1. Introduction
An increasingly popular interdisciplinary area of research concerns the question of how non-human animals (“animals”) make economic decisions. While there is no question that this is an inherently interesting and important area of research (Santos and Rosati Reference Santos and Rosati2015; Kalenscher and van Wingerden Reference Kalenscher and van Wingerden2011), the study of animal decision-making is also said to be of use in economics (for humans). So, Kalenscher and van Wingerden (Reference Kalenscher and van Wingerden2011, 8) note that “comparative research [can] uncover inconsistencies in choice behavior between humans and animals that allow for an improved, more comprehensive description of choice behavior and possibly force us to re-think the basis of economic theory in the light of the evolutionary roots of choice.” Other authors, too, argue for the appeal to data on animal economic decision-making in the investigation of economic questions (Santos and Chen Reference Santos, Chen, Paul, Camerer, Fehr and Poldrack2009; Brosnan et al. Reference Brosnan, Grady, Lambeth, Schapiro and Beran2008; Glimcher et al. Reference Glimcher, Dorris and Bayer2005). It is this economic-focused use of data on animal decision-making that is the focus of this paper.
However, it turns out that establishing exactly what role such data can play in debates in contemporary economics is not as straightforward as it first may appear. In particular, this paper shows that this role does not concern the corroboration of economic theories of choice—as might have been expected, and as is suggested by some of the above authors (including Kalenscher and van Wingerden Reference Kalenscher and van Wingerden2011). Rather, it concerns the delineation of the proper domains of application of different economic models of choice. The paper goes on to show that, in order to play this role, data on animal economic decision-making need to be embedded in detailed phylogenetic comparative analyses.
To make this clearer, the paper begins, in section 2, by laying out the nature of the (for present purposes) key contemporary economic debates concerning decision-making. Section 3 presents some of the major data on animal economic decision-making. Section 4 details the structure of evolutionary biological phylogenetic comparative analyses, and uses this structure to develop a compelling methodology for linking the data on animal economic decisions from section 3 to the debate described in section 2. Section 5 concludes.
2. Economics and decision-making
While it is uncontroversial that one of the major areas of inquiry in economics is human decision-making, it is not uncontroversial how this study is to be conducted. Two such controversies are particularly important to mention here.
First, there is the question of whether economics should study all kinds of human decision-making (see, e.g., Robbins Reference Robbins1932; Fumagalli Reference Fumagalli2016b), or just a subset of this—e.g., and most famously, merely choices resulting from a desire for wealth (and only in so far as these choices stem from this desire or its opposite, the desires for leisure and consumption) (Mill [1836] Reference Mill and John1967). While there may well be reasons to sometimes treat the economic study of choice as having a narrow domain, in this context, it is the former, broader perspective that is relevant.
On the one hand, this is inherently compelling; after all, much contemporary work in economics, either explicitly or implicitly, does take on this broader perspective (Fumagalli Reference Fumagalli2016b). On the other hand, reading economics narrowly would greatly complicate appealing to data from animals in economics—but in a way that is perpendicular to the issues at stake here. In particular, we would need to identify an evidentially relevant alternative trait to (e.g.) a desire for wealth in animals (such as a motivation to acquire status, mates, or offspring), and then use this trait to bridge the human and non-human cases. This, though, concerns an entirely separate suite of considerations, and can be left for another occasion. Instead, the present paper adopts a broad view of economics, according to which economics studies how agents—whether human or not—allocate scarce resources among competing ends (Robbins Reference Robbins1932). Whether the arguments of the paper carry over to a narrower reading of economics will not be further considered.Footnote 1 That said, the fact that the paper adopts this broad view should not be taken to imply that it assumes that a particular economic model of choice applies to some group of animals. On the contrary, as will be made clearer below, determining whether or not a given model is a useful way to approach the behavior of some animal is a key part of the proposal defended in this paper.
Somewhat related points hold for the second major area of controversy concerning the economic study of choice: whether economics studies the decisions that humans do or would make in various circumstances (including hypothetical ones), or whether it studies the decisions that it would be rational for humans to make in these circumstances. In what follows, it will be the first, descriptive and non-rational, reading that will be central.
On the one hand, this is again inherently plausible. While rationality is undoubtedly central to much work in economics (see, e.g., Robbins Reference Robbins1932; Hausman Reference Hausman2012), it should not be seen as being central to all such work. Much of economics tries to determine how agents actually make economic decisions (see, e.g., Kahneman and Thaler Reference Kahneman and Thaler1991; Thaler Reference Thaler2000, Reference Thaler1980; Glimcher et al. Reference Glimcher, Dorris and Bayer2005; Rabin Reference Rabin1998; Hutchinson and Gigerenzer Reference Hutchinson and Gigerenzer2005, 105, 110–11, 119; Fehr and Camerer Reference Fehr and Camerer2007; Camerer Reference Camerer2007). For example, Thaler (Reference Thaler2000, 137) writes: “[T]o attempt richer characterizations of economic agents via a better understanding of human cognition […] will be a major area of effort over the next two decades.”
On the other hand, even if economics is seen as providing a sort of “logic of choice” (Jeffrey Reference Jeffrey1983; Savage Reference Savage1954; Robbins Reference Robbins1932), then this would indeed complicate efforts to appeal to data from animals in economics. For example, we would need to find a way of bridging the is/ought gap.Footnote 2 However, these complications would again be in addition to the issues raised in this paper: even if a particular way of bridging this gap is accepted, it would still need to be shown how the data from animals can be related to the human case. That is, even if we can show that what humans ought to do is related to what they in fact do, it is not obvious how this relates to what animals do. The goal in what follows is to determine how we can relate the psychology and behavior of animals to humans. If and how to use this relation to ascertain appropriate standards of rationality can be left for another occasion.Footnote 3
Given this view of economics as the descriptive study of all kinds of (human) decision-making, a major area of dispute and investigation concerns the fact that there is a multitude of accounts of decision-making that have been developed, each of which has some empirical support. Some of the major candidates include the following (many further such theories could be mentioned, and each of the above theories comes in several different versions as well; for good overviews, see, e.g., Heap et al. Reference Heap, Lyons, Hollis, Sugden and Weale1992; Resnik Reference Resnik1987; Bradley Reference Bradley2017; Angner Reference Angner2016):Footnote 4
Expected utility theory (EUT): Agents are assumed to maximize the expected utility they obtain upon completing action A i in state S i , weighted by the probability of S i : max[EU(A k ) = $\sum\nolimits_{i = 1}^n {{\rm{P}}(} {S_i})\;{{u}}({A_k}\ {{ \& }}\ {{{S}}_{\rm{i}}}) $ ], with P(x) being a probability function ranging over S and u(x) a (set of positive affine transformations of a) real-valued utility function ranging over (A × S) (see, e.g., Jeffrey Reference Jeffrey1983; Savage Reference Savage1954; Hausman Reference Hausman2012; Resnik Reference Resnik1987).
Prospect theory (PT): Agents are assumed to maximize their expected prospects—i.e., the change in the status quo—upon completing action A i in state S i , weighted by the decision weight of S i (which is partially similar to the probability of S i , but diverges towards the extrema of the probability function): max[EP(A k ) = $\mathop \sum \nolimits_{i = 1}^n {\rm{\pi }}\;{\rm{(}}{S_i}{\rm{)\;}}{v_C}{\rm{(}}{A_k}{{\;\& \;}}{S_i}{{)}}$ ], with π(x) being a decision weight function ranging over S and v C (x) a value function ranging over (A × S) that is concave in gains and convex in losses (see, e.g., Barberis Reference Barberis2013; Kőszegi and Rabin Reference Kőszegi and Rabin2006; Tversky and Kahneman Reference Tversky and Kahneman1992; Kahneman et al. Reference Kahneman, Slovic and Tversky1982; Kahneman and Tversky Reference Kahneman and Tversky1979).
Regret theory (RT): Agents are assumed to minimize the expected regret—i.e., the utility they counterfactually could have gotten, if they had chosen differently—upon completing action A i in state S i , weighted by the probability of S i : ER(A k , A l ) = $\mathop \sum \nolimits_{i = 1}^n {\rm{P}}\;{\rm{(}}{S_i}{{)\;Q[u}}({A_k}{{\;\&\;}}{S_i}{\rm{)\; - \;u(}}{A_l}{\rm{\;\&\;}}{S_i}{\rm{)]}}$ , with P(x) being a probability function ranging over S and Q(x) a regret function ranging over (A × S) that is symmetric, convex for positive values of Q(x), and concave for regrets (among other features) (Bleichrodt and Wakker Reference Bleichrodt and Wakker2015; Loomes and Sugden Reference Loomes and Sugden1982, Reference Loomes and Sugden1987).
Theory of simple heuristics (SH): Agents are assumed to rely on many rules that are domain-specific, easy to apply, and satisficing, such as “Take the Best,” where agents make decisions by focusing on just one, predictively highly relevant, decision attribute, or “Take the First,” where agents make decisions by considering only the first feasible option that comes to mind (among many other such rules) (Gigerenzer and Selten Reference Gigerenzer and Selten2001; Todd Reference Todd, Gigerenzer and Selten2001; Gigerenzer Reference Gigerenzer2007; Lieder and Griffiths Reference Lieder and Griffiths2020).
There has been much discussion of many aspects of these theories. For present purposes, though, the issue at center stage is the fact that all of these accounts have some empirical evidence that speaks in their favor. While a full review of this evidence goes beyond the confines of this paper (and is not necessary here either), some key data points include the following.
-
• EUT has generally been found to be an accurate account in many situations where people have time to learn about the payoffs and adjust their behavior accordingly (see, e.g., Binmore Reference Binmore2007; Glimcher et al. Reference Glimcher, Dorris and Bayer2005). In such cases, people tend to act in ways that are captured well by a model that assumes they probabilistically evaluate the outcome space, and then look for the optimal value in this space.
-
• As shown by the behavioral economic revolution, there are also many cases in which people diverge from the predictions of EUT and instead act in ways that are well predicted by PT (Kahneman and Tversky Reference Kahneman and Tversky1979; Kahneman et al. Reference Kahneman, Slovic and Tversky1982; Kahneman and Thaler Reference Kahneman and Thaler1991; Barberis Reference Barberis2013). Some of the well-known findings here concern the documentation of framing effects (where choices framed as losses are treated differently from choices framed as gains) and endowment effects (where the evaluation of an item an agent has in their possession differs from that of the same item when not in possession).
-
• In a related body of work, neuroscientific and behavioral data indicate that sometimes people anticipate the regret/rejoicing they feel from different actions, and then act accordingly (Zeelenberg Reference Zeelenberg1999; Coricelli et al. Reference Coricelli, Dolan and Sirigu2007). There are also considerable empirical data showing that people often act intransitively (Bleichrodt and Wakker Reference Bleichrodt and Wakker2015; Kalenscher et al. Reference Kalenscher, Tobler, Huijbers, Daselaar and Pennartz2010; Tsai and Bockenholt Reference Tsai and Böckenholt2006; Sopher and Gigliotti Reference Sopher and Gigliotti1993; Loomes et al. Reference Loomes, Starmer and Sugden1991), which is predicted by RT, given that it conceptualizes decision-making as dependent on what the contrast class is against which a given action is evaluated (Loomes and Sugden Reference Loomes and Sugden1987, Reference Loomes and Sugden1982; Loomes et al. Reference Loomes, Starmer and Sugden1991).
-
• SH draws support from classic cognitive scientific data (Gigerenzer and Selten Reference Gigerenzer and Selten2001; Gigerenzer et al. Reference Gigerenzer and Todd2000; Kruglanski and Gigerenzer Reference Kruglanski and Gigerenzer2011; Lieder and Griffiths Reference Lieder and Griffiths2020), including the fact that in various decision-making tasks, people often forgo using all of the available information and are instead drawn particularly to decision options that they are familiar with (Henrich and McElreath Reference Henrich, McElreath, Dunbar and Barrett2007; Goldstein and Gigerenzer Reference Goldstein and Gigerenzer2002), or that they often use a lexicographic hierarchy of simple cues and stop assessing options once a cue is found that discriminates among them (Dhami Reference Dhami2003; Broder Reference Broder2000).
When considering these data, it is important to note that the issue here is not one of accommodation, but prediction. All of the above models can accommodate all the data; however, they do not all predict all of it (Hitchcock and Sober Reference Hitchcock and Sober2004).Footnote 5 While there is nothing in EUT that rules out the possibility of endowment effects, intransitive choices, or familiarity-based decision-making, there is also nothing in the theory that should lead us to expect such findings. The same goes for the other theories and findings. Put differently, any of the above theories can be retrospectively fitted to all of the above data, but different such data are differentially well predicted by the different theories ex ante.Footnote 6 It is in this sense that the different accounts are empirically well supported in different contexts.
This leads to the major point that needs to be emphasized concerning the dispute about economic decision-making. The continued empirical success of all these different decision models suggests that it is a serious possibility that they need to all be taken seriously. Of course, it is possible that new data will turn up that clearly show one of them to be superior to the rest, or that a new account is developed that is an improvement over all of the above. However, given that the above accounts have had empirical successes for several decades, it is looking ever more plausible that neither of these options will come to pass. Instead, we may need to accept that all of these accounts are here to stay. Put differently, it increasingly looks like Rabin and Thaler (Reference Rabin and Thaler2001, 230) were wrong when they said, some 20 years ago: “it is time for economists to recognize that expected utility is an ex-hypothesis, so that we can concentrate our energies on the important task of developing better descriptive models of choice under uncertainty.” On the contrary: unlike Monty Python’s ex-parrot, EUT is still very much alive. However—and this is the key point here—the same is true of PT, RT, and SH.
How can this be? One way is as follows. EUT, PT, RT, and SH need to be recognized as models (despite their name!) (Morrison and Morgan Reference Morrison, Morgan, Morgan and Morrison1999; Morgan Reference Morgan2012; Weisberg Reference Weisberg2013; Morrison Reference Morrison2015). These different accounts all get at some aspect of human economic decision-making, and thus have some predictive successes. In turn, this is due to the fact that human economic decision-making is a multifaceted trait: we sometimes do something like maximizing EUT, sometimes something like maximizing EP, sometimes something like minimizing regrets, and sometimes we rely on something like SH.Footnote 7 Because of this, the last two decades of work in neuroeconomics, behavioral economics, and cognitive science show support for all of EUT, PT, RT, and SH: they all get something right about how we make decisions—though none of them get all of it right.
This kind of absence of “one right account” of the phenomenon is characteristic of model-based methodologies in many contemporary sciences (see also Fumagalli Reference Fumagalli2016a). So, for example, in evolutionary biology, the question is not whether a phenotypic selection-based model, or a population genetic model, or a molecular genetic model, or an evo-devo model (etc.) is the right model of the evolutionary dynamics of some population. Each of these models foregrounds different aspects of these dynamics—natural selection, laws of genetic segregation and assortment, development, etc.—and as such, they can all be good models (Potochnik Reference Potochnik2017). They are simply different elements of our tool kit with which we can investigate the underlying biological reality, depending on which aspect of that reality we are looking to explore further (Morrison and Morgan Reference Morrison, Morgan, Morgan and Morrison1999; Morgan Reference Morgan2012; Weisberg Reference Weisberg2013; Morrison Reference Morrison2015). The same is true here: EUT, PT, RT, and SH (among others) all get at different aspects of the complex psychological reality underlying economic decision-making. They foreground different aspects of this reality—the calculation of expected utilities, weighted prospects, etc.—and they should all be seen to be part of the contemporary economic toolkit (Hochstein Reference Hochstein2022). This has two important implications.
On the one hand, pragmatic considerations such as analytical convenience or data availability are an important determinant of the choice of models (Potochnik Reference Potochnik2017; Levins Reference Levins1966; Orzack and Sober Reference Orzack and Sober1993; Weisberg Reference Weisberg2006; Morrison Reference Morrison2015; Samuelson Reference Samuelson2002; Massimi Reference Massimi2018; Parker Reference Parker2020). The present case is no exception (Binmore Reference Binmore2007). On the other hand, though, this does not mean that pragmatic considerations should be assumed to be all that is going on here. The fact that EUT, PT, RT, and SH are recognized to be models does not mean that we should take an “anything goes” purely instrumentalist attitude here (Friedman Reference Friedman and Friedman1953; Gul and Pesendorfer Reference Gul, Pesendorfer, Caplin and Shotter2008; Hausman Reference Hausman and Hausman2008). In particular, there is also an important set of representational questions to be asked. These questions concern the situations in which a given model foregrounds the right causal factors. That is, we want to know the proper domains of each of the models: which of the available models of choice are useful when, and why is this the case? Why do we make PT-like decisions here, and EUT-like decisions there? How can we characterize “here” and “there?” While the choice of a model partially depends on pragmatic factors and our interest, it also depends on the world: foregrounding cause C1 will not be empirically compelling if C1 is not a significant cause in the case at hand.
This is again similar to other model-based inquiries: while replicator-dynamics-based, population genetic, and evo-devo models all have their place in evolutionary biology, which is partly determined by pragmatic factors, this does not exhaust the debate there. These different models fore- and background different causal factors, and so it is also interesting to ask which empirical situations are well modeled with what (Potochnik Reference Potochnik2015, Reference Potochnik2010, Reference Potochnik2017). Which evolutionary scenarios feature few genetic and developmental constraints—making purely selection-based models adequate—and which feature prominent genetic or developmental processes that should be foregrounded—making other models representationally superior? The same is true in economics: the question of which situations feature mostly EUT-like decision-making, which mostly SH-like decision-making, etc., is an important, and so far unanswered, question.
Note that this question cross-cuts the familiar instrumentalism/realism debate surrounding economic models of choice (see, e.g., Friedman Reference Friedman and Friedman1953; Gul and Pesendorfer Reference Gul, Pesendorfer, Caplin and Shotter2008; Angner Reference Angner2018; Hausman Reference Hausman and Hausman2008). The question here is not whether we should see economic models as being psychologically neutral, purely behaviorist predictive devices, or as psychologically realistic descriptions of our neurocognitive architecture (though this is an interesting question, too: Fumagalli Reference Fumagalli2016a). Rather, the question here is how to best model the neuro-psychological mechanisms underlying economic decision-making, while acknowledging that none of the above models can be expected to be fully accurate, and yet still making room for these models to be differentially empirically compelling in different contexts.Footnote 8
Note also that answering this question is not straightforward, especially in economics, where there are many experimental constraints. For example, it is hard to experimentally test the idea that EUT-like decision-making is particularly common in cases where the stakes are high and time constraints low: for ethical reasons, it is hard to induce truly high-stakes decision situations in a lab. More than that: even if it were determined that a given model is a good fit to situations A, B, and C, it is not obvious how to step back and assess what these situations have in common (other than that they fit the parameters of the model in question). What do the many cases where EUT-like decision-making is observed have in common?
It is therefore in this context that the (for present purposes) key contemporary dispute surrounding economic decision-making is located. Once it is recognized that all of EUT, PT, RT, and SH (etc.) are in-principle adequate models, and that the choice between them is context-dependent and not to be made once and for all, it becomes clear that, from a representational perspective, the main issue to be settled is to understand their individual causal domains better. It is one thing to know that there are many different ways to model economic decision-making in different circumstances. It is another thing to know which situations are empirically particularly well-modeled with EUT, which with PT, and so on. Determining this is an important task, and one that does justice to both the model-based nature of EUT, PT, RT, and SH and the non-purely instrumentalist nature of economics.
3. Animal economic decision-making
Over the last few years, researchers have used variants of the kinds of choice situations typically used in economic experiments to investigate the ways in which a wide variety of animals make decisions. For example, animals have been given choices between safer and riskier options, between investing in value streams with different reward patterns, and between different goods when they had control over them and when not (good overviews are here: Santos and Rosati Reference Santos and Rosati2015; Kalenscher and van Wingerden Reference Kalenscher and van Wingerden2011; Brosnan et al. Reference Brosnan, Grady, Lambeth, Schapiro and Beran2008). While there is much that can be said about this work, for present purposes, the major point to note is that a key overall upshot of it is the finding that many of the same kinds of results familiar from human studies can also be found in animals. (Some more details about some of these studies will be noted in the next section.)Footnote 9
So, a number of different animals, including macaques, rats, and pigeons, have been shown to adjust their behavior in ways that is well predicted by assuming they are assessing the expected value of the relevant actions’ consequences (Kagel et al. Reference Kagel, Battalio and Green1995; Santos and Chen Reference Santos, Chen, Paul, Camerer, Fehr and Poldrack2009; Glimcher et al. Reference Glimcher, Dorris and Bayer2005). For example, Glimcher et al. (Reference Glimcher, Dorris and Bayer2005, 230) summarize some cognitive neuroscientific studies of decision-making in macaques by noting that “something very close to an economic choice variable was indeed being carried by the firing rates of these neurons [in the LIP area of macaque brains]. […] Together, these results suggested an interesting possibility, that the topographic map in area LIP encodes something like the relative expected value, or perhaps even the relative expected utility, of each possible eye movement under the conditions we had been studying.” However, this is not all that has been shown concerning non-human economic decision-making.
It is also the case that that a highly varied group of animals comprising (among others) chimpanzees, gorillas, orangutans, Capuchin monkeys, rats, honeybees, and starlings has been shown to be subject to framing effects (Chen et al. Reference Chen, Lakshminarayanan and Santos2006; Brosnan et al. Reference Brosnan, Grady, Lambeth, Schapiro and Beran2008; Kacelnik and Bateson Reference Kacelnik and Bateson1996; MacDonald et al. Reference MacDonald, Kagel and Battalio1991; Marsh and Kacelnik Reference Marsh and Kacelnik2002; Shafir et al. Reference Shafir, Reich, Tsur, Erev and Lotem2008). A significantly smaller group of these animals—a fact that will become very important below—has even been shown to display an endowment effect; these animals include chimps, gorillas, orangutans, and capuchin monkeys (Lakshminarayanan et al. Reference Lakshminarayanan, Chen and Santos2008; Brosnan et al. Reference Brosnan, Jones, Lambeth, Mareno, Richardson and Schapiro2007; Drayton et al. Reference Drayton, Brosnan, Carrigan and Stoinski2013; Flemming et al. Reference Flemming, Jones, Mayo, Stoinski and Brosnan2012).
Even this does not exhaust the findings in this context, though. Other studies have found evidence of RT-like decision-making—including intransitive decision-making—in macaques, chimpanzees, bonobos, grey jays, and honeybees (Waite Reference Waite2001; Shafir Reference Shafir1994; Rosati and Hare Reference Rosati and Hare2013; Lee et al. Reference Lee, McGreevy and Barraclough2005; see also Hayden et al. Reference Hayden, Pearson and Platt2009; Abe and Lee Reference Abe and Lee2011). On top of this, much work in behavioral ecology and cognitive ethology is based on the fact that many animals act in ways that suggest reliance on simple heuristics, such as choosing what is familiar instead of engaging in exhaustive search of the option space, or picking the first option that hits a minimal threshold in a fixed hierarchy of cues (Goulson Reference Goulson2000; Karsai and Penzes Reference Karsai and Zsolt Pézes2000; Shettleworth Reference Shettleworth2009; Gibson Reference Gibson1996; Kacelnik Reference Kacelnik, Hammerstein and Stevens2012).
Now, there is no question that the details of the above studies (and others like them) deserve to be closely scrutinized. Did the studies control for all the necessary confounders? Are the conclusions drawn by the studies’ authors warranted, or are other conclusions more reasonable? However, these questions can be sidestepped here. In the present context, what is relevant is just the conditional: if these studies hold up, how can they be used in the economic investigation of decision-making (as laid out in the previous section)? This question retains its interest, even if some of the above studies turn out to be problematic for one reason or another.
In order to see how to answer this question, it is best to begin by considering the full passage from Kalenscher and van Wingerden (Reference Kalenscher and van Wingerden2011) that was partially quoted in the introduction:
In the worst case, results obtained in animals will corroborate those obtained from humans, strengthening the existing theory. Preferably, though, comparative research will uncover inconsistencies in choice behavior between humans and animals that allow for an improved, more comprehensive description of choice behavior and possibly force us to re-think the basis of economic theory in the light of the evolutionary roots of choice.
(Kalenscher and van Wingerden Reference Kalenscher and van Wingerden2011, 8)
What is important about this quote is that it expresses a thought that deserves to be looked at in more detail. This is the idea that the data on animal economic decision-making can be disambiguating of the empirical work in economics, in that it can lead to the corroboration or disconfirmation of existing economic theories. This is not a prima facie implausible thought: theoretical disputes in various contexts have been usefully illuminated in precisely these ways. For example, there has been a long-standing dispute over whether dinosaurs were warm- or cold-blooded, with data pointing in different directions. Here, data suggesting a high metabolic rate can be seen to corroborate the former theory (Wiemann et al. Reference Wiemann, Menéndez, Crawford, Fabbri, Gauthier, Hull, Norell and Briggs2022). It is not unthinkable—and in fact quite reasonable—to presume something similar is true in economics.
However, as made clearer in the previous section, this is not in fact the best way of reading the dispute in economics. Data on animal economic decision-making cannot “corroborate [the results] obtained from humans,” thus “strengthening the existing theory”—as Kalenscher and van Wingerden (Reference Kalenscher and van Wingerden2011) suggest—simply because there is no one existing theory in economics. Indeed, as also noted in the previous section, this is the wrong way to think about the dispute in economics to begin with. EUT, PT, RT, SH (and others like them) are models of choice. They can all be adequate; the question is which of them is best used when.
Given this, it becomes important to ask how the data on animal decision-making can be used to address this dispute. Making this clearer is the aim of the next section.
4. A methodology for interspecies economics
To see how to advance the economic study of choice (as detailed in section 2) using data on animal economic decision-making, a brief detour into evolutionary biological reasoning is needed. In evolutionary biology, comparative analyses are used in the investigation of the factors that shaped the evolution of a trait. They work by determining correlations between a focal trait and other traits and aspects of the organisms’ environments (Harvey and Pagel Reference Harvey and Pagel1991; Harvey et al. Reference Harvey, Read and Nee1995a,b; Felsenstein Reference Felsenstein1985; Fisher and Owens Reference Fisher and Owens2004). If we know that trait A and feature of environment B tend to co-evolve (across a set of species), then this provides us with a narrowed hypothesis space as to what determines this evolution. For example, it may be that A is a selective response to B, or that B is a causal consequence of A, or that A and B are developmentally, genetically, or epigenetically linked in some way. While distinguishing among these possible explanations requires further studies, a phylogenetic comparative analysis provides the constraints within which these further studies are situated (Schulz Reference Schulz2013). These studies make precise the extent to which backgrounding the consideration of B is plausible when investigating the evolution of A.
There is an important wrinkle to the use of comparative methods that needs to be noted, though: the presence of a phylogenetic signal. In general, the correlation between A and B is mediated by phylogenetic inheritance (and may in fact be entirely due to the latter). So, both A and B may be (partially or fully) inherited from a common ancestor, so that the correlation between them may be (partly or fully) due to this inheritance. If so, the phylogenetically meaningful evolutionary correlation between A and B will be less than a standard regression analysis would suggest (Fisher and Owens Reference Fisher and Owens2004). This is a well-known issue in phylogenetic comparative analyses, and various methods exist that correct for it (Fisher Owens Reference Fisher and Owens2004; Freckleton et al. Reference Freckleton, Harvey and Pagel2002; Harvey and Pagel Reference Harvey and Pagel1991; Harvey et al. Reference Harvey, Read and Nee1995a,b; Felsenstein Reference Felsenstein1985). The details of these methods are not greatly important here, but it is important to note that this is an issue to which phylogenetic comparative analyses need to pay attention.
All of these points are important here, as they make for the other half of the formulation of a compelling methodology for linking the data on animal economic decision-making to human economics. Recall that the first half of this formulation consisted in showing that the relevant theoretical and empirical conflict surrounding the contemporary economic study of choice concerns the question of which models of decision-making foreground the right causal factors in which circumstances. Putting this insight about the nature of the dialectic in economics together with the structure of phylogenetic comparative analyses just laid out yields the core idea behind the methodology defended in this paper: data on animal economic decision-making can be used to do a phylogenetic comparative analysis of specific forms of economic decision-making, thus helping to determine the latter’s appropriate domains of application.Footnote 10
This works in two related, mutually reinforcing, ways. On the one hand, by seeing which models are predictively useful in which groups of organisms, we can triangulate on the environmental circumstances that make up the domain of application of these models. If we learn that EUT decision-making is found in organisms O 1 to O n , we can look for choice contexts C 1 to C m that these organisms have in common. We can then use this insight to narrow the domain of application of EUT by focusing on C 1 to C m . On the other hand, and in addition, we can consider different environmental circumstances directly, especially if we have ancillary reasons to think that some such are particularly important. For example, from studies in humans, we may suspect that EUT decision-making is found in circumstances C 1 to C m . We can then use phylogenetic comparative analyses to test this suspicion by seeing whether being in circumstances C 1 to C m phylogenetically correlates with EUT in organisms O 1 to O n . Looking at non-human animals thus broadens the data we can access to check on our suspicions about domains of application of models of choice, compared to just looking at humans.Footnote 11
In short: data on animal economic decision-making are important precisely because they allow us to find correlations between different ways of making economic decisions and different environmental circumstances. In turn, this can then help determine which ways of making decisions are usefully foregrounded in which circumstances. It is important to emphasize from the get-go, though, that the upshot of this method is just evidence. It is possible that different organisms make decisions differently in different circumstances (e.g., humans may make EUT decisions in different circumstances compared to other organisms). Still, all organisms are related. This makes phylogenetic correlations important to consider—here and in evolutionary biology more generally. I return to this issue below, but for now, it is useful to consider an example to make these points clearer.
Consider the endowment effect. As pointed out in the previous section, this effect, while found in some animals, is quite limited in its phylogenetic instantiation. In particular, it seems to be a primate-specific trait: humans show it in a wide variety of circumstances, other great apes (chimps, orangutans, gorillas) do so in a more limited manner, and capuchin monkeys in a very narrow range of cases only. Beyond this, though, the effect has not been clearly established (Brosnan et al. Reference Brosnan, Jones, Lambeth, Mareno, Richardson and Schapiro2007, Reference Brosnan, Jones, Gardner, Lambeth and Schapiro2012; Lakshminarayanan et al. Reference Lakshminarayanan, Chen and Santos2008; Drayton et al. Reference Drayton, Brosnan, Carrigan and Stoinski2013; Flemming et al. Reference Flemming, Jones, Mayo, Stoinski and Brosnan2012; Kanngiesser et al. Reference Kanngiesser, Santos, Hood and Call2011). Of course, this could be due to the fact that, outside these primates, suitable experimental studies have not been done (e.g., because they are difficult to do). However, it is at least a reasonable empirical presumption that the endowment effect is primate-specific (and uniquely widespread in the human case).
This matters, as it suggests that parts of the domain of PT (which, as noted in section 2, yields the endowment effect as one of its key predictions) is tied to features of primate living. A key—though, as noted momentarily, not the only—set of candidates for these features are the rich, dynamically changing psychological textures underlying primate social environments. The endowment effect seems to express the idea that when exchanging items (of biological value—Jaeger et al. Reference Jaeger, Brosnan, Levin and Jones2020), I appreciate that you may be tempted to keep your part and take mine. That’s a rich psychological thought and famously a classic aspect of primate social living (Humphrey Reference Humphrey1986; Whiten and Byrne Reference Whiten and Byrne1997).
There are two caveats concerning this conclusion that need to be noted, though. First, so far, it is not clear exactly which aspects of primate lifestyles correlate with the PT-like, endowment effect-yielding economic decision-making. There are many aspects to these lifestyles, and not all of them need to be correlates with the endowment effect. Furthermore, it is also not clear—as is true for phylogenetic comparative analyses in general—whether PT-like, endowment-effect-yielding decision-making is an evolutionary consequence of primate lifestyles, or whether the opposite is true (or whether both are due to a third evolutionary, genetic, or developmental factor). For this, further studies will need to be done that can help isolate the specific features of primate ecologies that PT-like, endowment-effect-yielding decision-making correlates with. I return to this point momentarily; for now, it is just important to note that the above conclusion about the domain of PT-like decision-making as having something to do with primate lifestyles needs to be recognized as relatively abstract, and as something in need of further investigation.
The second caveat concerning this conclusion that needs to be noted here concerns the need to correct this conclusion for phylogenetic inertia. Given that chimps, gorillas, orangutans, and humans are very closely related—with capuchins a little further off, but not drastically so—there is every reason to think that some of the correlation in PT-like, endowment-effect-yielding decision-making and primate ecologies is due to the fact that the decision-making mechanisms and ecologies of these species have been (partly) inherited from a common ancestor. Again, the exact details need to await a closer scrutiny of the exact features of primate ecologies that are correlated with the endowment effect. Here, it just needs to be noted that the corrected correlation is likely to be weaker than it may at first appear—and thus, that it yields a less clear signal about the domain of application of PT-like, endowment-effect-yielding decision-making.
These points are usefully contrasted with the situation concerning EUT-like decision-making. As noted in the previous section, a wide variety of animals have been shown to act in ways that are well predicted by a model based on expected utility maximization: other than humans, macaques, rats, and birds have been shown—sometimes using neurophysiological measures—to make decisions by evaluating which outcomes yield the highest expected rewards (Kagel et al. Reference Kagel, Battalio and Green1995; Santos and Chen Reference Santos, Chen, Paul, Camerer, Fehr and Poldrack2009; Glimcher et al. Reference Glimcher, Dorris and Bayer2005). In turn, this suggests that responding to primate sociality is not the main domain of EUT, but rather something common to a wider variety of organisms. In the present context, this is an important point to note, as it marks for effective and illuminating contrast to the previous case.
On the one hand, finding the specific set of environmental features that EUT-like decision-making is correlated with—i.e., determining the exact domain of application of EUT—is harder than when it comes to PT-like decision-making and primate ecology. While the latter lacks some precision, there are at least a number of clear possibilities to consider (Sterelny Reference Sterelny2003; Tomasello Reference Tomasello2022). When it comes to EUT-like decision-making, though, this is more difficult: finding meaningful environmental circumstances in common among the ecologies of humans, rats, and some birds (among others) is not straightforward. In turn, this makes it harder to do specific phylogenetic comparative analyses here: it is not easy to find candidate environmental circumstances that EUT-like decision-making may be correlated with.
On the other hand, though, since these organisms are so different and since their average degree of relatedness is relatively low, there is at least less reason to think that any phylogenetic corrections will be major here. Whatever exact set of environmental features turns out be well correlated with EUT-like decision-making, this correlation is unlikely to be much skewed by phylogenetic inheritance. In this way, it becomes clear that there is a tradeoff when engaging in phylogenetic comparative analyses of economic decision-making: analyses involving more distant—i.e., relatively less closely related—animals are less subject to the need for phylogenetic corrections, but these analyses are made harder by the fact that it is harder to determine the exact domain of application of this way of making economic decisions.
Stepping back from these particular cases yields two more general points concerning the phylogenetic comparative approach towards the economic study of decision-making (as laid out in section 2). First, as has been noted throughout this section, so far, the work needed to put the approach into practice is only in its infancy. There are few concrete conclusions to provide here yet, as the relevant specific correlations have not been investigated yet. This is something that it will take the combined effort of primatologists, economists, anthropologists, comparative psychologists, and philosophers to achieve.
Second—and this is the key upshot of this paper— it is now at least clearer that this is the right methodology for linking data on animal economic decision-making to contemporary economics. To link the data on animal economic decision-making to the human case, we need to do two things. First, we need to correlate these data with specific other traits and environments: which ways of making decisions link up with which situations? In turn, this requires investigating what these traits and environments might be. Depending on the details of the case, this can be more or less difficult, but is a key element of the approach. Second, we need to strip these correlations of their phylogenetic signal, in line with standard comparative methods. In turn, this requires assembling systematic data sets on sufficiently large classes of species, and then applying the appropriate methods.
In this way, the right methodology here is, in some ways, the opposite of Kalenscher and van Wingerden’s claim. The goal of this analysis is not to “uncover inconsistencies in choice behavior between humans and animals” that can “force us to re-think the basis of economic theory in the light of the evolutionary roots of choice.” Rather, the goal is to find correlations between different ways of making economic decisions and other traits and environments, whether these are shared or not shared. (In line with the discussion of section 2, it is not clear what it even means to “re-think the basis of economic theory in the light of the evolutionary roots of choice.”) Put differently: phylogenetic comparative studies can extend the economic methodological tool kit to clarify the domains of application of the different models.
It is important to emphasize this last point: the proposal laid out here is meant to supplement, not supplant, other methodological approaches—including other evolutionary tools, such as formal evolutionary (game theoretic) models (see, e.g., Cooper Reference Cooper1987; Okasha Reference Okasha2011, Reference Okasha2018; Okasha and Binmore Reference Okasha and Binmore2012; Güth and Kliemt Reference Güth and Kliemt1998; Samuelson Reference Samuelson2002; Huck et al. Reference Huck, Kirchsteiger and Oechssler2005). As noted earlier, the phylogenetic approach just provides evidence. Other studies are needed to strengthen this evidence. Turning this around, the comparative analysis can advance the discussion here—but it cannot resolve it.
That said, the phylogenetic approach has some specific advantages. On the one hand, it can make suggestions as to what to look for in our other studies. As made clear earlier, it can suggest specific environmental circumstances to further test when assessing the domains of application of different models of choice. On the other hand, the phylogenetic approach does not need to commit to a specific evolutionary modeling framework (e.g., concerning the payoffs of different ways of making economic decisions, or the nature of evolutionary dynamics) and opens up a whole new source of data and experiments. However, other tools—including other evolutionary approaches—are very useful as well and have their own strengths, such as being able to easily consider different modeling assumptions, or making predictions about the kind of environments that foster certain kinds of decision-making.
For example, there are models suggesting that RT-like, alternative option-dependent decision-making is adaptive in autocorrelated environments (Fawcett et al. Reference Fawcett, Fallenstein, Higginson, Houston, Mallpress, Trimmer and McNamara2014): if two successive states of the environments are more likely to be similar to each than ones further apart in time, it can be adaptive to treat seemingly irrelevant alternatives as “insurances” that can be relied on in times of need. In turn, this can yield intransitive decision-making across time. However, the nature of the modeling assumptions (such as the degree of autocorrelation, the nature of the options and payoffs, etc.) matters greatly here, with different such assumptions leading to different outcomes (Fawcett et al. Reference Fawcett, Fallenstein, Higginson, Houston, Mallpress, Trimmer and McNamara2014). In turn, this makes it useful to compare these insights with the results of phylogenetic comparative analyses: we can test whether autocorrelated environments phylogenetically correlate with RT-like decision-making, and perhaps also use the results of the latter analyses to triangulate on the kinds of autocorrelation needed for RT-like decision-making to be adaptive. In this way, the phylogenetic approach can be seen to be a fruitful expansion of traditional economic methodology.
5. Conclusion
An increasingly popular interdisciplinary area of research concerns the question of how animals make economic decisions. What is less clear is what role this research can play in economics. This paper has argued that—contrary to some recent claims otherwise—data on animal economic decision-making do not help in corroborating or falsifying existing economic theories of choice. In turn, this is due to the fact that, first, such corroboration and falsification is not looked for. The real question is not which of the different approaches towards choice—EUT, PT, RT, and SH—is the right one; rather, it is what the representationally proper domains are of the different models of choice.
Once it is clear that this is the question we are trying to answer, the role of data on animal economic decision-making becomes clearer. This role can now be shown to consist in enabling the establishment of correlations between the different ways of making economic decisions and different features of the environment. In order to establish these correlations, though, the data on animal economic decision-making need to be placed in a phylogenetic comparative analysis: they need to be correlated with specific features of the environment, and these correlations need to be corrected for the possible presence of distorting phylogenetic signals. While the precise upshot of such a phylogenetic approach towards economics is not yet clear, it is hoped that the paper has at least made clearer why this is the right approach to take.
Acknowledgments
I would like to thank the referees for this journal and audiences at the universities of Kansas and Rochester, as well as from the Metropolitan Area Philosophy of Science group in New York City, for very helpful comments on previous drafts of this paper.