ng bb 33 hypothesis testing basics
DESCRIPTION
TRANSCRIPT
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
National GuardBlack Belt Training
Module 33
Hypothesis Testing Basics
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
CPI Roadmap – Analyze
Note: Activities and tools vary by project. Lists provided here are not necessarily all-inclusive.
TOOLS•Value Stream Analysis•Process Constraint ID •Takt Time Analysis•Cause and Effect Analysis •Brainstorming•5 Whys•Affinity Diagram•Pareto •Cause and Effect Matrix •FMEA•Hypothesis Tests•ANOVA•Chi Square •Simple and Multiple Regression
ACTIVITIES
• Identify Potential Root Causes
• Reduce List of Potential Root Causes
• Confirm Root Cause to Output Relationship
• Estimate Impact of Root Causes on Key Outputs
• Prioritize Root Causes
• Complete Analyze Tollgate
1.Validate the
Problem
4. Determine Root
Cause
3. Set Improvement
Targets
5. Develop Counter-
Measures
6. See Counter-MeasuresThrough
2. IdentifyPerformance
Gaps
7. Confirm Results
& Process
8. StandardizeSuccessfulProcesses
Define Measure Analyze ControlImprove
8-STEP PROCESS
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
3Hypothesis Testing - Basic
Learning Objectives
Review the terms “Parameters” and “Statistics” as they relate to Populations and Samples.
Introduce Confidence Intervals for expressing the uncertainty when predicting a population parameter using a sample statistic, and how to calculate CI’s for some common situations for different sample sizes.
Show how the Central Limit Theorem and the Standard Error of the Mean applies to the use of Confidence Intervals and Tests
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
4Hypothesis Testing - Basic
Learning Objectives (Cont.)
Introduce statistical tests for some common tests and introduce the t-distribution with testing
Learn about Hypothesis Testing to prove a statistical difference in process performance in applications of Minitab
Understand the tradeoffs and influences of sample sizes on statistical tests.
Apply knowledge of different classes of statistical errors to the decisions used in process improvement to minimize risk.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
5Hypothesis Testing - Basic
Application Examples
Transactional – A Black Belt has just finished a pilot of a new process for handling blanket Purchase Orders and wants to know if it has a statistically significant: a) shorter cycle time and b) increased accuracy over the old process.
Administrative – The manager of an AAFES1 order entry department wants to compare two order entry procedures to see if one is faster than the other.
Service – Medical diagnostic imaging services are provided from two different medical treatment facilities to a central hospital which wants to know if there are differences in the quality of service, particularly: a) the number of lost records and re-takes, and b) average waiting time for MRIs and X-rays.
1AAFES, Army and Air Force Exchange System
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
6Hypothesis Testing - Basic
Since it is not always practical or possible to measure/query every item/person in the population, you take a random sample.
Population vs. Sample
25 appraisals chosen at random from a given month
All appraisals completed that month
3,000 people are given a new treatment in a clinical study
All sufferers of a certain disease that might be given the new treatment
10,000 people are asked who they will vote for President
All U.S. registered voters
SamplePopulation
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
7Hypothesis Testing - Basic
Terms and Labels: Population vs. Sample
~
Count of items Mean Median Standard Dev.
N
m
m
s
n
x
x
S
Estimators =m
s
x
s
~
Population =
ParameterTerm
Sample =
Statistic
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
8Hypothesis Testing - Basic
Population Parameters;
Mean, m (mu), and Standard Deviation, s (sigma)
Sample Statistics; Mean, x-bar,
and Standard Deviation, s
Population
Random Samples
of Size, n = 4
x s1 1,
x s2 2,
x s3 3,
x s4 4, m s,
Population Parameters vs. Sample Statistics
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
9Hypothesis Testing - Basic
If:x1, x2, …, xn are independent measurements (i.e., a random sample of size n)
from a population, where the mean of x is m, when
the standard deviation of x is given as s,
Then:
The distribution of x
has mean and standard
deviation given by:
In addition, when n is sufficiently large, then the distribution of x- bar is approximately normal (“bell-shaped curve”). More on sample sizes later...
Central Limit Theorem
Standard Error
of the Meann
XX
ssmm and
n
XXX nX
21 x3
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
10Hypothesis Testing - Basic
Variability of Means
Sample statistics estimate population parameters by inference:
For a given sample ( x, s, n ), we can estimate population
parameters of m s by inference.
As the sample size increases we are more confident that our sample statistic is a more valid estimator of the population parameter.
n
n
n
=
=
= 1
3
5
nxxss
sx
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
National GuardBlack Belt Training
Confidence Intervals
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
12Hypothesis Testing - Basic
What Is a Confidence Interval?
We know that when we take the average of a sample, it is probably not exactly the same as the average of the population.
Confidence intervals help us determine the likely range of the population parameter.
For example, if my 95% confidence interval is 5 +/-2, then I have 95% confidence that the mean of the population is between 3 and 7.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
13Hypothesis Testing - Basic
Usually, confidence intervals have an additive uncertainty:
Estimate ± Margin of Error
Sample Statistic ± [ ___ X ___ ]
Confidence
Factor
Measure of
Variability
What Is a Confidence Interval? (Cont.)
Example:
x, s
Note: Detailed formulas may be found in the appendix.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
14Hypothesis Testing - Basic
Why Do We Need Confidence Intervals?
Sample statistics, such as Mean and Standard Deviation, are only estimates of the population’s parameters.
Because there is variability in these estimates from sample to sample, we can quantify our uncertainty using statistically-based confidence intervals. Confidence intervals provide a range of plausible values for the population parameters (m and s).
Any sample statistic will vary from one sample to another and, therefore, from the true population or process parameter value.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
15Hypothesis Testing - Basic
Exercise
Let’s look at a population that has a normal distribution with:
known mean value = 65
standard deviation = 4
(This has been generated in dataset Confidence.mtw)
Each member in the class will randomly sample 25 data points from this population. (In Minitab, use Calc>Random Data>Sample from Columns.)
Sample 25 rows of data from C1 and store the results in C2.
Use graphical descriptive statistics to calculate the 95% confidence interval for the mean and sigma based on your sample of 25 data points. Do they include the mean, 65, and the sigma, 4?
Based on a class size of 25, we would expect 1 confidence interval to not contain 65 for the mean, and 1 that does not include 4 for sigma.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
16Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Population Standard Deviation (s) Known
Example
A random sample of size, n = 36, is taken and the distribution of x is normal. We are given that the population standard deviation (s) is 18.0. The value of x-bar is an estimator of the population mean (m), and the standard error of x-bar is:
From the properties of the standardized normal distribution,
there is a 95% chance that m is within the range of ( x-bar + and - 1.96 times the Standard Error of x-bar).
0.336/0.18/ nbarx ss
This is known as the Standard Error of the Mean
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
17Hypothesis Testing - Basic
1m1m - 1.96(3.0) 1m + 1.96(3.0)
.95
.025 .025
95% of all x-bars will fall into the shaded region, defined by m ± 1.96(3.0)
Standard Error of the Mean
What Values of x-bar Can I Expect?
Distribution of x-bar
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
18Hypothesis Testing - Basic
1m1m - 1.96(3.0) 1m + 1.96(3.0)
Observed sample mean, x-barsample C
But I Don’t Know m, I Only Know x-bar!
We can turn it around.
x-bar lying in the interval m ±1.96(3.0) is the same thing as m lying in the interval x-bar ±1.96(3.0).
Because there is a 95% chance that x-bar lies in the interval m ± 1.96(3.0), there is a 95% chance that the interval x-bar ± 1.96(3.0) encloses m.
The interval we construct using the observed sample mean is called a 95% confidence interval for m.
(---------- x-barsample C-----------)
(----------- x-barsample A -----------)
(---------- x-barsample B-----------)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
19Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Population Standard Deviation (s) Known
Another Example
An airline needs an estimate of the average number of passengers on a newly scheduled flight. Its experience is that data for the first month of flights is unreliable, but thereafter the passenger loading settles down.
Therefore, the mean passenger load is calculated for the first 20 weekdays of the second month after initiation of this particular new flight. If the sample mean (x-bar) is 112.0 and the population standard deviation (s) is assumed to be 25, find a 95% confidence interval for the true, long-run average number of passengers on this flight.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
20Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Standard Deviation (s) Known
Solution
We assume that the hypothetical population of daily passenger loads for weekdays is not badly skewed. Therefore, the sampling distribution of x-bar is approximately normal and the confidence interval results are approximately correct, even for a sample size of only 20 weekdays.
For a 95% confidence interval, we use z.025= 1.96 in the formula to obtain
We are 95% confident that the long-run mean, m , lies in this interval.
59.520
25
0.112bar-x
bar-x
ss
s
96.122 to 04.101 59.596.1112 or
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
21Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Population Standard Deviation (s) Unknown
A very important point to remember is that for this example we assumed that we knew the population standard deviation, and many times that is not the case. Often, we have to estimate both the mean and the standard deviation from the sample.
When s is not known, we use the t-distribution rather than the normal (z) distribution. The t-distribution will be explained next.
In many cases, the true population s is not known, so we must use our sample standard deviation (s) as an estimate for the population standard deviation (s
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
22Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Standard Deviation (s) Unknown (Cont.)
Since there is less certainty (not knowing m or s ), the t-distribution essentially “relaxes” or “expands” our confidence intervals to allow for this additional uncertainty.
In other words, for a 95% confidence interval, you would multiply the standard error by a number greater than 1.96, depending on the sample size.
1.96 comes from the normal distribution, but the number we will use in this case will come from the t-distribution.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
23Hypothesis Testing - Basic
What Is This t-Distribution?
The t-distribution is actually a family of distributions.
They are similar in shape to the normal distribution (symmetric and bell-shaped), although wider, and flatter in the tails.
How wide and flat the specific t-distribution is depends on the sample size. The smaller the sample size, the wider and flatter the distribution tails.
As sample size increases, the t-distribution approaches the exact shape of the normal distribution.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
24Hypothesis Testing - Basic
An Example of a t-Distribution
3210-1-2-3
0.4
0.3
0.2
0.1
0.0
t
freq
uenc
y
2.78
0.025
Area =
t-distribution
(n = 5)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
25Hypothesis Testing - Basic
Sample Size t-value (.025)*
2 12.71
3 4.30
5 2.78
10 2.26
20 2.09
30 2.05
100 1.98
1000 1.96
* For a 95% CI, = .05. Therefore, for a two tail distribution: /2= .05/2= .025
Some Selected t-Values
Here are values from the t-distribution for various sample sizes (for 95% confidence intervals):
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
26Hypothesis Testing - Basic
Confidence Interval for the Mean (m) with Population Standard Deviation (s) UnknownExample
The customer expectation when phoning an order-out pizza shop is that the average amount of time from completion of dialing until they hear the message indicating the time in queue is equal to 55.0 seconds (less than a minute was the response from customers surveyed, so the standard was established at 10% less than a minute). You decide to randomly sample at 20 times from 11:30am until 9:30pm on 2 days to determine what the actual average is. In your sample of 20 calls, you find that the sample mean, x-bar, is equal to 54.86 seconds and the sample standard deviation, s, is equal to 1.008 seconds.
The actual data was as follows:
54.1, 53.3, 56.1, 55.7, 54.0, 54.1, 54.5, 57.1, 55.2, 53.8,54.1, 54.1, 56.1, 55.0, 55.9, 56.0 ,54.9, 54.3, 53.9, 55.0
What is a 95% confidence interval for the true mean call completion time?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
27Hypothesis Testing - Basic
We’re 95% confident that the actual mean call completion time is somewhere between
54.389 seconds and 55.331 seconds,based on our sample of 20 calls.
n
stx 1nα/2,
20
008.109.2860.54
331.55,389.54
95% Confidence Interval for Mean Call Completion Time
x = 54.860
s = 1.008
n = 20
t.025,19 = 2.09 our sample of 20 calls
Luckily, we don’t have to worry about the details of how to calculate the t-value. Minitab takes care of
that for us.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
28Hypothesis Testing - Basic
1. Open the Minitab file PizzaCall.mtw
Now Let Minitab Calculate the Confidence Interval
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
29Hypothesis Testing - Basic
2. Select Stat> Basic Statistics> Graphical Summary
Now Let Minitab Calculate the Confidence Interval (Cont.)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
30Hypothesis Testing - Basic
Now Let Minitab Calculate the Confidence Interval (Cont.)
3. Double click on C-1 to place it in the Variables box
4. Click on OK
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
31Hypothesis Testing - Basic
Now Let Minitab Calculate the Confidence Interval (Cont.)
95% Confidence Interval
for Mean (m:
54.388 55.332
95% Confidence Interval
for Standard Deviation (s:
0.767 1.472
We’re 95% confident that the actual mean is
between 54.388 and 55.332
We’re also taking a 5%chance that we’re wrong.
57565554
Median
Mean
55.5055.2555.0054.7554.5054.2554.00
A nderson-Darling Normality Test
V ariance 1.016
Skewness 0.560026
Kurtosis -0.509797
N 20
Minimum 53.300
A -Squared
1st Q uartile 54.100
Median 54.700
3rd Q uartile 55.850
Maximum 57.100
95% C onfidence Interv al for Mean
54.388
0.60
55.332
95% C onfidence Interv al for Median
54.100 55.582
95% C onfidence Interv al for StDev
0.767 1.472
P-V alue 0.105
Mean 54.860
StDev 1.008
95% Confidence Intervals
Summary for C1
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
32Hypothesis Testing - Basic
Other Types of Confidence Intervals
There are other types of confidence intervals that are based on the same principles we have learned:
Standard Deviation
Proportions
Median
Others
We will discuss some of these later.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
National GuardBlack Belt Training
Hypothesis Testing
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
34Hypothesis Testing - Basic
Extending the Concept of Confidence Intervals
Extending the concept of confidence intervals allows us to set-up and interpret statistical tests.
We refer to these tests as Hypothesis Tests.
One way to describe a hypothesis test:
Determining whether or not a particular value of interest is contained within a confidence interval.
Hypothesis testing also gives us the ability to calculate the probability that our conclusion is wrong.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
35Hypothesis Testing - Basic
The New Car
You buy a one-year old car from the Lemon Lot in order to save money on gas. The previous owner still had the original features sticker and you were pleased to note that the EPA mileage estimate indicated that the car should get 31 miles per gallon overall.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
36Hypothesis Testing - Basic
The New Car (Cont.)
As soon as you buy the car, you fill up the tank so that you’ll be ready to take the family for a drive and to go to work the next day. A few days later, you fill up again and calculate your gas mileage for that tank. After you push the “=“ key on your calculator, the number 27.1 appears.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
37Hypothesis Testing - Basic
The New Car (Cont.)
Should you send the car to a mechanic to check for problems?
Do you conclude that the EPA estimate is simply wrong?
Do you leave cruel messages on the seller’s answering machine?
What ARE your conclusions?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
38Hypothesis Testing - Basic
Continuing the Car Situation
At what value of gas consumption should you become alarmed that you are experiencing anything more than just random variation?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
39Hypothesis Testing - Basic
The Car Situation (Cont.)
What if we knew this?
s = 3.46
Distribution of gas consumption for this
car
12.8 %
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
40Hypothesis Testing - Basic
Hypothesis Testing
Hypothesis Testing:
Allows us to determine statistically whether or not a value is cause for alarm (or is simply due to random variation)
Tells us whether or not two sets of data are different
Tells us whether or not a statistical parameter (mean, standard deviation, etc.) is statistically different from a test value of interest
Allows us to assess the “strength” of our conclusion (our probability of being correct or wrong)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
41Hypothesis Testing - Basic
Hypothesis Testing (Cont.)
Hypothesis Testing Enables Us to:
Handle uncertainty using a commonly accepted approach
Be more objective (2 persons will use the same techniques and come to similar conclusions almost all of the time)
Disprove or “fail to disprove” assumptions
Control our risk of making wrong decisions or coming to wrong conclusions
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
42Hypothesis Testing - Basic
Hypothesis Testing (Cont.)
m
Population Mean
Sample BTrue
Population Distribution
Sample A
Sample C
Sample D
Some Possible Samples
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
43Hypothesis Testing - Basic
Sample Size Concerns
If we sample only one item, how close do we expect to get to the true population mean?
How well do you think this one item represents the true mean?
How much ability do we have to draw conclusions about the mean?
What if we sample 900 items? Now, how close would we expect to get to the true population mean?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
44Hypothesis Testing - Basic
Sample Size (Cont.)
m
Population
Likely value of x-bar with a small sample
size
Likely value of x-bar with a large sample
size
x
The larger our sample, the closer x-bar is likely to be to the true population mean.
x
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
45Hypothesis Testing - Basic
Standard Deviation
What effect would a lot of variation in the population have on our estimate of the population mean from a sample?
How would this affect our ability to draw conclusions about the mean?
What if there is very little variation in the population?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
46Hypothesis Testing - Basic
Standard Deviation (Cont.)
Population with a lot of variation
Population with less variation
m
m
Likely value of x-bar with sample size, n
Likely value of x-bar with sample size, n
x
x
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
47Hypothesis Testing - Basic
Statistical Inferences and Confidence
How much confidence do we have in our estimates?
How close do you think the true mean, m, is to our estimate of the mean, x-bar?
How certain do we want/need to be about conclusions we make from our estimates?
If we want to be more confident about our sample estimate (i.e., we want a lower risk of being wrong), then we must relax our statement of how close we are to the true value.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
48Hypothesis Testing - Basic
Statistical Inferences and Confidence (Cont.)
m
Population
If we want to have high confidence in our conclusions, we must
relax the range in which we say the true
mean lies
As we tighten our estimate of the mean, our risk of being wrong increases. Thus, our confidence decreases.
x
x
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
49Hypothesis Testing - Basic
Three Factors Drive Sample Sizes
Three concepts affect the conclusions drawn from a single sample data set of (n) items:
Variation in the underlying population (sigma)
Risk of drawing the wrong conclusions (alpha, beta)
How small a Difference is significant (delta)
)(n
Risk
Variation Difference
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
50Hypothesis Testing - Basic
Three Factors: Variation, Risk, Difference
These 3 factors work together. Each affects the others.
Variation: When there’s greater variation, a larger sample is needed to have the same level of confidence that the test will be valid. More variation reduces our confidence interval.
Risk: If we want to be more confident that we are not going to make a decision error or miss a significant event, we must increase the sample size.
Difference: If we want to be confident that we can identify a smaller difference between two test samples, the sample size must increase.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
51Hypothesis Testing - Basic
Three Factors (Cont.)
Larger samples improve our confidence interval.
Lower confidence levels allow smaller samples.
All of these translate into a specific confidence interval for a given parameter, set of data, confidence level and sample size.
They also translate into what types of conclusions result from hypothesis tests.
Testing for larger differences between the samples, reduces the size of the sample. This is known as delta (D).
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
52Hypothesis Testing - Basic
An Example
A unit has several quick response forces, QRF. Some forces have over 700 members, with at least 300 on the site at any time.
By regulation, all forces must have a quick response plan, the critical first phase of which is required to be completed in 10 minutes (600 seconds) or less.
There are two teams that are vying for “most responsive.” They have taken somewhat different approaches to implementing their quick response plans and management wants to know which approach is better: Team 1 or Team 2
Each one has 100 data points for actual responses and drills (Minitab file Response.mtw)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
53Hypothesis Testing - Basic
The Data from Team 1
598.0 598.8 600.2 599.4 599.6
599.8 598.8 599.6 599.0 601.2
600.0 599.8 599.6 598.4 599.6
599.8 599.2 599.6 599.0 600.2
600.0 599.4 600.2 599.6 600.0
600.0 600.0 599.2 598.8 600.0
598.8 600.2 599.0 599.2 599.4
598.2 600.2 599.6 599.6 599.8
599.4 599.6 600.4 598.6 599.2
599.6 599.0 600.0 599.8 599.6
599.4 599.0 599.0 599.6 599.4
599.4 599.8 599.6 599.2 600.0
600.0 600.8 599.4 599.6 600.0
598.8 598.8 599.2 600.2 599.2
599.2 598.2 597.8 599.8 599.4
599.4 600.0 600.4 599.6 599.6
599.6 599.2 599.6 600.0 599.8
599.0 599.8 600.0 599.6 599.0
599.2 601.2 600.8 599.2 599.6
600.6 600.4 600.4 598.6 599.4
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
54Hypothesis Testing - Basic
The Data from Team 2
601.6 600.8 599.4 599.8 601.6
600.4 598.6 598.0 602.8 603.4
598.4 600.0 597.6 600.0 597.0
600.0 600.4 598.0 599.6 599.8
596.8 600.8 597.6 602.2 597.8
602.8 600.8 601.2 603.8 602.4
600.8 597.2 599.0 603.6 602.2
603.6 600.4 600.4 601.8 600.6
604.2 599.8 600.6 602.0 596.2
602.4 596.4 599.0 603.6 602.4
598.4 600.4 602.2 600.8 601.4
599.6 598.2 599.8 600.2 599.2
603.4 598.6 599.8 600.4 601.6
600.6 599.6 601.0 600.2 600.4
598.4 599.0 601.6 602.2 598.0
598.2 598.2 601.6 598.0 601.2
602.0 599.4 600.2 598.4 604.2
599.4 599.4 601.8 600.8 600.2
599.4 600.2 601.2 602.8 600.0
600.8 599.0 597.6 597.6 596.8
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
55Hypothesis Testing - Basic
Descriptive Statistics – Team 1
600.75600.00599.25598.50597.75
Median
Mean
599.70599.65599.60599.55599.50599.45599.40
1st Q uartile 599.20
Median 599.60
3rd Q uartile 600.00
Maximum 601.20
599.43 599.67
599.40 599.60
0.54 0.72
A -Squared 0.84
P-V alue 0.029
Mean 599.55
StDev 0.62
V ariance 0.38
Skewness -0.082566
Kurtosis 0.745102
N 100
Minimum 597.80
A nderson-Darling Normality Test
95% C onfidence Interv al for Mean
95% C onfidence Interv al for Median
95% C onfidence Interv al for StDev
95% Confidence Intervals
Summary for Team 1
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
56Hypothesis Testing - Basic
Descriptive Statistics – Team 2
603.0601.5600.0598.5597.0
Median
Mean
600.6600.4600.2600.0599.8
1st Q uartile 599.00
Median 600.20
3rd Q uartile 601.60
Maximum 604.20
599.86 600.60
599.80 600.60
1.65 2.18
A -Squared 0.29
P-V alue 0.615
Mean 600.23
StDev 1.87
V ariance 3.51
Skewness 0.051853
Kurtosis -0.518286
N 100
Minimum 596.20
A nderson-Darling Normality Test
95% C onfidence Interv al for Mean
95% C onfidence Interv al for Median
95% C onfidence Interv al for StDev
95% Confidence Intervals
Summary for Team 2
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
57Hypothesis Testing - Basic
Example
The average cycle time for Team 1 is 599.55 seconds.
The average cycle time for Team 2 is 600.23 seconds.
The target cycle time for Phase 1 response is 600 seconds.
Is the difference between the two average cycle times statistically significant?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
58Hypothesis Testing - Basic
Example (Cont.)
The unit wants to determine if the true averages of the two teams are really different.
The unit thinks that the 600.23 average of team 2 is little too high, so there is a need to determine if the data indicates that the true average is really not equal to the target of 600 seconds.
The unit will use hypothesis testing to answer these questions.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
59Hypothesis Testing - Basic
Example
The first hypothesis test to be performed is to determine whether there is a statistically significant difference between the means of the two teams. This is called a 2-Sample t Test.
The real question is whether or not the means are different enough to indicate that the approaches taken by the two teams really are centered differently, or are they close enough that the difference could simply be a result of random variation?
After that, hypothesis testing can tell us if there is evidence indicating whether or not each team’s average is different from the target of 600 seconds.
First, we need to introduce some terminology.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
60Hypothesis Testing - Basic
The Null Hypothesis for a 2-Sample t Test
The 2-Sample t Test is used to test whether or not the means of two populations are the same.
The null hypothesis is a statement that the population means for the two samples are equal.
Ho: μ1 = μ2
We assume the null hypothesis is true unless we have enough evidence to prove otherwise. We say – we “fail to reject the null”.
If we can prove otherwise, then we “reject the null” hypothesis and accept the Alternative Hypothesis
HA: μ1 ≠ μ2
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
61Hypothesis Testing - Basic
Null Hypothesis for 2-Sample t Test (Cont.)
This is analogous to our judicial system principle of “innocent until proven guilty”
The symbol used for the null hypothesis is Ho:
0: OR : 210210 mmmm HH
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
62Hypothesis Testing - Basic
The Alternative Hypothesis for a 2-Sample t Test
The alternative hypothesis is a statement that represents reality if there is enough evidence to reject Ho.
If we reject the null hypothesis then we accept the alternative hypothesis.
This is analogous to being found “guilty” in a court of law.
The symbol used for the alternative hypothesis is Ha:
0: OR : 2121 mmmm aa HH
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
63Hypothesis Testing - Basic
Our Emergency Response Team Example
In our example, the first hypothesis test will take this form:
21
21
:
:
mm
mm
a
o
H
H
0:
0:
21
21
mm
mm
a
o
H
H
We can rewrite it in this form:
Reminder:We are conducting a 2-Sample t test to determine if the average cycle time of the
Phase 1 response from our two teams are different.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
64Hypothesis Testing - Basic
Our Emergency Response Team Example (Cont.)
If we wanted to specifically test only whether or not there was enough evidence to indicate that team 2’s average was greater than team 1’s, it would take this form:
0:
0:
21
21
mm
mm
a
o
H
H
This is still a 2-Sample t-Test
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
65Hypothesis Testing - Basic
Our Emergency Response Team Example (Cont.)
The second hypothesis test will be a 1-Sample t. It will take this form for each team:
600:
600:
1
1
m
m
a
o
H
H
When you are testing whether or not a population mean is equal to a given or
Target value, you use a 1-Sample t
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
66Hypothesis Testing - Basic
Hypothesis Test in Minitab
We will use Minitab to conduct our hypothesis tests.
Open the Minitab file Response.mtw
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
67Hypothesis Testing - Basic
Hypothesis Test in Minitab:2-Sample t-Test
Select Stat> Basic Statistics> 2-Sample tto compare Team 1 to Team 2
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
68Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
Team 1 and Team 2 are in different columns, so select Samples in different columns
Double click on C1-Supp1Then double click onC2-Supp2 to place them In First and Second boxes
Select Graphs to get the Graphs dialog box
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
69Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
In the Graphs dialog box, check both Boxplots of dataand Dotplots of data
Click OK here, and then click on OK in the previous dialog box
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
70Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
Team 2Team 1
605
604
603
602
601
600
599
598
597
596
Da
ta
Boxplot of Team 1, Team 2
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
71Hypothesis Testing - Basic
Team 2Team 1
605
604
603
602
601
600
599
598
597
596
Da
ta
Individual Value Plot of Team 1, Team 2
Hypothesis Test in Minitab (Cont.)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
72Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
This descriptive output shows up in your Session Window
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
73Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
The null hypothesis states that the difference between the two means is zero
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
74Hypothesis Testing - Basic
Hypothesis Test in Minitab (Cont.)
We will cover p-values in more detail a little later
The p-value here is less than 0.05, so we can reject the null hypothesis
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
75Hypothesis Testing - Basic
Assumptions
The Hypothesis Tests we have discussed make certain assumptions:
Independence between and within samples
Random samples
Normally distributed data
Unknown Variance
In our example, we did not assume equal variances. This is the safe choice. However, if we had reason to believe equal variances, then we could have checked the “Assume equal variances” box in the dialogue box.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
76Hypothesis Testing - Basic
The Risks of Being Wrong
Conclusion Drawn
Accept Ho
The
True
State
Ho True
Ho False
Type I
Error
-Risk)
Type II Error
-Risk)
Correct
Correct
Reject Ho
Error Matrix
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
77Hypothesis Testing - Basic
Type I and Type II Errors
Type I Error
Alpha Risk
Producer Risk
The risk of rejecting the null, and taking action, when no action was necessary
Type II Error
Beta Risk
Consumer Risk
The risk of failing to reject the null when you should have rejected it.
No action is taken when there should have been action.
I’ve missed a significant effect!
I’ve discovered something that really
isn’t here!
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
78Hypothesis Testing - Basic
Type I and Type II Errors (Cont.)
The Type I Error is determined up front.
It is the alpha value you choose.
The confidence level is one minus the alpha level.
The Type II Error is determined from the circumstances of the situation.
If alpha is made very small, then beta increases (all else being equal).
Requiring overwhelming evidence to reject the null increases the chances of a type II error.
To minimize beta, while holding alpha constant, requires increased sample sizes.
One minus beta is the probability of rejecting the null hypothesis when it is false. This is referred to as the Power of the test.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
79Hypothesis Testing - Basic
Type I and Type II Errors (Cont.)
What type of error occurs when an innocent man is convicted?
What about when a guilty man is set free?
Does the American justice system place more emphasis on the alpha or beta risk?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
80Hypothesis Testing - Basic
Exercise
Draw the Type I & II error matrix for airport security.
Do you think the security system at most airports places more emphasis on the alpha or beta risk?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
81Hypothesis Testing - Basic
The p-Value
If we reject the null hypothesis, the p-value is the probability of being wrong.
In other words, if we reject the null hypothesis, the p-value is the probability of making a Type I error.
It is the critical alpha value at which the null hypothesis is rejected.
If we don’t want alpha to be more than 0.05, then we simply reject the null hypothesis when the p-value is 0.05 or less.
As we will learn later, it isn’t always this simple.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
National GuardBlack Belt Training
Power, Delta and Sample Size
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
83Hypothesis Testing - Basic
Beta, Power, and Sample Size
If two populations truly have different means, but only by a very small amount, then you are more likely to conclude they are the same. This means that the beta risk is greater.
Beta only comes into play if the null hypothesis truly is false. The “more” false it is, the greater your chances of detecting it, and the lower your beta risk.
The power of a hypothesis test is its ability to detect an effect of a given magnitude.
Minitab will calculate beta for us for a given sample size, but first let’s show it graphically….
1Power
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
84Hypothesis Testing - Basic
Beta and Alpha
1m
95% Confidence Limit (alpha = .05) for mean, m1
(critical value)
Here is our first population and its corresponding alpha risk.
Alpha Risk
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
85Hypothesis Testing - Basic
Beta and Alpha (Cont.)
1m
95% Confidence Limit (alpha = .05) for mean, m1 (critical
value)
We want to compare these two populations. Do you think that we will easily be able to determine if they are different?
D2m
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
86Hypothesis Testing - Basic
Beta and Alpha (Cont.)
1m
Beta Risk
95% Confidence Limit (alpha = .05) for mean, m1 (critical
value)
If our sample from population 2 is in this grey area, we will not be able to see the difference. This is called Beta Risk.
D
2m
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
87Hypothesis Testing - Basic
Beta and Delta
If we are trying to see a larger change, we have less Beta Risk.
2m
Beta Risk
D1m
95% Confidence Limit (alpha = .025) for
mean, m1 (critical value)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
88Hypothesis Testing - Basic
Beta and Sigma
Now we’re back to our original graphic. What do you think happens to Beta Risk if the standard deviations of the populations decrease?
1m
Beta Risk
95% Confidence Limit (alpha = .05) for mean,
m1 (critical value)
D2m
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
89Hypothesis Testing - Basic
Beta and Sigma (Cont.)
If the standard deviation decreases, Beta Risk decreases.
Reducing variability has the same effect on Beta Risk as increasing sample size.
1m
Beta Risk
D2m
95% Confidence Limit (alpha = .05) for mean, m1
(critical value)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
90Hypothesis Testing - Basic
How Can Power Be Increased?
Power is related to risk, variation, sample size, and the size of change that we want to detect.
If we want to detect a smaller delta (effect), we typically must increase our sample size.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
91Hypothesis Testing - Basic
Example:
Power
Let’s use Minitab to determine the beta risk of the hypothesis test we performed on the two teams.
First, we’ll have to make some assumptions.
We don’t know the TRUE difference in the means, so we’ll assume that it’s 0.682, the differences in the sample averages.
A variance hypothesis test shows that the variances are not equal.
We will average the variances from Minitab to determine the combined variance using the following formula:
2
2
2
2
1 sss
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
92Hypothesis Testing - Basic
Example:
Power (Cont.)
Select; Stat> Power and Sample Size>2-Sample t...
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
93Hypothesis Testing - Basic
Example:
Power (Cont.)
To calculate Power, we need three things;1. Sample Size2. The Difference between the two Means3. The Average Standard Deviation of the two samples
We can get all this information from our 2-Sample t-Test conducted earlier:
1. Sample Size = 100
2. Difference Between Means = 0.682 (600.230 – 599.548 = 0.682)
3. Average Standard Deviation ??(See Next Slide)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
94Hypothesis Testing - Basic
Example:
Power (Cont.)
To Calculate Average Standard Deviation
Remember that Standard Deviations are the Square Roots of the Variance. Since square roots are not additive (we cannot add them and divide by two) we have to convert them back to Variances which are additive.
StDev Squared = VarianceTeam 1 0.619 squared = 0.3832Team 2 1.870 squared = 3.4969
Sum = 3.8801Divide by 2 to get Average = 1.9401
And Square Root of Average = 1.3929
So the Average Standard Deviation for the two samples is 1.3929
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
95Hypothesis Testing - Basic
Example:
Power (Cont.)
1. Type in Sample Size of 100 here
2. Type in Difference Between Means of 0.682 here
3. Type in Average Standard Deviationof 1.393 here
4. Click on OK
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
96Hypothesis Testing - Basic
Example:
Power (Cont.)
If the TRUE difference between the two support orgs. was 0.682, we would have a 6.88% chance of not observing this
and therefore concluding that they are the same.
The Power = 0.9312And since Beta = (1 –Power)
Beta = 0.0688.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
97Hypothesis Testing - Basic
Example:
Power (Cont.)
In practice, we evaluate the power of a test to determine its ability to detect a difference of a given magnitude that we deem important, or practically significant.
For example, we could calculate the power of a hypothesis test to see if we could measure a one minute difference in responsiveness between the two teams.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
98Hypothesis Testing - Basic
Example:
Power (Cont.)
Let’s say that if the two support organizations’ cycle times differ by as little as 0.4 seconds, then we need to analyze the reasons for the differences.
What is the power of our test to detect this difference?
What is the probability of making a type II error (concluding that there is no difference when one exists)?
Use Minitab to individually answer these questions.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
99Hypothesis Testing - Basic
Exercise:
Sample Size
Now that we understand the relationship between Beta, Power, Delta, and Sample Size, we can use this information to calculate the sample size necessary to give us the information we want.
We simply use the same function in Minitab to solve for sample size rather than power.
This is a very useful and common extension of Hypothesis Testing.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
100Hypothesis Testing - Basic
Exercise:
Sample Size (Cont.)
Here we enter the Difference (delta) we wish to detect, and the minimum Power value that we are willing to live with.
We leave Sample sizesblank.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
101Hypothesis Testing - Basic
Exercise:
Sample Size
Let’s extend our response team cycle time example
Determine what sample size we would need to detect a difference of 0.4 seconds at a power of 0.90.
What about at a power of 0.95?
What about at a power of 0.95 and an alpha of 0.025?
Hint: Click the Options button in the Power and Sample Size dialogue box.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
102Hypothesis Testing - Basic
Other Power and Sample Size Scenarios
We can perform these calculations not only for the difference between
two means, but for other tests as well.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
103Hypothesis Testing - Basic
1-Sample t-test in Minitab
Now, we will return to Minitab to test the following hypothesis about our two support organizations cycle times:
600:
600:
1
1
m
m
a
o
H
H
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
104Hypothesis Testing - Basic
Back to the Support Organization Example:One Sample t-Test
1-Sample t-test in Minitab
Choose Stat>Basic Statistics>1-Sample tto test the mean of each response team against a standard or spec
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
105Hypothesis Testing - Basic
1-Sample t-test in Minitab
Double click on C1 Team 1 and C2 Team 2 to place them in the dialog box here.
Type in the Hypothesized mean, or standard we are comparing to. Here it is 600.
Click the Graphs button to get to the Graphs dialog box.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
106Hypothesis Testing - Basic
1-Sample t-test in Minitab (Cont.)
Select Histogramof data andBoxplot of data
Click OK here and onthe previous Screen
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
107Hypothesis Testing - Basic
601.0600.5600.0599.5599.0598.5598.0
35
30
25
20
15
10
5
0
-5
X_
Ho
Team 1
Fre
qu
en
cy
Histogram of Team 1(with Ho and 95% t-confidence interval for the mean)
1-Sample t-test in Minitab
This shows the Target we are testing, along with the Average and theConfidence Intervalfrom the data.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
108Hypothesis Testing - Basic
1-Sample t-test in Minitab (Cont.) - adj
601.5601.0600.5600.0599.5599.0598.5598.0
X_
Ho
Team 1
Boxplot of Team 1(with Ho and 95% t-confidence interval for the mean)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
109Hypothesis Testing - Basic
Team 2
Fre
qu
en
cy
603.0601.5600.0598.5597.0
15.0
12.5
10.0
7.5
5.0
2.5
0.0X_
Ho
Histogram of Team 2(with Ho and 95% t-confidence interval for the mean)
1-Sample t-test in Minitab (Cont.)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
110Hypothesis Testing - Basic
1-Sample t-test in Minitab (Cont.)
605604603602601600599598597596
X_
Ho
Team 2
Boxplot of Team 2(with Ho and 95% t-confidence interval for the mean)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
111Hypothesis Testing - Basic
1-Sample t-test in Minitab (Cont.)
Here is the descriptive output for the 1-Sample t-Test found in Session Window
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
112Hypothesis Testing - Basic
2-Sided and 1-Sided Hypothesis Tests
We have concentrated on 2-sided hypothesis tests.
2-Sided tests determine whether or not two items are equal or whether a parameter is equal to some value.
Whether an item is less than or greater than another item or a value is not sought up front. A 2-sided test is a less specific test.
The alternative hypothesis is “Not Equal”.
Everything we have learned also applies to 1-sided tests.
1-Sided tests determine whether or not an item is less than (<) or greater than (>) another item or value.
The alternative hypothesis is either (<) or (>).
This makes for a more powerful test (lower beta at a given alpha and sample size).
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
113Hypothesis Testing - Basic
More Detailed Information
Remember to use the Stat Guide button to learn more about the results and to help you interpret them.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Hypothesis Test Summary Template
Hypothesis Test (ANOVA, 1 or 2 sample t - test, Chi Squared,
Regression, Test of Equal Variance, etc)
Factor (x)
Testedp Value Observations/Conclusion
Example: ANOVA Location 0.030
Significant factor - 1 hour driving time from DC
to Baltimore office causes ticket cycle time to
generally be longer for the Baltimore site
Example: ANOVA Part vs. No Part 0.004
Significant factor - on average, calls requiring
parts have double the cycle time (22 vs 43
hours)
Example: Chi Squared Department 0.000
Significant factor - Department 4 has digitized
addition of customer info to ticket and less
human intervention, resulting in fewer errors
Example: Pareto Region n/a
South region accounted for 59% of the defects
due to their manual process and distance from
the parts warehouse
Describe any other observations about the root cause (x) data
Optional BB Deliverable- Example -
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
No P
art
Part
0
50
100
150
Part/No Part
Net H
ours
Call
Open
Boxplots of Net Hour by Part/No
(means are indicated by solid circles)
Analysis of Variance for Net Hour
Source DF SS MS F P
Part/No 1 7421 7421 8.65 0.004
Error 69 59194 858
Total 70 66615
Individual 95% CI's For Mean
Level N Mean StDev --+---------+---------+---------+----
No Part 27 21.99 19.95 (--------*---------)
Part 44 43.05 33.70 (------*------)
--+---------+---------+---------+----
Pooled StDev = 29.29 12 24 36 48
After further investigation, possible reasons proposed by the team are OEM backorders, lack of technician certifications and the distance from the OEM to the client site. It is also caused by the need for technicians to make a second visit to the end user to complete the part replacement. Next step will be for the team to confirm these suspected root causes.
Boxplot: Part/ No Part Impact on Ticket Cycle Time
Because the p-value <= 0.05, we can be confident that calls requiring parts do have an impact on the ticket cycle time.
One-Way ANOVA Template
Optional BB Deliverable
- Example -
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Linear Regression Template
95% confident that 94.1% of the variation in “Wait Time” is from the “Qty of Deliveries”
Deliveries
Wa
it T
ime
353025201510
55
50
45
40
35
S 1.11885
R-Sq 94.1%
R-Sq(adj) 93.9%
Fitted Line PlotWait Time = 32.05 + 0.5825 Deliveries
Optional BB Deliverable
- Example -
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
117Hypothesis Testing - Basic
Takeaways
Since it is not always practical or possible to measure every item in the population, you take a random sample.
A basic understanding of the terms: Population, Sample, Population Parameter, Sample Statistic, Sample Mean, and Sample Standard Deviation
How to calculate a confidence interval with the population standard deviation known
How to calculate a confidence interval with the population standard deviation unknown
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
118Hypothesis Testing - Basic
Takeaways (Cont.)
How Hypothesis tests help us handle uncertainty
The role of sample size, variation, and confidence level
The null and alternative hypotheses
Type I and Type II errors
Hypothesis tests in Minitab
Stat Guide
p-value
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
119Hypothesis Testing - Basic
Takeaways (Cont.)
How to conduct a 1-way and 2-way t-test
How to conduct a Variance test (see Appendix)
How to conduct a Paired t-test (see Appendix)
Understanding of 1-way and 2-way test of proportions (see Appendix)
Understanding the relationship between Power and sample size and detectable difference (delta)
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
What other comments or questions
do you have?
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
121Hypothesis Testing - Basic
References
Hildebrand and Ott, Statistical Thinking for Managers, 4th Edition
Kiemele, Schmidt, and Berdine, Basic Statistics, 4th Edition
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
National GuardBlack Belt Training
APPENDIX
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Step 1: Define the problem objective
Step 2: Determine what data to collect (continuous or attribute)
Step 3: Based on data type, determine the appropriate hypothesis test to use
Step 4: Specify the null (H0) hypothesis and the alternative (H1) hypothesis
Step 5: Select a significance level (degree of risk acceptable), usually 0.05
Step 6: Execute Data Collection plan from step 2
Step 7: From the sample, conduct the hypothesis test using a statistical tool
Step 8: Identify the p-value
Step 9: Compare the p-value to the significance level - if the p-value is less than or equal to your acceptable risk (your alpha), then the null hypothesis is rejected
Step 10: Translate the decision to the situation
Hypothesis Testing - Steps
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree Matrix
Data Type(Step 2)
Hypothesis to be Tested (Step 3) Tree
Variable Testing equality of population MEAN (average) to a specific value 1
Variable Testing equality of population MEANS (averages) from two populations 2
Variable Testing equality of population MEANS (averages) from more than two populations 3
VariableTesting equality of population VARIANCES (standard deviation) from more than two
populations 4
Attribute - Binomial "Go/No-Go"
"Pass/Fail" or "Defective" Data
Testing equality of population PROPORTIONS (binomial data; e.g., pass/fail, go/no go, is/is not, etc.) from one or more populations 5
Attribute - Poisson"Count" or
"Defects" data
Testing equality of population PROPORTIONS (Poisson data; i.e., frequency of occurence in time or space) from two or more populations 6
Attribute (Contingency Table Data)
Testing for ASSOCIATION (not necessarily causal)Note: For use with attribute data only. For variable data, use correlation
or regression. No decision tree required.7
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 1
Start
Is
n > 30?1-Sample Z-test
Stat > Basic Statistics > 1-Sample ZYes
Ispopulationnormally
distributed?(Anderson-
Darling)
No
1-Sample Wilcoxon testrandom sample from a continuous,
symmetric population
Stat > Nonparametrics > 1-Sample Wilcoxon
No
1-Sample t-test(reasonably robust
against normality assumption)
Stat > Basic Statistics > 1-Sample t
Yes
Testing Equality of Population Mean
to a Specific Value
Variable (Continuous)
Has the average button diameter from the welder
changed from its historical value?
Application:
Type of Data:
Example:
Example1.vsd 6-1-00
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 2
2-Sample Mann Whitney(independent, random variables from two
populations with same shape, same variance)
Stat > Nonparametrics > Mann-Whitney
Note: If the two populations have different shapes
or different standard deviations, then use:
2-Sample t-testwithout pooling variances
Start
2-Sample Z-testStat > Basic Statistics > 2-Sample t
Yes
Are bothpopulations
normallydistributed?(Anderson-
Darling)
No
2-Sample t-testwith pooled variances
(reasonably robust against normality
assumption)
Stat > Basic Statistics > 2-Sample t
Assume equal variances
Testing Equality of Means
from Two Populations
Variable (Continuous)
Is the average button diameter from Welder A
different from that of Welder B?
Application:
Type of Data:
Example:
Are the
two samples
dependent?
EqualVariances?
(F-test)
Paired t-test(samples from normal distribution)
Stat > Basic Statistics > Paired t
Yes
2-Sample t-testwithout pooling variances
Stat > Basic Statistics > 2-Sample t
(Do not assume equal variances)
No
Do
n1 and n2
both exceed
30?
Yes
No
Yes
No
Example2.vsd 6-1-00
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 3
Start
Do samplescontain outliers?
(Box Plot)
Mood's Median test(independant, random samples f rom continuous
distributions hav ing same shape)
Stat > Nonparametrics > Mood's Median test
Yes
No
Testing Equality of Means from
More than Two Populations
Variable (Continuous)
Do the average button diameters from
Welders A, B and C differ from one another?
Application:
Type of Data:
Example:
Kruskal-Wallis(independant, random samples
f rom continuous distributions
hav ing same shape)
Stat > Nonparametrics > Kruskal-Wallis
Are thepopulations
normallydistrubted?(Anderson-
Darling)
One-Way Analysis of Variance
(ANOVA)(reasonably robust against assumptions
of normality and equal v ariances)
Stat > ANOVA > One-way
Yes
No
Did the test showsignif icance?
Tukey's testto conduct pairwise comparisons
Stat > ANOVA > One-way
Comparisons: Tukeys
No
Yes
Stop
Note: Use Dunnett's Method ifcomparing treatments to a control.
Example3.v sd 6-1-00
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 4
Start
Levene's testStat > ANOVA > Test for Equal Variances
2
More than 2
No
Bartlett's testStat > ANOVA > Test for Equal Variances
Yes
Testing Equality of Variances
Variable (Continuous)
Do the variances in button diameter
from the three welders differ from one another?
Application:
Type of Data:
Example:
Example4.v sd 6-1-00
How manypopulations are
being compared?
Are the
populations
normally
distributed?
(Anderson-
Darling)
Are the
populations
normally
distributed?
(Anderson-
Darling)
F-testStat > Basic Statistics >2 Variances
Yes
NoLevene's test
Stat > Basic Statistics > 2 Variances
Note: The F-test and Bartlett's test are not robust
against the normality assumption.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 5
Case 2:
Testing Equality of
Proportions from Two
Populations
Example: Are Lines 1 and 2
running at the same
% defective rate?
Testing Equality of Population
Proportions
Attribute (Discrete) - Binomial Distribution
Application:
Type of Data:
Stat > Basic Statistics > 1-Proportion
Stat > Basic Statistics > 2-Proportions
Ho:P1=P2 no difference in popluation
proportions
M iniTab - Options select pooled p
Case 1:
Testing Population Proportion
Against a Specific Value
Example: Has the % defective rate
on Line 1 changed
from its historical value?
Use Chi-Square testMiniTab
Stat>Tables>Chi-square test
Case 3:
Testing Equality of
Proportions from More than
Two Populations
Example: Are Lines 1, 2 and 3
running at the same
% defective rate?
Example5.vsd 5-10-01
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 6
Testing Equality of Population
Defect Rates
Attribute (Discrete) - Poisson Distribution
Application:
Type of Data:
Comparing more than two Poisson
Distributions
1) Is the number of errors on invoices different
between Dept. A and Dept. B?
2) Does the number of seat defects
differ among shifts 1, 2 and 3?
Examples:
Example6.vsd 5-10-01
Comparing two Poission
Distributions
Use One-Way Analysis of Variance
Stat > ANOVA > One-way
Use 2 Sample t-test
Stat > Basic Stat > 2 Sample t
Caution
No Extreme Outliers
Stat > Basic Stats > 2-sample Poisson Rate
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
Decision Tree # 7
Testing for Association
Attribute (Contingency Table Data)
Application:
Type of Data:
Chi-square test
Minitab:Stat > Tables > Chi-square test
Does the type of defect that occurs
depend on which product is being produced?Example:
Example7.vsd 6-1-00
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
132Hypothesis Testing - Basic
Group 1:
Hypothesis Tests for Variation
Use the Minitab electronic docs, stat guide, and help to learn about performing hypothesis tests for equality of variance among two populations. You may also use your textbooks if you wish.
Prepare a 10-15 minute teachback on hypothesis tests for variation.
Be sure to work an example in your teachback.
Hint: Using Minitab to conduct a hypothesis test to determine if there is a difference in the amount of variation exhibited by each support organization would be a good example to use.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
133Hypothesis Testing - Basic
Group 2:
Paired t-Test
Use the Minitab electronic docs, stat guide, and help to learn about performing paired t-tests. You may also use your textbooks if you wish.
Prepare a 10-15 minute teachback on paired t-tests.
Be sure to illustrate the difference between a standard 2-way t-test and a paired t-test.
Be sure to work an example in your teachback.
Go through a sample size calculation in your example.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
134Hypothesis Testing - Basic
Group 3:
Hypothesis Tests with Proportions
Use the Minitab electronic docs, stat guide, and help to learn about performing hypothesis tests with proportions. You may also use your text books if you wish.
Prepare a 10-15 minute teachback on hypothesis tests with proportions.
Be sure to illustrate the main difference between hypothesis tests of proportions and the other hypothesis tests we have talked about.
Include both 1-way and 2-way proportion hypothesis tests.
Be sure to work an example in your teachback.
Go through a sample size calculation in your example.
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
135Hypothesis Testing - Basic
Confidence Interval Formulas
Confidence Intervals for:
Mean (s Known) Mean (s Unknown)
Standard Deviation
Proportions (Approximate)
n
σZx α/2
n
stx 1nα/2,
2
1nα/2,1
2
1nα/2,
1nsσ
1ns
n
ppZpp
n
ppZp
ˆ1ˆˆ
ˆ1ˆˆ
2/2/
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
136Hypothesis Testing - Basic
Table of Normal Curve Areas
Source: Statistical Thinking for Managers, Hildebrand and Ott, 4th Edition, page 800.
z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.00 0.0000 0.0040 0.0080 0.0120 0.0160 0.0199 0.0239 0.0279 0.0319 0.0359
0.10 0.0398 0.0438 0.0478 0.0517 0.0557 0.0596 0.0636 0.0675 0.0714 0.0753
0.20 0.0793 0.0832 0.0871 0.0910 0.0948 0.0987 0.1026 0.1064 0.1103 0.1141
0.30 0.1179 0.1217 0.1255 0.1293 0.1331 0.1368 0.1406 0.1443 0.1480 0.1517
0.40 0.1554 0.1591 0.1628 0.1664 0.1700 0.1736 0.1772 0.1808 0.1844 0.1879
0.50 0.1915 0.1950 0.1985 0.2019 0.2054 0.2088 0.2123 0.2157 0.2190 0.2224
0.60 0.2257 0.2291 0.2324 0.2357 0.2389 0.2422 0.2454 0.2486 0.2517 0.2549
0.70 0.2580 0.2611 0.2642 0.2673 0.2704 0.2734 0.2764 0.2794 0.2823 0.2852
0.80 0.2881 0.2910 0.2939 0.2967 0.2995 0.3023 0.3051 0.3078 0.3106 0.3133
0.60 0.3159 0.3186 0.3212 0.3238 0.3264 0.3289 0.3315 0.3340 0.3365 0.3389
1.00 0.3413 0.3438 0.3461 0.3485 0.3508 0.3531 0.3554 0.3577 0.3599 0.3621
1.10 0.3643 0.3665 0.3686 0.3708 0.3729 0.3749 0.3770 0.3790 0.3810 0.3830
1.20 0.3849 0.3869 0.3888 0.3907 0.3925 0.3944 0.3962 0.3980 0.3997 0.4015
1.30 0.4032 0.4049 0.4066 0.4082 0.4099 0.4115 0.4131 0.4147 0.4162 0.4177
1.40 0.4192 0.4207 0.4222 0.4236 0.4251 0.4265 0.4279 0.4292 0.4306 0.4319
1.50 0.4332 0.4345 0.4357 0.4370 0.4382 0.4394 0.4406 0.4418 0.4429 0.4441
1.60 0.4452 0.4463 0.4474 0.4484 0.4495 0.4505 0.4515 0.4525 0.4535 0.4545
1.70 0.4554 0.4564 0.4573 0.4582 0.4591 0.4599 0.4608 0.4616 0.4625 0.4633
1.80 0.4641 0.4649 0.4656 0.4664 0.4671 0.4678 0.4686 0.4693 0.4699 0.4706
1.90 0.4713 0.4719 0.4726 0.4732 0.4738 0.4744 0.4750 0.4756 0.4761 0.4767
2.00 0.4772 0.4778 0.4783 0.4788 0.4793 0.4798 0.4803 0.4808 0.4812 0.4817
2.10 0.4821 0.4826 0.4830 0.4834 0.4838 0.4842 0.4846 0.4850 0.4854 0.4857
2.20 0.4861 0.4864 0.4868 0.4871 0.4875 0.4878 0.4881 0.4884 0.4887 0.4890
2.30 0.4893 0.4896 0.4898 0.4901 0.4904 0.4906 0.4909 0.4911 0.4913 0.4916
2.40 0.4918 0.4920 0.4922 0.4925 0.4927 0.4929 0.4931 0.4932 0.4934 0.4936
2.50 0.4938 0.4940 0.4941 0.4943 0.4945 0.4946 0.4948 0.4949 0.4951 0.4952
2.60 0.4953 0.4955 0.4956 0.4957 0.4959 0.4960 0.4961 0.4962 0.4963 0.4964 z area
2.70 0.4965 0.4966 0.4967 0.4968 0.4969 0.4970 0.4971 0.4972 0.4973 0.4974 3.5 0.49976737
2.80 0.4974 0.4975 0.4976 0.4977 0.4977 0.4978 0.4979 0.4979 0.4980 0.4981 4.0 0.49996833
2.90 0.4981 0.4982 0.4982 0.4983 0.4981 0.4984 0.4985 0.4985 0.4986 0.4986 4.5 0.49999660
3.00 0.4987 0.4987 0.4987 0.4988 0.4988 0.4989 0.4989 0.4989 0.4990 0.4990 5.0 0.49999971
Source: Computed by P.J. Hildebrand.
0 z
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
137Hypothesis Testing - Basic
Calculation of t Test Statistic
The t test statistic is calculated as follows:
where D0 is the hypothesized difference between the two population means.
For an assumption of unequal variances:
21
021
XXs
XXt
D
)(
2
2
2
1
2
1
21 n
s
n
ss
XX
UNCLASSIFIED / FOUO
UNCLASSIFIED / FOUO
138Hypothesis Testing - Basic
Calculation of t Test Statistic
For an assumption of equal variances:
where
21
1121 nn
ss pXX
2
)1()1(
21
2
22
2
11
nn
snsns p