Table of Contents
List of Abbreviations
2 Overview on Decision Making
2.1 Introduction to Decision Making
2.2 Common Biases in Decision Making
2.3 Mental Processes in Decision Making
3 Implications of Aging
3.1 Age-related Effects on Deliberative Processes
3.2 Age-related Effects on Experiential Processes
3.3 Development of Interaction between Deliberative and Experiential Processes
4 Age-related Differences in Decision Making
4.1 Negativity Bias
4.2 Framing Bias
4.3 Sunk Cost Fallacy
4.4 Priming Effect
4.5 Overconfidence Bias
4.6 Non-Bias Related Differences
5 Implications of Age Differences in Decision Making
List of Abbreviations
Abbildung in dieser Leseprobe nicht enthalten
The world population gets older as life expectancy increases. The population of people over the age of 60 is expected to be over two billion in 2050 (cf. United Nations, 2017, pp. 13-14). Older adults (OA) still face decisions, which might be even more significant or complex than decisions of younger adolescents. For example, the average age of C-suite members in American firms is 54 (cf. Korn Ferry, 2007). On a more individual level, the population aged 55 and higher has a significantly higher mean net worth than the younger population (cf. Bricker et al., 2017, p. 13). Especially, if one accounts for the fact that research has shown that declines in cognitive functioning are seen before the age of 50 (cf. Salthouse, 2004, p. 141), research on aging and decision making (DM) is of high importance.
In recent years numerous papers on this topic were published. This thesis helps existing literature by creating a literature overview to summarize current findings. Because of the reasons above, this thesis concentrates on age-related differences between young and OA, but does not analyze research about DM in childhood. It adds to current literature by being the first work with an extensive overview of age-related changes in the respective analyzed biases with a short subsequent view on financial DM.
The thesis deals with the research question, whether biases of DM literature differ between younger and OA. To answer this question, I examine differences in typical DM biases. The chosen biases are the ones being investigated the most thoroughly in existing literature on aging and DM and wherever possible or necessary I will focus on DM in economic contexts.
The remainder of this paper is organized as follows: section two gives a short overview about the foundations of DM. After examining age-related effects on processes, which underly DM, in chapter three, the fourth chapter deals with age-related differences in DM. Afterwards, I shine a light on potential implications resulting from respective age differences. Chapter six concludes.
2 Overview on Decision Making
2.1 Introduction to Decision Making
In the early years of research, the DM process is described as rational. Hence, literature strongly focuses on rational DM theories.
A rational decision is a decision, which tries to maximize the target, which is usually money or utility (cf. von Neumann & Morgenstern, 1953, pp. 8-9). Since I examine DM in uncertain environments which contain risk, a rational decision maker maximizes the “expected value” or, respectively, the “expected utility” (Edwards, 1954, p. 391). Hence, a good decision maximizes the probability of achieving the desired outcome (cf. p. 396).
In order to be able to maximize utility, it has to be a quantitative measure (cf. von Neumann & Morgenstern, 1953, p. 16). This is true, if two axioms are fulfilled. On the one hand, the preferences of any rational decision maker have to be complete, which means that the individual has a clear preference between two alternatives (cf. pp. 26-27). On the other hand, the “transitivity” (p. 27) axiom has to hold, which says that, if option A is preferred over option B and option B is perceived better as option C, it holds that the decision maker also prefers A over C (cf. p. 26). Furthermore, rational DM is based on other assumptions such as “complete information” (p. 30).
The rational DM theory is further developed and expanded. For example, based on actual risk preferences of decision makers, Kahneman & Tversky (1979) develop the prospect theory (cf. p. 263). The theory captures the difference in risk preferences in gain and loss domains via a value function. While individuals are risk prone in gains, they act risk averse in loss domains (cf. pp. 278-280)1. The value function can also explain the loss aversion. Loss aversion reflects the perception that losing an amount of money is more utility-diminishing than winning the same amount (cf. Kahneman & Tversky, 1984, p. 342). In short, that means that “losses loom larger than gains” (p. 346).
Simon (1955) introduces bounded rationality as an important assumption to be added to rational DM, which brings the theory of von Neumann & Morgenstern closer to actual DM (cf. Simon, 1955, p. 114). Bounded rationality means, that individuals need to simplify aspects of the DM process, e.g. the payoff function or information gathering (cf. pp. 104-108), in order to be able to process complex situations (cf. p. 104). For example, following Simon’s theory, people tend to look only for a good option instead of the best one (cf. p. 108).
One way to reduce complexity is by applying heuristics (cf. Tversky & Kahneman, 1974, p. 1124). Heuristics are a way of substituting complex target attributes with another related attribute, which is more readily available (cf. Kahneman & Frederick, 2002, p. 53). In a nuthsell, this means that individuals tend to use information readily available, which seems at least partly fitting to answer complex questions, instead of thinking hard for a perfect answer. This is in line with the theory of Payne et al. (1993), who see strategy selection as a trade-off between accuracy and effort (cf. p. 73).
All in all, heuristics are a method of DM, which neglects a part of the available information in order to make decisions with less effort, quicker or in some cases even more precise than normative strategies (cf. Gigerenzer & Gaissmaier, 2011, p. 454). Due to the trade-off of accuracy and effort, heuristics are definitively useful for DM, but also serve as source for systematic biases (cf. Tversky & Kahneman, 1974, p. 1124).
2.2 Common Biases in Decision Making
The aim of this section is to introduce a selective choice of the mentioned systematic biases in DM. The choice is based on the biases examined in the fourth chapter dealing with age-related differences in the respective biases. The order in which I introduce those typical errors is random and of no specific association with their importance. It is mirrored in the fourth chapter. The biases will be only briefly defined in this section, as an extensive view including empirical works would be beyond the scope of this master thesis.
Differences in influence of negative and positive stimuli on various domains have been analyzed thoroughly. In 2001, Rozin & Royzman publish a literature overview trying to study, whether a general bias can be assumed due to the similar findings in individual domains (cf. p. 297). In fact, they find support for four different types of negativity biases, from which I will focus on two in perspective of the relevance for chapter four. Rozin & Royzman contrast “negative potency” (p. 298) from “negative dominance” (p. 298). According to negative potency negative events are more salient than positive events of the same objective magnitude (cf. p. 298). This effect can also be observed in the different valuation of gains and losses in prospect theory (cf. Kahneman & Tversky, 1979, p. 297). Negative dominance means, that a combination of events with subjectively equal magnitude in opposite directions will be negative (cf. Rozin & Royzman, 2001, p. 199). It is based on the work of Rokeach & Rothman (1965, cf. p. 130).
A negativity bias can be found in various domains. Among others, one can see the effect in physiological activity (e.g. Suls & Mullen, 1981, pp. 32-33 and Gormly, 1974, p. 662), attention in information processing (e.g. Fiske, 1980, p. 903 and Pratto & John, 1991, p. 388) and DM, e.g. via loss aversion in prospect theory (Kahneman & Tversky, 1979, p. 279).
Rozin & Royzmans’ (2001) results are in line with complimentary research (cf. p. 298) from Baumeister et al. (2001), who also finds a stronger effect of bad than good information across a variety of phenomena (cf. p. 354).
Tversky & Kahneman (1981) introduce the well-known Asian disease problem in which people should decide on two different methods to combat the disease. The problem and the emerging options are hereby described in two ways. On the one hand, the first option will certainly save one third of the infected population while on the other hand, it will certainly kill two thirds of the population. The second option gives the decision maker a probability of one third to save everyone, respective a probability of two thirds that everyone dies. The majority of people will choose the first “safe” option when it is positively framed (save one third) and the second “risky” option if the first option is negatively framed (kill two thirds). From a rational standpoint, the choices should not differ due to the change in the description. However, the framing bias comes into effect, leading the decision maker to different, inconsequent choices for the same problem. It is defined at the decision maker’s perception of choices and their outcomes and acts (cf. p. 453).
In 1954, Edwards argues that a rational decision maker maximizes their targeted outcome with the alternatives available to them (cf. p. 382). Hence, historical foregone investments should not be considered as they do not play any role in the current state. However, Thaler (1980) suggests that a sunk cost effect can be seen in the positive relation of usage rates of services or products and the respective payments respondents made for them (cf. p. 47). Arkes & Blumer (1985) find empirical support for the sunk cost hypothesis in a series of ten different experiments (cf. pp. 126-136).
The sunk cost fallacy (SCF) can be partially explained by loss aversion. Thaler (1980) states that people feel the pain of the loss of their former investment (cf. p. 49) which they would like to avoid. Arkes & Blumer (1985) stress the effect of the reference point dependence. They suggest, that people in sunk cost situations are in the loss area, in which potential gains weigh more than the probable small losses one would face when deciding to continue the (unprofitable) investment due to the sunk costs (cf. p. 131).
Another way how the perception of information can influence subsequent judgements is called the priming bias. Firstly mentioned in 1932, Bartlett states a connection between the individual’s perception and categorized information stored in memory (e.g. comparisons, judgements and images) depending on its presence (cf. p. 196). Since perception, recognition and recall are components to the same psychological process, priming can influence judgement in the short- and long-term (cf. p. 187).
Bruner (1957) further investigates the mentioned necessary presence of information by introducing the concept of accessibility (cf. pp. 129-130), arguing that a higher accessibility will lead to lower input needed for a categorization of the information, more characteristics seen as suitable for judgement in question as well as the ignorance of better information. These effects lead several researchers to examine the priming bias. For instance, Higgins et al. (1977) find support for an influence of prior activated information on subsequent evaluations (cf. pp. 150-151), when exposing respondents to unrelated personality traits before they have to evaluate the character of a person based on a short essay (cf. pp. 144-145).
The last bias described in this section is widely accepted across research. Overconfidence can have two meanings in literature (cf. Glaser & Weber, 2010, p. 242). The first states a miscalibration of probabilities while the second meaning is called “better-than-average effect” (p. 242). I will focus on the miscalibration effect, as it will be more important for the literature review on age differences in chapter four.
In 1981, Lichtenstein et al. updated their work from 1976 giving insights to the calibration of probabilities. They analyzed subjective probabilities and with such a respective (mis-)calibration of probabilities. A perfect calibration would be if the amount of actual correct answers to a series of questions fit the assumed amount (cf. p. 307). Calibration curves are defined as a plot of the amount of correct answers on the respective confidence ratings in percent (cf. pp. 307-308). Overconfidence can be seen by the extent of the actual calibration lying below the perfect calibration curve (cf. p. 308). In other words, people being sure that 80% of their judgements are correct, while actually only 50% are right, are overconfident.
The overconfidence bias can be seen across various tasks and settings. For example, Nickerson & McGoldrick (1965) find support for overconfidence (cf. pp. 314-315) in a multiple-choice questionnaire (cf. p. 312). Furthermore, the results of Fischhoff et al.’s (1977) provide evidence for overconfidence across different questions and measurements (cf. pp. 553-559).
2.3 Mental Processes in Decision Making
Over the past 120 years, broad research has examined the underlying processes of DM. Since elaborating on the whole history of research on mental processing would go beyond the scope of this master thesis, I will concentrate on the pieces of literature frequently referred to in research examining age differences in mental processing and the corresponding effect on DM.
The majority of the referred research assumes a dual process (e.g. cf. Peters et al., 2007a, pp. 4ff. and Strough et al., 2018, pp. 8-9). Hence, despite the existence of single continuum theories, e.g. the expected utility theorem (cf. Sanfey et al., 2006, p. 111), and neurological support for a multi-system process (cf. p. 114), I solely focus on literature dealing with dual processing.
Additionally, as mentioned by Epstein (1994), most of the work on dual process models finds converging results. The models differ mainly from the specific angle used or the varying naming of the respective processes (cf. p. 714).
The first mentioning of a dual process of thinking can be found in Freud’s famous work on the interpretation of dreams (1900). Freud specifies the first system as unconscious and responsible for the reception of stimuli of perception but lacking any form of memory. A second conscious system transforms the receptions of the first system into lasting memories. The systems are interconnected and influence each other (cf. Freud, 2010, pp. 540-543). Freud distinguishes the first system as “primary process” and the second system as “secondary process” (p. 597). Among other tasks the secondary process is responsible for correcting potential misbeliefs of the first system (cf. p. 597).
Epstein’s cognitive-experiential self-theory (1973, 1994) builds upon Freud’s work and introduces the idea of an experiential, emotion-driven system (cf. p. 715). Epstein’s theory assumes two main systems as well – a rational one and an experiential one. The rational system is similar to Freud’s secondary process while the experiential system is the successor of Freud’s primary process. Similar to Freud (cf. Freud, 2010, p. 599), Epstein assumes an excessive evolution of the experiential system. This evolutionary process can be applied to age-related differences, especially if one takes into account the rather brief evolution of the rational system (cf. Epstein, 1994, p. 715)2.
The experiential system is complex. In its basic functions it “automatically, rapidly, effortlessly, and efficiently processes information” (p. 715), whereas it can also serve as source for creativity and wisdom. On the contrary, the rational system is “deliberative, effortful, abstract” (p. 715). While it may be inefficient in various situations of everyday life, it can delay gratification and think highly abstract. (cf. p. 715)
Epstein’s Theory assumes that all behavior is the result of the interaction of the two systems mentioned above, e.g. evidence for the influence of the rather unconscious experiential system on the rational system can be found in various studies about the priming effect (cf. section 3.3) (cf. p. 715-716).
Epstein argues, that the influence of the experiential system is unaware of the individual causing a lack of potential control by the rational system (cf. p. 716). He goes even one step further, declaring the experiential system as the “default option” (p. 716) in most situations, as it is associated with less effort and a higher efficiency compared to the rational system.
Consistent with the view of the dual processes as two interacting systems, Sloman (1996, cf. p. 6) argues for them to have “complementary functions” (p. 18). Sloman further elaborates that the systems can be seen as experts working together - despite the possibility that the systems have contrasting goals (cf. p. 6). He finds evidence for the dual processing hypothesis in instances in which people believe in contradicting responses at the same time (cf. p. 11), which is also in line with Stanovich & West (2000, cf. p. 659).
While Sloman (1996) explores the interaction of both systems on a more general level, Kahneman (2003) describes accessibility as a specific factor regulating the collaboration. He defines accessibility as the effort necessary for mental contents to become available to use (cf. p. 699). Consistent with the view on bounded rationality from Simon (1955) examined in section 2.1, Kahneman states that the overall mental capacity is finite (cf. p. 698). In line with Stanovich et al. (2000, cf. p. 659) and Epstein (1994, cf. p. 711), Kahneman (2003) defines the intuitive system as effortless and the rational system as effortful (cf. p. 699), which is in line with the work of Ennis et al. (2013, cf. p. 498). According to Kahneman’s earlier work (1973), the demanding processes of the rational system can interfere with each other, while the effortless processes of the intuitive system are unrestricted (cf. pp. 201-202).
These interferences can be seen as a reason for the affect heuristic mentioned in Slovic, et al. (2002). The affect heuristic describes the tendency of people to rely on the experiential system and affective information (cf. p. 332). Objects, memories and events are tagged with affect. During the decision process this affect – positive or negative – is considered (Slovic et al., 2007, cf. p. 1335-1336). If cognitive demanding processes might interfere with each other, it is better to deal with as many problems as possible without the usage of effortful processes. This can be easier and more efficient than deliberatively thinking about the problem and retrieving information from memory (cf. p. 1336). The affective heuristic, which is also supported by Stanovich et al. (2000, cf. p. 659), is thus another proof for dual process theories.
Slovic et al.’s theory is supported by the “risk-as-feelings hypothesis” of Loewenstein et al. (2001, p. 271). As the rise of feelings does not need any cognitive support (cf. p. 272), Loewenstein et al. argue that behavior is, at least partly, determined by feelings (cf. pp. 272-273).
Additionally, Epstein (1994) mentions the role of affect in the experiential system. An emotionally meaningful situation activates the experiential system to search for memories related to the current situation. That is how the processed memories influence the behavior of the individual (cf. p. 716). Affect and the experiential system also play a significant role in age-related differences in DM (cf. section 4).
Reyna (2004) goes even one step further and emphasizes the role of experiences and memory in DM in his “fuzzy-trace theory” (p. 60). He states that intuition and hence system one is at the apogee of development. In line with the literature viewed above, Reyna assumes that people can use two different processes for DM but rather rely on “fuzzy traces of experience” (p. 61). However, in contrast to the affect heuristic, Reyna puts less importance on the role of emotions. Emotions are still important but are not seen as purposeful and infallible signal as one might think from other research (cf. p. 61).
Coming from another angle which is based on the measurement of intelligence, Cattel (1943) brings up his theory of fluid and crystallized intelligence which add up to mental capacity. Fluid intelligence can be seen as equivalent to the rational system mentioned by Epstein (1994), while crystallized intelligence resembles the experiential system of Epstein (cf. Cattell, 1943, p. 178). Similar to the ideas of Epstein, fluid and crystallized intelligence work closely together, and thus are hard to separate (cf. Cattell, 1963, pp. 2-5).
On the one hand, similar to the rational system of Epstein, fluid ability, in its basic form mentioned, in Cattell has the task to adapt to new situations in which crystallized ability has no advantages (cf. p. 3). It is only “physiologically determined” (p. 4). On the other hand, crystallized ability reflects skilled patterns emerging from learning applications, or in other words experience (cf. p. 3). In contrast to fluid ability, crystallized ability is formed by the environment and the experiences the individual has made in them (cf. p. 4).
Cattell also sheds a light on the development of both abilities, mentioning an early peak of the fluid ability, while crystallized ability may increase continuously with age. This results in an increasing gap between the two nodes (cf. p. 3). Despite the increasing gap, Cattell assumes crystallized and fluid ability to be strongly correlated as crystallized ability is a complex function consisting also of the factor for fluid ability among other factors (cf. pp. 4-5).
This correlation is specified by Horn & Cattell (1967), who argue that crystallized intelligence is partly related and dependent on fluid intelligence. Any individual needs at least basic fluid abilities to be able to learn from experiences. These experiences are one of the factors being largely independent from the fluid capacity (cf. pp. 111-112).
In summary, one can distinguish two types of research strings of dual processing in the light of age-related differences in DM. Both strings come to equal conclusions and can almost be substituted with each other. The first one is concentrating on neural processes based on the work of Freud (1900). The second string is derived by Cattel (1943, 1963) based on intelligence tests.
While system one is associative, automatic, effortless and associated with affect, system two is deliberate, analytic and associated with higher effort. Both systems interact with each other and can support each other. However, they can also come to contrasting responses (cf. Osman, 2004, p. 991).
To be consistent, I call system one from now on experiential system and system two deliberative system, which is in line with literature dealing with age-related differences in DM (e.g. cf. Strough et al., 2015b, p. 239, Peters & Bruine de Bruin, 2012, p. 114).
1 Risk aversion is displayed by a concave function and risk seeking by a convex function (cf. Kahneman and Tversky, 1984, p. 342).
2 More details on how the processes develop can be found in chapter three.