Professional Documents
Culture Documents
A. CHAPTER OBJECTIVES
B. INTRODUCTION
C. SAMPLING
D. FUNDAMENTALS OF IMPROVEMENT
E. MEASURES OF CENTRAL TENDENCY
F. MEASURES OF VARIABILITY
G. NORMAL PROBABILITY TEST
H. HANDLING NON-NORMAL DATA
I. SAMPLING DISTRIBUTIONS
J. CENTRAL LIMIT THEOREM
K. STANDARD NORMAL PROBABILITY
DISTRIBUTION
CHAPTER OBJECTIVES
INTRODUCTION
We want information on the entire population.
Since this is not always possible nor economically
realistic, we use statistics.
Statistics enable us to take a sample from the population
and estimate a characteristic about the population.
It is important to differentiate the sample statistics
(estimations) from the population parameters.
SAMPLING
X = Sample Mean
= Population Mean
s = Sample
Standard Deviation
Statistics
= Population Standard
Deviation
Estimate
Parameters
FUNDAMENTALS OF IMPROVEMENT
Variability & stability are used to determine the status of a
process.
We use the mean () to determine if process is on target.
We use standard deviation () to determine variability.
Stability helps us to determine how well a process
performs over time.
Stability is represented by a constant mean and
predictable variability over time.
Every process displays variation; some display controlled
variation while others display uncontrolled variation (Walter
Shewart).
VARIATION
Control chart A displays controlled variation; stable &
consistent pattern of variation over time.
Control chart B displays uncontrolled variation; variation
that changes over time.
X-Bar
Chart for Process A
X - B a r C h a r t fo r P ro c e s s A
U C L = 7 7 .2 0
80
U C L = 7 7 .2 0
U CL = 7 7 .2 7
75
M ea n
X = 7 0 .9 1
70
X = 7 0 .9 1
70
X =70.98
70
S a m p le
S a m p le M e a n
S a m pl e M e a n
75
L C L = 6 4 .7 0
60
65
L C L = 6 4 .6 2
65
L C L = 6 4 .6 2
5
0
10
5
15
10
20
15
S a m p l e N u mb e r
25
20
50
0
25
10
15
2 0
25
S a m p le N u m b e r
S a m p l e N u mb e r
VARIATION - cont.
Variation will be present in any process and can be tolerated if:
The variation of the output is relatively small compared to the
process specifications and the process is on target.
The process is stable over time.
USL
USL
Old
Old
Thinking
Thinking
Cost
Acceptable
LSL
LSL
Cost
Nom
Nom
Nom
Nom
USL
USL
New
New
Thinking
Thinking
Mean
Median
Mode
10
MEAN
Mean ( - population, X - sample) is the arithmetic average of the data
values (X 1, X2, X3. XI) which is expressed as follows.
Population - =
Sample - X
Brian Sr.
11
MEDIAN
Single value from data set that measures the central item
in the data.
Single item is the middle most or most central item in data
set.
Half the items lie above this point; other half lie below it.
Reflects 50% rank or center number after data set has
been sorted from high to low.
Median is Robust to extreme values.
12
MODE
Brian Jr.
13
M e a n , M e d ia n ,
M ode
Mean is more
representative of the
location of the
distribution.
S y m m e tr ic a l
d is tr ib u tio n
M ode
M e d ia n
M ean
S k e w e d to le ft
M ode
M e d ia n
M ean
S k e w e d to r ig h t
14
APPLICATION EXERCISE
15
Dialog Window
Click OK.
16
SESSION WINDOW
Results for: DISTSKEW.MTW
Descriptive Statistics: Norm, Pos Skew, Neg Skew
Variable
Norm
Pos Skew
Neg Skew
N
500
500
500
N*
0
0
0
Variable
Norm
Pos Skew
Neg Skew
Maximum
103.301
130.366
77.106
Mean
70.000
70.000
70.000
SE Mean
0.447
0.447
0.447
StDev
10.000
10.000
10.000
Minimum
29.824
62.921
1.866
Q1
63.412
63.647
67.891
Median
69.977
65.695
73.783
Q3
76.653
72.821
76.290
18
Dialog Window
Select
Simple
OK
Click OK.
19
HISTOGRAMS
Histogram of Neg Skew
250
140
120
200
Frequency
Frequency
100
150
100
80
60
40
50
20
0
12
24
36
Neg Skew
48
60
72
70
80
90
100
Pos Skew
110
120
Histogram of Norm
70
Project Manager
Select all 3 Graphs
Right click Selected Graphs
Select Tile
50
Frequency
Graphs
60
40
30
20
10
0
30
40
50
60
70
Norm
80
90
100
20
130
MEASURES OF VARIABILITY
Mean, median & mode tell us only part of what we need to
know about the characteristics of data.
We must also measure dispersion (spread) or variability.
Three measures of variability will be reviewed. Range,
variance and standard deviation.
21
RANGE
22
VARIANCE
Variance (2 population, s2 sample) - sum of the squared
distances between the mean and each item divided by the
total number of elements in the population.
Formulas:
Population
( X )
N
Sample
(X X )
n 1
23
STANDARD DEVIATION
Standard deviation ( population, s sample) - quantifies
data variability and is the square root of the variance.
Enables us to determine where the values of a frequency
distribution are located relative to the mean.
This can be shown with a normal curve and related
probability areas.
24
STANDARD DEVIATION
Formulas:
Population
Sample
( X )
( X X )
n 1
An important Six Sigma principle indicates that the total variation (variance) of
a process output variable can be partitioned into the variation due to the
process inputs of the process as follows:
2
26
27
28
29
Normal
Normal
Mean
StDev
N
AD
P-Value
99
Percent
95
90
70.00
10.00
500
0.418
0.328
p=
.328
80
70
60
50
40
30
20
99.9
95
90
P<
.005
80
70
60
50
40
30
20
10
10
0.1
Mean
70.00
StDev
10.00
N
500
AD
44.491
P-Value <0.005
99
Percent
99.9
0.1
30
40
50
60
70
Norm
80
90
100
110
20
40
60
Neg Skew
80
100
Mean
70.00
StDev
10.00
N
500
AD
46.489
P-Value <0.005
99
Percent
95
90
80
70
60
50
40
30
20
P<
.005
10
5
1
0.1
40
50
60
70
80
90
Pos Skew
100
110
120
130
30
MYSTERY DISTRIBUTION
Following the previous procedure, generate a normal probability plot for the
Mystery variable in C4 column of the Distskew.mtw file.
31
Graphical Summary
32
Click OK
33
RESULT
Minitab provides a histogram along with the related
descriptive statistics.
Summary for Mystery
Anderson-Darling Normality Test
A-Squared
P-Value <
Mean
StDev
Variance
Skewness
Kurtosis
N
40
60
80
100
120
140
Minimum
1st Quartile
Median
3rd Quartile
Maximum
160
27.11
0.005
100.00
32.38
1048.78
0.00716
-1.63184
500
41.77
68.69
104.20
130.81
162.82
102.85
117.66
30.49
34.53
Mean
Median
80
90
100
110
120
34
Typical examples:
Two different machines provide a bimodal distribution. As a result,
analyze the data for each machine separately.
Data comes from an unstable process. As a result, the process must
be stabilized before reliable statistical results can be obtained.
35
In instances where the process is stable and predictable and the data proves to be
nonnormal there are a couple of options
Normalize the data via a transformation (transformations are beyond the scope of
our training)
Utilize a nonnormal probability model (weibull, lognormal, exponential, etc) to
analyze overall capability (Pp, Ppk, PPU, and PPL)
36
Mean
StDev
N
AD
P-Value
99
95
90
Percent
80
70
60
50
40
30
20
10
5
1
0.1
-2
-1
2
Dist3
37
0.9100
0.8654
500
19.095
<0.005
INDIVIDUAL DISTRIBUTION
IDENTIFICATION
Using Cltest.mtw, perform an Individual Distribution
Identification test as follows:
1. Stat > Quality Tools > Individual Distribution Identification
2. In the dialog window
Enter C3 Dist3 into single column.
Select Specify
Use the default distributions ( Normal, Exponential,
Weibull, Gamma).
Note: We will use these settings to simplify the example.
We could have used the Use all distributions option to
look at 10 additional distributions. However, in this instance
we know that one of these will Best Fit the distribution.
OK
38
39
CAPABILITY ANALYSIS
Now that we are comfortable with the probability that our data fits the
Exponential distribution we are able to perform a capability analysis.
Using Cltest.mtw, we will perform a capability analysis for Nonnormal
data and fit data with Exponential distribution.
1. Stat > Quality Tools > Capability Analysis > Nonnormal
2. In the dialog window
Enter C3 Dist3 into single column.
In Fit data with
Select Distribution
Select Exponential
Lower spec Enter 0
Upper spec Enter 3
Click OK
Next slide has graphs
40
Ppk = 0.44
USL
Process Data
LSL
0.00000
Target
*
USL
3.00000
Sample Mean 0.90997
SampleN
500
Mean
0.90997
O verall Capability
Pp
0.50
PPL
1.00
PPU
0.44
Ppk
0.44
Exp. O verall Performance
PPM<LSL
0.0
PPM>USL 37000.5
PPMTotal
37000.5
O bserved Performance
PPM<LSL
0
PPM>USL 32000
PPMTotal
32000
0.0
0.8
1.6
2.4
3.2
4.0
4.8
41
SAMPLING DISTRIBUTIONS
If we selected 10 groups of 25 samples from a continuous process
& computed the mean length and standard deviation of the length of
each sample group, the mean and standard deviation of each
sample group would be different.
Sampling distribution of the mean - a probability distribution of all
the possible means of the samples.
Sampling distributions can be partially described by its mean and
standard deviation.
Rather than say standard deviation of the distribution of sample
means we call it the standard error of mean.
Standard error indicates size of the chance error and the accuracy
we will likely get if we use it a sample statistic to estimate a
population parameter.
42
B.
C.
A.
43
sample size ( X
).
n
The equation for the standard error (standard deviation) of the mean
for an infinite population is:
44
45
GRAPHICAL EXERCISE
Turn to the population graphical
exercise located on the next page of
your student manual and proceed as
follows:
1. Select 2 dots (at random).
2. Using the selected dots, draw a
new dot in-between the two.
3. Repeat steps 1 and 2 until all
preprinted dots are used only
once.
4. Circle the new dots, ignoring the
original dots.
47
GRAPHICAL EXERCISE
Questions:
1.
Is the spread of the new population different from the original?
2.
What about the shape?
3.
What differences are there between the original population of
dots and the population resulting from the subgroup?
48
1.
2.
Variables: select C1
B. Choose Stat > Control Charts > Variables Charts for Subgroups > Xbar.
Select C1 Output
Select Point Plotted stores the subgroup mean in worksheet for analysis
C.
3.
Click Graphs
49
I Chart of Output
100
UCL=80.70
80
UCL=96.59
90
70
_
_
X=68.28
65
Individual Value
Sample Mean
75
80
_
X=68.28
70
60
50
60
LCL=55.86
55
3
12
15
18
Sample
21
24
27
30
40
LCL=39.97
1
15
30
45
60
75
90
Observation
105
120
135
150
N
150
30
N*
0
0
Descriptive Statistics
Mean SE Mean StDev Minimum
68.280
0.776 9.498
43.000
68.280
0.858 4.701
58.000
Median
68.000
67.600
Maximum
92.000
80.800
50
51
N
500
500
100
100
N*
0
0
0
0
Mean
0.90016
0.90005
0.90016
0.90005
SE Mean
0.00445
0.00291
0.00408
0.00311
StDev
0.09952
0.06497
0.04082
0.03106
Minimum
0.56399
0.62989
0.79356
0.82731
Median
0.89696
0.91351
0.90219
0.90221
Maximum
1.24185
0.99842
0.97541
0.97392
52
Normal
Normal
99.9
Mean
StDev
N
AD
P-Value
99
80
70
60
50
40
30
20
0.9002
0.09952
500
0.213
0.852
95
90
80
70
60
50
40
30
20
10
10
0.1
Mean
StDev
N
AD
P-Value
99
Percent
Percent
95
90
99.9
0.1
0.5
0.6
0.7
0.8
0.9
Dist1
1.0
1.1
1.2
1.3
P VALUE = .852
0.80
0.85
0.90
Mean1
0.95
1.00
1.05
P VALUE = .348
53
0.9002
0.04082
100
0.404
0.348
Normal
Normal
99.9
Mean
StDev
N
AD
P-Value
99
80
70
60
50
40
30
20
0.9001
0.06497
500
10.132
<0.005
95
90
80
70
60
50
40
30
20
10
10
0.1
Mean
StDev
N
AD
P-Value
99
Percent
Percent
95
90
99.9
0.1
0.6
0.7
0.8
0.9
1.0
1.1
Dist2
P VALUE = <0.005
0.80
0.85
0.90
Mean2
0.95
1.00
P VALUE = .232
54
0.9001
0.03106
100
0.478
0.232
In Chapter 8, you were introduced to the standard normal table for determining
the area under the normal curve and how to determine a Sigma level (Z value).
We can also use the normal table to compute the probability (area under the
curve) of being within a certain distance (ie. Spec limits) from the mean in units
of standard deviation (Z values).
Z standard transform equation produces a value from a distribution where
mean=0 and =1.
(X X )
Z
Z value indicates how far the number isfrom the mean in units of standard
deviations (Z).
For estimating a process yield, we can substitute the upper and lower spec
limit for X in the equation. We can calculate the proportion of product that is
out-of-spec.
55
Z TRANSFORM EXAMPLE
56
Z TRANSFORM EXAMPLE
To perform the calculations, lets proceed as follows.
1. Calculate the Z score for each specification limit (upper and lower).
Z
( LSL X )
(USL X )
(.9 1.03)
.0573
(1.1 1.03)
.0573
Z 2.27
Z 1.22
2. Calculate the areas below the lower specification and above the upper
specification using the normal table.
Table A (Area Under the Standardized Normal Curve) located in the Gbdata
file gives us an area of .0116 (1.16%) for a Z value of 2.27 (disregard the
negative sign) and it gives us an area of .1112 (11.12%) for a Z value of 1.22.
If we add these 2 area under the curve together, we get 12% (.0116 + .1112 =
.1228 or 12%).
57
Z TRANSFORM EXAMPLE
This is shown graphically below.
.0116
.1112
+
.9
LSL
Z=-2.27
1.03
x
1.1
USL
Z=1.22
58
59
60
.1228
-1.161
61
62
RESULTS OF INDIVIDUAL
Z TRANSFORM EXERCISE
Results of A are as follows:
First, calculate the Z value for X = $6 million.
(X X )
6 10.1
Z
2.6
Where; Z
Z 1.57
63
RESULTS OF INDIVIDUAL
Z TRANSFORM EXERCISE
The area below is between Z=-1.57 and left-hand tail
(disregard minus sign).
.0582
+
6
Z=-1.57
10.1
64
RESULTS OF INDIVIDUAL
Z TRANSFORM EXERCISE
Results of B are as follows:
First, calculate the Z values for X = $6 million.
Z = -1.57 (as indicated in A)
Then, calculate the Z value for X = $14 million as follows.
Z
(X X )
(14 10.1)
2.6
Z 1.5
Next, using Table A, calculate the probabilities for each of the Z values.
Z = -1.57 is equal to an area of .0582. This is the area between Z = -1.57 and
the left-hand tail.
Z = 1.5 is equal to an area of .0668. This is the area between Z = 1.5 and the
right-hand tail.
65
RESULTS OF INDIVIDUAL
Z TRANSFORM EXERCISE
Lastly, add the 2 probabilities together and subtract from 1 to determine
the area between Z = -1.57 and Z = 1.5.
.0582 (area between Z = -1.57 and the left-hand tail)
.0668 (area between . Z = 1.5 and the right-hand tail)
.1250 (total area below Z = -1.57 and above Z = 1.5)
.875
.0668
+
6
Z=-1.57
10.1
14
Z=1.5
66