OR 681/SYST 573
Fall,� 2007
Decision Theory And Analysis
����������������������������������������������� ���� ������������������������������������������������������ �Instructor: Dave Schum
------------------------------------------------------------
Course Description And Overview
����������� This course concerns a variety of
issues encountered in research on decision
processes and the bearing of this research on the design of procedures and
systems for assisting people as they perform decision tasks in the face of
uncertainty. In many contexts such tasks involve difficult probabilistic
inferences and equally difficult value or worth assessments. That people need,
and frequently request, assistance in the performance of these tasks is no
secret. What is at issue is the manner in which such assistance should be given
and the various forms such assistance might take. Sensible recommendations for
assisting people in decision and inference tasks require careful analyses of
the tasks themselves as well as careful examination of the many judgments these
tasks require. This course concerns research on inference and decision tasks
themselves as well as on the judgmental requisites of these tasks. Many
attempts to provide assistance in inference and decision tasks miscarry because
of inadequacies in the manner in which the task is structured or because of
innocence about the often-difficult judgmental requirements of these tasks.
Research on methods for enhancing our inferential and decisional skills forms
an important element in the Volgenau School Of Information Technology And
Engineering, of which I am a member.
����������� Before I tell you about the specific
topics we will discuss in this seminar, I must make sure that you are aware of
the essential objectives of this course so that you can make an informed choice
about whether or not you wish to enroll in it or stay enrolled. A major
objective of this course is to provide you with a careful and useful account of
the essential ingredients of inference and decision tasks and how these
ingredients might be combined. There is now a very substantial amount of
research on a wide array of decision-related topics. This research occupies
persons from many different disciplines. No single discipline contains a
repository of all useful information about human inference and choice.
Naturally, we can't cover all of this research in a single semester and so I
have had to exercise some selectivity.
����������� This is definitely not a
vocationally oriented course. I make no claim that, as a result of taking this
course, you will instantly be able to solve some decision-related problem you
now face in your work or in your everyday activities. Such a claim would be
very difficult for me to make since I have no present awareness of the
particular decision problems you currently face. But what I will guarantee is
that the topics we will cover in this course have direct relevance to inference
and choice tasks regardless of the context in which they occur, including the
ones of current interest to you. Be assured that in this course you will have
ample opportunity to make us all aware of the particular decisions your present
work may require. Students frequently complain that a course is too
theory-oriented and offers no direct prescriptions for application. However,
sensible application of some procedure or technique assumes that we understand
the capabilities and limitations of these procedures. One major trouble is that
people often make choices without being aware of all the ingredients these
choices require. So, one of my major objectives in this course is to provide
you with a thorough assessment of what ingredients are required in every choice
made in the face of uncertainty.
����������� As you notice, this course is
cross-listed as Systems Engineering 573 and Operations Research 681. I welcome
all of you and hope that I can provide a learning experience that is congenial
to your interests regardless of what disciplinary affiliation you now have.
Matters we will discuss in this course certainly cut across the areas just
mentioned and many others as well. As I proceed, I will try to provide specific
examples of particular issues that arise in these disciplines. As I will
mention again below, I will provide you with a number of assignments to test
your mastery of the matters we discuss. As you will see, there are certain
algorithms employed in the analysis of inference and choice. In other words,
there is a computational element of these tasks. Some decision courses focus
almost entirely on the application of these algorithms in particular
situations. We will indeed examine some of these algorithms but my emphasis
will be upon how these algorithms arise and on the consequences of employing
them in various situations. Many of you have an interest in the topic of risk
and its assessment and management. At various points in this course we will
dwell on how risk is characterized and assessed by various means.�
����������� This course will concern, in roughly
equal proportions, structural, probabilistic, and value-related elements of
human choice. These elements are rarely, if ever, provided for us; they have to
be discovered or generated by imaginative thought. To date there is no machine
that can generate decision options and all their possible consequences and then
assess the value of these consequences and determine how likely they are to
occur. There are decidedly non-trivial measurement and judgmental issues that
arise in human inference and choice, many of which we will discuss in this
seminar. Inference and decision problems never spring forth in well-posed form;
they have to be formulated or structured in the certain knowledge that such
structures may have to be altered in light of new evidence and corrected or
revised insights. Until quite recently, attention to the structural attributes
of human inference and choice were overlooked in all but a small amount of
scholarship. Also overlooked has been the role of imaginative reasoning
in generating options and consequences, hypotheses or possibilities giving rise
to these consequences, and defensible evidentiary tests of these possibilities.
The discovery of essential decision problem ingredients is of crucial
importance and will command our special attention.
����������� In our discussion of the
probabilistic inferential elements of choice we will pay careful attention to
the evidence upon which inferences are made and to alternative conceptions of
the process of assessing and combining the "weight" or
"force" of evidence in reaching a conclusion. This is quite an
interesting time to be studying probabilistic reasoning. New systems of
inference and probability have shed considerable light on some age-old problems
associated with the task of drawing conclusions from incomplete, inconclusive,
ambiguous, dissonant, and unreliable information. But there is still quite a
bit of life left in conventional views of probability and inference when these
views are expanded to incorporate the complex attributes of human inference
observable in so many real life circumstances.
����������� Our discussion of the value-related
ingredients of human choice will focus upon the generation of options and their
consequences as well as upon judgmental problems associated with placing a
value on these consequences. In many situations, identified consequences have
many attributes, some of which may be in conflict and that require us to
consider various trade-offs we must make. Identification of consequence
attributes and assessing their relative importance are usually not simple
tasks. We will examine, in some detail, a variety of measurement issues that
arise during the process of assessing human preferences. Study of the relation of
these measurement issues to other forms of behavioral measurement is quite
instructive. Of obvious concern to us is consideration of the various
strategies that exist for combining our uncertainties and values in making a
choice. Although there is no uncontroversial strategy for combining assessments
of probabilities and values, there are some that prove very useful in many
situations.
����������� The word rationality is very commonly used with reference to the evaluation
of human inference and choice. We frequently hear someone say that another
person's inference or choice was not a "rational" one. The term
"rationality" slips off the tongue rather easily, especially in
connection with the inferences and choices made by others. However, just what
constitutes a "rational" inference or choice is far from being
settled. Indeed, what rationality means is just one among many controversial
topics we will consider. In spite of this controversy about what rationality
means, there is some well-publicized behavioral research being used in support
of the claim that we are all deficient in our basic inferential and decisional
capabilities. In other words, the claim is made that we are all frequently
irrational. Taken seriously, this research generates little confidence in the many
societal institutions and organizations whose activities rest upon human
inferences and decisions. But there are other views about the suitability of
such research that offer a more charitable assessment of our basic intellectual
competence. We will have a brief look at both sides of this current
controversy.
����������������������������������������������� Outline Of Major Topics
����������������������� ����� [Extensive Notes provided on each of
these topics]
Section I: Preliminaries.
����������� A. Essential Ingredients Of Human
Inference And Choice.
����������� B. On The Adequacy Of Decisions:
What Constitutes "Rational" Inference and ��� ����������� ����������� �����Choice?
����������� C. Normative-Descriptive Issues.
����������� D. Judgment and Measurement Issues.
����������� E. Some Examples of Choice
Paradigms.
����������� F. Decision Analysis: Divide and
Conquer
Section II: On the Imaginative and Structural Elements of
Inference and �������� ����������� �������Choice
����������� A. Three Forms Of Reasoning To
Consider.
����������� B. Probabilistic Reasoning and the
Construction of Inference Networks
����������� C. Decision Structuring: Decision
Trees and Influence Diagrams
Section III: Evidence and Probabilistic Reasoning.
����������� A. On the Credentials of Evidence:
Relevance, Credibility, and Inferential
����������� ���
Weight or Force.�
����������� B. Logically-Distinguishable Forms
And Types Of Evidence.
����������� C. Alternative Views Of
Probabilistic Judgment and Reasoning.
����������� D. A Survey Of Evidential And
Inferential Subtleties.
����������� E. Probabilistic Reasoning and the
Science of Complexity
�����������
Section IV: Decision Consequences and Values.
����������� A. Value-Utility: A Brief Historical
Summary.
����������� B. Basic Formal Issues In The
Assessment Of Value/Utility.
����������� C. Single-Attribute Utility:
Concepts And Assessment Methods.
����������� D. Multiattribute Utility
Assessment.
�����������
����������������������������������������������� Reading Materials
����������� As I noted, no single discipline can
lay claim to all the matters of interest in this course. I now extend this idea
even farther. No single instructor can tell you all there is to be said about
the complexities of human inference and choice; neither can any single
textbook. Human inference and choice draw upon intellectual processes of the
highest order. Consequently, those who study these processes soon learn to take
all the help they can get, from wherever it comes. Textbooks come and go. For
several years I used two books [both paperback] that were quite well accepted
by students enrolled in this course. Unfortunately, they are both now out of
print. So, I have had to decide on another book for this course. The one I have
chosen is:
����������� Clemen, Robert T. Making Hard Decisions: An Introduction to
Decision
����������������������� Analysis. Duxbury Press, Belmont, CA 1996 [2nd edition].
����������� Why would I ask you to read a book
for which there are later editions? You may be aware of the fact that in the
state of Virginia, legislators have exhibited great concern over the costs to
students of the textbooks they are required to purchase for their courses. You
all have great experience in such matters. The cost of textbooks is larcenous
as we all know. Instructors are now required to adopt older versions of
textbooks, if they are available, and for which there are likely to be used
copies. The later editions of the Clemen book are certainly more
attractive and� come with a CD. But they
do not cover essential matters any better than the older version I have asked
you to purchase.� But what I have
discovered, much to my regret, is that the older edition of this book is still
larcenously priced. ����
����������� The edition of the Clemen book I
have asked you to purchase has been well received and covers many topics in
decision analysis. It is a long book and I will not ask you to read all of it.
Particular assignments from this book appear below. I will try my best to make
our discussions in class correspond with the order in which topics are
discussed in this book. This is not always easy to do since I will tell you
about many things Clemen does not mention in his book. If all I told you in
class was also said in this book, you should complain very loudly to my
Departmental Chairperson or to my Dean. On some topics, you will find your
instructor in disagreement with what your author says. I cannot possibly
overemphasize how controversial are many essential topics in human inference
and choice. Many matters we discuss will continue to be subjects for debate and
discourse for a very long time.
����������� In addition to Clemen's book, I will
regularly provide you with extensive notes and handouts on all matters
we discuss in class. I will have these notes for you in advance of the time
they are to be discussed in class so that you can think about these matters and
challenge me about what I have told you. You cannot challenge Clemen, but you
can certainly challenge me each time we meet in class. The discussions we
have in class are most important. Be assured that in class I will not
simply go over the notes I have given you in advance. In a few cases this will
be necessary because of the nature of the topic or the particular notes I will
give you. The notes I give you simply form a major vehicle for getting our
discourse started during each class. You can provide other vehicles by raising
particular decision problems you now face in your work or that are currently
being faced by persons in various governmental military, business, or other
organizations. In short, my major objective in class will be to draw ideas
out of you. I am a scholar, not a preacher, and so you should not expect to
hear me say that the ideas I will mention concerning inference and choice have
been settled for all time. Your own ideas and the experiences upon which they
are based are so important. I will do all I can to create an atmosphere in
class that makes you willing to share your ideas and experiences with others in
the class.
����������������������������������������������� Reading Assignments
����������� These reading assignments are keyed
to the outline of topics given above. Notice that I have not assigned all the
chapters and have assigned only parts of others.
Section I: Chapters 1 and 2, pages 1-40.
Section II: Chapter 3, pages 41 - 90,
����������� ����
Chapter 4, pages 101-116; 118-35
����������� ����
Chapter 6, pages 187-215
����������� ���������
Section III: Chapter 7, pages 219-264 [optional if you
have had a recent course
����������� ������ in probability]
����������� ������ Chapter 8, pages 265-298
Section IV:�
Chapter 13, pages 461-497
����������� ������ Chapter 14, pages 503-526
����������� ������ Chapter 15, pages 530-561
����������������������������������������������� Method Of Evaluation
����������� Your grade in this course will
depend upon two methods of evaluation: a paper and some assigned exercises. As
you will observe as we proceed, we will cover a fair number of different issues
and problems in the study and analysis of decision tasks. In short, the breadth
of coverage of decision-related topics in this seminar will be substantial. You
should naturally be inclined to ask the question: where does depth of
coverage come in? The answer to this question depends upon you and your
interests. It is natural to expect that you will find at least one topic
that captures your interest or that is especially relevant to the work you may
now be doing. Here is where depth comes in: on the topic you choose to
examine in as much detail as your interest and time allows. Naturally, I will
be prepared to help you find as much information as I can about this
topic.��
����������� 60% of your grade in this course
will be based upon a paper you write or project you perform (and describe) on some
topic of your choice, provided that this topic bears directly upon human
inference or choice. This paper must be an original work; i.e., papers
written to satisfy requirements of other courses are not acceptable. You
have every right to know what I expect of your papers or projects. The first
thing I will tell you is that if you are interested in your chosen topic and
have examined it carefully, the odds are strong that I will enjoy reading what
you have to say about it. As Samuel Johnson once remarked: "What is
written without interest is, in general, read without pleasure". As you
work on your paper or project, remember that you do this mainly for yourself.
Students frequently ask me: what do you want in this paper? The appropriate
question is: what do you [the student] want? My major objective in this
assignment is to give you the opportunity to satisfy your own curiosity about
some decision-related matter that captures your interests and about which you
wish to have some greater depth of understanding. I will of course be very
pleased to discuss with you ideas you may have for papers you could write or
projects you could perform.�
����������� In the 22 years I have offered this
course at GMU I have received papers on many different topics. Here is a brief
listing of some of the general themes of papers I have found quite interesting
and enjoyable to read.
����������� 1) An analysis of some past,
current, or future decision you and perhaps others face in the work you now do
in a business, military, or other governmental context. It would be quite
appropriate for you to apply what you have learned in this course to these
kinds of decisions. There is one caution here that I mention on the basis of
considerable prior experience with papers/projects of this sort. In some
instances a person will spend the entire paper describing the particular
institutional context in which decisions are required and end up saying little
or nothing about how these decisions could be formulated and analyzed using any
methods discussed in class or in the text I have assigned. The result is that I
am left completely in the dark about how much this person has learned in this
course. An additional problem I face is that, being usually innocent of the
particular details of the contexts in which you presently find yourself, I have
very little basis for evaluating what you have done. The moral here is quite
simple. If you choose to present an analysis of some current job-related
decision problem, make sure that you relate it to one or more concepts or
methods that have been of concern in this seminar.
����������� 2) An analysis in greater depth of
some particular inference or decision topic that has captured your interest in
this seminar.
����������� 3) As a result of your reading or
our discussions in class, you may wish to criticize current positions taken on
a particular issue either by your instructor or by other persons who have
written on the topic. As I have noted, very little about human inference and
choice is regarded as fixed for all time. As we proceed in this seminar, you
will find many topics about which there is unending controversy. Where there is
controversy, there is usually a graduate paper lurking. Your ideas about the
matters at issue may be every bit as valuable as those that now exist.
����������� 4) An analysis of some past,
current, or future choice you face in your personal life. People face decisions
all the time concerning employment opportunities, purchases, and many other
matters. It would be quite appropriate to employ the concepts and methods of
analysis we discuss to personal decisions you now face or have faced in the
past.
����������� These are just some of the general
topic areas that might get you started thinking about your paper or project.
There are other options of course. The final matter here concerns grades I
assign to papers or projects. Most students have more than a passing interest
in what grading standards a professor has in the back of her/his mind as far as
assignments, such as I have described, are concerned. Here, very generally, are
some characteristics of what I regard as A, B, and C papers/projects.
����������� Grade of C: A graduate
paper should get a grade of C if it reads essentially like a high school book
report. In some cases a student's paper reads like: " X says this, Y says,
this, and Z says that". In such an effort there has been no attempt by the
student to be critical of what X, Y, and Z said, nor has there been any attempt
to integrate what these persons have said in any meaningful or useful manner. A
report of a project or an analysis also gets a grade of C if I cannot find any
connection in the paper to any decision or inference concept or analytic method
we have discussed in this course.
����������� Grade of B: A critical and
well-integrated analysis of what others have done and said gets a grade of B on
a graduate paper. Similarly, a report of a project in some ongoing
context gets a grade of B if there is at least some attempt to relate it to the
essential topics in this seminar.
����������� Grade of A: If your analysis
of some topic is both critical and well-integrated and, in addition, if you
have provided your own thoughts about this matter in a defensible way,
this paper begins to look like an A paper. In graduate work what counts most is
the application of your own imaginative reasoning and critical thought to the
task at hand. The same idea extends to papers/projects on specific analyses of
decisions you encounter in your jobs or elsewhere. In short, what YOU think about the matters of concern
to you counts the most and is something on which all of us who teach graduate
courses place the highest value.
����������� The remaining 40% of your grade will
be based on four assignments I will give you. Some of these assignments will be
problems for you to solve. Others will require your critical and imaginative
thought about a decisional or inferential issue in some situation. Some will
come from your text.
����������� ***I would greatly appreciate
receiving your paper and other assignments in hard copy rather than in
electronic form. I would much rather spend the time reading your work
carefully, and providing you with comments, than waiting to download and print
your work.***
����������������������������������� Where To Find Your Instructor.
����������� I hope you will feel free to call
upon me at any time; do not hesitate to call me at home if you can't
find me at the office. I am a member of the full-time faculty at GMU. What this
means is that you can get all the help you need from me, when you need it. I
can usually be found at one of the following locations:
����������� Office:� Room 111-A, Science & Technology Bldg II.
GMU Phone: 703-993-1694�������������
����������������������� Home: 2219 Chestertown
Dr., Vienna, Va. Phone: 703-698-9515
����������� e-mail: Please use my home e-mail which
is: <[email protected]>
I do hope you enjoy
this course and I will do all I can to help make this a valuable experience for
you.
�����������