Causal Explanations

Abstract

There are several forms of explanation, one of which being causal explanations. Causal explanations are often descriptions of mechanisms, i.e., descriptions of how a state change in one object, labelled ‘the cause’, is transmitted through a number of intermediate objects to the final effect, i.e., a state change in another object. So the fundamental structure of mechanistic explanations is that of chained cause-effect relations.

The main points of this chapter are:

• Causal explanation is one kind of explanation beside several other kinds.

• A causal explanation often consists of describing the mechanism by which the cause produces its effect.

• Reasons for human actions are often viewed as causes of those actions, but that is controversial.

• Three types of causal explanations in terms of mechanisms are confounder mechanisms, feedbacks, and bifurcations.

1 Explanation-Seeking Questions

Scientific research aims at answering questions (Van Fraassen, 1980). As these answers accumulate, the knowledge grows, is corrected or completely revised. The starting point of research is usually a set of questions raised by practical concerns, or by earlier scientific research. Scientists’ goal is to find convincing and correct answers to these questions. An important research skill is the ability to formulate these questions in a fertile manner. Part of this skill is the ability to break up the original set of questions into more concrete questions that can be answered by means of empirical research. Another part of the skill is research imagination, which allows the researcher to see which questions her methods and data can answer. Finally, the third part consists of the ability to design and conduct the research in a manner that convinces the audience that the answers presented by the researcher are better justified than competing accounts.

Most everyday research questions are descriptive. They ask about the facts regarding the world. Researchers seek to find out new facts about the world, but just as often they aim to check, correct or challenge what we believe to be facts. These questions usually start with words like Who? What? When? Where? Which? How many? How much? For example, the researchers studying Pleistocene extinctions ask questions like: Which mammal species went extinct during the late Pleistocene period? Where did these species live? What were their habitats? How large were their populations? When did they go extinct? Etc. These questions are often extremely tricky to answer conclusively.

Answering descriptive questions like these is the backbone of all scientific inquiry. But the scientific ambition is not limited to answering these questions; scientists also wish to answer explanatory questions. These questions often start with words like Why? or How? These questions take the answer to a descriptive question as their starting point and ask why that fact is the way it is. Many facts are puzzling to us, and we want to know why they happened or why they are one way rather than some other way. The fact to be explained is called the explanandum, and what explains it, the explanans.

Not all why-questions are explanation-seeking questions. Sometimes we ask ‘Why?’ when we want justification for a belief (Hempel, 1965) or for an action. For example, when someone claims that there was a mass extinction of large mammals during the late Pleistocene, it is reasonable to ask for some reasons to believe this claim. Because we do not yet believe the claim, we ask a justification-seeking why-question. However, once we take the claim to be a fact, we probably want to know what caused those extinctions. We want to understand why they happened. In this case, we would be asking an explanation-seeking why-question.

In the case of an explanation, the factuality of the explanandum is a presupposition of the explanation-demand, without which the question does not make sense. So, for example, if we ask (Barnosky et al., 2004), ‘Why did most species of megafauna go extinct during the late Pleistocene (50,000–10,000 BC)?’, we are presupposing that such a mega-extinction really happened. Sometimes the presuppositions of the question are not as obvious. For example, our question can be either read as assuming that there is a single cause for the mass extinction, or it can be read more loosely as allowing multiple independent causes. The first reading incorporates a quite strong assumption that easily can be false. It is possible that megafauna perished from different continents due to independent causes.

It is not always obvious that everybody shares the same presuppositions. A well-known anecdote about the famous 1930s bank robber Willie Sutton captures this. When a journalist asked Sutton why he robbed banks, Sutton responded, ‘Because the money is there.’ Clearly, he had a different contrast in mind than the journalist who was asking about his career choice.

One cannot expect that one answer to an explanation-seeking question could explain everything about the explanandum, e.g., the late Pleistocene extinctions. Typically we can explain only a certain aspect of a complicated event. A useful way to make the explanandum more precise is to articulate the intended contrast. The contrast describes an alternative state of affairs (the foil) that could have occurred instead of the fact. For example, we could ask why did the extinctions happen during the late Pleistocene rather than some earlier or later period? Alternatively, we could ask why did the extinctions happen almost simultaneously rather than stretched over a longer period of time? We could also ask why the extinctions concentrate on megafauna rather than species of different sizes or all species? An explanans that provides an insightful answer to the last question might be quite uninformative about the other two questions, and vice versa. The contrast helps to pick up a causal difference-makerfrom the complicated causal history of Pleistocene ecology, and different contrasts can highlight very different difference-makers. Thus it makes sense to split the general explanation-seeking question into a series of more precise questions. The articulation of contrasts is a useful way to make explanation-seeking questions more precise (Van Fraassen, 1980; Garfinkel, 1981; Lipton, 1991).

Often our curiosity arises when we observe something unexpected, and we ask why things did not turn out as expected. We are quite curious when a person behaves in an unexpected manner, for example when he pours his coffee over his own head, but we are not usually asking for an explanation for his ordinary coffee drinking. The origins of our expectations might be in what we typically observe, theoretical predictions, or normative ideals. In everyday life, we usually explain surprising things, but in scientific research, also obvious things can become objects of curiosity (Hesslow, 1983). Why is the grass green rather than any other colour, or why there are two, rather than three biological sexes, are both meaningful scientific questions that are not raised outside science, except maybe by small children.

More generally, explanation-seeking questions are typical in theoretically oriented basic research. However, it would be a mistake to assume that such questions can be ignored by more practically oriented researchers. Reliable answers to such questions are usually descriptions of mechanisms and such descriptions are needed for the expansion of both theoretical and practical knowledge.

2 Explanations

The word ‘explanation’ can refer both to the activity of providing an explanation and to the product of that activity. While most discussions of explanation is focused on the latter, it is good to remember that explanation-seeking and explanation-giving are continuous social activities; we rarely provide complete explanations. Typical explanations, even in science, are limited by pragmatic contexts, they are more like sketches of explanations, leaving out relevant components that the reader can be assumed to be aware of in advance. Usually, in a given context we only highlight the salient features of the explanation and leave many background conditions unarticulated. This means that an explanation might have problematic presuppositions that we are not fully aware of. Furthermore, many of the explicit assumptions might be promissory: we believe that the facts that we have assumed are indeed the case, but we do not have sufficient evidence to support them. So, if these presumptions turn out to be false, we have to reject or at least revise the explanation demand.

An answer purporting to be an explanation is expected to be true. An explanation that relies on false facts cannot be the proper explanation of an empirical observation. However, consisting of true statements is not enough, it also has to be relevant. First, it must answer the question by relieving the audience of the puzzlement the explanandum gave rise to (Lipton, 1991). Furthermore, this has to be done in a correct manner: it is not enough that the audience just thinks that they have understood or have a sense of understanding, as these metacognitive states are quite often unreliable. The audience might not get the explanation, for example, because it lacks sufficient background knowledge. It is also possible that an explanation is great in providing understanding but is unfortunately not true. Cases like this are called possible explanations (Hempel, 1965; Lipton, 1991). They are explanations that would have been satisfactory if their assumptions were true. Possible explanations are often an important element of the explanatory inquiry. For example, in the case of late Pleistocene extinctions, it is important to articulate a set of possible alternative explanations and then proceed to find evidence that discriminates between the alternatives (Barnosky et al., 2004; Stuart, 2014). Without the set of alternative explanations, we could easily mistakenly accept our first explanation as the correct one.

3 Different Kinds of Explanations

Answering a demand for an explanation is sometimes to provide a cause, or several causes, for the explanandum. But there are several other kinds of explanations that at least at first sight do not provide causes. We will here briefly discuss four such kinds before we delve into causal explanations: constitutive explanations, teleological explanations, functional explanations and intentional explanations.

3.1 Constitutive Explanations

In a constitutive explanation the capacities of a whole are explained by capacities of its components and their organisation (Ylikoski, 2013). The relation between the parts and the whole is not causal, hence this in not a case of causal explanation, which usually relates events to each other. However, it should be recognised that basically the same ideas about explanation apply to constitutive explanation that applies to causal explanation. Furthermore, constitutive explanation relates the causal capacities of the whole to the causal capacities of the parts, and, furthermore, changes in capacities are causal processes. So it would be highly misleading to say that constitution is completely unrelated to causation. Constitutive relations are an integral part of a causal picture of the world. The reason we have to recognise their difference is that confusing part-whole relations with causal relations can lead to confused causal analysis.

3.2 Teleological Explanations

Another candidate for non-causal explanation is teleological explanation, which explains a process by an imagined goal, rather than by its causes. However, all forms of ‘teleological’ explanations in the sciences are actually subspecies of causal explanation (Elster, 1989). Biology is full of teleological explanations, but modern biology only accepts those that are supported by appropriate causal mechanisms. Natural selection is the prime example of such a mechanism. Consider, as an example, the human sclera, the white of the eye, which is a rare feature among great apes. According to the cooperative eye hypothesis, humans have white sclera because they facilitate telling the direction of gaze, which greatly facilitates non-verbal communication and coordination of action. The hypothesis explains the colour of the sclera by its beneficial consequences. However, for the hypothesis to be true, the claim has to be true about the past: the colour of sclera must be a heritable trait, and it must have given a relative fitness advantage to its carriers in earlier phases of human lineage because it facilitates cooperation. For example, if the colour is a by-product of some other trait, then the hypothesis is false. Thus when unpacked, the teleological claim is, in fact, a claim about a causal history.

3.3 Functional Explanations

Functional explanations are sometimes used in the social sciences. However, there is a quite broad consensus that they require an underlying causal mechanism. Finding such mechanisms are quite demanding, so proper functional explanations are quite rare in the social sciences.

In passing one may observe that term ‘function’ sometimes is meant to express a causal relation, sometimes only a mathematical or logical relation, the latter being common in natural and social sciences. One variable can be a function of another variable, but that can be the case without there being any causal link, as we discussed in the preceding chapter. This is not restricted to quantitative variables; if one boolean variable (i.e. having only two values, e.g. male-female) is correlated to another one (for example, do/do not enter higher education) one may correctly say that the second variable is a (probabilistic) function of the former one. But whether the first variable is a cause of the second one is a further question. If there is little or no evidence for there being a causal link between functionally related variables, one can hardly say that one explains anything just by pointing out that one is a function of the other.

3.4 Intentional Explanations

Finally, there are intentional explanations, which play an important role in the social sciences. Intentional explanations are not properly teleological explanations, because the explanatory work is done by a representation of a mental state (consisting of desires and beliefs) that precedes the outcome. It should be observed that the mental state of the actor precedes what is to be explained, while the content of that mental state is an imagined future event or state of affairs. Furthermore, the outcome causally depends on the agent having that mental state; if the mental state had been different, the action would have been different, which would have made a difference to the outcome. Intentional explanation has an extensive list of causal background conditions without which the connection between the mental state and the outcome will not hold. People cannot bring about, i.e., cause, things in the world just by having thoughts about them.

Intentional behaviour, i.e. actions, are usually explained by giving the agent’s beliefs and desires; they are assumed to be the immediate causes of actions. From the point of view of everyday reasoning, the causal role of beliefs, desires and other mental states is quite obvious. We usually assume that beliefs and desires, i.e., reasons, can make a difference to the way we behave. Furthermore, in communication we attempt to influence each others’ mental states and thus influence their behaviour.

The role of interpretive understanding in causal explanation of action highlights the importance of qualitative research. While much of it is descriptive, it describes what different people think, experience, and strive for, hence it lays the ground for causal explanations of their actions. To causally explain action, we have to get people’s desires, feelings, and beliefs right. Similarly in institutional contexts we have to get right both the rules people follow and why they follow them.

4 Causal Explanation and Mechanisms

Apart from relevance and truth, an explanation requires the right kind of dependence between the explanans and the explanandum. Something is the case because it explains how facts are the way they are. For example, suppose that megafauna extinctions occurred because of human hunting. This is a claim about causal dependence: if there had not been extensive hunting of large mammals, the mass extinction would not have happened. Thus the explanatory claim is a claim about counterfactual dependence: if the cause had been different, the outcome would have been different too. Here the relevance criterion for the explanation is causal difference-making (Lipton, 1991; Woodward, 2003). While there is a huge number of things in the causal history of any event, the difference-making criterion helps us to pick the explanatorily relevant part of that causal history. If we have correctly identified the right contrastive explanandum, we now have an informative answer to our explanation-seeking question.

While this kind of a simple causal statement might sometimes be enough to explain an event, we often want additional information. First, all causal claims hold only when certain background conditions hold (Mackie, 1974), see Sect. 5.6. If the background conditions do not hold, the cause would not be able to make the difference to be explained. So the first piece of additional information concerns the relevant background conditions. A better grasp of the background conditions helps us to see how the change we cite as the cause is embedded into a larger causal configuration. It might also help us to understand how fragile the causal connection is. It might well be that the cause can bring about the effect only in very rare circumstances. Understanding the relevant causal configuration helps to answer questions about the preconditions of the causal relation and possibly about alternative causes for the effect.

Another additional piece of information concerns causal mechanisms, that is, how the cause brought about the effect (Craver, 2007; Hedström and Ylikoski, 2010). This involves the idea that causation is a process, and describing that process increases explanatory understanding. One could say that information about the causal mechanism answers the how-question behind the causal why-question.

Knowledge of causal mechanisms is valuable for multiple reasons. First, evidence about mechanisms can help to justify the causal claim. A causal claim is more credible if there is a known mechanism by which the cause could bring about the effect and there is evidence that this particular mechanism has been present in the case at hand. Second, together with knowledge about the background conditions, understanding of the causal mechanism helps to understand how robust or fragile the causal relation is and what kinds of factors could prevent or modify the effect. Third, the mechanism helps to organise the causal explanation to a narrative that is easier to comprehend than individual claims about causal dependencies. Fourth, general knowledge in human and biological sciences is often formulated in the form of mechanism-schemes rather than general law-like generalisations. A mechanism-scheme outlines what kind of cause and causal configuration can bring about a certain type of effect. The outline has to be filled in for any particular explanatory use, but it provides useful guidance for the search for causes. It is often the case that there are alternative mechanisms that could bring about a similar effect. In cases like these, it is useful to have a toolbox of possible mechanisms that helps to find evidence that discriminates between alternative mechanistic scenarios.

Is there a general way to define what a mechanism is? In the literature there are many competing definitions. The entities and processes studied by different sciences are quite heterogeneous, so it is difficult to provide a definition that is both informative and covers all examples of mechanisms. One widely cited definition is the following:

“A mechanism is a structure performing a function in virtue of its component parts, component operations and their organisation. The orchestrated functioning of the mechanism is responsible for one or more phenomena.” (Bechtel and Abrahamsen, 2005, 423)

This definition might work in some areas of biology, but its application to, for example, SES research is difficult. While it is easy to recognise the importance of parts, operations, and organisation, the definition does not give much guidance for the construction of mechanism-based explanation. For example, it does not solve the problem of relevance: which entities, activities and their relations should be included in the explanation? A crucial aspect is that a mechanism-based explanation describes the causal process selectively. It seeks to capture the crucial elements of the process by abstracting away the irrelevant details. But how do we determine what is relevant and less relevant?

While a general definition is impossible, it is possible to say something general about mechanisms (Hedström and Ylikoski, 2010). First, a mechanism is always a mechanism for something, so the explanandum plays an important role in its identification. Second, mechanism is an irreducibly causal notion. It refers to the entities of a causal process that produces the effect of interest. A correlation between cause and effect is not enough for a mechanism, as it is based on the idea that there is a continuous process by which the causal influence is transmitted from the cause to the effect. Third, when a mechanism-based explanation opens the black box, it discloses this structure. In other words, it makes visible how the participating entities and their properties, activities, and relations produce the effect of interest. For this reason, the suggestion that a mechanism just is a chain of intervening variables misses an important point, each link in the mechanism must be a causal link. Conceptualising the mechanisms requires theoretical thinking. However, it also generates a series of additional hypotheses that can be tested, thus opening additional avenues for confirming causal claims.

While the idea of mechanism-based explanation is appealing, it is good to recognise some dangers associated with the idea. First, while explanation in terms of mechanisms comes naturally to us, we quite often end up with mechanistic storytelling. This means that we are satisfied with the first sketchy mechanism-story that we could come up with and do not bother to consider alternative mechanisms or to check whether our story agrees with empirical evidence. Second, quite often people just name a mechanism rather than describe how it is supposed to work. While this kind of intellectual laziness is understandable, it is not supported by the core idea of mechanism-based explanation. The goal of mechanism-based theorising is not to create illusory understanding, but to fight it.

5 Some Special Mechanisms

Causal mechanisms can consist of several different types of structures. Here we will briefly discuss confoundermechanisms, feedback mechanisms and bifurcations, which are of particular interest in SES.

5.1 Confounder Mechanisms

Confounders are common in empirical sciences. A confounder is a non-observed variable which is a common cause of two observed and correlated variables. A strong correlation between two observed variables can be due to three possible causal connections, according to Reichenbach’s principle (see Sect. 7.1); variable X is one the causes of variable Y, or vice versa, or there is an unobserved common cause Z. This is often called a confounder.¹

The two mechanisms where a common cause, the confounder, produces a correlation between the two observed variables X and Y can be visualised by the following directed graph (Fig. 8.1):

Fig. 8.1

The variables X and Y are correlated because both depend on a common cause Z. N.b: there is no arrow between X and Y, since there is no causal mechanism going from X to Y!

The structural equation system for this situation is:

$$\displaystyle \begin{aligned} \begin{cases} X=: k_1Z \\ Y=: k_2Z \end{cases} \end{aligned}$$

where $k_1$ and $k_2$ are the coefficients of correlation, if there are no other common causes of X and Y. Since correlation is a transitive relation the correlation between X and Y is the product $k_1k_2$. We may thus infer that in order to observe even a weak correlation of e.g., r = 0.3 between X and Y the coefficients $k_1$ and $k_2$ must be rather substantial.

This in turn means that if there are unknown causes other than Z independently affecting X and Y, the coupling coefficients $k_1$ and $k_2$ will be weak, hence the correlation between X and Y will in such cases be very weak and often not discernible.

The time arrow must be interpreted with caution. It tells us that individual events of the type ‘variable Z has the value $z_i$’ occur earlier than the individual events of the types ‘variable X has the value $x_i$’ and ‘variable Y has the value $y_i$’. But the variables themselves, which are map**s from events to numbers, cannot be attributed times, since they are abstract entities.

5.2 Feedback Mechanisms

Feedback mechanisms are common in complex systems, see e.g. the following quote:

The essential element in any SES with persistent structure (e.g., an ecological community and its human dependents) is feedback (Csete and Doyle, 2002; Carlson and Doyle, 2002). In SESs, these feedbacks take the form of information-action loops wherein human individuals or groups extract information about the state of a system (e.g., an ecosystem), decide how to act on the system (e.g., which species to protect and which to harvest), and undertake the action, generating a response from the ecosystem (e.g., changing population size or distribution), that over time triggers system change and restarts the cycle (loop) (Anderies et al., 2007, 2019). (Anderies et al., 2022a, 3)

As discussed in Sect. 5.4, a feedback mechanism does not contradict the fundamental idea that an individual effect cannot precede its cause. When we talk about feedback mechanisms we always talk about relations between variables. Variables are abstract entities, sets of values (in the case of quantitative variables), or attributes of concrete events, objects or states of affairs (in the case of category variables). To repeat, abstract things such as sets or attributes do not exist in space and time. Hence in Fig. 8.2 there can be no time line.

Fig. 8.2

The feedback from Y to X goes via the variable Z

But we have a natural tendency to interpret figures of this type as if there are time relations between the items, and taking arrows as indicating processes in time. This is not correct for feedback diagrams! We have discussed use of diagrams more extensively in (Banitz et al., 2022b).

The structural equation system (cf. Sect. 6.5.2) for this case is

$$\displaystyle \begin{aligned} \begin{cases} X=: k_1Z \\ Y=: k_2X\\ Z=: k_3Y \end{cases} \end{aligned}$$

A variation of dz in Z will result in a variation $k_1dz$ in X, which will produce a variation $k_1k_2dz$ in Y, etc., hence the ‘strength’ of the feedbackmechanism is the product $k_1k_2k_3$. It is obvious how to generalise to any number of intermediate variables making up the feedback. Furthermore, if we have observed a system with an effective feedback, each of the three coupling coefficients must be high. It follows that if one is able to manipulate one of the couplings by an interventionone might break down the feedback.

5.3 Bifurcation Mechanisms

Functional relations (see Sect. 5.2.2) state not only that there are relations between variables, but also provide a detailed account of properties of these relations. Take for example a system of ordinary differential equations that describes a freshwater lake ecosystem exposed to nutrient runoff:

$$\displaystyle \begin{aligned} \begin{aligned} \frac{db}{dt} &=r_b \frac{n}{n+h_0}b-c_b b^2-k_1\frac {b^2p}{b^2+h_1^2} , \\ \frac{dp}{dt} &=k_2\frac {b^2p}{b^2+h_1^2}\frac {v}{v+h_2}-c_pp^2 - m_pp. \end{aligned} \end{aligned} $$

(8.1)

Letters b, p and v denote bream, pike and vegetation respectively and represent the key species in the lake ecosystem. Parameters in differential equations, denoted by $k_1$, $h_1$ and $k_2$ define the strength of interactions between bream and pike, parameters $r_b, c_b, c_p, m_p$ define ecological processes of each species and parameter n define bream response to the amount of nutrients in the lake water.

Figure 8.3 shows changes in the lake ecosystem dynamics due to changes in the amount of nutrients (i.e. values of parameter n). For small values of parameter n, the lake water is clear and bream levels stay low, but more intensive nutrient load (represented by higher values of n) can lead to eutrophication of the lake, increased bream levels, decreased pike levels and changed structure and functioning of the ecosystem. This creates a bistable region, where depending on the initial conditions, the lake can evolve toward clear or turbid state. Further increase in nutrients leads to turbid lake state. The shift from clear to turbid lake state due to increase in nutrient load (and parameter n) is an example of a regime shift, a phenomenon that can be explained by bifurcation mechanisms.

Fig. 8.3

Bifurcation diagrams for bream and pike in relation to the nutrient level. Red lines represent stable regimes, grey lines are unstable states. Arrows indicate nutrient increase and consequent regime shift from clear to turbid state

Bifurcation mechanism means that qualitative properties of system dynamics change due to changes in strength of individual interactions or drivers. We have discussed this at some length in (Radosavljevic et al., 2023).

6 Summary

Explanations in general, and scientific explanations in particular, are highly context-dependent because a number of background assumptions are usually made without being explicitly stated. Scientific explanations are in most cases explanation sketches, not complete explanations.

There are several types of explanations, one of which being causal explanations. Teleological and functional explanations are, on closer inspection, causal explanations.

Causal explanations are often given by describing a mechanism which tells us how the cause produces its effect.

Discussion Questions

• 1. How do you respond to an explanation request that has a tacit and false background assumption?

• 2. What is the difference between functional description and functional explanation?

• 3. Is there a way of describing a causal mechanism that is not built out of chains of variables coupled via structural equations?

• 4. Is it possible to discern any ultimate constituents of a chain of elements constituting a mechanism, constituents that need no further analysis? If so, why are these the ‘bottom’?

• 5. If a person believes that she will achieve a goal G by performing a certain action A, and if that person desires G, she will perform action A. Do you think that an explanation of this person’s action A consists of a description of this particular belief and desire? If so, is it a causal explanation?