2. Revealed preference approaches

Revealed preference approaches are a set of frameworks within the rational choice theoretic approach to modelling choice behaviour. To set up the discussion of revealed preference approaches, I will first introduce some basic building blocks of rational choice theory and then discuss the revealed preference relation.

Within classical rational choice theory, there are two main approaches to modelling choice behaviour. The first approach, which I will not discuss, takes preferences as primitive, and imposes certain rationality requirements on an individual’s preferences. The second approach takes choice behaviour as primitive, and imposes rationality requirements on choice behaviour.Footnote ²

To model choice behaviour, the following are taken as primitive:

• A non-empty set X of mutually exclusive objects of choice. Elements include x, y, z …

• A non-empty set ${\cal K}$ of contexts, where each element $K \in {\cal K}$ is a non-empty subset $K \subseteq X $ of feasible options. K is often taken to be the set of all non-empty subsets of X.Footnote ³

• A choice function $C:K \in {\cal K} \to C(K) \in {\cal K}$ , where the choice function assigns a non-empty set of options to every context.

In addition to the above primitives, revealed preference approaches then define a revealed preference relation. This binary relation imposes additional structure through an ordinal ranking of alternatives across contexts. One benefit of this additional structure, which I will return to in Section 5 and in the conclusion, is that the ordinal ranking more efficiently systematizes the relevant pattern of choice behaviour, compared with the choice function. Importantly for what follows, the revealed preference relation does not represent any additional information, beyond this ordinal ranking; in particular, it does not represent any information about causes. Instead, it more efficiently encodes the information in the choice function. By efficiently summarizing patterns in choice behaviour, the revealed preference relation serves an epistemic function, as it provides a more efficient tool for organizing information about such patterns.

There are many ways to construct the revealed preference relation; the construction method does not matter for present purposes, so I will adopt the following definition for the so-called weak revealed preference relation ‘≿*’. As above, x and y are options, K is a context, and C(K) is the choice function.

$$x\, \scsim^{\ast\ } y \ {\rm iff \ for \ some}\ K,x \in C(K) \, \rm{and} \, y \in K$$

In other words, x is weakly revealed preferred to y if and only if x is chosen from a context that contains y (note that this is compatible with y being chosen over x in a different context).Footnote ⁴

The weak revealed preference relation is a formal object in a theory, one that needs to be interpreted. The interpretation under consideration here takes the revealed preference relation to be a mere summary of patterns in choice behaviour.Footnote ⁵ Here, I follow Clarke (Reference Clarke2016) in maintaining that the claim that preferences are summaries of an agent’s choice behaviour is separable from behaviourist and other positivist claims that have often been adopted by proponents of revealed preferences (going back to Samuelson (Reference Samuelson1938) and still prevalent among some practicing economists). As discussed in the Introduction, the revealed preference relation, so interpreted, is inaptly named, since preferences, understood as mental states attributable to agents, are not represented by the revealed preference relation. Instead, the revealed preference relation is shorthand for information about choice behaviour. For example, say that Monique is represented as having a weak revealed preference for pineapples over bananas. This revealed preference relation is shorthand for a longer claim, of the following form: ‘In context K₁, Monique chose pineapples over bananas; in context K₂, Monique chose pineapples over bananas …’

There are two areas of economics where revealed preference approaches have been most influential, and thus where the arguments of this paper have the most significant ramifications for economic practice. The first is general equilibrium theory, particularly the study of the demand side of markets. The study of demand often proceeds by starting from individual demand functions and aggregating those functions into a market-wide demand function.Footnote ⁶ The second is an empirical research programme that Wade Hands calls contemporary revealed preference theory, which studies choice behaviour by constructing a utility function that rationalizes a set of finite choice data. The bulk of this work has been done in applied research on consumer choice, which uses data about prices and quantities and the associated opportunity sets (see Blundell Reference Blundell2005), but ‘contemporary revealed preference theory’ covers ‘those who are using it as the primary theoretical tool in their applied research’ (Hands Reference Hands, Lehtinen, Kuorikoski and Ylikoski2012: Ch. 7, Section 3).

A caveat is in order here. There is widespread disagreement about how to interpret economic practice here, made more difficult by the following three-fold distinction. There is what economists actually do: how they construct theories, build and test models, etc. There is also what they say about what they do: claims about how they build and test models, etc. One difficult question facing those interested in the methodology of economics is the question of how much weight the latter should be given in interpreting the former. Complicating the interpretation of economic practice is the further question of what economists should do. This latter question, for example, is one that Hausman (Reference Hausman2012) addresses in a recent book, arguing that economists should (and do) interpret preferences as total subjective comparative evaluations.

Many philosophers interested in the question of what economists actually do have interpreted economists as, by and large, working with a mentalist interpretation of preference.Footnote ⁷ Accordingly, they interpret contemporary revealed preference approaches as a methodology to discover mental preferences (see footnotes 5 and 7 for examples). Here I agree with Angner (Reference Agner2018) that a better interpretation of economic practice is one on which economists take the concept of preference as a primitive – as neither capable of or in need of further analysis – with two key properties, namely, completeness and transitivity.Footnote ⁸ At the same time, economists engaged in empirical research need to identify the entities in the world that are represented by their models. There, I take the hypothetical revealed preference interpretation, discussed in Section 4, as the prevailing interpretation used by some communities of economists engaged in empirical research (see Section 3 for discussion of contemporary revealed preference approaches). That being said, I am happy to grant that the mentalist interpretation of preference prevails in economics, for dialectical purposes. The arguments of this paper primarily address the question of how economists should interpret revealed preference frameworks, in light of the two explanatory objections discussed in Sections 3 and 4.

With that background in place, let’s move on to the two explanatory objections.

3. The no self-explanation objection

The criticism that the identification of preferences (or utility) with choice behaviour produces a circularity goes back at least to J.E. Cairnes’ (Reference Cairnes1872: 20) criticism of W.S. Jevons’ discussion of so-called ‘exchange-value’, i.e. marginal utility.Footnote ⁹ Indeed, the circularity seems so obvious as to need no argument: A revealed preference relation is defined in terms of choice behaviour, and is then used to explain choice behaviour.

A mere circularity, of course, might not worry the revealed preference theorist. Jevons, Sen, and others, however, have argued that the circularity is a vicious one, either because of issues around the measurement of utility (see Cairnes Reference Cairnes1872: 20) or because it trivializes the concept of utility (see Sen Reference Sen1977: 322–3). Thus far, the problems that this circularity raises for measurement and for the meaningfulness of economic theory have received the most attention in the literature.

As I will argue below, this circularity also raises a distinctly explanatory problem, which seems unavoidably vicious. Because of the circularity, explanations containing the revealed preference relation violate a plausible principle of explanation, the No Self-Explanation Principle (taken from Miller Reference Miller2015). The No Self-Explanation Principle says that no fact (partially) explains itself. If this principle doesn’t immediately strike you as plausible, think about a putative explanation of why a projectile travelled 10 metres that appeals, among other facts, to the fact that it travelled 10 metres, and about whether you would find that explanation satisfying.

Unfortunately, as I will argue below, this violation of the No Self-Explanation Principle is needed to deal with a particularly gnarly version of the problem of induction. Furthermore, an attempt to avoid the circularity yet deal with this version of the problem of induction only escapes the circularity in letter, not spirit.

To better understand the explanatory viciousness of the circularity, it will help to have an example purported explanation at hand. Say that a revealed preference theorist is trying to explain why Olivia chose olive pants over jeans or yellow pants. The revealed preference theorist would say that Olivia has a revealed preference for olive pants over jeans or yellow pants. So, the constructed revealed preference relation for Olivia predicts, and thereby explains, that she chooses the olive pants.Footnote ¹⁰ However, recall that the revealed preference relation is a mere summary of patterns among choices. So, information about choice behaviour is used to explain a particular choice.

So far, it seems, so good: it is good scientific practice to use information about patterns among phenomena of some type to explain token behaviour of that type. For example, an explanation of why Taib’s ice cream melted on a hot day may give information about the melting point of ice cream, heat transfer between the ice cream and the surrounding air, etc. – information that applies to many token melting ice creams. However, grouping together thermodynamic and revealed preference explanations is slightly disingenuous. Taib’s melting ice cream is not explained by mere reference to the fact that past ice cream cones have melted under similar conditions. Instead, the explanation cites robust generalizations that apply to a range of objects. In the case of revealed preference explanations, an agent’s past choices of an option explain her future choices of that very same option (in conjunction with additional information about her budget, physical constraints she is under, etc.), rather than any appeal to more widely applicable generalizations.

Even the fact that choice behaviour explains choice behaviour, however, does not yet lead to circularity. Circularity only threatens if the particular choice to be explained is contained within the revealed preference relation. If the revealed preference relation includes the choice itself, then a particular choice partially explains itself.Footnote ¹¹ Thus, a closer look at the domain of the revealed preference relation is an important next step. Following Hands (Reference Hands2013), I will divide revealed preference approaches into two major camps, each of which posits a different domain: more theoretical approaches, and contemporary revealed preference approaches, an empirical research programme.

The more theoretical tradition of revealed preference frameworks, one that goes back to Samuelson (Reference Samuelson1938), assumes an infinite domain of choice acts for the revealed preference relation. In such frameworks, the circularity is straightforward, as the domain of the revealed preference relation includes the choice that is explained thereby.

By contrast, contemporary revealed preference approaches focus on the construction of a revealed preference relation on the basis of a finite set of observed choice data (see Afriat (Reference Afriat1967), Varian (Reference Varian1982), as well as an overview in Hands (Reference Hands2013: Section 2)). Such empirical approaches seem to posit a revealed preference relation with a domain of actual, observed choices.Footnote ¹² In that case, there is no circularity, since the finite set of observed choice data does not contain the choice to be explained.

Such a solution, however, avoids an explanatory circularity by running up against a particularly serious version of the problem of induction. The restriction of the domain of the revealed preference relation to past choices raises the following inductive puzzle: why are we justified in using information about past choices to predict future ones? If the revealed preference relation were understood as a mere summary of past choices, it is not clear why past behaviour would be a good guide to future behaviour. Absent certain assumptions, my choice of a cappuccino this morning, for example, is not a good guide to my beverage choices of the afternoon. (Perhaps I am a caffeine sensitive person who only drinks coffee in the morning, for example.)

At this point, you might be thinking something like the following: ‘So what? The problem of induction is everyone’s problem.’ Agreed – but, the problem of induction is particularly pressing in the case of contemporary revealed preference approaches. That is because, as Hands (Reference Hands, Lehtinen, Kuorikoski and Ylikoski2012) says:

Recall that, per the characterization of Section 2, contemporary revealed preference approaches do not represent the causal mechanisms that produce choice behaviour. Thus, the theorist does not have evidence – evidence which would take the form of causes or causal mechanisms – of the stability of the relevant preferences which she can point to in order to justify her assumption of that stability. To do so would be to violate the relevant interpretation of the revealed preference relation. Of course, inductive worries can be raised about such causal mechanisms. Still, even the inductive sceptic should admit that different low degrees of confidence that the future will resemble the past are warranted by different amounts and types of information about the past. The identification of a single causal mechanism across different circumstances of choice behaviour should increase one’s confidence that the future will resemble the past, compared with the mere observance of a pattern in choice behaviour across such circumstances.

To avoid the inductive sceptical puzzle, the theorist could take the domain of the revealed preference relation to be past, future, and perhaps possible choice behaviour. On this solution, the revealed preference theorist who gloms on to the correct revealed preference relation is justified in using it to predict future choices, since the revealed preference relation already represents the same choice that it aims to predict. Note that, for this solution to work, the revealed preference relation must contain the choice to be explained. Otherwise, the inductive puzzle can be raised again: what justifies the prediction of my coffee choice this morning on the basis of past coffee choices? However, if the revealed preference relation already represents the choice to be explained, then the deep inductive puzzle set out by Hands above is, in a sense, stipulated away. ‘Look’, the revealed preference theorist might say to the sceptic, ‘if you grant that I’ve inferred the correct revealed preference relation from an agent’s past choices, then you have to grant that I am no less justified in using it to predict the future than a scientist who uses a psychological interpretation of preference, as the revealed preference relation already represents that future choice’.Footnote ¹³

A second solution, by contrast, maintains that the domain of the revealed preference relation is a set of (observed) actual choice behaviour. It then adds the additional assumption that the agent has a stable revealed preference, i.e. that the pattern of choice does not change in future, assuming that the agent is presented with the same options.Footnote ¹⁴ In the previous case of my morning cappuccino, this solution would assume that my beverage of choice does not change as the time of day changes, and predict that I order a cappuccino in the afternoon when faced with the same options.

This assumption does prevent the letter of the circularity objection, since the revealed preference relation does not contain the choice itself.Footnote ¹⁵ And yet, the distinction between the two solutions is a distinction without much of a difference. The package of a revealed preference relation defined over a domain of actual, past choices and the assumption of a stable pattern of choice gets around the problem of predicting future choices in a similar way to the first solution, namely, by adding the information that the choice to be explained is part of the pattern represented by the revealed preference relation. The only difference is that the relevant work is done by a background assumption, rather than the revealed preference relation itself.

To sum up: in order to avoid a particularly pressing inductive puzzle, the revealed preference relation must be understood either to contain the choice to be explained, or as paired with a background assumption of stable preferences. This avoidance comes, however, at the cost of a violation of the No Self-Explanation Principle, and an associated, seemingly vicious circularity.

4. The causal explanation objection

A second major locus of criticism centres around revealed preference approaches’ purported inability to causally explain choice. The lack of causal explanation by revealed preferences seems to follow straightforwardly from the interpretation of the revealed preference relation as a mere summary of choice behaviour. The revealed preference theorist, for example, does not seem able to say that Annie chose the slice of chocolate cake because (in the causal sense of ‘because’) she prefers chocolate cake. To do so would be to say something like ‘Annie chose the slice of chocolate cake because she chose slices of chocolate cake in similar situations in the past’, or, even worse, something like ‘Annie chose the slice of chocolate cake because she chose the slice of chocolate cake’. Neither gloss seems to tell a compelling causal story. The claim that choices cause other choices except by way of a mental intermediary seems like a highly implausible view, and the view that an event could cause itself seems even worse. And so, since revealed preference approaches do not represent the causes of choice behaviour, revealed preference frameworks cannot be used to give causal explanations.

Of course, there are cases of scientific explanations that give generic information about causes but do not represent the particular causes of some outcome of interest. Some explanations in biology are of this sort. A biologist might explain the presence of a trait in a population due to certain selective pressures, thereby ruling out that the trait is a spandrel. Explanations such as these represent causes of a certain type (in this case, selective pressures) but do not pick out particular causal mechanisms.

Such causal abstraction is not, however, what is going on in the case of explanations that appeal to revealed preferences, on the interpretation of revealed preferences given by Section 2. For the analogy with evolutionary explanations to help the revealed preference theorist, revealed preference explanations would have to be interpreted as representing types of causes of choice behaviour, while not picking out any particular underlying causes or mechanisms that produce choice, such as psychological or neurological mechanisms (see Section 2).

There is, however, a better defence of the causal explanatory power of revealed preferences, one that deserves much more exploration than this paper can undertake. There is a package of views on which the revealed preference relation seems to causally explain choice behaviour. The first component of this package is an interpretation of the revealed preference relation as a behavioural disposition. On this interpretation, the revealed preference relation entails a series of counterfactuals of the following form: if the agent were in context K, she would choose option x. The second component is a particular view of causal explanation, namely, interventionism (see e.g. Pearl Reference Pearl2000; Woodward Reference Woodward2003). Interventionist accounts of causation and causal explanation represent causes and effects as variables in a causal model, and take a variable C to be cause of another variable E if and only if a suitable intervention on C that changes its value (and the value of no other variables) produces a corresponding change in the value of E (or the probability distribution of E). For the interventionist, the counterfactuals entailed by the revealed preference relation are more perspicuously understood as claiming that an intervention on the context variable causes a change in the choice variable.

Thus, while the causal explanation objection appears devastating, there may be a package of views that saves the causal explanatory power of revealed preferences. The rest of this paper will focus on a unificationist defence of the explanatory power of revealed preference approaches. However, I will return to this interventionist defence in the conclusion.

Suppose that you are convinced by the causal explanation objection. Is this lack of causal explanation by revealed preference approaches a theoretical benefit, or a damning problem? Developments in economic methodology over the last forty years or so seem to support the latter position. In particular, the discipline’s move away from the formal, idealized theory-building of individuals such as Samuelson and towards causal investigation, as well as the incorporation of more psychologically realistic assumptions by behavioural economists, may cause you to doubt the Introduction’s assertion that revealed preference approaches are still common in current theoretical and empirical investigations. Those methodological developments alone, however, do not cause problems for revealed preference approaches, as the revealed preference relation could be important for purposes other than causal explanation, such as prediction.

The addition of two assumptions, however, turns this seeming theoretical benefit into a damning problem. The first assumption is that something is an explanation in virtue of providing information about dependence relations, such as the causal relation or laws of nature, and that this account of explanation is the single, correct account employed across the sciences. The second assumption is that, all other things equal, a framework that can provide explanations is better than one that cannot.

Interestingly, vocal defenders of revealed preference approaches, such as Binmore (Reference Binmore2009), have exclusively denied the second assumption. The denial focuses on epistemic trade-offs in science. As Kuhn (Reference Kuhn1977: Ch. 13), among many others, has argued, there are different epistemic virtues that scientific theories may possess, and scientists often make trade-offs among these virtues. In the case of revealed preference approaches, Binmore (Reference Binmore2009: 20), for example, argues that revealed preference approaches give up on the virtue of causal explanation, and gain ‘a theory that is hard to criticize because it has little substantive content’.Footnote ¹⁶ The argument from trade-offs is particularly powerful if paired with the claim that these trade-offs are inevitable, as no single scientific theory can realize all of the epistemic virtues (see van Fraassen Reference van Fraassen1989, among others). A defence such as Binmore’s concedes the heart of the criticism, that revealed preference approaches cannot provide causal explanations, and defends revealed preference approaches on other grounds. The Kuhnian-inspired defender, however, has given up the game a bit too quick, here. This trade-off among epistemic virtues only gets going against a certain background assumption about explanation, namely, the first assumption above. By contrast, the more successful defence of revealed preferences against the no causal explanation objection rejects this first assumption, and denies that there is indeed a trade-off for revealed preference approaches between explanation and other virtues. Let’s now turn to that defence.

5. A fundamental disagreement about explanation

I want to pause and cast an eye over the dialectic thus far. We’ve seen claims of generalizations as mere patterns in the phenomena, rather than something more modally robust; explanatory circularities; charges that a philosophical theory results in explanatorily inert generalizations … if this reminds you of the disagreement between Humeans and anti-Humeans over laws of nature, then I whole-heartedly agree. The analogy with the debate over laws of nature reveals a principled response to both the no self-explanation objection and the causal explanation objection: unificationism about explanation.

Three notes about terminology before I proceed. First, the analogy draws on a common epistemic property of Humean laws and the revealed preference relation, namely, their systematicity. In the case of laws, this systematicity, and the Humean account of laws more generally, is usually paired with other background commitments, such as a denial of fundamental necessities. However, as Demarest (Reference Demarest and Jacobs2017) and Hicks (Reference Hicks2018) discuss, the Humean view of laws is compatible with other background commitments, such as a commitment to fundamental necessities. This point is important for the plausibility of revealed preference approaches, as contemporary defenders of revealed preferences accept the existence and causal influence of mental states on choice (see Binmore (Reference Binmore2009: 19–20) for one example). To separate the discussion here from such orthogonal background Humean commitments (which are difficult to mentally jettison), I will follow Hicks (Reference Hicks2018) in using ‘regularity theorists’ instead of ‘Humeans’ and ‘modalists’ instead of ‘anti-Humeans’.

The second note concerns ‘unificationism’. As Colombo and Hartmann (Reference Colombo and Hartmann2017) discuss, ‘unification’ is a heterogeneous notion in science, one which often comes apart from explanation. For present purposes, two notions of unification are relevant. The first notion of unification is that of modelling a set of diverse phenomena within a single framework or with a few mathematical equations (see Colombo and Hartmann Reference Colombo and Hartmann2017: Section 4). The second notion of unification is the notion of a systematization of a set of facts by showing how those facts are instances of a more general pattern (see Friedman Reference Friedman1974; Kitcher Reference Kitcher, Kitcher and Salmon1989).

Finally, these two different notions of ‘unification’ are the basis for two distinct roles that unification plays in science. First, scientists often prefer theories that are more unifying, i.e. a theory that can model a larger set of diverse phenomena within a single framework or set of equations. Call this the criterion of theory choice use of ‘unification’. Thus far, the literature on revealed preferences and unification has focused on the criterion of theory choice use of ‘unification’.Footnote ¹⁷ By contrast, this paper argues that revealed preference theorists should adopt a unificationist theory of explanation, where a set of information counts as an explanation in virtue of showing how the fact to be explained is an instance of a more general pattern.

6. Where to now?

This paper has argued for the following conditional conclusion: if the revealed preference theorist adopts a background unificationist theory of explanation, then she avoids the two explanatory problems posed in Sections 3 and 4. The pairing of revealed preference approaches with unificationism shows that it is wrong to consider the revealed preference modeller as tasked with the construction of models that capture underlying (causal) constraints. Instead, the revealed preference modeller attempts to find an efficient systematization of a set of choices. In other words, she looks for mere patterns among choice behaviour. By her own lights, a proponent of revealed preference approaches succeeds by doing so, even if she does not provide causal explanations, and she violates the No Self-Explanation Principle.

I’ll end by discussing two natural next steps in the dialectic. An important question that arises out of this paper is whether revealed preference approaches are, indeed, more unifying. That question, of course, requires a comparison – more unifying than what alternatives? In the sense of efficient expression, a choice-based framework with a revealed preference relation is clearly more unifying than a choice-based framework without one. The harder question for the proponent of revealed preferences is whether revealed preference approaches are sometimes more unifying than frameworks that represent psychological facts. Proponents of revealed preference approaches, such as Binmore, certainly take them to be so. Here is Binmore (Reference Binmore2009: 8–9):

Note that the defence of revealed preference approaches cannot merely claim that bracketing causes at the individual level is more unifying, for reasons related to the discussion of evolutionary explanations in Section 4. There, I pointed out that many causal explanations do not explicitly represent causes or causal mechanisms at the individual level. However, the biologist would not deny that these explanations stand in for specific causal mechanisms in particular populations. By analogy, a mentalist about preferences can maintain that economic explanations are better in virtue of bracketing causes, but that these explanations represent mental entities which cause choice behaviour. The revealed preference theorist thus has to argue that her framework is more unifying in virtue of refusing to represent mental preferences at any stage in the modelling process.

Binmore has articulated a compelling thought here, but it is one that requires more defence. There are two complementary routes to pursuing such a defence, one more epistemic and the other more metaphysical. The more epistemic defence argues that, due to cognitive limitations of human beings, the diversity of their cognitive interests, and facts about their representational and technological capabilities, unificationist explanations that do not aim to represent the psychological determinants of choice allow us to better explain, and thereby understand, large-scale explanatory patterns that would be less readily apparent at the psychological level. This defence can look defeatist: in some cases, our cognitive and technological capacities are too limited to get at the causal structure of the world, so we seek efficient summaries of patterns instead. However, one might lean more heavily on the diversity in our cognitive interests, rather than the limitation in our capabilities, in order to present a less defeatist version of this defence.Footnote ²³ Furthermore, the epistemic defence appears less defeatist when paired with a metaphysical defence that starts from the fact that some systems in the world, such as the large social systems studied by economists, are very complex. This complexity sometimes grounds the explanatory superiority of unificationist over constraint-based explanations, as simple pattern representation allows inquirers to abstract away from much of the complexity.

Of course, these two routes are nicely complementary. And indeed, together, they offer an argument for explanatory pluralism, one starting from a case study of revealed preference approaches. Such a two-pronged strategy fits nicely, for example, with arguments from Kellert et al. (Reference Kellert, Longino and Waters2006: Introduction) that such case studies ‘establish the possibility that the world is too complicated or too indeterminate and our cognitive interests too diverse for the monist ideals’. However, as they stand, the arguments of the paper do not yet support explanatory pluralism. A successful argument for explanatory pluralism requires an unconditional defence of the pairing of unificationism and revealed preference approaches. Furthermore, the question of whether interventionism about explanation equally well or better accounts for the explanatory practice of revealed preference approaches is still an important open question, raised but not settled here. If, indeed, it does, such a fact speaks against revealed preference approaches as a case study that supports explanatory pluralism.

This last point brings us back to a very different avenue of explanatory defence for revealed preference approaches: interventionism. Section 4 suggested that, on an interpretation of revealed preferences as behavioural dispositions, revealed preferences are sometimes causes, at least against a background interventionist view of causation. If so, the combination of the interpretation of revealed preferences as a behavioural disposition and an interventionist view of causal explanation offers a potential second defence of the explanatory power of revealed preference approaches. This defence has two main advantages. It avoids theoretical challenges, such as the problem of causal explanation. It also reconciles revealed preference approaches with economics’ current methodological focus on causal modelling. The next step in the dialectic is to address the following two questions. First, if interventionist views of causation do deem revealed preferences to be causes, is that an objection to interventionist views, or a plausible result? Second, even if revealed preferences are causes, are such putative revealed preference explanations (good) explanations? But, I’m afraid that such questions will have to wait for another time.