Singular Clues to Causality and Their Use in Human Causal Judgment
It is argued that causal understanding originates in experiences of acting on objects. Such experiences have consistent features that can be used as clues to causal identification and judgment. These are singular clues, meaning that they can be detected in single instances. A catalog of 14 singular clues is proposed. The clues function as heuristics for generating causal judgments under uncertainty and are a pervasive source of bias in causal judgment. More sophisticated clues such as mechanism clues and repeated interventions are derived from the 14. Research on the use of empirical information and conditional probabilities to identify causes has used scenarios in which several of the clues are present, and the use of empirical association information for causal judgment depends on the presence of singular clues. It is the singular clues and their origin that are basic to causal understanding, not multiple instance clues such as empirical association, contingency, and conditional probabilities.
“Causal understanding” is a term that could cover a wide variety of things. The impression of causality we have when we observe a moving billiard ball contact a stationary one and the latter then moving off; the impression we have of making something happen ourselves, such as when we lift a cup of tea or, more indirectly, press a button on a remote control to activate a television; the inference that a bad cold that someone has caught was caused by a long walk on a cold and rainy day or, more generally, the belief that colds can be caused by such walks; the inference that our bumper crop of tomatoes this year was caused by the new fertilizer we added to the soil; it could be said that causal understanding is involved in all of these different kinds of processes and impressions. Our understanding of causality could include the potential use of various cues or clues that guide inferences about both specific occasions and general causal mechanisms: The cues or clues may function as guides to the detection of causal mechanisms that are themselves covert, unobservable. Certainly, many and perhaps all causal mechanisms are unobservable, so there is a need for inferences about causality to be guided by what can be observed, and that is where indirect indicators can be useful.1
The literature on causal inference has long been dominated by the hypothesis that the main guides to identifying causes are a set of cues proposed by Hume (1978): regularity of association (in modern terms, contingency or covariation information or conditional dependence), spatial and temporal contiguity, and temporal priority (the rule that a cause cannot be temporally subsequent to its effect). That literature includes models of causal inference from contingency information (Allan, 1993; Cheng, 1997; Cheng & Novick, 1992; De Houwer & Beckers, 2002; Hattori & Oaksford, 2007; Perales & Shanks, 2007; Shanks, 1995) and models based on Bayesian analyses of causal structure (Gopnik et al., 2004; Griffiths & Tenenbaum 2005, Griffiths & Tenenbaum, 2007, 2009; Holyoak & Cheng, 2011; Lu, Yuille, Liljeholm, Cheng, & Holyoak, 2008; Perales & Catena, 2006). Authors such as Cheng (1997) have argued that the use of multiple instance information, accompanied by a small amount of innate knowledge, supports causal induction and therefore functions as the source of causal knowledge. Vast quantities of research studies have been carried out to test models and hypotheses formulated within this general framework (see, e.g., Hattori & Oaksford, 2007; Holyoak & Cheng, 2011; Perales & Catena, 2006; Perales & Shanks, 2007). Other analyses of causal knowledge have been proposed (White, 1989; Wolff, 2007; Wolff & Song, 2003; Wolff et al., 2010) but have been, to date, less influential.
In this article, I put the case for a different set of clues to causality. I propose a specific point of origin for the understanding of causality; that point of origin has a set of consistent associated features; and those features are then abstracted from the point of origin and used as guides to identifying causes in other areas. Specifically, I argue that knowledge of causality originates with single experiences of acting on objects. The idea that causal understanding originates in human action has a long history in psychology (De Biran, 1942; Heider, 1944; Michotte, 1963; Piaget, 1954; Uzgiris, 1984; White, 1988). Most prominent in that history is the argument made by Piaget (1954) that causal understanding originates in a feeling of efficacy that occurs at the point of culmination of an action. Recent developments in the understanding of voluntary motor output and haptic feedback support a more precise and detailed account of the experience of causality in action (White, 2012a,b), and recent research findings favor action over other contenders as the point of origin of causal knowledge (Muentener & Carey, 2010; Muentener & Lakusta, 2011).
I show that consistent features of actions on objects function as clues to causality that can be used for causal judgment in general. I shall refer to these features as singular clues to causality because they can be detected in single instances. I also argue that other kinds of clues to causal judgment such as mechanism and regularity clues can be viewed as derived from the basic singular clues; that the use of regularity information in causal judgment is actually dependent on the presence of multiple singular clues; and that singular clues to causality function as heuristics aiding judgments made under conditions of uncertainty.
I begin with a summary of my proposal about the origins of causal understanding. Then comes the catalog of singular clues to causality, the derivation from these of other kinds of clues such as mechanism and regularity, and implications for studies of the use of contingency and conditional probability information in causal judgment.
2. Origins of causal knowledge
I have argued that the origins of causal knowledge lie in experiences of our own actions on objects. This view has been articulated in full elsewhere (White, 2009a, b, 2012a, b), so I shall present a brief summary here, enough to set the scene for a consideration of clues to causality, which are the central focus of this article.
When someone carries out a planned action, the motor activity is guided in part by an internal model of the action (Blakemore, 2003; Blakemore, Frith, & Wolpert, 1999; Blakemore, Wolpert, & Frith, 2002; Desmurget & Sirigu, 2009; Frith, 2002; Grush, 2004; Haggard, 2005; Wolpert & Flanagan, 2001). The model includes an efferent copy, which is a model of the planned motor output, and a forward model, which is a model of the anticipated sensory consequences of the action. These models include information about, among other things, the forces to be exerted by the actor as the action unfolds and the anticipated forces that the object acted on will exert on the actor. Anticipation of forces encountered as the interaction unfolds is necessary to ensure successful completion of the action (Flanagan & Wing, 1997; Johansson & Flanagan, 2009). When we turn over a piece of paper, the forward model represents the anticipated amount of resistance the paper will offer to being turned over and the force exerted in the action is guided by that. Thus, if what appears to be paper is in fact made of some heavy metal, the action will fail because the force exerted will not be adequate.
Actual sensory feedback about the forces involved in the interaction is obtained through the haptic system. The haptic system comprises receptors in the skin, joints, and muscles that are stimulated by movement of the body and by contact or interactions with other objects (Carello & Turvey, 2004; Gandevia & Burke, 1992; Gelfan & Carter, 1967; Gibson, 1962, 1966; Goodwin, McCloskey, & Matthews, 1972; Johansson & Flanagan, 2009; Matthews, 1982). The haptic system is a sensory system that registers information about forces. These give rise to patterns of neural impulses that result in perceptual impressions of forces (and higher perceptual constructs such as object properties) in the brain (Dijkerman & de Haan, 2007). This is particularly important if the forces exerted by the object acted on do not match those represented in the forward model. Haptic input, like visual input, is coupled to motor output through the forward model and the comparison processes that detect discrepancies between anticipated and actual sensory feedback. In that case, haptic input can guide appropriate adjustment to the action (Desmurget & Sirigu, 2009; White, 2012a,b). (Readers can experience this kind of adjustment for themselves by picking up an object of unknown weight with eyes closed.)
As far as causality is concerned, actions on objects provide us with the experience of causing. We experience a causal relation when acting on an object, not only because of our knowledge of what we are putting into the action but also because of haptic input about the effects of our action on the object. If we push a ball with our hand, for example, we feel ourselves move the ball, in part because of the internal model of the motor output and in part because of haptically mediated information about the change in state of the ball while it is in contact with the hand. What the haptic system provides us with in that respect is most clearly revealed by the problems encountered by those whose haptic systems are not functional because of neuropathy (Cole & Sedgwick, 1992; Gallagher & Cole, 1995; Rothwell et al., 1982). These people are severely handicapped in their interactions with objects and have no experience of bringing about effects in objects by their actions on them. They do still have efferent and forward models and judgments about observed actions can still be guided by those (Lafargue, Paillard, Lamarre, & Sirigu, 2003), but they lack the experience of causality that is provided by haptic feedback of outcome information, closing the loop between planning and knowledge of execution.
What we learn about causality by acting on objects does not appear to be dependent on any other kind of knowledge. Causality is part of what the motor and sensory information about acting on an object gives us. Even if we had no pre-existing idea of causality, we could acquire one from experiences of acting on objects. We have such experiences right from the start of life. The haptic system is functional at least from a few hours after birth and can be used even at that age to ascertain object properties (Rochat, 1987; Streri & Gentaz, 2003, 2004). Neonates can also act on objects, albeit with a limited repertoire of actions (Bushnell & Boudreau, 1993; Hernandez-Reif, Field, Diego, & Largie, 2001; Jouen & Molina, 2005; Molina & Jouen, 2004; Sann & Streri, 2008). It is therefore possible that, if we are born without any idea of causality in the world, we can acquire one from actions on objects from earliest times.
When we act on an object, Newton's third law applies and the object exerts as much force on us as we exert on the object. However, as long as the sensory feedback is consistent with the anticipated sensory feedback in the forward model, the force exerted by the object on us is neglected. The experience is of efficacious force operating in only one direction, from actor to object. Two points should be noted about this. One is that actions can be efficacious, despite Newton's third law, because there is net transfer of momentum from actor to object. The object acted on is usually much less massive than the actor, and moreover the application of force can be sustained by the actor because it is powered by an internal source of energy. In addition, the relatively low mass of the object acted on means that the effect of the object on the actor tends to be smaller than the effect of the actor on the object, which adds to the impression of force acting just in one direction. If we push a ball, the hand accelerates the ball a lot but the ball decelerates the hand only a little, and that deceleration is overcome by extended application of force. The second point is that sensory feedback is heeded when it is discrepant with the representation in the forward model, but that happens only when the action does not proceed as planned, such as when the action fails in its goal of moving the object, and of course that would not be an experience of causation because the action is not efficacious. The appearance of force acting in just one direction is therefore common, if not universal, in efficacious actions on objects.
In the remainder of the article, I seek to defend three main claims: (a) that actions on objects are not only the source of the idea of causality but also the main source of clues to causal judgment; (b) that there are many more clues to causal judgment than has previously been thought; and (c) that, in explicit causal judgments, clues function as heuristics, guiding judgment under uncertainty. As such, they tend to be applied too widely, leading to errors in causal inference. That is mitigated, to some degree, by a preference for using clues in combination rather than individually. In essence, an interaction will tend to be identified as a causal relation, or a cause of some outcome will be identified, according to the extent to which the features of the interaction resemble the features of actions on objects.
3. A catalog of singular clues to causality
I have argued that actions on objects are the main source of clues to explicit causal judgment. The derivation of clues to causality would happen via processes of abstraction of features of actions on objects from their original context. Abstraction of features is an emergent property of episodic trace models of memory. For example, Hintzman (1986) proposed a computational simulation as a model of memory in which experiences yield individual, episodic memory traces. A retrieval cue tends to activate memory traces according to their degree of similarity to the cue. Although all features of a given memory trace are equally activated, the resultant representation predominantly represents features that are shared by most or all of the traces, and traces that are unique to one or a small proportion of traces are more weakly represented. Hintzman showed that the model was capable of generating concepts in a schema-abstraction task. I would not claim that Hintzman's model is the last word on the subject (see also Jamieson, Crump, & Hannah, 2012), but it does illustrate how relatively abstract representations such as categories can fall out of a process of activation of episodic memories. The abstract representations are then available to explicit judgment. Thus, features common to many actions on objects are likely to be abstracted as related to causality. They are then available to explicit causal judgment processes where they function as clues to the identification of causes. I therefore begin the catalog by listing features shared by most if not all actions on objects.
I do not seek to argue that explicit causal judgment is narrowly bound by adherence to the features of actions on objects. For example, temporal immediacy is a feature of direct actions on objects, but as adults we do not suppose that all causes produce effects immediately: We can learn that some causal relations involve slow operations. In addition, once we have the idea of causality, we can acquire novel causal beliefs, for example, through education, and causal judgment can then be influenced by these. Education in Newtonian physics can result in correct answers being given on naive physics problems where a judgment based on an understanding of actions on objects would be in error. Causal reasoning by scientists is liberated, to some degree, from dependence on singular clues by education and professional experience. I do argue, however, that, other things being equal, a cause possessing a given feature of actions on objects will tend to be preferred over one that does not possess that feature; that even those who have been educated in the relevant physics will tend to fall back on an understanding of actions on objects under some circumstances (Kozhevnikov & Hegarty, 2001) and that, the more physical theories deviate from an understanding of actions on objects, the harder they are to learn and apply correctly. Development of causal understanding involves, in part, gradual movement away from strict adherence to the features of actions on objects, but there is a persistent tendency to prefer causes that possess those features over ones that do not.
- Human action as cause. Originally, causality is experienced as our own actions on objects. As described above, experiences of actions on objects tend to be stored in long-term memory as episodic traces, subject to forgetting. These traces comprise temporally co-ordinated representations of several kinds: the efferent motor commands, or their representation in the efferent model, the forward model of anticipated sensory consequences of action, sensory feedback in several modalities, principally visual and haptic, temporal correspondences between these as the action unfolds, and antecedent mental states such as intentions, beliefs, and desires. Part of that information is kinematic information: information about the motions of body parts, particularly those parts directly in contact with the object, and about the motion of the object. As described more fully in White (2012a, b), visually perceived kinematic information about interactions between objects, including other animate beings, is matched to kinematic information in the stored episodic traces. These traces are activated and the remainder of the information in them functions to fill in gaps in the available information about the interaction. Critically for present purposes, the filled-in information includes specification of forces and causality. This process may help us to interpret and predict the actions of other humans, but it is also involved in the interpretation and prediction of any interaction if there is sufficient resemblance between the perceived kinematics of the interaction and stored episodic traces of our own actions on objects. Thus, we tend to perceive interactions between objects as being like our own actions on objects, if the kinematics match. To that extent, human action is a model of physical causality (White, 2012a, b).
- Two perceived objects. This applies specifically in the case where no human actor is involved; an example would be a perceived collision between two billiard balls. I am using “object” with fairly broad reference: For the purposes of the questionnaire study reported later, it may be taken to include such things as light rays, quantities of water, and avalanches, as well as more conventional objects such as automobiles and billiard balls.
- Prior activity of actor (causal entity). In an action on an object, the actor is potentially active, and actually active when initiating the action. The precise definition of activity is not easy to pin down. In actions, it refers to internally generated motor activity directed at the object. In causal relations not involving human actors, the motion need not be internally generated: An object might, for example, roll down a hill under gravity and strike a stationary object at the bottom. It is likely that this would be perceived as a causal relation, because it is known that impressions of causality do occur when stimuli are presented in which a moving object contacts a stationary one and the latter then moves; this is known as the launching effect (Michotte, 1963; Scholl & Tremoulet, 2000; White, 2009a,b, 2012a,b). There can be kinds of activity other than motion of an entire object. These may include changes of state, such as water boiling, or motion of parts of an object (Rakison, 2006). Any could fall under the definition of activity. The main implication, however, is a preference for identifying active objects as causes. The search for a cause is, to some degree, a search for an active object.
- Prior to the action, the object acted on is not active. In most cases, this means that the object is stationary and that none of its constituent parts is in motion. Of course, we can act on moving objects, as when we catch or bat away a projectile, but it is not clear that such interactions are experienced or understood as causal relations. When an object is in motion prior to the interaction, its motion is of obvious relevance to the success of the action. Successful interception of a projectile requires a representation that extrapolates the projectile's motion beyond the point at which it is currently perceived (Zago & Lacquaniti, 2005). The mass of the projectile must also be anticipated: If the projectile is heavier than expected, not only will the interception attempt fail, but the extremity it contacts may even be damaged. This implies that such interactions are experienced as two-way, the object acting on the actor as well as the actor on the object. It remains possible that they have features that are influential in perception and judgment of other interactions in which both objects are in motion prior to contact, such as a collision between two vehicles, but that is an open question at present. The main point is that they are likely to be experienced as interactions, not as straightforward actions on objects. So, for understanding causality, the object acted on is not active.
- Contact between actor and object. Contact is a feature of every direct action on an object. There are examples of indirect actions on objects that occur more frequently later in development, and that may also be influential for explicit causal judgment. These include tool use and certain kinds of action across a gap, such as blowing out a candle flame. I shall argue later that these are important to the understanding of generative transmission between physical systems.
- Influence perceived as operating in one direction only. I shall use the term “monodirectional” for this. I have already argued that actions on objects are usually experienced as monodirectional, even though they are not actually monodirectional according to the laws of physics. Whether perceived or judged monodirectionality functions as a clue to causality, or is a quality attributed to an interaction that has already been interpreted as a causal relation, is not clear at present. It is, however, a striking feature of our understanding of causal relations that we regard them as monodirectional (see, e.g., Pearl, 2000). By that I mean specifically that, when someone believes that A is causing something in B, she tends to believe that B is not simultaneously causing something in A. B may, of course, be believed to react back or otherwise influence A subsequently (this is a common misinterpretation of Newton's third law, according to diSessa, 1993) but that is something different. The hypothesis that causal understanding is based on experiences of acting on objects provides a simple explanation for this erroneous belief in the monodirectionality of causation.
- Change in the object acted on at contact. When we act on an object, change in that object (e.g., initiation of motion) is detected through the haptic system and is contemporaneous with contacting and acting on the object. This is not to say that change cannot continue to happen after contact, just that contact marks the time at which it starts. Change can also be detected visually, but perceptually interpreting change as an effect of the action is more problematic through the visual modality than through the haptic modality because the haptic modality registers contact forces and the visual modality does not (White, 2012a, b).
- Property transmission. Particularly in cases where the outcome is kinematic, the motion properties of the actor tend to be transmitted to the object. Thus, the faster the hand is moving toward the object when contact occurs, the faster the object moves on contact. The direction of motion of the extremity also tends to be transmitted to the object. This is not a correct Newtonian view, at least for objects that are in motion prior to contact (diSessa, 1982), but the correct Newtonian view is generally applied to classical collisions that occupy little time and involve objects that do not noticeably deform. The actor can use manipulations of extremities to impose a direction of movement on an object through a temporally extended contact. It is possible that this is the source of the erroneous tendency reported by diSessa (1982) to judge that a moving object will go in precisely the direction of the force exerted on it.
- Brief duration of interaction. The simplest and earliest actions on objects are of brief duration, consisting of kicking, pushing, and brief grasping and manipulating motions (Bushnell & Boudreau, 1993; Greco, Rovee-Collier, Hayne, Griesler, & Earley, 1986; Hernandez-Reif et al., 2001; Jouen & Molina, 2005; Molina & Jouen, 2004; Rovee-Collier, 1991; Sann & Streri, 2008). It could be argued that a single action on an object is always of brief duration, and that longer acts of manipulation are understood as multiple actions. When playing tennis, for example, the racquet is held in the hand and manipulated for a considerable period of time, but this may be understood as a series of different but connected actions, each with its own efferent and forward models. Executing a forehand drive, for example, is an action that comes to an end at the end of the follow-through, and then a separate action commences. In any case, the move to more complex and connected series of actions is gradual during development, and early actions are brief and simple.
- Occurrence of a force impression, as described earlier.
- Occurrence of a causal impression, also as described earlier. The force impression and the causal impression are not identical. The evidence for this comes mainly from studies of visual stimuli involving collisions between moving and stationary objects (White, 2009a, 2012a, b). The difference can be approximately categorized as the difference between bringing about some kind of outcome in or for an object and exerting some amount of force on an object.
- Amount of perceived force exerted by the cause. Some actions are experienced as involving greater exertion of force than others (White, 2009a, 2012a, b).
- Outcome magnitude information. Some information about the magnitude of the outcome for the object is provided by haptic information: We can detect haptically the amount of acceleration or deformation produced in an object while in contact with it, because we experience ourselves producing that outcome. More information about outcome magnitude may be obtained through the visual modality after contact has ceased. In that case, there is more of a problem with knowing whether the behavior of the object is an effect of the action or not. Other causes may act on the object, such as gravity or contact with other objects, so judging what is and is not an effect of the action may not be straightforward, especially early in life. Outcome magnitude information could also be conveyed through other sensory modalities, such as audition.
- Cross-modal correspondences. The full perceptual representation of an action on an object encompasses information yielded by the haptic and visual modalities, usually the auditory modality as well, and in principle any sensory modality (e.g., temperature sensors). Input from different sensory modalities is temporally co-ordinated, so the representation of the action incorporates information about temporal correspondences between information from different sensory modalities. For example, the haptic sensation of contact with an object is usually accompanied by visual information about contact (not invariably, because the object may sometimes be visually occluded by the extremity used to contact it). Cross-modal temporal correspondences may therefore be used as clues to causal identification (Körding et al., 2007; Parise, Spence, & Ernst, 2012).
- Exclusivity. When I carry out an action on an object, I experience the production of the outcome through haptic feedback (or through the absence of a discrepancy with the information in the forward model). This carries the important implication that I know through experience that nothing else produced the outcome. The experience of producing the outcome automatically eliminates the possibility of alternative causes. This makes actions on objects a powerful tool for causal learning. It may also be responsible for the faith causal learners have in the value of interventions for causal learning (Perales & Catena, 2006).
The main problem with exclusivity is that it is confined to immediate outcomes of actions on objects, meaning those that can be ascertained haptically (with or without additional feedback in other sensory modalities). Knowledge of any other consequences of action is always uncertain. This will be discussed in the section following the report of the questionnaire study.
As an initial test of the hypothesis, I carried out a questionnaire study in which participants read brief verbal descriptions of interactions and judged whether they were causal relations. The descriptions were chosen such that the events or states of affairs described in them could be visually perceived (with or without perceptual impressions of force or causality). That rules out such things as viruses causing disease and fertilizers making plants bloom. More will be said about those later. Even so, the use of verbal descriptors means that this is more a study of beliefs about causal relations than of perceptions. However, it is far from clear that participants could articulate the criteria by which they make their judgments. All they have to be able to do is to recognize causal relations when verbally described to them, and this could be accomplished by means of implicit knowledge that is not itself accessible to verbal report. That is an open question. Of course, a simple questionnaire study cannot provide proof of the hypothesis: The aim is just to make a case that the hypothesis is worthy of further investigation.
The participants were 40 first-year undergraduate students of psychology with English as their first language. They received course credit for their participation. None had been taught any psychology of relevance to this topic. One participant did not finish the questionnaire and was excluded prior to analysis, so n = 39.
4.1.2. Stimulus materials
The materials comprised a questionnaire with an initial instruction sheet followed by 40 descriptors for judgment. The instructions read as follows: “This questionnaire is concerned with people's beliefs about what sorts of events are causal relations and what sorts of events are not causal relations. You will see a series of short sentences describing simple events, and under each sentence are two questions for you to answer. The first question just asks whether the event is a causal relation or not, and you just answer yes or no to that. If you answer yes to that question, the second question asks you to identify the cause and the effect, briefly and in your own words. If you answer no to the first question, you do not have to answer the second. At the end, please check to make sure you have answered question 1 for all the events and question 2 for all events where you answered yes to question 1.”
The 40 descriptors are now listed. Terms in parentheses after each descriptor are identifying labels used for convenience in the text and Table 1.
Table 1. Features possessed by each descriptor and numbers of participants endorsing each one as a causal relation
|Stone hits vase||2||3||4||5||6||7||8||9||10||9||38|
|Car hits lamppost||2||3||4||5|| ||7||8||9||10||8||37|
|Plate onto plate||2||3||4||5|| ||7||8||9||10||8||36|
|Two moving balls||2||3|| ||5|| ||7||8||9||10||7||36|
|Bucket of water||2||3|| ||5||6||7||8||9||10||8||34|
|Two cars collide||2||3|| ||5|| ||7||8||9||10||7||30|
|Headlights||2||3|| ||5||6||7||8||9|| ||7||30|
|Car skids||2|| || ||5||6||7|| ||9|| ||5||30|
|Ball rebounds||2|| || ||5||6||7|| ||9||10||6||27|
|Salt||2|| ||4||5||6||7||8|| || ||6||22|
|Dam||2|| ||4||5||6||7|| || || ||5||21|
|Mirror||2|| || ||5||6||7|| ||9|| ||5||19|
|Cake||2||3||4||5||6||7|| || || ||6||16|
|Elevator||2||3||4||5||6||7||8|| || ||7||15|
|Ball down slope||2|| ||4||5||6||7|| || || ||5||12|
|Bottle smashes||2|| || ||5||6||7|| ||9||10||6||12|
|Weight on spring||2|| ||4||5|| ||7||8|| || ||5||10|
|Dye||2|| ||4||5||6||7||8|| || ||6||10|
|Cross-beam||2|| ||4||5|| || || || || ||3||9|
|Parachute||2|| || ||5|| || || || || ||2||9|
|Satellite||2|| || || ||6|| || || || ||2||8|
|Water soaks||2||3||4||5||6||7||8|| || ||7||8|
|Avalanche||2|| ||4||5||6||7|| || || ||5||8|
|Clouds|| || || || || || || || || ||0||5|
|Wave||2|| || || || || || || || ||1||5|
|Boat sinks||2|| ||4||5||6||7|| || || ||5||4|
|Paper airplane||2|| || ||5|| || || || || ||2||4|
|River||2|| || ||5||6||7|| || || ||4||4|
|Pendulum||2|| ||5|| || ||7|| || || ||3||3|
|Ball passes ball||2||3||4|| || || || || || ||3||2|
|Plate rests||2|| ||4|| || || || || || ||2||1|
- A paper airplane glides through the air. [Paper airplane]
- A plate rests on a table. [Plate rests]
- A bucket of water lands on the roof of a moving car; the roof of the car gets wet. [Bucket of water]
- Two moving cars collide and rebound. [Two cars collide]
- A cake bakes in an oven. [Cake]
- A ball rolls down a slope. [Ball down slope]
- The headlights of an oncoming car dazzle a motorist. [Headlights]
- A boat gradually sinks in water. [Boat sinks]
- Clouds build up. [Clouds]
- A train pulls some carriages along a track. [Train]
- A car skids on some black ice. [Car skids]
- A dam across a river holds back water. [Dam]
- A hypodermic syringe, having fallen off a table, pierces a cushion on the floor. [Syringe]
- A weight oscillates (moves up and down) on the end of a vertical spring suspended from the ceiling. [Weight on spring]
- Some salt dissolves in some water. [Salt]
- Wind blows some leaves on a tree. [Wind]
- A stone hits a vase and the vase shatters. [Stone hits vase]
- A knife cuts some paper. [Knife]
- Some dye diffuses into some water. [Dye]
- A parachute descends to earth. [Parachute]
- A car runs into a lamppost; both car and lamppost are dented. [Car hits lamppost]
- Light reflects off a mirror. [Mirror]
- A satellite orbits the Earth. [Satellite]
- A ball, having been thrown, is traveling through the air. [Thrown ball]
- A pendulum in a grandfather clock moves back and forth. [Pendulum]
- A bottle smashes on a rock. [Bottle smashes]
- Lightning strikes a tree which is set alight. [Lightning]
- A moving billiard ball passes by a stationary ball and does not touch it; the stationary ball does not move. [Ball passes ball]
- A ball rolling along the ground rebounds from a wall. [Ball rebounds]
- A pen makes a mark on a sheet of paper. [Pen]
- An elevator carries people from one floor of a building to another. [Elevator]
- Water soaks into some clothes. [Water soaks]
- One plate is dropped onto another that is sitting on a table; both plates break. [Plate onto plate]
- Water flows in a river. [River]
- A moving billiard ball contacts a stationary billiard ball and sets it in motion. [Launching]
- A snowplough moving at constant speed pushes snow along a road. [Snowplough]
- An avalanche goes down a mountainside. [Avalanche]
- A cross-beam supports the roof of a building. [Cross-beam]
- Two moving billiard balls come into contact and rebound off each other. [Two moving balls]
- A wave travels across the sea. [Wave]
Each descriptor was accompanied by two questions. The first question asked, “Is this a causal relation? Please underline whichever you think is more likely to be the correct answer:” The words “YES” and “NO” then followed. The second question asked, “If you answered “yes” to question 1, please identify the cause and the effect.” This was followed by the words “CAUSE” AND “EFFECT” on separate lines.
The dependent measure of interest was the number of participants who answered “yes” to the causal relation question. This would then be correlated across descriptors with the number of features from the first 10 in the list given above that each descriptor possessed. The list of features possessed by each descriptor is given in Table 1.
Not all of the features were included. The first feature, human action as cause, was not included because there were no human actions in the descriptors. The quantitative criteria, amount of force perceived and outcome magnitude, were not included because the descriptors provided no information about either of these. Cross-modal correspondences and exclusivity were not included because it was judged too difficult to assess these for most of the descriptors. Occurrence of a causal impression was not included because there is a danger of circular reasoning in the use of occurrence of a causal impression to judge whether something is a causal relation. The remainder was included. The criterion of two perceived objects was deemed to be satisfied if two objects were explicit in the descriptor and not if otherwise. All descriptors met this criterion except for “clouds.”
Participants took part in small groups, supervised by an experimenter. The questionnaire for this experiment was included among a set of materials for experiments on unrelated topics. Participants were told that they should ask questions if anything in the instructions was not clear. None had any questions about the materials for this experiment. Participants then proceeded through the tasks at their own pace. At the end of the session participants were given course credit and debriefed about the aims of the research.
Numbers of participants endorsing each descriptor as a causal relation are given in Table 1. As a preliminary note, the second question was included mainly to ensure that participants attended to the descriptors and gave identifications that were not obviously inappropriate. However, one descriptor, “thrown ball,” was endorsed as a causal relation by 23 participants and 22 of those identified the cause as the act of throwing the ball. As that was not the intended interpretation, this descriptor was considered flawed and was excluded from further analysis.
The correlation across descriptors between frequency of endorsement and number of features possessed was +0.76 (p < .001).
The results show a high degree of correlation between the number of features possessed by a descriptor and the likelihood of endorsement as a causal relation. This supports the hypothesis. Some caveats should be noted.
First, the correlation between likelihood of endorsement and number of features possessed is not perfect, and there are some notable anomalies. “Car skids” and “ball rebounds” have five and six of the critical features, respectively, yet were endorsed by far more participants (30 and 27, respectively) than “bottle smashes” (12), which also has six of the features. Of the 12 who endorsed “bottle smashes,” six identified the bottle hitting the rock as the cause, three the force of the bottle, and one each the hardness of the rock, the bottle falling, and the combination of the force of the bottle falling and the hardness of the rock. Possibly, the difficulty of identifying a clear single cause deterred some participants from identifying “bottle smashes” as a causal relation. But, objectively, the same ambiguity holds for the other two descriptors, yet participants were not deterred.
The low rates of endorsement for “knife,” “train,” and “elevator” are also noteworthy, given the number of features they possess. “Train” and “elevator” were chosen as examples of pulling and carrying (entraining), respectively. There is evidence that visual impressions of one object pulling one or more others do occur with suitable stimuli (Michotte, 1963; White, 2012b; White & Milne, 1997), and pulling and carrying are common actions on objects. In typical pulling stimuli used in the visual perception research, motion of the causal object begins before motion of the effect object(s) begins, whereas that would not be the case for trains and elevators. This could indicate that activity of the causal object prior to contact (or prior to onset of interaction) is a relatively important criterion, perhaps because it helps to assign causal roles to the objects. That could even explain the conundrum noted in the previous paragraph: The car (in “car skids”) and the ball (in “ball rebounds”) could both be regarded as active in that they are engaged in horizontal motion which, in the former case at least, is self-propelled motion. But the bottle may not be regarded as active prior to the interaction because it is just falling: There is evidence that, for an object in free fall, motion is not seen as internally caused but just as natural, as if it required no further explanation (Kozhevnikov & Hegarty, 2001; Ogborn, 1985; Pittenger, 1990). Other descriptors involving objects falling or sinking under no force other than gravity (“avalanche,” “boat sinks,” “pendulum,” and possibly “water soaks”) received low rates of endorsement, which is consistent with the possibility that an interaction tends not to be interpreted as a causal relation if it involves an object just falling.
Readers might be surprised to see that two participants identified “ball passes ball” as a causal relation. Both of them gave the cause as no contact and the effect as no movement. If this sounds odd, it is possible that they imagined that the moving ball had been set in motion by a human agent attempting to hit the stationary ball, so that the miss was an unintended outcome. Then, the stationary ball failed to move because the actor failed to hit it, a violation of an expectation (Weiner, 1985).
The assignment of features to descriptors as shown in Table 1 is problematic in some cases because it is not known what sort of understanding people have of the mechanisms involved. This is particularly acute in the case of the “clouds” descriptor. This was deemed to have none of the critical features on the grounds that only one object was mentioned in the descriptor. The five who endorsed “clouds” as a causal relation tended to identify a single specific antecedent such as rising air. This suggests that they had an explicit model of the system as involving a single causal factor and an effect. In that case, the clouds descriptor as they understand it could possess several of the key features, including two perceived entities (air rising and cloud), entity A (air) being active prior to the interaction, entity B (clouds) being inactive, monodirectionality, interaction at contact, and change in entity B at contact. Those who did not endorse this descriptor as a causal relation could have had a different understanding of it, but there is no evidence either way. Future research could usefully assess participants’ beliefs about the mechanisms involved in various simple causal systems, and it would then be possible to ascertain whether the likelihood of endorsement as a causal relation is related to the number of critical features that individuals’ subjective models of the system possess. For present purposes, it must be noted that there is a degree of uncertainty over the extent to which the list of features possessed by each descriptor in Table 1 matches the implicit or explicit beliefs of participants.
There are several methodological reasons why the correlation could have been less than perfect that do not count against the hypothesis. One, as just discussed, is that the judgments about features possessed by each descriptor in Table 1 may not in all cases match the features of the implicit understanding of these interactions in the heads of some or all participants, and a more thorough investigation would be needed to ascertain those features empirically. Another is that amount of force was not included in the specification of features in Table 1, and it is likely that some judgments (such as the high endorsement of “lightning” as a causal relation) were influenced by that. A third is that, as discussed earlier, the study takes no account of the possibility that features may not be equally weighted. In view of these possibilities, the high correlation obtained is sufficient to indicate that the hypothesis of singular clues is worthy of further investigation.
Most participants made appropriate identifications of cause and effect for most descriptors. Of the few identifications not intended by the author, several invented a specific causal antecedent to the effect in question. These included the act of pouring salt into water as a cause of the salt dissolving (four participants), the act of putting dye into water as a cause of the dye diffusing (one), setting a ball in motion as a cause of it rolling down the slope (one), being put into orbit as a cause of the satellite orbiting (one), and people pushing a button as a cause of the elevator moving (one). It is noteworthy that all of these instances involved human actions, or events brought about by humans, as causes. Therefore, even though humans were not mentioned in any descriptor, there is still some evidence for the hypothesized tendency to identify humans as causes.
5. Clues that emerge from multiple experiences of actions on objects
Individual actions on objects provide experience of causal relations. Further clues to causality can be derived by extending out from this starting point in various directions. These are more complex clues, derivative from the basic set, because they involve putting together more than one event. These include the direction of consequences (or perceived consequences) of individual actions, with tool use as a special case of this; generalization from specific (repeated) experiences to attribution of enduring causally relevant properties of things; connection of individual causal relations, understood in terms of actions on objects, into generative transmission events and mechanisms; and replication of actions as a means of elucidating causal structures.
5.1. Actions and secondary and tertiary consequences
When acting on an object, knowledge of causality comes from the combination of knowledge of motor output and haptic feedback, as described earlier. Usually, the outcome for the object is perceived through other modalities as well, mainly vision but in principle through any sensory modality. These other sensory modalities do not respond to information about forces in interactions between objects, however, so they do not directly give rise to causal knowledge. While the hand is in contact with the object, there are spatio-temporal correspondences between haptic and visual information about the object. (This applies to other modalities as well but for convenience I shall discuss only vision here.) For example, if the hand makes the object move, then there is haptic information about the object's motion and also visual information about the object's motion, and information about the correspondence between these two kinds of information as the action unfolds. The stored representation of the action on the object includes information about these temporal correspondences. Thus, particularly if the action is repeated, the visually perceived outcome for the object can be interpreted as an effect of the action because of the replicable temporal correspondence with the haptic feedback. Information in other sensory modalities obtained while physical contact with the object is maintained, and interpreted as indicating effects of the action, can be called secondary consequences of action.
Once contact with the object has ended, haptic information about the outcome for the object is no longer available. Information in other sensory modalities may continue to be available. One may see the object continue to move, for example, after the hand has finished acting on it. At this point, things become problematic. There is cross-modal correspondence between haptic and visual information while contact is maintained, and there is continuity in the visual information before and after the loss of physical contact. Continuity of visually perceived motion is a cue to the persistence and identity of the object (Spelke, 1994). So it is perhaps not difficult to learn that the continued motion of the object was caused by the efficacious action on it: The action initiated the motion, so the action was the cause of the visually perceived continuation of the motion. Information in other sensory modalities, obtained after physical contact with the object is lost, and interpreted as indicating effects of the action, can be called tertiary consequences of action. This kind of causal knowledge is far from guaranteed to be accurate, however. Once the body has lost contact with the object, it is difficult to distinguish effects of the action from effects of other causes, such as causes internal to the objects, gusts of wind, and so on. It is only haptic feedback that underwrites the distinction between effects of the action and effects of other things. Causal knowledge may be extended by interpreting further sensory input about the object acted on as indicating an effect of the action, but such inferences, whether made in perceptual processing or in cognitive processing, are subject to error because of the lack of haptic information.
Studies by Wegner and Wheatley (1999) and Aarts, Custers, and Wegner (2005) have shown that people can judge that they themselves brought about, or contributed to the bringing about of, an event closely following an action of theirs when in fact the event was determined by a confederate or a computer. To illustrate, in the study by Aarts et al. (2005) the primary consequence of the participant's action was that a key was depressed. This also was transmitted through the inner workings of the computer and was converted, or appeared to be converted, into the position of a black square on the computer screen. Participants overestimated the extent to which their action determined the position of the black square. That was an error in relation to a supposed tertiary consequence of action, because the visual information about the position of the black square was spatially and temporally separate from the haptic information, which is just information about the pressing of the key. It could be argued that replicability would reduce the error in causal inference in such cases. Under everyday circumstances that might be the case, but the possibility of error can never be eliminated. In the study by Aarts et al. (2005), the error cannot be eliminated because the computer has, in effect, the role of a malicious demon, always interfering between the key press and the location of the black square. There is no way to be sure that acts of intervention yield valid causal inferences beyond the level of haptic involvement.
That is not meant to imply that haptic feedback ensures the correctness of impressions of causing in actions on objects. There are haptic perceptual illusions, such as the haptic size-weight illusion (Amazeen & Turvey, 1996; Charpentier, 1891), so the haptic system is not completely trustworthy, and there is always the theoretical possibility of error in a system that involves communication of peripheral stimulation to central processing (Dijkerman & de Haan, 2007). There is, however, no room for a malicious computer of the sort used by Aarts et al. (2005) to interfere between the initiation of an action and haptic feedback about the action's direct effect on an object through physical contact, so that whole category of error is ruled out.
5.2. Generalization from specific experiences to enduring causally relevant properties
Different kinds of actions on different kinds of objects have different kinds of outcomes. These differences are not random. To the extent that there are consistencies in relations between actions and outcomes, conditional perhaps on the kind of object acted on, Hintzman's (1986) model predicts the emergence of abstracted schemas of different kinds of interactions with objects. A simple example concerns the distinction between rigid and compressible objects. It has been shown that infants at a few months of age can quickly learn that some objects are rigid and others are compressible. They then respond with surprise if a supposedly rigid object is fitted into a container smaller than the object, but they do not respond with surprise if the same is done with a compressible object (Aguiar & Baillargeon, 1998; Schweinle & Wilcox, 2004). This indicates some level of understanding of compression as a kind of causal relation: Not only do the infants experience compression in their own actions on the objects, they must understand in some way that the object can be fitted into the container by compressing it. At the same time, they understand that not all objects are compressible. Having established by acting on it that an object is not compressible, they do not expect that it can be fitted into the container.
Evidently, the infants must be attributing stable, enduring causally relevant properties to the objects; otherwise they would not have any expectation for what would happen when the rigid object was fitted into the container. This is the first point to be made in this section: Young infants appear to be adept at filling the gaps in experience by inferring enduring properties of things. If an object is rigid at time 1, and then following a short interval it is also rigid at time 2, infants appear to infer that it continues to be rigid between times 1 and 2, and then beyond time 2 as well. A substantial body of research has shown that infants operate with a general understanding of persistence in concrete material objects (Baillargeon, 2008; Baillargeon, Li, Luo, & Wang, 2006; Spelke, 1994; Spelke, Breinlinger, Macomber, & Jacobson, 1992; Spelke, Katz, Purcell, Ehrlich, & Breinlinger, 1994; Spelke, Phillips, & Woodward, 1995). Thus, at a basic level, an object and its properties are assumed to continue to exist between gaps in perception of them (unless made to change by some action); objects are solid, in the sense that any two objects cannot interpenetrate or occupy the same space; objects follow connected spacetime paths.
Specifically for the concerns of this article, this generalization tendency applies to causally relevant properties of things. Infants learn not merely that objects and their properties tend to persist over time, they learn that causally relevant properties of things persist as well. This is what the studies by Aguiar and Baillargeon (1998) and Schweinle and Wilcox (2004) show. From an experience of acting on an object, with haptic experience of the production of an outcome for that object, it is apparently only a short step to an inference that the object possesses the capacity to have that outcome produced in it, not just at the time of the action but at all times. It is likely that repeating the action on the object supports this kind of inference. The infants in the studies by Aguiar and Baillargeon (1998) and Schweinle and Wilcox (2004) had the opportunity for multiple actions on the objects before the container-fitting was presented to them. So, from early times, we can use actions on objects to acquire beliefs about enduring causal powers and liabilities of things.
Infants also learn from actions on objects, not just that causally relevant properties are general across the span of existence of an object, but also that they are specific to certain objects and kinds of objects. One object is compressible; another is not. Not far beyond that, some kinds of objects are compressible and others are not. Of course, there is much to learn about the generalization of causally relevant properties of things, and development of knowledge about generalization continues in an uneasy course throughout childhood (Karmiloff-Smith, 1992). But young infants are already quite sophisticated in this respect, as a study by Bourgeois, Khawar, Neal, and Lockman (2005) shows. Infants aged 6 months were given the opportunity to relate a manipulable object—a hard or soft cube—to various kinds of surface. The kind of operation the infants performed varied with the surface. For example, they pressed the cubes into a yielding surface more than into a liquid or a rigid surface. They also tended to use different cubes for different operations. They banged the hard cube more often on the rigid surface than on the yielding or liquid surfaces, but not the soft cube.
This shows an already sophisticated understanding of the relations between the causal powers of one thing and the liabilities of another: The outcome depends on both the object acted on and the object acted with. The infants are learning that a hard cube can elicit a different noise from a hard surface than a soft cube can; and they learn that the noise elicited depends on what kind of surface is being acted on. Presumably, this involves learning of co-ordinated properties of hardness (assessed mainly haptically) and sound. These are simple causal structures: Bang a hard surface with a hard cube to get a clonking noise; press a yielding surface with a cube to deform it. And they are tied to actions on objects. The manipulating of the cube involves the experience of forces through the haptic system and the forward model. Perceptually, the execution of these actions is spatio-temporally co-ordinated with outcome information that may be perceived through the haptic, visual, or auditory modalities. The differentiation of the cubes is particularly revealing in this regard: One cannot have a preference for an object that produces one kind of outcome over an object that produces another kind of outcome without understanding that the outcomes are produced by the action of the object on the surface. This supports learning of stable, enduring causally relevant properties of things.
5.3. Tool use in causal discovery
The role of tool use in the acquisition of causal knowledge has been underestimated. From the age of 4 months, infants grasp objects and manipulate them with repetitive finger and hand movements (Bushnell & Boudreau, 1993; Thelen, 1995). An example would be banging an object on another object (Bourgeois et al., 2005). Actions of that sort have a key feature for causal learning: The contingency between the banging of the object and the noise elicited (or whatever sensory consequence is salient for the infant) is specifically a contingency between an action and an outcome directly and immediately generated by the action. Thus, the infant has an understanding of causation in mechanoreceptor-mediated experiences of acting on objects, in this case wielding the grasped object, and the outcome, the noise, is grafted onto that understanding. The infant learns that the noise is produced by the banging action, because the noise is both contingent upon and temporally associated with the banging. The temporal co-ordination between mechanoreceptor feedback of contact with the surface or object banged and auditory feedback of the noise is precise. The occurrence of the noise is under the control of the infant.
It could be argued that the occurrence of a noise need not be represented by the infant as an outcome for the object acted on. However, the study by Bourgeois et al. (2005) described above suffices to show genuine causal learning. This is because infants showed that they understood the specificity of the relation between the tool (the cube) and the surface acted on with it. Infants are acquiring knowledge about causally relevant properties of things, and knowledge about the limits on generalizability of that knowledge, from acting on things with tools. Hard tools can elicit outcomes from objects that soft tools or hands alone cannot.
Further development in tool use shows a progressive liberation of causal discovery from the tight strictures of the use of tools in direct actions on objects. By 7 months, infants can pull on a cloth to retrieve an object that is resting on the cloth but out of reach (Kolstad, 1994; Willatts, 1999). Beyond that, infants can learn novel operations with tools by emulation. Emulation differs from imitation in that mastery of the task requires understanding the causal structure of the situation, rather than just being able to copy the behavior of a model (Thompson & Russell, 2004). This is shown by studies in which infants around 18 months old learned novel accomplishments with tools from models that were not human (Danish & Russell, 2007; Huang & Charman, 2005; Thompson & Russell, 2004). Danish and Russell (2007) showed that emulation can help infants learn more indirect links between action and outcome. In the study by Danish and Russell (2007), infants learned to produce an outcome that was both spatially and temporally distal from the action and the object acted on, and human and non-human models were equally effective in this.
The extent to which tool use is responsible for these developments in causal learning is not certain: It could be argued that the development of the capacity to learn about causal discovery through tool use is determined by development in other areas, such as contingency learning. I shall argue against that position later, when I focus on contingency information.
5.4. Connection of actions and consequences into sequences: Generative transmission and mechanisms
Generative transmission was proposed as a primary source of causal knowledge by Shultz (1982; Shultz, Fisher, Pratt, & Rulf, 1986). Shultz (1982) described it as “some sort of transmission between materials or events by virtue of which one acts to change or produce the other” (p. 3). In the experiments Shultz conducted to investigate whether children and adults used generative transmission information for causal inference, he studied systems such as a vibrating tuning fork making a box resonate, a bellows extinguishing a candle flame, and a flashlight making the vanes of a Crookes’ radiometer rotate.
Generative transmission cues can be understood as connected events that are themselves understood in terms of actions on objects. The advance comes in understanding how one action may be connected to a subsequent one. Consider one of the experimental stimuli used in the research by Shultz (1982): A bellows extinguishes a candle flame. The sentence describes this as if it was a single causal relation but in fact it is three. First, an action by the experimenter compresses the bellows. Then, the compression of the bellows produces a gust of wind. Then the gust of wind extinguishes the flame. Each of these is a separate causal relation and is understood as such from an early age. This claim is supported by the research on the development of tool use just cited. I am assuming that children learn at an early age what can be achieved by blowing on things, and they learn that the gust of air produced by blowing is the cause of an outcome for an object with which one is not in contact; there does not appear to have been any research on this. In summary, generative transmission should be seen, not as a fundamental clue to causality, but as a higher order clue representing a developing understanding of connecting actions in series.
The same can be said of mechanism clues. Mechanisms have been defined as “a system of connected parts that operate or interact to make or force an outcome to occur” (Ahn & Kalish, 2000, p. 201) or “a system of parts that change as the result of interactions among them that transmit force, motion, and energy” (Thagard, 2000, p. 262). There is evidence that people tend to seek mechanism clues in preference to empirical association information when making causal attributions (Ahn, Kalish, Medin, & Gelman, 1995), and there is evidence that children understand causal relations involving intermediate mechanisms at least from 3 years of age (Bullock, Gelman, & Baillargeon, 1982; Shultz, Pardo, & Altmann, 1982). Mechanism beliefs may also have a role in the interpretation of temporal information in relation to causality (Buehner & McGregor, 2006).
The idea that a mechanism is a system of connected parts helps to reveal how mechanisms are understood. A mechanism is understood, at least in early years, as a connected network or series of actions on objects. To illustrate, consider the apparatus used by Shultz et al. (1982). The apparatus involved a Y-shaped tube with starting points at the tips of the branches of the Y. Balls that could be rolled down the tube were placed at each starting point. Partway down each branch was an arch. In one branch, a golf ball was placed on the near side of the arch, so that it could not pass through, and in the other branch it was placed on the far side. In the latter case, the ball at the starting point could contact the golf ball through the arch, even though it could not itself pass through the arch. Where the branches converged was another ball. Participants had to choose which ball to roll down the tube from the starting point to make this final ball move. Children aged 3 and 5 years succeeded in this task. In this case, the mechanism can be understood in terms of a series of connected actions. The child acts on the ball to set it in motion. The ball then strikes the golf ball and sets it in motion; this is understood in terms of actions on objects. The golf ball then strikes the final ball and sets it in motion; this too is understood in terms of actions on objects. The finding that children chose the correct ball to push shows that they had an appropriate representation of the mechanism in advance of their choice of action.
It is noteworthy that the developmental studies of mechanism understanding have tended to use simple mechanical devices of the sort used by Shultz et al. (1982). Children have not been required to solve problems involving mechanisms that cannot be readily interpreted in terms of actions on objects. An example of a mechanism given by Ahn and Kalish (2000) involved germs, the person's immune system, and lack of sleep. Of course, young children are not likely to possess the domain-specific knowledge necessary to comprehend mechanisms of that sort. Even so, I would predict that, if domain-specific knowledge can be equated, young children should show a superior understanding of mechanisms the components of which can be understood in terms of actions on objects over mechanisms that cannot be understood in that way.
5.5. Replication and other empirical association cues
An action on an object can be repeated, and this may rapidly lead to a large number of stored representations of similar actions with similar outcomes. Replicability is greatly facilitated by tool use. If we push an object away, the object has to be retrieved in order for the action to be replicated. Once we are able to grasp and manipulate an object, however, a rapid rate of replication can be achieved, as anyone who has watched an infant repeatedly banging a mug on a table will know. In Hintzman's (1986) model, the acquisition of multiple similar episodic traces results in abstraction of the common features of the event as a schema, in effect a causal belief: not just, “I banged the mug and produced this noise,” but “banging the mug produces this noise reliably.” There is much to learn about how general causal beliefs of this sort are acquired and I do not wish to propose that this is the single and basic way of doing so. Nevertheless, replication is a powerful tool of causal learning, and I do want to propose that it is the basis for the use of other empirical patterns as clues to causality.
Any action on an object can be construed as an intervention, so replication could form the basis for competence in using intervention to ascertain causal structures. An origin of this competence in one's own actions on objects would explain why children have difficulty in understanding relations between interventions and spatially distal (tertiary) outcomes at an age when they can understand relations between interventions and spatially proximal outcomes (Bonawitz et al., 2010). The conditional probabilities are the same in both cases, but we understand spatially proximal outcomes earlier because we experience ourselves causing spatially proximal effects through haptic input, but we do not experience ourselves causing spatially distal effects. For that, replication is necessary.
It is plausible to suggest that competence in using empirical information in causal judgment develops gradually. Although young children can use interventions to work out the causal structures of physical systems, they are limited by factors other than conditional probabilities. An example would be spatial separation of cause and effect (Bonawitz et al., 2010); I shall argue later that this is an example of dependence on singular clues to causality. They also appear to be limited to simple systems with few alternative causal explanations. In a study by Gopnik et al. (2004), there were only three relevant entities (an experimenter and two puppets), and only two causal hypotheses, which the children were explicitly informed about. Other studies have used systems scarcely more complex than that (Gopnik et al., 2004; Gopnik & Schulz, 2004; Gopnik, Sobel, Schulz, & Glymour, 2001; Gweon & Schulz, 2012; Kushnir & Gopnik, 2005; Schulz, Bonawitz, & Griffiths, 2007; Schulz, Gopnik, & Glymour, 2007; Sobel & Kirkham, 2006, 2007; Sobel & Munro, 2009). Even older children fail to appreciate the importance of control of variables in causal hypothesis testing (Chen & Klahr, 1999; Kuhn, Schauble, & Garcia-Mila, 1992; Schauble, 1990; Zimmerman, 2007). This indicates that, even though children can devise interventions to test causal hypotheses (as they did in the study by Schauble, 1990, for example), they do not have an understanding of the basic principles of experimental design. That is consistent with an account in which children draw causal inferences without a proper appreciation of methodological principles relevant to causal inference.
6. Use of the singular clues in causal judgment
The hypothesis that causal understanding is ultimately derived from experiences of actions on objects has yielded a large set of possible clues to causal judgment. I have identified 14 singular clues in the form of features of actions on objects, and several ways of extending causal knowledge by building on what experiences of actions on objects supply. This does not imply that explicit causal judgment is rigidly tied to the use of these clues. Acquired knowledge can to some degree liberate us from narrow dependence on clues related to human action. However, I propose that these clues retain their influence as a pervasive set of biases in causal judgment. Especially under conditions of uncertainty, we seek and preferentially identify as causes things that possess the singular clues to causality over things that do not.
As an example, consider the first clue in the list. Clearly, we do not interpret every outcome as an effect of human action. We are capable of inferring non-human and non-animate causes of many events and occurrences in the physical world. Our first preference would be for physical causes that possess as many of the clues to causality as possible, about which I shall say more in a moment. Under conditions of uncertainty, however, we still have a bias toward favoring human actions as causes. Support for this comes from research by McClure, Hilton, and Sutton (2007) and Hilton, McClure, and Sutton (2010). They found that adults judged human actions to be both better and more important causes of outcomes in causal chains than physical events with equivalent power. For example, a person fanning flames was judged to be a better explanation for a forest fire than a wind fanning the flames, even though the wind would arguably be more efficacious because of being stronger and more sustained.
Hart and Honoré (1985) argued that, in legal cases, voluntary human actions tend to be preferred as causal explanations over physical events. A tendency to accept teleological (purposive) explanations of both biological and non-biological events is common among young children and persists in adults (Kelemen, 1999; Kelemen & DiYanni, 2005; Kelemen & Rosset, 2009). The pathetic fallacy is a widespead tendency, not just to attribute human mental states to inanimate nature but also to interpret natural occurrences and phenomena in terms of human actions and feelings.2 As an appropriate example, Michotte (1963) commented that observers of the launching effect “had an amazing tendency, in describing their impressions, to make comparisons with human or animal activity” (p. 280). He reported that one observer of a launching effect stimulus described it as follows: “A ‘gave B a kick in the seat of the pants and sent him flying’” (p. 280). A common, but incorrect, version of Newton's third law states that, for every action, there is an equal and opposite reaction. This implies a kind of understanding in which one object acts on another and the latter then reacts back. There is an enlightening discussion of this in an educational context by diSessa (1993). Among other things, diSessa reported that students will sometimes say, “One can tell the action from the reaction according to who acts first” (p. 155).
It is not easy to set up equal contests between human actions and inanimate causes, which is why the studies by McClure et al. (2007) and Hilton et al. (2010) are so important in this regard. In addition, as McClure et al. (2007) pointed out, causal attributions in real life tend to be made in a context of practical concerns, where there is often greater utility in identifying a human action than a physical event as a cause. There is therefore a need for much more research to investigate the extent to which a heuristic of interpreting events in terms of human causes pervades our understanding of physical events.
A second example is the property transmission heuristic, item no. 7 in the catalog of singular clues. Under conditions of uncertainty, we seek causes that satisfy the property transmission clue, or we infer property transmission when judging effects, and this leads to predictable biases in causal judgment. White (2009b) described several lines of research that support this claim, including a series of studies on magical contagion (Nemeroff, 1995; Nemeroff & Rozin, 1989, 1994; Rozin, Millman, & Nemeroff, 1986; Rozin, Nemeroff, Wane, & Sherrod, 1989). As an example, Nemeroff and Rozin (1989) found that judged personality characteristics of fictitious groups of people tended to resemble distinguishing characteristics of animals they ate, as if the characteristics of the animals were transmitted to the people who ate them. Under conditions of uncertainty, judgments about personality characteristics were guided by the idea of property transmission, in this case from characteristics of animals in the people's diet.
It may be objected that the singular clues are not valid guides to causal identification because they are not exclusively associated with actual causal relations. Taking the contact clue as an example, one thing may come into contact with another but that does not guarantee that the former is the cause of something that happens to the latter. There are two replies to this. One is that the singular clues are used as guides to causal identification because they are empirically associated with something that is a valid guide to causal identification, namely the actual experience of a causal relation in acting on an object. The other is that they are not meant to be used in isolation. Things are more likely to be identified as causes as they possess more of the clues; a thing possessing just one or two clues is not likely to be identified as a cause. In fact, the clues provide a solution to a traditional problem of causal identification, the selection problem (Hesslow, 1988; Wolff & Song, 2003; Wolff et al., 2010). If something happens, there are, a priori, indefinitely many possible causes of it, and the problem for any plausible mechanism of causal induction is that it cannot process information about all those possible causes. Various ways of limiting the field of causal candidates have been proposed (Cheng & Novick, 1992; Griffiths & Tenenbaum, 2009; Hart & Honoré, 1985; Hume, 1978; Wolff et al., 2010), but the singular clues solve the problem automatically without the need to define limitations to the field. Any causal candidate can be rapidly assessed for the number of singular clues it possesses, and in fact the clues guide the search for causal candidates by focusing attention on certain areas at the expense of others. The contact and temporal immediacy clues, for example, serve to eliminate possible causes that are distant in space or time from the outcome of interest. Thus, the solution to the causal selection problem is that only candidates with the greatest number of singular clues are considered as possible causes. These can, of course, be subject to further testing, and other candidates may be sought if none proves satisfactory.
The ontogenesis of many things that have been shown in previous research to function as cues to causality can be seen in the features of actions on objects. Temporal contiguity tends to function as a clue to causal judgment (Shultz & Kestenbaum, 1985; White, 1988), and it originates in clue no. 6, change in the object acted on being contemporaneous with the action on it. Spatial contiguity also tends to function as a clue to causal judgment (Bullock et al., 1982; White, 1988), and it originates in clue no. 4, contact between actor and object. Temporal priority tends to function as a clue to causality (Shultz & Kestenbaum, 1985; White, 1988), and it originates in clue no. 2, prior activity of the actor. All three have traditionally been regarded as Humean cues to causation because they were explicitly specified by Hume (1978) as giving rise to the illusion of causality, along with constant conjunction. This is a historical artifact. They are in fact clues derived from actions on objects, along with several others in the list above that were not mentioned by Hume, and it is time that they were divorced from their purported philosophical ancestry.
6.1. Implications for the use of contingency and conditional probability information in causal judgment
As was mentioned early in the article, there have been many proposals in psychology about how causal knowledge, valid or not, may be acquired from observation of regularities in events. They are rooted in a philosophical position that causality cannot be directly observed, and that multiple instances are required for any kind of belief about causality to emerge (Hume, 1978; Sosa & Tooley, 1993). This approach has dominated the study of causal judgment, to the extent that some prominent reviews of the causal judgment literature do not even mention other possibilities (Allan, 1993; De Houwer & Beckers, 2002; Holyoak & Cheng, 2011; Perales & Shanks, 2007).
In this section, I argue that singular clues cast this approach in a different light, one in which contingency is a derived clue to causality, not the source of causal knowledge. Specifically, I argue that the use of contingency and other empirical cues depends on the presence of singular clues to causality; it is redundant where there are adequate singular clues for causality to be identified in single instances; there is evidence that the use of contingency information in causal judgment is an acquired skill and not innate; there is evidence that contingency information is actually used for hypothesis testing, in other words in a context of acquired beliefs and clues to causality; and there is evidence that identical contingencies are treated differently in terms of causal inference. I shall discuss each of these in turn.
The most serious problem for testing any model or hypothesis in which contingency information (or any form of empirical regularity) is fundamental to human causal judgment is that of separating contingency from singular clues to causality. All situations in which the use of empirical information for causal judgment has been studied are confounded by the presence of numerous singular clues. I shall first illustrate this with an analysis of the puppet study by Gopnik et al. (2004).
In the puppet study, children aged 4 years were shown a stage with two puppets on it, and they saw the puppets move together and stop moving at the same time. The experimenter told them that one of the puppets was special and always made the other one move. The children had to guess which one was the special puppet. After a short training session, they were exposed to a series of trials in which the puppets both moved together and stopped together. Then, the experimenter intervened and moved one of the puppets (Y) and the other puppet (X) did not move. Then both puppets moved together again. The children tended to say that X was the special puppet. This shows an inference being made from the experimenter's intervention. On trials where the experimenter appears not to intervene, both puppets move. This is, of course, ambiguous because either one could be making the other one move. If Y was the special puppet then, when the experimenter intervened on Y, X should have moved as well. X did not move. The experimenter's intervention, therefore, ruled out the hypothesis that Y was the special puppet, leaving only the hypothesis that X was special. This was a key experiment in an argument made by Gopnik et al. (2004) that children detected causal structure in a manner conforming to normative causal Bayes nets analyses.
The puppet study does not in fact support that claim, because the scenario in the puppet study provides several singular clues that could have guided children's judgments: no. 1, human action (not only intervention by the experimenter but also the puppets were introduced as actors); no. 3, no prior activity of entity in which the outcome occurred; no. 4, contact between cause and outcome object, though in this case contact was mediated by the physical connection between the puppets; no. 5, monodirectionality, which is explicit in the instructions; no. 6, change in the object acted on at contact, because there is no temporal interval between the motions of the two puppets; no. 7, property transmission, in the form of resemblance between the motions of the two puppets; no. 8, brief duration of interaction. The intervention of the experimenter on one of the puppets possesses all those clues and more: no. 2, prior activity of causal object, because the experimenter moves to the puppet in acting on it, and no. 4, contact between experimenter and puppet is direct. Note that the children would not be in any doubt that the experimenter moved the puppet and the puppet did not move the experimenter. This is so obvious that it has never been pointed out, but it is causal knowledge that is a prerequisite for ascertaining the causal structure of the puppet situation. Clearly, a great many singular clues to causality are present in that experimental situation. Similar analyses can be provided of any study of causal structure elucidation. The claim that participants are identifying causal structures just from conditional probabilities is untenable. A proper test of that claim would require a situation in which there were no singular clues.
Studies of causal judgment from contingency information also typically provide several of the singular clues. Usually, a scenario is set up with specific content, and judgments are made about contingency information within the context of the scenario. For example, in Lober and Shanks (2000) and Collins and Shanks (2002), a research scenario was used in which the possible cause was irradiation and the outcome was mutation in the DNA of the irradiated organism. This scenario satisfies several of the clues: no. 1, human action (the radiation was activated by a researcher), no. 2, prior activity of possible cause (irradiation); no. 3, the organism in which the outcome occurred was not active prior to the application of the radiation; no. 4, contact between radiation and the organism; no. 5, monodirectionality, because it is unlikely that participants would have believed that DNA can affect radiation; no. 6, change in the organism at contact, because no delay between irradiation and occurrence of mutation was mentioned; and no. 8, brief duration of interaction. In addition, the higher order clue of mechanism was satisfied, because the scenario was designed to evoke pre-existing beliefs about causal mechanisms connecting radiation and mutation. Scrutiny of any scenario used in causal judgment research will reveal a similar list of singular clues to causality. Use of causal language in instructions and materials is also routine.
If contingency is the fundamental means by which causal learning occurs, then it should be possible to demonstrate that it occurs under conditions where the singular clues are absent. Temporal contiguity can be excepted from this, because it is arguable that a fairly brief window of time is required for the detection of contingency (Shanks & Dickinson, 1987). No published experiment has met those conditions. It can therefore be argued that the use of contingency information in causal judgment is dependent on the presence of at least some of the singular clues, not to mention the causal language used by the experimenter.
It could be argued that singular clues and contingency information are just used for different purposes: Singular clues are used for problems of causal attribution, where the issue is to identify the cause of one particular outcome, and contingency information is used to establish general causal beliefs. However, as I showed earlier, infants and young children are inclined to generalize causally relevant properties revealed by brief interactions (e.g., Aguiar & Baillargeon, 1998), and older children tend to induce novel causal hypotheses from properties of single instances (e.g., Schauble, 1990). Singular clues derived from experience of single instances therefore form a source of hypotheses about causal generalities that may or may not be subject to further testing. In that way, they support inference of general causal beliefs as well as addressing causal attribution problems.
In the present account, causal learning begins with actions on objects and gradually progresses to a point where reasonably sophisticated use can be made of contingency information (I say “reasonably” because of the requirement for other singular clues to causality to be satisfied.). If that is the case, then one would expect both developmental and individual differences in the use of contingency information in causal judgment. There is evidence for both.
Shaklee and Elek (1988) compared American junior high school students and college students (no age data were published) and found superior performance by normative standards in the latter, in terms of proportions of participants who identified covariates as causes. Shaklee and Goldston (1989) sampled participants aged 8, 12, and 21 years, and found improvement with age in identifying covariates as causes. Individual differences have been found among adults (Anderson & Sheu, 1995; White, 2000). Anderson and Sheu (1995) and White (2000) both found that some participants used only cause-present information for causal judgment and ignored cause-absent information in the stimulus materials. That finding, which disconfirms the predictions of any model that postulates normative competence at causal judgment, has never been satisfactorily explained by advocates of contingency-based models (Cheng, 1997; Cheng & Novick, 2005; White, 2005). In addition, White (2000) found self-reported individual differences in the use of cause-absent information. Tendencies in causal judgments matched the reports made by the participants, indicating a degree of insight into individual use of information. Some participants reported what White (2000) called the “closer to zero rule”: If the previous judgment had been below zero (on a scale from −100, preventing, to +100, causing), then an instance of cause-absent information led them to raise their judgment, and if the previous judgment had been above zero, then the same cause-absent information led them to lower their judgment. A minority of participants reported using that rule and their judgments were consistent with the use of it. These tendencies can be interpreted as representing attempts by participants to deal with information that is not in a form that they find natural for causal judgment. It is clearly not the case that everyone makes judgments in the same way, and neglect and idiosyncratic use of cause-absent information are common.
Most models of causal judgment from contingency information have treated causal learning as inductive. That is, given a few starting assumptions, causal judgments and beliefs emerge from the accumulation of empirical information without the prior formulation of hypotheses to direct the search for information. An example of this approach is the power PC theory (Cheng, 1997). In that theory, it is assumed that humans are born with the knowledge that there is such a thing as causality in the world, and the inferential mechanism modeled in the theory is also present as a form of innate competence. Causal beliefs are acquired simply from the mechanism operating on input information about contingencies without the need for directed searches for particular kinds of information, or any kind of formulation and testing of causal hypotheses. Contrary to that approach, there is considerable evidence that contingency information is used, not inductively, but for hypothesis testing (Ahn et al., 1995; Luhmann & Ahn, 2011; Schauble, 1990; Zimmerman, 2007). That is, causal judgment begins with the formulation of one or more hypotheses, and information is then sought that might have confirmatory or disconfimatory value for the hypothesis under test. This is a sophisticated use because the formulation of a hypothesis requires some domain-specific knowledge. It could be argued that hypotheses are generated from accumulated contingency information, but research findings do not strongly favor that possibility. At the very least, it is just one among many ways in which hypotheses could be formulated. Schauble (1990) found that even when children were collecting data that would enable contingencies to be assessed and scrutinized for possible hypotheses, they tended to generate hypotheses on the basis of individual instances, not on patterns detected across a sample of instances.
Moreover, if humans have a hypothesis-testing orientation, they are not restricted to contingency information in testing their hypotheses. Any of the singular clues could be used, as could any of the kinds of information derived from the singular clues. Ahn et al. (1995) argued that mechanism clues are favored. Participants in their experiments sought clues to causal mechanisms and had little interest in contingency information. The materials included event descriptions that were nonsense sentences, and things about which people could not have had preconceived beliefs. In an experiment where people had enough information to conduct a covariation analysis to identify some factor in the stimulus information as the cause, they did not do so; instead, they identified a mechanism not explicit in the description of the event. There is a good reason for this: “the focus of the mechanism analysis would be on discovering the process underlying the relationship between the cause and the effect” (Ahn et al., 1995, p. 304). Their point is that contingency analysis can only identify a cause, or give justification for changing one's estimate of the probability of a given cause, whereas mechanism analysis can answer the “how?” question, explaining how a given effect is generated. That is more informative and more useful. The force of the argument is illustrated with a real-world example in the supplementary materials.
Luhmann and Ahn (2011) have taken this argument further, showing that even the confirmatory and disconfirmatory status of individual instances of contingency information is not fixed. White (2000) had already shown that the status of cause-absent information could vary depending not only on the individual participant but also on the value of the previous causal judgment. Luhmann and Ahn (2011) showed that the status of cause-present information is open to interpretation as well. Objectively, occurrences of an outcome in the presence of a candidate cause are confirmatory for the cause because they tend to increase the contingency, and non-occurrences of the outcome in the presence of the cause are disconfirmatory because they tend to decrease the contingency. Luhmann and Ahn showed, however, that the interpretation of such instances varied depending on the kinds of instances that had been encountered earlier. If previous instances were predominantly confirmatory, for example, then non-occurrences of the outcome in the presence of the cause were most commonly interpreted either as coincidence or as indicating that, for some reason, the cause failed to produce the outcome on that occasion. This is reminiscent of the reasoning about objectively disconfirmatory instances in the study by Schauble (1990): Instead of being treated as disconfirmatory for the hypothesis, they tended to be explained away as exceptions or reinterpreted as not disconfirmatory. Equivalent tendencies were found in the interpretation of objectively confirmatory instances if the previous instances had been predominantly disconfirmatory. Luhmann and Ahn also showed that adding confirmatory instances after a history of largely disconfirmatory instances could lead to lower, not higher, causal judgments if a second candidate cause was present. The presence of the second cause provided a means of interpreting the apparently confirmatory instances in a way that maintained the hypothesis about the first cause.
The final problem to be discussed here concerns the findings of a study by Muentener and Carey (2010). Several studies with infants at a few months of age have presented stimuli in which a moving object contacts a stationary one and the stationary one then moves. The consensus of the findings is that infants have a causal impression, an impression that the moving object makes the stationary object move, that appears to resemble the launching effect that occurs with adult observers (Cohen, Amsel, Redford, & Casasola, 1998; Cohen & Oakes, 1993; Leslie, 1982, 1984; Leslie & Keeble, 1987; Newman, Choi, Wynn, & Scholl, 2008; Oakes, 1994; Oakes & Cohen, 1990; Saxe & Carey, 2006). Muentener and Carey used a variation on this stimulus in which a screen concealed the space in which contact would occur. The target object was partly occluded by the screen; the causal object (a toy train) moved toward the target and disappeared behind the screen, whereupon the target moved off. Infants aged 8 months perceived this as causal: They were surprised if, when the screen was removed, the causal object did not contact the target object, but not surprised if it did. In a variation on this, Muentener and Carey presented changes of state in the target object: The target object either emitted musical sounds or broke into pieces. Infants did not perceive those stimuli as causal; whether the causal object contacted the target or not when the screen was removed made no significant differences to their looking times. In the critical experiment, the toy train was replaced with a human hand. This time, infants apparently perceived both motion and change of state in the target as caused by the hand. Thus, they perceive contact from a human hand as causing a change of state in another object at an age when they do not perceive contact from an inanimate object as causing the same change of state.
Muentener and Carey pointed out that the conditional probabilities in those two events are exactly the same: The outcome always occurs when the moving object contacts the stationary one. If infants were inferring causality from detected conditional probabilities, therefore, they should represent both events as causal relations. The finding that they represent only one of the events as a causal relation therefore counts against the hypothesis that empirical cues are fundamental to causal learning. Instead, the findings support the hypothesis advocated here that human action is fundamental to causal learning.
In summary, contingency information is one among a large number of clues to causality. It cannot be used alone and must be supplemented by singular clues such as temporal contiguity, which itself has to be meaningfully defined. It cannot be the origin of causal understanding partly because its use depends on the presence of singular clues to causality and partly because it cannot be used purely inductively. Instead, it is used for the testing of hypotheses, and its use is significantly biased by the hypothesis under test. The use of contingency information in causal judgment is a sophisticated accomplishment confined to those who have already acquired knowledge about causality from other sources.
Clues to causality are ultimately derived from the kinds of experiences that are formative for our understanding of causality, namely actions on objects. During development more sophisticated clues, including contingency information and mechanism information, can be built on those foundations. Despite an increase in sophistication, partly due to education, we retain a pervasive bias to prefer candidate causes that possess the singular clues to those that do not. Our understanding of causality continues to be, at bottom, an action-like understanding, even when education in Newtonian physics tells us otherwise (Muentener & Lakusta, 2011; diSessa, 1993). Advocates of varieties of empirical association as the source of causal learning have assumed that causal relations cannot be perceived and must be inferred (Cheng, 1997; Perales & Catena, 2006). Instead, it can be argued that actions on objects provide a sufficiently close approximation to direct experience of causality that they can function as a source of the idea that there is such a thing as causality in the world and, more specifically, they function as a source of singular clues through the use of which causal knowledge can be extended far beyond its locus of origin.
Evidence in support of some of the clues has been discussed. Not just studies showing use of the postulated cues, but more important, studies that pose what are in effect contests between different kinds of cues, are most revealing. In this respect, the most important studies are given in the following: (a) those by McClure et al. (2007) and Hilton et al. (2010), showing preference for human action over inanimate physical causes; (b) the study by Bonawitz et al. (2010) showing that use of contingency information for causal inference is profoundly affected by the presence of absence of singular clues to causality; and (c) the study by Muentener and Carey (2010) showing that young infants perceive a causal relation from a given pattern of contingency when the cause is a human action but not when the cause is contact from an inanimate object. Clearly much more research is needed to test the key propositions I have put forward about the singular clues, but these studies show a promising degree of support for them.
This is not a philosophical study, and I do not argue that causal understanding is founded on a general, abstract concept of causality. Nor am I committed to a particular philosophical position on causality. I could say, pragmatically, that causal understanding concerns anything that people perceive, recognize, or judge as a causal relation and leave it at that, but that problematic word “causal” appears in both parts of that characterization and therefore remains unexplained by it. I share with Hume (1978) the view that causality is a construction of the human mind, and I think it does not precisely correspond to anything in the laws of physics. But I come to that view through a very different route, namely the hypothesis about the origin of causal understanding that is summarized in this article. It is a construction that is reasonably consistently related to certain kinds of events in the world, in that certain kinds of events are likely to result in the occurrence of a perceptual causal impression and others are not, but it may not be possible to define exactly a class of physical events for which that is the case (see, e.g., Cartwright, 1999, 2004, 2007). It is possible to obtain somewhere near a defined class of events by referring to generative or productive relations (e.g., Bunge, 1963; Harré & Madden, 1975). I think it likely that everything that people perceive as a causal relation involves generation or production of something, such as change in object properties or motion. However, it is not clear that everything that involves generation or production of something is perceived as a causal relation. When a bottle is dropped onto a rock and shatters, the shattering of the bottle is produced by the impact, but most people do not believe that it is a causal relation, as the study reported here shows. Moreover, the things that people may believe to be causal relations go well beyond their perceptual impressions, in categorical terms. People can believe that non-events are causes (e.g., Wolff, Barbey, & Hausknecht, 2010), and they can seek causes of things staying as they are (e.g., saying that the kitchen wall continues to be white because the decorator failed to turn up). Nonetheless, favoring generative relations is sufficient to distance me from Hume's radical empiricist philosophy. If causality is just a construction of the human mind, there is more to it than the mere empirical conditions that give rise to it.
The term “pathetic fallacy” was first used by the art critic John Ruskin to refer to the attribution of strong, usually negative feelings to inanimate nature or to living things such as plants that lack the capacity for such feelings. The meaning of the term has broadened since then and now refers to any attribution of human qualities or capacities to living or non-living things that do not possess them.