Tuesday, March 13, 2012

Are mechanisms complex?

Source: D. Little, "Causal explanation in the social sciences" (link)

How can we distinguish between causal mechanisms and extended causal processes? Is the difference merely a pragmatic one, or is there some reason to expect that mechanisms should be compact and unitary in their workings? Is the children's story leading from the "want of the nail" to the loss of the kingdom a description of an extended mechanism or a contingent causal process?

My preferred definition of a social causal mechanism runs along these lines (“Causal Mechanisms in the Social Realm” (link)):
A causal mechanism is (i) a particular configuration of conditions and processes that (ii) always or normally leads from one set of conditions to an outcome (iii) through the properties and powers of the events and entities in the domain of concern. 
This captures the core idea presented in the Machamer-Darden-Craver (MDC) definition of a causal mechanism (link):
Mechanisms are entities and activities organized such that they are productive of regular changes from start or set-up to finish or termination conditions. (3)
There is also an ontological side of the concept of a mechanism -- the idea that there is a substrate that makes the mechanism work. By referring to a nexus between I and O as a "mechanism" we presume that there is some underlying ontology that makes the observed regularity a "necessary" one: given how the world works, the input I brings about events that lead to output O. In evolutionary biology it is the specifics of an ecology conjoined with natural selection. In the social world it is the empirical situation of the actor and the social and natural environment in which he/she acts.

So mechanisms reflect regularities of input and output. In this respect they correspond to pocket-sized social regularities: observed and sometimes theoretically grounded conveyances from one set of circumstances to another set of circumstances.  Take free riding as a mechanism arising within circumstances of collective action:
When a group of individuals confront a potential gain in public goods that can be attained only through effective and non-enforcible collective action, enough individuals will choose to be free riders to ensure the good is not achieved at the level desired by all members of the group.
This states a regularity (conditioned by ceteris paribus clauses): groups of independent individuals are commonly incapable of effective collective action. And it is grounded in a theory of the actor; rational individuals who pay attention to private costs and benefits but not public costs and benefits can be predicted to engage in free riding.

Now consider the mechanism described in social psychology as "stereotype threat" (link):
When subjects are exposed to signs of negative stereotypes of their group with respect to a given kind of performance, the average performance of the group declines.
This is a mechanism that can be identified in a number of different settings, both observational and experimental; and it can be combined with other mechanisms to bring about complex results. The substrate here is a set of hypothetical cognitive structures through which individuals process tasks and influence each other. 

Now consider an instance of concatenation. Suppose we are interested in military mistakes -- weighty decisions that look in hindsight to be surprisingly poor given the facts available to the decision makers at the time. Our theory of the case may involve three separate mechanisms that interfere with good reasoning: stereotype threat, inordinate hierarchicalism, and the effects of agenda setting. These are independent social cognitive mechanisms that impair group decision making. And our theory of the case may attempt to document the workings of each on the eventual outcome and the ways they aggregated to the observed decision. 

Are mechanisms thought to be simple, or can we consider composite mechanisms -- mechanisms composed of two or more simpler mechanisms? Our definition above required that a mechanism should link I to O with a sufficiently high probability to count as "likely". This puts a practical limit on the degree to which simple mechanisms can be composed into composite mechanisms. Take the sequential case: iMj (prob=.90) and jNk (prob=.90). Then let V be the sequential composite mechanism "M then N". Then we have iVk (prob=.81). The probability of the final endstate given the initial starting condition drops with each additional mechanism that we insert into the composite mechanism.  So eventually concatenation will bring the probability of an antecedent leading to a consequence below the threshold of likelihood required by the definition of a mechanism. 

McAdam, Tarrow and Tilly refer to a concatenation of mechanisms in a concrete instance as a process, not a higher-level mechanism.  The reason for this, it would seem, is that processes are highly contingent in their workings precisely because they incorporate multiple mechanisms in series and parallel, all of whose causal properties are probabilistic.  So there is no reason to expect that processes describe reliable associations between beginnings and endings.

So this implies that mechanisms should be conceived at a fairly low level of compositionality: to preserve the likelihood of association between antecedent and consequent, we need to identify fairly proximate mechanisms with predictable effects.  This doesn't mean that a mechanism has little or no internal structure; rather, it implies that the internal structure of a mechanism fits together in such a way as to bring about a strong correlation between cause and effect.  The mechanism of stereotype threat mentioned above presumably corresponds to a complex set of functionings within the human cognitive system. The net effect, however, is a strong correlation between cause (expressing a stereotype about performance to an individual) and effect (suppressing the level of performance of the individual).


  1. Dear Dan,

    thank you for your deliberations. Here my positions on this:

    If I understand you correctly you would say that social processes, which – starting from specific initial conditions X – consist of chains/several steps of causally connected actions or even interactions/communications and lead to specific outcomes Y, are rather no mechanisms. In other words: there are no social mechanisms like (specific) social processes because there are no regularities as reliable associations but only events which are highly contingent. Mechanisms are causal mechanisms and these causal mechanisms are cognitive mechanism, that is individual entities, which are or may be in some way or other social relevant (as event in groups etc.). These cognitive mechanisms then are – in your words – „a strong correlation between cause and effect“.

    Considering the discussion about social mechanism this is an important point. At first glance I see two consequences which seem relevant (not meant as an objection but to try a clarification): First, because many positions – Hedström, Mayntz, George/Bennett, Gross, Demeulenaere, Falleti/Lynch etc. – think of social mechanisms (as regularities) as specific social processes. If you are right this then would be an error (with regard to a clarification I deliberately intensify a little bit). Elias’ „Oscillation of power“, which often is discussed as a social mechanism and just as well Mertons „Self fullfilling prophecy“ then are as social processes no mechanisms. Second, because „normally“ mechanisms are described not as „mere“ correlations, but as a relationship with several links (and often the argument is: to spell out these links). In the words of George/Bennett: mechanisms are processes like „,X leads to Y through (causally connected, R.G.) steps A, B, C’“. And again with regard to social processes – as social entities in the sense outlined above – this would be an illusion.

    If you are right respectively if my conclusions are right I think the discussion about causal respectively social mechanisms ought to run in another direction.

    A short remark to McAdam, Tarrow and Tilly. I interprete there position in „Contention“ about the concept „process“ a little bit different. They describe processes as „frequently recurring causal chaines“ (27) and say that „(m)echanisms and processes form a continuum“ (27). So I think processes for them are not in that way contingent as you assume. Moreover, an indication therefore is their statement „We employ mechanisms and processes as our workhouses of explanation, episodes as our workhouses of description“ (30). Explanation I doubt requires a sort of regularity (the „frequently recurring“).

    So far my deliberations. I’m curious about your answer!

    Rainer Greshoff

  2. Clear, informative, simple. Like your post! Don’t stop writing, you’ve given me lots of good info!

  3. Thanks, Rainer. Several points:

    I've describe a "process" as a concatenation of several independent mechanisms. Because each is probabilistic in effect, after a few links the correlation between starting pont and end point will drop pretty low, so the concatenation as a whole does not satisfy the definition which requires "strong likelihood". I suppose you could imagine a case where two independent mechanisms whose preconditions tend to co-occur, and one of whose effects is an antecedent to the second. In that case I don't object to calling the composite a "mechanism", but this seems uncommon.

    I don't want to say that the only mechanisms we can refer to are individual or "cognitive" mechanisms. They have micro foundations, but they are social complexes. For example, "run on the bank" is a complicated social phenomenon with clear microfoundations.

    It is true that any given mechanism has a complicated set of internal workings. This is true in molecular biology, the adaptation of a population to an environment following a major ecological change, and in sociology when a period of protest comes to an unexpectedly rapid end. Many components contribute to the overall workings of the mechanism as a whole.

    I don't think there is much disagreement between my approach and those of Mayntz and Hedstrom. At the molar level of the mechanism, each requires something analogous to the idea of a regular relationship between input and output. This doesn't mean, however, that explanations take the form of covering-law accounts linking outcomes to general social laws. The regularities associated with mechanisms are "pocket-sized".

    In short, my account of a mechanism holds both things: a mechanism is a specific, concrete set of activities linking together to have a specific effect; and second, that a mechanism needs to be something that recurs in multiple settings with similar results.

    You're right about what McAdam, Tarrow, and Tillly say about processes. But we need to dig more deeply into the question, what social circumstances would create an environment in which similar sequences of mechanisms recur. Take the mechanisms of grievance, spontaneous protest, and escalation. Grievance occurs when a population's interests or sensibilities are offended. Sometimes protest occurs and sometimes it does not. We would like to discover some of the factors that make protest more likely. And once protests occur, we will sometimes witness escalation and sometimes not. The sequence "grievance -> protest -> escalation" sometimes occurs: this is a process. But it may be uncommon, relative to the set of circumstances in which such sequences could in principle have occurred. So I wouldn't call this a mechanism leading from provocation to escalated protest. Each of the mechanisms works independently.

    I suppose there is an generic understanding of "social process" that is different from this specification. Perhaps: "A process is a flow of events and activities, influenced at various points by structures and mechanisms, leading from one social situation to another social situation." For example: "A rapid process of social change occurred in England following the extension of the factory system in the 1820s."

  4. Thank you Dan. I will answer with regard to the background of my comment:

    I became uncertain about the range of the concept "social mechanism". Thus far I thought (and I think) that the core of social mechanisms – in the sense of Hedström, Gross, Mayntz – would be a process of social aggregation which has a certain regularity. So the crucial point in my opinion is: to conceptualize social mechanisms as – starting from initial conditions – aggregation-processes, namely as processes which consist of causally connected chains of interactions/communications etc. by which are produced structural results (= outcomes like for instance norms, rules etc.). To put it another way: the „charm“ or the special claim of the concept "social mechanism" I thought would be to conceptualize a regularity of an aggregation phenomenon that is a regularity of a social process as a “whole”.

    Now for me this idea is questioned. I have difficulties to imagine in which way one is able to conceptualize this regularity. Does in fact exist such regularity? Whereon relies it, where does it come from? Of course the regularity must result from the actions of the relevant actors who are part of the mechanism process and at once produce the process. Or rather: the regularity results from in a special way oriented actions that are oriented by special expectations (which are social valid). For the moment I experimented with the idea that this orientation occurs because the expectations are "triggered" in certain social situations by the interpretations of the actors (interpretations of the situation) and these interpretations occur with a certain likelihood in certain situations and then the thereof resulting expectations orient also with a certain likelihood the follow-ing actions. So roughly is the direction of my idea.

    But perhaps this idea cannot function. Perhaps social processes as a whole can never be grasped as mechanisms with certain regularities. This is the point of Dan. But perhaps some social pro-cesses consist – in the sense of Dan – of several “pocket-sized mechanisms”, which under certain social circumstances are „triggered“ by or rather through the actors (their definitions of the situation). So one might say that a social process XY occurs with a certain likelihood if there is a likelihood that the pocket sized mechanisms a, b, and c are triggered by/through the actors. The pocket-sized mechanisms mentioned a moment ago then had to be conceptualized – that in my opinion is a crucial point – as parts of a social aggregation process.

    But the latter is a different conception compared with that of – for instance – Peter Hedström. Hedström distinguishes between social mechanisms and elementary (individual) mechanisms (Dissecting 26f.). Elementary mechanisms are part of social mechanisms. Social mechanisms he defines in the following way: “A social mechanism, as here defined, describes a constellation of entities and activities that are organized such that they regularly bring about a particular type of outcome. We explain an observed phenomenon by referring to the social mechanism by which such phenomena are regularly brought about“ (Dissecting 25). The outcomes are for instance supraindividual entities like distributions and aggregate patterns, typologies of networks, informal rules or social norms (Dissecting 5f.). They are produced by chains of interactions (Dissect-ing 5f., 26f., 70). The production has a certain regularity. Now my read is that the elementary mechanisms of Hedström correspond the pocket sized mechanisms of Dan. But the claim of the concept “social mechanism” is more comprehensive: to have knowledge of social mechanisms as autonomous social regularities (“autonomous” of course not within the meaning of: independent from certain individual doings, expectations). And then again there is – among others – the question (for Hedström etc., not Dan) whereon relies this autonomy, where does it come from?

    Rainer Greshoff