This chapter introduces you to the basics of decision theory and how it can be applied to improve your decisions. It starts with a review of material in Smart Choices, a book written by John Hammond, Ralph Keeney, and Howard Raiffa, experts from the Harvard Business School in the field of decision theory and its application in business. (An MBA program is largely the study of how to improve decisions important to business, and decision theory is a key component of this study.) It continues with a discussion of our goals for models intended to help us predict and explain social behavior. We see that our modeling focuses only upon the most important factors necessary to predict and explain, even if we must sacrifice a complete or accurate description of the individual’s actual decision making process.
Components of a Good DecisionEdit
We want the models that we will eventually build throughout this course to help us in predicting and explaining social behavior around us. Since we will be examining a class of models based upon individuals choosing alternatives that promote their self interest, it will be important for us to know what it means exactly for a choice to be rational and to which situations different levels of rationality might fruitfully be applied. To help us understand “rationality” better, to help us in modeling what decisions are actually made, we will first consider advice on improving the rationality of your decisions, or what decisions should be made. How should you decide to best meet your goals? What constitutes a rational choice?
Some decisions in your life will be fairly easy, but most of the important ones will be tough and complex with no obvious choice as the best. Even though a great deal may be at stake, for you and for those around you that you care about, with these tough decisions we risk anxiety from confusion and doubt and the fear of error, regret, and embarrassment. We want to provide you with an anxiety-reducing structure or process to help you with these important, tough decisions.
Find feasible alternativesEdit
Decision theory separates the selection of choices to be considered feasible from how to choose among these choices. It, thus, separates the analysis of what is feasible from the analysis of what is good. Each of these two key steps can also be broken into key components. This divide-and-conquer approach is an important benefit from the use of decision theory. Your assumptions, data, and logic become clear, where they might not be otherwise. As we will see later, separating the analysis of the feasible from analysis of the good is a major benefit from the use of game theory as well.
Ideally, you want to consider all possible feasible alternatives, rank them in order of your preferences, and then choose the best. In practice, the set of all possible alternatives may not be obvious. Further, many feasible options may be clearly sub-optimal, so for simplicity’s sake you might omit them from consideration. Even so, the alternatives in your model represent the range of potential choices you will have for pursuing your objectives. Since you’ll eventually be choosing the best among the alternatives that you consider, make sure that you include all possible best alternatives.
Reconsider the scope of your problemEdit
“A good solution to a well-posed decision problem is almost always a smarter choice than an excellent solution to a poorly posed one.” (16) Be creative about defining your problem; it is often broader than it first appears. The trigger that initiated the decision may suggest a relatively narrow problem, but upon reconsideration you learn that broader issues are involved and should be considered. For example, changes considered to deal with the trigger may open up complementary changes that would not be considered otherwise. (see p.18 for example)
In defining the problem consider the following: What is the trigger? What are the constraints to the problem? What are the essential elements of the problem? What decisions impinge on or hinge upon this decision? Can this problem be broken into smaller parts? After you have asked and answered these questions, talk with others about this problem, either experts that have a similar problem before or non-experts that have your interests at heart. You may gain insight just from your explaining your thinking, as well as from any advice itself.
Crafting an appropriate definition takes time, and it often can’t be done effectively in one sitting. Like the crafting of an introductory paragraph for a Freshman Studies paper, the problem definition often needs reworking as the body is developed. Like that introductory paragraph, keep revisiting it as your analysis proceeds. Reconsidering the problem definition is especially important when circumstances change rapidly or new information becomes available. Remember, if you have an inappropriate problem definition, at best, you may go very fast—but in the wrong direction.
- use your objectives and work backwards — ask “how?”
- Make sure that each of the constraints that limit your alternatives is real. If not, eliminate it. Even if they are real, can you find any ways around these constraints?
- challenge constraints
- Make sure that each of the constraints that limit your alternatives is real. If not, eliminate it. Even if they are real, can you find any ways around these constraints?
- set aspirations high
- “One way to increase the chance of finding good, unconventional alternatives is to set targets that seem beyond reach.” (51) Start with an alternative with a great outcome, but one that is also infeasible, and then modify it to make it feasible.
- give your subconscious time to operate
- Your mind doesn’t work just by direct attack. If you start on the problem early, you give yourself time to be reconsidering the problem in a “low” mode while doing something else (a shower? drifting off to sleep? working on another problem?). Sometimes good ideas come. Start early and give yourself the chance.
learn from othersEdit
Find out what you and others have done in similar situations. What lessons do you learn from these other situations? What could have been done better?
Others may be able to provide some insights from earlier experience with similar situations or they may suggest new paths that you haven’t considered. Also, keep an open mind during your conversations with others. The primary benefit may not be the specific ideas that others provide, but simply the insights that you gain from organizing your thoughts into explanations and from answering questions.
At the same time, talking with others first may limit you to their framework of the problem. Think about the problem first yourself, since you may find a new approach.
consider more alternativesEdit
- consider process alternatives
- Especially when you need to justify your choice to others or have their approval, the best alternative may be a process for making a choice, rather than the choice itself. Could you use a voting mechanism? an auction? set standards that would then be applied after gathering more information?
- search for win-win alternatives
- Again, especially when you need to justify your choice to others or have their approval, you should consider alternatives that make them better off as well as you. Put yourself in their shoes, and ask yourself how they might choose. Create alternatives where you both win.
- consider information-gathering alternatives
- You may be dealing with important uncertainties. You might want to gather more information that would reduce this uncertainty. Another possibility is a choice that allows you to defer a complete decision that may allow you to gather useful information. Remember, though, that deferring a decision may mean other alternatives may disappear, and thus, have a price. (Consider how to deal with uncertainties and whether to gather more information with Chapters 7-9.)
For the most part you want to generate as many alternatives as possible to evaluate later. Be careful not to eliminate seemingly bad alternatives if they might be transformed into good ones. Even alternatives with a fatal flaw might be modified to eliminate the flaw and yield an exceptional candidate. Also, throughout the decision-making process keep looking for alternatives.
At the same time, some balance is needed here in considering new alternatives. You don’t want to spend time, money, and energy if there is no potential payoff. The model of your decision does not need to include all possible alternatives that may be taken—just all the possible best ones. Is the time, effort, and energy of this decision-making process with it? How much value and how much cost is likely to come from the extra effort?
Choose best among the feasibleEdit
Now that you’ve considered your decision problem and the alternative choices available to you, what do you really want to achieve? What are your goals? You need to know these goals to be able to evaluate your various alternatives.
a procedure for uncovering your objectives
write down all your concerns that you hope to address through your decision convert your concerns into succinct objectives; use short phrases with verb and object (e.g. maximize profit, raise chance of re-election, reduce pollution) separate ends from means to establish your fundamental objectives; these fundamental objectives will be used to evaluate the alternatives clarify what you mean with each objective test your objectives against several alternatives; are they consistent with your underlying feelings?
understand the consequences of each alternativeEdit
From the choice of any alternative will come consequences, first the physical consequences, and then the consequences in terms of all of the objectives of your decision problem.
- Describe the physical consequences with appropriate accuracy, completeness, and precision
- Describe the consequences in terms of each of the objectives with a consequences table
- as much as possible, experience the consequences — try before you buy
- use common scales as much as possible
- use scales with the appropriate level of precision
- don’t rely exclusively on hard data; subjective assessments are important too
- make the most of available information
- use experts wisely; understand their assumptions as well as their conclusions
- address uncertainty head on
consider the tradeoffs between different objectives of each of the alternativesEdit
|Goals||Job A||Job B||Job C||Job D||Job E|
|(4) computer||(1) manage people,
|(5) organization||(2) time management,
|health, dental||health||health, retirement||health, dental|
eliminate dominated alternativesEdit
With two alternatives, if the first is better in some objectives and no worse in any, the second alternative cannot be the best choice, and thus, can be eliminated. In the job example below, Job B is better than Job E with four objectives and equal with two. Therefore, eliminate Job E.
make tradeoffs clearer by subtracting equalsEdit
In any comparison, since we are looking for differences to determine which is best, we can eliminate anything of equal value from all columns. We change the decision problem by substituting components of equal value, and then focus on differences, ignoring all parts of the outcomes that are of equal value. In the job example, we may eliminate $1800 of salary, health benefits, and 10 vacation days from all columns.
Try to subtract the most easily identified equals first. When evaluating the differences, is there a common scale for some of the objectives? When the difference in the number of vacation days is small, one additional vacation day may be worth $100 to Vincent. This would then mean that the monthly salary and vacation days rows can be combined by adding the salary and the value of the extra vacation days. With this change, Job A dominates Job D, and Job D can be eliminated.
Similarly, Vincent can hire someone to be with his ill father for those times that he could not be with him for $100 per month to make up the difference between low and high work schedule flexibility. After swapping $100 plus low flexibility for Job B with high flexibility for Job C, we find Job B now dominates Job C, and Job C can be eliminated.
Now all that remains of Vincent’s decision problem is to choose between Job A and Job B. Vincent needs to decide if the higher enjoyment, retirement benefits, and extra work schedule flexibility is worth more or less than an extra $400 in salary and better skill development. Vincent determines that the retirement benefits and extra work schedule flexibility are worth $200, so he must compare whether the remainder, the higher enjoyment with A is worth more or less than an extra $200 in salary and better skill development.
Often subtracting the most easily identified equals will allow enough alternatives to be eliminated that difficult tradeoffs may not be necessary to establish. Sometimes not, though. The tradeoffs may remain difficult. Unfortunately, no matter what choice you make, you are implicitly placing a value on these tradeoffs. You might as well confront the choices directly. Following the procedures above allows you to simplify your decision-making problem as much as possible to allow you to better understand the tradeoffs that you must face in your decision.
Modeling Behavior as Rational DecisionsEdit
Now we want to shift our focus. Instead of examining how some person should make a decision, we want to examine what decisions people actually make. We are most interested in predicting and explaining decisions that we see made in some social behavior of interest. To do this, we want to construct a model that incorporates the most important features of some social behavior of interest that would help us predict and explain what we see. In doing this, we are certainly willing to sacrifice a complete description of some people, their environment, or the resulting behavior, if we can describe in as simple a way as possible what will happen and why.
Create models that predict and explain behaviorEdit
One must always remember that a model is not intended to describe the fullness of actual behavior. It is only a stripped down version of the real thing, but one designed to determine how important factors influence observed behavior. The modeler consciously includes only those factors that he feels are most important. Especially at the beginning of an investigation, he keeps the model extremely simple. Once the effect of those few factors of an early model are understood, then he includes another to determine its effect, then another, and then another. To test the resulting logic, he adds and subtracts various factors, and applies his new models to various environments. And he compares his predicted results to actual choices made in similar environments. But even at the end, factors that at times may have some influence over some behavior may be left out. For the questions asked, these factors may add nothing of importance in predictive power and, in their inclusion, add such difficulty to obscure the essence of an appropriate explanation. These factors should not ever be included; the simplifying assumptions that omit these factors are to be encouraged. Note that a model that predicts and explains best always remains a simplified version of the world; it is not intended to be a description of the real thing.
Keeping a model simple not only improves the model's ability to explain, it also reduces the probability that mere chance explains the success of the model. A more parsimonious model, one that is more concise and simple, is more likely to be correct. We seek models that have parsimony.
A model with power, one that explains a large number of phenomena, also reduces the probability that mere chance explains the success of the model. A model with more power is more likely to be correct. We seek models that have more power.
At the same time, these models are not used within a vacuum. Each discipline shares common views and assumptions that they find most useful for making progress in their discipline. Some of the disciplines that study social behavior share paradigms based upon models of individuals furthering their own self interest. Some do not.
An objective of this course is to demonstrate that, compared to other views that could be used in each discipline that studies social behavior, models of individuals furthering their own self interest appear to predict observed behavior reasonably well in many contexts, they seem to explain why this behavior is occurring relatively simply and with great power. These models have sufficiently wide applicability that they may offer a common paradigm for the study of all social behavior. For especially those disciplines where practioners do not now generally share this view, much work is needed to test this view against others and to reassess which views are likely to be the most useful for making progress in these disciplines. For these disciplines, we may be in the midst of paradigmatic revolution.
All models are wrong, but some are useful if . . .Edit
As would be expected if texts are needed to teach people how to make rational decisions, many actual decisions are made that do not meet the standards of a rational choice model. “Virtually no one believes that anything approximating such a procedure is observed in any individual or organization . . . for any number of . . . decision tasks that confront them. . . . Pure rationality strains credulity as a description of how decisions actually happen.” (March, 1994, 5)
In many actual decisions, we find (March, 1994, 8-9):
- Instead of considering all alternatives simultaneously, some decision makers appear to consider only a few of these alternatives and then look at them sequentially.
- Some decision makers do not consider all of the consequences of their alternatives: they focus on some and ignore others.
- Relevant information about consequences is not sought, and available information is often not used.
- Some decision makers implicitly use preferences that are incomplete and inconsistent.
- Some decision makers don’t choose their best alternative possible, but choose a choice that is good enough.
To make models more consistent with these concerns, modern decision theory sometimes modifies the rational choice model by changing the knowledge of the decision makers, the number of decision makers, loosening the restrictions on their preferences, or changing the decision rules that they use. Most importantly, they allow limited rationality. With this limited rationality, decision makers intend to be rational, but they are constrained by limited cognitive capabilities and incomplete information, and thus their actions may be less than completely rational, at least when compared to those that would be selected with costless decision making, in spite of their best intentions and efforts. The internal costs of making a decision lead them to decide in ways inconsistent with the rational choice model that does not include these internal costs.
Are these differences important for our purposes? Sometimes, sometimes not. For three reasons (or for combinations of these three reasons), the rational choice model can still remain appropriate.
In some decisions, the rational choice model may model the decision making process quite well. When a great deal is at stake, when the decision making environment is one in which the decision maker has a great deal of experience, and when the decision making environment is relatively simple, decisions may be made rationally. For these situations, certainly, the rational choice model is perfectly appropriate to predict and to explain observed behavior. Determining the situations to which this model applies then becomes a critical question.
In some other decisions, even when a decision maker does not decide as the rational choice model describes, she may act as if she does. For example, as a driver of an automobile, she may not know or understand all of the mechanics and physical forces at work in her car. Nevertheless, she has learned that when she moves her hands and feet in certain ways in certain situations, the car reacts in certain ways. With experience, she still may choose actions that maximize consistent preferences as if she understood all of this knowledge. The rational choice model would be perfectly appropriate to predict behavior in such situations. At the same time, it may be less valuable at explaining what is happening. The appropriateness of this model would then depend upon why we want to use it.
In other decisions, the rational choice model may not even be able to predict individual decisions very well, but aggregate behavior or average behavior may be quite accurate. For example, for some economic good, the quantities demanded by some individual consumers in this market may not always match the predictions from the rational choice model, but the aggregate over all consumers, the market demand, may be quite accurate. Some consumers may choose a bit more than the model predicts, while others choose a bit less, so that the aggregate is quite accurate. If our concern is this aggregate or average behavior, the rational choice model is useful. A more complete, limited rationality model may allow the observed “noise” around the rational-choice prediction, but if introducing this limited rationality doesn’t introduce any bias in the central tendency, the greater simplicity of the rational choice model may be valuable itself.
Remember that formal models have two purposes: We want to use them both to predict and to understand behavior under specified circumstances. These two purposes also have conflicting needs. Prediction usually improves if more and more factors that have any impact at all are included within the model. Understanding usually improves if fewer factors are included within the model. The goal then is to include only those factors that have an important effect without significantly changing the reasons behind the prediction. This means we always want to include any assumption that omits a factor that does not change the prediction of the model, a simplifying assumption, but for other assumptions, how "important" the effect has to be and how "significant" a change in the reasons behind the prediction for some factor to be included within the model depends upon the questions to be asked of it.
In any event, we will return later to consider how observed behavior sometimes does not match up with the predictions from our rational choice models. We reconsider situations where rational choice models seem more likely to apply, and how this might alter the models that we might want to use and when we might want to use them. Also, since we will be examining social behavior in many different contexts, we will look for similarities across many different types of social behavior. As a first attempt, we will assume that when reasonable people who have the same goals are put into similar situations, they tend to make similar choices.
- Hammond-Keeney-Raiffa (1999), Smart Choices, Chapters 1-6
- Kuhn (1962), Scientific Revolutions, 10-34
- Olson (1982), The Rise and Decline of Nations, 12-16
- Shepsle-Bonchek (1997), Analyzing Politics, Chapters 1-2
- Mankiw (2004), Principles of Microeconomics, 19-23
- Belsky-Gilovich (1999), Why Smart People Make Big Money Mistakes
- Skinner (1999), Introduction to Decision Analysis, Chapters 1-5, 7
- Luce-Raiffa (1957), Games and Decisions, 12-19
- March (1994), Primer on Decision Making, Chapters 1, 2, and 5