Let us assume selective realism about properties and relations: certain predicates and relational terms refer to entities that exist as part of a mind-independent reality. A major question at this point concerns the nature of these entities. What is the relationship between these entities and the behaviour of their bearers? If they themselves ensure that objects possessing them behave in certain ways in certain circumstances, they are irreducibly dispositional. If they do not – if something additional is required – then they are not irreducibly dispositional. They are, we might say, ‘categorical’. This view that all properties (and relations – I take this extension for granted in what follows) are irreducibly dispositional I shall call [following Molnar (2003)] Pan-Dispositionalism; the view that all are categorical [following Armstrong (1997)] I call Categoricalism. There are two middle positions. First, there is the view that properties have both an irreducibly dispositional and a categorical side, or aspect – a view endorsed recently by Heil (2003). Second, there is the view that some properties (perhaps most) are irreducibly dispositional, but some are categorical. This latter position is Moderate Dispositionalism.

In this paper, I want to focus on the reasons philosophers have given for preferring Moderate Dispositionalism to Pan-Dispositionalism. Why accept that some properties are irreducibly dispositional, but not all? Brian Ellis and George Molnar are Dispositionalists who claim that certain kinds of properties are clearly categorical. I will argue that they are wrong and that nothing they say provides good grounds for rejecting Pan-Dispositionalism. This, of course, is no reason in itself to prefer Pan-Dispositionalism to Moderate Dispositionalism.

There are good reasons for preferring Pan-Dispositionalism or Moderate Dispositionalism to Categoricalism. For example, Armstrong (1983) takes laws to be nomic relations between categorical properties. But while the move away from a ‘regularity’ view of laws is to be applauded, his account struggles to accommodate the intuition that there might be laws involving simple properties that, as a matter of contingent fact, are never instantiated. It seems that the only way to accept such laws, on a relations-between-categorical-properties account, is to take properties to be transcendental entities, i.e. entities that are not part of the spatiotemporal world. The Dispositionalist does not have to make this move. He or she can endorse an immanent view of properties (whereby properties are part of the spatiotemporal world) and accept laws involving uninstantiated properties because laws, according to Dispositionalism, are not themselves entities: they are truths derived from the natures of properties, and a property may well have a nature such that if it were instantiated and in certain circumstances, a certain property – one which is actually never instantiated – would be instantiated. There is a second reason for preferring Dispositionalism: Categoricalism fails to accommodate the thought that it is the objects themselves, with their properties, which fix what happens in the circumstances, not something existing ‘over and above’ those objects.

Reasons for preferring Pan-Dispositionalism to Moderate Dispositionalism are harder to find. But there is at least one: it is the simpler theory. Simplicity is an important consideration when choosing one’s metaphysical theory, and when it comes to the choice between these two theories, both of which share benefits over non-Dispositionalist accounts and both of which adequately explain the phenomena, it may be the only reasonable way there is left to choose between them. The choice between these two theories is the choice of which best explains the phenomena of law-like regularity.Footnote 1

In what sense is Pan-Dispositionalism simpler? Not in the sense that it takes the world to contain fewer properties: Moderate Dispositionalism may endorse the same properties as Pan-Dispositionalism – the difference being only that it takes some of them to be categorical, some irreducibly dispositional. Gauging ontological simplicity is generally taken to be a matter of counting number of kinds, not number of members of kinds. And Pan-Dispositionalism posits fewer kinds of property. There are only irreducibly dispositional properties (one kind), not both irreducibly dispositional properties and categorical properties (two kinds). One might argue that Moderate Dispositionalism posits the same number of kinds as Pan-Dispositionalism, namely one: properties. But irreducibly dispositional properties and categorical properties are so different, despite their commonality, that I believe they are relevant kinds for counting here. We should prefer the scientific theory that (all else being equal) posits fewer kinds of particle to help explain some phenomenon. Analogously, we should prefer the metaphysical theory that (all else being equal) posits fewer kinds of property to help explain law-like regularity.Footnote 2

I shall proceed as follows. First, I describe in some detail what an irreducibly dispositional property is (Section 1). I then look at two regress arguments often taken as fatal to Pan-Dispositionalism (Section 2). Finally, I turn to the specific arguments of Ellis (Section 3) and Molnar (Section 4) for Moderate Dispositionalism.

1 Irreducibly Dispositional Properties

As indicated, I shall assume realism about properties. If x is F, and F is a property, then x has a mind-independent entity. There are two options at this point. If y is also F, and properties are universals, then x and y instantiate the numerically identical entity F. If y is also F, and properties are tropes, then y has an entity that is similar enough, qualitatively, for it also to be called ‘F’. Armstrong endorses universals, Ellis endorses universals and tropes, and Molnar endorses tropes. For the most part, I shall assume that properties are universals – partly for stylistic reasons and partly because I think properties are universals. Nothing of importance here hangs on this choice, however, and the advocate of tropes is free to re-interpret talk of properties in what follows as talk of sets of qualitatively similar or identical tropes.

The realism assumed is a selective one. Not all predicates pick out a separate natural property (e.g. ‘is in pain’ may not pick out any one property that is being in pain). Some predicates pick out no property at all (e.g. ‘being phlogiston’). There might also be currently undiscovered natural properties that, as a result, have no predicate term referring to them. Dispositionalism, strong or moderate, is naturally allied to an a posteriori account of properties and relations according to which mature science is the best guide to what properties and relations exist.

Irreducibly dispositional properties (‘powers’) stand in contrast to categorical properties (or ‘non-powers’ or ‘non-dispositional properties’), and the contrast can be characterisedFootnote 3 as that between those properties that are themselves enough to ensure the truth of one or more counterfactuals about how an object bearing it would behave, and those that are not.

Take, for example, the property of having mass m, which we can assume scientific realism would endorse. If an object x has mass property M, then that property is irreducibly dispositional iff x having it is enough to ensure that x would behave in certain ways in certain circumstances. The mass property is categorical, on the other hand, iff x having it is not enough to ensure that x would behave in certain ways in certain circumstances. If the property is categorical, what is also needed is a law – as an extra entity over and above x and its mass property – governing how objects with mass m would behave in certain circumstances.

We might try and cash out ‘ensuring’ in terms of possible worlds: if M is irreducibly dispositional, then x behaves the same way (in the same circumstances) in all metaphysically possible worlds; if M is categorical, x does not. But this definition fails to allow that M may be categorical and be part of the same set of laws across all metaphysically possible worlds. Evan Fales (1990) holds just such a view, taking laws to be nomic relations between categorical properties where those relations hold between properties as a matter of metaphysical necessity. Since Fales’ view appears to harbour no contradiction, I reject this possible worlds characterisation of ‘ensuring’. In this paper, I shall cash out the notion of M ensuring certain behaviour involving x as a matter of M making true counterfactual facts concerning x and leave the notion of truthmaking an intuitive one.

It is clear that we can, and do, predicate dispositional terms of various objects. We say, for example, that the sugar cube is soluble, or that the rubber band is flexible. But this in itself is not enough for it to be the case that there are irreducibly dispositional properties. Since property-terms such as ‘being soluble’ are unlikely to be used in the statement of our best scientific theories, we might say that each time we correctly attribute solubility to an object, there is some (though perhaps not always the same) scientifically respectable property that we pick out, e.g. having molecular structure s. But the property picked out in this way need not be irreducibly dispositional. We have to ask what makes it true that an object with that property would (for example) dissolve when placed in water. If it is simply the object having that property, then the property is irreducibly dispositional. If it is the object having that property and that property being part of a law connecting it to dissolving in water, then the property is categorical. In this way someone who endorses Categoricalism, the view that all properties are categorical, can still hold onto the common sense view that certain dispositional claims are true.

Both Moderate Dispositionalism and Pan-Dispositionalism give centre stage to the notion of an irreducibly dispositional property. The counterfactuals made true by objects bearing such properties are often conditional counterfactuals. Take Shoemaker’s (1980) example of a knife. An object being knife-shaped, or being knife-sized, is not enough to make it true that the object will cut paper when pressed close – it might be made of something that crumbles on touching the paper. Neither is an object being made of steel enough either – it might be the wrong shape to cut anything. Nevertheless, it seems true, of each of these, that if it were instantiated along with the other two, the object of which it is a property would cut paper if it were pressed close. In other words, each of the three properties, instantiated by any x, makes true a claim about x of the following form:

if x were to instantiate other properties F, G…, then if x were in circumstances C there would be behaviour B.

Take x being knife-shaped. The conditional counterfactual made true by this state of affairs, given the irreducible dispositionality of being knife-shaped, takes as F the property of being knife-sized, G the property of being made of steel, C is being pressed against paper in a certain fashion and B is the paper being cut by x. The counterfactual is conditional because another counterfactual is embedded within it – a counterfactual that would be true if a certain condition (the instantiation by x of certain properties) were to obtain.

It should be clear from what I have said so far how irreducibly dispositional properties ground metaphysically necessary laws. For example, the essential properties of an electron are its mass, charge and spin. Suppose that these are irreducibly dispositional, and take any one. One of the counterfactuals this property grounds, when instantiated by any x, will be that if x were to instantiate the other two properties, x would behave in certain ways in certain circumstances. If any x were an electron, for example, then it would behave in certain ways in certain circumstances. Here are our laws concerning electrons, grounded by any one of the irreducibly dispositional properties essential to electrons.

When you have the irreducibly dispositional properties essential to a given natural kind of object, you have the laws involving those kinds.Footnote 4 In any possible world with those properties, you have those laws. This is because the properties that make something a member of a certain natural kind cannot vary across worlds. And the dispositional nature of those properties cannot vary across possible worlds either – a property’s dispositional nature is what makes it the property it is. Therefore, the same laws hold of objects of a particular natural kind no matter what possible world such objects are in. Laws, in other words, are metaphysically necessary, not contingent.

There is an alternative. One could take each irreducibly dispositional property to support some counterfactuals concerning how objects bearing it – and others – would behave in certain circumstances, and thereby ground some laws, but not support all such counterfactuals, and so not ground all the laws it is involved in. I shall not concern myself with this position here. The view of irreducible dispositional properties I shall assume throughout is one in which they ground all the ‘general nomic counterfactuals’ they are involved in, not just some.

One might object that if properties are entities that support counterfactuals, they must be at least partly categorical. But I see no contradiction in the concept of irreducibly dispositional, non-categorical entities. The contrast, I have indicated, is between the entity (which is not itself a law, but is a property or relation) that is enough to ensure that various nomic counterfactuals are true of the object instantiating it, and the entity that is not. The former is dispositional in nature. Is there a problem saying that the nature of such an entity is exhausted by its dispositionality, the ways it leads those objects instantiating it to behave in certain circumstances given the instantiation of certain other universals? I cannot see why. There is the view that all properties are categorical, which Armstrong endorses. There is the view that properties are both categorical and dispositional, which Heil endorses. I see no reason to think the next view along – the view that all properties are dispositional – is incoherent or contradictory.

Pan-Dispositionalism is not currently in favour. Mumford (2004) seems open to the position but remains uncommitted. Molnar (2003) cites Shoemaker and Mellor as exponents, but this cannot be right: Shoemaker certainly gave the impression that he was in his 1980 paper but, in a later paper (Shoemaker 1998), is at pains to distance himself from the position; and while Mellor has claimed that all properties are dispositional because they support some counterfactuals, Pan-Dispositionalism conflicts with Mellor’s conviction [aired, for example, in Mellor and Oliver (1997)] that laws are contingent, as well as with other claims he has made, for example that properties are not ‘in themselves’ either dispositional or categorical (Mellor 2003). Even Swoyer, taken by many as a key exponent of Pan-Dispositionalism, has informed Armstrong, in personal communication, that he only endorses Moderate Dispositionalism (Armstrong 1997: 76).

A defence of Pan-Dispositionalism is long overdue.Footnote 5

2 Regress Arguments

2.1 The Ontological Regress

Armstrong has consistently denied that any property is a power. But he sets out the following ontological regress argument against the claim that all properties are powers:

If a property is nothing but its capacity to enter into nomic relations to further properties, the same must be said of these further properties and so on indefinitely unless we return in a circle to the original property or properties. No property is anything in itself but only in its relations to other properties as given by the laws of nature. But how can a system of things, each logically nothing in itself independently of the system, be made into something by incorporation in the system? (1983: 162)

…The power is constituted the power it is by the sort of actualisations it gives rise to in suitable sorts of circumstance. But what are these sorts of actualisation and sorts of circumstance? They themselves can be nothing but powers, and so again they can only be constituted by the sorts of actualisation which they give rise to in suitable circumstances. The power to produce A is nothing but the power to produce the power to produce B…and so on. Nor will the situation be relieved by bringing the powers around in a circle. (1983: 123)

This argument does not threaten Moderate Dispositionalism, since the regress can be said to terminate with those properties that are not powers. It is aimed specifically at Pan-Dispositionalism. Fortunately, however, the argument is no real threat to Pan-Dispositionalism either, since it presupposes a view of powers that Dispositionalism does not share.Footnote 6

Armstrong is right: if properties are taken to be powers and if powers are construed as ‘nothing but’ what will happen in certain circumstances, then we get ourselves in a vicious regress (or circle). If powers are ‘nothing in themselves’, as Armstrong says, then to say that x has a power P is to say merely that certain counterfactuals are true of x. These counterfactuals, in turn, will involve other properties – and therefore powers – in their antecedent and consequent clause, and these powers will also be ‘nothing in themselves’. To say that any particular, x, has these further powers is, again, to say that certain counterfactuals are true of x. And so on. We never get, from this procedure, to a stage where the powers are something in themselves, and since that is what we are trying to get, the regress is vicious.

But Dispositionalism does not – and should not – take powers to be ‘nothing in themselves’. Powers are entities that support counterfactuals. To say that x has power P is to say that x instantiates a property – a mind-independent entity – which makes certain counterfactuals true of x. If powers are like this, there is no question of an ontological regress, since the existence of powers is not threatened. At most, we would have a characterisation regress, namely that the nature of P has an infinite complexity because, by its nature, it supports certain counterfactuals, and the powers described in those counterfactuals by their nature support further counterfactuals, and so on ad infinitum.

However, there is no threat to the characterisation of properties, at least in principle, since, contra-Armstrong, bringing the powers around in a circle is of use. The nature of a property is not linear in the way an infinite regress dictates, but is circular. More specifically, what we have is a network of properties. We have a number of properties that are each nomically related to the others through the counterfactuals they support. That number may be infinite – there may, for example, be a continuum of values for temperature properties. But there is no problem, on the face of it, with the idea that a property may have a nature that involves certain other properties, even all properties. The nature of a bride is such that, to be one, there must be a groom, and vice versa, but that does not raise questions about the existence or characterisation of brides and grooms. The same applies to the nomic network as to the marriage network.Footnote 7

2.2 The Epistemological Regress

A second regress argument occasionally levelled at Pan-Dispositionalism seems to have been first advanced by Swinburne (1980) in connection with Shoemaker’s (non-current) view of properties as powers. This regress is epistemic rather than ontological: it does not threaten the existence of any property F, only our knowledge of it – our ability to identify or recognise F. Swinburne puts it like this:

For if [Shoemaker] is right, we could never come to know or even have a reasonable belief about what properties objects have – and often we do have reasonable beliefs about this. (1980: 316)

…if properties are nothing but potentialities to contribute to powers, one could only justifiably attribute such properties to objects if one had observed their effects. And so on ad infinitum. The regress is vicious. (1980: 317)

Swinburne also seems to be presupposing a ‘nothing but’ conception of powers. And the response is the same as it was for the first regress: if powers are ‘nothing but’ what will happen in certain circumstances, then there is indeed a regress. But powers are entities that ensure objects behave in certain ways in certain circumstances.

Suppose, however, that we recast the regress argument with this ‘entity’ conception of powers in mind. It would go as follows. All properties, as powers, support various counterfactuals, some (perhaps most) of which are conditional. To recognise an instance of F, we need to know that the object instantiating it, given the other properties it instantiates, would, if it were in circumstances C, exhibit behaviour B. But both C and B involve properties, and we must recognise these if we are to recognise F (not to mention the object’s other properties that determine what powers F bestows). But these properties also support further counterfactuals involving properties and to know these further properties…and so on. F also supports counterfactuals concerning the circumstances in which F would be instantiated. But again, those circumstances involve the instantiation of other properties we need to be able to recognise if we are to recognise F. In either direction, we find ourselves involved in a regress.

Endorsing a network view will not solve the problem of how we can know any one of these properties. Take a network-cum-circle of three properties, F, G and H. I only know F if I know G, I only know G if I know H and I only know H if I know F. It does not seem, in that case, that I can know any of the three properties, and the case we are considering – the world’s network – is only different in terms of the number of properties involved. We might know F if we knew all the laws that constitute the network. But knowledge of laws depends on our already recognising at least some properties. If we cannot get to know what properties there are, we cannot begin to establish what laws there are.

Fortunately, there is a problem with this second version of the epistemological regress argument. On the ‘nothing but’ conception of powers, it was clear why knowledge of one power depended on knowledge of others. On the ‘entity’ conception, it is not. As entities, there is no reason why the Pan-Dispositionalist cannot allow that some properties are known directly, without our having to have knowledge of other properties that cause them to be instantiated. Pan-Dispositionalism claims that properties, of their nature, are caused to be instantiated by certain other properties and, in turn, cause, in certain circumstances, other properties to be instantiated. But this metaphysical claim in itself says nothing about how we become acquainted with properties.

Evan Fales has suggested that the properties of our perceptual experience can be used to avoid the epistemic regress (1990: 224). We have a perceptual experience of a triangular shape, for instance, and the property of triangularity is recognised directly. Fales is happy to take such properties of our phenomenal experience to be non-physical. The physicalist will need to do a little more work but, in principle, there seems no reason why some plausible story cannot be told of how the regress is broken by (for example) direct knowledge of physical brain-excitation properties presented to us from the first-person perspective. It may be implausible (given multiple realisability) to think that only one brain-excitation property is presented to us by our perceptual experience of triangularity. But we can say something like this: with each such experience, one of a range (the ‘triangularity’ range) of brain-excitation properties is presented to us, and, using the hypothesis that a mind-independent world is causally responsible for our perceptual experiences, we can build outwards from knowledge of these brain-excitation properties to knowledge of those properties that are causally relevant to that excitation.

2.3 On Powers as Something in Themselves

In the Ontological Regress argument, Armstrong seems to be taking object x having power P to amount to no more than x having certain counterfactuals true of it. There is no entity to identify with P, nor are there entities to identify with the properties (powers) mentioned in the counterfactuals true of P, and so we have a regress.

As I have said, I think Armstrong was wrong to construe powers in this way – Dispositionalism is a something in themselves view of powers. Here are my reasons for claiming this.

First, if a power is nothing in itself, Armstrong might as well have dispensed with Dispositionalism on general truthmaking grounds before even considering regress objections. Take an object that is magnetic at time t and not magnetic at time t + 1. What makes the relevant counterfactuals concerning magnetic attraction true at t, but not t + 1? The object is the same. It must, therefore, be a matter of the object having different properties at t and t + 1. And for these properties to be truthmakers, they must be entities. Armstrong is a realist about properties and takes them to be universals and categorical in nature. If he wants to defend Categoricalism against its non-categorical alternative, he needs to defend it against another realist account and not one that by his own views on truthmaking is doomed from the start.

Second, the debate between Categoricalism and Dispositionalism is waged between realists about properties or those who at least accept property realism for the sake of the argument. And within that debate, the Dispositionalist’s conception of powers is usually taken as one in which powers are something in themselves: entities that help explain the behaviour of objects. It certainly appears that Armstrong is attributing to the Dispositionalist a leaner conception of powers when he raises the regress objection. But elsewhere this isn’t the case. He contrasts ‘Categoricalism’ with ‘Dispositionalism’ in A World of States of Affairs (1997), and I take the contrast there to be between the view that all properties are categorical and the view that all properties are dispositional, not between the view that properties are categorical and the view that there are no properties, just predicates attributable to objects because certain counterfactuals concerning those objects are true. Armstrong talks, for instance, of the Dispositionalist’s properties being like Ryle’s ‘inference tickets’, only they “exist in nature” (1997: 79). He also says they are truthmakers for dispositional truths (1997: 81). Truthmakers, of course, cannot be ‘nothing in themselves’.

Key Dispositionalists also see powers as something in themselves. For example, Molnar (2003) – as we shall see in more detail later – presents Dispositionalism as a realist account of properties, defending a moderate account in which properties are tropes and most, but not all, are irreducibly dispositional in nature. Swoyer (1982) argues that properties are irreducibly dispositional, and the nomic relation between two properties “is not some third entity over and above the two properties themselves” but is “a far more intimate, internal relationship grounded in the properties themselves” (1982: 217). And Ellis (2002: 74) agrees that dispositions such as elasticity and brittleness must be grounded in real properties but argues that these real properties are irreducibly dispositional, not categorical.

Not everyone has taken Dispositionalism to be a something in themselves view of powers. Richard Holton (1999), for example, seems to argue that, even though the view that all properties are powers entails that all truths are counterfactual, this does not mean the view is incoherent. But only a view in which powers are nothing in themselves could entail that all truths (or at least all truths about properties, given the identification of properties with powers) are counterfactual. If powers are entities, there are many truths about the properties objects have (e.g. a has power P), and these truths are not counterfactual. It therefore seems that Holton accepts the idea that Dispositionalism takes properties to be nothing in themselves. He makes clear that he is not an advocate of Dispositionalism, however, as he thinks it may suffer on other counts.

Heil (2003) also seems to have a blind spot where the something in themselves view is concerned. He argues for his ‘Identity Theory’ of properties via the consideration and rejection (2003: Chapter 10) of the view that properties are “pure powers”, and pure powers, he claims, bestow powers upon an object “but are nothing in themselves” (2003: 98, 109). If this were the only way to construe powers, then he would be right to criticise the position on the grounds that “a non-qualitative world is a world devoid of concrete objects” (2003: 102) and “a rejection of intrinsic qualities looks like a rejection of a substance-attribute model of objects” (2003: 108).

There is no reason to construe powers in this way, however, and there is every reason not to. Taking powers to be entities, we are not led from Dispositionalism to either the view that the world is devoid of objects or the falsity of the substance-attribute model of objects. And taking powers to be entities locates Dispositionalism and Categoricalism as opposing views within the general realist debate about the nature of properties.

It is not clear to me that either Holton or Heil has actually considered the idea that powers are something in themselves. But it is only if powers are taken in this way that Dispositionalism is both the right theory to contrast with Categoricalism and a serious alternative to it.

3 Ellis and Structural Properties

Ellis rejects Pan-Dispositionalism. He thinks that structural properties are clearly non-dispositional.

We need to be clear from the outset what Ellis means by ‘structural properties’ here, since this term is used in at least two different ways in the literature. Armstrong, for example, takes being H 2 O to be a structural property because, according to him, it has as its ontological constituents properties (being hydrogen and being oxygen) and component relations (e.g. of bonding) which hold between the atomic parts instantiating these properties. Others deny that being H 2 O is itself structured in this way, but still call it a structural property because, in order for it to be instantiated, its bearer must have certain parts in certain relations and with certain properties. Either way, these are not the structural properties Ellis is referring to. We might put the distinction as that between those properties that are structured, or depend on structure, and those that are properties of having a certain structure (e.g. being triangular). When Ellis says that structural properties are categorical, it is the latter kind of property he has in mind.

Ellis distinguishes between two sorts of structural properties (2002: 69–70):

Block structural properties

If S is a block structure (and so S has a block structural property), S has independently identifiable constituent objects.

Intrinsic structural properties

If S is an intrinsic structure (and so S has an intrinsic structural property), S has parts that are incapable of independent existence.

Molecular structures are examples of block structures. S is a molecular structure iff S has as constituents independently identifiable atoms related in the appropriate way. An electro-magnetic field structure, on the other hand, is an intrinsic structure; its constituents are electrical and magnetic forces spatiotemporally distributed in a configuration as determined by Maxwell’s equations, and these forces, or potentials, cannot exist independently of the structure.

Ellis is not explicit about his reason for taking intrinsic structural properties to be categorical. Regarding block structural properties, however, he says the following:

Now these block structural properties are clearly not just dispositional. It may be true that an atomic or a molecular structure of a given kind exists if and only if there is some atom or molecule that is disposed to behave in a certain way in appropriately specified conditions. But that is not what makes it an atom or molecule of this kind. Its essence is structural, not dispositional. It is, of course, only from the behaviour of an atom or molecule that we can infer its structure. But the structure exists independently of the disposition to behave this way. (2002: 69)

Ellis’s point seems to be this. Even if there could not be a block structural property without there being an atom or molecule disposed to behave in certain ways, this does not mean that the structural property must itself be dispositional. It is not.

The problem with this, however, is that no argument has been provided for this latter assertion. Why is it not? His further claim – that we infer an atom or molecule’s structure from its behaviour, but that its structure is independent of its disposition to behave certain ways – can hardly be challenged. It is obviously true that an atom or molecule’s block structural property exists independently of the disposition to behave in all the ways that atom or molecule behaves, given that atoms or molecules of different natural kinds, with different behaviours, can nevertheless have the same structure. But that alone is not enough to show that block structural properties are categorical. If it were shown that there are no dispositions shared by all metaphysically possible atoms or molecules with the same block structural property, S, and which could be identified as the dispositional essence of S, then S would be shown to be categorical. But Ellis has not shown this to be the case and, indeed, cannot take this to be the case. Structural properties are involved in laws. But if laws are metaphysically necessary, as Ellis believes, then there must be some counterfactuals true of all actual and metaphysically possible objects with the same structure, and which might be said to set out the dispositional essence of the corresponding structural property.Footnote 8

Consider methane (CH4) and carbon tetrachloride (CCl4), which have the same molecular structure, call it S. A property P of object x is a power iff certain counterfactuals are true of x and made true simply by x having P. Now block structural property S (and any block or intrinsic structural property) might be said to make certain counterfactuals true of its bearers. For instance, we might say that when an object x has S, this makes true counterfactuals of the following two forms:

  1. 1.

    If x were to have a CH4 combination of parts, then if x were in C, B

  2. 2.

    If x were to have a CCl4 combination of parts, then if x were in C, B

Methane and carbon tetrachloride molecules, in virtue of their instantiating S, make true conditional counterfactuals of both forms. This does not stop molecule x being CH4 making true different counterfactuals to x being CCl4. But we can say that x being S makes true one set of counterfactuals, x being CH4 makes true another set of counterfactuals, and x being CCl4 makes true a third set.

Ellis is free to deny that structural properties ground counterfactuals of these forms. He can say, for instance, that while counterfactuals of forms 1 and 2 are true of x, what makes them true is the dispositional nature of the properties borne by the parts of the object with structure S. What makes claims of form 2 true of methane molecule x are the properties being C and being Cl, because it is of the nature of these properties that anything instantiating them in a CCl4 configuration (and so instantiating S) is such that if x were in C, then B. In other words, being C and being Cl are powers, but the structural property being S is not. Ellis can say this, and so take there to be both categorical and dispositional properties. But still lacking here is an argument for making this ontological distinction.

There is, however, an argument – which I flagged up in the introduction – for taking S to be a power. With Pan-Dispositionalism, all properties are powers; with Moderate Dispositionalism, properties are either powers or non-powers. Pan-Dispositionalism is simpler in terms of the kinds of entity it posits and therefore (all else being equal) is to be preferred.

One might try to counter this argument by pointing out that the powers posited by Ellis are enough to explain the way objects behave; taking structural properties to be powers on top of this merely multiplies explanations of behaviour beyond necessity. However, first of all, Occam’s razor is concerned, at least primarily, with reducing the number of types of thing. The number of token explanations would seem of less concern than the number of types of entity. Second, whilst it may well be true that non-structural properties are all we need to explain the behaviour of objects, this cannot support Ellis’s claim that structural properties are categorical. This is because structural properties alone, taken as powers, also seem able to do the explanatory job equally well. According to Pan-Dispositionalism, each structural property S has a nature making true various counterfactuals of the following form, where x is the object with S: if the parts of x were to have certain properties and be in certain relations to one another, then if x were in circumstances C, there would be behaviour B. Because these counterfactuals are conditional on the instantiation of various properties other than S, the counterfactuals supported by any one property cover all the other properties it might be instantiated alongside. For (powerful) structural properties to then be sufficient to explain the behaviour of all objects, we need only assume that all objects either have a structure or are part of some structure – and this is surely hard to deny.

4 Molnar and S-Properties

Molnar (2003) also rejects Pan-Dispositionalism. Some properties, he thinks, are non-powers (he prefers the term ‘non-powers’ to ‘categorical’). These are positional properties – the properties of spatial/temporal location (being at spatial position s, being at temporal position t) and spatial/temporal orientation (being spatial distance S away from x, being temporal distance T away from event e) – and properties concerning the numerical identity of parts (having p, q… as mereological parts) (2003: 160).

Molnar calls these ‘S-properties’. They are properties associated with a symmetry operation, which he defines as “an operation on a particular that at its conclusion leaves all but one of the salient physical properties of the particular unchanged” (2003: 160). For example, if you move an object in space or time (the symmetry operation), the particular retains all its physical properties apart from its spatial or temporal position (the S-property). If you interchange qualitatively identical parts (the symmetry operation), the particular changes its physical numerical-identity-of-parts property (the S-property).

Molnar seems to think it is just obvious that S-properties are non-powers:

Both common sense and science recognise the existence of symmetry operations. In symmetry operations, some property of an object undergoes change (the ‘S-property’), but all the powers of the object are left unchanged. It is this fact which gives us an epistemic handle on non-powers: one discovers empirically what symmetry operations there are, what the relevant S-properties are, and so one distinguishes between the actual powers and non-powers…Pan-Dispositionalism is not impossible but it is false. (2003: 181)

But Molnar’s description of events here, in which all the powers remain the same but the S-property changes, clearly begs the question against Pan-Dispositionalism. He gives us no reason in this quotation to deny that the existence of S-properties is compatible with Pan-Dispositionalism. And the two are compatible, as I shall now show.

Let me deal with numerical-identity-of-parts properties first. Take object x, which has three parts p, q and r. Molnar’s argument seems to be that if we swap p with some qualitatively identical part, p*, there is a change of truths about x and so we should say there is a change in its properties: first, x had the property of having p as a part, now it has the property of having p* as a part. And these properties are clearly not powers. The powers of x remain the same.

The problem with this line of argument is that it is just not clear that if you change a part of x, x now possesses a different property. If part p is swapped for part p*, it is true that x has a part p*, but it seems more plausible to say that this truth is made true by x and its parts, rather than by x now having the property has a part p*. As Molnar has himself made essentially this sort of move with regard to relational dispositional properties – claiming that there are no such properties, only relational dispositional predicates – it is not at all clear why he does not employ it here and say that there are no numerical-identity-of-parts properties.

I do not propose to deal with positional properties in the same way. It seems clear that some positional properties (or relations) are needed if we are to have truthmakers for positional truths. It is less clear, however, that these properties need to be non-powers. They are certainly non-powers if we accept Molnar’s definition of powers as properties with the following five features:

  1. 1.

    Directedness – a power ‘points to’ its ‘characteristic’ manifestation (e.g. the characteristic manifestation of solubility is dissolving)

  2. 2.

    Independence – a power must exist for there to be a manifestation, but not vice versa

  3. 3.

    Actuality – powers are actual entities, and not reducible to possibilities

  4. 4.

    Intrinsicality – powers are intrinsic properties; if x has a power P, its having P is not dependent on the existence of any object other than x

  5. 5.

    Objectivity – powers are mind-independent entities

Positional properties do not have feature 4. They are extrinsic, not intrinsic: whether x bears positional property P is dependent on the existence of other particulars apart from x. Object x cannot have the property of being two miles away from y if y does not exist, and event e cannot have the property of being two days away from event e* if e* does not/did not exist. Similarly, x cannot bear the property of being at spatial/temporal position s/t if either some other object does not exist (given a relational view of space and time) or some space/time point s/t does not exist (as with a non-relational view of space and time). Therefore, if Molnar’s characterisation of powers is correct, positional properties are not powers.

Molnar denies that a conditional characterisation of powers such as the one I have assumed thus far will work. Powers do not entail conditionals, he claims, and so we need to look elsewhere for a characterisation of powers. Fortunately for Pan-Dispositionalism, this is not true.

Molnar’s consideration of conditional accounts focuses on two proposed definitions of what it is for an object to possess a disposition. Both begin with: ‘x is disposed at time t to give response r to stimulus s iff…’ The Naive Conditional Analysis adds ‘if x were to undergo stimulus s at time t, then x would give response r.’ The Causal Conditional Analysis adds ‘x has some property F that would cause x to give r if x were to undergo stimulus s at time t.’

Molnar has a number of arguments against these analyses, but his main argument, and the important one here, is that the terms on each side of the ‘iff’ are not equivalent. This is a familiar objection that comes from Martin (1994). Suppose x is disposed to give you a shock if you touch it. This is not equivalent to the proffered conditional – if you were to touch x, you would get an electric shock – because though x is disposed to give you a shock if you touch it, it may be fitted with a device (an ‘electro-fink’) that stops it from being live the instant you do touch it, and so renders the conditional false. Similarly, a device may be attached to an object x, which makes it live only at the instant you touch it; x would then not be disposed to give you a shock (i.e. it is not live), but the conditional is true of x (i.e. if you were to touch x, you would get a shock). The problem is the same whether we consider the Naive Conditional Analysis or the Causal Conditional Analysis (with F, in the latter, the property of having an electrical current running through it).

I agree with Molnar so far, but not on the conclusion he draws regarding powers entailing (rather than being analysed in terms of) conditionals. Molnar seems to think that to be able to characterise powers in terms of their entailing conditionals, we should be able to take specific powers and point to what conditional or conditionals each entails. And as we cannot do this, for the reason just given, he thinks that any conditional characterisation of powers is doomed.

However, nothing this strong is required. Perhaps we cannot state the conditionals entailed by any specific power. The problems Molnar raises with finks may indeed show that much. But this fact about our cognitive limitations does not show that there are no conditionals entailed by x having power P i.e. that there is no set of conditionals entailed by x having P which are of the form if x were to instantiate other properties F, G…, then if x were in circumstances C then there would be behaviour B. The property of being disposed to give you a shock will not, instantiated by x, make it true of x simply that if you were to touch it you would get a shock; it will make true a number of highly complex conditionals according to some of which you would not get a shock if an ‘electro-fink’, of whatever construction, were to be switched on at the moment you touch x.Footnote 9

The set of conditionals entailed by the instantiation of any specific power will be complex. But this does not rule out a definition of powers in general as those entities that entail conditionals – more specifically, counterfactual conditionals – since this does not require precise details of those conditionals entailed by any particular power.

If CB Martin had been right, Dispositionalism characterised in terms of entailed counterfactuals would be doomed. This is not to say that Dispositionalism itself would be doomed: its key, defining feature is the idea that objects behave the way they do in virtue of the nature of the properties they have, and while I have tried to flesh out this nature using counterfactuals, there are other ways, as Molnar’s account demonstrates.

But Martin is not right. He points out that there are so many ways a ‘finkish’ device could work that we cannot possibly amend our candidate counterfactual to accommodate them. The moral: we cannot analyse dispositional ascriptions in terms of counterfactuals. So far, we have a semantic claim – a claim about the meaning of dispositional terms. But Martin then goes further, claiming that it shows that dispositions themselves cannot entail the truth of any counterfactual.

It is this move from the semantic to the metaphysical that is unjustified. Just as epistemological arguments for metaphysical conclusions are often considered dubious (why should what we can or cannot know tell us what reality is like?), it is hard to see why this semantic argument about the meaning of dispositional terms should lead to a metaphysical conclusion about dispositional properties (and, given Pan-Dispositionalism, to all properties). Why does Martin’s example not simply show us that x having property P entails the truth of counterfactuals about x that are just extremely complex and quite impossible to state given the limitations of our knowledge?

One might try and mount an objection to this idea in the following way. According to Dispositionalism as I have characterised it, if x has irreducibly dispositional property P, this entails that there are certain counterfactuals true of x. The antecedents of these counterfactuals will be complex, involving the instantiation of many other properties. Since there are no negative universals, the antecedent will involve a total positive specification of the circumstances that will ensure that the state of affairs type given in the consequent is instantiated. In many cases, the properties involved in these circumstances will be at some point in time instantiated and so (given immanent realism) exist. But there may well also be counterfactuals we want to say are true of an object in virtue of its instantiating P, but which involve a non-existent property in the antecedent or the consequent. For instance, it seems plausible that there might be true counterfactuals involving anything with P such that, if it were in certain circumstances, a simple property F would be instantiated – but since these circumstances never (as a matter of contingent fact) obtain, and nothing else can cause an instance of F, then F is never instantiated.

With this in mind, one might claim that we should at least give up on the idea that there is a set of complex counterfactuals entailed by an object’s instantiating P and which catalogues its complete nature. For what can make true counterfactuals involving uninstantiated and (assuming immanent realism) therefore non-existent entities? If there are no truthmakers for these counterfactual truths, then there are no such counterfactual truths.

However, even if we accept the claim of this last sentence, Dispositionalism is unaffected. Armstrong has a problem. He takes truths about powers to be grounded in (and reducible to) truths about categorical properties, and counterfactuals about how an object would behave in certain circumstances to be made true by nomic relations between properties. But there cannot be relations where one of the relata is non-existent, and so there cannot be a truthmaker for a counterfactual about what (as a matter of fact, never instantiated) simple property would have been instantiated if certain circumstances, never actual, had in fact obtained. This is the problem of uninstantiated universals that Armstrong confronts in What is a Law of Nature? (1983), and which I briefly mentioned in the introduction. Anyone endorsing Categoricalism and immanent realism will have to deny that there are such counterfactual truths, and this denial seems counter-intuitive.

Someone endorsing Dispositionalism, on the other hand, faces no such problem. A property is the truthmaker for many general counterfactuals about how objects instantiating it would behave given certain circumstances. But while properties must be instantiated if they are to be the truthmakers for counterfactual truths, this does not mean that all the properties mentioned in those counterfactual truths need also be instantiated. A truth, or fact, is simply that which a true statement, possible or actual, states (see Mellor 1995: 161). And there can be true possible statements about metaphysically possible but contingently non-existent properties.Footnote 10

I conclude that Molnar’s attack on the claim that powers are those properties that entail counterfactuals is unsuccessful. As a consequence of this, his alternative characterisation of powers, in terms of the ‘five features’, can be resisted.

This is not to say that someone who endorses the counterfactual characterisation of powers outlined in Section 1 must deny that powers have any of the five features Molnar highlights. Objectivity and Actuality are an integral part of the characterisation. And as long as one defines the ‘manifestation’ of powers in terms of entailed conditionals, both Directedness and Independence seem acceptable. The Pan-Dispositionalist must reject Intrinsicality, since positional properties are extrinsic. But this is no great loss, since anyone allowing that relations might be powers must reject Intrinsicality – the claim that powers are intrinsic properties – anyway. While it is true that we usually talk about a power of something, and this may lead one to think that all powers are properties, I have used the term power only to mean an irreducibly dispositional entity: an entity that makes true certain counterfactuals and which has no parts that do not make true certain counterfactuals. Nothing has been said so far that prevents there from being relations that fit this category.Footnote 11

Molnar also denies that positional properties have Directedness and Independence. If this were so, Pan-Dispositionalism – taking positional properties to be powers – would then need to deny that powers possess these two features. But positional properties can be said to have both, I think, given a counterfactual account of powers.

Take object x, at point p, which is near object y. According to Molnar, x has the positional location property being at point p and the positional orientation property being distance D from y. The Pan-Dispositionalist can claim that each of these is a power. Object x having being at point p can be said to make true a number of counterfactuals about what would happen if certain other objects were to be at various spatial points in the vicinity of p: one being, for example, that if x were magnetised, and at point p, then if another magnet z were placed nearby, at p + 1, there would be a certain force F between the x and z. Object x having being distance D from y can also be said to make true a number of counterfactuals: for example, that if x were magnetised and if y were magnetised, there would be a certain force F between x and y. Whilst neither property has a ‘characteristic’ manifestation, as Molnar takes powers to have, each can be said to have Directedness in this sense: each power supports various conditional counterfactuals, and the outcome presented in the consequent of each embedded counterfactual (the second if, then) is a possible manifestation of the power. Both properties can also be said to have Independence, in the sense that no one manifestation needs to be manifest for the property to be instantiated. There does not need to be a magnetic force between x and y, for example, for being distance D from y to be instantiated by x.Footnote 12

Positional properties, I have argued, can reasonably be construed as powers. And there is no reason to think that there are numerical-identity-of-parts properties. It appears, therefore, that Molnar has not given us sufficient grounds to reject Pan-Dispositionalism.

5 Conclusion

I have argued that Pan-Dispositionalism fares favourably against Moderate Dispositionalism. The regress arguments levelled against Pan-Dispositionalism can be resisted. Both Ellis and Molnar’s candidates for non-powers can be construed as powers. And Pan-Dispositionalism is actually simpler than Moderate Dispositionalism, inasmuch as it posits only one kind of property: the powerful kind. In the absence of further, more compelling arguments against it, and in light of this greater simplicity, Pan-Dispositionalism should be considered the better theory.