… they sat with ruled papers in their hands, whereon they set down the coups, calculated the chances, reckoned, staked, and—lost exactly as we more simple mortals did who played without any reckoning at all.
The Gambler, Dostoyevsky
- 1 We should also mention John C. McKinsey (1908-1955), who could rightly be considered to be the pers (...)
1Following Guala (2008), the overall picture of experimental economics in the United States during the 1950s could be described as hinging around three major research centres: The first located at the University of Pennsylvania, Philadelphia, and led by psychologist Sidney Siegel (1916-1961) while the second, also run by a psychologist, Ward Edwards (1927-2005)—along with Amos Tversky (1937-1996), who would later take the head of the research program initiated by Edwards—at the University of Michigan. The third research centre, at Stanford, California, was run by Suppes (1922-2014) and Davidson (1917-2003).1
2Geographic distance between the centres should not lead us to conclude that the three projects were unconnected. The 1957 publication Decision Making, An Experimental Approach tells us about the way the three projects overlapped. For the book, the Stanford team (Suppes and Davidson) worked jointly with Sidney Siegel, taking into account most of Edwards’ comments presented in the experiments he conducted in 1953 and 1954. One of the comments, which Edwards coined “the wording effect” in 1954, is particularly relevant to the purpose of this paper.
3In Edwards’ mind, the wording effect corresponds to a test in which the experimenter changes the way options are formulated by reversing the proposition that describes both gains and losses. Broadly speaking, the idea is to check whether a reversal in preference results from the reversal in wording. In other words, does the wording influence the preferences expressed by subjects? Is it a parameter that needs to be taken into account? As we shall see, Edwards showed little interest in such an effect but cast light on the effects that could influence choice in experimental decision by paving the way of experimenters regarding a significant issue, namely: how subjects interpret outcomes or consequences. This posed a major challenge to the expected utility theory.
4Choosing this effect is not without purpose: this article aims at showing that the kind of treatment proposed for this singular “effect” implies a particular conception of the interactions between economics and psychology.
5Starting from the initial formulation of the wording effect proposed by Edwards (Section 1), this paper is going to examine the identification of a recurring issue in decision theory, the interpretation problem, through the prism of thought of two major authors, Amos Tversky and Donald Davidson (Section 2). This analysis will highlight how the expected utility theory is very much challenged. Moreover, the way the wording effect is approached by the two authors will shed light on the relationship between economics and psychology (Section 3). Indeed, discussing this effect raises the question of interpreting its outcomes and consequences as well as the words the experimenter has used. For Tversky, a solid interpretation should be added to the theory right from the start. The expected utility theory should be abandoned in favour of a more sophisticated theory using methods and results of psychology to gain access to a more reliable theory of human behaviour. For Davidson, interpretation should not be taken for granted. The latter is derived from choices since they are expressed verbally. One has access to meanings just as desires and beliefs derive from choices. A revised version of the canonical decision theory, allied to a theory of radical interpretation will lead us to a unified theory of thought, meaning and action. Implicitly, the role of psychology is confined to an instrumental use, without departing too much from the initial paradigm of economists.
6The focus on the wording effect is twofold: first, because one should consider the effect as a deviation from the “objective” model described by Edwards, which prescribes to “choose the bet with the larger positive expected value” (Edwards, 1953, 350); and secondly, because the underlying problem raised by this effect is that of interpretation, which decision theory must address. Depending on the answer to this problem, a specific connection between economics and psychology will be put forward.
7Without looking for the genealogy of the concept, one might agree that the first explicit reference to the wording effect can be found in Edwards’ work of the 1950s. This does not mean that previous experiments are of no interest. The protocol and the realism of the experiments performed are two aspects that best describe the concerns of experimenters. The criticisms formulated against these earlier experiments have opened the way to consider globally all the factors that might affect experimentation. Thus, they are presented here before analysing the way Tversky and Davidson took over the wording effect.
8In an attempt to understand the context in which Edwards introduced the wording effect, we will briefly describe three major steps in the early stages of experimental economics (1.1.) and then address the new challenge he proposed (1.2.).
It would in short be preferable at least in principle, to interrogate the person, not literally through his verbal answer to verbal questions, but rather in a figurative sense somewhat reminiscent of that in which a scientific experiment is sometimes spoken as an interrogation of nature. (Savage, 1954, 28)
9According to Roth (1995), three phases can illustrate the early stages of experimental economics.
10First, one might mention Thurstone’s experiments (1931), generally considered to be one of the first reports on experiments to collect data allowing the representation of choices made by individuals with indifference curves. The idea was to propose combinations of goods and then derive from the choices made the exchange rate required to deduce the indifference curves, according to the Pareto tradition (see Moscati, 2004). The well-known criticism of Wallis and Friedman (1942) focuses on the way experiments were conducted, denouncing their lack of realism.
11The second phase corresponds to the work of Preston and Baratta (1948) from the University of Pennsylvania. The two authors aimed at proving the existence of a “psychological scale of probabilities” through experimentation (1948, 185). More specifically, they were interested in all the “other considerations” that could influence the objective calculations of mathematical expectations. These “other considerations” included “unsupported convictions about the dependence of chance events” or “the peculiar value which often attaches to a given winning because it completes a larger whole” (Ibid., 183). The most striking consideration emphasised by these authors was what they called the “player’s notion of the meaning of a given probability” (I underline). This meaning is immediately derived from the behaviour observed in gambling: “Players for whom a probability of 0.01 means that they have a fair chance of winning are behaving as if a probability of 0.01 was a probability of 0.10 or more” (Ibid., 184). In other words, the observable behaviour reveals attitudes such as “beliefs” or “meanings” assigned to events.
- 2 See Davidson, Suppes and Siegel (1957, 22-23).
12The objection comes this time from Davidson, Suppes and Siegel (1957). Not only did Preston and Barrata not manage to measure utility by interval scales2 as Von Neumann and Morgenstern were requesting (1947, second edition), but they were also unable to avoid biases linked to participation in the game or not.
- 3 Mosteller and Nogee (1951, 399).
13The third relevant phase relates to the experiments conducted by Mosteller and Nogee (1951). In line with the analysis of Friedman and Savage (1948), the Harvard authors conceived a game session in which repetitive bets were proposed to subjects in order to derive a utility curve. The latter was then used to predict the future behaviour of these subjects. If the results obtained did not correspond to expectations3, there were good reasons to be optimistic in future about the predictive power of the expected utility theory.
14In the early stages of experimental economics, there were no debates on the way options were put into words. Edwards’s contribution to decision theory consists in determining factors influencing choices in gambling situations. His research focuses on every psychological factor influencing the gambler when he is playing, including the wording effect.
15Edwards could be considered to be the first experimental psychologist to describe in detail all the variables that could affect a subject’s choices, in one way or another. By showing how subjects systematically diverge from the objective model, Edwards opened the way to consider globally the psychological factors that influence choices.
- 4 See Edwards (1953).
- 5 It should be noted that Edwards is reluctant to use the word “utility”, generally used by the econo (...)
16The starting point of Edwards’ reasoning is the “objective model”4 where every experimental subject in a gambling situation should “always choose the bet with the larger positive expected value (EV)” (Edwards, 1953, 350).5 Edwards proposes specific experiments designed to test whether subjects are following this code of conduct. The results are undisputable—the experimental subjects did not meet the objective model requirements :
Neither in the experiment of Preston and Baratta nor in real life do people follow the objective model. Why not? Several possibilities exist: it may be wrong to assume that people try to maximize monetary returns (or anything else, for that matter) when they gamble. Or, it may be that people try to maximize returns but don’t know how. They may misinterpret the probabilities, the amounts of money involved, or both. Or, they may not know how to combine probabilities and amounts of money to determine the best bet. (Edwards, 1953, 350)
17To explain such deviations, Edwards suggested the assumption that “preferences among probabilities do determine choices in this experiment”. Moreover, Edwards raised the issue of the methodology used in the experiments. In 1954, he outlined the role of extraneous variables on the choices themselves. Among these “extraneous variables”, he mentioned: “the effects of wording, [and] previous rolls” (Edwards, 1954, 76). As for the impact of the effect of wording during experiments, he raised the following questions:
What are the effects of wording of the bets? Would similar results occur with a different number of bets? What effect does the outcome of previous bets have on a decision? (Edwards, 1954, 68)
18To test whether the wording of bets have an impact on results, Edwards offers to reverse the sequence in which gains and losses are presented. In the usual case, “the possibility of winning” (1954, 77) comes first, before “the possibility of loss”. By inverting this order (i.e. presenting first the possibility of loss), Edwards observed non-significant differences: “It is clear that the results of reversed wording are very similar to the results of the previous wording”. The conclusion is unambiguous: “so it is reasonable to conclude that the order of wording does not make much difference” (Ibid.). However, Edwards conceded that subjects
simply pay more attention to statements about what they can win than to statements about what they can lose, regardless of wording or location, at least so far as properties of wording and location are changed by this experiment. (Edwards, 1954, 77)
19Edwards put forward another explanation: subjects focused on the amount of money they could have earned without comparing it with what they could lose. Similarly, the amount of money mentioned at the beginning of the gamble influenced subjects, whether it referred to a gain or a loss. Here again, this hypothesis was rejected: “no evidence was found to destroy confidence in probability-preferences as the explanation for the results obtained” (Ibid., 81). Yet, does it mean the wording has no influence at all? As Edwards pointed out himself “Would similar results occur if the EV-levels chosen, and therefore the amounts of money involved in the bets, were different?” (Ibid., 68-69).
20This first mention of the wording effect leads to a negative result. Compared to “probability-preferences”, the impact of the wording effect does not appear to be a critical factor. From Edwards’s point of view, the wording corresponds to the order in which gain or loss is presented. “Wording” is only considered in a narrow sense. The “wording” of a sentence corresponds to the way it is phrased, not to the meaning of the sentence.
21In other words, according to Edwards, no interpretation problem arose from the reverse wording. Reversing the order of presentation of gain or loss in the sentence has no impact on choice itself. According to Edwards, this means that changing the wording does not result in a different interpretation.
22The sections below will highlight the extension of this effect proposed by Tversky and Davidson. The wording matters because the meaning and the outcome of each bet—notably in terms of monetary payoffs—influence the subject’s real choices.
23One of the interesting aspects of Edwards’s analysis is the increased emphasis on the “extraneous” variables that influence choice.
24In 1957, Davidson, Suppes, and Siegel addressed the “effects of cumulative and immediate reinforcement” and more generally on any distortion arising during the experiments. Edwards was mentioned twice. While there is no specific mention of the wording effect, an extensive analysis of all the elements that could limit the results of the experiments was proposed. The most significant progress achieved by the Stanford Team consists in “operational interpretation”, proposed to support the formal model (2.1.). Tversky (2.2.) went a step further: without a real interpretation, the decision theory is to be questioned.
25Keeping in mind Edwards’ hypothesis concerning a possible preference for specific probabilities, Davidson, Suppes and Siegel (1957) decided to use the same chance event to avoid any distortion: “Whether or not Edwards’ evidence can be conclusively interpreted this way [a preference for certain probabilities], no distortion of this kind can affect the results in the experiments reported here” (1957, 17). Indeed, the Stanford team first started using a coin, then a die and finally chose to use a “specifically made die” (Ibid., 51) where the usual numbers had been replaced by a “nonsense syllable”.
26The second reference to Edwards points out the scepticism of the Stanford authors regarding Edwards’ thesis on probability-preferences. They underline that he is not postulating that the utility is linear for money (Ibid., 25). More generally, the Stanford team expresses doubt on the meaning that could be drawn from the hypothesis formulated by Edwards concerning a possible preference for specific probabilities:
It is not clear whether this merely means that subjective probabilities are not equal to objective probabilities, or something more; nor is it obvious that the conclusion really can be drawn at all independently of utility considerations. (Davidson, Suppes, and Siegel, 1957, 25)
27While the wording effect is not specifically mentioned in Decision Making, the need to control possible distortions that may occur during the experiments is frequently put forward.
28Davidson, Suppes, and Siegel (1957) discovered that winning or losing several times in a row made subjects respectively optimistic or pessimistic. As a result, this had an impact on the subjects’ subsequent responses to similar offers. The increasing (or decreasing) amount of money at play also influenced choices. This means that one must identify these distortions and find a way of avoiding contamination of all the choices made (Ibid., 53).
29Another issue mentioned is the “recency effect”: even with the special die used, subjects got attached to specific nonsense syllables, for example “if the same syllable came up three times in succession” (Ibid., 54). To overcome the issue, the authors postponed the payoffs and used three dice instead of one and “the die in use was changed after each toss” (Ibid.).
30Beyond this analysis designed to control possible distortions during the experiments, what is striking is the special attention given to “problems of empirical interpretation” (Ibid., 1). The introductory chapter of Decision Making deals with this issue. The Stanford team introduces the idea of “operational interpretation” (Ibid., 4) to support the formal model. By operational interpretation, they mean an empirical interpretation specifically designed to test the theory. The underlying objective is to reconcile the normative conception of rational decision, which outlines norms of rationality in decision making, with the descriptive conception that attempts to describe as accurately as possible how someone rationally decides in real life:
A normative theory of rational decision which is to be of any practical use must, then, be capable of empirical application; and if it is capable of empirical application, it is possible that It is true as a descriptive theory. (Davidson, Suppes, and Siegel, 1957, 4)
31In the same vein, Davidson, Suppes, and Siegel provide what they call an “experimental hypothesis”, defined as the “logical consequences of the model and interpretation” (Ibid.).
32This reference to an “experimental hypothesis” is somewhat reminiscent of what Friedman and Savage called a “scientific hypothesis” in 1952:
the function of a scientific hypotheses is to enable us to ‘predict’ phenomena not yet observed, that is, to make statements about phenomena not yet observed that are (a) capable of being contradicted and (b) will not in fact be contradicted ... The wider the range of observable phenomena capable of contradicting the hypothesis, the greater its potential fruitfulness, because this is equivalent to greater precision of prediction. (Friedman and Savage, 1952, 465)
33The major step forward proposed by the Stanford team consists in submitting these hypotheses to experimental testing. In experimental situations, one should be able to tell whether or not theoretical statements are met, or in others words, what it means in practice to act in accordance with canons of rationality for example. In this context, interpreting means translating.
34At this stage, the authors are seeking a “behaviouristic” solution dwelling neither on the words used by the experimenter nor on the verbal answers from subjects. In section 3, we shall see how Davidson’s conception of interpretation evolved towards an interpretation of language itself.
35This need for an interpretation attached to decision is echoed by Tversky. Referring to the rephrasing trick used by Savage to justify his own failure with the famous Allais paradox, Tversky demonstrated how important it was to take the wording into account.
- 6 From a descriptive point of view, the decision theory “describe[s] and explain[s] how actual decisi (...)
- 7 Which means, another way of presenting the problem: decomposing outcomes by lottery tickets.
- 8 This kind of analysis was generalised by the end of the 1970’s through the “prospect theory” jointl (...)
36Tversky, Edwards’ PhD student, revived the idea of a wording effect during the 1970’s but made no reference to it. In 1970, he published an article in which he clearly stressed one of the major issues raised by the modern utility theory: the “psychological interpretation of axioms” (Tversky, 1970, 124). Seen from a descriptive point of view6, decision theory might be considered to be a branch of psychology. As such, one should expect to find attached to the theory, a detailed explanation of how to interpret all the variables that affect choices. As an argument to support his claim, he uses the example of Savage reformulating the famous “Allais paradox”. Adopting “another way of looking at the problem” (Savage, 1954, 102-103)7 after showing there was an inconsistency in his own preferences in Allais’ dilemma, Savage demonstrated how he reversed his preferences on gambles, correcting his previous “error” at the same time. Broadly speaking, according to Tversky, the wording of gambles themselves could be misleading for experimental subjects: “if gambles are displayed in terms of their immediate rather than final outcomes, the relationship between the compound and the simple gambles may very well escape the subject” (Tversky, 1970, 124).8 The lack of such psychological interpretation has an immediate impact on the “explanatory power” (Ibid., 124) of the theory.
37In 1975, Tversky was more incisive. According to him, the “standard” (1975, 163) interpretation incorporated into the utility theory is inappropriate. As a descriptive theory, all the conditions and axioms on which the theory is based can be disputed experimentally. As a normative theory, Savage’s example shows that the experimental results (observable data) are not compatible with the theory itself: from Savage’s point of view, the preferences expressed in Allais’ paradox are just erroneous in this particular case. The theory could be seen as a “guideline or a corrective tool” (1970, 128). But for Tversky, the problem is that “in the absence of any constraints, the consequences can always be interpreted so as to satisfy the axioms”. Instead of being conceived prior to the experiment, the interpretation of consequences might be adapted depending on the context: “Thus, the theory could be valid in one interpretation and invalid in another” (1975, 171). As such, both the predictive power and the descriptive dimension of the theory are dismissed.
38In the last section, we aim to show that the wording effect, in a broader sense, leads to an interpretation problem. Tversky and Davidson did not just criticise the lack of interpretation in decision theory, they provided their own solutions through two distinct theories.
39It is not just a matter of word order but rather a matter of implementing an interpretation theory to support the decision theory. Let us start with an example to illustrate the conceptions of Tversky and Davidson. Imagine a subject is proposed a game that consists in obtaining a certain amount of money if a specific card is drawn from a non-rigged pack of cards, and losing money otherwise. The instructions include sentences of the form “if you draw a club, you win α $, if you draw anything else, you lose β $”. Of course, such statements could be endlessly modified: “if you draw a club, you win α $, if you draw anything else, you win nothing” or “if you draw a club or a spade, you win ϕ $, if you draw anything else, you lose δ $”. The expected value of each bet could be positive, negative or equal to zero.
- 9 Tested with a zero expected value.
40From Edwards’ perspective, the wording effect9 consists in reversing the order in which gain or loss are proposed. In line with the example, this means placing the possible loss before the possible gain. While, for Edwards, this particular effect is only one among others, a singular examination is proposed by Tversky: an interpretation should be attached to the theory because the wording counts. By “interpretation”, Tversky (3.1.) means both an interpretation of the outcomes described in monetary payoffs and an interpretation of “all non-monetary considerations” (Tversky, 1975, 171). In this broader perspective, the experimenter needs to interpret, i.e. to find the cause of, all the deviations from the expected experimental results: how can we explain that subjects are systematically violating the axioms of utility theory? What kind of mechanisms are at stake when they are involved in experiments or in everyday life?
41Tversky offers two analyses: first, the subjects are using heuristics i.e. quick intuitive mental operations that influence their probabilistic inferences; second, and more generally, the prospects upon which the subjects make their choice distort the evaluation that precedes the choice itself.
42To illustrate Tversky’s conception in our example, we might say that subjects might overestimate the probability for a club to appear because they have not seen any club in the last ten minutes or because they have noticed two or three times that a club appears after a heart. In Tversky’s words, we would say that the similarity of situations pushes people to imagine some kind of regularity, or that the salience of a particular event increases with the ease “with which instances or occurrences can be brought to mind” (Kahneman and Tversky, 1974, 1127). In other words, practical examples of errors in probability assessments or in the prediction of values reveal the mental operations that influence probabilistic inferences. Keeping in mind such heuristics and integrating them into the overall framework helps getting a better view of the actual choices subjects will make.
- 10 Among the recent discussions of this principle, we could mention Bonnay and Cozic (2011).
- 11 That use of the charity principle is dictated by an epistemological requirement. See Davidson (1967 (...)
43For Donald Davidson, the need for integrating an interpretation in the theory is also emphasised but in a different way (3.2.): the interpretation should not be imposed by but merely derived from choices. In our example, we should not take the interpretation for granted when describing heuristics. We should start from scratch without anticipating the results. Subjects have their own interpretation of the outcomes. They attach specific meanings to words, not only to outcomes. As a first step, the experimenter has to understand these meanings: “To understand the speech of another, I must be able to think of the same things …” (Davidson, 1982, 105). The interpreter’s understanding involves making propositional attitudes (such as desire, hope, expectation...) intelligible to the person, by considering he would have made the same errors “if he had been in [the subject’s] shoes” (Davidson, 1986, 69). To ensure the success of the interpretation, Davidson uses the charity principle10 that enjoins us to maximize the agreement11: “Torn between the need to make sense of a speaker’s words and the need to make sense of the pattern of his beliefs, the best we can do is choose a theory of translation that maximises agreement” (Davidson, 1968, 101). The interpretation does not precede the theory but is a condition of it.
- 12 For a complete presentation of Davidson’s conception, see Engel (1994).
44Contrary to Tversky’s view, the decision theory should not become a branch of psychology but needs to extend its field to the interpretation of language because meanings are linked to both desires (utilities) and beliefs (probabilities). In the tradition of analytical philosophy, the analysis of language comes first but is also, according to Davidson, immediately related to beliefs and desires because of the “holism of the mental”12: one cannot understand the “meanings” without understanding the “desires” and the “beliefs” of subjects.
45While Tversky considers observed behaviour as a sign of mental states, endorsing in a sense a form of psychologism13, Davidson puts forward a radical interpretation that does not take meaning as granted. Rejecting any reductionism, what he proposes is the opposite of Tversky’s proposition.
46Davidson and Tversky agree on this point: the theory of expected utility is invalid whatever the dimension. Both of them have conceived an alternative that implies an interpretation theory but their theories are based on two different conceptions of psychology.
47Here is the final conclusion drawn by Tversky in 1975:
The theory of expected utility is formulated in terms of an abstract set of consequences that are the carriers of utilities. The axiomatic theory, by its very nature, leaves the consequences without interpretation. Any application of the theory, of course, is based on a particular interpretation of the outcomes. Thus, the theory could be valid in one interpretation and invalid in another. The appropriateness of the interpretation, however, cannot be evaluated within the theory. How then should the consequences be interpreted in any particular application? (Tversky, 1975, 171)
48From that moment on, Tversky searched for alternatives to the expected utility theory, “alternatives” in the sense of proposing a concurrent theory that could explain the deviations from the objective model. Ultimately, Tversky felt that the expected utility theory had to be replaced by this new theory.
49The first draft of this project dealt with heuristics “employed [by people] to assess probabilities and to predict values” (Kahneman and Tversky, 1974, 1124). These heuristics enable people to simplify situations in which probability judgements are complex: “people rely on a limited number of heuristic principles which reduce the complex tasks of assessing probabilities and predicting values to simpler judgmental operations” (Ibid.). Three of these are of particular interest for Kahneman and Tversky: “representativeness”, “availability” and “adjustment and anchoring”.
50Representativeness means that people are inclined to consider A to be representative of B if A resembles B. The similarity at play biases people’s evaluation. For example, they will overestimate the probability that object A belongs to class B (Ibid.). The heuristic “availability” corresponds to situations “in which people assess the frequency of a class or the probability of an event by the ease with which instances or occurrences can be brought to mind” (Ibid., 1127). Finally, “adjustment and anchoring” means that “in many situations, people make estimates by starting from an initial value that is adjusted to yield the final answer. The initial value, or starting point, may be suggested by the formulation of the problem, or it may be the result of a partial computation” (Ibid., 1128).
51Without going into details, the main idea that could be extracted is that these heuristics correspond to a form of regularity in behaviour that explains deviations from the objective model described by Edwards. According to Kahneman and Tversky, these heuristics should be considered as tools to understand and therefore interpret the behaviour of subjects. They give an explanation, an interpretation of the deviations observed; they enrich the debate and modify the border between economics and psychology.
52Starting in 1979, the collaboration with Kahneman took another turn. The two authors offered an alternative model to the expected utility theory: prospect theory.
- 14 See Kahneman and Tversky (1981, 453).
53The starting point was the same: understanding the underlying reasons why rationality axioms are systematically violated (consistency for example). In 1981, Kahneman and Tversky went one step further by making reference to “psychological principles that govern the perception of decision problems and the evaluation of options”. They stopped using the word “heuristic” and referred instead to a “decision frame” to describe “the decision-maker’s conception of the acts, outcomes, and contingencies associated with a particular choice” (Kahneman and Tversky, 1981, 453). Two factors influence and determine the “decision frame”: first “the formulation of the problem” and second “the norms, habits and personal characteristics of the decision-maker” (Ibid.). With the prospect theory, Tversky viewed the formulation issue as the worthy successor of the wording effect. This time, the idea was not to reverse the order in which the options are described but mainly to rephrase these options whilst keeping the literal meaning unchanged. However, the question remained: do we observe a reversal of preference following a change in the formulation of a decision problem? To highlight and test the effect of formulation, one can act on the elements that pervade the decision frame before the evaluation itself. Two sets of factors were traced: “variations in the framing of acts, contingencies, and outcomes, and the characteristic nonlinearities of values and decision weights” (Ibid., 457). For example, the effect of variations in framing can be tested by proposing two decisions only diverging by their formulation.14
54As such, the prospect theory was thought to be an alternative model to the expected utility theory. It consists in using psychological principles as a central component in order to investigate decision problems. This idea was shrewdly captured by the metaphor that the authors used:
If while traveling in a mountain range you notice that the apparent relative height of mountain peaks varies with your vantage point, you will conclude that some impressions of relative height must be erroneous, even when you have no access to the correct answer. Similarly, one may discover that the relative attractiveness of options varies when the same decision problem is framed in different ways. Such a discovery will normally lead the decision-maker to reconsider the original preferences, even when there is no simple way to resolve the inconsistency. (Kahneman and Tversky, 1981, 457)
55Depending on the perspective from which one considers two mountain peaks, one’s evaluation of the relative height of each of them might change since one’s perspective frames and determines one’s evaluation.
To learn the preferences of an agent, particularly among complex gambles, it is obviously necessary to describe the options in words. But how can the experimenter know what those words mean to the subject? The problem is not merely theoretical: it is well known that two descriptions of what the experimenter takes to be the same option may elicit quite different responses from a subject. (Davidson, 1974b, 315)
56This quote by Davidson is of great interest for at least two reasons: first because without specifying who is speaking, one might easily imagine that Tversky is the author of this text, and second because if the starting point is the same in Davidson’s mind, the conclusions drawn from this passage are diametrically opposed to Tversky’s.
57There are many ways of underscoring the divergence between Davidson and Tversky. Yet, for the purpose of our discussion, we will limit our focus on, first, the unified theory of action and meaning developed by Davidson in the 1980s and, second, Davidson’s overall conception of psychology.
58As explained earlier, Tversky offered a theory aiming to think outside the frame of the traditional theory of expected utility. The border between economics and psychology moved towards psychology. The prospect theory is a descriptive theory in which decision theory is a branch of psychology.
59Davidson’s assessment consists in an extension rather than a replacement of the paradigm. In this regard, it seems relevant to describe the way Davidson introduced a theory of communication at the heart of decision theory (i) and presented his own conception of psychology (ii).
60Davidson begins by an argument resembling Tversky’s theory:
In testing a theory of decision making under uncertainty it is usually assumed that subjects understand the words used by the experimenter to describe the alternatives being offered. Or, perhaps more accurately, the experimenter assumes he knows how the subject understands his words. Given the complexity of wagers, the obscurity of the connection that is supposed to exist between states of nature and outcomes, and the many other ways misunderstandings can occur, the assumption of perfect communication between subject and experimenter is often not satisfied. Even if it were, the assumption is unsatisfactory from a theoretical point of view. (Davidson, 1985, 87)
- 16 This difference is reminiscent of the influence Hempel exerted on Davidson, who obviously borrowed (...)
61The conclusions drawn are however quite different16: there is no need to change the paradigm since economists and psychologists have heretofore neglected a significant feature, namely meanings.
- 17 “Broadly stated, my theme is that we should think of meanings and beliefs as interrelated construct (...)
- 18 For further developments, see Sahlin (1990).
- 19 For a detailed presentation, see Harnay (2008). For a detailed and complete presentation of Davidso (...)
62Choices in decision theory are explained by two “psychological factors”: “the relative values the chooser places on the outcomes, and the probability he assigns to those outcomes, conditional on his choice” (Davidson, 1974b, 145). Similarly, understanding a language means having access to both beliefs and meanings, in other words, being able to determine both in the same process17. The aim is to duplicate Ramsey’s stroke of genius consisting in determining simultaneously desires and beliefs via the use of “ethically neutral propositions”18. One the one hand, one has preferences expressed by subjects in decision theory, and on the other hand, in the interpretation theory, one has the primitive “holding true”19 that applies to propositions rather than objects, and provides access to both beliefs and meanings: “A speaker who holds a sentence to be true on an occasion does so in part because of what he means, or would mean, by an utterance of that sentence, and in part because of what he believes.” (Davidson, 1974b, 141) It therefore seems pointless to infer the belief attached to a proposition without knowing its meaning. At any stage of the process, none of the three majors components (desires, beliefs and meanings) are taken for granted. As such, the entire initiative comes from the interpreter: “Theory of interpretation is the business jointly of the linguist, psychologist and philosopher” (Ibid.).
63Surprisingly, one of the arguments used by Davidson in favour of a unified theory—which combines a revised version of the expected utility theory with an interpretation theory—is Edwards’ wording effect:
There is not just an analogy between decision theory and interpretation theory, there is a connection. Seen from the side of decision theory, there is what Ward Edwards once dubbed the ‘presentation problem’ for empirical applications of decision theory. To learn the preferences of an agent, particularly among complex gambles, it is obviously necessary to describe the options in words. But how can the experimenter know what those words mean to the subject? The problem is not merely theoretical: it is well known that two descriptions of what the experimenter takes to be the same option may elicit quite different responses from a subject. (Davidson, 1974b, 147)
64More than twenty years after Decision Making (1957), Davidson proposed a unified theory of decision and interpretation during the 1980’s.
65The unified theory of decision and interpretation proposed by Davidson can be seen as a response to the experimental issues encountered by the Stanford Team:
Theory of meaning as I see it, and Bayesian decision theory, are evidently made for each other. Decision theory must be freed from the assumption of an independently determined knowledge of meaning; theory of meaning calls for a theory of degree of belief in order to make serious use of relations of evidential support. But stating theses mutual dependencies is not enough, for neither theory can be developed first as a basis for the other. There is no way to add on to the other in order to get started, each requires an element drawn from the other. What is wanted is a unified theory that yields degree of belief, utilities on interval scale, and interpretation of speech without assuming any of them. (Davidson, 1980, 158)
66Among the expected results of such a theory, we could mention the enhancement of content due to the introduction of meanings as raw data as well as the explanations available to clarify cases of potential irrationality. There is no need to change the paradigm or to increase the role of psychology: the unified theory grounds its justification in the holism of the mental (presented later in (iii)).
67Guided by Richard Jeffrey’s work, The Logic of Decision ( 1983), Davidson acted on his intuitions about the connection between the decision theory and the interpretation theory by proposing a slightly different version of Jeffrey’s model:
We owe to Richard Jeffrey a version of Bayesian decision theory that makes no direct use of gambles, but treats the objects of preference, the objects to which subjective probabilities are assigned, and the objects to which relative values are assigned uniformly as propositions. (Davidson, 1980, 160)
68Indeed, Jeffrey ( 1983) conceived a particular form of expected utility theory which only deals with propositions, i.e. a set of sentences to whom the subject assigns preferences and subjective probabilities (drawing upon Ethan Bolker’s axioms (1967) and propositional logic). By introducing this idea, Jeffrey is expanding beyond the canonical decision theory inherited from Ramsey and Savage.
- 20 For a detailed presentation, see Harnay (2008).
69While Davidson decided to use Jeffrey’s model, it was not without making slight modifications. Indeed, in line with one of Davidson’s most important statements, one of the first propositions is that one should not take meaning for granted. This amounts to using non-interpreted sentences instead of propositions because from his point of view, the latter could be assimilated to meanings themselves20.
- 21 Davidson mentions a third reason: externalism.
70One should emphasise that the entire structure of the theory is built upon standards of rationality. Among them, Davidson mentioned “the assumption of a rational distribution of probabilities over propositions, and of a proportioning of degrees of belief in accord with the conditional probabilities” (Davidson, 1995, 126). However, even if this “unified theory” has a formal structure that meets the scientific requirements, it cannot be considered as a science like physics. At least two reasons21 can be put forth to explain this fact: first because the basic concepts used in the unified theory cannot be reduced to those used in physics.
There are no precise bridging laws that firmly and reliably relate events or states described in the psychological vocabulary with events or states described in the vocabulary of a physical science. (Davidson, 1995, 129)
71In other words, psychology cannot “emulate” physics. The second reason is the holism of the mental which is not only a holism of attitudes, but also of the concepts expressed in their contents.
- 22 “There are so many concepts that we must have in order to talk about or describe thinking, acting o (...)
72Because of the holism of the mental22, Davidson sees the interpretation of language as a process that determines beliefs and meanings simultaneously. Nevertheless, his view is that the existing solution in decision theory is “neat and satisfying” and that “nothing as good is available in the theory of meaning” (Davidson, 1974b, 145). To illustrate both the holism of the mental and the way Davidson defends an interconnection between desires, beliefs and meanings, an experiment that exemplifies Davidson’s thought must be mentioned.
73In 1974, Davidson described an experiment he carried out with Merrill Carlsmith, a well-known specialist of cognitive dissonance23, in which subjects were asked to make pairwise choices within a small field of alternatives. The set of options was fixed. The idea was to check the consistency of choices by neutralizing learning or conditioning effects. To do so, alternatives were complex and pay-offs were deferred to the end of the experiment (1974a, 234).
74What was not expected according to the theory was the increasing consistency of subjects. What we learn from this experiment is clear: “Merely making choices (with no reward or feedback) alters future choices”. As Davidson pointed out, “from the start there were underlying and consistent values which were better and better realized in choice” (Ibid., 235). In spite of all the safeguards used to control the whys and wherefores of the experiment, this singular effect had not been predicted.
- 24 For a detailed presentation of this issue, see Engel (1989 and 2007).
75A similar argument is used by Davidson about language: “Meaning, as we might loosely use the word, is contaminated by theory, by what is held to be true” (1974c, 187). This argument is put forward by Davidson in his criticism of conceptual relativism. The idea is twofold: first, one postulates the existence of conceptual schemes thought as ways of organising sense-data, systems or point of views that are specific to people, from which they consider reality. Secondly, when the argument is pushed as far as it can go, this relativism entails a failure in the translation: “Reality itself is relative to a scheme: what counts as real in one system may not in another” (Ibid., 183). In other words, thought, language and all the mental representations are relative to those schemes. Everything is perceived from a perspective, the mental content is always relative to a scheme that is not shared with others. The universality of mental contents is denied. Applied to “the truth” for example, it means that the truth is related to a point of view, and is ultimately conditional and relative, not objective24.
76From this argument, it is easy to see the distance between Tversky and Davidson. Indeed, in Tversky’s theory, the frame precedes the evaluation and influences it. The objection raised by Davidson is the following: How can one put oneself in the place of others if there is no translation from one scheme to another, if a scheme cannot be translated into another?
77There is also an epistemological reason for introducing a theory of the interpretation of language at the heart of decision theory. Here, the conception of rationality itself is at stake.
78Indeed, from Davidson’s point of view, the normative and descriptive aspects of the theory cannot be separated: describing, understanding and interpreting one’s behaviour means attributing rationality to others. It would be meaningless to interpret someone without thinking he is rational, at least to the same extent as the interpreter himself:
the description of the intention with which he acted explains why he acted as he did, and at the same time rationalizes the action, that is, reveals a framework within which the action was reasonable from the point of view of the agent … Familiar theories of decision do, of course, idealize the normative element by giving it holistic scope; but even in positing a consistent pattern in the totality of preferences and choices of an agent, decision theories depart from common sense only in degree and precision. Thus (to mention the simplest sort of case) we hesitate to attribute outright violations of transitivity of preference to an agent, and search instead for an explanation that rationalizes the apparent violation, perhaps by redescribing the choices or the alternatives from the agent’s point of view. (Davidson, 1985, 89-90)
- 25 For a detailed presentation, see Harnay (2010).
79Finally, the reason why we should not take interpretation for granted is what Davidson calls “the first person authority”25:
We know in a way no one else can what we believe, fear, want, value, and intend. We know how things seem to us, how they look to us, feel to us, smell and sound to us. We know these things in a way we can never know about the world around us. Whether or not we are sometimes wrong about the contents of our own minds, whether or not we can be in doubt about our own sensations and thoughts, one thing is certainly true of such beliefs: they cannot be generally mistaken. If we think we have a certain thought or sensation, there is a strong presumption that we are right. (Davidson, 1991, 193)
80This privileged access to our own mind does not refer to any form of introspection. The reasoning is as follows: since we have privileged access to our own mind, the subject knows better than anybody else the meaning he attaches to sentences, how he is evaluating objects and events, and the principle of his own desires. We cannot have the same access from outside. The only remaining solution lies in interpreting how these propositional attitudes are expressed in the subject’s language. Having an interpretation in mind, prior to this assertion, means having already explained what should be explained.
81The wording effect presented by Edwards in 1954 was taken over by Tversky and Davidson to question decision making as proposed by the theory of expected utility.
82While each of them agrees on the need to introduce an interpretation theory, the way it is combined to the theory of expected utility differs. For Tversky, psychological principles should be used to implement an interpretation theory at the heart of the decision theory, whereas for Davidson, interpreting should be an operation performed in parallel with the detection of preferences and probability inferences.
83This discrepancy hides two different conceptions of the relationship between economics and psychology. The prospect theory proposed by Tversky is clearly seen as a branch of psychology whereas Davidson’s theory of thought, meaning and action is an extension of and an improvement on the standard expected utility theory.
84From this distinction, two research fields open up: research in neuroeconomics on the one hand, in order to give scientific support to psychological intuition and, on the other hand, research on the paradox of irrationality as explored by Davidson in which the weakness of the will and self-deception play an important role.
I would like to thank the reviewers for their relevant feedbacks and Pascal Engel, Pétronille Rème and Mélaine Harnay (Tufts University) for their useful comments, their expertise and encouragement. A special thought goes to Audrey, Héloïse and Mathilde who gave me the strength to complete this adventure.