Hedstrom Swedberg as 1996

Social Mechanisms

Peter Hedström and Richard SwedbergStockholm University

In this article it is argued that the search for ’social mechanisms’ is ofcrucial importance for the development of sociological theory. With thisconcept - which is occasionally used in the sociological literature buthas received little systematic attention - attention is called to anintermediary level of analysis in-between pure description and story-telling, on the one hand, and universal social laws, on the other. Whilethe search for universal laws and grand theory has a great deal ofappeal, we do not believe that this type of theorizing is likely to fosterthe development of a useful body of explanatory theory. Drawing on theheritage of Robert Merton and James Coleman, it is argued that theessential aim of sociological theorizing should be to develop fine-grainedmiddle-range theories that clearly explicate the social mechanisms thatproduce observed relationships between explanans and explanandum.We provide a tentative typology of social mechanisms, and we illustrateour argument by showing that three well-known theories in sociology -the self-fulfilling prophecy (Robert Merton), network diffusion (JamesColeman), and threshold-based behavior (Mark Granovetter) - all arefounded upon the same social mechanism.

Peter Hedstrom and Richard Swedberg, Department of Sociology,Stockholm University, S-10691 Stockholm, Sweden© Scandinavian Sociological Association 1996

1. IntroductionEven though the concept of ’social mechanisms’ is occasionally used in thesociological literature, it has received little systematic attention. Our mainambition with this article is therefore to examine more closely the notion ofmechanisms, in an attempt to evaluate its potential role in explanatorysociological theory. The main message of the article is that further advancesof sociological theory call for explanations that systematically seek to

explicate the generative mechanisms that produce observed associationsbetween events.

It might appear obvious that every sociological theory, worthy of itsname, should be explanatory. But upon closer examination, it turns out thatwhat often goes under the rubric of general sociological theory, should moreproperly be viewed as conceptual or sensitizing schemes, and not as

explanations proper. Much of Anthony Giddens’s work, for example,

282

exemplifies this tendency (e.g. Turner 1986; cf. Giddens 1987). And accordingto Jeffrey Alexander, writing in Handbook of Sociology, sociology should payless attention to ’explanation’ and more to ’discourse’ (Alexander1988:78-81). In an insightful article by someone who has devoted most ofhis academic career to general social theory, G6ran Therborn (1991:178)notes: ’Absent in or marginal to currently prevailing general sociologicaltheorizing is any ambition to explain. Little interest can be found in

contributing to answering questions like: Why do these people act in thisway? Why does that social order change in that way?’ The type ofmechanism-based theorizing advocated here focuses exactly on these typesof why-questions.

The focus on mechanisms advocated here should not be confused with apurely descriptive sociology that seeks to account for the unique chain ofevents that led from one situation or event to another. All properexplanations explain the particular by the general, and as will bedemonstrated below, there are general types of mechanisms, found in arange of different social settings, that operate according to the same logicalprinciples. Our vision of an explanatory sociology, contains an ensemble ofsuch fundamental mechanisms that can be used for explanatory purposes ina wide range of social situations.

Given the present, practically non-existent state of serious discussion ofthe notion of mechanism in sociology, we approach our task in a somewhatroundabout way.’ First, we briefly relate Robert Merton’s discussion of therole of explanatory mechanisms in middle-range theorizing, and thereafterwe consider two recent discussions of explanatory mechanisms and their rolein the social sciences, by Jon Elster and Arthur Stinchcombe. This is followedby brief discussions of a number of topics that we believe need to be clarifiedand elaborated. These include the interdisciplinary nature of the concept ofmechanism; the role of mechanisms in sociological explanations; and acritique of a variable-centered type of theorizing that is inherent in theapproach. Thereafter we employ the work of Merton (1968) on the self-fulfilling prophecy, of Coleman et al. (1966) on the diffusion of a new drug,and of Granovetter (1978) on collective behavior, in order to illustrate ournotion of a general social mechanism, namely that the same mechanism oftenunderlies different phenomena and different sociological theories. The articleends with a brief typology of such social mechanisms.

2. The use of mechanisms in sociologyAmong the classics the term ’mechanism’ is rarely used,2 even if the ideaoften is present.3 An explicit use of the concept of ’mechanism’ does not seemto have appeared in sociology until after World War II. The most suggestivediscussion of the concept is in our opinion to be found in the writings ofRobert Merton, who brought together the idea of mechanism with that ofmiddle-range theorizing (Merton 1967). Merton firmly rejected all attemptsto develop general systems of sociological theory and instead advocated thatsociological theory should deal with ’social mechanisms’.4 The point was tolocate a middle ground between social laws and description, Merton said, and’mechanisms’ constitute such a middle ground.

283

Merton defined social mechanisms as ’social processes having designatedconsequences for designated parts of the social structure’, and argued that itwas the main task of sociology to ’identify’ mechanisms and to establishunder which conditions they ’come into being’, ’fail to operate’ and so on(Merton 1968:43-44). Merton briefly discussed concrete mechanisms thatdetermine reference groups, create dissonance, and articulate role-sets,5 butthe most important contribution of his essay, in our opinion, was his view ofmechanisms as elementary building blocks of middle-range theories.6

During the last few decades the emphasis on mechanisms that hadplayed such a vital role in Columbia Sociology has largely been pushed asideby a resurgence of general social theories, on the one hand, and of a variable-centered mode of theorizing, on the other. A few successful attempts todevelop sociological theory founded upon the notion of explanatory mechan-isms can be found in the literature, but these represent the exception ratherthan the rule. Before discussing these efforts, however, we will outline part ofthe recent meta-theoretical discussion on the role of explanatory mechanismsin social theory. This discussion has been dominated by Jon Elster, thoughArthur Stinchcombe has also made an important contribution.7

7

3. Elster and Stinchcombe on the use of mechanisms in social scienceJon Elster’s work has for some time been infused with a strong sense that itis imperative to use the notion of mechanism in the social sciences. Throughmechanisms, Elster argues, it will be possible to bypass the old oppositionbetween laws and description in the social sciences and to focus research onan intermediary and much more flexible level. A view that Elster attributesto French historian Paul Veyne is probably also held by himself: ’progress inthe social sciences consists of knowledge of ever-more mechanisms ratherthan ever-better theories’ (Elster 1989:173).

In his book Nuts and Bolts for the Social Sciences (1989), Elster gives themost complete account of what he means by this concept. He enumerates fivethings that a mechanism is not, and when taken together, these can be saidto add up to a negative definition of the concept. They are the following:~ First, causal explanations must be distinguished from true causal

statements. To cite the cause is not enough: the causal mechanism mustalso be provided, or at least suggested;

~ Second, causal explanations must be distinguished from assertions aboutcorrelations;

~ Third, causal explanations must be distinguished from assertions aboutnecessitation;

~ Fourth, causal explanations must be distinguished from story-telling; and. Finally, causal explanations must be distinguished from predictions.Sometimes we can explain without being able to predict, and sometimespredict without being able to explain. (Elster 1989:4-8)

In Nuts and Bolts, as well as in other works, Elster gives a number ofconcrete examples of mechanisms, from the natural and the social sciences.As an example of a ’psychological mechanism’, he mentions the example of

284

’sour grapes’ or the fact that one often ceases to desire what one cannot get.There also exists an opposite tendency: ’forbidden fruits’, which refers to asituation where one desires what one cannot get. Two other such opposing’mechanisms’, mentioned by Elster, are ’the bandwagon effect’ and ’theunderdog effect’ (Elster 1993:13-15).~ 8

That Elster’s work on mechanisms is still at an early and evolving stageis, however, also clear. One indication of this is that he has, up until now,failed to give a formal definition of what constitutes a mechanism (seeespecially Elster 1989:3-10; 1993:2-3).9 Another is that he has changed hisopinion of what characterizes a mechanism in general. While in Nuts andBolts Elster says that mechanisms imply ’explanations of ever finer grain’, ina later work he maintains that mechanisms, as opposed to laws, only havelimited generality (Elster 1989:7; cf. Elster 1991:7-8).

Like Elster, Stinchcombe argues that ’we do not have a sufficientlysupple armory of mechanisms for making social science theory’, and the mainpurpose of his contribution to the current discussion of mechanisms is to helpremedy this situation (Stinchcombe 1993:24). It is important to note,however, that Stinchcombe advocates a much more restrictive definition ofmechanisms than Elster:

Mechanisms in a theory are defined here as bits of theory about entities at adifferent level (e.g. individuals) than the main entities being theorized about(e.g. groups), which serve to make the higher-level theory more supple, moreaccurate, or more general. (Stinchcombe 1991:367)

Examples that Stinchcombe uses to illustrate his definition include

maximizing individuals (on the lower level) who create a market throughtheir actions (on the higher level); and molecules (on the lower level) thatunder certain conditions turn into gas (on the higher level).

Unlike Elster, Stinchcombe seems mostly concerned with establishingwhat constitutes a purely sociological mechanism. While in economics, aswell as in much of sociology, the analysis often starts with individuals andends up with outcomes on a group level, there also exists another possibilityfor sociologists, according to Stinchcombe. This is to view cz situation as amechanism (’situational mechanism’; Stinchcombe 1993:28-30). Stinchcomberefers at this point of his argument to Erving Goffman’s definition of a’situation’, where the general idea is that people monitor one another inpublic places and adjust their behavior to the situation at large.10

The works of Elster and Stinchcombe raise a number of questions thatneed to be examined more closely in order to arrive at a useful and

sufficiently precise definition of a ’social mechanism’. One such question, tobe discussed in the next section, is how the notion of mechanisms has beenused in other sciences, from physics to economics. We can obviously onlyprovide a first and incomplete sketch of what is a huge and difficult topic.Nonetheless, one interesting aspect of the notion of mechanisms is exactly itsinterdisciplinary nature and the prospect that sociology might be able toincorporate certain ideas about or types of mechanisms from the othersciences.

285

4. The concept of mechanism as used in other sciencesIn physics the word ’mechanism’ is rarely used. One important reason forthis is of a historical or accidental nature and has to do with the fact that inphysics the word ’mechanism’ is connected to the scientific world view of the17th century (e.g. Dijksterhuis 1986). It should also be remembered that inthe 19th century thermodynamics popularized the notion of a system, whichis broader than that of ’mechanism/machine’ and allows the analyst to choosethe environment of the system according to the purpose of the study. Theattempt to conceptualize all phenomena according to the elementary laws ofmechanics became impossible after the emergence of field physics in themiddle of the 19th century. From then on, the word ’mechanism’ has beenused sparingly in modern physics.

The 17th century notion of mechanism spread from physics andastronomy to a number of sciences - such as chemistry and biology -where the term ’mechanism’ is still used, though with different meanings.The Cartesian notion that organisms can be conceptualized as machinesturned out to be very useful and became central to a new biologicalphilosophy called ’mechanism’, which is usually contrasted to that of’vitalism’ or the doctrine that life cannot be reduced to mechanics (e.g.Beckner 1967). In the 19th century the term mechanism was disconnectedfrom the metaphor of the machine1 and instead became linked to that of thesystem.

12

The concept of ’law’ is only used occasionally in the biological sciences -typically in connection with genetics and on the understanding that they arenever without exceptions. According to Francis Crick, who shared the NobelPrize in 1962 for his discovery of the molecular structure of DNA, 20thcentury biologists prefer to think in terms of’mechanisms’ and not ’laws’. Thereason for this is that the notion of ’laws’ is generally reserved for physics,which is the only science that can produce explanations based upon powerfuland often counterintuitive laws with no significant exceptions. ‘What isfound in biology is mechanisms, mechanisms built with chemical componentsand that are often modified by other, later, mechanisms added to the earlierones’ (Crick 1989:138).

In the social sciences, the prevalence of explicitly stated mechanism-based explanations varies widely between the disciplines. These types ofexplanations are rarely used in history, sometimes in sociology, and quitefrequently in economics and psychology. Particularly in cognitive psychology,the notion of mechanism plays a key role. The basic approach here is one of’information-processing’, where the cognitive process is divided up intodifferent stages with special ’mechanisms’. To cite a well-known work, ’Theinformation-processing approach [in cognitive psychology] assumes that

perception and learning can be analyzed into a series of stages during whichparticular components (’mechanisms’) perform certain transformations orrecoding of the information coming into them’ (Bower 1975:33).

Economists often see themselves as thinking in terms of mechanisms, asopposed to sociologists and historians, who are believed to be more interestedin institutions (e.g., Schumpeter 1989:293). The one mechanism thateconomists relate most of their analyses to - their master mechanism, so-to-speak - is clearly the market, conceptualized according to marginal utility

286

theory. That the market can be seen as a ’mechanism’ goes back to the 18thcentury, when economics (via e.g. Adam Smith) became influenced by theNewtonian-Cartesian worldview; and it has become so self-evident to

contemporary economists that the market is a mechanism, that they oftenuse the terms ’market’ and ’market mechanism’ synonymously.

The basic approach in economics, to repeat, is to analyze a phenomenonby relating it in some manner to the market mechanism. Much ofneoclassical economics in the 20th century can be understood as an attemptto explain ever more aspects of the economic process through the mechanismof the market: production as well as consumption and distribution.Economists’ talent for thinking in terms of mechanisms, however, oftenbecomes clear to non-economists only when they go beyond the traditionalboundaries of economics. Examples of this can be found in Hirschman’s Exitand Voice (1970)13 as well as in much of Thomas Schelling’s work (e.g. 1978).

As economists gradually have expanded the boundaries of their

discipline to include a range of topics traditionally considered the domainof sociologists, such as the family and organizations, the difference betweenthe disciplines to an increasing extent have come to concern the types oftheories being used. One such difference, but by no means the only one,centers exactly on the importance attributed to explanatory mechanisms.Comparing labor market sociology with labor economics, Aage Sorensen hasnoted that most labor market sociologists think of theory

as having to do with which variables should be included in the equations andhow these variables relate to other variables - and not as something which isabout which mechanisms produce the observed associations in the variables.This is where there is a huge difference between sociological research andeconomic research in this area; and the difference is very much to thedisadvantage of the sociologist. (Sorensen 1990:308)

5. The explanatory importance of social mechanismsThe core argument of this article is that the identification and analysis ofsocial mechanisms is of great importance for the progress of the sociologicalenterprise. But what exactly is a mechanism and why should we focus onmechanisms rather than on statistical associations or other forms of

relationships between the entities of interest?It is far from trivial to provide a precise, yet sufficiently general

definition of a social mechanism that captures the essence of the concept. Assuggested by philosopher of science Rom Harr6 (1970), one key definingcharacteristic of an explanatory mechanism is the function it performs in anexplanatory account. When we have observed a systematic relationshipbetween two entities, say I and 0, in order to explain the relationshipbetween them we search for a mechanism, M, which is such that on theoccurrence of the cause or input, I, it generates the effect or outcome, O. Thesearch for mechanisms means that we are not satisfied with merelyestablishing systematic covariation between variables or events; a satisfac-tory explanation requires that we are also able to specify the social ’cogs andwheels’ (Elster 1989:3) that have brought the relationship into existence.

287

This generative view of causal explanations differs in important respectsfrom the classic covering-law model as advocated by Carl Hempel and hisfollowers (see Hempel 1942, 1962). According to Hempel, an explanation ofan event entails subsuming the event under a general law. A satisfactoryexplanation therefore must specify the general covering law and theconditions which make the law applicable in the specific case. 14 Accordingto Hempel, deterministic laws are quite unlikely in the social and thehistorical sciences. The general ’laws’ that can be invoked in the socialsciences are instead of a probabilistic nature, i.e., they state that theoccurrence of a particular event will come about with such and suchprobability if certain specified conditions are at hand.

Since this form of explanation simply entails applying a general law to aspecific situation, the insights offered by the exercise are directly propor-tional to the depth and robustness of the ’probabilistic law’. If this law is onlya statistical association, which is the norm in the social and historicalsciences according to Hempel, the specific explanation will offer no moreinsights than the law itself and will usually only suggest that a relationshipis likely to exist, but it will give no clue whatsoever as to why this is likely tobe the case. Covering-law explanations in the social sciences thereforenormally are ’black-box’ explanations and they do not attempt to reveal anymechanisms that might have generated the observed relationships. We areinclined to agree with von Wright’s position that it is better ’not to say thatthe inductive-probabilistic model [of Hempel] explains what happens, but tosay only that it justifies certain expectations and predictions’ (von Wright1971:14).15

The main reason for advocating explanations that directly refer togenerative mechanisms is, in our opinion, that they provide (or encourage)deeper, more direct, and more fine-grained explanations. The search forgenerative mechanisms consequently helps us distinguish between genuinecausality and coincidental association, and it increases the understanding ofwhy we observe what we observe.

The role that the search for mechanisms plays in distinguishing betweenspurious and real associations can be illustrated by the recent controversysurrounding possible health effects of electromagnetic fields. Some epide-miological studies have found an empirical association between exposure toelectromagnetic fields and childhood leukemia (see Feychting & Ahlbom1993). However, the weight of these empirical results is severely reduced bythe fact there exists no known biological mechanism that can explain howlow-frequency magnetic fields could possibly induce cancer (ORAU 1992).According to Bennett (1994), it is furthermore extremely unlikely that amechanism will ever be found, because such a mechanism would have toviolate well-established physical principles. The lack of a plausible mechan-ism increases the likelihood that the weak and rather unsystematicempirical evidence reported in this epidemiological literature, simply reflectsunmeasured confounding factors rather than a genuine causal relationship(Hedstr6m 1994a).

The distinction between black-box explanations and explanations thatrefer to explicit and generative mechanisms can be illustrated in moregeneral terms with the following example that is adopted from the work of

288

philosopher of science and physicist Mario Bunge (1967). Assume that wehave observed a systematic (non-random) relationship between two types ofevents or variables, I and O. The way in which the two sets of events orvariables are linked to one another is expressed with the mechanism, M:

What characterizes a black-box explanation is that the link between inputand output, or between explanans and explanandum, is assumed to be devoidof structure or, at least, whatever structure there may be, is considered to beof no inherent interest (perhaps because it cannot be observed). In a black-box explanation, Bunge maintains, the ’mechanism’ linking input and outputis a purely syntactical link between a column of values for the input, I, and acolumn of values for the output, O.

In sociology, the most systematized form of black-box explanation can befound in the so-called causal modeling approach (see Duncan 1975), whichwill be discussed more fully in the next section. In the causal-modelingtradition, the explanatory ’mechanism’ simply is a regression coefficientlinking I and 0, and this regression coefficient (if the model includes allrelevant variables) is supposed to describe the causal influence of I upon O.

One important difference between black-box explanations and the typeof explanation advocated here concerns the information content of theproposed mechanism. Since the alleged ’mechanism’ in a black-box explana-tion (e.g. a regression coefficient) exclusively is derived from I and 0, itcontains no information that is not already contained in the events orvariables themselves. The approach advocated here does not rest withdescribing the form of the relationship between the entities of interest butaddresses a further and deeper problem: how, i.e. through what process, wasthe relationship actually brought about? 16

Consider the example of poisoning (Bunge 1967). It would be possible toestimate the parameters of an equation describing the relationship betweenthe intake of, say, strychnine and the risk of dying. If the model had thecorrect functional form, we might even have established a ’covering law’ ofthe dose-response relationship which could be used in subsequent explana-tions of other occurrences of strychnine intake. But as long as we have notspecified the mechanisms that link strychnine intake to morbidity andmortality, the explanation is clearly wanting. By pointing to how strychnineinhibits the respiratory centers of the brain and to the biochemical processesresponsible for this paralysis, we provide a mechanism that allows us notonly to describe what is likely to happen, but also to explain why it is likely tohappen.

A comparison of behaviorist and rational-choice accounts of individualbehavior illustrates the same basic point. Explanations in the behavioristtradition, such as that represented in sociology by the work of George Homans,resemble what we have referred to above as black-box explanations; their aimis to arrive at general propositions relating stimuli to response withoutinvoking any (unobserved) generative mechanism that might have broughtabout the relationship. Homans’s so-called ’success proposition’ is a represen-tative case in point: ’For all actions taken by persons, the more often a

289

particular action of a person is rewarded, the more likely the person is toperform the action’ (Homans 1974:16). Rational-choice theory does not restwith observing and describing the form of the relationship between reward andbehavior, but tries to explain precisely those associations which a behavioristlike Homans is content with only establishing. The rational-choice explanationpostulates the existence of a more supple generative mechanism built uponhypothetical constructs such as preferences and intentions (see Hedstrom,forthcoming). By spelling out a detailed mechanism linking individualbehavior and rewards, the rational-choice theory provides a deeper and morefine-grained explanation than behaviorist theory.

One line of sociological research that illustrates the shortcomings of black-box explanations is research on class and its individual correlates. In

empirically oriented sociology, individuals’ class-belonging has become apopular explanation for various individual-level phenomena such as income(e.g. Kalleberg & Berg 1987) and health (e.g. Townsend & Davidson 1982). Theconcept of class might be useful for descriptive purposes where it serves as ashorthand for various aspects of individuals’ socio-economic living conditions,and research in these traditions has produced informative empirical researchdescribing the living conditions of different ’classes’. Whether the empiricalexercise of relating variables describing class and income or class and healthalso has an explanatory value - in the deeper sense of saying something aboutwhy we observe what we observe - is much more doubtful.

Despite the common sociological rhetoric of describing class as a

’determinant’ of various individual traits and behaviors, class in and ofitself obviously cannot influence an individual’s income or health. A ’class’cannot be a causal agent because it is nothing but a constructed aggregationof occupational titles. A statistical association between ’class’ and income, or’class’ and health, tells us that individuals from certain ’classes’ have lowerincomes or worse health than others, but it says nothing about why this isthe case. To answer such questions it is necessary to introduce and explicatethe generative mechanisms that might have produced the observeddifferences in average income or health between the occupational groupsthat the researchers have assigned to different ’classes’. A statistical ’effect’of a class-variable in contexts like these is essentially an indicator of ourinability properly to specify the underlying explanatory mechanisms. Theworse we do in specifying and incorporating the actual generativemechanisms into the statistical model, the stronger the ’effect’ of the class-variable will appear to be.

It is important to note that both the biochemical mechanism and therational-choice mechanism referred to above are mechanisms of some

generality, and it is this generality that gives them their explanatorypower. Simply making up an ad hoc story tailored to a specific case does notconstitute an acceptable sociological explanation. Even moderately talentedjournalists are able to make up these sorts of ad hoc stories, and, as ArthurStinchcombe once noted, ’a student [of sociology] who has difficulty thinkingof at least three sensible explanations for any correlation that he is reallyinterested in should probably choose another profession’ (Stinchcombe1968:13). Serious, non-commonsensical, sociological explanations requiremechanisms of some generality.

290

Generative explanations usually invoke some form of ’causal agent’(Bhaskar 1978) or ’causal power’ (Harre 1970), which is assumed to havegenerated the relationship between the entities being analyzed. It is byexplicitly referring to these causal agents that the relationship is madeintelligible. In the natural sciences, causal agents come in a variety of formssuch as organic reactions in chemistry and natural selection in biology. Insociology, however, the elementary ’causal agents’ are always individualactors, and intelligible social mechanisms should, in our opinion, alwaysinclude explicit references to the causes and consequences of their actions. 17This principle of methodological individualism is motivated by the same coreidea as is advocated throughout this article: understanding is obtained orenhanced by making explicit the underlying generative mechanisms thatlink one state or event to another, and in the social sciences actionsconstitute this link.

It is important to note that explanatory mechanisms, in the natural aswell as in the social sciences, often are unobserved or only observable in theireffects. Weinberg (1993) emphasizes the important role that unobservedentities have played in physics. For example, the existence of both theelectron and the neutrino was conjectured and their roles in various physicalprocesses were usefully theorized, long before they actually were observed.Similarly, the social sciences routinely postulate the existence of unobservedexplanatory mechanisms. Assumptions of intentions, discounting, andpreferences have proven extremely useful for the analysis of individualaction even though they never can be observed. Mechanisms thus aretheoretical constructs that provide hypothetical links between observableevents. In many situations the operation of a postulated mechanism can onlybe tested by logically deriving the effects that should be observed if themechanism was operating as assumed in the theory, and then comparingthese theoretical expectations with what actually is being observed.

Elster (1989) appears to have a slightly different view of the role ofunobserved mechanisms in scientific explanations. He argues that mechan-isms should provide an account of what happened as it actually happened,and not as it might have happened. No one would dispute the correctness ofthis proposition if it were possible to realize, but accounting for something ’asit actually happens’ is always problematic. As argued by Bunge (1967),simply describing all the events, microscopic and macroscopic, that takeplace in a room during one second would - if it were technically possible -take centuries. Thus, when describing ’what happened’, we can never cite allevents immediately preceding the event to be explained (not to mention moredistant events), but we must be highly selective in what events we chose toinclude in our description of what ’actually happened’ (see also Weber 1949).Hence, mechanisms as well as general theories can refer only to a subset ofthe potentially important events, and they can therefore only give plausibleaccounts of what happened as it could have happened.&dquo;3

The belief in explanations that provide accounts of what happens as itactually happens has pervaded the sociological literature for decades and hasproduced an abundance of detailed descriptive narratives, but few explana-tory mechanisms of any generality. It is through abstractions and analyticalaccentuation, however, that general mechanisms are made visible. But these

291

abstractions also distort by their nature the descriptive account of whatactually happened, by accentuating certain aspects of the situation andignoring others. Francis Crick’s characterization of the process throughwhich good biological theories are arrived at is, in our opinion, equally validfor the social sciences: ’To produce a really good biological theory one musttry to see through the clutter produced by evolution to the basic mechanismslying beneath them’ (Crick 1989:138; see also Van Parijs 1981). And as RomHarr6 has argued, making hypothetical models of unknown mechanisms isusually what moves a scientific field forward:

Generally speaking, making models for unknown mechanisms is the creativeprocess in science, by which potential advances are initiated, while the makingof models of known things and processes has, generally speaking, a moreheuristic value. (Harré 1970:40)

6. Variables versus social mechanisms .

The widespread use and knowledge of survey analysis and the statisticaltechniques needed for analyzing such data have clearly improved the abilityof sociologists to describe social conditions and to test sociological theories.But the increasing use of these techniques has also fostered the developmentof a variable-centered type of theorizing that pays only scant attention toexplanatory mechanisms. Coleman (1986) has aptly described this type ofsociology as a form of ’individualistic behaviorism’. The guiding principlebehind this type of theorizing - usually referred to as ’causal-modeling’ - isthe notion that individual behavior can and should be explained by variousindividual and environmental ’determinants’, and the purpose of the analysisis to estimate the causal influence of the various variables representing thesedeterminants.19

According to Coleman, this emphasis on ’causal’ explanations of behaviorrepresented a considerable change from the type of explanatory account usedin the earlier tradition of community studies: ’One way of describing thischange is to say that statistical association between variables has largelyreplaced meaningful connection between events as the basic tool of

description and analysis’ (Coleman 1986:1327-1328). In the causal-modelingtradition, variables and not actors do the acting (Abbott 1992).20

The tension between a variable-centered causal approach to sociologicaltheorizing and a generative view emphasizing the importance of socialmechanisms came to the fore in an exchange between Robert Hauser andRaymond Boudon in the mid-1970s. The context of this exchange was areview by Hauser of Boudon’s (1974) book on education and inequality. Inthis book Boudon developed a theoretical model which he hoped would makeintelligible a number of apparent paradoxes reported by empirical researchon social mobility.21 Hauser suggested numerous changes to Boudon’s model,but the main message of his article was a strong disbelief in the very ideathat had motivated Boudon to write the book, i.e., that an importantdistinction should be made between statistical and theoretical models, andthat theoretical models are needed to explain the results of an empiricalanalysis:

292

Boudon dismisses several standard representations of the mobility process asbeing ’basically statistical’. I can only guess what this means - perhaps thatthey are rich in formal properties or that sampling distributions of theirparameters are known. Neither of these properties strikes me as undesirable,and these models do have coherent and intuitively meaningful interpretationsrelative to the mobility process. (Hauser 1976:923)

Boudon responded by noting that descriptive models of the sort advocated byHauser are undoubtedly useful for many purposes, but that their usefulnessfor causal analysis is considerably more restricted than assumed by Hauser.According to Boudon, understanding normally is achieved not by the meansof descriptive models such as path-models, but through theoretical modelswhich show the logic of the process being analyzed. In order to understandthis logic, Boudon argued ’we must go beyond the statistical relationships toexplore the generative mechanism responsible for them’ (Boudon 1976:117).

One way of describing the difference between Boudon and Hauser isthrough the distinction between descriptive and structural models. Hauser,being one of the leading figures in the causal-modeling tradition, had astrong commitment to the idea that it is possible to estimate true structuralmodels on the basis of non-experimental data, i.e., models expressinginvariant causal relationships between different sociological and demo-graphic variables. Boudon, on the other hand, viewed regression models aspurely descriptive and therefore in need of explanation. As Boudon expressedit in a different context,

causal analysis does not explain the [statistical] chart. It simply summarizes it.Understanding a statistical structure means in many cases building a

generating theory or model ... that includes the observed empirical structureas one of its consequences. (Boudon 1979:51-52)

The position taken by Hauser and others in the causal modeling traditionhas also been challenged by sociologists such as Mark Granovetter, who hasargued that these sorts of correlational analyses can be useful in ruling outtheories and arguments previously taken seriously, but that they arewanting because they cannot tell ’by which mechanisms these correlationshave their effects, or what broader historical and socio-economic forces haveset these mechanisms in motion’ (Granovetter 1982:259-260).

The statistician David Freedman (1991, 1992a, 1992b) has discussed thestatistical foundations of the causal modeling approach in some detail.According to Freedman, the belief of some social scientists in the possibilityof estimating true structural models is both naive and counterproductive.The basic tools of causal modelers such as path analysis and its latentvariable counterparts, Freedman argues, are based upon a network of highlyrestrictive stochastic assumptions that rarely, if ever, are fulfilled. The basicthrust of Freedman’s argument is that social scientists need to think moreabout underlying social processes, and they also need to look more closely atdata without the distorting lens of conventional and, according to Freedman,largely irrelevant stochastic models: ’In my opinion, the confusion betweendescriptive and structural models pervades the social-science scholarlyliterature of the past 20 years, and has distorted the research agenda of ageneration’ (Freedman 1992b:123). According to Freedman, investing even

293

more time and intellectual energy in trying to estimate non-existingparameters would not be a particularly worthwhile activity. At best, thesetypes of statistical models can provide compact descriptive summaries of thedata - but they cannot in themselves provide causal explanations.22

The epidemiological research tradition has produced some importantand duly celebrated examples of non-experimental empirical researchleading to the identification of genuine causal mechanisms. Snow’s researchwhich suggested that cholera was a water-borne disease, and Doll’s researchon the relationship between smoking and lung cancer are perhaps the mostnotable examples (see Doll & Peto 1981; Hamlin 1990). But these successstories appear to be the exceptions that probe the rule (and Freedman nodoubt would attribute part of their success to the fact that they did not usestructural equation models). Despite some 30 years of causal modelingcarried out by some of the most gifted sociologists of this generation, nosimilar findings have been reported in the sociological literature. Given thispaucity of results, it might be time for causal modelers to contemplate theimplications of David Freedman’s (1992b) advice that causal inferencesshould always ride on the strength of the argument, and not on ’the magic ofleast squares’. 23

So where does this leave us? We do not wish to suggest that quantitativeempirical research is of minor importance for the sociological enterprise.Quite the contrary: Quantitative research is essential both for descriptivepurposes and for testing sociological theories. We do, however, believe thatmany sociologists have had all too much faith in statistical analysis as a toolfor generating theories, and that the belief in an isomorphism betweenstatistical and theoretical models, which appears to be an integral feature ofthe causal modeling approach, has hampered the development of sociologicaltheories built upon concrete explanatory mechanisms.

Over the past few years, one can discern a movement away from the’hard core’ position represented by Hauser. Nevertheless, the way in whichquantitative sociologists still allocate their time and intellectual energybetween statistical and theoretical modeling reveals a strong preference fordescription and testing of hypotheses formulated by others. 4 As suggestedby Stinchcombe (1993:27-28), sociologists in the multivariate modelingtradition still ’make only rhetorical use of the language of mechanisms’ andthey rarely show any serious intellectual commitment to developing thetheoretical foundation of the discipline themselves.

7. Social mechanisms: some selected examplesIn order to concretize the idea of general social mechanisms underlying arange of different social processes, we briefly examine three well-knowntheories in sociology - the self-fulfilling prophecy (Robert Merton), networkdiffusion (James Coleman), and threshold-based behavior (Mark Granovet-ter) - and we suggest that all are founded upon the same basic belief-formation mechanism.

The self-fulfilling prophecy is perhaps the most famous of all mechan-isms-based theories in sociology and was formulated in 1948 by RobertMerton in a seminal article (Merton [1948] 1968). The basic idea is that an

294

initially false definition of a situation evokes behavior that eventually makesthe false conception come true. The key example that Merton uses toillustrate his argument is a run on a bank. If a rumor of insolvency somehowgets started, some depositors will withdraw their savings. Their withdrawalwill strengthen the belief in the rumor, partly because the withdrawalsactually may hurt the financial standing of the bank, but more importantlybecause the act of withdrawal in itself signals to others that somethingindeed might be wrong with the bank. This produces even more withdrawals,which further reduces the trust in the bank, and so on. Because of theoperation of this mechanism, even an initially sound bank may go bankruptif enough depositors withdraw their money in the (initially) false belief thatthe bank is insolvent.

The study of network diffusion processes is an important area of

sociological research (cf. Burt 1987; Marsden & Podolny 1990; Strang &Tuma 1993; Hedstrbm 1994b). To a considerable extent this research is

inspired by Coleman, Katz & Menzel’s classic study of the diffusion of a newdrug (see Coleman, Katz & Menzel 1957, 1966). Their main finding was thatthe physicians’ positions in various professional networks influenced thediffusion process, particularly during the period immediately after the newdrug had been introduced on the market. Their explanation for this finding isreminiscent of Merton’s argument about the self-fulfilling prophecy:

Why should these sociometric ties to colleagues who have used the drug beinfluential during the first months of the drug’s availability, but not later? Onepossible answer lies in the greater uncertainty about the drug that must haveprevailed when it was new ... We know from work in the tradition of Sherifthat it is precisely in situations which are objectively unclear that socialvalidation of judgments becomes most important. More generally, thisexplanation implies that a doctor will be influenced more by what hiscolleagues say and do in uncertain situations, whenever and wherever theymay occur, than in clear-cut situations. (Coleman, Katz & Menzel 1957:268-269)

The core of their argument is consequently that networks are importantbecause information about innovations, in this case a new drug, diffusethrough them, and that an individual’s propensity to adopt the innovation isinfluenced by what others do, particularly when there is a great deal ofuncertainty about the true value of the innovation.

Our final example is Granovetter’s threshold theory of collectivebehavior (see Granovetter 1978; Granovetter & Soong 1983). Granovetterargued that an individual’s decision whether or not to participate incollective behavior often depends in part on how many other actors alreadyhave decided to participate. He further argued that actors differ in terms ofthe number of other actors who already must participate before they decideto do the same, and he introduced the concept of an individual’s ’threshold’ todescribe this individual heterogeneity. An actor’s threshold denotes theproportion of the group which must have joined before the actor in question iswilling to do so.

An important qualitative result of Granovetter’s analysis was that evenslight differences in thresholds can produce vastly different collective

295

outcomes. Assume, for example, a group of 100 people. One individual has athreshold of 0, another has a threshold of 1, a third has a threshold of 2, andso on up to the last individual who has a threshold of 99. Initially only theperson with a threshold of 0 will participate. His/her participation willactivate the person with a threshold of 1 and this person’s participation, inturn, will activate the person with a threshold of 2. This process will continueuntil all 100 people participate. If the distribution of thresholds changedslightly - for example if the person with a threshold of 1 was replaced with aperson with a threshold of 2 - the collective outcome would changedramatically. The initiator, of course, will participate also under these newconditions. But since there is no one with a threshold of 1, the process ends atthis point; only one person will participate in the ’collective’ movement. Thepopulations are virtually identical, but the collective outcomes are vastlydifferent (see Granovetter 1978).

Granovetter gives a range of examples of threshold-based behavior, butthe following example illustrates particularly well the logic behind this sortof conditional behavior:

Suppose you are in an unfamiliar town and enter an unknown restaurant onSaturday evening at seven o’clock. Whether or not you decide to take a mealthere will depend in part on how many others have also decided to do so. If theplace is nearly empty, it is probably a bad sign - without some minimal numberof diners, one would probably try another place. (1978:1438-1439)

The reason that the number of visitors at the restaurant is likely to influencean individual’s choice of restaurant (or more generally, why the number ofparticipants in a collective action is likely to influence an individual’sdecision whether or not to join the action), Granovetter argues, is that insituations of uncertainty, the number of diners (movement participants) is asignal about the likely value of the action (e.g. the quality of the food beingserved or the benefits that are likely to accrue to the individual if he/she joinsa particular group for collective action), and this signal may be decisive forthe individual’s choice of action.

In order more clearly to see the logical structure of the argumentsadvanced by Merton, Coleman, and Granovetter it is useful to adopt aslightly more formalized language. Let

Pit = propensity of individual i to perform the act being analyzed at time t, e.g.withdrawing savings from the bank, adopting a new drug, visiting a

restaurant, or joining an organization for collective action, andbit = the strength of individual i’s belief in the value or necessity of performingthe act in question at time t.

All three authors assume that individuals are goal directed and that anindividual’s propensity to perform the act being analyzed is an increasingfunction f of the individual’s belief in the value of performing the act:

Plt = f (bIt). However, the core mechanism that gives Merton’s, Coleman’s,and Granovetter’s analyses their counter-intuitive appeal concerns the waysin which individual beliefs are being formed. More specifically, their

proposed mechanism states that individual i’s belief in the value or necessityof performing the act is a function of the number of other individuals who

296

performed the act at time t-1. Merton’s bank customers based theirjudgments about the solvency of the bank on the number of other customerwithdrawing their savings from the bank; Coleman’s physicians based theirevaluations of the possible effect of the new drug on the doings of theircolleagues; and Granovetter’s restaurant visitor based his/her decision on thenumber of diners already in the restaurant. That is, they all assumed that

bit = g(nt-1)

where

nt-, = number of individuals performing the act time t-1, andg is an increasing function.

Inserting this expression into the former one, we arrive at P,t = f(g(nt-1))which suggests that an individual’s propensity to withdraw savings from thebank, adopt a new drug, visit a restaurant, or join an organization forcollective action is an increasing function of the number of other individualswho already have performed the same act.

The main difference between the three theories considered here centerson the function g, which provides the fine-grained details of the link betweenb,t and nt-1, and the details of this link will influence the aggregate dynamicsof the system.25 But the core characteristic of these theories that gives themtheir non-obvious character and appeal is the general belief-formationmechanism which states that the number of individuals who perform acertain act signal to others the likely value or necessity of the act, and thissignal will influence other individuals’ choice of action. It is this belief-formation mechanism that is at the heart of the self-fulfilling prophecies ofMerton, the network effects of Coleman, and the bandwagon effects ofGranovetter.26 On the fundamental level of mechanisms, the run on thebank, the prescription of the drug, and the emergence of the collectivemovement, all are analagous. 27

8. Social mechanisms: a typologyIn order to specify in more general terms the types of mechanisms we believeto be of particular importance for sociological theory, we propose a tentativetypology. This typology takes its departure from James Coleman’s wellknown model for how to conceptualize social action, the so-called macro-micro-macro model (see Figure 1).

The general thrust of this model is that proper explanations of macro-level change and variation entail showing how macro-states at one point intime influence the behavior of individual actors, and how these actions addup to new macro-states at a later time.28 That is, instead of analyzingrelationships between phenomena on the macro level, one should always tryto establish how macro-level events or conditions affect the individual (Step1), how the individual assimilates the impact of these macro-level events(Step 2), and how a number of individuals, by their actions and interactions,generate macro-level outcomes (Step 3). This way of conceptualizing social

297

Figure 1. Macro-micro-macro relations, according to James Coleman.Source: James S. Coleman, ’Social Theory, Social Research, and a Theory of Action’,American Journal of Sociology, May 1986, pp. 1309-1335.

action lends itself in a very natural way to a typology of mechanisms: macro-micro mechanisms, micro-micro mechanisms, and micro-macro mechanisms- and a few words will be said about each of these. 29

The first of these three types of mechanisms covers the macro-to-microtransition, and following the suggestion of Stinchcombe (1993) we refer to itas a situational mechanism. Erving Goffman’s (1963) work on behavior inpublic places and Karl Popper’s form of situational analysis (cf. Popper 1994)have these sorts of mechanisms at their core. The belief-formationmechanism discussed above, opportunity-generating mechanisms such asWhite’s (1970) vacancy chains, and preference-formation mechanisms suchas those expressed in the idea of reference groups (see Merton & Rossi 1968;Boudon 1988), are prototypical examples of general social mechanisms thatin a systematic and reasonably precise way link a social structure or othermacro-sociological state to the beliefs, desires, and opportunities ofindividual actors.

The second type of mechanism is a micro-level mechanism, and we referto it as an individual action mechanism. This type of mechanism shows howa specific combination of individual desires, beliefs, and action opportunitiesgenerates a specific action. A plurality of psychological and socialpsychological mechanisms operates at this level. General decision theoriesas well as more specific theories such as Leon Festinger’s (1957) theory ofcognitive dissonance and George Ainslie’s (1992) on discounting illustratedifferent types of action mechanisms. One concrete example that Colemanuses to illustrate individual action mechanisms comes from his reading ofThe Protestant Ethic (see Figure 2). The micro-to-micro transition has in thiscase to do with the individual believer’s realization that his or her values alsoimply a change in orientation towards economic activities, followed by actioninspired by these values. 30

The third type of mechanism covers the micro-to-macro transition, andwe propose to call it a transformational mechanism. Here a number ofindividuals interact with one another and the specific mechanism (which

298

Figure 2. Weber’s ProtestantEthic, conceptualized according to Coleman’s model.Source: James S. Coleman, ’Social Theory, Social Research, and a Theory of Action’,American Journal of Sociology, May 1986, p. 1322.

~ 1

depends upon the type of interaction) describes how these individual actionsare transformed into a collective outcome, sometimes unintended andunexpected by all actors. Several of the theories mentioned elsewhere inthis paper - such as Schelling’s tipping model, standard game-theoreticmodels such as the tragedy of the commons, and neoclassical market models- are built upon and illustrate specific transformational mechanisms.

9. Concluding remarksIn this concluding section we briefly summarize the main thrust of ourargument (see also Hedstrbm & Swedberg, forthcoming (b)). We have arguedthat the notion of social mechanisms is essential for sociological theory, andthat mechanism-based explanations are characterized by three core features:

1. The principle of direct causality;2. The principle of limited scope;3. The principle of methodological individualism.

The first of these principles - direct causality - has essentially to do withopening up the black-box, and that one should always strive to narrow thegap or lag between input and output, cause and effect. A mechanism-basedexplanation seeks to provide a fine-grained and tight coupling betweenexplanans and explanandum.

The second principle - limited scope - captures the essence of middle-range sociology and expresses the idea that sociology should not prematurelytry to establish universal social laws (which are unlikely to even exist) - but

299

should instead aim at explanations specifically tailored for a limited range ofphenomena.

The third principle - methodological individualism - captures the ideathat in the social sciences, actors and not variables do the acting. Amechanism-based explanation is not built upon mere associations betweenvariables, but always refers directly to causes and consequences of individualaction oriented to the behavior of others. A corollary to this principle statesthat there exist no macro-level mechanisms; macro-level entities or eventsare linked to one another via combinations of situational, individual action,and transformational mechanisms, i.e., all macro-level change should beconceptualized in terms of three separate transitions (macro-micro, micro-micro, and micro-macro).31

A general social mechanisms can now be defined in the following way: asocial mechanism is an integral part of an explanation which (1) adheres tothe three core principles stated above, and (2) is such that on the occurrenceof the cause or input, I, it generates the effect or outcome, O.

We realize that this definition may seem excessively general andabstract. But it is not so much the definition per se that is important, as thetype and style of theorizing it encourages. We believe that there is a naturaland important affinity between Robert Merton’s idea of middle-range theoryand the idea of social mechanisms, in the sense that social mechanisms arethe elementary building blocks of such theories. Other types of sociologicaltheory tend to make only rhetorical use of the notion of social mechanism;and this goes for grand theory (which often ignores all three principles statedabove) as well as for variable sociology (which ignores the principle ofindividual action). A focus on explanatory mechanisms helps sociology toavoid the trap of mindless empiricism on the one hand, and conventional andempty theorizing on the other.

First version received October 1995Final version accepted April 1996

AcknowledgementsWe thank Mario Bunge, Cecilia Gil-Swedberg, Mark Granovetter,Barbara Hobson, Ole-Jorgen Skog, Arthur Stinchcombe, MichaelTahlin, Lars Ud6hn, seminar participants at the Department of

, Sociology, Stockholm University, participants at the conference on’ Social Mechanisms held at the Royal Swedish Academy of Sciences,

June 6-7, 1996, and two anonymous referees for their usefulcomments on an earlier version of this article. We owe specialthanks to Carl-Gunnar Janson for his detailed written comments,and to Alejandro Gil-Villegas and Aage Sorensen for providingvaluable background information.

Notes1 The term ’mechanism’ is often used in sociology but in a casual

manner and hence constitutes what Merton calls a ’proto-concept’.Merton explains: ’A proto-concept is an early, rudimentary, particu-

300

larized, and largely unexplicated idea (which is put to occasional usein empirical research and, indeed, often derives from it); a concept isa general idea which once having been defined, tagged, substantiallygeneralized, and explicated can effectively guide inquiry into

seemingly diverse phenomena’ (Merton 1984:267).2 Weber, for example, rarely used the term ’mechanism’

(Mechanismus) with the exception of his analysis of bureaucracy,where it is more or less synonymous with ’machine’ (see, e.g., Weber[1921-1922] 1978:961, 967, 988; Weber as cited in Marianne Weber1975:416-417). In Zwischenbetrachtung, Weber makes the followingstatement, which sums up the situation brought about by Descartesand Newton: ’The tension between religion and intellectual knowl-edge definitely comes to the fore wherever rational, empiricalknowledge has consistently worked through to the disenchantmentof the world and its transformation into a causal mechanism(kausalen Mechanismus)’ (Weber 1946:350; emphasis added).

3 Simmel’s work is rich - and still relatively unexplored - insuggestions on this score. A well-known example is his analysis ofthe ways in which the number of actors influences the structure ofinteraction. Durkheim had a powerful sense for social mechanismsthat unites the individual and the group, as his various hypothesesin Suicide make clear. Some of Durkheim’s analyses of mechanismsare, however, marred by his functionalism, such as the argumentthat society has a tendency to maintain itself in the face of attacks orthreats of dissolution. As for Weber, the work that is usually citedapropos mechanisms is The Protestant Ethic with its argument thatthe religious idea of vocation (Beruf) came to influence economicbehavior in a methodical, capitalist direction. A number of interest-ing mechanisms can also be found in Weber’s later work, especiallyin Economy and Society and The Economic Ethics of the World

Religions.4 Merton’s work on middle-range theory goes back to his critiqueof Parsons at the 1947 meeting of the American SociologicalAssociation (see Merton 1948). Also Parsons discussed the conceptof mechanism, especially in his work from the early 1950s (see, e.g.,Parsons 1951:201-325; Parsons & Shils 1951:125-149). Parsons’view, however, was marred by his functionalism as well as by hisattempts at grand theory; and the function of social mechanisms wasbasically reduced to that of maintaining the social system when thiswas threatened in some manner. One of the first sociologists to showmore than a casual interest in the concept of mechanisms is GeorgeLundberg in Foundations of Sociology from 1939. Lundberg arguedfor a common sense approach to the notion of mechanism, often withfunctionalist overtones. (We thank Lars Udéhn for the reference toLundberg’s work.)

5 To this can be added a few other more general mechanismsthat were to emerge from Merton’s own work as well as fromColumbia Sociology in general: the two-step model of communica-tion, the self-fulfilling prophecy, the Matthew Effect, and thediffusion mechanism of Medical Innovation (1966).

6 Though Merton should be praised for discussing ’socialmechanisms’ in his essay on middle-range theory, it is also truethat Merton did not succeed in singling out this concept as much as

301

one would have wished. See in this context also Social Mechanisms:Studies in Sociological Theory (1958) by the Swede Georg Karlsson.Karlsson’s main source of inspiration was Nicolas Rashevsky, whosework he had come into contact with in Chicago and who Lazarsfeldhad introduced at Columbia. Karlsson explicitly rejects systems-oriented sociological theory and opts for middle range sociology(Karlsson 1958:10, 16). His work is centered around three families ofmechanisms: ’social diffusion’, ’group choice mechanisms’, and’general mechanisms of interaction’. When Karlsson returned toSweden, he soon changed orientation since the intellectual climate inSwedish sociology was hostile to this type of theorizing (interviewwith Karlsson on 19 September 1995).7 Of Elster’s publications so far, the place where one can find themost complete statement on this topic is in the introduction to Nutsand Bolts for the Social Sciences, entitled ’Social Mechanisms’(Elster 1989:3-10). The clearest statement of Stinchcombe’s views onthese matters can be found is an article from 1991 (republished inmore or less identical form in 1993). See also e.g. Sayer (1984),Pawson (1989), and Kiser & Hechter (1991) for some recentdiscussions of the role of mechanisms in the social sciences.

8 Tocqueville is a true master of mechanisms, according to Elster(1993), who cites a number of examples from Democracy in Americaand Ancien Régime.

9 For a furious attack on Nuts and Bolts for precisely thisreason, see Humphreys (1991).

10 Goffman writes in Behavior in Public Places: ’By the termsituation I shall refer to the spatial environment anywhere withinwhich an entering person becomes a member of the gathering that is(or does then become) present. Situations begin when mutualmonitoring occurs, and lapse when the second-last person has left’(Goffman 1963:18).

11 The word ’mechanism’ originally comes from the word’machine’ (cf. e.g. Oxford English Dictionary 1989). That theassociation to ’machines’ and ’mechanics’ is still popular amongthose who write on social mechanisms is clear from the title ofElster’s book: Nuts and Bolts of the Social Sciences.

12 A famous example of a 19th century mechanism is that ofhomeostasis, discovered by Claude Bernard and usually defined as aself-regulating process through which a biological system maintainsits stability. Sociologists will recall that the term ’homeostatsis’ wascoined by W. B. Cannon, whose work The Wisdom of the Body (1932)deeply influenced Parsons. (In chemistry the same idea is known asthe principle of Le Chatelier.) Parsons may also have picked up theidea from Cannon.

13 The problem that Hirschman addresses has to do with whathappens when an organization (including a firm) begins to decline.According to Hirschman, two ’mechanisms of recuperation’ are

usually triggered off in this situation, one that is discussed primarilyin economics (’exit’) and one that primarily political scientists focuson (’voice’).

14 Hempel (1942) uses the example of an automobile radiator

cracking during a cold night to illustrate the logic of his proposal.The general laws cited in the explanation would need to refer to how

302

the pressure of water changes with changes in temperature andvolume, and the specific circumstances referred to would beconditions such as the temperature during the night and thebursting pressure of the radiator. A proper explanation has beenproposed if, and only if, the proposition about cracking of theradiator can be logically deduced from the sentences stating the lawsand the specific circumstances.

15 Mark Granovetter, who is a student of Hempel, disagreeswith our argument here and notes that ’If [only] correlations wereproduced he [that is, Hempel] would no doubt point out that moredetailed covering laws are needed to explain the correlations’(personal communication).

16 It should be emphasized that the distinction between ’black-boxes’ and ’mechanisms’ to some extent is time-bound. In the wordsof Patrick Suppes (1970:91): ’From the standpoint of either scientificinvestigation or philosophical analysis it can fairly be said that oneman’s mechanism is another man’s black box. I mean by this that themechanisms postulated and used by one generation are mechanismsthat are to be explained and understood themselves in terms of moreprimitive mechanisms by the next generation.’

17 For the sake of clarity, it should be noted that we also include’intentions’ among the possible ’causes’ of individual action.

18 In this respect, our position is closer to that of Stinchcombewho explicitly argues against those who do not accept that one oftenhas to make simplifying assumptions about the units on the lowerlevel (e.g. homo economicus) in order to be able to generateinteresting results on the higher level. In other words, a mechanismdoes not necessarily have ’to be true to be useful’ - but it has to’produce interesting hypotheses or explanations at the higher level’(Stinchcombe 1993:27).

19 The affinity between behaviorism and structural equationmodeling was also noted by O. D. Duncan himself: ’In [structuralequation] models that purport to explain the behavior of individualpersons, the coefficients [of the structural equation] could well takethe form of units of response per unit of stimulus strength; thestructural equation is, in effect, a stimulus-response law’ (Duncan1975:162-163).

20 Throughout his career, Coleman was a strong proponent of a

generative view of causality and he often expressed serious doubtsabout the usefulness of the type of causal analysis referred to above.Already in Introduction to Mathematical Sociology he wrote: ’Note,however, that there is nowhere the proposal simply to engage incurve-fitting, without an underlying model which expresses a socialprocess. If the data happen to fit a simple curve, this may provide aneconomical statement of the data, in terms of the one or two

parameters of the distribution curve. But if there is no underlyingmodel with a reasonable substantive interpretation, little has beengained by such curve fitting’ (Coleman 1964:518).

21 As these terms are being used here, a ’theoretical model’differs from a ’mechanism’ in primarily two ways: (1) a theoreticalmodel is always formalized, something which mechanisms onlysometimes are, and (2) a mechanism is a more elementary entity

303

than a theoretical model; (formalized) mechanisms often constitutethe building-blocks of models.

22 Another important reason for why the causal modelingtradition has had such moderate success in sociology is the characterof the data being used. The inductive approach adopted in thistradition is modeled upon experimental research, but the data beingused are normally of a non-experimental kind. As emphasized byLieberson (1985), unmeasured selection is a serious obstacle tocausal inferences when using non-experimental data.

23 As exemplified in the work of John Goldthorpe (forthcoming),there is an emerging consensus being established even amongempirically oriented sociologists that causality in social processescannot be established from quantitative analyses alone.

24 In areas such as labor market sociology, the hypotheses beingtested have furthermore usually been formulated by members ofother scientific disciplines, i.e., by economists. There are numerousempirical tests being reported in the sociological literature of humancapital, internal labor markets, dual labor markets, and transactioncosts, all theories or hypotheses developed by economists, but

exceptionally few tests of mechanisms proposed by sociologists;White’s (1970) vacancy chain mechanism being one noteworthyexception.25 Coleman assumed that g was a function of the sociometricties, Granovetter assumed that it was a function of individualthresholds, and Merton left the functional form unspecified. Whenmechanisms are expressed in mathematical language they appear asfunctions transforming variables. These functions can be distin-guished from one another on the basis of their functional form andtheir parameter values (see Hernes 1976).26 A process like this might also arise in a situation where not allactors form their beliefs on the basis of the mechanisms discussedabove. Consider the following situation: (1) I do not believe inproposition A; (2) I know (or believe) that many others believe in Aand their belief in A will have negative consequences for me; (3)other’s belief in A will induce me to act as if A were true; and this (4)will signal to others the correctness of proposition A. We believe thatrecent instability on various financial markets can partly beattributed to these types of processes. But also social processes,such as the fear of magnetic fields referred to earlier, exhibit thesecharacteristics. If this fear threatens to undermine the propertyvalues of houses close to power lines, houseowners (even those whohave realized that the fear is groundless) are induced to put theirhouses on the market, and this ’escape’ sends a signal to others thatthere indeed must be some substance to the view that living close toa power line is a potential health hazard. This threatens to furtherreduce property values, and so on.

27 In addition to this belief-formation mechanism, there are ofcourse other action and transformation mechanisms that are

involved in Merton’s, Coleman’s, and Granovetter’s analyses, butthese mechanisms are commonplace and tangential to the coreprocesses they analyze.

28 It is important to note that this form of methodologicalindividualism does not imply that macro-level factors are unim-

304

portant or somehow theoretically inappropriate when it comes toexplaining individual action. It only implies that these extra-individual entities also in principle should be accounted for withreference to the intended and unintended consequences of indivi-duals’ past behavior (see Hedström & Swedberg, forthcoming (a)).

29 The logic of Coleman’s argument also suggests that any kindof continuous social action can be conceptualized as a long chain ofsuccessive macro-micro-macro transformations, where, in manycases, only the peaks, so-to-speak, (’macro-macro’) are visible to theresearcher - but where the analytical point is precisely to explainthis cumulative social action as a result of a large number of macro-micro-macro transitions.

30 Coleman’s interpretation of Weber in this example has beenchallenged by several authors, but this does not affect the logic ofColeman’s argument (see Swedberg, forthcoming).

31 In addition to these basic characteristics of social mechan-

isms, the ideal mechanism, it seems to us, should also be simple andnon-obvious.

References

Abbott, A. 1992. What Do Cases Do? Some Notes on Activity inSociological Analysis. In C. C. Ragin & H. S. Becker (eds.), What isa Case? Exploring the Foundations of Social Inquiry, pp. 58-82.Cambridge: Cambridge University Press.

Ainslie, G. 1992. Picoeconomics: The Strategic Interaction ofSuccessive Motivational States within the Person. Cambridge:Cambridge University Press.

Alexander, J. C. 1988. The New Theoretical Movement. In N.Smelser (ed.), Handbook of Sociology, pp. 77-102. London: SAGE.

Beckner, M. O. 1967. Mechanism in Biology. In P. Edwards (ed.), TheEncyclopaedia of Philosophy, Vol. 5, pp. 250-252. New York: TheMacmillan Company and the Free Press.

Bennett, W. R. Jr. 1994. Health and Low-Frequency ElectromagneticFields. New Haven: Yale University Press.

Bhaskar, R. 1978. A Realist Theory of Science. Sussex: TheHarvester Press.

Boudon, R. 1974. Education, Opportunity, and Social Inequality.New York: John Wiley & Sons.

Boudon, R. 1976. Comment on Hauser’s Review of Education,Opportunity, and Social Inequality. American Journal of Sociol-ogy, 81, 1175-1187.

Boudon, R. 1979. Generating Models as a Research Strategy. In R. K.Merton, J. S. Coleman & P. H. Rossi (eds.), Qualitative andQuantitative Social Research: Papers in Honor of Paul F.

Lazarsfeld, pp. 51-64. New York: The Free Press.Boudon, R. 1988. The Logic of Relative Frustration. In M. Taylor

(ed.), Rationality and Revolution, pp. 245-267. Cambridge: Cam-bridge University Press.

Bower, G. H. 1975. Cognitive Psychology: An Introduction. In W. K.Estes (ed.), Handbook of Learning and Cognitive Processes, pp.25-80. New York: John Wiley & Sons.

Bunge, M. 1967. Scientific Research. Volume 3 in Studies of the

305

Foundations, Methodology, and Philosophy of Science. Berlin:

Springer-Verlag.Burt, R. S. 1987. Social Contagion and Innovation: Cohesion VersusStructural Equivalence. American Journal of Sociology, 92,1287-1335.

Cannon, W. B. 1932. The Wisdom of the Body. New York: W. W.Norton & Co.

Coleman, J. S. 1964. Introduction to Mathematical Sociology. NewYork: The Free Press.

Coleman, J. S. 1986. Social Theory, Social Research, and a Theory ofAction. American Journal of Sociology, 91, 1309-1335.

Coleman, J. S. 1990. Foundations of Social Theory. Cambridge, MA:Harvard University Press.

Coleman, J. S., Katz, E. & Menzel, H. 1957. The Diffusion of anInnovation Among Physicians. Sociometry, XX, 253-270.

Coleman, J. S., Katz, E. & Menzel, H. 1966. Medical Innovation.Indianapolis: Bobbs-Merril.

Crick, F. 1989. What Mad Pursuit: A Personal View of ScientificDiscovery. London: Penguin Books.

Dijksterhuis, E. J. 1986. The Mechanization of the World Picture:Pythagoras to Newton. Princeton: Princeton University Press.

Doll, R. & Peto, R. 1981. The Causes of Cancer. Oxford: OxfordUniversity Press.

Duncan, O. D. 1975. Introduction to Structural Equation Models.New York: Academic Press.

Elster, J. 198?. Explaining Technical Change: A Case Study in thePhilosophy of Science. Cambridge: Cambridge University Press.

Elster, J. 1989. Nuts and Bolts for the Social Sciences. Cambridge:Cambridge University Press.

Elster, J. 1991. Patterns of Causal Analysis in Tocqueville’sDemocracy in America. Rationality and Society, 3, 277-297.

Elster, J. 1992. Local Justice: How Institutions Allocate ScarceGoods and Necessary Burdens. New York: Russell Sage Founda-tion.

Elster, J. 1993. Political Psychology. Cambridge: Cambridge Uni-versity Press.

Festinger, L. 1957. A Theory of Cognitive Dissonance. Stanford:Stanford University Press.

Feychting, M. & Ahlbom, A. 1993. Magnetic Fields and Cancer inChildren Residing Near Swedish High-voltage Power Lines.American Journal of Epidemiology, 138, 467-481.

Freedman, D. A. 1991. Statistical Analysis and Shoe Leather.Sociological Methodology, 21, 291-313.

Freedman, D. A. 1992a. As Others See Us: A Case Study in PathAnalysis. In Juliet P. Schaffer (ed.), The Role of Models inNonexperimental Social Science: Two Debates, pp. 3-30. Washing-ton, DC: American Educational Research Association.

Freedman, D. A. 1992b. A Rejoinder on Models, Metaphors, andFables. In J. P. Schaffer (ed.), The Role of Models in Nonexperi-mental Social Science: Two Debates, pp. 108-125. Washington, DC:American Educational Research Association.

Ghins, M. 1990. Méchanisme. In S. Auroux (ed.), Les Notions

306

Philosophique - Dictionnaire, pp. 1582-1583. Paris: PressesUniversitaires de France.

Giddens, A. 1987. Social Theory and Modern Sociology. Stanford:Stanford University Press.

Goffman, E. 1963. Behavior in Public Places: Notes on theOrganization of Gatherings. New York: The Free Press.

Goldthorpe, J. H. Forthcoming. The Quantitative Analysis of Large-scale Data Sets and Rational Action Theory. European SociologicalReview.

Granovetter, M. 1978. Threshold Models of Collective Behavior.American Journal of Sociology, 83, 1420-1443.

Granovetter, M. 1982. Review of Christopher Jencks et al., Who GetsAhead? The Determinants of Economic Success in America. Theoryand Society, II, 257-262.

Granovetter, M. & Soong, R. 1983. Threshold Models of Diffusionand Collective Behavior. Journal of Mathematical Sociology, 9,165-179.

Hamlin, C. 1990. A Science of Impurity: Water Analysis in Nine-teenth Century Britain. Bristol: Adam Hilger.

Harré, R. 1970. The Principles of Scientific Thinking. Chicago: TheUniversity of Chicago Press.

Harré, R. 1984. The Philosophies of Science. 2nd ed. Oxford: OxfordUniversity Press.

Hauser, R. M. 1976. Review Essay: On Boudon’s Model of SocialMobility. American Journal of Sociology, 81, 911-928.

Hedström, P. 1994a. Magnetic Fields and Cancer Risks: ShouldRecent Epidemiological Results be a Cause of Concern? OccasionalPaper Series. Department of Sociology, Stockholm University.

Hedström, P. 1994b. Contagious Collectivities: On the SpatialDiffusion of Swedish Trade Unions, 1890-1940. American Journalof Sociology, 99, 1157-1179.

Hedström, P. Forthcoming. Rational Choice and Social Structure: OnRational-choice Theorizing in Sociology. In B. Wittrock (ed.), SocialTheory and Human Agency. London: Sage.

Hedström, P. & Swedberg, R. Forthcoming (a). Rational Choice,Empirical Research, and the Sociological Tradition. EuropeanSociological Review.

Hedström, P. & Swedberg, R. (eds). Forthcoming (b). SocialMechanisms: An Analytical Approach to Social Theory. Cam-bridge : Cambridge University Press.

Hempel, C. G. 1942. The Function of General Laws in History.Journal of Philosophy, 39, 35-48.

Hempel, C. G. 1962. Explanation in Science and in History. In R. G.Colodny (ed.), Frontiers of Science and Philosophy, pp. 9-33.

Pittsburgh: University of Pittsburgh Press.Hernes, G. 1976. Structural Change in Social Processes. AmericanJournal of Sociology, 82, 513-547.

Hirschman, A. O. 1970. Exit, Voice, and Loyalty: Responses to

Decline in Firms, Organizations, and States. Cambridge: HarvardUniversity Press.

Homans, G. C. 1974. Social Behavior: Its Elementary Forms. NewYork: Harcourt Brace.

307

Humphreys, P. 1991. Review of Jon Elster, Nuts and Bolts for theSocial Sciences. Philosophy of the Social Sciences, 21, 114-121.

Kalleberg, A. & Berg, I. 1987. Work and Industry: Structures,Markets, and Processes. New York: Praeger.

Karlsson, G. 1958. Social Mechanisms: Studies in SociologicalTheory. Stockholm: Almquist & Wicksell.

Kiser, E. & Hechter, E. 1991. The Role of General Theory inComparative-historical Sociology. American Journal of Sociology,97, 1-30.

Lazarsfeld, P., Berelson B. & Gaudel, H. 1944. The People’s Choice.New York: Columbia University Press.

Lieberson, S. 1985. Making it Count: The Improvement of SocialResearch and Theory. Berkeley: University of California Press.

Lundberg, G. A. 1939. Foundations of Sociology. New York: TheMacmillan Company.

Marsden, P. V. & Podolny, J. 1990. Dynamic Analysis of NetworkDiffusion Processes. In H. Flap & J. Weesie (eds.), Social NetworksThrough Time, pp. 197-214. Utrecht: ISOR.

Merton, R. K. [1948] 1968. The Self-fulfilling Prophecy. In SocialTheory and Social Structure, pp. 475-490. New York: The FreePress.

Merton, R. K. 1948. Discussion [of Talcott Parsons, ’The Position ofSociological Theory’]. American Sociological Review, 13, 164-168.

Merton, R. K. 1967. On Sociological Theories of the Middle Range. InOn Theoretical Sociology, pp. 39-72. New York: The Free Press.

Merton, R. K. 1984. Socially Expected Durations: A Case Study ofConcept Formation in Sociology. In W. W. Powell & R. Robbins(eds.), Conflict and Concensus, pp. 262-283. New York: The FreePress.

Merton, R. K. & Rossi, A. S. 1968. Contributions to the Theory ofReference Group Behavior. In Social Theory and Social Structure,pp. 279-334. New York: The Free Press.

ORAU. 1992. Health Effects of Low-Frequency Electric and MagneticFields. Oak Ridge Associated Universities Panel. Washington, DC:U.S. Government Printing Office.

Oxford English Dictionary. 1989. Mechanism. In Vol. IX, pp.536-537. Oxford: Clarendon Press.

Parsons, T. 1951. The Social System. New York: The Free Press.Parsons, T. & Shils, E. (eds.). 1951. Toward A General Theory ofAction. Cambridge: Harvard University Press.

Pawson, R. 1989. A Measure for Measures: A Manifesto for EmpiricalSociology. London: Routledge.

Popper, K. R. 1994. Models, Instruments, and Truth: The Status ofthe Rationality Principle in the Social Sciences. In K. R. Popper,The Myth of the Framework: In Defence of Science and Rationality,pp. 154-184. London: Routledge.

Sayer, A. 1984. Method in Social Science. London: Hutchinson.Schelling, T. C. 1978. Micromotives and Macrobehavior. New York:W. W. Norton & Company.

Schumpeter, J. A. 1989. Essays on Entrepreneurs, Innovations,Business Cycles, and the Evolution of Capitalism. New Brunswick:Transaction Publishers.

308

Sørensen, A. B. 1990. Interview. In R. Swedberg, Economics andSociology, pp. 303-315. Princeton: Princeton University Press.

Stinchcombe, A. L. 1968. Constructing Social Theories. Chicago: TheUniversity of Chicago Press.

Stinchcombe, A. L. 1991. The Conditions of Fruitfulness of Theoriz-ing about Mechanisms in Social Science. Philosophy of the SocialSciences, 21, 367-388. (See also Stinchcombe 1993).

Stinchcombe, A. L. 1993. The Conditions of Fruitfulness of Theoriz-ing about Mechanisms in Social Science. In A. Sørensen & S.Spilerman (eds.), Social Theory and Social Policy: Essays in Honorof James S. Coleman, pp. 23-41. Westport, Conn.: Praeger. (Thispaper is identical, save for a few items, to Stinchcombe 1991).

Strang, D & Tuma, N. B. 1993. Spatial and Temporal Heterogeneityin Diffusion. American Journal of Sociology, 99, 614-639.

Suppes, P. 1970. A Probabilistic Theory of Causality. Acta Philoso-phica Fennica, Fasc. XXIV. Amsterdam: North-Holland PublishingCompany.

Swedberg, R. Forthcoming. Analyzing the Economy: On theContribution of James S. Coleman. In J. Clark (ed.), JamesColeman. London: The Falmer Press.

Therborn, G. 1991. Cultural Belonging, Structural Location andHuman Action: Explanation in Sociology and in Social Science.Acta Sociologica, 34, 177-191.

Townsend, P. & Davidson, N. 1982. Inequalities in Health: The BlackReport. Baltimore: Penguin Books.

Turner, J. D. 1986. Review Essay: The Theory of Structuration.American Journal of Sociology, 91, 969-977.

Van Parijs, P. 1981. Evolutionary Explanation in the Social Sciences:An Emerging Paradigm. Totowa: Rowman and Littlefield.

Weber, M. 1975. Max Weber: A Biography. Trans. Harry Zohn. NewYork: John Wiley & Sons.

Weber, M. [1921-1922] 1978. Economy and Society: An Outline ofInterpretive Sociology. Edited by G. Roth & C. Wittich. Trans.Ephraim Fischoff et al. Two vols. Berkeley: University ofCalifornia Press.

Weber, M. 1946. From Max Weber: Essays in Sociology. Translatedand edited by H. Gerth & C. Wright Mills. New York: OxfordUniversity Press.

Weber, M. 1949. The Methodology of the Social Sciences. Translatedand edited by E. A. Shils & H. A. Finch. New York: The Free Press.

White, H. C. 1970. Chains of Opportunity: System Models of Mobilityin Organizations. Cambridge: Harvard University Press.

Weinberg, S. 1993. Dreams of a Final Theory: The Search for theFundamental Laws of Nature. London: Vintage.

Wright von, G. H. 1971. Explanation and Understanding. Ithaca:Cornell University Press.