Cognition xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Cognition journal homepage: www.elsevier.com/locate/COGNIT
Cognitive science contributions to decision science Jerome R. Busemeyer Psychological and Brain Sciences, Indiana University, 1101 E. 10th St., Bloomington, IN 48705, United States
a r t i c l e
i n f o
Article history: Available online xxxx Keywords: Utility theory Prospect theory Heuristics Sequential sampling models
a b s t r a c t This article briefly reviews the history and interplay between decision theory, behavioral decision-making research, and cognitive psychology. The review reveals the increasingly important impact that psychology and cognitive science have on decision science. One of the main contributions of cognitive science to decision science is the development of dynamic models that describe the cognitive processes that underlay the evolution of preferences during deliberation phase of making a decision. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Is cognitive science having an impact on decision science, and if so, when did this happen, and what is this impact? The answers are yes, and recently, and it is the ability to describe the dynamical nature of the decisionmaking process. However, to understand the answers to these questions, we need to look back into the history of decision research and see exactly where cognitive science really enters the picture in an important and unique way. 2. Economic influence on decision science Decision science is a very large field comprised of researchers from economics, engineering, marketing, statistics, philosophy, psychology, and finally, cognitive science. Decision science has a very long and venerable history going back as far as the 17th and 18th century with initial theoretical formulations by Pascal (1671/1950), Cramer in 1728, and Bernoulli (1738/1954) and others. Arguably the most important and influential contribution in decision science was the axiomatic formulation of expected utility (EU) theory for decisions under risk in the 1940s by Von Neumann and Morgenstern (1947/ 1970), and the later extension to subjective expected utility for decisions under uncertainty by Savage (1954). E-mail address: [email protected]
The axioms of EU theory are a small (3 or 4 depending on the version) set of behavioral properties that a decision maker is supposed to obey. For example, one axiom is dominance – if action A is at least as good or better than action B under all states, then action A should be chosen. Another axiom is transitivity – if action A is chosen over B, and action B over C, then action A should be chosen over C. A third axiom is independence – if two actions involve the same consequence under a given state, then this common consequence should not matter. These axioms strike many decision scientists as intuitively compelling and the rational way to make decisions. Therefore, these behavioral axioms form the definition of rational decision-making: a rational decision maker is a person who obeys the axioms of EU theory. The way to guarantee obedience to these axioms is by using the EU formula to make choices, which is actually a theorem derived from the axioms. The EU formula assigns a utility to each action, by computing a probabilityweighted average of the utilities of outcomes produced by an action. The rational decision maker chooses the action with the maximum EU. By using this rule, one is guaranteed to obey the axioms. Furthermore, for anyone who obeys the axioms, their behavior can be reproduced by this formula, that is, their behavior can be described ‘‘as if’’ they used the EU formula. Using Marr’s (1982) levels of analysis, the EU formula is the computational goal of the rational decision maker. According to EU theory, decision-making boils
http://dx.doi.org/10.1016/j.cognition.2014.11.010 0010-0277/Ó 2014 Elsevier B.V. All rights reserved.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
2
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
down to the use of only two central concepts: the probability weight and the utility value that a decision maker assigns to each outcome. Economists and business researchers working in decision sciences adopted EU theory wholeheartedly, greatly expanding this axiomatic foundation, and delving into deep and important applications using this theory. Economic theories and applications usually begin with the assumption that all agents are rational, that is they are all EU maximizers. However, informal surveys in the 1960s by behavioral economists, such as Allais (1953) and Ellsberg (1961), and subsequent experiments in the 1970s by psychologists such as Kahneman and Tversky (1979) provided convincing evidence that people systematically violate one of the axioms of EU theory – the independence axiom. This prompted some decision scientists, such as for example Machina (1987) and Wakker (2010), to question the applicability, and even the rationality, of the independence axiom. New and more general axiomatic formulations of utility theory were developed, such as those used in rank dependent utility (RDU) theory, proposed by economists such as Quiggin (1982) and Schmeidler (1989). (It is worth noting that these ideas were anticipated by psychologists, Birnbaum and Stegner (1979), Lopes (1987) and Luce (2000).) Although this seems like a revolution in decision theory, the basic ideas did not change: decision-making still boiled down to the use of two concepts: weights and values (but now in more general forms).
3. Psychological influence on decision science The psychological study of behavioral decision making started in the early 1950s, initiated by psychologists such as Coombs (1964), Edwards (1954) and Peterson and Beach (1967). They introduced the simple gambling paradigm, that is, giving people choices among pairs of simple gambles. They initiated a program of research to explore the hypothesis that ‘‘man was an intuitive statistician.’’ In other words, this research designed to experimentally determine how well the EU rule could in fact predict human decision-making behavior. This early work claimed some limited success in the sense that the EU rule turned out to be a fairly robust first approximation to human decision-making. Soon afterwards, the ‘‘man is an intuitive statistician’’ program of research radically changed direction under the influence of Tversky and Kahneman (1974). The revised program was now aimed at showing that in fact people systematically violate the axioms of rational decision theory in fundamental ways. Kahneman and Tversky (1979) were very effective at demonstrating various types of violations, including common consequence, common ratio, and reference point effects. Their work culminated in the formulation of a more descriptively accurate (as opposed to strictly rational) theory of decision making called prospect theory. Prospect theory is essentially a relaxed version of EU theory that builds in some psychological features such as non-additive probability weights and the introduction of loss aversion into the utility function. However, once again, decision-making boiled down to the use of only
two concepts: weights and values (but now more psychologically descriptive). The failure of EU theory prompted other psychologists to look for completely different rules that were simpler, less optimal, and not strictly rational. This lead to the exploration of toolboxes of simple heuristics as originally suggested by Simon, Augier, and March (2004) in the 1950s, and Kahneman and Tversky in the 1980s, and this idea was pursued more programmatically in the 1990s and 2000s by psychologists such as Payne, Bettman, and Johnson (1988) and Gigerenzer and Todd (1999). For example, within a decision environment that is appropriate for the tool, a simple rule such as the lexicographic rule, also known as the ‘‘take the best rule,’’ produces decisions that come close to matching an optimal rule, but with much less cognitive effort. The lexicographic rule evaluates options one attribute at a time, starting with the most important, and working down to less important attributes. If one action exceeds all others on the most important attribute, then it is immediately chosen without considering other attributes; if several actions are approximately equal on the first attribute, then the second attribute is evaluated, and so on. Referring again to Marr’s levels of analysis, heuristic toolbox models attempt to understand the simple algorithms people use to achieve their computational goals. The development of toolboxes of heuristics is perhaps the first fundamentally psychological contribution to decision science. These rules have been implemented within cognitive production rule systems such as the Adaptive Control of Thought (ACT-R) model (Anderson & Lebiere, 1998). The ‘‘tool box of heuristics’’ approach is one that clearly departs from the basic weighted average rule of EU theory, and it is one in which decision-making is based cognitive principles of limited information processing that go beyond the two concepts of weights and values. Heuristic decision rules, such as the lexicographic rule, represent a very drastic departure from the EU rule. The EU rule is compensatory in nature – disadvantages along one dimension (possibility of loss) can be compensated by advantages along another (high possibility of a large gain). Heuristics, such as the lexicographic rule, are usually non-compensatory in nature, e.g., if the options differ on the first dimension that is evaluated, then later dimensions are not evaluated at all, no matter how good or bad! This can lead to violations of dominance and transitivity, which many decision scientists consider to be unacceptable ‘‘irrational’’ properties of a decision theory.
4. Cognitive science influence on decision science Starting in the 1950s, cognitive scientists were busy developing their own theories of decision making for cognitive tasks such as perception, memory recognition, and categorization. The earliest and most prominent was the signal detection model promoted by Green and Swets (1966). The purpose of signal detection theory was to describe how decision makers make inferences about an uncertain state of nature based on a noisy sample of state information (e.g., decide whether an X-ray image is sampled from a patient that has a benign or cancerous tumor). The model
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
was used to predict both accuracy (the probability of taking the appropriate action for a given state of nature), as well as the confidence rating for the action that was chosen. A major limitation of the signal detection model was it provided no mechanism for describing the time required to make the decision. Decisions take time, and the time taken to make a choice can change the decision. Decision time is one of the most important performance measures in cognitive psychology. It is clearly a critical feature of many decisions such as urgent medical decisions or operational military decisions or even product choices by busy consumers. A basic finding in cognitive science is a relation called the speed – accuracy tradeoff: a decision maker can increase accuracy at the cost of increasing decision time. To account for this relation, the signal detection model needed to be expanded to include a mechanism to produce decision time as well as accuracy and confidence. This requirement to simultaneously model all three cognitive performance measures (choice probability, confidence, and decision time) led to the development in 1970s of the sequential sampling class of models by Laming (1968), Link (1975), Ratcliff (1978), Vickers (1979), and others. The basic ideas were actually borrowed from Bayesian sequential hypothesis testing models (DeGroot, 1970) originally developed initially by Wald (1947). The basic idea is that the decision maker sequentially samples information across time until the accumulated evidence becomes sufficiently strong that it crosses a threshold. The first action to cross the threshold determines the choice, the number of samples required to reach the threshold determines the decision time, and the strength of evidence accumulated after the choice determines the confidence rating. Speed – accuracy tradeoffs are accounted by the threshold in these models – increasing the threshold for making a decision increases the amount of evidence that needs to be accumulated, which then increases the time that is required to reach a decision. Sequential sampling models were originally developed to accumulate evidence regarding hypotheses about the state nature that was generating the sample information. Now these sequential sampling models are being widely used to model the accumulation of preferences across time to make risky and multi-attribute decisions (Bhatia, 2013; Busemeyer & Townsend, 1993; Krajbich & Rangel, 2011; Otter, Allenby, & van Zandt, 2008; Stewart & Simpson, 2008; Trueblood, Brown, & Heathcote, 2014; Usher & McClelland, 2004). When applied to decision-making problems, the basic idea is that the decision maker sequentially samples evaluations for each course of action over time until the strength of preference for one action exceeds a threshold. Interestingly, this sequential sampling process provides a simple recursive algorithm for online estimation of the expected utility of each action. At each point in time, a new preference state is obtained by anchoring on the previous preference state and adjusting for the new evaluation. In this way, the current preference state represents an estimate of the utility of action, and this estimate is gradually improved by sequential sampling evaluations over time during the accumulation process. What is being sampled during this preference accumulation process? There are several answers to this question.
3
One idea, called attention switching, is that the decision maker’s attention switches from one attribute or state to another across time accumulating the advantages and disadvantages contributed while attending to each attribute or state across time. Another idea, called decision by sampling, is that the decision maker retrieves sample values from memory for past real world experiences (e.g., evaluates the current price of a product by recalling previous purchases) to estimate the utility of an action. These sampling mechanisms provide opportunities to incorporate cognitive principles of attention and memory more systematically into judgment and decision processes (Dougherty, Gettys, & Ogden, 1999; Weber & Johnson, 2009). Sequential sampling models for preference attempt to derive traditional principles of decision theory, such as risk aversion or loss aversion, by the dynamic processes used during the deliberation process. The advantages of using these cognitive models are that they provide mechanisms that account for puzzling findings regarding preferential choice behavior such as similarity, attraction, compromise, and reference point effects that violate rational choice axioms such as strong stochastic transitivity and independence of irrelevant alternatives. Importantly, they also account for the time taken to make decisions and changes in preferences under time pressure Diederich (2003). Sequential sampling models have also had a large impact in the field of decision neuroscience. Neuroscientists (e.g., Gold & Shadlen, 2007; Schall, 2001), have used multiple cell recordings from the brains of primates to measure the online accumulation of neural activation that leads to a decision. In these applications, sequential sampling models provide accurate accounts of the accumulation of neural activation in the brain. In other applications with humans, EEG and fMRI methods are used to monitor the evidence or preference accumulation process during decision-making. Eye-movements recorded during decision-making have been used to track how attention guides the accumulations process. In summary, both cognitive and neuroscientists seem to be converging on a common basic principle that underlies human decision-making: decisions dynamically evolve from a sequential sampling and accumulation of evidence/ preference process that continues until a threshold criterion is reached. These ideas, which originated in cognitive science, and then spread to neuroscience, are now having an important impact on decision science as well.
References Allais, M. (1953). Le comportement de l’homme rationnel devant le risque: Critique des postulats et axiomes de l’école américaine. Econometrica: Journal of the Econometric Society, 503–546. Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Lawrence Erlbaum Associates Publishers. Bernoulli, D. (1954). Specimen theoriae novae de mensura sortis (Exposition of a new theory on the measurement of risk). Econometrica: Journal of the Econometric Society, 23–36. Bhatia, S. (2013). Associations and the accumulation of preference. Psychological Review, 120(3), 522–543. Birnbaum, M. H., & Stegner, S. E. (1979). Source credibility in social judgment: Bias, expertise, and the judge’s point of view. Journal of Personality and Social Psychology, 37(1), 48–74.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
4
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
Busemeyer, J. R., & Townsend, J. T. (1993). Decision field theory: A dynamic-cognitive approach to decision making in an uncertain environment. Psychological Review, 100(3), 432–459. Coombs, C. H. (1964). A theory of data. Wiley. DeGroot, M. H. (1970). Optimal statistical decisions. Wiley Classics Library. Diederich, A. (2003). Mdft account of decision making under time pressure. Psychonomic Bulletin and Review, 10(1), 157–166. Dougherty, M. R., Gettys, C. F., & Ogden, E. E. (1999). MINERVA-DM: A memory processes model for judgments of likelihood. Psychological Review, 106(1), 180–209. Edwards, W. (1954). The theory of decision making. Psychological Bulletin, 51(4), 380–417. Ellsberg, D. (1961). Risk, ambiguity, and the savage axioms. The Quarterly Journal of Economics, 74(4), 643–669. Gigerenzer, G., & Todd, P. M. (1999). Fast and frugal heuristics: The adaptive toolbox. Oxford University Press. Gold, J. I., & Shadlen, M. N. (2007). The neural basis of decision making. Annual Review Neuroscience, 30, 535–574. Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics (Vol. 1). New York: Wiley. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica: Journal of the Econometric Society, 263–291. Krajbich, I., & Rangel, A. (2011). Multialternative drift-diffusion model predicts the relationship between visual fixations and choice in valuebased decisions. Proceedings of the National Academy of Sciences, 108(33), 13852–13857. Laming, D. R. J. (1968). Information theory of choice-reaction times. Academic Press. Link, S. W. (1975). The relative judgment theory of two choice response time. Journal of Mathematical Psychology, 12(1), 114–135. Lopes, L. L. (1987). Between hope and fear: The psychology of risk. Advances in Experimental Social Psychology, 20, 255–295. Luce, R. D. (2000). Utility of gains and losses: Measurement-theoretical and experimental approaches. Lawrence Erlbaum Press. Machina, M. J. (1987). Decision-making in the presence of risk. Science, 236(4801), 537–543. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. New York, NY: Henry Holt and Co., Inc., pp. 2–46. Otter, T., Allenby, G. M., & van Zandt, T. (2008). An integrated model of discrete choice and response time. Journal of Marketing Research, 45, 593–607.
Pascal, B. (1950). Pensées. Pantheon Books. Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection in decision making. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(3), 534–552. Peterson, C. R., & Beach, L. R. (1967). Man as an intuitive statistician. Psychological Bulletin, 68(1), 29–46. Quiggin, J. (1982). A theory of anticipated utility. Journal of Economic Behavior & Organization, 3(4), 323–343. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59–108. Savage, L. J. (1954). The foundations of statistics. Wiley. Schall, J. D. (2001). Neural basis of deciding, choosing and acting. Nature Reviews Neuroscience, 2(1), 33–42. Schmeidler, D. (1989). Subjective probability and expected utility without additivity. Econometrica: Journal of the Econometric Society, 57(3), 571–587. Simon, H. A., Augier, M., & March, J. G. (Eds.). (2004). Models of a man: Essays in memory of Herbert A. Simon. MIT Press. Stewart, N., & Simpson, K. (2008). A decision-by-sampling account of decision under risk. In N. Chater & M. Oaksford (Eds.), The probabilistic mind: Prospects for Bayesian cognitive science (pp. 261–276). Oxford, England: Oxford University Press. Trueblood, J. S., Brown, S. D., & Heathcote, A. (2014). The multiattribute linear ballistic accumulator model of context effects in multialternative choice. Psychological Review, 121(2), 179–205. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. Usher, M., & McClelland, J. L. (2004). Loss aversion and inhibition in dynamical models of multialternative choice. Psychological Review, 111(3), 757–769. Vickers, D. (1979). Decision processes in visual perception. Academic Press. Von Neumann, J., & Morgenstern, O. (2007). Theory of games and economic behavior (60th anniversary commemorative edition). Princeton University Press. Wakker, P. P. (2010). Prospect theory: For risk and ambiguity. Cambridge University Press. Wald, A. (1947). Sequential analysis. New York: John Wiley and Sons. Weber, E. U., & Johnson, E. J. (2009). Mindful judgment and decision making. Annual Review of Psychology, 60, 53–85.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
Contents lists available at ScienceDirect
Cognition journal homepage: www.elsevier.com/locate/COGNIT
Cognitive science contributions to decision science Jerome R. Busemeyer Psychological and Brain Sciences, Indiana University, 1101 E. 10th St., Bloomington, IN 48705, United States
a r t i c l e
i n f o
Article history: Available online xxxx Keywords: Utility theory Prospect theory Heuristics Sequential sampling models
a b s t r a c t This article briefly reviews the history and interplay between decision theory, behavioral decision-making research, and cognitive psychology. The review reveals the increasingly important impact that psychology and cognitive science have on decision science. One of the main contributions of cognitive science to decision science is the development of dynamic models that describe the cognitive processes that underlay the evolution of preferences during deliberation phase of making a decision. Ó 2014 Elsevier B.V. All rights reserved.
1. Introduction Is cognitive science having an impact on decision science, and if so, when did this happen, and what is this impact? The answers are yes, and recently, and it is the ability to describe the dynamical nature of the decisionmaking process. However, to understand the answers to these questions, we need to look back into the history of decision research and see exactly where cognitive science really enters the picture in an important and unique way. 2. Economic influence on decision science Decision science is a very large field comprised of researchers from economics, engineering, marketing, statistics, philosophy, psychology, and finally, cognitive science. Decision science has a very long and venerable history going back as far as the 17th and 18th century with initial theoretical formulations by Pascal (1671/1950), Cramer in 1728, and Bernoulli (1738/1954) and others. Arguably the most important and influential contribution in decision science was the axiomatic formulation of expected utility (EU) theory for decisions under risk in the 1940s by Von Neumann and Morgenstern (1947/ 1970), and the later extension to subjective expected utility for decisions under uncertainty by Savage (1954). E-mail address: [email protected]
The axioms of EU theory are a small (3 or 4 depending on the version) set of behavioral properties that a decision maker is supposed to obey. For example, one axiom is dominance – if action A is at least as good or better than action B under all states, then action A should be chosen. Another axiom is transitivity – if action A is chosen over B, and action B over C, then action A should be chosen over C. A third axiom is independence – if two actions involve the same consequence under a given state, then this common consequence should not matter. These axioms strike many decision scientists as intuitively compelling and the rational way to make decisions. Therefore, these behavioral axioms form the definition of rational decision-making: a rational decision maker is a person who obeys the axioms of EU theory. The way to guarantee obedience to these axioms is by using the EU formula to make choices, which is actually a theorem derived from the axioms. The EU formula assigns a utility to each action, by computing a probabilityweighted average of the utilities of outcomes produced by an action. The rational decision maker chooses the action with the maximum EU. By using this rule, one is guaranteed to obey the axioms. Furthermore, for anyone who obeys the axioms, their behavior can be reproduced by this formula, that is, their behavior can be described ‘‘as if’’ they used the EU formula. Using Marr’s (1982) levels of analysis, the EU formula is the computational goal of the rational decision maker. According to EU theory, decision-making boils
http://dx.doi.org/10.1016/j.cognition.2014.11.010 0010-0277/Ó 2014 Elsevier B.V. All rights reserved.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
2
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
down to the use of only two central concepts: the probability weight and the utility value that a decision maker assigns to each outcome. Economists and business researchers working in decision sciences adopted EU theory wholeheartedly, greatly expanding this axiomatic foundation, and delving into deep and important applications using this theory. Economic theories and applications usually begin with the assumption that all agents are rational, that is they are all EU maximizers. However, informal surveys in the 1960s by behavioral economists, such as Allais (1953) and Ellsberg (1961), and subsequent experiments in the 1970s by psychologists such as Kahneman and Tversky (1979) provided convincing evidence that people systematically violate one of the axioms of EU theory – the independence axiom. This prompted some decision scientists, such as for example Machina (1987) and Wakker (2010), to question the applicability, and even the rationality, of the independence axiom. New and more general axiomatic formulations of utility theory were developed, such as those used in rank dependent utility (RDU) theory, proposed by economists such as Quiggin (1982) and Schmeidler (1989). (It is worth noting that these ideas were anticipated by psychologists, Birnbaum and Stegner (1979), Lopes (1987) and Luce (2000).) Although this seems like a revolution in decision theory, the basic ideas did not change: decision-making still boiled down to the use of two concepts: weights and values (but now in more general forms).
3. Psychological influence on decision science The psychological study of behavioral decision making started in the early 1950s, initiated by psychologists such as Coombs (1964), Edwards (1954) and Peterson and Beach (1967). They introduced the simple gambling paradigm, that is, giving people choices among pairs of simple gambles. They initiated a program of research to explore the hypothesis that ‘‘man was an intuitive statistician.’’ In other words, this research designed to experimentally determine how well the EU rule could in fact predict human decision-making behavior. This early work claimed some limited success in the sense that the EU rule turned out to be a fairly robust first approximation to human decision-making. Soon afterwards, the ‘‘man is an intuitive statistician’’ program of research radically changed direction under the influence of Tversky and Kahneman (1974). The revised program was now aimed at showing that in fact people systematically violate the axioms of rational decision theory in fundamental ways. Kahneman and Tversky (1979) were very effective at demonstrating various types of violations, including common consequence, common ratio, and reference point effects. Their work culminated in the formulation of a more descriptively accurate (as opposed to strictly rational) theory of decision making called prospect theory. Prospect theory is essentially a relaxed version of EU theory that builds in some psychological features such as non-additive probability weights and the introduction of loss aversion into the utility function. However, once again, decision-making boiled down to the use of only
two concepts: weights and values (but now more psychologically descriptive). The failure of EU theory prompted other psychologists to look for completely different rules that were simpler, less optimal, and not strictly rational. This lead to the exploration of toolboxes of simple heuristics as originally suggested by Simon, Augier, and March (2004) in the 1950s, and Kahneman and Tversky in the 1980s, and this idea was pursued more programmatically in the 1990s and 2000s by psychologists such as Payne, Bettman, and Johnson (1988) and Gigerenzer and Todd (1999). For example, within a decision environment that is appropriate for the tool, a simple rule such as the lexicographic rule, also known as the ‘‘take the best rule,’’ produces decisions that come close to matching an optimal rule, but with much less cognitive effort. The lexicographic rule evaluates options one attribute at a time, starting with the most important, and working down to less important attributes. If one action exceeds all others on the most important attribute, then it is immediately chosen without considering other attributes; if several actions are approximately equal on the first attribute, then the second attribute is evaluated, and so on. Referring again to Marr’s levels of analysis, heuristic toolbox models attempt to understand the simple algorithms people use to achieve their computational goals. The development of toolboxes of heuristics is perhaps the first fundamentally psychological contribution to decision science. These rules have been implemented within cognitive production rule systems such as the Adaptive Control of Thought (ACT-R) model (Anderson & Lebiere, 1998). The ‘‘tool box of heuristics’’ approach is one that clearly departs from the basic weighted average rule of EU theory, and it is one in which decision-making is based cognitive principles of limited information processing that go beyond the two concepts of weights and values. Heuristic decision rules, such as the lexicographic rule, represent a very drastic departure from the EU rule. The EU rule is compensatory in nature – disadvantages along one dimension (possibility of loss) can be compensated by advantages along another (high possibility of a large gain). Heuristics, such as the lexicographic rule, are usually non-compensatory in nature, e.g., if the options differ on the first dimension that is evaluated, then later dimensions are not evaluated at all, no matter how good or bad! This can lead to violations of dominance and transitivity, which many decision scientists consider to be unacceptable ‘‘irrational’’ properties of a decision theory.
4. Cognitive science influence on decision science Starting in the 1950s, cognitive scientists were busy developing their own theories of decision making for cognitive tasks such as perception, memory recognition, and categorization. The earliest and most prominent was the signal detection model promoted by Green and Swets (1966). The purpose of signal detection theory was to describe how decision makers make inferences about an uncertain state of nature based on a noisy sample of state information (e.g., decide whether an X-ray image is sampled from a patient that has a benign or cancerous tumor). The model
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
was used to predict both accuracy (the probability of taking the appropriate action for a given state of nature), as well as the confidence rating for the action that was chosen. A major limitation of the signal detection model was it provided no mechanism for describing the time required to make the decision. Decisions take time, and the time taken to make a choice can change the decision. Decision time is one of the most important performance measures in cognitive psychology. It is clearly a critical feature of many decisions such as urgent medical decisions or operational military decisions or even product choices by busy consumers. A basic finding in cognitive science is a relation called the speed – accuracy tradeoff: a decision maker can increase accuracy at the cost of increasing decision time. To account for this relation, the signal detection model needed to be expanded to include a mechanism to produce decision time as well as accuracy and confidence. This requirement to simultaneously model all three cognitive performance measures (choice probability, confidence, and decision time) led to the development in 1970s of the sequential sampling class of models by Laming (1968), Link (1975), Ratcliff (1978), Vickers (1979), and others. The basic ideas were actually borrowed from Bayesian sequential hypothesis testing models (DeGroot, 1970) originally developed initially by Wald (1947). The basic idea is that the decision maker sequentially samples information across time until the accumulated evidence becomes sufficiently strong that it crosses a threshold. The first action to cross the threshold determines the choice, the number of samples required to reach the threshold determines the decision time, and the strength of evidence accumulated after the choice determines the confidence rating. Speed – accuracy tradeoffs are accounted by the threshold in these models – increasing the threshold for making a decision increases the amount of evidence that needs to be accumulated, which then increases the time that is required to reach a decision. Sequential sampling models were originally developed to accumulate evidence regarding hypotheses about the state nature that was generating the sample information. Now these sequential sampling models are being widely used to model the accumulation of preferences across time to make risky and multi-attribute decisions (Bhatia, 2013; Busemeyer & Townsend, 1993; Krajbich & Rangel, 2011; Otter, Allenby, & van Zandt, 2008; Stewart & Simpson, 2008; Trueblood, Brown, & Heathcote, 2014; Usher & McClelland, 2004). When applied to decision-making problems, the basic idea is that the decision maker sequentially samples evaluations for each course of action over time until the strength of preference for one action exceeds a threshold. Interestingly, this sequential sampling process provides a simple recursive algorithm for online estimation of the expected utility of each action. At each point in time, a new preference state is obtained by anchoring on the previous preference state and adjusting for the new evaluation. In this way, the current preference state represents an estimate of the utility of action, and this estimate is gradually improved by sequential sampling evaluations over time during the accumulation process. What is being sampled during this preference accumulation process? There are several answers to this question.
3
One idea, called attention switching, is that the decision maker’s attention switches from one attribute or state to another across time accumulating the advantages and disadvantages contributed while attending to each attribute or state across time. Another idea, called decision by sampling, is that the decision maker retrieves sample values from memory for past real world experiences (e.g., evaluates the current price of a product by recalling previous purchases) to estimate the utility of an action. These sampling mechanisms provide opportunities to incorporate cognitive principles of attention and memory more systematically into judgment and decision processes (Dougherty, Gettys, & Ogden, 1999; Weber & Johnson, 2009). Sequential sampling models for preference attempt to derive traditional principles of decision theory, such as risk aversion or loss aversion, by the dynamic processes used during the deliberation process. The advantages of using these cognitive models are that they provide mechanisms that account for puzzling findings regarding preferential choice behavior such as similarity, attraction, compromise, and reference point effects that violate rational choice axioms such as strong stochastic transitivity and independence of irrelevant alternatives. Importantly, they also account for the time taken to make decisions and changes in preferences under time pressure Diederich (2003). Sequential sampling models have also had a large impact in the field of decision neuroscience. Neuroscientists (e.g., Gold & Shadlen, 2007; Schall, 2001), have used multiple cell recordings from the brains of primates to measure the online accumulation of neural activation that leads to a decision. In these applications, sequential sampling models provide accurate accounts of the accumulation of neural activation in the brain. In other applications with humans, EEG and fMRI methods are used to monitor the evidence or preference accumulation process during decision-making. Eye-movements recorded during decision-making have been used to track how attention guides the accumulations process. In summary, both cognitive and neuroscientists seem to be converging on a common basic principle that underlies human decision-making: decisions dynamically evolve from a sequential sampling and accumulation of evidence/ preference process that continues until a threshold criterion is reached. These ideas, which originated in cognitive science, and then spread to neuroscience, are now having an important impact on decision science as well.
References Allais, M. (1953). Le comportement de l’homme rationnel devant le risque: Critique des postulats et axiomes de l’école américaine. Econometrica: Journal of the Econometric Society, 503–546. Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Lawrence Erlbaum Associates Publishers. Bernoulli, D. (1954). Specimen theoriae novae de mensura sortis (Exposition of a new theory on the measurement of risk). Econometrica: Journal of the Econometric Society, 23–36. Bhatia, S. (2013). Associations and the accumulation of preference. Psychological Review, 120(3), 522–543. Birnbaum, M. H., & Stegner, S. E. (1979). Source credibility in social judgment: Bias, expertise, and the judge’s point of view. Journal of Personality and Social Psychology, 37(1), 48–74.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
4
J.R. Busemeyer / Cognition xxx (2014) xxx–xxx
Busemeyer, J. R., & Townsend, J. T. (1993). Decision field theory: A dynamic-cognitive approach to decision making in an uncertain environment. Psychological Review, 100(3), 432–459. Coombs, C. H. (1964). A theory of data. Wiley. DeGroot, M. H. (1970). Optimal statistical decisions. Wiley Classics Library. Diederich, A. (2003). Mdft account of decision making under time pressure. Psychonomic Bulletin and Review, 10(1), 157–166. Dougherty, M. R., Gettys, C. F., & Ogden, E. E. (1999). MINERVA-DM: A memory processes model for judgments of likelihood. Psychological Review, 106(1), 180–209. Edwards, W. (1954). The theory of decision making. Psychological Bulletin, 51(4), 380–417. Ellsberg, D. (1961). Risk, ambiguity, and the savage axioms. The Quarterly Journal of Economics, 74(4), 643–669. Gigerenzer, G., & Todd, P. M. (1999). Fast and frugal heuristics: The adaptive toolbox. Oxford University Press. Gold, J. I., & Shadlen, M. N. (2007). The neural basis of decision making. Annual Review Neuroscience, 30, 535–574. Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics (Vol. 1). New York: Wiley. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica: Journal of the Econometric Society, 263–291. Krajbich, I., & Rangel, A. (2011). Multialternative drift-diffusion model predicts the relationship between visual fixations and choice in valuebased decisions. Proceedings of the National Academy of Sciences, 108(33), 13852–13857. Laming, D. R. J. (1968). Information theory of choice-reaction times. Academic Press. Link, S. W. (1975). The relative judgment theory of two choice response time. Journal of Mathematical Psychology, 12(1), 114–135. Lopes, L. L. (1987). Between hope and fear: The psychology of risk. Advances in Experimental Social Psychology, 20, 255–295. Luce, R. D. (2000). Utility of gains and losses: Measurement-theoretical and experimental approaches. Lawrence Erlbaum Press. Machina, M. J. (1987). Decision-making in the presence of risk. Science, 236(4801), 537–543. Marr, D. (1982). Vision: A computational investigation into the human representation and processing of visual information. New York, NY: Henry Holt and Co., Inc., pp. 2–46. Otter, T., Allenby, G. M., & van Zandt, T. (2008). An integrated model of discrete choice and response time. Journal of Marketing Research, 45, 593–607.
Pascal, B. (1950). Pensées. Pantheon Books. Payne, J. W., Bettman, J. R., & Johnson, E. J. (1988). Adaptive strategy selection in decision making. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(3), 534–552. Peterson, C. R., & Beach, L. R. (1967). Man as an intuitive statistician. Psychological Bulletin, 68(1), 29–46. Quiggin, J. (1982). A theory of anticipated utility. Journal of Economic Behavior & Organization, 3(4), 323–343. Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59–108. Savage, L. J. (1954). The foundations of statistics. Wiley. Schall, J. D. (2001). Neural basis of deciding, choosing and acting. Nature Reviews Neuroscience, 2(1), 33–42. Schmeidler, D. (1989). Subjective probability and expected utility without additivity. Econometrica: Journal of the Econometric Society, 57(3), 571–587. Simon, H. A., Augier, M., & March, J. G. (Eds.). (2004). Models of a man: Essays in memory of Herbert A. Simon. MIT Press. Stewart, N., & Simpson, K. (2008). A decision-by-sampling account of decision under risk. In N. Chater & M. Oaksford (Eds.), The probabilistic mind: Prospects for Bayesian cognitive science (pp. 261–276). Oxford, England: Oxford University Press. Trueblood, J. S., Brown, S. D., & Heathcote, A. (2014). The multiattribute linear ballistic accumulator model of context effects in multialternative choice. Psychological Review, 121(2), 179–205. Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. Usher, M., & McClelland, J. L. (2004). Loss aversion and inhibition in dynamical models of multialternative choice. Psychological Review, 111(3), 757–769. Vickers, D. (1979). Decision processes in visual perception. Academic Press. Von Neumann, J., & Morgenstern, O. (2007). Theory of games and economic behavior (60th anniversary commemorative edition). Princeton University Press. Wakker, P. P. (2010). Prospect theory: For risk and ambiguity. Cambridge University Press. Wald, A. (1947). Sequential analysis. New York: John Wiley and Sons. Weber, E. U., & Johnson, E. J. (2009). Mindful judgment and decision making. Annual Review of Psychology, 60, 53–85.
Please cite this article in press as: Busemeyer, J. R. Cognitive science contributions to decision science. Cognition (2014), http://dx.doi.org/ 10.1016/j.cognition.2014.11.010
