Using Attribute Data for Root Cause Analysis: The Dinner Party You Want to Miss

© 2017 RealKnowledge, Inc.
Using Attribute Data for Root Cause Analysis
The Dinner Party You Want to Miss
Scott Leek
Continuous Improvement Leader & Coach

The Dinner Party You Want to Miss
A group of ten volunteers worked
hard all year for their common cause.
To celebrate their outstanding
contributions to the community a
dinner party was planned in a private
room at a local restaurant. To keep
things simple, they all agreed to a
common menu that included:
• Pate
• Caesar Salad
• Asparagus
• Beef Wellington
• Chocolate Mousse

The Problem
Much to the group’s dismay, six of the
volunteers developed food poisoning.
A root cause investigation is initiated
to discover the source of the food
poisoning. The investigation intends to
answer the following:
• Which menu item is the source of
the food poisoning?
• What is the pathogen that
contaminated the menu item?
• What is the vector of infection for
the contamination?

Root Cause Analysis
Finding the root cause of the food
poisoning starts with collecting
attribute data about the diners and
the menu items:
• A list of each diner suffering food
poisoning (yes) and, a list of each
diner not suffering food poisoning
(no)
• A list of each menu item each diner
consumed (yes) and, a list of each
menu item each diner did not
consume (no).

Root Cause Analysis
• Data analysis is conducted using a simple set of methods developed
by John Stuart Mill.* The methods are:
– Method of agreement
– Method of difference
– Joint method of agreement & difference.
• Not only do these methods work, but they have the added benefits
of being simple and graphical, only requiring a simple set of tables
using color coding (Note: Mill obviously did not have the benefit of
spreadsheets and the attendant color coding).
• After identifying the offending menu item, further analysis is
required on the specific pathogen causing the illness, and the
vector of infection.

Organizing the Data
The data are organized in a simple table:
1. The instances and non-instances of food poisoning are
setup in the rows, each row is represented by a diner
2. The menu items are setup in the columns
3. The phenomenon (food poisoning) is placed in the
right most column
4. The table is populated with “yes” when the menu item
is consumed, or when the food poisoning is present
5. Cells are left blank when the menu item or food
poisoning was not present (no)
6. The menu items are sorted in descending order from
right to left based on the frequency of occurrence (yes)

The Data Table
Instances
Caesar
Salad Pate
Beef
Wellington Asparagus
Chocolate
Mousse
Food
Poisoning
Trish Yes Yes Yes Yes Yes Yes
Petra Yes Yes Yes Yes
Randy Yes Yes Yes
Kumar Yes Yes Yes Yes
Patty Yes Yes Yes
Edward Yes Yes Yes Yes Yes
Prunella Yes Yes Yes
Jake Yes Yes Yes Yes Yes
Gracie Yes Yes Yes Yes
Mitch Yes Yes Yes Yes Yes

Method of Agreement
• The method of agreement is one of Mill’s methods of scientific
induction. Quoting Mill on the method of agreement:
“If two or more instances of the phenomenon under
investigation have only one circumstance in common, the
circumstance in which alone all the instances agree, is the
cause (or effect) of the given phenomenon.”
• In terms of the food poisoning investigation, this means that if one
of the menu items is perfectly correlated with the occurrences and
nonoccurrences of food poisoning, it is the likely cause.

Method of Agreement
Instances
Caesar
Salad Pate
Beef
Chocolate
Mousse
Food
Poisoning
Randy Yes Yes Yes
Patty Yes Yes Yes
# Agree 10 4 4 6 6
# Disagree 0 6 6 4 4
Agree in the negativeAgree in the affirmative Disagree in the affirmative or negative

Method of Agreement
Instances
Caesar
Salad Pate
Beef
Chocolate
Mousse
Food
Poisoning
Randy Yes Yes Yes
Patty Yes Yes Yes
# Agree 10 4 4 6 6
# Disagree 0 6 6 4 4
A quick scan of the Method of Agreement table leads to the following conclusions:
• All ten diners ate Chocolate Mousse but four did not get sick
• Eight diners ate Asparagus but three did not get sick, one got sick who did not eat
Asparagus
• Six diners ate Beef Wellington but three did not get sick, three got sick who did not
eat Beef Wellington
• Six diners ate Pate but three did not get sick, and three got sick who did not eat
Pate

Conclusion: Method of Agreement
Instances
Caesar
Salad
Food
Poisoning
Trish Yes Yes
Petra
Randy
Kumar Yes Yes
Patty
Edward Yes Yes
Prunella
Jake Yes Yes
Gracie Yes Yes
Mitch Yes Yes
• Only the Caesar Salad is perfectly
correlated with the instances of
food poisoning:
– All six instances when diners
consumed the Caesar Salad, they also
suffered food poisoning
– All four instances when diners did not
consume the Caesar Salad, they did
not suffer food poisoning
• Due to the perfect correlation,
there is a reasonably high degree
of belief that the Caesar Salad is
the cause of the food poisoning.

The Method of Difference
The method of difference is another one of Mill’s methods of scientific
induction. Quoting Mill on the method of difference:
“If an instance in which the phenomenon under investigation
occurs, and an instance in which it does not occur, have every
circumstance save one in common, that one occurring only in
the former; the circumstance in which alone the two instances
differ, is the effect, or cause, or a necessary part of the cause,
of the phenomenon.”

The Method of Difference
• In terms of the food poisoning investigation, we compare one observed instance of
when the problem occurred, with a “nearly similar” instance when the problem
did not occur. The menu item causing the food poisoning should be present in the
instance when the poisoning occurred, and not present in the nearly identical
instance when it did not occur.
• Examining the Method of Difference table, we see that Trish’s and Petra’s choice of
menu items are nearly identical. Trish consumed every item on the menu, while
Petra consumed every item except for the Caesar Salad. Therein lies the
difference. The fact that Trish suffered food poisoning and Petra did not, provides
further evidence implicating the Caesar Salad.
Instances
Caesar
Salad Pate
Beef
Chocolate
Mousse
Food
Poisoning

Joint Method of Agreement & Difference
• If the method of agreement or the method of difference implicates the
Caesar Salad, there is a reasonable degree of belief to suspect the Caesar
Salad is the source of the food poisoning.
• The fact that the methods of agreement and the method of difference
both implicate the Caesar salad, increases the degree of belief that the
Caesar Salad is the source of the food poisoning.
• Mill called the consistent finding of both methods, the joint method of
agreement and difference.

Drilling Down to Root Cause
• Using the method of agreement, method of difference, and the joint method
of agreement and difference, the cause of the food poisoning is narrowed
down from five menu items to one, the Caesar Salad.
• The root cause however, is still unknown. Two additional questions require
investigation:
1. What is the pathogen that contaminated the Caesar Salad?
2. What is the vector of infection for the contamination?
• Answering the first question requires testing the Caesar Salad in a laboratory
for common pathogens (e.g., E.coli, Listeria, etc.). Assuming the laboratory
tests reveal that Listeria is the pathogen, the next level of investigation focuses
on discovering the vector of infection for the contamination. Once that is
known, the root cause is near, or at hand. Whether the former or latter
depends on the degree of belief required in the answer.

Summary
In this case study, attribute data (menu items consumed) are
used to discover the most likely cause of another attribute
(suffering from food poisoning). Mill’s methods of agreement,
method of difference, and joint method of agreement and
difference are used to narrow down the potential causes for
further investigation.

A Technical Note
• Some practitioners are eager to suggest a more “vigorous” method to
analyze the data, using for example, some type of hypothesis test. This is
questionable for three reasons:
– Context
– Approach
– Technicality
• Contextually this case is an analytic study, not an enumerative study.**
The purpose of an analytic study is to understand a dynamic system of
cause and effect. The understanding is used to act on the system to
improve future outcomes. The essence of an analytic study is the
“problem of prediction,” that is, future outcomes.

A Technical Note
• The purpose of an enumerative study is to describe a static population.
The description is used to act on the population described, not on future
outcomes. Clearly the purpose here is to prevent future outbreaks of food
poisoning, not describe the six diners suffering food poisoning. Thus, this
is an analytic not enumerative study therefore hypothesis tests are not
appropriate.
• In any situation, the simplest approach to answer the questions at hand,
are most desirable. Mill’s methods fit this bill. They are simple enough for
anyone who can create a table of results. The analysis is also simple,
requiring no more than a tally of results in four categories (i.e., consumed
and got sick, consumed and did not get sick, did not consume and got sick,
did not consume and did not get sick).

A Technical Note
• Technically Mill’s methods are superior for two reasons. First, the sample
size is small. Most hypothesis tests require more data than provided by
the ten instances and five menu items. Second, while the assumptions for
hypothesis tests vary depending on the specific test, they all assume that
samples are random and independent, clearly not situation in the food
poisoning case. The ten diners, six of whom suffered food poisoning, do
not represent a random sample. Further, the six poisoned diners suffered
food poisoning because they consumed the same menu item, hardly
independent.

About the Author
Scott is a Continuous Improvement Leader & Coach, and the CEO of
RealKnowledge, Inc. He has more than 30 years of experience helping leaders
and coaching practitioners in the improvement arts and sciences. Scott can be
contacted here.

Footnotes
* Mill’s methods are outlined in “A System of Logic,” 1843.
** Deming, W. Edwards, “On Probability as a Basis for Action,” The American
Statistician, Vol. 29, No. 4, 1975, pp. 146 – 152.

Credits
• Icons below used in the presentation made by Freepik from Flaticon are licensed
under Creative Commons BY 3.0

Using Attribute Data for Root Cause Analysis: The Dinner Party You Want to Miss

More Related Content

Similar to Using Attribute Data for Root Cause Analysis: The Dinner Party You Want to Miss (20)

Recently uploaded (20)

Using Attribute Data for Root Cause Analysis: The Dinner Party You Want to Miss