Levels

I have some thoughts.

Theory
Sociology
Author

andrés castro araújo

Published

January 25, 2024

Metaphors about ‘levels’ are used to describe the world all the time.

We cannot avoid them. Scientists use them all the time. And so rejecting these metaphors outright just because some obscure sociologist made poor use of them is wrongheaded.

Nevertheless, the levels metaphor is also vague and invites confusion.

Problems

Most problems stem from being committed to the following:

These commitments seem innocuous and maybe obvious at times. For example, Andrew Abbott relies on this metaphor to offer an explanation as to why interdisciplinarity seems more straightforward in the natural sciences than in the social sciences.

Political science, we say, is about power, economics about choice, anthropology about ethnography, and so on. In the natural sciences, these axes of cohesion are aligned, to some extent, in a hierarchy of “levels of analysis.” Physics concerns the atomic and subatomic levels, chemistry the molecular level, biology the supermolecular level of living things. But in the social sciences and humanities, axes of cohesion are not aligned… a fact that has made interdisciplinarity in the social sciences much more complicated than the simpler, linear interpenetration of the natural sciences.

Abbott (2001, p. 140)

This sounds insightful, but it is wrong.

Most disciplines cast a wide net. The relationship between scientific disciplines and “ontological levels” is many-to-many. For example, evolutionary biology encompasses genes, organisms, populations, and environments. Take a look at virtually any scientific discipline and you’ll find the same.

The rest of this entry focuses on more legitimate ways of using the levels metaphors in different contexts. I will focus on two kinds of levels that I find useful: (1) levels of description, in which we look at the same phenomena from different perspectives; and (2) levels of organization, in which we look at relationships between “wholes” and their relevant “component parts.”


Note. I will ignore three other commonly used kinds of levels: scale, processing, and disciplinary topic.2

Levels of Description

Levels of description indicate different ways of looking at the same problem or phenomena. Here, the differences between levels are a matter of perspective.

Here, talk of different “levels” usually means we are looking at the same phenomena from different perspectives. As such, these levels don’t raise issues of “emergence” of “higher” levels (or at least they shouldn’t).

These levels are legitimate insofar they make important (non-redundant) contributions to understanding whatever it is we want to understand.

I will focus on two well-established levels of description:

  • David Marr’s hierarchy for the analysis of information-processing systems, in which levels refer to realization (or implementation).

  • The distinction between social structure and culture, in which levels stem from the classic philosophical distinction between form and content.

Realization

Marr’s famous hierarchy of analysis consists of three levels, summarized in Table 1. It is supposed to help us understand complex information-processing “machines.”

  • The computational level.

    What is being computed and why?

    In this framework, the “what” is usually a mathematical calculation (e.g., optimization, prediction); and the “why” is usually determined by looking at the environment in which the information-processing system is embedded.

  • The representation and algorithm level.

    Information-processing systems must represent their inputs and outputs in accordance to some formal scheme.3 This is important because “any particular representation makes certain information explicit at the expense of information that is pushed into the background and may be quite hard to recover” (Marr [1982] 2023, p. 60).

    Results at this level can be established without regard to the physical entities actually instantiating the algorithm—e.g., is the computation tractable (e.g., Kleinberg 2000; Van Rooij 2008)?

  • The physical implementation level.

    An account of physical computation is outside the scope of this entry (see Piccinini 2015).

These are levels of realization. A computation is implemented by an algorithm, which in turn is implemented by some kind of physical system. The analysis at each level can be partially decoupled from the others because of the general notion of multiple realizability—i.e., the same computation can be realized by a variety of algorithms; and the same algorithm can be realized by a variety of physical entities.

The choice of an algorithm is influenced for example, by what it has to do and by the hardware in which it must run. But there is a wide choice available at each level, and the explication of each level involves issues that are rather independent of the other two.

Marr ([1982] 2023, p. 63)

Computational theory Representation and algorithm Hardware implementation
What is the goal of the computation, why is it appropriate, and what is the logic of the strategy by which it can be carried out? How can this computational theory be implemented? In particular, what is the representation for the input and output, and what is the algorithm for the transformation? How can the representation and algorithm be realized physically?
Table 1: The three levels at which any machine carrying out an information-processing task must be understood.

More importantly, the three levels also constrain each other.

Working from the top down, knowing the structure of the environment and the information it makes available to the organism limits the types of information-processing algorithms that can utilize that information. Likewise, having identified a mode of organization and the conditions under which it will generate a form of behavior can guide the search for the components that implement the design. Constraints also arise from the bottom up. Knowing features of the implementation can put constraints on the search for algorithms. Some algorithms might not be implementable, given the components available, and alternatives must be sought. Likewise, knowing the algorithm that seems to be functioning can guide investigations into the environment and reveal different features of its structure that are relevant to the organism.

Bechtel and Shagrir (2015, p. 321)

Structure and Culture

Many sociologists—following in the footsteps of Georg Simmel—like to distinguish between structure and culture (e.g., Rawlings et al. 2023). These perfectly overlapping “levels” stand in opposition to the level of individuals (a distinction that roughly corresponds to a parts-to-whole relationship). But the difference between structure and culture roughly corresponds to a distinction between form and content.

Perspectives that privilege structure have produced many important ideas such as the strength of weak ties, structural holes, and vacancy chains. Perspectives that privilege culture emphasize ideas such as roles, rules, categories, schemas, and legitimacy.

The idea is that the analysis of each level can be partially decoupled from the other, even though each level is just a different way of representing the same phenomena. They are different sides of the same coin. Each level makes up the blind spot of the other.

Some novel forms of network analysis attempt to incorporate both levels simultaneously (e.g., Lizardo 2023a, 2023b).

Levels of Organization

The levels metaphor is most useful when it distinguishes between wholes and their component parts.4

The main idea behind mechanistic explanations is to figure out how the organized activities of parts produce some property or behavior of a larger whole (Machamer et al. 2000; Craver 2001; Craver and Bechtel 2007; Craver and Darden 2013). Here, the levels metaphor relates some activity or feature of a whole to the properties and organized activities of relevant component parts. This is what it means to describe a mechanism: the component parts work together to accomplish something that they cannot do on their own. Mechanisms— by definition—are more than the sum of their parts.

This provides for a view of levels that is very different from the notion of ontological “layers”; “Levels of mechanisms cannot be read off a menu of levels in advance” (Craver 2014, p. 18).

For example, a researcher might be interested in explaining how human groups accomplish complex tasks—e.g., spatial learning and navigation (Hutchins 1995; Vaughan 2021). Or they might be interested in figuring out how status hierarchies enhance the performance of task-oriented groups (Willer 2009). Or they might simply be an organizational sociologist.

Mechanism levels are not spooky

A mechanism approach to levels also guards against “spooky emergentism” or ontologically unmoored macro abstractions—i.e., what happens when sociologists commit to the existence of higher-level phenomena that have no explanation in terms of the organized activities of their component parts. Nothing “emerges” from levels of mechanisms except in the trivial sense that organized parts do things that they cannot do in isolation.

Importantly, mechanistic explanations are not committed to methodological individualism—i.e., the view according to which good explanations in the social sciences must always point towards individuals and their choices.

It is not part of our view that all explanations must bottom out in some privileged set of fundamental entities and activities…

Nor is it part of our view that all mechanistic explanations must go down in a hierarchy of mechanisms to be explanatory. In order to explain why a given strand of DNA is on a particular island in the Galapagos, one would do better to investigate the migratory patterns of birds and the selective effects of a recent drought, rather than the molecular components of its DNA. One must, in such cases, ascend from the part to the whole to understand the relevant mechanisms. If one wants to understand why a cyclist’s cellular glucose metabolism suddenly accelerated, the right answer might well be that the cyclist has started the climb.

Craver and Darden (2013, p. 25)

This is compatible with the view of multi-level explanation in sociology.

…explanations at one level may refer to causal processes at another. If a soldier shoots another soldier, it may have nothing to do with their individual properties or relations: it may simply be that two armies are at war, and thus it might be pointless to study causality as arising from the personalities involved. On the other hand, causal explanation of a societal process—say, a decline in a birth rate—may properly invoke little more than changes in the individuals involved (e.g., a rise in education produces women who prefer to have fewer children). The point here is that the content of an explanation must be worked out empirically, considering possible connections at multiple levels of analysis, sorting out which connections are present and substantial, and which are not.

Jepperson and Meyer (2021, p. 199)

References

Abbott, Andrew. 2001. Chaos of Disciplines. University of Chicago Press.
Bechtel, William, and Oron Shagrir. 2015. The Non-Redundant Contributions of Marr’s Three Levels of Analysis for Explaining Information-Processing Mechanisms.” Topics in Cognitive Science 7(2): 312–22.
Craver, Carl F. 2001. Role Functions, Mechanisms, and Hierarchy.” Philosophy of Science 68(1): 53–74.
———. 2014. Levels.” in Open MIND, edited by T. Metzinger and J. M. Windt. Frankfurt am Main: MIND Group.
Craver, Carl F., and William Bechtel. 2007. Top-down Causation Without Top-down Causes.” Biology & Philosophy 22(4): 547–63.
Craver, Carl F., and Lindley Darden. 2013. In Search of Mechanisms: Discoveries Across the Life Sciences. University of Chicago Press.
Hutchins, Edwin. 1995. Cognition in the Wild. MIT press.
Jepperson, Ronald L., and John W. Meyer. 2021. Institutional Theory: The Cultural Construction of Organizations, States, and Identities. Cambridge University Press.
Kleinberg, Jon. 2000. “The Small-World Phenomenon: An Algorithmic Perspective.” in Proceedings of the 32nd Annual ACM Symposium on Theory of Computing.
Lizardo, Omar. 2023a. An Analytical Approach to Culture.” Philosophy of the Social Sciences 00483931231169313.
———. 2023b. “Culture and Networks.” in The Sage Handbook of Social Network Analysis, edited by J. McLevey, J. Scott, and P. J. Carrington. 2nd edition. 2nd edition. SAGE Publications Ltd.
Machamer, Peter, Lindley Darden, and Carl F. Craver. 2000. Thinking about Mechanisms.” Philosophy of Science 67(1): 1–25.
March, James G., and Herbert A. Simon. 1993. Organizations. John wiley & sons.
Marr, David. [1982] 2023. “Vision.” in Mind Design III: Philosophy, Psychology, and Artificial Intelligence, edited by J. Haugeland, C. F. Craver, and C. Klein. The MIT Press.
Piccinini, Gualtiero. 2015. Physical Computation: A Mechanistic Account. Oxford University Press.
Rawlings, Craig M., Jeffrey A. Smith, James Moody, and Daniel A. McFarland. 2023. Network Analysis: Integrating Social Network Theory, Method, and Application with R. Cambridge: Cambridge University Press.
Van Rooij, Iris. 2008. The Tractable Cognition Thesis.” Cognitive Science 32(6): 939–84.
Vaughan, Diane. 2021. Dead Reckoning: Air Traffic Control, System Effects, and Risk. University of Chicago Press.
Willer, Robb. 2009. Groups Reward Individual Sacrifice: The Status Solution to the Collective Action Problem.” American Sociological Review 74(1): 23–43.

Footnotes

  1. Some social scientists find this “ontological stratification” of the world very attractive because they want to protect themselves against incursions from nearby disciplines—e.g., psychologists protecting themselves against the neuroscientists, sociologists protecting themselves against the psychologists economists. They further complicate matters by developing some goofy view about the autonomy or irreducibility of their preferred level.↩︎

  2. I will also ignore the related (and important) notion of units of analysis—i.e., levels are individuated (made distinct) by units of measurement (e.g., individuals, organizations, groups, populations).↩︎

  3. “To say that something is a formal scheme means only that it is a set of symbols with rules for putting them together—no more and no less” (Marr [1982] 2023, p. 60).↩︎

  4. “Levels of parts and wholes must also be correlated with size differences because parts can be no larger than the wholes they compose. But the size differences between levels of parts and wholes are an accidental consequence of the part-whole relationship itself, not part of defining what it is for things to be at different part-whole levels” (Craver 2014, p. 12).↩︎