The purpose of theory in the social sciences is (1) description, (2) explanation and (3) prediction – or some combination thereof (Singer 1961). Milton Friedman (1953) argues that a good theory is a theory that generates good predictions, that is, predictions that are better than those generated by rival theories. In this view, the purpose of economics and economic theory is prediction, not explanation. In some deeper sense, prediction is explanation. There is nothing over and above explanation than the degree of accurate prediction. Prediction is explaining from such a positivist point of view. Most historians, by contrast, seek to generate descriptions and explanations, but rarely if ever predictions. But what exactly constitutes an explanation?
Traditionally both philosophers of social science and social scientists have distinguished between Verstehen (understanding) and Erklären (explaining) (Hollis & Smith 1991; Hollis 1994). This is a distinction that goes back Wilhelm Dilthey and Heinrich Rickert and was later picked up by Max Weber. At risk of oversimplification, understanding involves the interpretation of human action, while explaining is generally focused on objective or objectifiable social facts. Understanding often makes use of counterfactual theories of causation rather than a deterministic (or statistical) regularity view of causation, as does explanation. Understanding typically involves the meaning of an action from the agent’s point of view and focuses on the subject and subjective meaning. This requires what Weber calls ‘interpretive understanding’ of the world on the part of the analyst. In cultural anthropology and verstehende Soziologie, for example, this approach is associated with empathic understanding (for explanatory purposes) and participatory observation (for descriptive purposes).
Erklären is generally less comfortable with subjectivity how agents view the world and seeks to make use of objective or quantifiable social facts or objectifiable social phenomena. Understanding looks for reasons, explaining for causes. Understanding is more closely associated with individualism, explaining generally with holism. Understanding frequently uses 'thick description' instead of abstract, controlled quantification. It tends to generate idiographic rather than nomothetic knowledge. Understanding typically features prominently in historical writing, especially the history of unique, exceptional events. Explaining is geared towards regularity and a certain degree of replicability. It is also associated with social science. Historians lean lean towards understanding and social scientists, and especially behaviorists of the Skinnerian persuasion, lean towards explaining. Understanding is associated with Geisteswissenschaft (humanities) and explaining with Naturwissenschaft (natural science). Explaining is tied to positivism and understanding to hermeneutics.
The symbolic anthropology of Clifford Geertz (1973) would represent one end of the spectrum. Large-N quantitative studies relying on highly aggregated variables the other. Again history as an intellectual pursuit leans towards thick description, meaning and ideas (Carr 1961, Collingwood 1946, Evans 2001, Winch 1958), tough there are notable exceptions such as quantitative history and the Écoles des Annales, not to mention postmodernists. Understanding is often associated with intentionality. According to Weber, for example, intentional actions can be (1) zweckrational, (2) wertrational, (3) affective or (4) traditional. To make sense of individual and social behaviour, an analysts needs to make sense of a agent's actions and how an agent interprets the world. In Weber's view, this allows for ‘explanatory understanding’ and even ‘causal interpretation’ (for a thoughtful discussion, see Nusser 1986). Positivists would take exceptions to assigning causal power how an agent attributes meaning to the world. Even in Weber's verstehender Soziologie, the analysts needs to understand agents' action in light of how they see the world.
Overlaying the understanding/ explaining distinction with the issue of causation complicates the picture further. This is not the place to get into debates about contingency/ necessity and counterfactual/ regularity/ process theories of causation or retrace the debate initiated by Hume, Kant and Descartes. Suffice it to observe that it can be somewhat unsatisfying to identify agency, and especially individual agency, as a cause of a historical or social event (Kincaid 1997). What caused WWII? Adolf Hitler. On the other hand, what 'caused' the building to burn down? An arsonist (as opposed to an electrical short-circuit). Often, and especially in the case of singular, unique and important historical events, one often expects to be given an answer to a why question. An answer to a why question often involves intentionality. In the case of understanding, a reason (motive) can offer a satisfactory answer. In the case of explaining, a cause can offer a satisfactory answer, that is, do without interpretation, meaning and intentionality. Explaining often involves subsuming a case under an observed or established or statistical regularity or general law. (As an aside, this creates a reference problem. True, if a large number of cases is analysed, a multi-factorial regression analysis may provide a satisfactory answer. But if one wishes to explain/ predict a specific event, the outcome is necessarily binary and the probability is either zero or one. An analyst making use of a detailed understanding of a specific case may (may!) be better positioned to make a prediction, even if the epistemological foundation of such a prediction appears weak. Again, this suggests that explaining pursues different epistemic goals than understanding.)
The regularity and the counterfactual theory of causation do different things. Finding regularities and using regularities to make predictions often involve statistical statements. Counterfactual approaches to causation, more compatible with understanding, are capable of dealing with individual factors, events, often individual action as the 'cause' of another event. (Admittedly, this distinction is perhaps somewhat too stark, but not indefensible.) Another related important difference is that the regularity view generates a probabilistic forecast, but does not necessarily identify an underlying causal mechanism (sometimes it does). In case of a counterfactual analysis of causation, an underlying causal mechanism is posited or at least implied. With respect to causal analysis, understanding and explaining approaches tend to differ, too.
The regularity and the counterfactual theory of causation do different things. Finding regularities and using regularities to make predictions often involve statistical statements. Counterfactual approaches to causation, more compatible with understanding, are capable of dealing with individual factors, events, often individual action as the 'cause' of another event. (Admittedly, this distinction is perhaps somewhat too stark, but not indefensible.) Another related important difference is that the regularity view generates a probabilistic forecast, but does not necessarily identify an underlying causal mechanism (sometimes it does). In case of a counterfactual analysis of causation, an underlying causal mechanism is posited or at least implied. With respect to causal analysis, understanding and explaining approaches tend to differ, too.
Starting from the assumption of purposeful, rational intentions and actions, whatever the ends of the purposed actions (e.g. security gains, economic gains, affective gains), is an approach widely used in policy analysis. Policy analysis is closer to an understanding than an explaining approach. They do however often include, at least implicitly, insights from an explaining approach in terms of regularity. In much of International Political Economy, it is not possible to explain and predict individual policy decision without at least implicit assumptions about the distributional-economic consequences of such decision. Often knowledge about the distributional consequences is nomothetic and is derived from models or statistical analysis (or both). Thanks to game theory, an approach combining understanding and explaining allows for the explanation and prediction of agents' behaviour in an interactive context. If interactive behavior is involved, one needs to determine what type of game the various players play. Do the seek absolute gains or do they seek relative gains, that is, do they engage in a zero- or non-zero-sum game? Microeconomic analysis is a variation of this approach. Where it deals with perfectly competitive markets that lead to the elimination of agents in case of incompatible behaviour, it has clear associations with Darwin's theory of evolution, Waltzian neo-realism and perhaps even more so Gilpinian realism (Gilpin 1981, Waltz 1979, see Wolforth 2011). Bueno de Mesquita (2010) claims that a game-theoretical framework has significant predictive power. One only needs to know the following: Who are key actors? How influential are they? How committed are they to the issue at hand? How resolved are they? It is often the only sensible approach to take. Simply subsuming a case under a (arbitrary) reference class will generally be of listed usefulness given that each individual case has important idiosyncratic elements that make it difficult to feel comfortable about reference class choice.
Individuals (often) respond to structural incentives and system-level pressures influence behaviour. An agent-centred sociology, as is often used in policy analysis, focuses on the costs and benefits of various courses of actions under given constraints. Rationality and interest maximization offer an explanation, while behaviour incompatible with systemic constraints will prove unsuccessful and disappear. In the case of atomistic, perfectly competitive markets, the constraints are a given. In a less structured context (e.g. oligopolies), the systemic constraints can potentially be manipulated by the agents. The age-old question whether it is the agent or the structure that determines outcomes is akin to the economic question whether supply creates demand or demand creates supply. Anthony Giddens (1984) has proposed the term ‘structuration’. A firm operating in perfectly competitive markets will experience the market structure as given and forces the firm into adaptive behaviour. A firm operating in oligopolistic or monopolistic markets confronts a situation where it can potentially manipulate the systemic constraints it faces (game theory often comes in handy in this situation, analytically speaking). A firm operating in a monopolistic market has even fewer constraints. Systemic constraints can be more or less binding and in some cases they can be manipulated, but not in others. An understanding of an agent's intention and whether an agent acts in zweckrational or wertrational way requires understanding - at least in a particular historical situation. Alternatively, agent preferences can simply be postulated, behaviour measured and then compared whether it matches the model's (explaining theory) prediction.
Individuals (often) respond to structural incentives and system-level pressures influence behaviour. An agent-centred sociology, as is often used in policy analysis, focuses on the costs and benefits of various courses of actions under given constraints. Rationality and interest maximization offer an explanation, while behaviour incompatible with systemic constraints will prove unsuccessful and disappear. In the case of atomistic, perfectly competitive markets, the constraints are a given. In a less structured context (e.g. oligopolies), the systemic constraints can potentially be manipulated by the agents. The age-old question whether it is the agent or the structure that determines outcomes is akin to the economic question whether supply creates demand or demand creates supply. Anthony Giddens (1984) has proposed the term ‘structuration’. A firm operating in perfectly competitive markets will experience the market structure as given and forces the firm into adaptive behaviour. A firm operating in oligopolistic or monopolistic markets confronts a situation where it can potentially manipulate the systemic constraints it faces (game theory often comes in handy in this situation, analytically speaking). A firm operating in a monopolistic market has even fewer constraints. Systemic constraints can be more or less binding and in some cases they can be manipulated, but not in others. An understanding of an agent's intention and whether an agent acts in zweckrational or wertrational way requires understanding - at least in a particular historical situation. Alternatively, agent preferences can simply be postulated, behaviour measured and then compared whether it matches the model's (explaining theory) prediction.
Much (but not all) said and done. Using an understanding approach to account for individual decisions, idiosyncratic phenomena or unique events often leads to the charge of a 'just-so explanation' at best and a highly subjective, arbitrary, solipsistic story at worst. However, small, random events can and do cause (sic!) large, important events in deterministic (non-linear) systems. And such large-scale outcomes are unpredictable in principle even if the processes that lead from the small, random event to the significant outcome are completely deterministic (Strogartz 2014). Moreover, to the extent that unique events are the effects of a cause at all (something the regularity view of causation/ explanation would deny), such events may also be accounted for by causal complexity. Metaphysically speaking, there is no reason to believe that different combinations of independent variables (or background conditions) cannot have the same effects on a selected dependent variable (Pearl 2018, Ragin 2000). Causal complexity may be difficult to model and prove, but that does not mean that it can be dismissed out of hand. Social scientists have made use of conjunctural causation for ages (e.g. Barrington Moore, Theda Skocpol and Charles Tilly). The point is that causation is a more difficult concept than positivist and the explainers assume. In short, the explaining camp overreaches when it dismisses the understanding epistemic framework as subjective and non-causal in principle. At the same time, it is far from obvious that an explaining approach relying on quantification and correlation rests on a sounder epistemic foundation given well-known issues related to statistics, correlation and causation.
Blaise Pascal’s suggestion that the history of the world would have been different had it not been for Cleopatra’s nose (Marc Anthony and Caesar falling for her and all that) appears perfectly sensible in light of our most sophisticated scientific models. After all, complex systems and chaos theory do point to the importance of small random events in terms of large outcomes (Miller & Page 2007). Then again, if replicability and regularity are seen as the hallmarks of science, Pascal's insight will not find much resonance with the 'explaining camp'. There is room for understanding and explaining approaches once one reminds oneself that they tend to have different epistemic objectives. Ultimately, they do form a symbiotic relationship as meaning and fact, individual and system, idiosyncratic and nomothetic knowledge are often invariably interconnected. Yet an awareness of how and where they differ undoubtedly makes one a better, more thoughtful and self-aware analyst.
Blaise Pascal’s suggestion that the history of the world would have been different had it not been for Cleopatra’s nose (Marc Anthony and Caesar falling for her and all that) appears perfectly sensible in light of our most sophisticated scientific models. After all, complex systems and chaos theory do point to the importance of small random events in terms of large outcomes (Miller & Page 2007). Then again, if replicability and regularity are seen as the hallmarks of science, Pascal's insight will not find much resonance with the 'explaining camp'. There is room for understanding and explaining approaches once one reminds oneself that they tend to have different epistemic objectives. Ultimately, they do form a symbiotic relationship as meaning and fact, individual and system, idiosyncratic and nomothetic knowledge are often invariably interconnected. Yet an awareness of how and where they differ undoubtedly makes one a better, more thoughtful and self-aware analyst.
