The last decades have seen the rise of experimental and other empirical methods in almost all branches of economics. Pace many skeptics, proponents of experimental methods frequently promise to substantially transform the discipline. Two traditions of experiments in microeconomics are especially prominent, namely cognitive psychology experiments in the heuristics and biases tradition of Daniel Kahneman and Amos Tversky (H&B-experiments) and experimental economics in the tradition of Vernon Smith. While H&B-experiments emphasize the importance of studying cognitive mechanisms for improving economic models, experimental economics highlights the importance of institutions and social environments. These two experimental paradigms have been frequently portrayed as rivals by observers and their respective proponents alike (e.g., Loewenstein 1999, Dekker & Remic, 2019, Smith & Wilson, 2019).

However, on closer inspection, it is hard to pin down what the disagreement is supposed to be about. To illustrate, Smith & Wilson (2019), two prominent experimental economists, want to explain how human choice-behavior depends on our ability to plug into specific environments via our sentiments as envisioned by Adam Smith (1759/2010). They motivate this project in opposition to the literature on social preferences developed by allies of the H&B-tradition such as Fehr & Schmidt (1999).Footnote 1 This approach aims at explaining certain behavioral patterns observed in laboratory experiments by introducing further terms (e.g., inequity aversion) into the agents’ utility functions. While Smith and Wilson set up experimental economics in contrast to social preferences research, it remains unclear whether the two are conflicting or complementary. While the social preference approach can be seen as an attempt of explaining with preferences (i.e., explaining the behavior of agents via their preferences), the approach of Smith and Wilson looks more like an attempt at what Guala (2019) would call explaining preferences (i.e., explaining why agents have certain preferences). In the light of this, one may be tempted to conclude that the tension between the experimental paradigms is not based on substantial issues and better explained in other terms. However, in this paper, I resist this conclusion and instead offer a novel conceptualization of the two experimental paradigms that highlights a plausible source of genuine disagreement, namely their vision of preferences.

In the first half of the paper, I argue that experimental economics can be reconstructed as holding that the constituents of preferences can be partially located in agents’ environments, while H&B-experiments tend to implicitly assume that the constituents of preferences are entirely located within the agents’ bodies (Sect. 1).Footnote 2 I then illustrate how these different commitments about the location of the constituents of preferences shed light on the disagreement between the two experimental paradigms (Sect. 2).

To further support my reconstruction, the second half of the paper argues that locating part of the constituents of preferences in the environment might seem counterintuitive but is in fact plausible. More specifically, I first survey the three prominent conceptions of what preferences in economics are, i.e., behaviorism, mentalism, and dispositionalism (see Hausman 2011, Guala 2019, Thoma, 2021). I then argue that these three conceptions are, in principle, all compatible with locating part of the constituents of preferences in the environment and demonstrate how my reconstruction can be articulated under each of them (Sect. 3).Footnote 3 Thereby, I also highlight a neglected dimension in the debate about the nature of preferences: instead of only offering accounts of what preferences are, philosophers of economics should also ask where preferences are. Before concluding, I address objections to my proposals. In particular, I discusses whether my reconstruction strays too far from ‘commonsense’ and why I take it to be more advantageous than construing the disagreement in terms of explaining via preferences vs. explaining via incentives (Sect. 4).

1 Two Experimental Paradigms

Historically, H&B-experiments have their origin in the idea that neoclassical economics is largely based on psychologically inadequate assumptions about human rationality. H&B-experimenters, therefore, hold that economics needs a robust injection of psychological theory and method to uncover the mechanisms (e.g., heuristics and biases) that really guide our choice-behavior (Rabin, 2002, also see Guala 2019). Experimental economics, on the other hand, holds that economics proper is the study of institutions like markets. Therefore, it urges a focus on the study of social environments, and its proponents frequently dismiss work of the H&B-tradition as too narrow (e.g., Smith 2008).Footnote 4 Yet, both paradigms agree on the importance of experiments to determine how we should apply and modify choice-theoretic models, which represent economic agents via their beliefs and preferences. Given that both paradigms rely on beliefs and preferences as their most fundamental concepts, it appears puzzling how they can fiercely disagree over what experimental research is needed to decide how to model a particular class of agents. In the following, I briefly expand on the two experimental paradigms and propose that their disagreement can be reconstructed in terms of different commitments about the location of the constituents of preferences.

1.1 H&B-experiments

The aim of H&B-experiments is to collect data on human choice-behavior in laboratory settings and use this information to uncover the mechanisms that underly certain (structural) features of preferences (e.g., Tversky 1969, Slovic et al., 2002, Busemeyer et al., 2006). Knowledge of these mechanisms is then employed to construct allegedly better choice-theoretic models. During the relevant experiments, individuals are usually separated from their familiar environments and presented with several choice-problems in psychological laboratories. Based on their choices, a set of preferences is ascribed to the individuals, and mechanisms for explaining the (structural) features of these preferences are hypothesized.

An early example of this line of research is Tversky’s (1969, 1972) work on intransitive preferences. One mechanism that is meant to account for the intransitive preferences observed in various experiments is attribute-by-attribute-comparison. Attribute-by-attribute-comparison holds that a decision-maker compares her options along their individual attribute dimensions in her mind and counts over how many dimensions of one option it beats the corresponding dimensions of another one. She will then choose the option that beats its rivals on most attribute dimensions. Tversky argues that humans tend to engage in such one-dimensional evaluations before trying to aggregate across different attributes of an object. To illustrate how this mechanism can account for intransitive preferences, consider an agent who must choose between three different winter jackets (X, Y, Z). Let us assume that the agent ranks the attributes of the jacket as follows:

  1. 1.

    Comfort: X > Y > Z.

  2. 2.

    Style: Z > X > Y.

  3. 3.

    Price: Y > Z > X.

In this example, X beats Y 2 to 1, Y beats Z 2 to 1, and Z beats X 2 to 1. Hence, the resulting preference relation of the agent in the example will be intransitive (X > Y > Z > X).

What I want to highlight is that the H&B-tradition has a strong tendency to make the implicit assumption that the mechanisms that constitute preferences are solely located within the agent’s body. For instance, Tversky (1969, 1972) holds that attribute-by-attribute-comparison should be interpreted as a cognitive heuristic and, therefore, takes place within the agents’ minds (see Guala 2019, p. 395). A further example of this implicit assumption is given by Busemeyer et al., (2006) who identify the “state of preference” with the aggregate of evaluations across different times and argue that this value has a neural basis. Moreover, Slovic et al., (2002) argue that preferences are sometimes constituted by affect, which they define as a feeling state. In addition to that, Bettman et al., (1998) hold that individuals often do not have well-defined preferences because there is no context-independent cognitive mechanism that could constitute such preferences.

Note that while I hold that H&B-experimenters frequently make the implicit assumption that preferences are solely located within the agent’s body, contextual influences on choices are clearly of interest to experimenters in the H&B-tradition. For instance, the paradigm extensively studies the influence of so-called framing effects and choice-architectures, i.e., how the presentations of options can influence what is chosen. However, those influences are usually conceptualized as triggering different internal states. Following Bacharach (2003), frames can be understood as a set of concepts used in agents’ representations of options. Different choice-architectures can lead agents to draw on different conceptual resources in their representation of options. What H&B-experimenters have argued is that while agents may prefer one option over another under one frame, this may not be the case under other frames, i.e., different choice-architectures can trigger different preferences. Yet, nothing in this line of thought requires one to include parts of the environment into the constituents of agents’ preferences. It is, of course, possible to conceptualize framing effects in such a way that environments become parts of agents’ preferences. However, as the seminal account of Bacharach (2003) indicates, this is not how they are usually understood. Instead, preferences usually seem to be understood as internally stored rankings over options under a particular frame.

Moreover, even in cases in which H&B-experimenters investigate how environmental features like market signals influence preferences, there still appears to be a strong commitment to preferences being solely located within an agent’s body. Take, for instance, Baucells et al., (2011) who investigate the updating of reference points. In order to elicit the reference points of their subjects they asked them “at what selling price [they would] feel neutral about the sale of the stock, i.e., be neither happy nor unhappy about the sale” (Baucells et al., 2011, p. 508). This suggests that they conceptualize reference points in terms of internal feeling states.

Finally, in cases where environmental influences cannot plausibly be conceptualized in terms of updating or triggering different internal states, they are often considered arbitrary and supposed to demonstrate that agents do not genuinely have certain preferences. For example, Ariely et al., (2003, p. 77) argue that “consumers do not arrive at a choice or at a pricing task with an inventory of preexisting preferences.” They present several experiments that aim at demonstrating that agents’ choices are heavily influenced by “arbitrary” environmental anchors. Ariely et al., (2003, p. 102) take their experiments to, inter alia, indicate that even if consumers can be perfectly well described by an ordinal utility function, this cannot inform us about their “true preference in any meaningful sense of the term.” They hold that this follows from the possibility that consumers’ choices could have been influenced by the kind of arbitrary anchors they highlight in their experiments. In this regard, they seem to conceptualize “true” preferences as preexisting internal states with which agents could “arrive” at a particular choice-task. To sum up, the H&B-tradition encourages, albeit implicitly, the assumption that preferences are solely located within the agent’s body.

1.2 Experimental Economics

Experimental economics is interested mostly in cases where the outcomes predicted by choice-theoretic models are supposed to result from the interaction of individuals with their environment. In this regard, experimental economists build certain (institutional) environments into their experimental settings and study how changing these environments affects the accuracy of the predictions of choice-theoretic models. These studies assume that, depending on the details of the choice-tasks, material environments, and rules of interactions at hand, different choice-theoretic models will perform better or worse than their competitors. Experimental economics thus suggests that what economist should ultimately study is how variations in the environment affects the performance of choice-theoretic models (Smith 2003). This can help us to determine which circumstances can be adequately described by which choice-theoretic model.

To illustrate, experimental economists have argued that Nash-equilibrium predictions work well for the Fouraker-Siegel bargaining environment but break down in the ultimatum-game environment. Moreover, the extent to which the winners’ curse occurs in common-value auctions differs between scenarios involving 3–4 bidders versus situations with 6–7 bidders. Yet, simple game-theoretic models would predict that the phenomenon does not occur at all (Smith 1994, see Plott 1991).

My reconstruction of experimental economics seeks to establish a clear point of tension with the H&B-tradition. I start from the idea that preferences and beliefs, including those about other agents and available options, are the most fundamental concepts for constructing any choice-theoretic model. If we want to affect the outcome of a choice-theoretic model, we either have to change the preferences or the beliefs assigned to at least one of the modeled agents (including beliefs about available options and other agents). Because of this, it appears attractive to reconstruct the difference between the two experimental paradigms in terms of different commitments about preferences (and beliefs). In particular, as I shall argue, those commitments pertain to the location of the constituents of these states.

As highlighted above, the H&B-tradition frequently makes the implicit assumption that the constituents of preferences are solely located within the agent’s body. In contrast to that, experimental economics heavily emphasizes the importance of environments when assessing choice-theoretic models. This suggests that the emphasis on the environment in experimental economics can be reconstructed as implementing the idea that the constituents of choice-theoretic concepts like preferences are partially located in the agent’s environment. This idea provides us with a straightforward contrast between the two experimental paradigms in terms of which we can attempt to reconstruct their disagreement.Footnote 5

Importantly, I do not claim that experimental economists themselves necessarily conceive of their work like this. Nevertheless, some arguments by experimental economists suggest that my proposed reconstruction may not be too far off from how they think about their own work. Consider for instance Wilson (2010, p. 80), who wants to confute “the notion that the individual’s expression of his social preferences as an action can be represented as a set of separable and private utilitarian ‘preferences’ within him” (my emphasis).Footnote 6

2 Explicating the Disagreement

In this section, I illustrate how the two different commitments about where (the constituents of) preferences are can lead us to assign different preferences based on the same observations. In other words, I outline how exactly the commitments matter for interpreting the results of experiments. I, thereby, want to demonstrate that these commitments can indeed account for (part of) the disagreement between the two paradigms.

Let us start with a thought-experiment. Imagine that we are interested in modelling the choice-behavior of Bob, who is buying and selling second-hand clothing. After studying his buying behavior in the market, we are struggling to find a utility function that describes Bob’s choice-behavior well. Consequently, we become suspicious that Bob’s preferences do not satisfy the axioms required for the existence of a standard utility function. To get more information on the constituents of Bob’s preferences, we start studying his behavior in a laboratory.

Surprisingly, in the laboratory, Bob’s decisions only depend on the color of the clothing options offered to him. In the laboratory, Bob exhibited a complete and transitive preference relation over different colors and his choice between options was only influenced by the colors of the clothing. Consequently, we have no trouble coming up with a utility function that can capture Bob’s choices in the laboratory. However, once we try to employ this utility function to predict Bob’s behavior in the market for second-hand clothing, it again fails to capture Bob’s choice-behavior.

How can we make sense of this case? One possibility is that Bob, while not being studied in the laboratory, has access to certain parts of his environment that are essential for executing the decision mechanisms he employs in the market. For instance, following Tversky (1972), we could imagine that Bob’s preferences are constituted by a complicated version of attribute-by-attribute-comparison that requires him to track a large number of attributes. Yet, Bob might only be able to carry out this version of attribute-by-attribute-comparison provided that he has access to certain parts of the environment in which he usually makes his choices, for example, a pen and a notebook. Apart from the color, it may also matter for Bob whether the clothing was produced under fair conditions, whether eco-friendly material was used, and so on. We can easily continue this list. What is important here is that a larger number of attributes can make it quite demanding to track which option beats which on a greater number of attribute-dimensions. As a result, Bob might be unable to carry out attribute-by-attribute-comparison unless we provide him with a pen and a notebook which he can use to keep track of the beating relations. Yet, while he has access to these items in the market, Bob had no access to them in the laboratory and could, therefore, not execute attribute-by-attribute comparison.

As illustrated above, attribute-by-attribute-comparison can easily lead to intransitive preferences and, thus, explain why we cannot come up with a standard utility function for Bob in the market. However, if Bob fails to execute attribute-by-attribute-comparison because he lacks the pen and the notebook, he may resort to the mechanism we observed in the laboratory. For instance, without the notebook Bob may only be able to take one attribute into account. This can explain why we had no trouble assigning a utility function to Bob in the laboratory.

We now have different options to interpret this situation. Inspired by H&B-experimenters like Ariely et al., (2003), we could classify the notebook as an arbitrary influence on Bob’s choice-behavior. The presence of the notebook does not update Bob’s internal states in any substantial way. Moreover, it does not alter how Bob perceives the options and, therefore, does not lead to the triggering of different internal states. We could, therefore, hold that Bob does not really exhibit the relevant (intransitive) preferences in the market. At best, we can say that Bob has genuine preferences over different colors.

However, if we include the pen and the notebook into the constituents of Bob’s preferences, we can view the situations as Bob having different preferences in the two different environments in virtue of the presence or absence of the pen and notebook. Those features of his environment play an active role in determining his choice-behavior. They do not simply trigger different internal states or update them, but crucially feature in a back and forth between Bob’s internal states and his environment.Footnote 7

Hence, including the pen and the notebook into the constituents of Bob’s preferences would undermine the view of H&B-experimenters that agents do not genuinely exhibit certain preferences if their choices are malleable due to the presence or absence of environmental factors that cannot be understood as triggering different internal states or updating them (see also Bettman et al., 1998). If preferences can be partly constituted by environments, we can account for this kind of malleability by conceptualizing the agent as having different preferences in different environments in virtue of their distinct environmental constituents. Therefore, the influence of such environmental factors on choice-behavior does not necessarily undermine the attribution of genuine preferences.

In this regard, the commitment I assigned to experimental economics provides a clear point of tension with how the H&B-tradition uses the concept of preferences. Moreover, it also motivates a broader focus of experimental economics on environments that goes beyond studying how framing effect and choice-architectures trigger different preferences or how internal states are updated in the light of new inputs.Footnote 8

My toy example indeed reflects some real-world disagreement. Consider, for instance, Krupka & Weber (2013), two social preference researchers allied with the H&B-tradition, who want to explain agents’ behavior across several variants of the dictator-game via their preferences for norm compliance. To do so, they develop an incentivized method for eliciting norms with the help of simple coordination games. The rough idea is that subjects are rewarded if their own labeling of behavior as, for instance, “socially appropriate” or “socially inappropriate” matches the labeling of others. Krupka and Weber then rely on the idea of a general concern for the elicited norms (in combinations with a concern for money) to construct utility functions that are meant to capture choice-behavior of other subjects. Given their predictive success, they conclude that “subjects have a generally stable willingness to sacrifice money to take behaviors that are socially appropriate” (Krupka & Weber, 2013, p. 495).

Smith & Wilson (2019) focus on this work to highlight what they take to be wrong with social preference research. They argue that we cannot infer from the fact that one set of subjects labeled an action as “socially appropriate” or “socially inappropriate” that a general concern for norms should be incorporated into other agents’ preferences. Smith and Wilson hold that even if a norm is “is converted into a number and put on the right-hand side of a regression to explain the actual decisions of another set of subjects, that’s not causation/prediction. That’s correlation” (Smith & Wilson, 2019, p. 61). Yet, it is quite hard to see here what their general complaint is supposed to be. Surely, they are not just cautioning against hasty causal claims. So, what is the bigger picture?

Let us, therefore, turn to my proposed reconstruction. What Krupka and Weber seem to presuppose is that the influence of norms on choice-behavior necessarily must take the form of some internally stored concern for such norms. In this regard, they state that they simply “start with the assumption that individuals care about behaving in a manner consistent with social norms” (Krupka & Weber, 2013, p. 496). If we assume that constituents of preference are entirely located within the agent’s body, then having an internally stored concern for norms seems to be the most plausible way in which norms can be included into agents’ preferences. Under this assumption, no big leap of faith is required to infer that a general concern for norm compliance is part of the agents’ preferences based on the elicited norms and Krupka and Weber’s predictive success. How else should norms affect preferences?

However, under the assumption that constituents of preferences can also be located in an agent’s environment, this conclusion seems too hasty. Instead, it becomes possible that agents’ preferences for norm compliance (like in the case of Bob) only arise out of a back and forth between their internal states and the environment. That is, the agent only has the preferences whilst she is plugged into a specific environment. Under this alternative view, Krupka and Weber have indeed little reason to conclude that people’s preferences include a general concern for norm compliance. Instead, whether we should add the relevant term to their utility function could depend on specific features of the environments into which the agents are plugged in.Footnote 9 If we are not mindful of those environmental features, preference assignments, including those with potentially far-reaching implications like in the case of social preferences, could go easily wrong.Footnote 10 Hence, I take my reconstruction to be useful in elucidating a concrete disagreement between the two camps on how to interpret experimental results.

All in all, I hold that my reconstruction can successfully account for part of the disagreement between the camps. However, this claim presupposes, of course, that there is some initial plausibility to the commitment that the constituents of preferences can be partially located in an agent’s environment. Some may not be convinced of that and, therefore, hold that my reconstruction lacks plausibility. I grant that the relevant commitment can seem counterintuitive at first. Consequently, in the next section, I address this worry.

3 The Debate About Preferences

From here on, I will argue that the commitment I ascribe to experimental economics is indeed highly plausible. More specifically, I argue that the commitment can be plausibly defended under each of three major conceptions of what preference in economics are.

3.1 What Preferences Are

I start by outlining behaviorism, mentalism, and dispositionalism. This is done to show that even though there has been an extensive debate about what preference in economics are, this debate has remained silent regarding the commitments in virtue of which I have reconstructed the two paradigms, i.e., commitments about where (the constituents of) preferences are.

3.1.1 Behaviorism

According to behaviorism, preference relations are nothing more than concise descriptions of patterns in choice-behavior. For example, ascribing a preference X over Y to some agent is just to say that she would choose X in some circumstances, where Y is an available alternative (Clarke, 2016). Behaviorists make no ontological commitments apart from that a particular choice takes place in certain circumstances when assigning a preference to an agent.

Despite the long history of behaviorism in economics, also known as revealed preference theory (see Hands 2013), most philosophers nowadays reject it because of the following two arguments. The first argument highlights that we also need information on the agent’s beliefs to infer preferences from choice-behavior. Yet, if preferences were identical with choice, information on an agent’s choice alone would suffice to inform us about the agent’s preferences (Hausman 2011). The second argument holds that being identical with choice-behavior would prevent preferences from figuring into causal explanations of choice-behavior because an event cannot cause itself (cf. Vredenburgh, 2020). Yet, economists often refer to preferences when offering causal explanations of choice-behavior. While some rebuttals to these arguments have been put forward (e.g., Vredenburgh 2020, Thoma, 2021), most philosophers argue that we must identify preferences with something other than choice-behavior (e.g., Hausman 2011, Guala 2019).

3.1.2 Mentalism

The most common alternatives to behaviorism are mentalistic conceptions of preferences that identify them with some kind of mental state. There are several proposals concerning what kind of mental states preferences are. Some authors hold that preferences are (conscious) intentional states. Intentional states are mental states with propositional content. On this view, preferences are something like conditional intentions or conditional desires (see McDaniel & Bradley 2008). Others like Hausman (2011) understand preferences as total comparative evaluations. Total comparative evaluations are a kind of summary attitude that is formed by taking various beliefs and partial evaluations into account. Consequently, for Hausman, preferences in economics do not refer to a simple desire or intention but to the outcome of a sophisticated cognitive process. Moreover, some authors also advance specific views on what mental states in general are. Dietrich & List (2016) subscribe to functionalism. Functionalism holds that mental states should not be individuated by their intrinsic properties, but by the causal role they play in a certain system of inputs, outputs, and other mental states.

As this shows, there is no agreement within mentalism on what mental states are, let alone on what kinds of mental states preferences are. Yet, no matter the details, the following problem applies to all versions of mentalism (see Guala 2019). Economists frequently build choice-theoretic models that aim at describing entities we usually do not ascribe mental states to. For example, there are choice-theoretic models of mice and hermit crabs (e.g., Elwood & Appel 2009). More frequently, we find choice-theoretic models of firms, states, and political parties. To list a few examples, consider the Cournot-Nash model of firms in oligopolistic markets (see Mas-Colell et al., 1995) or the use of the Hotelling model to model a two-party system (Stokes, 1963). Yet, mentalism restricts the applicability of choice-theoretic models to entities with the relevant mental states. Consequently, so the argument goes, mentalism has trouble accounting for how economists employ the concept of preference.

3.1.3 Dispositionalism

Dispositionalism tries to combine the best of behaviorism and mentalism without their costs. According to dispositionalism, preferences are belief-dependent dispositions with multiply realizable supervenience bases (Guala, 2019). “Belief” here is meant to be a short-hand for an agent’s informational state. The supervenience base of a disposition is the set of (causal) properties that constitutes or instantiates the disposition. Roughly speaking, under dispositionalism, an agent prefers X over Y iff she is disposed (in a way that depends on her informational state) to choose X in some circumstance, where Y is also an available option. In contrast to behaviorism, ascribing a preference under dispositionalism incurs an ontological commitment regarding the existence of a particular disposition. Moreover, what is important for dispositionalism is that it assumes the supervenience bases of preferences to be multiply realizable, i.e., different sets of causal properties can constitute the same preference. Hence, entities with different causal properties like firms and humans can have the same preferences in the same circumstances. It, therefore, can account for the ascription of preferences to entities that lack mental states.

While the dispositional conception avoids many of the pitfalls of the other two conceptions (see Guala 2019 for details), it is also the least developed of the three accounts. For instance, it needs further explication how exactly preferences are supposed to be dependent on informational states. Moreover, one might worry that dispositionalism can hardly be separated from functionalism. Imagine that the causal roles for being a mental state of the type ‘preference’ would be so coarse-grained that any set of entities that is such that it constitutes the supervenience base of a preference under dispositionalism also realizes a mental state of the type ‘preference’ under functionalism. In this case, the two conceptions would ascribe the same preferences to the same entities. The only difference would be that functionalism would insist on calling them mental states (see also Clarke 2020).

3.2 The Where-Question

So far, I have outlined the main positions in the extant debate about preferences in economics. As should become apparent, the debate does not tell us much about the question of where the constituents of preferences are located. Yet, most of the literature tacitly sides with H&B-experiments in implicitly assuming that the constituents of preferences are entirely located within agents’ bodies. For instance, Guala (2019, p. 398) tells us that human preferences are “indeed mainly constituted by psychological mechanisms.” Moreover, Dietrich & List (2016) solely focus on neurophysiological processes when talking about the realizers of mental states. Hence, the commitment that I ascribed to H&B-experiments appears to be the commonsense position. Therefore, one might be skeptical of the plausibility of the commitment that I assigned to experimental economics.

Hence, I now argue that this commitment can be made plausible for all three conceptions of what preferences are. Within each of the next sub-sections, I focus on one of the three conceptions and demonstrate that my reconstruction can be plausibly defended under this conception.

3.2.1 Mentalism and the Where-Question

Let us start with mentalism. Initially, one might suspect that preferences – qua being mental states – cannot be constituted by environments. However, in contemporary philosophy of mind, Clark & Chalmers (1998) offer a forceful critique of the idea that mental states and cognitive processes are solely located within our bodies. Their so-called parity argument aims at establishing that we should not restrict mental states to being solely realized by events that take place within the human body because of the active role of our environments in cognition. They call this the extended mind hypothesis. I now show that, under mentalism, their argument can make plausible that the constituents of preferences are partially located in an agent’s environment. Their argument rests on the parity-principle, which can be stated as follows.

If an entity plays the same causal role as another entity that we already consider to be part of the constituents of some mental states or cognitive-process, we should also consider the first entity to be part of the constituents of a mental state or cognitive-process.

To support the parity-principle, Clark and Chalmers provide the example of Otto, who has Alzheimer’s. Otto heavily relies on a notebook to store and retrieve important information. Imagine Otto wants to visit his local history museum. He will consult his notebook, retrieve the information that the museum is located at 112 A-Caterer-Street, and walk to 112 A-Caterer-Street. We can explain Otto’s behavior by stating that he wants to go to the museum and believes that 112 A-Caterer-Street is the location of the museum. Now consider the case of Inga. Inga does not have the disease. She can successfully store a lot of information in her biological memory. Imagine she also wants to visit the history museum. She will consult her biological memory, retrieve the relevant information, and walk to 112 A-Caterer-Street. Like in the case of Otto, we can explain her behavior by stating that she wants to go to the museum and believes that 112 A-Caterer-Street is the location of the museum. The upshot of this example is that Otto’s notebook plays the same causal role in certain cognitive processes as Inga’s internal biological memory. Hence, according to Clark and Chalmers, if we classify Inga’s biological memory as the realizer of a mental state, we should also afford this classification to Otto’s notebook.

To see how the parity-principle applies to preferences under mentalism, recall the example of Bob. Here, the parity-principle implies that the pen and the notebook are part of the constituents of Bob’s preferences because the pen and the notebook play an essential causal role in bringing about Bob’s choice-behavior. In particular, they play the same role as storing the beating-relations in one’s biological memory.

It could be objected that the parity-principle varies with respect to its plausibility for different versions of mentalism. Clark and Chalmers are heavily influenced by functionalism when presenting their argument and it is often suggested that their argument depends on functionalism (e.g., Carter et al., 2016). However, given that the application of choice-theory to agents like firms implies that (coarse-grained) functionalism is the most plausible version of mentalism anyway, these worries do not have to bother us here. Under the most plausible version of mentalism, there is a viable strategy to articulate the idea that the constituents of preferences are partially located in the agent’s environment.

There is a vast literature on the status of the parity-principle (for a good overview see Menary 2010). Building on this literature one could, of course, argue against the plausibility of the idea that preferences as mental states can be partially realized by an agent’s environment. It could, for instance, be argued that a relevant difference between biological memory and the pen and notebook is that biological memory cannot be as easily decoupled from the agent as the pen and the notebook. Chalmers and Clark would respond here that even biological memory is sometimes decoupled from the agent, e.g., when the agent is asleep or knocked out. No matter how successful one thinks this rebuttal is in general, the corresponding objection is not very troublesome if we are concerned with preferences in economics. Economists want to model specific social interactions. Therefore, it does not matter whether the notebook is always accessible, but only whether the agent has access to the notebook in the kind of choice-situations that are the target of our model.

Assessing the plausibility of various other arguments surrounding the parity-principle would obviously go beyond the scope of this paper. However, my discussion already highlights that it has an initial plausibility in the case of preferences. Hence, under mentalism, my reconstruction of the disagreement between the two paradigms can be plausibly stated as experimental economists accepting the extended mind hypothesis for preferences, while H&B-experimenters reject it.Footnote 11

3.2.2 Dispositionalism and the Where-Question

Let me now show how the relevant commitment can be made plausible under dispositionalism. To do so, I introduce a co-opted version of the parity-principle that applies to dispositionalism:

If an entity plays the same causal role in bringing about choice-behavior as another entity that we already consider to constitute part of the supervenience base of some preference, we should also consider the first entity to constitute part of the supervenience base of a preference.

The co-opted parity-principle is not about cognitive processes or mental states. Consequently, it applies to preferences independently of whether they are mental states. Under dispositionalism, applying the co-opted parity-principle to the example of Bob gives us the same result as applying the parity-principle under mentalism. Hence, dispositionalism and the co-opted parity-principle allow us to hold that the constituents of preferences (i.e., their supervenience base) can be partially located in the agents’ environment.

One could object that I conflate the supervenience base of a disposition with the mechanisms that produce choice. While a supervenience base is the set of properties that, given certain background conditions and the relevant laws of nature, causally suffices for the manifestation of the disposition, the mechanism encompasses everything that leads to the production of choice-behavior (i.e., background conditions and laws of nature). However, what counts as merely part of the mechanisms (e.g., initial conditions) and what should count as the constitutive supervenience base is exactly what is at stake here. If we say Bob has intransitive preferences over the options in my example and we think that preferences are dispositions, we have to include the environment into the supervenience base of the disposition. If we decide that the environment is only part of the mechanism that produces choice-behavior, we could not attribute these preferences to Bob.

The objection may be fueled by the view that the supervenience base of a disposition must be entirely constituted by intrinsic properties. Lewis (1986, p. 61) distinguishes “intrinsic properties, which things have in virtue of the way they themselves are, from extrinsic properties, which they have in virtue of their relations or lack of relations to other things.” While he holds that the supervenience base of a disposition must be entirely constituted by intrinsic properties (cf. Lewis 1997), other authors like McKitrick (2003) hold that the supervenience bases can include extrinsic properties. An example of a plausible extrinsic disposition is the disposition of a key to open a door. It only has this disposition because of the properties of the lock. Similarly, Bob’s intransitive preferences would be an extrinsic disposition he only possesses because of features of his environment. Hence, under dispositionalism, my reconstruction of the disagreement between the two paradigms can be plausibly stated as experimental economists accepting that there are certain extrinsic dispositions, while H&B-experimenters reject this idea.

3.2.3 Behaviorism and the Where-Question

The case of behaviorism must be approached in a slightly different way. Behaviorists identify preferences directly with choice. Hence, under behaviorism, the where-question cannot pertain to constituents because preferences do not stand in a constitutive relationship to realizers or supervenience bases. Nevertheless, we can formulate a version of the where-question for behaviorism.

Under behaviorism, we usually have some leeway in terms of how we should individuate agents’ preferences, i.e., which preferences we assign to agents. Given this leeway, even a behaviorist may wish to argue that we should be wary of preference assignments that make their shapes crucially dependent on certain features of the environment.

To better appreciate what is at stake here, consider the following example inspired by Thoma (2021): if we observe someone eating a spoon of avocado cream, we have several ways of characterizing this choice. We could assign a ‘preference for avocado cream (over the other available options)’ or a ‘preference for green paste (over the other available options).’ Prima facie, both options seem legitimate under behaviorism. Yet, these assignments should not be treated as equally legitimate. Just imagine that we go for the ‘green paste’ option and, therefore, predict that the person will eat a full spoon of wasabi if we present her with a menu that swaps avocado cream with wasabi. In this instance, our prediction is likely to go wrong. Hence, not every way of individuating an agent’s preferences is equally acceptable under behaviorism. So, how exactly should we individuate preferences?

Thoma (2021) argues that we can solely focus on an agent’s beliefs to decide how to individuate preferences and black-box the “motivating factors” behind choice. She holds that this allows us to retain behaviorism. More precisely, she proposes a limited mentalism about beliefs according to which descriptions of options should be consistent with the agent’s beliefs and should include any feature for which “there are choice situations where the agent would make a different choice when she believes that feature is present or absent respectively” (Thoma, 2021, p. 173). As this demonstrates, plausible versions of behaviorism need to put restrictions on legitimate preference assignments.

In this spirit, even a behaviorist could agree with proponents of the H&B-tradition and argue that we should not assign preferences whose shapes crucially depends on what they consider arbitrary environmental influences. To give a simple example, if we could describe an agent’s choices or preferences in such a way that they are transitive, but the transitivity under this description would crucially depend on the presence of certain environmental features, we should not describe an agent’s preferences in this way.

Consequently, even a behaviorist can hold that certain environmental influences should make us wary about the assignment of certain preferences. Yet, other behaviorists might see no problem in describing an agent’s choices in a way that makes them dependent on the presence of those environmental factors. They might, for instance, hold that relevant patterns in choice-behavior are such that we need to assign preferences whose features crucially dependent on the presence of those environmental factors. Under my reconstruction of the two experimental paradigms, the second brand of behaviorist would have reason to side with experimental economics, while the first brand may side with the H&B-tradition. Consequently, the where-question is also relevant under behaviorism. Under behaviorism, my reconstruction of the disagreement between the two paradigms can be plausibly restated as a disagreement about how to individuate preferences.Footnote 12

To sum up Sect. 3, the commitment about preferences I ascribed to experimental economics can be made plausible under mentalism, dispositionalism, and behaviorism. This, in turn, supports the plausibility of my reconstruction of the two paradigms.

Moreover, I take my argument to highlight a new dimension in the debate about preferences in economics. The debate has only focused on what preferences are. Yet, my argument indicates that it should also focus on where (the constituents of) preferences are.

4 Objections

I will now address the following objections: my reconstruction strays too far from ‘commonsense’ (4.1), and experimental economics does not take preferences as seriously as the H&B-tradition (4.2).

4.1 Common Sense

One could argue that my proposed reconstruction of experimental economics breaks too much with commonsense to be acceptable to economists. The idea that constituents of preferences can be partially located in an agent’s environment stretches the concept beyond what economists would be willing to accept.

My reply follows Petracca & Gallagher (2020), who urge the application of extended and distributed cognition to the study of institutions and argue that those ideas – while prima facie clashing with commonsense – can, in fact, help us to make sense of everyday ways of thinking. In the present context, including an agent’s environment into the constituents of her preferences can help us, for instance, to scrutinize the meaning of the frequently encountered, yet usually vaguely specified claim of experimental economists that institutional features of markets account for the successful application of choice-theoretic models. That is, we can now further specify that they can do so in virtue of constituting parts of the preferences those models assign to agents. Hence, a departure from commonsense seems worth its costs.

Yet, one could argue that my reconstruction stretches commonsense unnecessarily. I have already outlined that behaviorist eschew talk of constituents. However, in the case of mentalism and dispositionalism, one may still insist that constituents of preferences are entirely located within the agent’s body. Consequently, one may hold that a reconstruction of the difference between the two experimental paradigms should assume that constituents of preferences are always located inside the agent’s body if such an interpretation is available. The difference between the two paradigms, so the argument could then continue, could also be interpreted merely in terms of how malleable we permit certain internal states to be to outside influences.

However, I hold that this way of looking at the issue tends to obscure rather than illuminate it. Reconsider the case of Bob. Of course, one could say that what really constitutes his preferences for, let us say, X over Y is just his inclination to choose the option that has more strokes in his notebook together with his internal representation of ‘option X having more strokes in the notebook than option Y.’ Assuming that Bob’s inclination to choose in line with his notebook is not itself dependent on environmental factors, we could then say that Bob’s preference is entirely constituted by internal factors. However, this perspective would totally obscure the disagreement about whether we can genuinely assign a preference for X over Y to Bob in cases where his decision-making is crucially dependent on a back and forth between his internal states and his environment.Footnote 13 That is, it would distract from the fact that some would say that Bob does not really have a preference for X over Y because of the influence of his environment, while others wish to say that he does so precisely because of his environment. Hence, I maintain that framing the issue in terms of the location of the constituents of preferences is more useful for highlighting a source of disagreement.

4.2 Taking Preferences Seriously

While it is most common to present preferences as playing an essential role in choice-theoretic explanations (e.g., Mas-Colell et al., 1995, Hausman 2011, Guala 2019), some authors hold that preferences are not part of the explanatory framework of those models. Those authors hold that, even though choice-theoretic models assign utility function and preferences to agents, what really does the explanatory work are the incentives and constraints embedded in the agents’ environment. However, the model’s assumptions about agents’ preferences, so the argument goes, can be safely bracketed (e.g., Alchian 1950). Hence, one may wonder why I did not attempt to reconstruct the difference between the two paradigms along those lines. On this alternative view, experimental economists are those who want to explain via reference to incentives embedded in environments, while the H&B-tradition wants to explain via reference to (internal) preferences.

First, this alternative view – sometimes called externalism about choice-theory (e.g., Satz & Ferejohn 1994, see also Binmore 2010) – has been extensively criticized by Hausman (1995). He argues that it is just another iteration of the narrow instrumentalism frequently attributed to Friedman (1953). Narrow instrumentalism holds that we should only be interested in a subset of the implications of economic models concerning the phenomena one is interested in. Yet, apparent falsities implied by the model that do not concern those phenomena can be ignored. I agree with Hausman that narrow instrumentalism is bad methodological advice. Bracketing a model’s assumptions on an agent’s preferences threatens to obscure the conditions under which certain incentives and constraints would lead to the choices the model is meant to predict or explain (see Hausman 1994). Second, incentives obviously play an important role in explaining choice-behavior. However, when it comes to conceptualizing incentives, it appears that some (implicit) reference to something like preferences is required even in the case of entities like firms.

Moreover, while Satz & Ferejohn (1994) – in their seminal proposal of externalism – suggest that a model’s assumptions about preferences can be bracketed, they also state that externalism relies on a social theory of ‘interests’ or ‘preferences’ which need to be “constructed out of the social environment.“ At a first glance, this appears contradictory. Should we care about what the model says about preferences or not? Prompted by his commitment to (non-extended) mentalism, Hausman (1995) opts for the reading that externalists think that we should not. However, a more charitable reading would be that externalism does not care much about the preferences we would ascribe to agents if we were to construe them as internal mental states. Yet, we should care about preferences construed as partially constituted by (social) environments. Viewed like this, committing to the idea that preferences can be partially located in an agent’s environment points us towards a realist reading of externalism about choice-theory that evades Hausman’s critique of narrow instrumentalism. Hence, since my proposed reconstruction avoids pushing experimental economics into the narrow instrumentalist corner, it gains an advantage over construing the disagreement along the incentives vs. preference divide.Footnote 14

5 Conclusions

In this paper, I have offered a novel reconstruction of the disagreement between the two experimental paradigms and, thereby, highlighted a neglected dimension in the philosophical debate about preferences in economics. Concerning the latter contribution, my argument indicates that philosophers of economics should broaden the set of questions they ask about preferences and draw on a wider set of philosophical tools for answering them. By relying on a wider array of ideas instead of only debating what preferences really are, the debate about preferences stands to improve its relevance for economics practice (cf. Angner, 2018, Clarke 2020). For instance, making the option that preferences can be partially constituted by environments salient raises novel questions about the robustness of social preferences and experimental design that experimentalist should explicitly grapple with (cf. Woodward, 2008). In this regard, it may become less interesting to ask whether agents really have social preferences and more important to consider in which environments they have them.

Concerning the first contribution, I take my argument to provide H&B-experimenters and experimental economists with novel resources for conceptualizing their disagreement. This may allow the two camps to realize that part of their disagreement is located at the level of commitments that are not immediately up for testing. Moreover, pinning down this source of disagreement and pointing towards the philosophical debates surrounding it will ideally help to reach a resolution. Situating the relevant philosophical debates within the context of the concrete epistemic aims of economists, thus, bears great promises for future research.