Test
Example 3: Increased Power
Problem
• The results of the first power analysis suggest that a larger sample
would be useful in evaluating the process.
• Six observations did not have enough power to detect a 2.5-gram
difference.
Data collection
• Engineers randomly select 12 boxes of cereal and weigh them.
1-Sample t-
t-Test
Example 3: Increased Power
Tools
• 1-Sample t
• Normality Test
Data set
L01_Dataset_CerealBx.MPJ
MoreObs Cereal box weight (g)
Adding a footnote to a graph

• When doing a similar analysis on multiple worksheets (such as in the
example Technical Support).
• It can be helpful to have the worksheet name stamped on the graphs.
• This allows you to quickly differentiate between the same type of
graph using different data sets.
Adding a footnote to a graph
Add a footnote
1. Choose Tools > Options
2. In the left pane, double-click Grahpics.
3. Under Grahpics, double-click Annotation.
4. Under Annotation, click My Footnote.
5. Under Information to include in my footnote, check Worksheet
name.
6. Click OK.

• Analyze the new sample to determine whether the process mean is
different from 365 grams.
1-Sample t
1. Open CerealBx.MPJ
2. Choose Window > MoreData.MTW.
3. Choose Stat > Back Statistics > 1-
Sample t.
4. Complete the dialog box as shown
below.
5. Click Graphs.
6. Check Individual value plot.

• The individual value plot is clearly from the new data set with 12
observations because the worksheet name is stamped in the lower
left corner.
• The individual value plot illustrates the findings of the test:
• The sample mean () is higher than the target value.
• The target value ( ) is outside the confidence interval.
Conclusion
• The difference between the process mean and the target value of 365
grams is significant at the 0.05 –level.
Individual Value Plot of MoreObs
_
X
Ho
362 363 364 365 366 367 368 369 370

MoreObs

T
The t-statistic (2.75) is:
• t = (sample mean – hypothesized mean) / SE Mean
• Where SE Mean is the standard error of the mean (a measure of
variability).
• For a given sample size, as the value of t increases, the p-value
decreases.
P
The t-test indicates that the difference between the process mean and
the target of 365 grams is significant at the 0.05 –level:
• The p-value (0.019) is less than (0.05).
• The 95% confidence interval does not include the target value.
One-Sample T: MoreObs
Test of μ = 365 vs ≠ 365

MoreObs 12 366.636 2.060 0.595 (365.327, 367.945) 2.75 0.019
• The 1-sample t-test assumes the data are sampled from a normally
distributed population.
normality is valid.

Normality Test
Normality Test.
shown below.
3. Click OK.
• Use the normal probability plot to verify that the data do not deviate
substantially from what is expected when sampling from a nor mal
distribution.
• If the data come from a normal distribution, the points should
roughly follow the fitted line.
• If the data do not come from a normal distribution, the points likely
will not follow the line.

The hypotheses for the Anderson-Darling normality test are:
• : Data are from a normally distributed population.
• : Data are not from a normally distributed population.
• The p-value from the Anderson-Darling test (0.545) assesses the
probability that the data are from a normally distributed population.
• At an of 0.05, not enough evidence exists to suggest that the data
are not from a normal population.
Conclusion
• Based on the plot and the test, you can assume that the fill weights
are normally distributed and a t-test is appropriate for testing the
mean. Probability Plot of MoreObs
Normal
99
Mean 366.6
StDev 2.060
95 N 12
AD 0.292
90
P-Value 0.545
80
70
Percent
60
50
40
30
20
10
5
1
362 364 366 368 370 372
MoreObs
• Based on a sample size of 12, = 0.05, and a p-value of 0.019, reject
and conclude that the process mean is not equal to the target
value of 365 grams.
The sample size played an important role in evaluating this process
mean:
• When the sample size was 6, the test found no statistical difference.
• When the sample size was 12, the test found a statistically significant
difference.
• Whenever possible, investigate the power and sample size before
collecting the data.
• Otherwise, if you discover later that the power is inadequate, it may
be difficult to obtain additional samples under the same conditions as
the original samples.
• A 1-tailed test provides greater power to detect the specified
difference than does a 2-tailed test.
• However, a 1-tailed test cannot detect a difference in the opposite
direction specified by the alternative hypothesis.
Detecting Changes in Bearing Diameter

Exercise D: Detecting Changes in Bearing Diameter
Problem
• A ball bearing manufacturer is concerned that bearing diameter has
shifted away from the target of 0.5 cm.
• Engineers consider a change of 0.01 cm important enough to warrant
adjusting the equipment.
• The standard deviation of the diameters is typically 0.004 cm.
• Use an -level of 0.05 for all tests.
Data collection
• Engineers randomly select and measure 10 bearings.

Instructions
1. Use Stat > Power and Sample Size > 1-Sample t to calculate the
sample size needed to detect a 0.01 cm difference with a power of
0.85.
2. Calculate the difference detectable with a power of 0.90 for five
and ten observations.
3. Determine whether the process is on target. Conduct a one-sample
t-test ( : μ ≠ 0.05) and display a boxplot of the data.
4. Test the sample for normality.
Data set
• L01_Dataset_Bearings.MPJ
Bearings Bearing diameter (cm)
Investigating Supplier Quality

Exercise E: Investigating Supplier Quality
Problem
• A cheese manufacturer suspects that one of their milk suppliers is
adding water to its milk to increase profits.
• Adding water to milk raises its freezing temperature, which is
normally -0.545 ℃.
Data collection
• Analysts measure freezing temperature for 10 randomly selected
batches of the supplier’s milk.
Instructions
1. With a sample size of 10, determine the power to detect 0.008, 0.010,
and 0.012 differences. In Options, under Alternative Hypothesis, choose
Greater than. Assume a standard deviation of 0.01.
2. Use Display Descriptive Statistics to summarize the sample.
3. Use a 1-sample t-test to determine whether the mean is greater than -
0.545 and display an individual value plot of the data. In Options, under
Alternative Hypothesis, choose Mean > hypothesized mean.
4. Check the data for normality.
5. If all the samples came from one shipment of milk, would that change
the analysis and conclusion?

Exercise E: Investigating Supplier Quality
Data set
• L01_Dataset_Cheese.MPJ
FrzTemp Freezing temperature for batch (℃)
2-Sample t-
t-Test
What are power and sample size for a 2-sample t-test
In a two sample t-test:
• Power is the probability that you will detect a difference between the
two means when they actually differ.
• Sample size is the number of samples per group that you need to
achieve a specified power.

2-Sample t-
t-Test
When to use power and sample size for a 2-sample t-test
Use the analysis either:
• Before collecting data, to determine the sample size
• After collecting data, to evaluate the power to detect a difference
between means
2-Sample t-
t-Test
Why use power and sample size for a 2-sample t-test
Power and sample size can determine:
• The sample size per group that you need to detect a difference
between means with a specified power.
• The power of a test to detect a difference between means based on a
specified sample size
• The size of a detectable difference with a specified power and sample
size.
Determining sample size for 2-

2-sample t-
t-test
Sample sizes
• Do not enter a sample size when you want to determine the sample
size.
Values of difference and standard deviation
• The power of a test depends on the difference you want to detect
relative to the standard deviation.
• To detect a 1-standard deviation (or 1-sigma) difference, enter a
difference of 1 and -1, and a standard deviation of 1.
2-sample t-
t-test
Power values
• Enter the desired power value(s).
• Power values higher than 0.80 are typically considered acceptable.

2-sample t-
t-test
Power and sample size for a 2-sample t-test
1. Choose File > New, select
Minitab Project, and then
click OK.
shown below.
4. Click OK.
Sample size
• A sample size of 17 plastic samples per supplier gives a power of
0.807037.
• A sample size of 23 plastic samples per supplier gives a power of
0.912498.

Target power versus actual power
• The target power is the power value(s) that you specified in the dialog
box.
• Because the sample size must be an integer, you will seldom obtain
the exact target power.
• The actual power is for the smallest sample size resulting in a power
at least as large as the target power.
2-Sample t Test
Testing mean 1 = mean 2 (versus ≠)

Calculating power for mean 1 = mean 2 + difference
α = 0.05 Assumed standard deviation = 1
Sample Target
Difference Size Power Actual Power
-1 17 0.8 0.807037
-1 23 0.9 0.912498
1 17 0.8 0.807037
1 23 0.9 0.912498
The sample size is for each group.

Power Curve
• In this example, the difference represents the number of standard
deviations between the supplier A process mean and the supplier B
process mean.
• The power is reasonably good at detecting a difference of 1 standard
deviation with sample size between 17 and 23.
Power Curve
1.0
Sample
Size
17
0.8 23
Assumptions
α 0.05
StDev 1
0.6 Alternative ≠
Power
0.4
0.2
0.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Difference
• The company should use between 17 and 23 samples per supplier to
detect a difference of 1 standard deviation (1) between supplier
means with reasonably high power.
• The larger the sample size, the higher the power of the test.
• Collect enough observations to achieve adequate power, but not so
many that you waste time and money on unnecessary
measurements.
• Sometimes a difference and standard deviation are not readily
available when computing power.
• Because the power calculation is based on the ratio of the difference
to the standard deviation, you do not need to specify them if you are
willing to accept the answer in terms of the number of standard
deviations that separate the two population means.
• For example, if the difference is 2 and the standard deviation is 2, the
power will be the same as if the difference is 1 and the standard
deviation is 1, because both are 1 difference.
• As shown in this example, you can specify a 1 difference by setting
the difference and the standard deviation equal to one another.
• Similarly, you can specify a 2 difference by specifying a difference
that is two times the standard deviation.
2-Sample t-
t-Test
Example 5: Plastic Strength
Problem
• A calculator manufacturer is selecting a plastics supplier.
• Based on the results of the Power and Sample size, the manufacturer
decides to use a sample size of 20 from each supplier to compare the
breaking strength of the suppliers of plastic pellets.
2-Sample t-
t-Test
Data collection
• A machine presses pellets from randomly selected batches of plastic
into wafers of uniform thickness.
• Technicians record the breaking strength (in Newtons) for each wafer.
2-Sample t-
t-Test
Tools
• Dotplot
• Graphical Summary
• 2-Sample t
• 2 Variances
• Layout Tool
• Send to Microsoft PowerPoint
2-Sample t-
t-Test
Data set
• L01_Dataset_Plastic.MPJ
SupplrA Breaking strength of wafers prepared from supplier A
plastic (Newfons)
SupplrB Breaking strength of wafers prepared from supplier B
plastic (Newfons)
Creating dotplots
• Graphs are an important preliminary step in data analysis.
• They enable you to examine the data and to identify patterns,
relationships, and potential problems.
Dotplots
• Use dotplots to examine the shape and spread of sample data.
• Begin the analysis by examining a dotplot of the breaking strength
data for each supplier.
Creating dotplots
Dotplots
1. Open Plastic.MPJ
2. Choose Graph > Dotplot
3. Under Multiple Y’s, choose
Simple, then click OK.
shown below.
5. Click OK.
• The graph shows the strength data from both suppliers.
• On average, the breaking strength from supplier A appears higher
than from supplier B.
• An unusually low value for supplier B appears on the far left side of
the plot.

Correcting the outlier
• Hover the cursor over the supplier B value of 0.
• The data tip identifies the value as row 10 in the worksheet.
• After checking with the person who entered the data, you discover
that the strength information for this wafer is missing due to operator
error.
• Instead of entering 0, you should enter an asterisk (*) to indicate that
the value is missing.
Dotplot of SupplrA, SupplrB
SupplrA
SupplrB
0 25 50 75 100 125 150 175
Data
Removing incorrect values

The strength data in row 10 for supplier B should be recorded as
missing, not as a zero.
Delete
• Use the Delete key to clear the contents of a cell.
• After you clear the cell, notice that the length of the supplier B
column is still 20 observations with 1 missing value and the
association among the data in the columns is maintained.
Missing value symbol (*)
• Minitab inputs an asterisk in the cell to indicate data are missing for
numeric or date/time data; for text data, Minitab leaves the cell
blank.
• Most commands exclude all rows with a missing value from analysis.
• You can display the number of missing values for a variable using
Display Descriptive Statistics.

Show Info
1. Click in the Project Manager toolbar.
2. In the SupplrB column, highlight the cell in column 2 and row 10.
3. Press Delete.
Updating a graph
• The icon in the upper left corner of a Graph window indicates the
current status of the graph.
• When data in the associated worksheet are modified, the icon turns
yellow.
• The table to the right explains the graph icon shapes and colors.
Updating a graph
Icon Indicates the graph is… Because…
Green check Current You manually updated the graph, have

specified Minitab to automatically update,
or haven’t changed the graph data.
Yellow triangle Not current, and data have Minitab is waiting for you to manually
changed update the graph.
Red “X” On hold (only occurs when the You are changing the data and some
graph is set to update condition exists that prevents the update.
automatically) Minitab waits for the condition to be
resolved.
Red “X” in a circle Not current, and is ineligible for The graph is not a core graph, or cannot
updating otherwise be updated.
Updating a graph
• Row 10 of the strength data for supplier B has been changed from a
zero to missing.
• Notice that the icon in the upper left corner of the graph is now a
yellow triangle, indicating that the data in the worksheet has been
modified, but the graph does not reflect the change.
• Graph update choices are:
Update Graph Now Updates an individual graph for current data changes
Update Graph Automatically Updates the selected graph and all other graphs
created with the same dialog box or command
whenever the worksheet changes
Update All Graphs Now (under Editor > Update Updates all open graphs for current data changes
or Window menu)
Updating a graph
1. To choose the dotplot, click in the Project Manager toolbar.
Click the graph to make it the active window.
2. Choose Editor > Update > Update Graph Now, or right-click the
graph and choose Update Graph Now, as shown below.
• The graph updates to reflect the modified data
• In the graph on the right, the x-axis automatically adjusts to represent the
current data and the icon in the upper left corner is now a green check.
• In general, supplier A has more spread or variation than supplier B.
• The look of a dotplot or histogram is controlled by the number of bins.
• With the missing value, the number of bins changed from the previous
graph resulting in plots that make it easier to see the center and spread.
• However, they no longer show bell-shaped curves.

SupplrA
SupplrB
154.0 157.5 161.0 164.5 168.0 171.5 175.0
Data
Changing number of bins
• The default binning of the x-axis can be changed by selecting a certain
number of intervals or by entering the actual positions of the
midpoints or endpoints.

Edit Scale
1. Double-click the x-axis on the dotplot.
2. Click the Binning tab.
3. Complete the dialog box as shown below.
4. Click OK.
Edit Scale
1. Double-click the x-axis on the
dotplot.
2. Click the Binning tab.
shown below.
4. Click OK.

• With 6 bins, the plots show the data as bell-shaped.
SupplrA
SupplrB
150 155 160 165 170 175
Data
Displaying a graphical summary
The graphical summary provides graphical representations of the data and
descriptive statistics.
The summary components include:
• Histogram with normal curve-Shows the distribution of the data overlaid with a
normal curve.
• Boxplot-Summarizes information about the shape, spread, and center of the data.
• Confidence intervals-Provide interval estimates of the population parameters of
the data.
• Anderson-Darling normality test – Tests whether the data follow a normal
distribution.
• Summary statistics – Describe important features of the data, such as the mean
and standard deviation.
Displaying a graphical summary

Graphical Summary
1. Choose Stat > Basic Statistics
> Graphical Summary.
shown below.
3. Click OK.
The graphical summaries for supplier A and Supplier B strengths are
shown on the right.
Histogram
• The histograms on each summary appear approximately bell-shaped,
indicating the data are normally distributed.
• You can use the Anderson-Darling test to confirm that the data are
normal.

• Confidence intervals are estimates of the true population parameters,
based on a sample.
• Parameters, such as the population mean, median, and standard
deviation, are commonly estimated, as shown in this output.
• Each interval provides a likely range of values for a population
parameter, given a certain confidence level.
• In this example, the confidence level is 95%.
Quartiles
• Examine the minimum, maximum, and quartile values to understand
the spread of data.
• By definition, 50% of the data fall between the first and third
quartiles.

Summary Report for SupplrA Summary Report for SupplrB
Anderson-Darling Normality Test Anderson-Darling Normality Test
A-Squared 0.25 A-Squared 0.52
P-Value 0.718 P-Value 0.164
Mean 163.82 Mean 160.14
StDev 5.66 StDev 3.26
Variance 32.04 Variance 10.65
Skewness -0.251625 Skewness -0.743309
Kurtosis 0.235105 Kurtosis 0.048938
N 20 N 19
Minimum 151.80 Minimum 152.70
1st Quartile 160.60 1st Quartile 157.80
Median 165.00 Median 161.10
3rd Quartile 167.55 3rd Quartile 162.20
Maximum 175.60 Maximum 165.00
95% Confidence Interval for Mean 95% Confidence Interval for Mean
156 162 168 174 161.17 166.46 152 156 160 164 158.57 161.71
95% Confidence Interval for Median 95% Confidence Interval for Median
160.92 166.96 157.97 162.20
95% Confidence Interval for StDev 95% Confidence Interval for StDev
4.30 8.27 2.47 4.83
95% Confidence Intervals 95% Confidence Intervals
Mean Mean
Median Median
161 162 163 164 165 166 167 158 159 160 161 162
Independent 2-
2-sample t-
t-test
What is an independent 2-sample t-test
• An independent 2-sample t-test helps determine whether two
population means are different.
• The test uses the sample standard deviations to estimate for each
population.
• If the difference between the sample means is large relative to the
estimated variability of the sample means, then the population
means are unlikely to be the same.
• You can also use an independent 2-sample t-test to evaluate whether
the means of two populations are different by a specific amount.
Independent 2-
2-sample t-
t-test
When to use an independent 2-sample t-test
• Use an independent 2-sample t-test with continuous data from two
independent random samples.
• Samples are independent if observations from one sample are not
related to the observations from the other sample.
• In this example, the strengths of pellets from supplier A’s process are
not influenced by the strengths of supplier B’s pellets; thus, samples
are independent.
Independent 2-
2-sample t-
t-test
When to use an independent 2-sample t-test
• The test also assumes that the data come from normally distributed
populations
unimodal, and reasonably symmetric.
Independent 2-
2-sample t-
t-test
Why use an independent 2-sample t-test
An independent 2-sample t-test answers questions such as:
• Are the means of a product characteristic between two suppliers
comparable?
• Is one formulation of a product better on average than another?
For example,
• Is the mean viscosity of oil from two different vendors similar?
• Is one ink formulation brighter on average than another?
Conducting 2-
2-sample t-
t-test
A 2-sample t-test will show whether the mean strengths of the plastic
from the two suppliers are different.
The hypotheses for the test are:

H : μ -μ = 0
H : μ -μ ≠ 0
Display boxplots to visualize the data.
Conducting 2-
2-sample t-
t-test
Assume unequal variances
• The t-test will be more powerful if the variances of the two
populations are equal.
• However, assuming the variances are equal when they are not may
give misleading results.
• Thus, if you have any doubt, do not assume equality.
Conducting 2-
2-sample t-
t-test
2-Sample t
> 2-Sample t.
shown below.
3. Click Graphs.
4. Check Boxplot.

The boxplot illustrates that:
• The sample average breaking strength of the plastic from supplier A
appears to be higher than the sample average breaking strength of
the plastic from supplier B.
• The sample variability in breaking strength from supplier A appears to
be larger than the sample variability in breaking strength from
supplier B.
• You need to look at the results of statistical hypothesis tests to
determine if these observations from the graph are significant.
Boxplot of SupplrA, SupplrB
175
170
165
Data
160
155
150
SupplrA SupplrB

Two-Sample T-Test and CI: SupplrA, SupplrB
Two-sample T for SupplrA vs SupplrB
N Mean StDev SE Mean

SupplrA 20 163.82 5.66 1.3
SupplrB 19 160.14 3.26 0.75
Difference = μ (SupplrA) - μ (SupplrB)

Estimate for difference: 3.67
95% CI for difference: (0.67, 6.68)
T-Test of difference = 0 (vs ≠): T-Value = 2.50 P-Value = 0.018 DF = 30
• The difference between the sample means (3.67) is an estimate of the
difference between the population means ( (SupplrA) - (SupplrB)).
• The confidence interval for the difference is based on this estimate
and the variability within the samples.
• You can be 95% confident that the Supplier A mean is greater than
the Supplier B mean by between 0.67 and 6.68 Newtons.

T-value and p-value
• The t-value for the test is 2.50 and the associated p-value is 0.018.
• The p-value (0.018) is the probability of arriving at a t-value with a
magnitude of 2.50 or greater if the null hypothesis of no difference is
true.
• Thus, reject the null hypothesis at the 0.05 -level, and conclude that
the mean strengths are different.
Comparing variances
• The 2-sample t-test compares the means of two populations.
• Often it is of interest to know whether the variances (or standard
deviations) of two groups are different.
• In this example, it may be important to know if one supplier’s
strength varies more than the other because it may be desirable to
choose the low variation supplier.
Comparing variances
• You already created boxplots that suggest that the variability in the
strength of the two plastics is different.
• Now, conduct a format hypothesis test.
• The hypotheses for the 2 Variances test are:
H : σ /σ = 1
H : σ /σ ≠ 1
• To increase the power of the equality of variance test, use a larger
type I error rate, such as = 0.10.
Comparing variances
2 Variances
> 2 Variances.
shown below.
3. Click Options and enter 90 in
Confidence level.

• Use confidence intervals to compare the variances (or standard
deviations).
• Since both confidence intervals on the ratio of the variances
(standard deviations) do not include 1, reject the null hypothesis that
the variances (or standard deviations) are equal.
Variance test
• Both Bonett’s and Levene’s are methods that can be applied to any
continuous distribution.
• The p-value for both tests are less than (0.10), so reject the null
hypothesis that the variance (standard deviations) are equal.
• The sample size calculation assumed equal variances.
• Because the variances are not equal, the sample size calculation is not
exact.
• If a difference is not found between the suppliers means, consider
increasing the sample size for the group with more variation (supplier
A) to ensure the power is at least 80%.

Test and CI for Two Variances: SupplrA, SupplrB
Ratio = 1 vs Ratio ≠ 1
95% CI for σ (SupplrA) / σ (SupplrB)

Bonett’s Test
Bonett
P-Value 0.048
Levene’s Test
Levene
P-Value 0.068
1.0 1.5 2.0 2.5 3.0
95% CI for StDevs

SupplrA
SupplrB
2 3 4 5 6 7 8
Boxplot of SupplrA, SupplrB
SupplrA
SupplrB
150 155 160 165 170 175

Testing the normality assumption
• The 2-sample t-test assumes the data are sampled from normally
distributed populations.
normality is valid.
• We already have the normality tests in the graphical summaries.
• Both p-values (0.718 for SupplrA and 0.164 for SupplrB) are greater
than the significance level = 0.05.

Conclusion
• Based on the plots and the tests, you can assume that the data are
from normally distributed populations.
Summary Report for SupplrA Summary Report for SupplrB
Anderson-Darling Normality Test Anderson-Darling Normality Test
A-Squared 0.25 A-Squared 0.52
P-Value 0.718 P-Value 0.164
Mean 163.82 Mean 160.14
StDev 5.66 StDev 3.26
Variance 32.04 Variance 10.65
Skewness -0.251625 Skewness -0.743309
Kurtosis 0.235105 Kurtosis 0.048938
N 20 N 19
Minimum 151.80 Minimum 152.70
1st Quartile 160.60 1st Quartile 157.80
Median 165.00 Median 161.10
3rd Quartile 167.55 3rd Quartile 162.20
Maximum 175.60 Maximum 165.00
95% Confidence Interval for Mean 95% Confidence Interval for Mean
156 162 168 174 161.17 166.46 152 156 160 164 158.57 161.71
95% Confidence Interval for Median 95% Confidence Interval for Median
160.92 166.96 157.97 162.20
95% Confidence Interval for StDev 95% Confidence Interval for StDev
4.30 8.27 2.47 4.83
95% Confidence Intervals 95% Confidence Intervals
Mean Mean
Median Median
161 162 163 164 165 166 167 158 159 160 161 162
Creating a multiple graph layout

Use the graph layout tool to place multiple graphs on the same page.
Rows
• Choose the number of rows for the layout.
Columns
• Choose the number of columns for the layout.
Adding graphs to the layout
• Select the graph you want to add, then click the cell in which you
want to place the graph.
• Click the right arrow button to place the graph in the layout.
• The left arrow removes a graph from the layout.
Graph Layout
1. Click , then double-click the Summary for SupplrA in the Project
Manager. Click the graph to make it the active window.
2. Choose Editor > Layout Tool.
3. Select Summary for SupplrB from the list of graphs and click to
place the graph in the layout window.

Adding graphs to the layout
• You can also double-click a graph to move it to the layout.
• When a graph is in the layout, you can click and drag it to a different
cell.

Graph Layout
4. In the layout window, click the bottom left square. Then, double-
click Dotplot of SupplrA, SupplrB.
5. In the layout window, click the bottom right square. Then, double-
click Boxplot of SupplrA, SupplrB.
6. Click Finish.
Interpreting your result
Editing graphs in the layout
• You can edit individual graphs in the layout.
• Double-click a graph to access the editing tools.
Updating graphs in a layout
• If the worksheet data change for graphs in a multiple layout, Minitab
cannot update the layout.
• You must update the individual graph and replace the outdated
version with the new one in the layout.
Interpreting your result

Sending graphs to Microsoft PowerPoint or Word
• To quickly build reports in Microsoft PowerPoint or Word, use the
Send to Microsoft PowerPoint or Send to Microsoft Word feature in
the right-click menu.
• This feature allows you to send selected Minitab output directly to
PowerPoint or Word without using multiple copy and paste actions or
creating a Minitab report first.

• You can send graphs or output directly from the Session or Graph
window.
• You can send several graphs or sections of Session window output at
the same time from the Session folder in the Project Manager.
• If PowerPoint or Word is open, Minitab sends the output to the
existing document.
• If Word or PowerPoint is not open, Minitab prompts the program to
open and adds the output to a new document.
Send to PowerPoint
1. Click to view the Graphs
folder.
2. Under Graph, right-click
Layout and choose Send to
Microsoft PowerPoint.

• The selected graph is sent to a new PowerPoint presentation.
• If more than one graph is selected, then each graph is presented in its
own slide.
• Minitab uses the following rules to divide output on the PowerPoint
slides:
• The title of the slide is the graph’s title.
• Graphs are always placed on their own slides.
• Text output always begins on a new slide. For text output that has more than
the specified maximum number of lines per slide (default is 20), Minitab
breaks the text as it does in the Session window.
• Supplier A’s plastic is significantly stronger on average but is more
variable than supplier B’s.
• If mean and variance are equally important, consider comparing the
Cpk statistics for the two suppliers.
• The Cpk statistic is a quality metric that takes into account both the
mean and variance along with the specification limits.
• If the Cpk statistics are close, the company may want to choose a
supplier based on cost.
When using a 2-sample t-test:
• Samples must be independent and random.
• Sample data must be continuous.
• Sample data are from normally distributed populations.
The t-test procedure is robust to violations of the normality

assumption, provided that the observations are collected randomly and
the data are continuous, unimodal, and reasonably symmetric.
Anode Height
Exercise F: Anode Height
Problem
• An electronics manufacturer must ensure that the electrical anode in each
capacitor is a certain distance above the surface of the capacitor’s ceramic
body.
• Recently, the manufacturer has produced many capacitors with anode
heights that violate the lower specification limit.
• A change has been implemented which they believe will increase anode
height.
• The engineers must determine how much data to collect, and then
compare the height measurements before and after the process change.
Anode Height
Exercise F: Anode Height
Data collection
• Samples from the anode insertion process are available but expensive
to measure, so the engineers must not collect more than necessary.
Anode Height
Anode Height (continued)
Instructions
1. Use Power and Sample Size to determine the number of observations required
to detect that the process has improved by at least 0.4 mm with a power of
0.80. Assume a standard deviation of 0.55 mm and specify a greater than
alternative hypothesis.
2. What is the required sample size if the t-test users a two-sided alternative
hypothesis?
3. Use Graphical Summary to check each sample for normality.
4. Check the two samples to verify the assumption of equal variance. Use alpha =
0.10.
Anode Height
Anode Height (continued)
Instructions
5. Use a 2-sample t-test to determine if the anode height has improved after the
process change. Specify the “greater than” alternative hypothesis, and create a
boxplot to visually compare the two sample’s data.
6. Use the Layout Tool to put the two graphical summaries and the boxplot on a
single graph for presentation.
7. Copy the output for the two sample t-test from the Session window. On the
Layout graph, click on the Graph Annotation toolbar to draw a text box in the
lower right corner that was left blank in step 6 and paste the Session Window
output for the 2-sample t-test.
8. Send the finished Layout graph to Microsoft PowerPoint.
Anode Height
Data set
• L01_Dataset_Anode.MPJ
Anode Height The height of an electrical anode above the
ceramic surface (mm)
Process Change Before or After the process change
Paired t-
t-test
Example 6: Parking Cars
Problem
• A consumer group wants to determine whether two popular cars
differ in handling ability.
• To measure the handling ability of the two cars, the group compares
the time it takes drivers to parallel park each car.
Data collection
• Twenty drivers park both cars in random order and parking time in
seconds is recorded.
Paired t-
t-test
Tools
• Open Excel
• Assistant – Paired t
• Calculator
• Boxplot
• Help
• Outlier Test
Paired t-
t-test
Data set
• L01_Dataset_Parking.xlsx
Car_A Park times driving car A (sec)
Car_B Park times driving car B (sec)
Importing data from Excel
Excel file structure
Before you import the data into
Minitab, consider the structure of
the Excel file:
• There are no variable names.
• The data begin in Row 1.
• The worksheet has no spaces or
blank rows.

Options and previews
• When opening non-Minitab files, the Option and Preview dialog
boxes are available.
• Use Options to specify column names, column delimiters, the row to
begin reading data, and any missing text or numeric variables to be
converted.
• Use Preview to see the column names, data types, and rows before
Minitab opens the file.
• You can also change column names and data types in Preview.
Files of type > Excel
1. Choose File > New, select Minitab
Project, and then click OK.
2. Choose File > Open Worksheet.
3. In File of type, choose Excel (*.xls;
*.xlsx).
4. Select Parking.xlsx, then click
Preview.

Files of type > Excel (continued)
5. Click OK, then click Options.
6. Complete the dialog box as shown.
7. Click OK, then click Open.
8. In the Data window in Minitab, click
the blank cell under C1 and type
Car_A. Click the blank cell under C2
and type Car_B.
• Examine the Minitab worksheet
to ensure that the data are in
the proper format.
• Notice that the data appear in
Minitab exactly as they did in
Excel.
• In addition, the variable names
now appear in the cells above
the data.
Paired t-
t-test
What is a paired t-test
• A paired t-test helps determine whether the mean difference
between paired observations is significant.
• Statistically, the paired t-test is equivalent to performing a 1-sample t-
test on the differences.
• A paired t-test also helps you evaluate whether the mean difference is
equal to a specific value.
Paired t-
t-test
Paired observations are related. Examples include:
• Weights recorded for individuals before and after an exercise
program.
• Measurements of the same part taken with two different measuring
devices.
Paired t-
t-test
When to use a paired t-test
• Use a paired t-test with a random sample of paired observations.
• The test also assumes that the paired difference come from a
normally distributed population.
unimodal, and reasonably symmetric.
Paired t-
t-test
Why use a paired t-test
A paired t-test answers questions such as:
• Does a new treatment result in a difference in the product?
• Do two different instruments provide similar measurements for the same
sample?
For example,
• Does treating lumber with certain chemicals increase its usable life?
(Assume that each sample piece of lumber is cut in half; one half is treated
with the chemicals while the other half is untreated.)
• Do two calipers differ in measuring parts? (Assume that several parts are
measured once with each caliper.)
Conducting a paired t-
t-test
• The consumer group wants to determine whether drivers can park
one car more quickly than the other.
• Because the data are paired (each individual parked both cars), use a
paired t-test to test the following hypotheses:
• : The mean difference between paired observations in the population is
zero.
• : The mean difference between paired observations in the population is
not zero.
Conducting a paired t-
t-test
Paired t
1. Choose Assistant > Hypothesis
Tests.
2. Under Compare two samples
with each other, choose
Paired t.
shown below.
4. Click OK.

Report Card
• The Report Card provides information about assumptions for the
procedure being used and contains a list of potential failure modes
that can be automatically detected.
• If a problem occurs, the report will tell you about the problem and
how to resolve it.
Report Card
• We can see that there is one paired difference that is showing as an
unusual observation.
• The sample size of 20 is large enough so that normality is not a
concern because the paired t-test is robust to violations of the
normality assumption.
• In addition, the Report Card shows that the sample size is large
enough to detect a difference between the means for the cars.

Report Card
Diagnostic Report
• The Diagnostic Report is designed to verify that the results are sound
so that you can trust the answer in the Summary Report.
• Again, there is one unusual difference identified.
• As a possible outlier, it is shown in red.
• With a sample size of 20, we have an 80% chance of detecting a
difference of 2.5332 seconds.

Paired t Test for the Mean of Car_A and Car_B
Diagnostic Report
Paired Data in Worksheet Order

Investigate any pairs with unusual differences (marked in red).
50 Car_A
Car_B
40
30
20
What is the chance of detecting a difference?

What difference can you detect with a sample
< 40% 60% Power 90% 100% size of 20?
Difference Power
2.0002 60%
2.2457 70%
2.5332 80%
2.0002 Difference 2.9322
2.9322 90%
For α = 0.05 and sample size = 20:
If the true means differed by 2.0002, you would have a 60% chance of detecting Observed difference = 1.965
the difference with a paired test. If they differed by 2.9322, you would have a
90% chance.
Power is a function of the sample size and the standard deviation. To detect smaller differences, consider
increasing the sample size.
Summary Report
• The Summary Report contains the overall conclusions from the data
analysis and is designed to be easy to explain to Management.
• The mean parking times are 34.86 seconds for car A and 32.9 seconds
for car B.
• The mean difference is about 2 seconds.
• You can be 95% confident that the difference in parking times
between car A and car B is at least 0.17 seconds, and at most 3.76
seconds.

Summary Report
• The p-value of the paired t test is 0.034.
• Thus, reject the null hypothesis at the 95% confidence level and
conclude that the average time required to park car A is greater than
the average time required to park car B.
Paired t Test for the Mean of Car_A and Car_B
Summary Report
Do the means differ? Paired Differences

0 0.05 0.1 > 0.5 *Paired
Statistics Differences
Sample size 20
Yes No
Mean 1.965
P = 0.034 95% CI (0.16988, 3.7601)
The mean of Car_A is significantly different from the mean of Standard deviation 3.8356
Car_B (p < 0.05).
*Difference = Car_A - Car_B
Individual Samples
Statistics Car_A Car_B
Mean 34.865 32.9
Distribution of the Differences Standard deviation 7.5916 7.2743
Where are the differences relative to zero?
0 Comments
• Test: You can conclude that the means differ at the 0.05
level of significance. The mean of the paired differences is
greater than zero.
• CI: Quantifies the uncertainty associated with estimating
the mean difference from sample data. You can be 95%
confident that the true mean difference is between 0.16988
and 3.7601.
• Distribution of Differences: Compare the location of the
differences to zero. Look for unusual differences before
interpreting the results of the test.
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 10.0
Boxplot
• One sample difference is identified as unusual.
• We want to investigate this difference using a boxplot and determine
whether it is statistically an outlier.
• Before doing these, store the pairwise differences in the worksheet.
Formulas
• Check Assign as a formula to assign the expression as a formula to the
column.
• This allows Minitab to recalculate the values automatically whenever
you add, remove, or change associated data.
Boxplot
1. Choose Calc > Calculator.
shown below.
3. Click OK.
4. Choose Graph > Boxplot.
5. Under One Y, choose Simple,
Click OK.
6. In Graph variables, enter Diff.
7. Click OK.

• When you assign a formula to a
column, a status icon appears in
the upper right corner of the
column heading in the
worksheet.
• This indicator tells you whether
the formula is properly defined
and whether the data need to be
updated by recalculating the
values.
• The boxplot shows the Boxplot of Diff
10
difference between the paired

observations. 5
• There is one unusual

observation.
Diff
0
-5
-10
Using Help to understand unusual observations

• The boxplot shows one unusual observation (the asterisk) with a large
negative difference between the parking times of car A and car B.
• By either hovering the cursor over the asterisk or by using the Brush
tool, it can be seen that this difference is in row 13.
• How does Minitab determine unusual observations on a boxplot or in
the Assistant for a Paired t-test.?
• We can use Minitab Help to easily see the rule.
• In the Help topic at the right, whiskers are defined using the quartiles
and the interquartile range: upper whisker Q3 + 1.5 IQR and lower
whisker Q1 – 1.5 IQR where the IQR = Q3 – Q1.
• Then, any data point that lies above the upper whisker or below the
lower whisker is considered an unusual observation or a potential
outlier.

Using Help
1. Click . Choose Graph Menu, then click Boxplot.
Conducting an outlier test
• One of the difference in the paired t-test is identified as a potential
outlier.
• We can use a formal test to verify whether the unusual observation is
indeed an outlier.
• The test is called Grubbs outlier test. It assumes that the data come
from a normal population.
• The hypotheses are the following:
: All data values come from the same normal population.
: The smallest or largest data value is an outlier.
• Use the default confidence level of 95%.
Conducting an outlier test

Outlier Test
> Outlier Test.
shown below.
3. Click OK.
• As seen in the Paired t-test the average difference is 1.965.
• The largest difference is 9.1 and the smallest difference is -7.6.
• The p-value of 0.130 indicates that we cannot reject the null
hypothesis at the 95% confidence level.
• The large negative difference of 7.6 is not a significant outlier.

Outlier Test: Diff
Method
Null hypothesis All data values come from the same normal
population
Alternative hypothesis Smallest or largest data value is an outlier
Significance level α = 0.05
Grubbs' Test
Variable N Mean StDev Min Max G P

Diff 20 1.965 3.836 -7.600 9.100 2.49 0.130
* NOTE * No outlier at the 5% level of significance

Outlier Plot of Diff
Grubbs' Test
Min Max G P
-7.60 9.10 2.49 0.130
-10 -5 0 5 10
Diff
• On average, drivers took about 2.0 seconds longer to park car A than
car B. This difference, although small, is statistically significant.
• Given the confidence interval for the mean difference, is a 0.17
second difference, or even a 3.76 second difference, of any practical
importance?
• If slightly longer parking time increases driver frustration, then the
difference may be important.
• This difference may also be more important to drivers who parallel
park frequently.
When using a paired t-test:
• Observations must be paired.
• The data must be continuous.
• The paired differences are from a normally distributed population.
• The t-test procedure is robust to violations of the normality
assumption, provided that pairs of observations are collected
randomly and the data are continuous, unimodal, and reasonably
symmetric.
Using paired observations eliminates the variability caused by
individuals.
For example,
• Driver 1 took 18.9 seconds to park car A and 18.2 seconds to park car B.
• In contrast, driver 18 took 43.4 and 37.3 seconds to park the same cars.
• Analyzing the differences for each driver eliminates the variability between
drivers from the calculations, increasing the power of the test.
Comparing Calipers
Exercise G: Comparing Calipers
Problem
• A company considers purchasing two different gages to measure ball
bearings: a caliper by EasyGage and a caliper by ToolIt.
• Compare the two brands to determine whether they yield different average
measurements.
• Use an -level of 0.05 for the test.
Comparing Calipers
Instructions
1. Use File > Open Worksheet to open the data. Be sure to change the
Variable Names option to None.
2. Name the columns (Bearing, EasyGage, ToolIt).
3. Use the Assistant paired t-test to determine whether the measurements
given by the calipers are different.
4. Use Calculator to create a Diff column. Check the differences for
normality.
Comparing Calipers
Data collection
• Analysts randomly sample 12 ball bearings from the process.
• A single operator measures each ball bearing with the EasyGage caliper
and with the ToolIt caliper.
Comparing Calipers
Data set
• L01_Dataset_Calipers.xlsx
Bearing Identifies the bearing measured
EasyGage Bearing diameter measured by EasyGage caliper (cm)
ToolIt Bearing diameter measured by TooIt caliper (cm)

Mit

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Mit

Caricato da

Copyright:

Formati disponibili

Hypothesis Tests:

Hypothesis Tests: Continuous Data

Hypothesis Tests: Continuous Data

Hypothesis Testing and Confidence Intervals

Hypothesis Testing and Confidence Intervals

Hypothesis Testing and Confidence Intervals

Variable N N* Mean SE Mean StDev Minimum Q1 Median Q3 Maximum

Hypothesis Testing and Confidence Intervals

Hypothesis Testing and Confidence Intervals

• Using information from a sample to make a conclusion about a

Testing the null hypothesis

Testing the null hypothesis

The logic of hypothesis testing

Interpreting your results

The t-statistic (1.74) is:

Interpreting your results

Test of μ = 365 vs ≠ 365

Variable N Mean StDev SE Mean 95% CI T P

Interpreting your results

Test of μ = 365 vs ≠ 365

Variable N Mean StDev SE Mean 95% CI T P

Testing the assumption of normality

Interpreting your results

Interpreting your results

Using the confidence interval

4. Click OK in each dialog box.

Interpreting your results

Individual Value Plot of BoxWeigh

363 364 365 366 367 368 369 370 371

Interpreting your results

Interpreting your results

Variable N Mean StDev SE Mean 95% CI T P

Individual Value Plot of BoxWeigh

363 364 365 366 367 368 369 370 371

The null hypothesis ( ):

When using a 1-sample t-test:

Exercise C: Filling Cereal Boxes: Minitab Assistant

Power and Sample Size

Determining the difference parameter

Power and Sample Size for 1-Sample t

Interpreting your results

Testing mean = null (versus ≠ null)

Interpreting your results

Interpreting your results

Interpreting your results

Power and Sample Size for 1-Sample t

Interpreting your results

Additional observations give you more power:

Interpreting your results

Testing mean = null (versus ≠ null)

Interpreting your results

Adding a footnote to a graph

Testing the null hypothesis

Interpreting your results

362 363 364 365 366 367 368 369 370

Interpreting your results

Interpreting your results

Variable N Mean StDev SE Mean 95% CI T P

Testing the assumption of normality

Interpreting your results

Detecting Changes in Bearing Diameter

Detecting Changes in Bearing Diameter

Investigating Supplier Quality

The null hypothesis ( ):