Statistical Process Analysis

with Minitab 18
Alexander Homberg
BASF OpEx Academy – OP0028/29
Corporate Operational Excellence
June 2019
1. Introduction

How to access the training files?

Did you receive an email on the 3rd of June?

https://basf-
my.sharepoint.com/:f:/r/personal/hombera_basfad_basf_net/Document
s/OpEx%20Academy%20-%20Statistical%20Process%20Control%20-
%20June%202019?csf=1&e=uUYuxM

4
1. Introduction

How to access the training files?

The OneDrive will also contains the presented slides

and all the exercise files, including the description file
on Cpk and sample size calculation.

5
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

6 12/5/2018
1. Introduction

What is Statistical Process Analysis ?

Definition:
▪ Statistical Process Analysis is a method, which employs
statistical methods to visualize and analyze process
data.
▪ Statistical Process Analysis is required if the
determination of parameters influencing the result is
complex. The human brain is still one of the best
devices for finding patterns in data...

Typical Application Fields:

▪ Quality Analysis Challenges:
• high complexity
▪ Batch Cycle Time • large data sets
▪ Yield, Raw Material Losses • quantitative predictions
• high speed and repetitive processing
▪ Admin Processes
▪ ….
7
1. Introduction

Example: m & m production

Quality Control:
▪ Date
data tags
▪ Operator
▪ Shortest diameter
▪ Longest diameter
▪ Colour metrics
▪ M intact and readable
▪ Coating intact
8
1. Introduction

Data Collection Plan

What data do we need? How do we collect data?

Step 1: Identify the metric

◼ Continuous or Attribute
◼ Ensure that you have clear operational definition

Step 2: Define categories (data tags)

◼ Determine which categories may be relevant (e.g. by shift,
operator, day, month, etc.)

Step 3: Create a Data Collection Plan

◼ Define Who measures What Where When and How
◼ Determine a representative sample

Step 4: Design the data collection form

Step 5: Design the measurement system
incl. operational definition
1. Pilot and evaluate your measurement system
2. Collect the data

# An operational definition is a detailed description of how to derive a reliable

measurement result
1. Introduction

Starting Minitab 17

Minitab has four main types of windows:

◼ The session window

Session Window: ◼ One or more data windows
This is the place where Minitab
displays the analysis output
◼ The project manager window

◼ Graph windows

Data Window:
This is a worksheet that contains one
or more columns of data

11
1. Introduction

MINITAB 17
Data Window - Worksheet
You can enter the data into the Data Window by:
◼ Typing it in
◼ Copying and pasting from Excel
◼ Importing it (Excel, ...) “File – Open Worksheet”

Rows Columns Cells Data Type D

(Date/ Time)
Data direction
arrow

Column Name Data Type T

(Text)

Row Number Data Type empty

(Numeric)
12
1. Introduction

Minitab File Types

Dataxyz.mtw
MTW is the file type for worksheets
◼ Worksheets = datasets
◼ Minitab projects must have at least one data window, but they may
also contain more than one
◼ “File” - “Open Worksheet” to open a worksheet
◼ Multiple worksheets must be saved as a project to keep them
together
MyProject.mpj

MPJ is the file type for projects

◼ Projects are containers for all of the Minitab windows that are on the
desktop while you are working; each project can contain a Session
Window, multiple Data Windows, and multiple Graph Windows
◼ Only one project can be opened in Minitab at the same time
◼ “File” - “Open Project” to open; “File” - “Save Project As” to save;
“File” – “New” - project or worksheet
13
13
1. Introduction

Menu Bar in Minitab

◼ In addition to the File menu, there are a number of other Pull Down Menu in Minitab
◼ The Menus that you will used most often are Data, Stat, and Graph

Data Menu

Stat Menu Graph Menu

15
1. Introduction

Short Cut Keys

▪ Ctrl + C / Ctrl + V
Copies data / Pastes data

▪ Ctrl + E
Same function as Ctr + E
Pulls up previous menu

▪ Alt + Tab
Moves you from one Windows application to
another (e.g. Minitab to PowerPoint for making
presentations, Excel to Minitab for copying data)

16
1. Introduction

MINITAB 17/18
Data Window
measured with measured with
gauge caliper gauge

letter “m” is
coating
intact &
is intact
readable

17
1. Introduction

How to use a Caliper Gauge

18
1. Introduction

Step 3: Data Collection Plan

The right sample size

▪ Precision in our sampling results increases as the sample size “N” increases

▪ Mathematically, it increases proportionately to √𝑁

▪ Doubling sample size does not double accuracy
▪ So generally we try to keep sample sizes relatively small

Sample Size

Continuous Data Attribute Data

◼ > 30 data points ◼ ~100 data points

19
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

22 12/5/2018
2. Framework on Statistics

Types of Data: Attribute versus Continuous Data

Not all data is the same !

Continuous variables can have

an infinite number of values,
Continuous data Attribute data but attribute variables can only
be classified into specified
▪ The values / observations belonging to it may ▪ The values / observations belonging to it are categories.
take on any value within a finite or infinite distinct and separate, i.e. they can be counted The advantage of continuous
interval. (1,2,3,....). measurements is that they
▪ You can count, order and measure continuous ▪ You can count but not necessarily order discrete usually give much more
data. data information. The advantage of
attribute data are that they are
▪ Examples: ▪ Examples usually easier to collect. A
– Weight (kg) – the colors of an item disadvantage of attribute data
– Temperature (°C) – blood group (O, A, B, AB) is that they are usually subject
to appraiser interpretation. For
– Time (seconds) – Good / Bad example, one appraiser may
– Pressure (psi) – Machine 1, Machine 2, Machine 3 define a chip defect differently
from other appraisers.
– the amount of sugar in an orange (g) – Shift number
– the time required to run a mile (min) – # of errors in a document
– # units shipped

You need different types of visualizations and statistical methods

2. Framework on Statistics

Descriptive statistics

◼ Measure for location

◼ Measure for spread

◼ Form of the distribution

Central limit theorem

The central limit theorem states that under certain
(fairly common) conditions, the sum of a large
number of a random variable will have an
approximately normal distribution.

24
2. Framework on Statistics

Measures of Location

Mean: Arithmetic average of a set of values

▪ Reflects the influence of all values
▪ Is the “balance point” or “center of
gravity” for the data
▪ Is strongly influenced by extreme values

Median:
Reflects the 50% rank & is the center number after a set of numbers
has been sorted
▪ Splits the data in half.
▪ 50% of the data points will be higher than the median
and 50% will be lower than the median.
▪ Is “robust” to extreme values. A large change in the data
may have very little, if any, effect on the median.

Mode: The most frequent number is a series.

25
2. Framework on Statistics

MEAN versus MEDIAN

You are asked to summarize the average starting salaries of graduates

€ 10, 20, 30, 40, 50 (€ in thousands)

What is the average income (mean)? What is the median income?

You consider including the starting salary of a top performer.

€ 10, 20, 30, 40, 5000 (€ in thousands)

What is the average income (mean)? What is the median income?

26
2. Framework on Statistics

Measures of Spread

▪ Range (R): Difference between highest and lowest number

▪ Variance (Var (x), s2, s2): The average, squared deviation of each data point from the mean

▪ Standard Deviation (s, s): The square root of the variance

Data: 2, 1, 3, 5, 4 5

4
X - X = -1 X- X =0
3 X =3
2
X - X = -2
1

27
2. Framework on Statistics

Sample versus Population Parameters

A data sample is a set of data selected (collected) from a

statistical population.
Sample statistics are used to estimate population parameters,
i.e. mean, median, range, standard deviation, variance

x = Sample Average  = Population Mean

s = Sample Standard s = Population Standard

Deviation Deviation

28
2. Framework on Statistics

Sample versus Population Parameters

sample population

sample average (mean) population mean

sample standard deviation population standard deviation

29
2. Framework on Statistics

Calculating the Standard Deviation

Problem: X X-X (X - X)2

Using the form on the 1
right, calculate the 2
standard deviation for
3
the numbers:
4
21354 5
6 N
7  (X − X )
i=1
i
2

8
n-1
9
10
S N

Mean  (X − X )
i=1
i
2

s2 n-1
s

30
2. Framework on Statistics

Calculating the Standard Deviation

X X-X (X - X)2
1 2 -1 1
2 1 -2 4
3 3 0 0
4 5 2 4
5 4 1 1
6 N
7  (X − X )
i=1
i
2

8
n-1
Why dividing by n-1? 9
The variance and standard 10
deviation has N − 1
S 15 10 N

 (X − X )
degrees of freedom, since 2
it is computed from N
random scores minus the
Mean 3 i =1
i

only 1 parameter estimated s2 2,5 n -1

as intermediate step, which
is the sample mean (=3). s 1,581139

31
2. Framework on Statistics

Properties of a Normal Distribution

A normal distribution can be described completely by knowing the:

▪ Mean
and
▪ Standard Deviation

Distribution 1

Distribution 2

Distribution 3

What is the difference among these three normal distributions?

32
2. Framework on Statistics

Standard Deviation increases with spread

33
2. Framework on Statistics

Properties of a Normal Distribution

Most natural phenomena and man-made processes are distributed normally, or can be represented as normally distributed.
The area under sections of the curve can be used to estimate the cumulative probability of a certain “event” occurring.

Cumulative probability
of obtaining a value
68,27% between two values
Probability of sample value

40%

30%
95,45%

20%

10% 99.73%

0%
-4 -3 -2 -1 0 1 2 3 4
Number of standard deviations from the mean
34
2. Framework on Statistics

Different distributions
Skewed distribution

▪ For centered distributions the mean and the median are close, whereas for skewed distributions they can differ significantly.
▪ In general: The Median is not so sensitive to extreme outliers

35
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

36 12/5/2018
3. Analyzing Data

Plotting data depends on the data type

Continuous data and attribute data

Continuous data
▪ Histogram
▪ Boxplot
▪ Scatter Plot
▪ Bubble Plot
Bubble Plot of Continouos data 2 vs Continouos data 1

▪ Time Series 7
Bubble siz e: Continouos data 3

5
Continouos data 2

0
0 1 2 3 4 5 6 7
Continouos data 1

Attribute data
▪ Pie Chart
▪ Pareto Chart
▪ Bar Chart

37
3. Analyzing Data

Plotting data depends on the data type

Continuous data and attribute data

38
3. Analyzing Data

How to create Histograms in Minitab? Language change

2
3

39
3. Analyzing Data

OWN EXAMPLE: Histogram

40
3. Analyzing Data

How to create a Graphical Summary in Minitab?

Graphical summary gives us visual insight on the Center, Spread and Shape of the data.

1
2 3

41
3. Analyzing Data

OWN EXAMPLE: Graphical Summary

42
3. Analyzing Data

Graphical Analysis with Minitab

◼ The Minitab Graph menu offers a
range of common graphs
◼ In order to remain in control of
the many graphical options of
For correlations Minitab:

◼ Rename your graphs, so

that the title reflects what the
graph is showing
For plotting distributions
◼ Save graphs you want to
keep for a long time as
separate files via File →
For comparing groups of data
Save Graph As… (file
extension .mgf)
For analyzing categories of data
◼ Close any unwanted graphs
For studying trends over time and minimize the ones to
keep
For 3D graphs

43
3. Analyzing Data

Plotting data depends on the data type

Pie Charts
For attribute data

When you are investigating the distribution of unique values in a first overview, the classical
pie chart will be a useful tool.

Practical hint:
In case you want to alter the color scheme of the pie slices,
left mouse click and then right mouse click on one of the slices
and select “Edit Pie”. In tab “Attributes”, select the option
“Custom” and choose a different background color.

44
3. Analyzing Data

Plotting data depends on the data type

Pareto Charts

For attribute data

A special type of bar chart where the plotted values are arranged from largest to smallest. A
Pareto chart is one of the basic quality control tools used to highlight the most frequently
occurring defects, the most common causes of defects, or the most frequent causes of
customer complaints.

80% of the
effect is linked
to 20% of the
causes

45
3. Analyzing Data

How to create Pareto Diagrams?

1
2

46
3. Analyzing Data

Plotting data depends on the data type

Scatterplots
For continuous data
When you are investigating a relationship between two factors, a Scatter plot is the best first
step for analyzing the data graphically.

Next steps after a Scatterplot:

In cases, where a relationship is not clear cut, correlation can be used to help decide if a
relationship exists. Regression techniques go a step further and define a relationship
(correlation) in a mathematical format.

Be careful to deduct to a cause and effect from a relationship of data: there are situations,
where a correlation exists, but there no cause and effect, e.g. sales of ice-cream and incidents
47of sunburn.
3. Analyzing Data

Plotting data depends on the data type

Boxplots - Overview

Median (Q2)
Q1 Q3

Box plots can be drawn

horizontally or vertically.
Sometimes the height of
the box represents the
sample size, but usually
it has no significance.
In some visualizations,
the dot on the box
represents the average.

2 3 4 5

Helpful hints:
Do not use box plots if you are looking at just one set of data, histograms are much better. Box
48
plots should be used when comparing several sets of data against each other.
3. Analyzing Data

Plotting data depends on the data type

Boxplots - Details

For continuous data

Analyzing the differences between subgroups of data can provide clues and insights into how
the process behaves. The process of dividing a dataset into subgroups is called stratification.

Why divide up the data?

The differences between the subgroups may be a substantial cause of variation in the
process. This can be used for root-cause-analysis, e.g. the best performing subgroup can be
analyzed for benchmarking purposes, the worst performing subgroups can be investigated
using 5 Whys, Fishbone diagrams, etc.
.

50
3. Analyzing Data

Plotting data depends on the data type

Bubble plots

For continuous data

While Scatter Plots only allow two factors to be mapped against each, Bubble Plots allow a
third variable to be added to the picture – enabling more complex relationships.

Next steps after a Bubble plot:

In cases, where a relationship is not clear cut, correlation
can be used to help decide if a relationship exists.
Regression techniques go a step further and define a
relationship (correlation) in a mathematical format.

The size of the bubble is significant, it is important to

ensure they are sized in the most effective way. Double-
clicking on any of the bubbles below opens the “Edit
Bubbles” function.
51
3. Analyzing Data

Plotting data depends on the data type

Histograms / Frequency Plots

For attributive data

Histograms are one of the most common Six Sigma tools. They show the shape of the
distribution and form an essential part of the “1st pass analysis” of any data.

Useful hints:
If you are creating several different histograms at the same time, you can choose to display
multiple graphs. The histograms can be either:
- Overlaid on the same graph: useful for comparing a 2-3 histograms
- In separate panels of the same graph: useful for comparing lots of histograms at the
same time
- On separate graphs: a rapid way of producing multiple histograms that you do not want
to compare.

52
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

53 12/5/2018
4. Process Stability

Time series plots

Introduction
Time series plots, combined with Histograms, are the two most important
graphical tools.

These plots help to spot time-based changes, that histograms cannot see,
such as: upwards or downwards trends, changes in the amount of variation,
differences between the short term and the long term, repeating patterns or
cycles, anything that does not appear random.

54
4. Process Stability

Time series plots

Continued
Things to watch out for with Time series:
1) Make sure your data is time-ordered:
Time plots require data that is in the order that actually happened, so that
time trends can be detected.
2) Mixed processes: the plot seems to have two different levels, which might
indicate, that two different processes have been mixed together.

3) Mixed trends and patterns:

55
4. Process Stability

Time series plots

Continued
Things to watch out for with Time series:
3) Mixed trends and patterns: Several different types of changes might
happen over time. In this case, there is a repeated pattern and an overall
upward trend. Trend analysis can be used to help identify overall trends
and Minitab’s decomposition can be used to separate seasonality, trends
and random variation.

56
4. Process Stability

Trend Analysis
Introduction
Real life processes often contain a combination of trends and
cycles, as well as seasonal and random variation. If a time series
plot suggests, that an overall trend exists, the Trend Analysis function
can help to identify, model and even forecast the trend.

Problem Statement: A LEAN project in a chemical company in

Ludwigshafen has been working on reducing production rework rates
(recorded as “absence rates”) due to missing equipment calibrations by
implementing a continuous improvement process. The %tage of
production lost due to rework has been recorded monthly over the last 5
years.

57
4. Process Stability

Trend Analysis
Introduction

If you want a forecast of how the fitted trend will behave in the future, select
“generate forecasts” and enter the number of required points.

58
4. Process Stability

Trend Analysis
Continued
In this case, the trend analysis fits a downward sloping curve, indicating that
the rework/absence rates have improved over the last five years and are now
levelling out.
The analysis also provides the exact mathematical equation of the fitted model, The accuracy measures
which in this case is a quadratic equation. (MAPE, MAD and MSD) all
aim to do the same thing:
provide a measure of how
accurate the fitted model
is. Lower values indicate a
better fit, and while the
values do not mean much
themselves, one can use
them to compare and
decide, which model fits
best.

59
4. Process Stability

Statistical Process Control (SPC) Methods

Statistical Process Control charts are essentially a sophisticated form of Time Series plot that enable the stability of the process,
and the type of variation involved , to be understood.

What do SPC charts detect?....Changes!

• Changes in process average
• Changes in process variation
• One-off changes such as special causes

time time

time

60
4. Process Stability

Basic Information about Control Charts

▪ Control Charts Provide information on the predictability of the process, specifically with regard to its:
▪ Accuracy (to a historical or process average value)
▪ Variation (both random and non-random)
▪ Control Charts are used in different DMAIC stages:
▪ In Measure stage to understand current process performance to determine the process stability
▪ In Improve/ Control stage for sustaining and monitoring process performance after process improvements have been implemented
▪ Control Charts only detect the fact that a process parameter is out-of-control, not why it is out of control

MONITORED CHARACTERISTIC
UCL Upper Control Limit
(Center Line +3s)

Center Line - Average X

LCL – Lower Control Limit

Data plotted over time (Center Line -3s) UCL = Upper Control Limit
LCL = Lower Control Limit
61
4. Process Stability

Common Cause versus Special Cause Variation

▪ Common Cause (controlled or noise)

▪ Is present in every process
▪ Is predictable
▪ Is produced by the process itself (the way we do business)
▪ Can be removed and/ or reduced but requires a fundamental change in the process
▪ A process is stable, predictable and In-Control when only Common Cause Variation
exist in the process
▪ Special Cause
▪ Is unpredictable
▪ Is caused by one-time disturbances or series of them
▪ May be a one-time event or a process change
▪ A process with special cause variation is said to be out-of-control and unstable

Plotting the data on a control chart is the best way to identify special causes

62
4. Process Stability

Specifications and Goals

▪ Most processes have specifications or/ and goals

▪ In the ideal world, they’re determined by customers, although sometimes they’re set by management, government
regulations, or others

▪ Examples:
▪ Call handling time (response time) between 1 and 2 minutes
▪ On-time delivery should be above 97%
▪ The cycle time for obtaining new customer contract should be less than six months

63
4. Process Stability

Control Limits versus Specifications and Goals

▪ Control Limits are natural process limits calculated based on the mean and standard deviation of observations, measuring
whether a process is in or out of control
▪ Upper Control Limits (UCL) and Lower Control Limits (LCL) is a statistical calculation without considering customer
requirements
▪ UCL and LCL are one representation of the Voice of Business

▪ Specification Limits (Specs) are customer defined allowable tolerances or requirements that have no connection to process
data

The most common mistake in the application of control charts is to confuse

specification limits with control limits

Control Limits and Specs are completely unrelated!!

Control limits are about VoB / Spec Limits are about VoC
64
4. Process Stability

Specification Limits ≠ Control Limits

Upper Control Limits = UCL Upper Specification Limits = USL

Lower Control Limits = LCL Lower Specification Limits = LSL

Is the process below, out of or under control?

Is it generating defects? ?
UCL

USL

LSL

LCL

65
4. Process Stability

Specification Limits ≠ Control Limits

Upper Control Limits = UCL Upper Specification Limits = USL

Lower Control Limits = LCL Lower Specification Limits = LSL

Is the process below, generating defects?

?
USL
UCL

LCL
LSL
66
4. Process Stability

Statistical basics

General model for Center Line, Upper and Lower Control Limits for variables chart

UCL = X + ks

+ ks

Center Line = X

- ks

LCL = X - ks

𝑋ത = the sample Mean

s = estimate of sample standard deviation
k = some distance of the control limits from the Center Line (usually 3). The value of k required to achieve 3
standard deviations is determined from the average range or standard deviation of the data

67
4. Process Stability

Elements of Control Charts

UCL versus LCL

What % of data points should fall between the UCL and LCL?
If a point falls beyond the UCL or LCL does this mean we are making a defect for the customer?

UCL USL

LCL LSL

68
4. Process Stability

Empirical rules for standard deviation

“Where do the data points lie?” ?

% of data points
99.7%
3 Sigma
95%
2 Sigma
The Item
we are 1 Sigma
68%
measuring 1 Sigma
2 Sigma
3 Sigma

69
4. Process Stability

Differentiate between two types of data

Continuous (Variable) :
▪ The data are continuous (measured)
▪ Results from the actual measuring of a characteristic such as diameter of a hose, electrical resistance, weight of a unit, etc.
▪ Types of charts:
▪ Individual-X and Moving Range Chart
▪ X-Bar / R chart and X-Bar / S chart
→ Tells how good it is
Attribute:
▪ The data are generally counted
▪ Results from using go/no-go gages, inspection of defects, quantity of missing parts, pass/fail or yes/no decisions, etc.
▪ Types of charts:
▪ p-Chart
▪ u-Chart
→ Tells whether it is good or not

70
4. Process Stability

Exercise: What is the data type?

1. Length of connector housings produced by a molding press V

2. Production rate of the manufacturing lines in a building V

3. Number of printing defects on a package label A

4. Number of typographical errors per sales contract A

5. Number units failing test in monthly production A

6. Percent failed units in monthly production A

7. Amount of time taken to close an Account Receivable V

8. Number of parts with defects per 1000 parts produced

A
71
4. Process Stability

Chart Selection – Choosing the proper Control Chart

& assuming normal

distribution

72
4. Process Stability

How to create Control Charts in Minitab

Continuous (Variable) Charts

(The I-MR chart is
also used for Attributes)

Attributes Charts

73
4. Process Stability

Rules that indicate out of control conditions

A set of standard rules has been created to help identify special cause events in our process

The rules we will use:

◼ Rule #1: One point beyond the UCL or LCL (3-sigma limit)
◼ Rule #2: Two of three consecutive outside the 2-sigma limit
◼ Rule #3: Four of five consecutive outside the one-sigma limit
◼ Rule #4: Eight / Nine consecutive on one side of the center line
◼ Trend: Six consecutive either increasing or decreasing
◼ Pattern: A cycle appears or a pattern repeats itself

Instruction on the application of the rules:

1. Start with rule #1 and Trend / Pattern detection
2. If high sensitivity required, go with rules #2, 3 & 4

The point with these rules is that if any of these conditions are identified in
your control chart, it is very likely that something special has occurred
74
4. Process Stability

Rules in Minitab

Rule #1
Rule #4
Trend

Rule #2
Rule #3

No specific “rule” for patterns

75
4. Process Stability

Control Charts Rules

Charts Description 1. Example 2. Example Interpretation

Process Chart points do not The process is stable, not
In Control form a particular UCL 20 UCL 20
changing. Doesn’t
pattern AND lie necessarily mean to leave the
15 15
within the upper and X X process alone. May be
lower chart limits. opportunities to improve the
LCL LCL 10
10
process and enjoy substantial
benefits.

Process Chart points form a Alerts you that the process is

Out of particular pattern OR UCL 20 UCL 20 changing. Doesn’t mean you
Control one or more points need to take corrective action.
lie beyond the upper X
15
X
15
May be related to a change
or lower chart limits. you have made. Be sure to
LCL LCL
10 10
identify the reason(s) before
taking any constructive action.

Run Chart points are on UCL 20 UCL 20

Suggests the process has
one side of the undergone a permanent
center line. The # of change (+ or -) and is now
15 15
points in a run is X X becoming more stable. Often
called the length of requires that you recompute
the run. LCL 10 LCL 10 the control lines for future
interpretation efforts.

76
4. Process Stability

Control Charts Rules (Cont.)

Charts Description 1. Example 2. Example Interpretation

Trend A continued rise or Often seen after some
fall in a series of UCL 20
7
UCL 20
1 change has been made.
6 2
points (7 or more 4
5
3
4
5
Helps tell you if the change(s)
2 3
consecutive points in X 15
1 X
15 6
7 had a + or - effect. May also
the same direction). be part of a learning curve
LCL 10 LCL 10
associated with some form of
training.

Cycle Chart points show Often relates to factors that

the same pattern UCL 20 UCL 20 influence the process in a
changes (e.g., rise or predictable manner. Factors
15
fall) over equal X
15
X occur over a set time period
periods of time. and have +/- effect. Helps
LCL 10 LCL 10
determine future work
load/staffing levels.

Hugging Chart points are Suggests a different type of

close to the center UCL 20 UCL 20
1/3
data has been mixed into the
line or to a control 1/2
subgroup being sampled.
15 15
limit line (2 out of 3, X 1/2 X Often need to change the
3 out of 7, or 4 out of 1/3
subgroup, reassemble the
LCL 10 LCL 10
10). data, and redraw the control
chart.

77
4. Process Stability

Basic Principles of Individual / Moving Range (I-MR) Charts

What is an I-MR Chart and when do we use it?

▪ Used when subgrouping can’t be done on a rational basis

▪ Sub-group size of 1, that is each data point is an individual measurement and not means of samples/sub-group

▪ Has two charts – Individual (I) and Moving Range (MR)

▪ Individuals (I) chart presents the measurements in time order

▪ Plots each individual data point
▪ Assesses stability of process average (means)

▪ Moving Range (MR) chart shows the short-term variability in the process
▪ Moving Range is the differences between successive observations
▪ Assesses stability of process variation

78
4. Process Stability

Calculating the Moving Range

I and MR Chart for C7

UCL=102.3
Individual Value

2.
100 3.
Mean=97.22
95 1.

LCL=92.12
1
90
Subgroup
Observation 0 10 20

Moving9Ranges calculation
1 8
Moving Range

1. 7 – 91.3| = 2.5
|93.8 % LOT Moving Range
6 93.8 * UCL=6.266
2. 5 – 99.2| = 7.9
|91.3 91.3 2.5
4 99.2 7.9
3
98.8 0.4
3. 2 – 98.8| = 0.4
|99.2 R=1.918
79
1
0 LCL=0
4. Process Stability

How to create an I-MR Chart in Minitab?

80
4. Process Stability

I-MR Chart – Example

81
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

82 12/5/2018
5. Measurement System Analysis

Measurement System Analysis (MSA)

Observed total variation can come from process and measurement

Total Variation

Process Variation Measurement Variation

◼ Due to the way the process ◼ Due to the way the process
works is measured
◼ Affects how our products and ◼ Affects our perception of
services appear to our process variation
customers

➢ the variation due to the measurement system must be

identified and separated from the process variation
➢ Look at “repeatability” and “reproducibility” as primary
contributors to measurement errors

83
5. Measurement System Analysis

The influence of measurement systems on process measurements

Measurement Systems are

the lens that we look through
to see our process performance

Step #1 Step #2 Step #3 Step #4

◼ We think what we see is process variation

◼ Actually, we see variation due to the process and the measurement system

84
5. Measurement System Analysis

MSA Terminology
Sources of Measurement System Variation
▪ Discrimination: Smallest detectable increment between two
measured values
▪ Accuracy (Genauigkeit, Lage) related terms
▪ True value - Theoretically correct value
▪ Bias (due to operator, due to instrument) - Difference between the
average value of all measurements of a sample and the true value
for that sample (Genauigkeit gemäß Standard / Kalibrierung)
▪ Precision (Präzision, Streuung)
related terms
▪ Repeatability - Variability inherent in the measurement system
under constant conditions
▪ Reproducibility - Variability among measurements made under
different conditions (e.g. different operators, measuring devices,
etc.)
▪ Stability - distribution of measurements that remains constant and
predictable over time for both the mean and standard deviation
(Stabilität über die Zeit)
▪ Linearity - A measure of any change in accuracy or precision over
the range of instrument capability (Linearität über den Messbereich)
85
5. Measurement System Analysis

How does measurement error appear?

15
LSL USL

Actual process variation –

Frequency
10

No measurement error
5

30 40 50 60 70 80 90 100 110
Process

15
LSL USL
10

Observed process variation –

Frequency
With measurement error
5

30 40 50 60 70 80 90 100 110
Observ ed

86
5. Measurement System Analysis

Accuracy vs. Precision

Short Exercise

Suppose we have a reference material with a ‘true’ value of 5.0 kg

▪ Method 1 gives the following results: 3.8 kg, 4.4 kg, 4.2 kg, 4.0 kg
▪ Method 2 gives the following results: 6.5 kg, 4.0 kg, 3.2 kg, 6.3 kg

Which method is more accurate?

Which method is more precise? ?

87
5. Measurement System Analysis

Attribute Measurement System Analysis (MSA)

88
5. Measurement System Analysis

Attribute Measurement System Analysis (MSA)

Piece Piece round Operator Operator Operator

18 A OP1 OP2 OP3
19 B 1 A C B
20 C 2 B A C
3 C B A
4 C A B
5 B C A
6 A B C
7 A C B
8 B A C
9 C B A

results on Flip Chart

round Operator Operator Operator
OP1 OP2 OP3
1
2
3
4
5
6
7
8
9
5. Measurement System Analysis

Attribute Agreement Analysis in Minitab

Stat, Quality Tools, Attribute Agreement Analysis

90
5. Measurement System Analysis

Attribute Agreement Analysis in Minitab

91
5. Measurement System Analysis

Attribute Agreement Analysis in XLS

92
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

93 12/5/2018
6. Process Capability

Capability Analysis
Capability analysis compares the performance of the process to the customer requirements or to
the project goals.

It describes the ability of a process to produce output within specification limits.#

Measurable Specification /
Performance of the
Customer Requirements:
process regarding:
▪ Longest diameter: 10 to 26 mm
▪ Shortest diameter
▪ M intact and readable
▪ Longest diameter
▪ Coating intact
▪ Color
▪ M intact and readable ▪ Weight 3 to 3,6 g (2,6 to 4 g)
▪ Coating intact
▪ Weight

#: requirement for capability analysis: process under control (no special causes), acceptable normal distribution
94
6. Process Capability

Depending on the data type there are two ways to calculate

process capability
1. Attribute Data
Results from inspection of visual defects, yes/no decisions (“m” and coating)
2. Continuous Data
Results from the actual measuring of a characteristic, such as time to complete a task, processing time (diameter, weight)

Define the Define the

Collect Process Data
Goals/Specifications Unit and the Defect

Determine Data Type

Attribute Data Continuous Data

Analyze Data Plot and Analyze Data
(Minitab or Excel) (Minitab)

State Capability State Capability

(DPU, DPMO, Sigma Level) (Cp, Cpk, Pp, Ppk)

95
6. Process Capability

Introduction to
Attribute Capability Analysis
Capability expressed as:
▪ Defects per Unit (DPU)
▪ Defects per Opportunity (DPO)
▪ Defects per Million Opportunities (DPMO)
▪ Sigma Level (s)
Unit (U):
▪ The output from your process
▪ Something that is inspected, evaluated, or judged by others to determine “suitability for use”
▪ Something delivered to customers or users
Defect (D):
▪ Anything that does not meet a critical customer requirement
▪ A defective unit is any unit containing a defect, there can be multiple defects in one defective unit

96
6. Process Capability

Defects per Million Opportunities

Expression Definition
Unit N: # of units
Defect D: # of defects
Defect opportunity O: # of defect opportunities
defective unit with one or more defects

DPU DPU = D / N

DPO DPO = D / (N x O)

DPMO DPMO = D / (N x O) x 1000000

Look up the Sigma Quality Level using

Process Sigma
Sigma Table
6. Process Capability

Graph, Bar Chart

Defects

COATING m
98
6. Process Capability

Defects per Million Opportunities and Sigma (s) Conversion

1. Determine number of defect opportunities per unit (O) 2 m, coating

2. Determine number of units processed (N) 198

3. Determine total number of defects made (D) 75

4. Calculate Defects per Opportunity (DPO) 75/(198*2) = 0.189

5. Calculate Yield (1-DPO) x 100 82%

6. DPMO 0,18x106 = 189000

7. Look up the Sigma Quality Level

2,48
using Sigma Table

99
6. Process Capability

Appendix – Sigma conversion Table

Sigma Level * -- Hundreths

0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
1.5 500,000 496,000 492,000 488,000 484,000 480,100 476,100 472,100 468,100 464,100
1.6 460,200 456,200 452,200 448,300 444,300 440,400 436,400 432,500 428,600 424,700 DPMO
1.7 420,700 416,800 412,900 409,000 405,200 401,300 397,400 393,600 389,700 385,900
1.8 382,100 378,300 374,500 370,700 366,900 363,200 359,400 355,700 352,000 348,300 180000
1.9 344,600 340,900 337,200 333,600 330,000 326,400 322,800 319,200 315,600 312,100
2.0 308,500 305,000 301,500 298,100 294,600 291,200 287,700 284,300 281,000 277,600
2.1 274,300 270,900 267,600 264,300 261,100 257,800 254,600 251,400 248,300 245,100
SIGMA 2.2 242,000 238,900 235,800 232,700 229,700 226,600 223,600 220,700 217,700 214,800
2.3 211,900 209,000 206,100 203,300 200,500 197,700 194,900 192,200 189,400 186,700
2.4 184,100 181,400 178,800 176,200 173,600 171,100 168,500 166,000 163,500 161,100
2.5 158,700 156,200 153,900 151,500 149,200 146,900 144,600 142,300 140,100 137,900
2.6 135,700 133,500 131,400 129,200 127,100 125,100 123,000 121,000 119,000 117,000
2.7 115,100 113,100 111,200 109,300 107,500 105,600 103,800 102,000 100,300 98,530
2.8 96,800 95,100 93,420 91,760 90,120 88,510 86,910 85,340 83,790 82,260
2.9 80,760 79,270 77,800 76,360 74,930 73,530 72,140 70,780 69,440 68,110
3.0 66,810 65,520 64,260 63,010 61,780 60,570 59,380 58,210 57,050 55,920
3.1 54,800 53,700 52,620 51,550 50,500 49,470 48,460 47,460 46,480 45,510
3.2 44,570 43,630 42,720 41,820 40,930 40,060 39,200 38,360 37,540 36,730
Sigma Level -- Tenths

3.3 35,930 35,150 34,380 33,630 32,880 32,160 31,440 30,740 30,050 29,380
3.4 28,720 28,070 27,430 26,800 26,190 25,590 25,000 24,420 23,850 23,300
3.5 22,750 22,220 21,690 21,180 20,680 20,180 19,700 19,230 18,760 18,310
3.6 17,860 17,430 17,000 16,590 16,180 15,780 15,390 15,000 14,630 14,260
3.7 13,900 13,550 13,210 12,870 12,550 12,220 11,910 11,600 11,300 11,010
3.8 10,720 10,440 10,170 9,903 9,642 9,387 9,137 8,894 8,656 8,424
3.9 8,198 7,976 7,760 7,549 7,344 7,143 6,947 6,756 6,569 6,387
100
4.0 6,210 6,036 5,868 5,703 5,543 5,386 5,234 5,085 4,940 4,799
4.1 4,661 4,527 4,396 4,269 4,145 4,024 3,907 3,792 3,681 3,572
6. Process Capability

Example

My son wants to eat 10 m&m‘s per day (each day of a year)

Last year we ran out of stock on 5 days.

Unit: 1 m&m
Defect: no m&m in stock
Opportunity: 1 (no m&m)
Number of units processed (N): 10 x 365 = 3650
Number of defects: 5 x 10 = 50
DPMO: 50/3650 x 1000000 = 13699
Sigma level: 3,7

102
6. Process Capability

Depending on the data type there are two ways to calculate

process capability
1. Attribute Data
Results from inspection of visual defects, yes/no decisions (“m” and coating)
2. Continuous Data
Results from the actual measuring of a characteristic, such as time to complete a task, processing time (diameter, weight)

Define the Define the

Collect Process Data
Goals/Specifications Unit and the Defect

Determine Data Type

Attribute Data Continuous Data

Analyze Data Plot and Analyze Data
(Minitab or Excel) (Minitab)

State Capability State Capability

(DPU, DPMO, Sigma Level) (Cp, Cpk, Pp, Ppk)

103
6. Process Capability

Capability Analysis

Capability analysis compares the performance of the process to the

customer requirements or to the project goals

Measurable Specification / There are three kinds of requirements

that we apply to data
Customer Requirements:
▪ Longest diameter: 10 to 26 mm Target: In general, we would like our
typical value to be equal the target
▪ Weight: 3 to 3,6 g
▪ M intact and readable Upper Specification Limit (USL): We
would like all of the diameters to be less
▪ Coating intact
than the upper specification limit
Lower Specification Limit (LSL): We
would like all of the diameters to be greater
than the lower specification limit

104
6. Process Capability

Basics of Continuous Capability Analysis

▪ Cp: Compares the voice of the customer (specification limits) to the
voice of the process (std deviation)
LSL USL
▪ Capability is expressed in terms of: Cp, Cpk and Pp, Ppk

Cp = 1:
The process is barely capable
(Just fits into the tolerance window)
Cp = 2
Voice of the The process is a six sigma process (The tolerance window
Customer is twice the process capability)

Voice of The Process Fixing values for minimum "acceptable" process capability
targets is a matter of personal opinion, it varies by industry
and the process under consideration, typically 1.00 up to
1.66
Voice of the Customer USL – LSL
= = Cp
Voice of the Process 6s

105
6. Process Capability

Illustration of Cp

LSL USL
Cp = 0.5

Cp = 1.0

Cp = 1.5

Cp = 2.0

106
6. Process Capability

Barely Capable Process: No Mean Shift

Lower Upper
Specification Specification
Limit
Cp = 1.00 Limit

Process +/-3s

Specif. +/-3s
107
6. Process Capability

Very Good Process: No Mean Shift

Lower Upper
Specification Specification
Limit
Cp = 2.00 Limit

Process +/- 3s

Specif. +/- 6s
108
6. Process Capability

Barely Capable Process: With Mean Shift

Lower Upper
Specification Specification
Limit Limit

109
6. Process Capability

Cp and Cpk for an Off-Center Process

LSL USL

Cp= 1.3
Cpk = 1.3

Cp= 1.3
Cpk = 0.8

Cp= 1.3
Cpk = 0.0

110
6. Process Capability

Process Capability Indices

Cp: ✓
Cpk: 

111
6. Process Capability

Process Capability Ratio – Cpk

Cpl Cpu

X - LSL USL- X
Cpk = Min( , )
3s 3s

X - LSL USL- X
Cpl = Cpu =
3s 3s

Cpk is minimum value of the two:

Cpu and Cpl, accounts for process centering and spread

112
6. Process Capability

Continuous Capability Analysis in Minitab

▪ Start with USL = 2.6mm and USL = 4mm

113
6. Process Capability

Process Capability „weight“

Measurable Specification /
Customer Requirements:
▪ weight: 3,0 to 3,6 g

114
6. Process Capability

Process Capability „weight“ and larger customer specification

Measurable Specification /
Customer Requirements:
▪ weight: 2,6 to 4,0 g

115
6. Process Capability

Process Capability „longest diameter“

Measurable Specification /
Customer Requirements:
▪ Longest diameter: 10,0 to 26,0 g

119
6. Process Capability

Appendix
For centered processes

120
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

123 12/5/2018
7. Hypothesis Testing

Objectives of 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)
◼ 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

124
7. Hypothesis Testing

Basic Terms of Hypothesis Testing

Ho: Null hypothesis – Pronounced: “H-Naught”, “H-Not”

▪ No difference exists between data sets
▪ No relationship exits between data sets

Ha: Alternative hypothesis

▪ There is a difference between data sets
▪ There is a relationship between data sets

p-value: Probability value

▪ Likelihood that differences or relationships are not due to chance
▪ The probability of being wrong if we accept the alternative hypothesis (reject the null)
▪ Used to check to see if the data provides evidence that there is a difference
▪ This is the quantitative measure that indicates the result of the test and helps us to make a decision

125
7. Hypothesis Testing

Basic Terms of Hypothesis Testing (cont.)

Test statistic
▪ The number that captures the information in the sample data. It is used to decide between the null and alternative hypotheses

α - Significance level
▪ The maximum probability (risk level) tolerated for rejecting a true null hypothesis

Rejection region
▪ Is the range of values of the test statistic that will lead us to reject the null hypothesis. It is defined by the critical value. The
area of the rejection region is α, the significance level.

Confidence level
▪ Defined as the region of acceptance or 1-significance level (1- α). Typical values are 99, 95 and 90%. Needs to be set at the
beginning of the test

126
7. Hypothesis Testing

The Null hypothesis vs Alternative Hypothesis

Characteristics of Null hypothesis Characteristics of Alternative hypothesis

◼ By convention, our default hypothesis is that ◼ The Alternative Hypothesis says that:
we have not found a significant X: ◼ The sample is not the same as the
◼ The sample is no different from the known standard or a population
known standard or population ◼ There is a difference between the two
◼ There is no difference between the two populations
populations ◼ There is a relationship between two
◼ There is no relationship between two variables
variables ◼ The Alternative Hypothesis:
◼ This default hypothesis is called the Null ◼ Is the opposite of the Null Hypothesis
Hypothesis. The null hypothesis: ◼ Always has an inequality sign: , , or 
◼ Always has an equality sign: =, , or  ◼ Is designated by the term Ha
◼ Is designated by the term H0
(pronounced H-oh)

Example, Ho:   3 Example: Ha:   3

127
7. Hypothesis Testing

Hypothesis tests in everyday life

◼ Ho: Age doesn’t matter in a company’s hiring practices

◼ Ha: Age does matter in a company’s hiring practices

◼ Ho: Data is Normal

◼ Ha: Data is not Normal

◼ Ho: Batch X Avg. Cycle Time = Batch Y Avg. Cycle Time

◼ Ha: Batch X Avg. Cycle Time  Batch Y Avg. Cycle Time

Ho = _______________________________________

Ha = _______________________________________
128
7. Hypothesis Testing

Defining Ho and Ha

Is the average amount of TV viewing for the population 12 hrs?

1. State the question statistically:  = 12
2. State the opposite statistically:   12
3. State the null hypothesis: H0:  = 12
4. Select the alternative hypothesis: Ha:   12

129
7. Hypothesis Testing

The p-value tells us to reject or not to reject a Null hypothesis

◼ After data is collected, we calculate a P-Value (the probability of Ho being true)

◼ The P-Value is based on a probability distribution (Normal, T-distribution, Chi-Square, F-distribution, etc.)

Small P-Value Large P-Value

Ho is Rejected Ho is Not Rejected

If P is Low, Ho Must Go!

!
How low must P-Value be?
◼ We would like there to be less than a 10% chance that these observations could have occurred randomly
( = .10)
◼ Five percent is much more comfortable ( = .05)
◼ One percent feels very good ( = .01)
◼ The selection of the alpha level is based on the consequences of an incorrect decision to reject the null
hypothesis and accept the alternative hypothesis

130
For most cases we will use an  -Value of .05
!
7. Hypothesis Testing

QC inspection risk example

Ho - Product meets spec
Ha - Product does not meet spec

QC Decision
Product in spec Product not in spec
Product
in spec

Type I Error Consequence:

Correct
(-Risk) Producer scraps in-spec product
The Truth

not in spec
Product

Type II Error
Correct
( -Risk)

Consequence: Consumer gets out of spec product

Note: Type II Error and  risk are often called Consumers risk
Type I error and α risk are often called Producers risk
131
7. Hypothesis Testing

Hypothesis Tests for Averages

Roadmap

You want to compare the averages of some samples of data to decide if they are statistically
different.

Are the samples Normally You could try transforming the

distributed? data using Data
Transformation.
Yes
How many samples do you
want to compare?

One Two Three or more

1 Sample t-Test 2 Sample t-Test One Way ANOVA
Comparing the average of a Comparing the averages of two Comparing the averages of three
sample against a specific target. samples against each other. or more samples.

Paired t-Test
Comparing the averages of two
samples that are linked in pairs.

132
7. Hypothesis Testing

1-Sample t Test is used to compare sampling data to some

target value

1-Sample t Test

Comparisons involving
NOTE:
1 level for the X
With only 1 group (1 level for X), we’re talking about comparing our sample data to
some target value (e.g.,  = 25 or s = 2 )
!
Study Stability (if applicable)

Study Shape
Example of 1 Level:
Study Study 110 measures of the longest diameter of our M&M. Can we check, if the planned diameter
OR
Spread Centering size of 18.6 mms produced as expected?

133
7. Hypothesis Testing

Hypothesis Testing

◼ Null Hypothesis (H0):

The mean value of the maximum diameter is
EQUAL the target value of 18.6 mm
◼ Alternative Hypothesis (Ha):
The mean value of the maximum diameter is
NOT EQUAL the target value of 18.6 mm

134
7. Hypothesis Testing

1-Sample t Test Roadmap

Data Type
1-Sample t Test Attribute (Discrete) Continuous (Variable)
compare sample mean against a target Control Chart: Control Chart:
e.g. Ho:  ≤ 25 Ha:  > 25 1. Assess Stability
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

Determine sample size required 3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Take sample data 5. Test Relationships

Chi Square Regression
(X Attribute) (X Continuous)

No Need to remove
Is the data stable?
‘special causes’ first Study Stability
Yes (Tool: Control Charts)

Is the data normal? Study Shape

(Tool: Normality Test
Yes No
e.g. Graphical Summary)

Use 1-Sample Use Non-Parametric Test:

t Test 1-Sample Wilcoxon Test for Median Study Centering
135
7. Hypothesis Testing

Minitab Exercise – 1-Sample t Test

For a Single Sample

Longest A practical question:

Diameter
(mm)
18.9
18.6
The longest diameter of the M&M should
always be 18.6 mm. Can we confirm this
from the given data?
?
15.2
16.5
21.5
17.9
21.9
…

136
7. Hypothesis Testing

Minitab Exercise – 1-Sample t Test

Open the Minitab 18 project file m_ms_measurements_wo_participants.mpj

Comparisons involving
1 level for the X

Study Stability (if applicable)

Study Shape

Study Study
OR
Spread Centering
The operator is not considered

137
7. Hypothesis Testing

Follow the Analysis Roadmap

Use Control Charts to assess Stability: Data Type

Attribute (Discrete) Continuous (Variable)

Control Chart: Control Chart:

1. Assess Stability
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Chi Square Regression

5. Test Relationships
(X Attribute) (X Continuous)

138
7. Hypothesis Testing

Follow the Analysis Roadmap

Use 1-Sample t Test to study Centering for

one sample against reference (target):

Data Type

Attribute (Discrete) Continuous (Variable)

Control Chart: Control Chart:

1. Assess Stability
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Chi Square Regression

5. Test Relationships
(X Attribute) (X Continuous)
Hypothesis
Ho:  = 18.6
Ha:  ≠ 18.6

139
7. Hypothesis Testing

How to conduct a 1-Sample t Test in Minitab?

140
7. Hypothesis Testing

Minitab Output – 1-Sample t Test

One-Sample T:
In this example, the required confidence
interval has been set to 95%, so the Alpha
Level is 5%.

? What decision do we make

The p value is 0.956, which is not lower than the alpha level, so you cannot reject
P-Value the Null. In other words, you cannot say, that the longest diameter is not 18.6 mm.

H0: Longest Diameter mean  = 18.6

Ha: Longest Diameter mean  ≠ 18.6

141
7. Hypothesis Testing

Minitab Exercise – 1-Sample t Test

longest Practical question:

diameter
for t
20.9
20.6
The longest diameter of the M&M should
now (after process improvement) be bigger
than 18.6 mm. Can we confirm this from
the given data?
?
17.2
18.5
23.5
19.9
23.9
22.6
...

142
7. Hypothesis Testing

Minitab Exercise – 1-Sample t Test

The p value is 0.000, which is lower than the alpha level, so you can
reject the Null in favor of H1. In other words, you can say, that the longest
diameter is > 18.6 mm.

143
7. Hypothesis Testing

Hypothesis Tests for Averages

Roadmap

You want to compare the averages of some samples of data to decide if they are statistically
different.

Are the samples Normally You could try transforming the

distributed? data using Data
Transformation.
Yes
How many samples do you
want to compare?

One Two Three or more

1 Sample t-Test 2 Sample t-Test One Way ANOVA
Comparing the average of a Comparing the averages of two Comparing the averages of three
sample against a specific target. samples against each other. or more samples.

Paired t-Test
Comparing the averages of two
samples that are linked in pairs.

144
7. Hypothesis Testing

Minitab Exercise – 2-Sample t Test

Daniela Alex Practical question:

18.9
18.6
15.2
16.5
16.1
19.9
19.6
18.0
How do the measuring results
of the longest diameter (mm)
from Daniela and Alexander
compare?
?
21.5 22.2
17.9 16.2
21.9 19.0
20.6 18.9
17.6 20.6
... ...
longest diameter mm
145
7. Hypothesis Testing

2-Sample t Test is used to compare some aspect of the X from

one level to another

2-Sample t Test

Comparisons involving NOTE:

2 levels for the X ◼ With 2 levels for the X, we will be comparing some aspect of the X from one level to the
other
◼ There are two ways to compare the means of two
Study Stability (if applicable)
samples of normally distributed data:
Study Shape

Study Spread
◼ 2-Sample t Test
◼ One-Way ANOVA
◼ We will examine just the 2-Sample t Test here
!
Study Centering
Example of 2-level X
Did Daniela and Alex have different measuring results ?
146
7. Hypothesis Testing

Hypothesis Testing

◼ Null Hypothesis (Ho):

The mean value of Daniela’s measuring of
the maximum diameter is NOT DIFFERENT
from Alex’s mean value.
◼ Alternate Hypothesis (Ha):
The mean value of Daniela’s measuring of
the maximum diameter is DIFFERENT from
Alex’s mean value.

147
7. Hypothesis Testing

2-Sample t Test Roadmap

Data Type
2-Sample t Test Attribute (Discrete) Continuous (Variable)
compare mean of 2 samples Control Chart: Control Chart:
1. Assess Stability
Ho: 1 = 2 Ha :1  2 I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

Determine sample size required 3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Take sample data 5. Test Relationships

Chi Square
(X Attribute)
Regression
(X Continuous)

No Need to remove Study Stability

Is the data stable?
‘special causes’ first (Tool: Control Charts)
Yes Study Shape
No (Tool: Normality Test
Is the data normal?
e.g. Graphical Summary)
Yes
Test for Equal Variances
(Use 2 Variances Test)

Variances Equal? Study Spread

(Tool: Test for Equal Variances)
Yes No

Use 2-Sample t Test Use 2-Sample t Test Use Non-Parametric Test:

Study Centering
(“assume equal var.”) (“assume unequal var.”) Mann-Whitney Test for Medians
148
7. Hypothesis Testing

Minitab Exercise – 2-Sample t Test

Open the Minitab 18 project file m_ms_measurements_wo_participants.mpj

Comparisons involving
2 levels for the X

Study Stability (if applicable)

Study Shape

Study Spread

Study Centering

149
7. Hypothesis Testing

Follow the Analysis Roadmap

Data Type

Attribute (Discrete) Continuous (Variable)

Use Control Charts to assess Stability: Control Chart: Control Chart:
1. Assess Stability
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Chi Square Regression

5. Test Relationships
(X Attribute) (X Continuous)

150
7. Hypothesis Testing

Follow the Analysis Roadmap

Data Type
Use e.g. Probality Plot* to study Shape: Attribute (Discrete) Continuous (Variable)

Control Chart: Control Chart:

1. Assess Stability
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

4. Study Centering
ANOVA

Chi Square Regression

5. Test Relationships
(X Attribute) (X Continuous)

151
7. Hypothesis Testing

Test for Normal Distribution

Anderson-Darling
hypothesis H0 :There is no difference between the data and a normal distribution (The data is normally distributed)
alternative hypothesis Ha :There is a difference between the data and a normal distribution

Stat – Basic Statistics – Graphical Summary Stat – Basic Statistics – Normality Test
Alexander
Daniela

p-value > 0,05 distribution is a normal distribution

152
7. Hypothesis Testing

Testing for Normality of the Data

Anderson Darling Normality Test - Overview Data Type

Attribute (Discrete) Continuous (Variable)

Control Chart: Control Chart:

1. Assess Stability

▪ Histograms with normal curves and

I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot
probability plots provide a graphical
3. Measure Spread N/A Test of Equal Variance

method of assessing Normality. The N/A Test of Means / Medians

4. Study Centering
Anderson Darling method provides a Chi Square
ANOVA

Regression
5. Test Relationships
Hypothesis test for assessing (X Attribute) (X Continuous)

Normality.

▪ The Null hypothesis is that the data is

Normally distributed
▪ The Alternative Hypothesis is that the
data is not Normally distributed
▪ Assuming an alpha level of 5% (which
is the probability of making the wrong
decision when the null hypothesis is
true), every p-value >5% assumes
153 Normal Distribution of the data.
7. Hypothesis Testing
Data Type

Follow the Analysis Roadmap 1. Assess Stability

Attribute (Discrete)

Control Chart:
Continuous (Variable)

Control Chart:
I-MR Chart or p-Chart Individuals or I-MR Chart

Descriptive Statistics, Histogram,

2. Study Shape Binomial or Poisson Distributions
Normal Probability Plot

3. Measure Spread N/A Test of Equal Variance

N/A Test of Means / Medians

Use 2-Sample t Test to study Centering for two samples: 4. Study Centering
ANOVA

Chi Square Regression

5. Test Relationships
(X Attribute) (X Continuous)

Hypothesis
Ho: 1 = 2
Ha: 1  2

154
7. Hypothesis Testing

Minitab Output –
Test for Equal Variances (2 Variances)

What decision do we make

155
?
7. Hypothesis Testing

Minitab Output – 2-Sample t Test

P-Value

What decision do we make? ?

156
7. Hypothesis Testing

Minitab Output – 2-Sample t Test – Alternative Measures

Later that day in the quality lab, Daniela and Alexander created a second dataset of the measurements from the
longest diameter of the M&Ms (→ column C11: longest diameter for t)

Let’s find out together, if these data are still having the same distribution in terms of mean.

157
7. Hypothesis Testing

Minitab Output – 2-Sample t Test – Alternative Measures (2/3)

The boxplot already shows some differences of the average measuring size for the diameter
(Daniela 20.58mm, Alexander 18.64mm)
7. Hypothesis Testing

Minitab Output – 2-Sample t Test – Alternative Measures (3/3)

The boxplot already shows some differences of the average measuring the size of the diameter for t
(Daniela 20.58mm, Alexander 18.64mm), and…

What decision do we make? ?

7. Hypothesis Testing

Summary – Hypothesis Testing Terms

1. Null Hypothesis (Ho): Statement of no change or difference. This statement is assumed true until sufficient evidence is
presented to reject it. Contains equal sign.
2. Type I Error: The error in rejecting Ho when it is in fact true, or in saying there is a difference when, if fact, there is no
difference
3. Alpha Risk: The maximum risk or probability of making a Type I Error.
◼ This probability is always greater than zero, and is usually established at 5%.
◼ The researcher makes the decisions to the greatest level of risk that is acceptable for a rejection of Ho
◼ Producers risk: Another name for Type I error and α risk
4. Alternative Hypothesis (Ha): Statement of change or difference. This statement is considered true if Ho is rejected
5. Type II Error: The error in failing to reject Ho when it is in fact false, or in saying there is no difference when there really is a
difference
6. Beta Risk: The risk or probability of making a Type II Error, or overlooking the solution to the problem
◼ Consumers risk: Another name for Type II error and β risk
7. Significant Difference: The term used to describe the results of a statistical hypothesis test where a difference is too large
to be reasonably attributed to chance

160
Training Objectives
BASF OpEx Academy
Statistical Process Analysis
1. Introduction 4
2. Framework on Statistics 20
3. Analyzing Data 34
4. Process Stability 51
5. Measurement System Analysis 80
6. Process Capability 90
7. Hypothesis Testing 120
8. Quiz 158

161 12/5/2018
8. Quiz

162
8. Quiz

Wer wird Millionär

▪ Which types of data exist?

▪ Which visualization plots exist? (name at least 3)
▪ What means UCL?
▪ What means MSA?
▪ Which name comes into mind with normal distribution?
▪ Which Hypothesis Tests do you know? (name 2)
▪ What means DPU?
▪ What means Cp?
▪ Which data types exist in Minitab? (D, T, numeric)

163
8. Quiz

OpEx-LEAN SixSigma Green Belt

Mehr Informationen zur OpEx Academy unter:
https://www.lean.europe.basf.net/portal/basf/en/dt.jsp?set
Cursor=1_655029

164
8. Quiz

OpEx-LEAN SixSigma Black Belt

Mehr Informationen zur OpEx Academy unter:
https://www.lean.europe.basf.net/portal/basf/en/dt.jsp?set
Cursor=1_655029

165