Transcript Attribute Measurement Systems Analysis

Chapter 3: Attribute Measurement
Systems Analysis (Optional)
3.1 Introduction to Attribute Measurement Systems Analysis
3.2 Conducting an Attribute MSA
1
Chapter 3: Attribute Measurement
Systems Analysis (Optional)
3.1 Introduction to Attribute Measurement Systems
Analysis
3.2 Conducting an Attribute MSA
2
Objectives



3
Introduce the basic concepts of an attribute
measurement systems analysis (MSA).
Understand operational definitions for inspection
and evaluation.
Define attribute MSA terms.
What Is an MSA?




4
A measurement systems analysis is an evaluation of
the efficacy of a measurement system.
It is applicable to both continuous and attribute data.
An attribute MSA evaluates whether a classification
system correctly sorts items.
Companies make decisions each day based on
classifications; it is necessary to evaluate the efficacy
of such classifications.
Operational Definitions
In order for a rater to decide if a product is defective
or not, he must have a clear description, or an
operational definition, of what constitutes a defect.
Such a definition might include the following:
 photographs
 physical specimens
 descriptions
 specifications.
5
Effectiveness
The effectiveness of an inspection process
 is the percentage of time that a rater, or other
measurement tool, is correct in its classification
of quality
 is often significantly low before any attempts at
improvement are instigated
 should be at least 95%.
Effectiveness = number of correct evaluations
number of total opportunities
6
3.01 Multiple Choice Poll
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is his effectiveness?
a. .67
b. .067
c. .1
d. .93
e. None of the above
8
3.01 Multiple Choice Poll – Correct Answer
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is his effectiveness?
a. .67
b. .067
c. .1
d. .93
e. None of the above
9
False Alarms
A false alarm is a non-defective item that is classified
as defective.
The probability of a false alarm, also known as Type I
error or producer’s risk, is given by:
P(False Alarm) =
10
number of false alarms
number of non-defective items
3.02 Multiple Choice Poll
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is the probability of a
false alarm?
a. .67
b. .067
c. .1
d. .93
e. None of the above
12
3.02 Multiple Choice Poll – Correct Answer
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is the probability of a
false alarm?
a. .67
b. .067
c. .1
d. .93
e. None of the above
13
Misses
A miss is a defective item that is classified as
non-defective.
The probability of a miss, also known as Type II
error or consumer’s risk, is given by:
P(Miss) =
14
number of misses
number of defective items
3.03 Multiple Choice Poll
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is the probability of a
miss?
a. .67
b. .067
c. .1
d. .93
e. None of the above
16
3.03 Multiple Choice Poll – Correct Answer
Suppose 100 windshields are inspected, and 10 are
defective and 90 are non-defective. If an inspector
decides that 6 non-defectives are defective, and 1
defective is non-defective, what is the probability of a
miss?
a. .67
b. .067
c. .1
d. .93
e. None of the above
17
Escape Rate
An escape rate gives the percentage of time a
customer is likely to see a defective item.
Escape Rate = P(Miss) × P(Defect)
where P(Defect) =
18
number of defects
number of items inspected.
Bias





19
Bias is the tendency of an inspector to classify items
either as defective or as non-defective.
Bias is defined as P(False Alarm)/P(Miss).
Bias =1 implies there is no bias.
Bias < 1 implies a bias towards accepting bad items.
Bias > 1 implies a bias towards rejecting good items.
3.04 Multiple Choice Poll
The bias is given by the probability of a false alarm
divided by the probability of a miss. In the windshield
example, the bias is given by .067/.1 = .67. What is the
interpretation of this value?
a. There is no bias.
b. There is a bias towards accepting bad items.
c. There is a bias towards rejecting good items.
21
3.04 Multiple Choice Poll – Correct Answer
The bias is given by the probability of a false alarm
divided by the probability of a miss. In the windshield
example, the bias is given by .067/.1 = .67. What is the
interpretation of this value?
a. There is no bias.
b. There is a bias towards accepting bad items.
c. There is a bias towards rejecting good items.
22
Rater Agreement
Rater agreement
 is a measure of how well raters agree with each other
 is not an indication of correctness
23
Kappa Statistic
The Kappa statistic
 is used to measure between-rater variability, or how
often two or more raters agree in their interpretations
 is a measure and not a test
 is given by:
kappa = po – pe
1 – pe
where po is the sum of observed proportions in diagonal
cells of the contingency table and pe is the sum of
expected proportions in diagonal cells of the contingency
table.
24
25
Chapter 3: Attribute Measurement
Systems Analysis (Optional)
3.1 Introduction to Attribute Measurement Systems Analysis
3.2 Conducting an Attribute MSA
26
Objectives


27
Examine the requirements for an attribute MSA.
Perform an attribute MSA in JMP.
Sample Size
To conduct an attribute MSA, the minimum recommended
sample sizes are given as follows:
Number of
Raters
28
Minimum Number
of Parts
Number of
Evaluations
1
40
3
2
30
3
3+
25
3
Attribute MSA Example


29
Suppose three inspectors, Henry, Matt, and Tom, are
independently going to classify each of 30 parts as
defective or non-defective in a random order. They will
evaluate each part three different times. Of the 30
parts, 13 are defective and 17 are non-defective.
The classification will be based on a predetermined
operational definition of defective and non-defective.
Attribute MSA
This demonstration illustrates the concepts
discussed previously.
30
31
Exercise
This exercise reinforces the concepts discussed
previously.
32