Normative and Descriptive Aspects of Decision Making

The problems of human behavior in decision processes are central in this chapter. The gaps between the requirements of decision methods and the possibilities of human information processing systems are analyzed. The qualitative model describing the decision maker’s behavior is proposed. The model defines the guidelines for the construction of decision methods justified from behavior point of view. 5.1 SPECIFIC FEATURES OF DECISION MAKING AS A BRANCH OF RESEARCH Decision Making, as a branch of research, has two main features distinguishing it from other research disciplines such as Economics and Operations Research: 1. For typical Operations Research problems, the information needed to solve a problem is given in the problem's statement. Contrary to the latter, the initial statement of any decision making problem has elements of uncertainty connected with a lack of information on general criterion of the solution's quality and/or the consequences of the decision's variants. This is why information from a Decision Maker (DM) and experts is required for the solution of a decision making problem. It is possible to say that a statement of any decision making problem includes a priori unknown preferences of the DM. This is why inside any decision making method there are some procedures of information elicitation from the DM (or a group of DM). 2. The primary step in many economical studies and in Operations Research is the construction of models representing the reality, small pieces of the real world having a mathematical description. In contrast to the latter, typical Decision Making problems imply the construction of a subjective model representing the personal perception of such a decision problem by the DM as the primary step. This subjective model reflects the DM's policy in the situation of a choice. The distinguishing features of Decision Making as a research branch make the DM the central figure of the decision making process. Therefore, behavioral aspects become the central features in this line of research. The goal of this chapter is to look at the Decision Making field of research from this point of view, to summarize the existing knowledge about human behavior in the decision processes, to analyze the existing gap between descriptive and normative approaches in decision making, and to draw some criteria for the construction of decision aiding tools and to demonstrate the importance of behavior aspects. 5.2 THE GAP BETWEEN DESCRIPTIVE AND PRESCRIPTIVE The source of the widely accepted gap between the requirements of decision methods to human beings and the possibilities of human information processing systems lies in the historical development of Decision Making as a research field. Decision Making has two roots: • economical utility theory; • operations research. NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING 5-3 5.2.1 Utilities and probabilities Each portion of a purchased commodity (e.g., bread or tea) has its consumer utility. The law of marginal utility reads as follows: the marginal utility decreases, that is, subsequent portions of a commodity are less valuable to the consumer than initially, which is quite understandable from our everyday experience. If there exists a need for several commodities, the consumer attempts to allocate money to support a constant ratio of the utility of a commodity to the general measurement unit (dollars, rubles, etc.). Stated differently, more money must be invested into commodities of higher utilities. The same human behavior is inherent to the problem of investments more money is put into more useful areas of activity. Economists believe this is the only correct behavior and refer to the person making such a choice as a rational person. It is assumed that rational persons are intrinsically consistent and that transitivity of choice is appropriate to them. Second, it is assumed that, when making a decision, a rational person maximizes his utility. To conclude, what does the rational person do? First of all, they list all possible decisions and their consequences for which utilities (or money values) are determined. For each variant of a decision, the probabilities of all its outcomes are determined (no matter how). Next, the expected utility of each variant is calculated by summing the products of utilities by corresponding probabilities. The best variant is that which has the maximum expected utility. J. von Neumann and O. Morgenstern laid the scientific foundation for the utility theory in their well-known "Theory of Games and Economic Behavior" [48]. The utility theory as presented in this book is axiomatic. The originators of the utility theory made use of so-called lotteries, where two results (outcomes) with respective probabilities exist, as simple problems of choice and demonstrated that if human preferences for simple problems (lotteries) satisfy some axioms, then human behavior can be regarded as maximizing expected utility. The axioms used by the authors of [48] assert, for example, that a person can compare all outcomes and he/she is transitive, due to possibility of determining the probabilities under which lotteries constructed on pairs of outcomes (out of three) are equivalent, etc. The axioms are required to infer the theorem of existence of the utility function for a person that agrees with the axioms. The internal utility function of the DM is used to measure the utility of any outcome. The theory presented in the classic book by J. von Neumann and O. Morgenstern needs a quantitative measurement of all utilities and probabilities. Von Neumann and Morgenstern’s theory assumes that probabilities are given as objectively known magnitudes. D.Savage [39] developed an axiomatic theory enabling one to measure simultaneously utility and subjective probability which gave rise to the model of Subjective Expected Utility (SEU) where the probability is defined as the degree of confidence in fulfillment of one or another event. Together with the development of utility theory and SEU some findings appeared concerning human behavior in the lotteries choice. Well known is the socalled Allais’ paradox that was the object of hot disputes for several years [37]. People repeatedly demonstrated contradictory numerical evaluations of utility in the 5-4 NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING tasks of choice. Inconsistent human behavior in lotteries choice was demonstrated [5]. Furthermore, it was demonstrated that people don't believe in Savage axioms [43]. The difficulties of checking axioms in the applications became evident. The entire research on Decision Making theory was greatly influenced by the psychological studies of P.Slovic, A.Tversky, B. Fischhoff [14], et al., who demonstrated the existence of human errors made when evaluating event probabilities. The main causes of these errors can be represented as follows [14]: • Judgement from representativeness: people judge about the membership of an object A to the class B only from its similarity to the typical representative of B disregarding the a priori probabilities. • Judgement from availability: people often evaluate the probabilities of events on the basis of their own meeting with such events. • Judgement from the anchoring point: if initial information is used as a reference point for determining probabilities, then it exerts significant influence on the result. • Superconfidence: people place too much confidence in their evaluations of event probabilities. • Tendency to eliminate risk: people try to eliminate risky situations as much as possible. These work may bring into question the possibility of practical application of utility theory and SEU theory. Clearly, it was the first demonstration of a gap between descriptive and normative. 5.2.2 Prospect theory Attempts were made to update utility theory so as to eliminate the most salient discrepancies between theory and real human behavior. The Theory of Prospects [4, 15] is the most conspicuous attempt of this kind. By prospect we mean a game with probabilistic outcomes. Prospect theory allows for three behavioral effects: • certainty effect, that is, the tendency to give greater weights to determinate outcomes, • reflection effect, that is, the tendency to change preferences upon passing from gains to losses, and • isolation effect, that is, the tendency to simplify choice by eliminating the common components of decision variants. All these effects being taken into consideration, the value of a lottery to gain outcomes x and y with respective probabilities p and q is defined by multiplying the utilities of the outcomes by the subjective importance of the probabilities of these outcomes. The function of the subjective importance of the probabilities has some specific features that allow one to avoid the Allais’ paradox and give some explanations to other disagreement between the theory and human behavior. The theory of prospects recommends to "edit" prospects before comparing them for example, to eliminate identical outcomes with identical probabilities, to merge in one prospects with identical outcome, and so forth. NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING 5-5 Despite the fact that the theory of prospects eliminates some paradoxes of choice stemming from the utility theory, it does not eliminate all problems and paradoxes appearing upon studying human behavior in the problems of choice. The possible paradoxes appear when editing the prospects. A solution of the problem depends very much on the way in which one “frames” it. Furthemore, the quite natural desire to round the probabilities and utilities leads to different results of the prospect comparison depending on the rounding [53]. The prospect theory, as well as the utility theory relies on an axiomatic basis. A common problem with all axiomatic theories is the validation of the axioms allowing one to use one or another form of the function of utility (value) of the theory. The common feature for the utility theory, SEU and the prospect theory is the same representation (model) of the decision problem: in the form of the holistic parameters of the utility (value) and probability. In the middle of the 70-ies a different model became more popular and promising: the multicriteria description of the positive and negative factors influencing the choice. The reason is that utilities and probabilities manifest itself in the multiple criteria of alternatives’ evaluation. 5.2.3 Multiattribute utility theory The next step in the evolution of the utility theory was marked by the transition to the multicriteria or Multiattribute Utility Theory “MAUT” [16]. The construction of a strict and harmonic mathematical theory of utility under multiple criteria was a great merit of R. Keeney and H. Raiffa. The theory is constructed axiomatically, where the general axioms of connectivity and transitivity on a set of alternatives, etc., are complemented by the axioms (conditions) of independence. There exist many conditions [12] which conceptually define the possibility of comparing alternatives in one criteria, while the estimates in other criteria are being fixed (at different levels). For example, the condition of the preference independence states that comparisons of alternatives in two criteria are valid if their estimates in other criteria are fixed at any level. If the conditions of such a kind are met for all pairs of criteria, then the existence of a utility function in different forms is proved. We note that the Multicriteria Utility Theory is directed to the problems where existence of many alternatives justifies great efforts that are required to construct a utility (value) function. After the development of “MAUT”, critical comments were made about possibilities of validating all axioms and conditions needed for the existence of a multicriteria utility function in one or other form. For example, the sum of the importance coefficients of the criteria is to be equal 1 for the existence of a utility function in the additive form [17]. The question is: if we take into account the possible small errors in the measurement, which value of the sum is close to 1? [51]. In the construction of one dimensional utility functions the lotteries were used as the preferences elicitation tool. But human behavior in a lottery choice is inconsistent [5]. Again there is an evidence of a gap between the requirements of the decision methods (normative) and the possibilities of the human information processing system (descriptive). 5-6 NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING 5.2.4 Multicriteria counterparts of the well-known problems of operations research The second root of Decision Making as a branch of research is Operations Research. The introduction of multiple quality criteria enables one to obtain multicriteria counterparts of the well-known problems of the operations research. For example, additional criteria are readily built into the generalized transportation problem [49] which can be formalized as a multicriteria problem of linear programming for which multitude of methods were developed [46]. Also, there are multicriteria assignment problems [22], multicriteria bin packing problems [24] which are counterparts of the well-known Operations Research problems. There exist a great deal of the man-computer procedures enabling DM to examine the domain of the admissible decisions and at the same time to establish a compromise between the criteria [46]. The man-computer procedure consists of alternating phases of analysis (performed by the DM) and optimization (performed by the computer). Each phase can consist of more than one step. Optimization phase (computer): • using the information received from the DM at the preceding step, a new decision is computed and auxiliary information for the DM is generated. Analysis phase (DM): • the presented decision (or decisions) is estimated and its admissibility is determined. If the answer is positive, then the procedure terminates; otherwise, auxiliary information is considered; • additional information is communicated to enable computation of a new decision. The man-computer procedures differ in content and execution of the above steps. Their efficiency depends mostly on the nature of the DM-computer interaction that is represented in terms of the quality and quantity of the information. Together with the development of many man-computer procedures, there appeared papers with the evaluation of such procedures from the behavioral point of view [18, 25]. The analysis demonstrated that many operations required from people in the framework of the man-computer procedures are difficult for the human information processing system. People show intransitivity in the process of choice, show suspiciously fast convergence to the solution and so on. Again we witness here the evidence of a gap between descriptive and normative. All the gaps mentioned above lead us to the question: what could be said about a human being as a DM? 5.3 THE QUALITATIVE MODEL OF THE HUMAN DECISION MAKER On the level of the existing knowledge it is possible to summarize the evidence about human behavior in the decision processes in the following way. 5.3.1 The features of the human information processing system A. Limited span of the working memory. According to cognitive psychology [45], human beings have a limited span NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING 5-7 of the short -term memory. In repeated tasks the span of the working memory could be enlarged [9] but it takes both time and efforts. That is why the DM cannot simultaneously pay attention to many factors (or evaluations of alternatives upon criteria) in the new decision tasks. As a matter of fact, for the new tasks DM has no possibility to create the internal structure of the necessary knowledge. This limitation manifests itself in such known facts as: The DM is trying to simplify the description of the decision situation by replacing some of the criteria by limitations, by eliminating some of the criteria, by grouping the alternatives and so on [25]. Such behavior is the unconscious desire to decrease the load on the short-term memory. Experienced DMs have usually the skill of simplifying the decision situation in the best possible way. For inexperienced DMs a significant increase in the number of contradictions for more complex decision tasks is typical [25] B. Limited exactness in quantitative measurements. According to the existing knowledge, a human being is not an exact measurement device producing quantitative measurements. The famous experiment of A.Tversky [47] demonstrated that people neglect small differences in the evaluations. It is the reason for the intransitive behavior in some problems of choice. Inability to take into account small differences in the evaluation leads to the elimination of the dominating alternatives by the conservation of the dominated ones [19]. The experiments demonstrated that people can poorly measure the probabilities in the quantitative way (see above). The change in the method of measurement, the transfer from the quantitative to the verbal probability allows one to decrease significantly the number of the preference reversals [13]. It was demonstrated in the experiments [28] that slightly different procedures of the quantitative measurements for the same variables give quite different results. C. Human errors and contradictions. It has been known since the time of antiquity that "To err is human". People err when processing information. There could be different reasons for such a behavior: weariness, lack of attention, habitual heuristics and so on. 5.3.2 The features of human behavior in the decision processes A. Absence of preconceived decision rules in new decision tasks. As many researchers supposed, the DM has no preconceived decision rules. As noted in [52] it can be hardly expected that the utilities and numbers expressing the subjective estimates of the objects and situations are just stored in our minds until elicited. To develop a decision rule the DM needs time and some learning procedures. Usually people use some kind of a "trial and error " approach in such procedures. B. Search of the dominance structure. At every step of the decision making procedures people pay attention to a 5-8 NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING limited number of objects. This is a possible explanation to the psychological theory of human behavior in the decision tasks-the search of the dominance structure [32]. According to the theory, in the case of the limited number of alternatives people make a preliminary selection of the potentially best alternative and compare it pairwise with other alternatives , trying to check the fact of dominance. In the case of a larger number of alternatives, people use initially the strategy of eliminating by aspects and after that utilize a more elaborated process (like the search of dominance) for a smaller number of alternatives. C. Minimization of human efforts. J. Payne et al. [35] suggested and substantiated another theory of human behavior upon choosing the best multicriteria alternative(s) that can be called the theory of the constructive processes. When comparing multicriteria alternatives, people can use various strategies. The studies of J. Payne et al. [35] have demonstrated that in the process of the decision making subjects often choose a strategy depending on the specific features of the alternatives under consideration (their evaluations by criteria). Here, the human preferences of the alternatives and criteria are very unstable. At the local stages of the comparison, rules (or their parts) can vary depending on the relation between the required human effort and the accuracy of choice. As J. Payne et al. notes, such a behavior is a characteristic of the untrained subjects. People experienced in the decision making, as well as regular decision makers have their preferable strategies for solving problems. 5.3.3 The features of human behavior in organizations A. Satisfactory decisions. The studies of economists and psychologists provided an insight into the human decision making in large organizations. Ch. Lindblom [31] notes the officers organizations try to make as small changes in the existing policy as possible to be able to adjust to the environmental changes. It is not only easier to work out such changes, but also to coordinate them within an organization. The sequence of changes is mostly the means for forming the current policy. Lindblom also believes that this way of solving problems is more realistic because it requires less effort and is more customary for the managers. On the other hand, this approach is more conservative and is not adjusted to dramatic changes in the policy. Similar discoveries were made by H.Simon [41] who introduced notion of satisfactory decisions as a counter to the optimal ones. In organizations, the life itself brings people to seek satisfactory decisions the environment is too complicated to be described by a model, the multiple criteria are defined incompletely, there are many active groups influencing the choice, etc. This natural behavior of the personnel resulted in the loss of the strategic objectives amid the petty, everyday routine. B. Taking the power in the hands. The desire to have the decision situation under control is typical for the behavior of a DM in organizations. It means that the DM is trying to control all stages of the decision making, all transformations of the information influencing a NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING 5-9 decision. Speaking differently, the DM is trying to have the power in the hands. In the case when it is necessary for him/her to take into account the interests of different active groups, the DM is looking for a mutually satisfying decision. [27] but he/she is always trying to implement the principal components of own policy. 5.4 HOW TO REDUCE THE GAP BETWEEN DESCRIPTIVE AND PRESCRIPTIVE The above features of human behavior define the numerous gaps between the requirements of the different normative methods and the possibilities of human beings to meet such requirements. The discrepancy manifests itself in human errors and contradictions badly influencing the results of an analysis, as well as in mistrust in the results of the DM to the results presented by an analyst and so on. There are several remedies to save the situation. First, it was the idea to improve the human performance in the process of choice: to teach people how to use the axiomatic methods or to train them to make the quantitative measurements. Unfortunately, we do not have any evident confirmation of success for this approach. Even more, it became clear that many features of human behavior could be explained by the basic organization of the human information processing system [42]. The second reaction to the gaps is the following: human behavior is not important factor in the decision processes. One could take evaluations in the qualitative form but transform them quickly in the qualitative form appropriate for many decision methods. On the final stage of the decision process it is possible to use the so-called sensitivity analysis to check the influence of the different factors on the output of an the analysis. Unfortunately, the task of sensitivity analysis is very complex. It is an independent difficult problem and only the skill of an analyst could shed light on the influence of the different factors. The approach we have taken [26, 30] differs from the others. From our point of view it is necessary to adapt the decision methods to human behavior. A possible way to close the gaps consists of taking the behavior finding concerning human behavior as constrains for the normative decision aiding methods [26]. By going on such a way it is possible to use the qualitative DM model described above as a base for the construction of the decision aiding tools and the decision support systems. In other words, on the basis of the behavior findings it is possible to formulate special requirements to the characteristics of the decision aiding methods [30]. 5.5 BEHAVIORAL REQUIREMENTS ON THE METHODS OF DECISION MAKING The knowledge about human behavior in the decision processes allows us to define the requirements for the methods for the decision making [26]. 5-10 NORMATIVE AND DESCRIPTIVE ASPECTS OF DECISION MAKING