A causal narrative?

source: Edward Tufte, edwardtufte.com

In a recent post I referred to the idea of a causal narrative (link). Here I would like to sketch out what I had in mind there.

Essentially the idea is that a causal narrative of a complicated outcome or occurrence is an orderly analysis of the sequence of events and the causal processes that connected them, leading from a set of initial conditions to the outcome in question. The narrative pulls together our best understanding of the causal relations, mechanisms, and conditions that were involved in the process and arranges them in an appropriate temporal order. It is a series of answers to "why and how did X occur?" designed to give us an understanding of the full unfolding of the process.

A narrative is more than an explanation; it is an attempt to “tell the story” of a complicated outcome. So a causal narrative will include a number of causal claims, intersecting in such a way as to explain the complex event or process that is of interest. And in my view, it will be a pluralistic account, in that it will freely invoke a number of causal ideas: powers, mechanisms, necessary and sufficient conditions, instigating conditions, and so forth.

Here is how I characterized a historical narrative in New Contributions to the Philosophy of History:

What is a narrative? Most generally, it is an account of the unfolding of events, along with an effort to explain how and why these processes and events came to be. A narrative is intended to provide an account of how a complex historical event unfolded and why. We want to understand the event in time. What were the contextual features that were relevant to the outcome — the settings at one or more points in time that played a role? What were the actions and choices that agents performed, and why did they take these actions rather than other possible choices? What causal processes—either social or natural—may have played a role in bringing the world to the outcome of interest? (29)

We might illustrate this idea by looking at the approach taken to contentious episodes and periods by McAdam, Tarrow, and Tilly in Dynamics of Contention. In their treatment of various contentious periods, they break the given complex period of contention into a number of mechanisms and processes, conjoined with contingent and conjunctural occurrences that played a significant causal role in the outcome. The explanatory work that their account provides occurs at two levels: the discovery of a relatively small number of social mechanisms of contention that recur across multiple cases, and the construction of complex narratives for particular episodes that bring together their understanding of the mechanisms and processes that were in play in this particular case.

We think what happens within a revolutionary trajectory can better be understood as the result of the intersection of a number of causal mechanisms. We do not offer a systematic account of all such mechanisms and their interaction in a sample of revolutionary situations. Instead, we use a paired comparison of the Nicaraguan revolution of 1979 and the Chinese student rebellion of 1989 to zero in on one processes in particular: the defection of significant elements from a dominant ruling coalition. (kl 2465)

The narrative for a particular case (the Mau Mau uprising, for example) takes the form of a chronologically structured account of the mechanisms that their analysis identifies as having been relevant in the unfolding of the insurgent movement and the government's responses. MTT give attention to "episodes" within larger processes, with the clear implication that the episodes are to some degree independent from each other and are amenable to a mechanisms analysis themselves. So a narrative is both a concatenated series of episodes and a nested set of mechanisms and processes.

Robert Bates introduces a similar idea in Analytic Narratives under the rubric of “analytic narrative”. The chief difference between his notion and mine is that his account is limited to the use of game theory and rational choice theory to provide the linkages within the chronological account, whereas I want to allow a pluralistic understanding of the kinds and levels of causes that are relevant to social processes.
Here is a brief account of what Bates and his collaborators mean by an analytic narrative:

The chapters thus build narratives. But the narratives are analytic narratives. By modeling the processes that produced the outcomes, we seek to capture the essence of stories. Should we possess a valid representation of the story, then the equilibrium of the model should imply the outcome we describe—and seek to explain. Our use of rational choice and game theory transforms the narratives into analytic narratives. Our approach therefore occupies a complex middle ground between ideographic and nomothetic reasoning. (12)
...
As have others, however, we seek to return to the rich, qualitative, and descriptive materials that narratives offer. And, as have others, we seek an explicit and logically rigorous account of the events we describe… We seek to locate and explore particular mechanisms that shape the interplay between strategic actors and that thereby generate outcomes. Second, most of these [other] literatures are structural: they focus on the origins and impact of alignments, cleavages, structures, and institutions. Our approach, by contrast, focuses on choices and decisions. It is thus more micro than macro in orientation. By delineating specific mechanisms and focusing on the determinants and impacts of choices, our work differs from our predecessors. (12-13)

A narrative typically offers an account of an historically particular event or process: the outbreak of a specific war, the emergence of ethnic conflict at a specific place and time, or the occurrence of a financial crisis. This places narratives on the side of particular social-science analysis. Is there a role for generalization in relation to narratives? I think that MTT would suggest that there is not, when it comes to large event groups like revolutions. There is no common template of revolutionary mobilization and regime collapse; instead, there are local and national interactions that constitute recurring mechanisms, and it is the task of the social scientist to discover the linkages and contingencies through which these various mechanisms led to revolution in this case or that. MTT try to find a middle ground between particularity and generalization:

Have we only rediscovered narrative history and applied to it a new, scientistic vocabulary? We think not. While convinced of the futility of deducing general covering laws of contention, we think our program -- if it succeeds -- will uncover recurring sets of mechanisms that combine into robust processes which, in turn, recur over a surprising number and broad range of episodes. (kl 3936)

In my view, anyway, a narrative describes a particular process or event; but it does so by identifying recurring processes, mechanisms, and forces that can be discerned within the unfolding of the case. So generalizability comes into the story at the level of the components of the narrative -- the discovery of common social processes within the historically unique sequence of events.

Causal narratives, mechanisms, and powers

A million termites move around industriously without supervisors or external coordination.  Some months later, a great structure has arisen — a termite cathedral mound. It is a structure that has apparent functionality (figure 2), it is oriented to the sun in a way that optimizes its ability to handle heat and cold, and the design plainly exceeds the cognitive or practical capacity of any single termite. How do they do it? (The BBC video below describes these mounds and their construction.)

This is a hard question because we know quite a bit about what the termites do not do. They do not have architects and project managers; they do not have blueprints guiding their work; they do not have a master plan. Instead, millions of independent organisms somehow coordinate their actions in ways that collectively result in the large structure.

 figure 1. The mound

 

figure 2. The structure of the mound

 

We would like to have an explanation for how this works; how the individual insects within this population behave in the ways that are necessary to create this vast complex structure.

One way of putting our explanatory needs here is to say that we are asking for a mechanism: what is the mechanism or ensemble of mechanisms that produce the collective behavior leading to the construction of the mound? What is it about the behavioral code of the insect that permits this collective behavior? This way of putting the problem is to highlight the mechanisms approach.

But we might better say, we are asking for an explanatory narrative, including elements like these:

• The insects have such-and-so behavioral routines (algorithms) embedded in their nervous systems.
• Behaviors are triggered by environmental circumstances and the activities of other insects around them.
• The triggered behavior in each insect contributes to a pattern of activity that leads to progressive “building” of the mound.

The force of the explanation hinges on the details we can learn about these powers and capacities of the termites as a species -- these behavioral algorithms. We want to know something crucial about the powers and capacities of the individual insects; we want to know how their routines are responsive to environment and other insects; and we want to know how the emerging structure of the mound leads to the modified activities of the insects over the process of construction.

This narrative highlights a topic we have considered several times before -- the idea of the causal powers of an entity. Most basically, we might look at the individual worker termite as robot controlled by a complex algorithm -- "when external circumstances X,Y,Z arise, carry out the Z routine." The causal powers of the individual worker termite are determined by its algorithm and its physical capacities -- salivation, moving around, carrying bits of mud, and so forth. And the task of explanation is to discover the nature of the algorithms and the ways in which the resulting behaviors aggregate to the observed physical structure of the mound.

 

We might observe, for example, that a certain kind of insect navigates a maze by following a simple rule: always keep the wall on your left. This rule will sometimes work well; sometimes it will not. But this observation suggests that the insect's central nervous system encodes the decision-making rule in this way. And we might also infer that "maze navigation" is important for the survival of the insect in its normal environment, and so its navigational algorithms will have been refined through natural selection.

 

We would also like to know something else about the insects and their powers: how did they come to have these particular capacities and algorithms? Here we have a well established explanation, in the form of the theory of the gene, natural selection, and the evolution of species characteristics through differential reproductive success. Here the explanatory challenge is to piece together the nature of the algorithms that would suffice to account for the observed collective outcomes.

It is also of interest in this example that there is a large field of research and discovery within complexity research that hinges on discovering the complex collective patterns that can emerge from simple routines at the level of the individual agent. For example, the "Game of Life" illustrates the power of cellular automata in generating complexity out of simple agent-level routines.

 

This example is a useful one, not primarily for entymologists, but for us as philosophers of social science. What would an explanation of this phenomenon look like? And a little bit of reflection seems to take us in the direction of some familiar ideas: the idea of things having causal powers that govern what they can do, the idea of the aggregation of complex outcomes from independent activities of large numbers of agents (agent-based models), and the idea that a good explanation gives us an empirically supportable understanding of how something works.

"How does it work" questions

Source: Karl Ove Moene in Alternatives to Capitalism, p. 85

One of the strengths of the causal-mechanisms approach to social explanation is how it responds to a very fundamental aspect of what we want explanations to do: we want to understand how something works. And a mechanisms account answers that question.

Let’s consider an example in detail. Suppose we observe that worker-owned cooperatives (WOC) tend to respond differently to price changes for their products than capitalist-owned firms (COF) when it comes to production decisions. The WOC firm will conform to this rule: “The higher the output price, the lower will be the supply” (85), whereas the COF firm will increase employment and supply. This is referred to as the Ward problem.

We would like to know how that comes about; what are the organization's processes and interactions that lead to the outcome. This means that we need to dig deeply into the specific processes that lead to production and employment decisions in both kinds of enterprises and see how these processes lead to different results.

The key part of the explanation will need to involve an analysis of the locus of decision-making that exists within the enterprise, and a demonstration of how the decision-making process in a WOC leads to a different outcome from that involved in a COF.

Here is how Karl Ove Moene analyzes this problem in “Strong Unions or Worker Control?” (Alternatives to Capitalism).

A production cooperative with worker control is defined somewhat restrictively as follows:

1. Productive activities are jointly carried out by the members (who in this case are the workers).
2. Important managerial decisions reflect the desires of the members, who participate in some manner in decision making.
3. The net income (income after expenses) is divide among the members according to some formula.
4. The members have equal rights, and important decisions are made democratically by one person, one vote. (84)

A capitalist firm acts differently:

1. Productive activities are carried out by wage laborers and directed by management controlled by the owners.
2. Important managerial decisions reflect the desires of the owners of the enterprise. 
3. Producers are paid a wage set by the labor market. The net income is assigned as profits to the owners.
4. Producers have no right of decision-making in production decisions.

The assumption is that decision-makers in both settings will make decisions that maximize their income — in other words, narrow egoistic economic rationality. In the assumptions used here for the cooperative, this implies that decision-making will aim at adjusting employment and production to the point where "marginal productivity (VMP) equals the net income per member (NIM)” (85). These quantities are represented in the graph above. Here is the reasoning:

What happens if the output price increases? In real terms, net income per member increases, because the fixed costs deflated by the output price decreases. Hence the NIM curve in Figure 5.1 shifts upwards, while the marginal productivity curve remains in place. As a consequence, the optimal number of members in the coop decreases and the firm's supply decreases the higher the output price. [Hence the coop lays off excess workers.] (85-86)

(Actually, this is what should happen in the long term. Moene goes on to show that the coop would not behave this way in the short run; but he acknowledges that the economic reasoning is correct. So for the sake of my example, let's assume that the coop behaves as Ward argues.)

The mechanism that distinguishes the behavior of the two kinds of firm is easy to specify in this case. The mechanism of individual decision-making based on rational self-interest is in common in the two types of firms. So the explanation doesn't turn on the mechanism of economic rationality per se. What differs across the cases is the collective decision-making process and the interests of the actors who make the decisions in the two cases. The decision-making mechanisms in the two cases are reflected in principles 2-4. The coop embodies a democratic social-choice rule, whereas the capitalist firm embodies a dictatorship choice rule (in Kenneth Arrow's sense -- one actor's preferences decide the outcome). A democratic decision about production levels leads to the reduction-of-output result, whereas a dictatorship decision about production levels leads to the increase-of-output result in these circumstances. And in turn, we are able to say that the phenomenon is explained by reference to the mechanism of decision-making that is embodied in the two types of firms -- democratic decision making in the coop and autocratic decision making in the capitalist firm.

 

This is a satisfying explanation because it demonstrates how the surprising outcomes are the foreseeable results of the differing decision processes. It identifies the mechanisms that lead to the different outcomes in the different circumstances.

 

This example also illustrates another interesting point -- that a given mechanism can be further analyzed into one or more underlying mechanisms and processes. In this case the underlying mechanism is the postulated model of action at the individual level -- maximizing of self-interest. If we postulated a different action model -- a conditional altruism model, for example -- then the behavior of the system might be different.

(I think this is a valid example of a mechanisms-based social explanation. Others might disagree, however, and argue that it is actually a deductivist explanation, reasoning from general characteristics of the "atoms" of the system (individual actors) to aggregate properties (labor-expelling collective decisions).)