GB_Core_07_Intro_to_Variation_2006-06

Introduction to Variation
Six Sigma Plus Core Green Belt

Honeywell International
DMAIC
DMAICModel
Modeland
andGreen
GreenBelt
BeltTools
Tools
D M A I C
Thought Process Map
SIPOC
Waste Identification and Elimination
Process Mapping
Intro to Variation - Control Charts
Value Stream Mapping
MSE
FMEA
Activity of X
Rapid Problem Solving
Mistake Proofing
Standardized Work
Visual Management
Control Plans
Note: This chart shows where the tools are typically used, but they can be used in any phase as it
2
makes sense.
gb_core_07_intro_to_variation_2006-06.ppt
Learning Objectives: Intro to Variation
z Define and quantify variation

z Learn how variation impacts our processes
z Identify opportunities to improve processes by
controlling and reducing variation
Variation Is the Enemy !

3
Concept of Variation
• No two objects are “identical”, therefore, variation exists

everywhere!
• It has been found that variation in a process can usefully

be categorized as being of one of two types:
- “Expected” or “Common Cause” variation – this variation is
measurable and predictable: “in control”
- “Unexpected” or “Special Cause” variation – this variation is
unpredictable: “out of control”
• To control and reduce variation of either of these types,

you must first understand, quantify, and interpret
variation in a data set
4
BASIC STATISTICS
• “Statistics” is the science of collecting, analyzing,

presenting and interpreting data
• Our goal is to convert data into knowledge and use

that knowledge to make good decisions
“Turning Data into Results ” 5

Statistics to Understand Variation
• How do we understand and quantify variation?

- We use the language and methods of Statistics
• How can Statistics help in understanding, quantifying,
and reducing variation?
- Statistics is used to quantify the amount of variation in our
input variables (X) and our output variables (Y), and to assess
whether the amount of variation is or is not acceptable
- The output variables are related to the input variables by the
process equation “Y = f(X)”. Statistical methods and tools can
be used to understand how variation in the input variables
leads to variation in key outputs
- This understanding leads to ways to reduce output variation
by controlling the important inputs (control plan)
6
Example of Statistics- Productivity Improvement
The collection, analysis and interpretation of data
Scrap
100K
$200K Improvement?
Improvement
Pr
oc
es
sC
$150K ha
ng
e
$100K
Old New
7
Example of Statistics- Productivity Improvement
The collection, analysis and interpretation of data
Scrap
100K
$200K Improvement?
Improvement
Pr
oc
es
sC
$150K ha
ng Range of data
e
$100K
Old New
Understanding our data yields good decisions

8
The Science of Statistics
• Describe data
- Graphical
Histogram, Dot Plot, Pareto Chart
- Analytical
Center (typical value),
Spread (variation)
• Examine differences
• Understand associations
• Describe the whole population by using a small
sample
- Estimate parameters of a population;
example: mean, variation
9
TYPES OF DATA
• Quantitative data
- Sometimes referred to as Variable data
- Types include:
Continuous data: temperatures, sales (pounds), sales
(dollars)
Discrete or Count data: number of defects, employees
hired, number of OSHA recordables per month
• Qualitative data
- Often referred to as Attribute
- Examples:
Good/Bad
Plant location
Machine A, B, C
Band 1, 2, … 6
Highly likely to buy … not likely to buy
• The distinction is important because the tools used for plotting and
analysis are different for different types of data.
- Obtain continuous data wherever you can. The statistical tools are more
powerful, and necessary sample sizes are smaller than for attribute data. 10
The Value of Project Data
“If you can’t measure it, you can’t improve it.”
Data:
- Identifies process baseline
- Quantifies how much improvement is required
- Confirms achievement of project goal
- Enables effective decision making
Data Failure Mode

11
QUIZ: Type of Data
A Green Belt examines the payment timing for a sample of

invoices and records the number of days taken before
payment is received. What type of data is measured?
{ Attribute
{ Quantitative: Continuous
{ Quantitative: Count
12
QUIZ: Type of Data
On a marketing questionnaire, one question asks for the

state in which the customer's business has its
headquarters. What type of data is this?
{ Attribute
13
QUIZ: Type of Data
The internal diameter of the nuts holding on the front

wheel of the bicycles are measured to the nearest
hundred of an inch. What type of data is this?
{ Attribute
14
PGA
When analyzing your data, you should think:
; Practically
; Graphically
; Analytically
15
SOME WAYS TO DESCRIBE DATA GRAPHICALLY
Dot Plot
Histogram
Pareto Charts
16
Dot Plots
• Primarily used with Quantitative data

• Example: 25 brake tests to stop a bicycle at 20 mph
– Data: 30.4 24.9 21.0 25.1 24.3 20.1 23.7 19.5 23.5
24.7 21.1 20.9 25.2 22.7 25.3 26.6 23.4 25.3 22.4
25.5 25.1 23.2 23.7 20.4 22.0
– Construction: plot a horizontal axis for the measurements, and
place a dot for each data point. If dots are too close, stack them
20 25 30
Stopping Distance in Feet
• Shows location and spread of data, identifies outliers

17
Histogram
• Characteristics of both a Bar Chart and a Dot Plot

• Counts the number of data points in specified intervals
• Shows distribution and shape of data, gaps and outliers
• More useful for larger data sets (100+ points) – use a dot plot for
smaller data sets.
5
Frequency
20 25 30
Stopping Distance (feet) 18
Pareto Charts
• Frequency counts by category, ordered from most

frequent to least frequent
• Prioritizes and focuses attention
• Cumulative counts usually also shown
2004 Annual Sales Per Market Segment

60000
100
50000
80
40000
Percent
60
$K
30000
20000 40
10000 20
0 0
Segment re a al ls iv v er
ca ar
m
de
r
hoo /Un Go th
th Ph e Sc s at
e O
al F ge
He ic l le St
bl
Pu Co
Count 19850 12020 8990 7310 3370 1845 940
Percent 36.5 22.1 16.5 13.5 6.2 3.4 1.7
Cum % 36.5 58.7 75.2 88.7 94.9 98.3 100.0
19
Describe Data Analytically
• Measures of Location, or Central Tendency

- Mean and Median
• Measures of Variation, or Dispersion, Spread

- Range, Variance and Standard Deviation
20
MEASURES OF LOCATION (CENTRAL TENDENCY)
• Where do the data “tend to center?”

• The two commonly used measures of central tendency
are:
- Mean
- Median
“Center” is somewhere
around here?
20 25 30
Stopping Distance in Feet

21
Mean
• The mean is the sum the of values, divided by the

number of values in the data set
- Bicycle stopping distances:
30.4 24.9 21.0 25.1 24.3 20.1 23.7 19.5 23.5 24.7
21.1 20.9 25.2 22.7 25.3 26.6 23.4 25.3 22.4 25.5
25.1 23.2 23.7 20.4 22.0
30.4 + 24.9 + 21.0 + ... + 20.4 + 22.0

Sample mean X =
25
= 23.60 feet
X 1 + X 2 + ... + X n
X =
n – The general formula, for “n” data
n points labeled X1, X2, … Xn
∑ Xi
– Capital Greek letter Sigma (Σ) means
= i =1
calculate a sum
n
22
Median
• The median has half the data values below and half
above (50th percentile)
- Midpoint value in the data set arranged from the lowest to
highest value
If the data set contains an odd number of values, the median is the middle value
If the data set contains an even number of values, the median is the average value
of the two middle values
- Stopping distance data sorted:
19.5 20.1 20.4 20.9 21.0 21.1 22.0 22.4 22.7 23.2
23.4 23.5 23.7 23.7 24.3 24.7 24.9 25.1 25.1 25.2
25.3 25.3 25.5 26.6 30.4
The 13th value in sorted order: median = 23.7

23
Mean vs. Median
4 points 4 points
70 75 80 85
median mean
Think of 1 pound balls on a weightless rod:

• The mean is the balancing point (center of gravity)
• The median has half the balls to the left and half the balls to the
right.
24
POPULATION VS. SAMPLE
• Population: The total set of units (for example,

people, parts, measurements, invoices, days, …)
about which we would like information.
• Sample: A subset of the population, usually much
smaller and drawn randomly from the population,
yielding data values that we can analyze “Practically,
Graphically, and Analytically.”
• Inference: The process of looking at characteristics of
the sample data values, for example, mean or
variation, to make conclusions about the likely values
of the corresponding characteristics in the population.
25
Determining Sample Size
The proper sample size depends on:

- Confidence required
- Variation in data
- Your purpose in sampling
Estimate parameters of the population
(The more precise the estimate, the larger the sample)
• What is the average expenditure on network support in a population of service companies?
Examine differences between groups in the population
(The smaller the difference, the larger the sample)
• Is there a difference in preference for FRAM filters between DIY oil filter purchasers in the
southern states and those in the other states?
Understand associations between characteristics of the population
• Relationship between customer’s perception of Honeywell’s technical support and their level of
purchasing from Honeywell
- Black Belts and MBB’s have techniques for calculating the proper sample
size.
26
Examples and Notation
• 50 invoices (sample) inspected for errors to estimate

the frequency of errors in all invoices (population)
generated by World Adventure Bicycles
• 25 stopping distances (sample) using a variety of riders
and ground surfaces to estimate the mean and
variation in stopping distance for the company’s
bicycles in actual use
∑X i
Population mean = μ = i =1
(N is population size)
N
n
∑X i
Sample mean = X = μ̂ = i =1
(n is sample size)
n 27
DESCRIBE DATA ANALYTICALLY: VARIATION
• Measures of Location, or Central Tendency

- Mean and Median
• Measures of Variation, or Dispersion, Spread

– Range, Variance and Standard Deviation
28
Measuring Dispersion
• A second set of 25 brake tests were done to discover the stopping distance
from 20 mph, but this time the tests were done using different riders and
road surfaces, rather than under standard conditions
Distance.2
Distance.1
10 20 30
• The dot plots show that both datasets have the same mean, but the
second set (the one plotted on top) has more variation
• We use several measures of spread or dispersion of measurements:
Range, Variance and Standard Deviation
29
Range
• The distance between the extreme values of

a data set
• Highest minus Lowest
Range: 34.1 – 10.1 = 24.0
Distance.2
Distance.1
10 20 30
Range: 30.4 – 19.5 = 10.9

30
Variance
• The “average” squared deviation of each data value from

the mean
- Subtract the mean from each measurement and square the result
- Add all of these squared deviations and divide by:
the number of observations minus 1
(34.1-23.6)2 = (10.5)2 =110.25

(19.1-23.6)2 = (-4.5)2 =20.25
10 20 30
mean
Distance.2
∑ (X i − X)
n
2
20.25 + 110.25 + ...(23 more)
Variance = = 36.48 ft 2
25 - 1 Var ( X ) = i =1
n −1
31
Standard Deviation
• Standard Deviation, the most common measure of

dispersion, is approximately the average distance of
the data values from the mean
• The Standard Deviation is defined as the (positive)
square root of the variance
• In practical terms the standard deviation can be used
to answer the questions:
1. How large is the spread of the data?
2. How does the spread of this data set compare with the spread
of other data sets?
SD( X ) = Var ( X )
SD(Distance.2) = 36.48 n
(Xi − X )
2
= 6.04
= ∑
i =1 n −1
32
Notation for Variance and Standard Deviation
Population Sample
Variance σ2 s 2 or σˆ 2
Standard Deviation σ s or σˆ
Frequently we are rather casual and use σ to represent the

standard deviation, and hope that the context makes it clear as to
whether we actually have the population standard deviation or an
estimate from a sample.
33
Range vs. Variance vs. Standard Deviation
• When might we prefer to use the range, variance or

standard deviation over one of the others?
- The range is useful for small data sets for hand calculation; its
principal use is with control charts which are traditionally done
using ranges (control charts date from the 1930’s – hand
calculated)
- The disadvantage of the range is that it typically grows larger
when your sample size is larger
- Variance is always the square of the standard deviation, so
these are equivalent in the sense that if you know one, then you
know the other
- Variance is used in many mathematical calculations in
statistics because variances of independent variables can be
added
- Standard deviation is most commonly used to express the final
measure of dispersion, because the physical units for standard
deviation are the same as the data values
34
Graphical View of Dispersion
Stopping distances for

standardized test
Stopping distances for

random riders and
conditions
10 20 30 40
Stopping Distance
The standardized test has less variation

35
THE “NORMAL” DISTRIBUTION
• There are mathematical reasons why the histograms or

dot plots of many data sets form a particular shape
• This shape is the familiar “bell-shaped” curve, more
properly called the “normal” distribution.
HistogramHistogram
of Diameter, with Normal Curve
of Diameter
100
90
80
70
Frequency
60
50
40
30
20
10
0
11.85 11.95 12.05 12.15 11.8 11.9 12.0 12.1 12.2

Diameter Diameter
36
Variation in a Normal Distribution
Total area
μ Mean • Also median
under the curve (average)
= 100 %
-∞ +∞
-6σ -5σ -4σ -3σ -2σ -1σ +1σ +2σ +3σ +4σ +5σ +6σ
68.26%
95.44%
99.72%
In a population which is normally distributed:

–About 68 % of the data values are within 1σ of the mean
–About 95 % of the data values are within 2σ of the mean
–About 99.7 % of the data values are within 3σ of the mean
37
Z-SCORES
• Often it is useful to measure distance from average

(mean) in terms of standard deviations
• This is called the Z-Score
• It is useful for:
- Looking up standard tables of probabilities to see how extreme
(how rare) a data value might be
- Expressing a specification limit in terms of number of standard

deviations from the process mean – this leads to the idea of
Benchmark Sigma Level or Benchmark Z-Score
38
Calculating Z-Score
• The Z-score measures the distance of a value from the

mean, in units of the standard deviation:
X −μ
Z= 2.82
σ
-4 -3 -2 -1 0 1 2 3 4
Z
Stopping Distance
20 25 30
mean
Z-score of largest value: ( 30.4 - 23.6 ) / 2.41 = 2.82
39
Estimating Probability
• The Z-score
- The distance between a single data value and the mean,
expressed in standard deviations
X −μ
Z=
σ
• Probability
- Use the Z-Score to look up the probability of obtaining a data
value that large:
Right tail area look-up in

table on next slide
Example: If Z = 2, then
-4 -3 -2 -1 0 1 2 3 4 tail probability = .0228 40
Z
Z-table: Calculates Normal Tail Probabilities
+ .00 + .01 + .02 + .03 + .04

0.00 .5000 .4960 .4920 .4880 .4840
0.10 .4602 .4562 .4522 .4483 .4443
0.20 .4207 .4168 .4129 .4090 .4052 etc.
The full table is 0.30 .3821 .3783 .3745 .3707 .3669
0.40 .3446 .3409 .3372 .3336 .3300
found in a file 0.50 .3085 .3050 .3015 .2981 .2946
emailed to you: 0.60 .2743 .2709 .2676 .2643 .2611
z-table.xls 0.70 .2420 .2389 .2358 .2327 .2296
0.80 .2119 .2090 .2061 .2033 .2005
0.90 .1841 .1814 .1788 .1762 .1736
1.00 .1587 .1562 .1539 .1515 .1492
1.10 .1357 .1335 .1314 .1292 .1271
1.20 .1151 .1131 .1112 .1093 .1075
1.30 .0968 .0951 .0934 .0918 .0901
1.40 .0808 .0793 .0778 .0764 .0749
1.50 .0668 .0655 .0643 .0630 .0618
-4 -3 -2 -1 0 1 2 3 4 1.60 .0548 .0537 .0526 .0516 .0505
Z
1.70 .0446 .0436 .0427 .0418 .0409
1.80 .0359 .0351 .0344 .0336 .0329
1.90 .0287 .0281 .0274 .0268 .0262
2.00 .0228 .0222 .0217 .0212 .0207
2.10 .0179 .0174 .0170 .0166 .0162
etc. 2.20 .0139 .0136 .0132 .0129 .0125 41
Z-score Example (part 1)
You are a VP for sales; since top line growth is critical,

you establish a monthly award for the sales representative
who has made the greatest impact on his or her region’s
sales performance
- Bill works in the Northeast region; in January, Bill had sales of
$45M, $20M (or 80%) over his region’s average of $25M
- Susan works in the Southern region; Susan had sales of $28M,
just $3M (or 12%) over her region’s average (also $25M)
- Who would get the award for outstanding improvement of the
sales process?
42
Z-score Example (part 2)
You’ve just returned from Green Belt training and
investigate the amount of variation in sales in each
region
- In January, Bill had sales of $45M, $20M (or 80%) over his
region’s average of $25M
The standard deviation is $25M
Z corresponding to $20M is ________
- Susan had sales of $28M, just $3M (or 12%) over her
region’s average (also $25M)
The standard deviation is $1M
Z corresponding to $3M is ________
- Who would get the award for outstanding improvement of
the sales process?
Understanding variation allows us to

make the right decisions and manage risk
43
QUIZ: Calculating Z-score
From analysis of Days of Sales Outstanding (DSO) month by month for a

baseline period, a Green Belt determines that DSO is normally distributed
with a mean of 40 days and a standard deviation of 5 days. A change is
made in the invoicing process to try to reduce DSO. After the change has
taken effect, in the next month the GB found DSO to be 32 days. What is
the Z-score for the DSO of 32 days?
{ 1.0
{ 1.6
{ 2
{ -2
{ -1.6
{ .32
44
QUIZ: Determining Probability
From analysis of Days of Sales Outstanding (DSO) month by month for

a baseline period, a Green Belt determines that DSO is normally
distributed with a mean of 40 days and a standard deviation of 5 days.
A change is made in the invoicing process to try to reduce DSO. After
the change has been implemented, in the next month the GB found DSO
to be 32 days. What is the probability that DSO would be this low (or
lower) if the process change had no effect?
{ 2.5 %
{ 5%
{ 10 %
{ 90 %
{ 95 %
45
Basic Statistics Summary
• Variation exists everywhere

• Data types:
- Quantitative/Variable data: continuous or discrete
- Qualitative/Attribute data: nominal or ordinal
• Looking at Data: PGA
- Practically, Graphically, Analytically
• Graphs can reveal and convey information
• Measures of location and dispersion quantify data
sets
• Probabilities can be calculated to measure how
extreme a data value is
• Understanding variation and probabilities allows
you minimize risk
46
MEASURING QUALITY
• Defect
- Any failure to meet a customer requirement
- Measured at points within the process
- There may be many defects within a single unit
• Defective
- A unit (a part, or a process or service event) that fails to meet
customer’s product or process requirements and contains one or
more defects
- Measured at the end of the process (too often by the customer!)
- A single unit is either defective or not
Defects: The Basic Measure 47

Defects and Defectives
• A completed bicycle is a complex product with many

parts
• It may fail to meet one or more of many customer
requirements:
- Missing something; a plastic handle cover, a reflector, a pedal,
…
- Something may not work; a broken brake cable, …
- Does not perform to specification; the brakes will not stop the
bicycle from 20 mph within a certain number of feet, …
• If a bicycle fails to meet a particular requirement, this
failure is a defect
• If a bicycle has one or more defects, it is defective;
conversely, it is not defective only if it has zero defects
48
Defects and Defectives: Transactional Processes
• The same definitions apply to transactional

processes:
- Shipment to customer: defects might include late shipment,
possibly early shipment, wrong number of units, invoice
incorrect (several opportunities here?), …
- What defects might occur in the invoicing process?
• Traditional difficulties in recognizing defects in
transactional processes:
- Failure to recognize a transactional process as a process
- Lack of metrics
- Failure to keep records
49
Opportunities for a Defect
• A complex completed product has many opportunities

for a defect:
- See bicycle example on the previous slides
• A standard measure of the frequency of defects is
Defects per Million Opportunities, or DPMO
- It is necessary to keep DPMO extremely low for complex parts and
processes in order that there be few defective units
• Counting opportunities:
- An opportunity is:
an action during a process step, or,
a specific feature
- In order to count an opportunity for a defect, there must be the
possibility to observe a defect, either an inspection or a
measurement taken
50
Counting Opportunities
• Internally, you may see many opportunities for

creating defects
• An external customer will see only one opportunity
for a defective part, product, or process.
• Sometimes a customer Sigma Score is calculated
rather than a product or process Sigma Score (recall
Sigma Scores from Introduction to Six Sigma Plus):
- The customer Sigma Score will be lower than the product or

process Sigma Score because there is only one opportunity
for a defect
51
DPMO
• Defects per Million Opportunities (DPMO) allows

defect data from products and processes of varying
complexity to be compared against one another
Number of Defects
DPMO = × 1,000,000
Number of Opportunities
52
Yield
• Yield is the fraction of units that emerge from a

process step without incurring a defect during that
step
Number of satisfactory units

Yield =
Total number of units
• Yield is often expressed as a percent yield instead

of a fraction (and may be called just “yield”):
%Yield = Yield × 100
53
Rolled Throughput Yield (RTY)
• Rolled Throughput Yield (RTY) is the probability that

a product will pass through the entire process
without rework* and without any defects
* “Hidden Factory” effect removed
RTY = Y1 × Y2 × Y3 × ... × Yn ( n = number of process steps)
Example:
RTY(bicycle) = Y (brakes ) × Y (gears) × Y (frame ) × ... × Y (assembly )
Unless the individual yields are very close to 100% at each step,
the rolled throughput yield may be very low for complex
assemblies and processes
54
INCORPORATING CUSTOMER REQUIREMENTS:
SIGMA SCORE
• A Sigma Score is a calculation which compares the
output of a process to customer requirements
- Sigma Score is a measure of Quality
- It compares the Voice of the Process (process variation) to the
Voice of the Customer (customer requirements)
• Sigma Scores enable us to:
- Quantify quality
- Compare the voice of unlike processes
- Focus on the right stuff
• The motivation for the calculation of Sigma Scores
comes from analysis of continuous data from a
process, with a customer specification limit for the
variable
55
Sigma Score for a Single Specification Limit
• Suppose we have measurements on a critical-to-quality

variable:
- with mean μ and standard deviation σ
- and an (upper) specification limit denoted as USL
• The Sigma Score is the number of standard deviations
between the mean and the specification limit
Sigma Score
USL - μ
Z= Zσ
σ
where Standard Deviation
μ is the process mean
σ is the standard deviation σ
USL is the upper spec limit Defects
μ − LSL μ Customer
Z= for a lower spec limit Specification
σ
Mean
Limit
56
Sigma Score --» Fraction Defective
• We can use the Z-table to see how often process

variation will result in measurements above the spec
limit
• See example on the next slide
Z=Sigma Score
Zσ
If we know Z for the USL, i.e.,

the Sigma Score, we can look
up the probability of a defect
Defects from the Z-table.
μ Customer
Mean Specification
Limit 57
Example: Sigma Score for an Upper Spec
• For the standardized braking test employed by World

Adventure Bicycles, we found the following stopping
distances for 25 bicycles (Intro to Variation module):
- 30.4 24.9 21.0 25.1 24.3 20.1 23.7 19.5 23.5 24.7 21.1 20.9
25.2 22.7 25.3 26.6 23.4 25.3 22.4 25.5 25.1 23.2 23.7 20.4
22.0
- Mean is 23.6 and SD is 2.41
• The customer (upper) Specification Limit is 30 feet
- What is the Sigma Score?
USL − μ 30 − 23.6
Sigma Score = Z = = = 2.66
σ 2.41
Probability of failing = .0039, or as a percentage, about 0.4%.
58
QUIZ: Benchmark Sigma for an Upper Spec
World Adventure Bicycles has recently been concerned about the quality of
the grease from its supplier, particularly with respect to certain
contaminants. It has analyzed the most recent 12 batches of grease for
moisture content, and found an average of 330 parts per million (ppm)
moisture, with a standard deviation for the 12 batches of 78 ppm. If the
(upper) specification limit for moisture is 500 ppm, what is the Benchmark
Sigma Level?
{ 1.3 (This is awful stuff!)

{ 2.2 (We've got trouble.)
{ 4.2 (Not too bad, but room for improvement.)
{ 6.4 (Superb!)
59
WHY SIX SIGMA IS THE GOAL
For complex products and systems, 6σ is necessary to

produce and perform defect-free more than 90% of the
time
Impact of Complexity on Rolled Throughput Yield (RTY)
Rolled Throughput Yield (RTY) (%)
100
6σ
80
60
5σ
40
20
4σ
0
0 2000 4000 6000 8000 10000
Number of Parts or Process Steps
60
Sigma Score (Z) vs. DPMO Conversion
Sigma Sigma Sigma

Score DPMO Score DPMO Score DPMO
0.0 933,193 2.0 308,538 4.0 6,210
Often defects are
0.1 919,243 2.1 274,253 4.1 4,661
counted for attribute 0.2 903,200 2.2 241,964 4.2 3,467
data. There is an 0.3 884,930 2.3 211,855 4.3 2,555
industry standard 0.4 864,334 2.4 184,060 4.4 1,866
table to convert 0.5 841,345 2.5 158,655 4.5 1,350
0.6 815,940 2.6 135,666 4.6 968
from DPMO to
0.7 788,145 2.7 115,070 4.7 687
Sigma Score in this 0.8 758,036 2.8 96,800 4.8 483
case (or in cases of 0.9 725,747 2.9 80,757 4.9 337
multiple specs for 1.0 691,462 3.0 66,807 5.0 233
continuous data): 1.1 655,422 3.1 54,799 5.1 159
1.2 617,911 3.2 44,565 5.2 108
1.3 579,260 3.3 35,930 5.3 72.3
1.4 539,828 3.4 28,717 5.4 48.1
1.5 500,000 3.5 22,750 5.5 31.7
1.6 460,172 3.6 17,864 5.6 20.7
1.7 420,740 3.7 13,903 5.7 13.3
1.8 382,089 3.8 10,724 5.8 8.5
1.9 344,578 3.9 8,198 5.9 5.4
2.0 308,538 4.0 6,210 6.0 3.4
61
Determining a Desired Sigma for Your Process
There is an optimum quality level beyond which the

costs of quality improvement exceed the expected cost
savings from a reduced number of defects
Impact of Quality Level on Cost
Cost of Implementation
Cost
Optimum
Cost of Quality
0 6
Sigma
62
Summary: Statistics for Quality
In this module you learned that:
• Sigma Score is an index of our ability to avoid

defects and disappointing the customer
• Defects can exist in any product or process
• Defects can be measured to reach objective

conclusions about how and where to focus to
improve quality
• Sigma Score relates this process knowledge to

customer requirements, and allows us to compare
quality across the businesses
63
Activity Breakout: The M&M Exercise
Purpose:
• In this exercise you will:
- Calculate detailed quality information for a fictitious M&M
production process:
Defects per Unit (DPU)
Yield
Defects per Million Opportunities (DPMO)
- Calculate overall process quality metrics for upper
management:
Total DPMO
Rolled Throughput Yield (RTY)
Sigma Score
- Prioritize which areas you will investigate to improve the
production process.
64
The M&M Exercise (cont.)
The Scenario:
• M&M’s sales are down significantly; focus group data reveals
that customers are dissatisfied with product quality, and are
turning to other candies to satisfy their craving for chocolate.
• We have been tasked to assess the quality of the M&M
production process based on sample product data, and to
determine where to focus our efforts to improve quality.
• Voice of the Customer data suggests five critical failure
modes:
- M&M’s taste bad
- M&M’s melt in hand, not in mouth
- M&M’s are deformed
- M&M’s have cracked shells We will focus on
- M’s are not marked properly these three defects
for our example.
65
The M&M Value Chain
• In this example, data is taken at the end of the value

chain.
• Some level of process knowledge is necessary to
interpret the data!
- SIPOC, Process Mapping and FMEA are useful tools.
Sample
We’re taking the data here! data
Raw
Wholesale Retail
Material Production Customers
Distributors Outlets
Suppliers
This is the process we’re interested in!
66
The M&M Exercise (cont.)
Instructions:
1. Open your bag and count the number of M&M’s inside (don’t eat them
yet!). Enter your number on the scorecard on the following page.
2. Using your best judgement, count the number of defects in your bag
for each of the following failure modes:
a. Deformed – M&M’s which are broken or grossly out of round

b. Cracked – M&M’s with cracked or chipped shells
c. Not properly marked – the “m” is missing, disfigured or off center
Come up with three numbers, one for each failure mode. If an M&M has
more than one defect, count it twice -- once for each defect. Write these
numbers on the scorecard on the following page.
3. The class will fill out the remaining part of the scorecard as a group.
67
The M&M Scorecard
# of Units (my bag):

# of Units (class):
My M&M’s Class M&M’s

# of # of
Failure Modes DPU Yield Opportunities DPMO
Defects Defects /Unit
Deformed
Cracked
Not Properly Marked
Totals
Rolled Throughput Yield (RTY):

The Scorecard provides a
Sigma Score: concise summary of the quality
of a product or process.
68
CAUSES OF VARIATION
Common Cause Special Cause

• Always present • Occasionally present
• Individual impact small • May come from outside the
- Impact may vary, standard process
- But variation is random • Impact of a single cause could
• Collective impact can be large be large or small
69
Example*: “In-Process Inventory”
Large % variance from plan!!!
* From Donald J. Wheeler, Understanding Variation: The Key to Managing Chaos

70
Example: The Data
In-Process Inventory for Department 17

(Hundreds of Pounds)
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Year One 19 27 20 16 18 25 22 24 17 25 15 17
Year Two 20 22 19 16 22 19 25 22 18 20 16 17
Year Three 20 15 27 25 17 19 28
• July is 12% above previous July

• July is 42% above plan
• What action to take?
• Does this signal stand out from the “noise”?
71
Time plot of inventory
• Plot inventory against time

• Is the last point “different”?
Plot of Inventory
35
UCL=32.89
30
Individual Value
25
_
20 X=20.39
15
10
LCL=7.89
Jan-04 Apr-04 Jul-04 Oct-04 Jan-05 Apr-05 Jul-05 Oct-05 Jan-06 Apr-06 Jul-06
Date
72
Add line at the overall average
• Making a control chart: add overall average
Plot of Inventory
35
UCL=32.89
30
Individual Value
25
_
20 X=20.39
15
10
LCL=7.89
Date
73
Final Control Chart
• Add upper and lower “control limits” – these are statistical

limits, not specifications
• Position is at ±3 standard deviations
I Chart of Inventory
35
UCL=32.89
30
Individual Value
25
_
20 X=20.39
15
10
LCL=7.89
Date
74
Conclusions?
• July’s inventory, although the largest ever recorded,

is still well within the range of normal variation for
inventory.
• So we are wasting our time if we focus on only

events that we think might be special for July.
• The proper approach is to analyze the drivers for

inventory in general, and:
- Bring the average inventory levels down
- Reduce variation where possible
• Control charts help us to identify the appropriate

reaction to a high or low data point
75
KEY FEATURES OF CONTROL CHARTS
• Identify potential root causes of

variation:
- Two categories of causes of variation:
“special causes” and “common causes”
- Key features of a control chart
- Use of the control chart to identify
special causes
- How to use control charts in the control
phase of the DMAIC model
- Selecting the appropriate control chart
for your data
• Next steps
76
Control Chart Features
Special Cause Variation

+ 3σ Upper Control Limit (UCL)
NUMERIC SCALE
Avg. Centerline
Lower Control Limit (LCL)

- 3σ
TIME
77
Quiz: Features of a Control Chart
Of the four items below, which is NOT a feature of a

control chart:
{ A plot of data points as a function of
time
{ Center line
{ Customer specifications
{ Control limits
78
Normal Distribution and Control Chart

UCL
± 3σ Common Cause Variation

(Well-Behaved)
LCL
79
One More Feature –The Range Chart
Upper Control Limit (UCL)
4
Individuals
2
Process Average
3
1
Lower Control Limit (LCL)

Moving Range
Upper Control Limit (UCL)
Moving Range Average
2-1 4-3 Lower Control Limit (LCL)

3-2
80
IDENTIFYING SPECIAL CAUSES

variation:
special causes
for your data
• Next steps
81
One Point Outside Limit - Special Cause
• “Out of Control Point”

Special Cause
• What’s Different?
UCL
Inflation Pressure
LCL
TIME
82
Inspection for Special Causes
Shifts
UCL
Trends
LCL
UCL
0 10 20
Cycles
LCL UCL
0 5 10 15
LCL
0 10 8320
Order Entry Application
• Debbie Miller: Order Entry Clerk

- Receives customer orders
- Assigns order number
- Logs order in database
- Sends confirmation letter
• Debbie Miller: Team Leader

- Must reduce cycle time from 5 to 1 day
- Interpreting control charts
84
Quiz: No Special Causes
Which chart shows no

evidence of special causes
of variation?
{ Chart A
{ Chart B
{ Chart C
85
Quiz: What Now?
Given this control chart of an improved process,

what should Debbie’s team do now?
{ A. Declare victory!
Special causes of
variation have been
removed.
{ B. Work with Black Belt

mentor to further
reduce both mean and
variation and meet
team goals.
{ C. Continue searching
for special causes
whenever the process
seems to change
86
USING CONTROL CHARTS IN CONTROL PHASE

variation:
special causes
for your data
• Next steps
87
Using Control Charts in the Control Phase
• Plot data immediately

• Identify and react to out of 35
UCL=34.21
control conditions
Individuals
30 MU=29.85
25 LCL=25.49
• Use Trouble Shooting Observation 0 20 40 60 80 100 120 140
Guides (Reaction Plans) 7

6
Moving Range
5 UCL=5.356
- Data on frequency of known 4

3
root causes 2
1
R=1.639
0 LCL=0.000
- Identify new root causes

• If the process is not out of
control, don’t make
adjustments
88
SELECTING CONTROL CHARTS

variation:
special causes
for your data
• Next steps
89
Types of Control Charts
• Variable control charts:

- X-Bar & R charts
(Averages & Range)
- X-Bar & S charts
(Averages and Standard Deviation)
- I & MR charts
(Individuals & Moving Range)
• Attribute control charts:

- P or NP charts
(Proportion or number of non-conforming units)
- C or U charts
(Count or proportion of defects)
90
Selecting the Right Chart
Sample Size
>1 >1
Type of data 1
Constant Changing
I& X-Bar X-Bar

Continuous
MR &R &S
Defective
P or P
nP
Defect
XXX c u
X 91
Control Chart Summary
• Two categories of causes of

variation: “special causes” and
“common causes”
• Key features of a control chart
• Use of the control chart to identify
special causes
• How to use control charts in the
control phase of the DMAIC model
• Selecting the appropriate control
chart for your data
92

GB_Core_07_Intro_to_Variation_2006-06

Uploaded by

Copyright:

Available Formats

GB_Core_07_Intro_to_Variation_2006-06

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GB_Core_07_Intro_to_Variation_2006-06

Uploaded by

Copyright:

Available Formats

Introduction to Variation

Six Sigma Plus Core Green Belt

z Define and quantify variation

Variation Is the Enemy !

• No two objects are “identical”, therefore, variation exists

• It has been found that variation in a process can usefully

• To control and reduce variation of either of these types,

• “Statistics” is the science of collecting, analyzing,

• Our goal is to convert data into knowledge and use

“Turning Data into Results ” 5

• How do we understand and quantify variation?

The collection, analysis and interpretation of data

The collection, analysis and interpretation of data

Understanding our data yields good decisions

“If you can’t measure it, you can’t improve it.”

Data Failure Mode

A Green Belt examines the payment timing for a sample of

On a marketing questionnaire, one question asks for the

The internal diameter of the nuts holding on the front

When analyzing your data, you should think:

• Primarily used with Quantitative data

Stopping Distance in Feet

• Shows location and spread of data, identifies outliers

• Characteristics of both a Bar Chart and a Dot Plot

• Frequency counts by category, ordered from most

2004 Annual Sales Per Market Segment

• Measures of Location, or Central Tendency

• Measures of Variation, or Dispersion, Spread

• Where do the data “tend to center?”

Stopping Distance in Feet

• The mean is the sum the of values, divided by the

30.4 + 24.9 + 21.0 + ... + 20.4 + 22.0

The 13th value in sorted order: median = 23.7

Think of 1 pound balls on a weightless rod:

• Population: The total set of units (for example,

The proper sample size depends on:

• 50 invoices (sample) inspected for errors to estimate

• Measures of Location, or Central Tendency

• Measures of Variation, or Dispersion, Spread

Range, Variance and Standard Deviation

• The distance between the extreme values of

Range: 34.1 – 10.1 = 24.0

Range: 30.4 – 19.5 = 10.9

• The “average” squared deviation of each data value from

(34.1-23.6)2 = (10.5)2 =110.25

• Standard Deviation, the most common measure of

Frequently we are rather casual and use σ to represent the

• When might we prefer to use the range, variance or

Stopping distances for

Stopping distances for

The standardized test has less variation

• There are mathematical reasons why the histograms or

11.85 11.95 12.05 12.15 11.8 11.9 12.0 12.1 12.2

In a population which is normally distributed:

• Often it is useful to measure distance from average

• This is called the Z-Score

- Expressing a specification limit in terms of number of standard

• The Z-score measures the distance of a value from the

Z-score of largest value: ( 30.4 - 23.6 ) / 2.41 = 2.82

Right tail area look-up in