We generally agree with Boyer & Petersen (B&P) regarding folk-economic beliefs (FEBs), including that economic cognition is largely terra incognita (sect. 6.3, para. 3), and that a computational, or “design stance,” framework is a viable alternative to intuitive person-level descriptions of rationality and utility – that is, the “intentional stance” (Dennett Reference Dennett1987). This might even be carried a step further: Person-level, commonsense “folk psychology” (FP) is robustly predictive in social encounters (Baron-Cohen et al. Reference Baron-Cohen, Tager-Flusberg and Lombardo2013), but it is an obstacle for less-intuitive concepts in psychology (Bloom Reference Bloom2004; Damasio Reference Damasio2005), especially in light of progress in computational and neurobiological research (Churchland Reference Churchland1981; Reference Churchland2013). In general, if computational models are better for carving psychology at its joints, then they should displace FP models where the two are at odds.
This disconnect between computational and FP perspectives is precisely the current state of affairs for competing theories of economic cognition and behavior. As B&P note, the main challenge is in describing “how [computational] models and findings could be integrated with classical, and often empirically successful, descriptions of economic behavior in terms of rationality […] and utility” (sect. 6.3, para. 5). The empirical success of classical descriptions is based, in part, on results in experimental economic games, which conventionally aim to isolate and test the forces of interest, namely, rationality and various forms of utility (sect. 6.3, para. 4) in the absence of environmental noise (Colman Reference Colman1982; Camerer & Fehr Reference Camerer, Fehr, Henrich, Boyd, Bowles, Camerer, Fehr and Gintis2005). They attempt to do this by omitting any environmental context except for the abstract rules of the game. When empirical results deviate systematically from the “standard model” of rationality and self-regarding utility, some researchers have attempted to rescue rationality by rejecting self-regarding utility in favor of other-regarding utility (e.g., Fehr & Schmidt Reference Fehr and Schmidt1999; Gintis Reference Gintis2007).
Environmental context is not noise, however; it is a signal. The decision-making machinery of all living organisms, including humans and human ancestors, evolved to make decisions based, in part, on cues of environmental context. This attempt to eliminate contextual cues from the experimental procedure therefore fails to control for an essential element of the decision-making process – that is, it increases rather than reduces noise. If empirical deviations from the “standard model” are based on a failure to control essential aspects of the decision-making process, then attempts to reconcile them with high-level FP concepts are misguided.
Taking a design stance, others argue that it is probably not computationally tractable for agents to search for optimal solutions in a novel decision-making task such as an economics experiment, and therefore suggest that agents rely on bounded rationality and a “toolkit” of ecologically valid computational heuristics (e.g., Gigerenzer Reference Gigerenzer2010; Gigerenzer & Selten Reference Gigerenzer and Selten2001; Kahneman Reference Kahneman2003; Simon Reference Simon, Eatwell, Milgate and Newman1990). If so, the cognitive problem for participants in economic experiments would not be utility maximization in their task environment (Simon Reference Simon1991), but rather, context identification prior to executing some associated heuristic – also known as the “frame problem.”
Solving the frame problem is especially difficult in experimental economics games that provide participants with only the rules and no context, and most “systematic” divergences from “subjective utility maximization predictions” (sect. 6.3, para. 3) occur in exactly these types of experiments. In general, we know very little about the cognitive mechanisms that “match input conditions” to “specific systems” (Figure 2 in B&P) in novel tasks such as experimental economic games. But what we do know is that when input conditions, or frames, are provided to participants, results can vary from the empirically robust classical findings. For instance, they can deviate subtly when subtle framing cues are presented (e.g., Cronk Reference Cronk2007; Cronk & Wasielewski Reference Cronk and Wasielewski2008; Eriksson & Strimling Reference Eriksson and Strimling2014; Gerkey Reference Gerkey2013; Keser & van Winden Reference Keser and van Winden2000; Liberman Reference Liberman2004; Leliveld et al. Reference Leliveld, van Dijk and van Beest2008), and they can deviate dramatically when even slightly more detailed framing scenarios are presented (Lightner et al. Reference Lightner, Barclay and Hagen2017). More important, these cited framing effects all deviate from standard findings in a way that reflects the social norms associated with the provided context in each experiment.
Theoretical models of economic cognition should therefore not take studies finding systematic deviations from standard economic theory at face value to begin with, especially when they are found in decontextualized experimental economic games (Hagen & Hammerstein Reference Hagen and Hammerstein2006). Interestingly, a version of this critique is raised by B&P themselves against the idea that the FEBs reveal an implicit theory of the economy (sect. 6.2). As they rightly note, there are likely as many different cognitive models of an economic scenario as there are individuals modeling that scenario. This exact line of reasoning can be applied to novel exchange scenarios such as experimental economic games, and as a consequence, there are likely as many different games being played as there are participants (Hagen & Hammerstein Reference Hagen and Hammerstein2006).
It is nonetheless tempting for researchers to continue taking behavioral “signals” from decontextualized game studies for granted – as B&P do in some parts of their target article (e.g., FEB 5 in sect. 5.4). The theoretical conclusions they extrapolate from them, however, are susceptible to the critique we raise here, and raised by B&P in section 6.2. If computational models of economic cognition will accommodate behavioral findings, then their radically different theoretical frameworks, each with radically different assumptions, must be addressed in future research. Productive steps in this direction would include reconsidering questionable results from decontextualized experimental economic games.
You have
Access
We generally agree with Boyer & Petersen (B&P) regarding folk-economic beliefs (FEBs), including that economic cognition is largely terra incognita (sect. 6.3, para. 3), and that a computational, or “design stance,” framework is a viable alternative to intuitive person-level descriptions of rationality and utility – that is, the “intentional stance” (Dennett Reference Dennett1987). This might even be carried a step further: Person-level, commonsense “folk psychology” (FP) is robustly predictive in social encounters (Baron-Cohen et al. Reference Baron-Cohen, Tager-Flusberg and Lombardo2013), but it is an obstacle for less-intuitive concepts in psychology (Bloom Reference Bloom2004; Damasio Reference Damasio2005), especially in light of progress in computational and neurobiological research (Churchland Reference Churchland1981; Reference Churchland2013). In general, if computational models are better for carving psychology at its joints, then they should displace FP models where the two are at odds.
This disconnect between computational and FP perspectives is precisely the current state of affairs for competing theories of economic cognition and behavior. As B&P note, the main challenge is in describing “how [computational] models and findings could be integrated with classical, and often empirically successful, descriptions of economic behavior in terms of rationality […] and utility” (sect. 6.3, para. 5). The empirical success of classical descriptions is based, in part, on results in experimental economic games, which conventionally aim to isolate and test the forces of interest, namely, rationality and various forms of utility (sect. 6.3, para. 4) in the absence of environmental noise (Colman Reference Colman1982; Camerer & Fehr Reference Camerer, Fehr, Henrich, Boyd, Bowles, Camerer, Fehr and Gintis2005). They attempt to do this by omitting any environmental context except for the abstract rules of the game. When empirical results deviate systematically from the “standard model” of rationality and self-regarding utility, some researchers have attempted to rescue rationality by rejecting self-regarding utility in favor of other-regarding utility (e.g., Fehr & Schmidt Reference Fehr and Schmidt1999; Gintis Reference Gintis2007).
Environmental context is not noise, however; it is a signal. The decision-making machinery of all living organisms, including humans and human ancestors, evolved to make decisions based, in part, on cues of environmental context. This attempt to eliminate contextual cues from the experimental procedure therefore fails to control for an essential element of the decision-making process – that is, it increases rather than reduces noise. If empirical deviations from the “standard model” are based on a failure to control essential aspects of the decision-making process, then attempts to reconcile them with high-level FP concepts are misguided.
Taking a design stance, others argue that it is probably not computationally tractable for agents to search for optimal solutions in a novel decision-making task such as an economics experiment, and therefore suggest that agents rely on bounded rationality and a “toolkit” of ecologically valid computational heuristics (e.g., Gigerenzer Reference Gigerenzer2010; Gigerenzer & Selten Reference Gigerenzer and Selten2001; Kahneman Reference Kahneman2003; Simon Reference Simon, Eatwell, Milgate and Newman1990). If so, the cognitive problem for participants in economic experiments would not be utility maximization in their task environment (Simon Reference Simon1991), but rather, context identification prior to executing some associated heuristic – also known as the “frame problem.”
Solving the frame problem is especially difficult in experimental economics games that provide participants with only the rules and no context, and most “systematic” divergences from “subjective utility maximization predictions” (sect. 6.3, para. 3) occur in exactly these types of experiments. In general, we know very little about the cognitive mechanisms that “match input conditions” to “specific systems” (Figure 2 in B&P) in novel tasks such as experimental economic games. But what we do know is that when input conditions, or frames, are provided to participants, results can vary from the empirically robust classical findings. For instance, they can deviate subtly when subtle framing cues are presented (e.g., Cronk Reference Cronk2007; Cronk & Wasielewski Reference Cronk and Wasielewski2008; Eriksson & Strimling Reference Eriksson and Strimling2014; Gerkey Reference Gerkey2013; Keser & van Winden Reference Keser and van Winden2000; Liberman Reference Liberman2004; Leliveld et al. Reference Leliveld, van Dijk and van Beest2008), and they can deviate dramatically when even slightly more detailed framing scenarios are presented (Lightner et al. Reference Lightner, Barclay and Hagen2017). More important, these cited framing effects all deviate from standard findings in a way that reflects the social norms associated with the provided context in each experiment.
Theoretical models of economic cognition should therefore not take studies finding systematic deviations from standard economic theory at face value to begin with, especially when they are found in decontextualized experimental economic games (Hagen & Hammerstein Reference Hagen and Hammerstein2006). Interestingly, a version of this critique is raised by B&P themselves against the idea that the FEBs reveal an implicit theory of the economy (sect. 6.2). As they rightly note, there are likely as many different cognitive models of an economic scenario as there are individuals modeling that scenario. This exact line of reasoning can be applied to novel exchange scenarios such as experimental economic games, and as a consequence, there are likely as many different games being played as there are participants (Hagen & Hammerstein Reference Hagen and Hammerstein2006).
It is nonetheless tempting for researchers to continue taking behavioral “signals” from decontextualized game studies for granted – as B&P do in some parts of their target article (e.g., FEB 5 in sect. 5.4). The theoretical conclusions they extrapolate from them, however, are susceptible to the critique we raise here, and raised by B&P in section 6.2. If computational models of economic cognition will accommodate behavioral findings, then their radically different theoretical frameworks, each with radically different assumptions, must be addressed in future research. Productive steps in this direction would include reconsidering questionable results from decontextualized experimental economic games.