hierarchical causation describes the influence of the microlevel, mesolevel and macrolevel on each other. The mesolevel is the level of analysis we are focusing on (also called the focal level), and we choose some scaling criteria for defining the hierarchy (Hölker & Breckling, 2002[1]). For example, if our focal level phenomenon is a bear, and we are interested in understanding it anatomically, its micro-level may be its cells, and its macrolevel may be its physical habitat. However, if we are more narrowly interested in the aggressive behavior of bears, the microlevel phenomenon might be the various hormonal states that accompany aggression, and the macrolevel might be the behavioral conditions that elicit aggression in that species of bear. The scaling criteria are physical in the first case, and behavioral in the second case.
Ecology requires concepts of hierarchical causation. The linear causation used in experimental science is less useful for ecology, because it is best observed when all variables are controlled but one. This suppresses the web of multiple interacting activities a subject might participate in, and that web is precisely what ecologists often want to study. Hierarchical causation is a conceptual framework that allows ecologists to illuminate how bottom-up, top-down and same-level forces interact to produce the phenomenon they want to understand.
One can generalize a bit about hierarchical causation and scaling criteria. Macro scale events typically occur over larger spatio-temporal spans than focal system events. They thus constitute the conditions which frame the activities at the focal level. Micro scale events typically involve the flux of material, energy and events which support the activities at the focal level. Their spatiotemporal frequencies are much higher that at levels above them. One can thus say that the spatiotemporal ‘grain’ is finer at lower levels, and that matter-energy fluctuations are faster at these levels.
Lower-level or upwards constraints are enabling constraints, in the sense that they make events at the focal level possible. Say that the focal level event is an episode of behavior: a bear fighting. For that focal level event, hormone levels would be a lower-level constraint – a necessary but not sufficient condition (Salthe, 1985[2]). Lower-level values generate possibilities and probabilities at the focal level without participating in focal level events. Hormones do not fight, for example. Lower-level constraints are necessary conditions for focal activity, but they are agnostic about sufficient conditions. Sufficient conditions exist only where goals and functions can be defined – at the focal level – as constrained by macro-level conditions.
Upper level constraints participate in focal-level events more indirectly. Upper level constraints are contextual or environmental constraints, and they reduce (or permit) the variety of options that systems of the focal level have for action. Cold weather for warm-blooded animals, for example, forces metabolic changes, changes in calorie intake or reductions in expenditure, which increases the value of enclosed shelter, etc. Upper level constraints in this example alter the cost structure at the focal level, but do not otherwise direct the activities at the focal level. Focal level systems themselves enjoy little or no upwards impact on these constraints. For example, we do not interact directly with the control parameters of seasonality, like the earth’s axis of rotation and orbital position around the sun. We cannot change these parameters as our strategy for managing seasonal temperature changes.
Buffering and emergence represent two other ways in which events at different scales can influence each other. For an example of buffering, the rain cycle over an ecosystem may not deliver water at sufficiently regular intervals to meet the water needs of many of its life forms. However, the structure of storages, reservoirs, channels and flows of water may be such that even with irregular rain, water is distributed predictably throughout the ecosystem at an essentially constant rate.
Emergent properties emerge in hierarchies when lower-level processes come together to produce focal-level properties that could not have been deduced from lower-level properties. Common examples of emergent behaviors include market interactions, which regulate the supply and pricing of goods around the world without any central entity to govern it, and flocking/herding/schooling formations among animals that allow them to benefit from the various properties of the collective as a unit. Both buffering and emergence are special instances of the kinds of downward and upward causality and constraint that characterize hierarchical systems.
Hierarchical causation describes many other systems at many scales, for example:
Animal Metabolism
• P-Level (bottom): Metabolic activities and pathways;
• A-Level: Cell structure, tissue/organ specialization;
• I-Level: Endocrine signaling, system-wide signaling of biological state changes;
• E-Level: Processing environmental signals of seasonal change, launching preparatory/anticipatory biological changes to be ready for seasonal change.
Strategic Purpose in Organizations
• P-Level (bottom): Transactions and discrete tasks;
• A-Level: Explicit plans for obtaining specific results with an allocation of resources, plans for allocating limited resources over a specified project set;
• I-Level: Stated goals of the organization for a time period, understood by each department, manager and employee, allowing work using to exercise creativity or collaboration in reaching goals in an organic fashion;
• E-Level: The mission, vision and strategic position of the organization, orienting decision-makers so that they can dispense with or change all of the lower-level, more concrete directives, in order to realize the core values of the organization.
Dramatic Structure
• P-Level: Beats, events, scenes – the transactions and interactions of drama;
• A-Level: Acts, chapters, sections, sequences – the ordering and organization of event presentation;
• I-Level: Character – local representations of values, goals and positions that may change as the circumstances of the story change;
• E-Level: Theme – The premise or argument the drama explores through the revelation of values in the interaction between all of its components and structural elements.
Hierarchical causation can be represented graphically as follows (adapted from Jørgensen, 1992, p. 210[3]):
This pattern of hierarchical causation represents lower-level component processes (P), within a dependency structure (A). These enable focal level activity that emerges between them and the downward constraints from the environment. However, patterns of downward constraint are variable, not fixed, and so lower-level ensembles need the flexibility to change to match a variety of circumstances (I). This match is never perfect, however. This means that there always remain fitness strategies yet to be explored, as well as longer-cycle or sporadic environmental changes that impact the risks and opportunities of a given niche. Exploratory action targets these unknowns at the E level.
The conceptual framework of hierarchical causation seems to be a promising model of event structure. It would explain why there are PAE and I parameters for all major tasks. As a result, we should be alert to hierarchical patterns in any of the models we might wish to catalog. Hierarchical causation may indicate that a concern structure may be discernable in such a model.
Hierarchies of Action
Different patterns of activity make sense depending on which level of hierarchical causation we need to operate upon. Take the following task hierarchy:
P – discrete tasks are the smallest microlevel,
A – dependencies between tasks are the next level up,
I – the coordination of people to do the tasks is the smallest macrolevel factor,
E – the landscape of contextual forces, opportunities and threats is the highest level.
The challenge of managing and coordinating tasks differs greatly depending on where they fall in the constraint hierarchy. Nickerson and Zenger (2004[4]) point out that if, if tasks are entirely decomposable into discrete and separate accomplishments (emphasizing P-style, lower-level constraints), there is no need to house them within a single firm or organization at all. A market can be set up, and this would be the most efficient way of coordinating how and where tasks would get done. Some tasks are not entirely decomposable however, but they are partly decomposable. A division of labor can be set up (emphasizing A-style differentiation and integration), and people in each area can work on their component of the overall solution. So long as each set of solutions can be developed without excessive impact on other components, an authority-based hierarchy can be established, with central decision-makers who define the projects, divide up the labor and coordinate the assembly of the solution. Central authorities can direct the search for the best solution.
High-interaction problems do not support this kind of decomposition, however, and they require a different kind of search for solutions. For a complex task like designing a new microprocessor, a broad group of people needs to be assembled, because nearly every aspect of the design has an impact on nearly every other aspect of the design (including engineering, financing, manufacturing, marketing, distribution etc.). A heuristic, E-style search is needed – one that supports high levels of I-style collaboration without too much overt direction – in order to find the best solutions to these kinds of problems. Nickerson and Zenger[4] describe how consensus-based hierarchies are the best way to govern these kinds of information searches within or among the participating firms.
This information-processing view of the firm echoes some fundamental principles of organizational theory – the distinction between specialists and generalists. In stable environments where common tasks repeat themselves, specialists can emerge. As the complexity and rates of change in the environment increase, however, generalists tend to predominate. In concern structure terms, PA concerns are more in the specialist domain, and EI concerns in the generalist domain. PA concerns focus on stable or concrete aspects of the task environment, EI concerns focus on adaptation to the more dynamic and unpredictable aspects of the task environment.
