In this new book, Jon Elster investigates the intellectual origins of a mechanism-based approach to social science theorizing, an approach that he has played a key role in developing (along with Randall Collins, Peter Hedstrom, Daniel Little, Wesley Salmon, Thomas Schelling, Arthur Stinchcombe, Richard Swedberg, Charles Tilly, and many others). Elster argues that the progenitor of the mechanismic approach to social reality is Alexis de Tocqueville, a thesis that the book ably defends.Footnote ¹

My comments will focus on Elster's broader vision, drawing on this book and his other writings about mechanisms.Footnote ² (All citations are to Alexis de Tocqueville unless otherwise noted.) I begin with the concept of a mechanism, as understood by Elster. The next section deals with the empirical and theoretical repercussions of his microfoundational agenda. The final section discusses his vision of a universal toolbox of explanations for social science, comprised of exportable causal mechanisms.

The Concept of a Mechanism. In Elster's work, one finds two definitions, side-by-side (and often intermingling). First, (a) there is the idea of mechanism as an intermediary variable, causal chain, or pathway from X to Y (the causal relationship of interest). Here, mechanism is opposed to X/Y “covariation” and to “black-box” arguments that pay no attention to what may have generated X's impact on Y. Without mechanisms, causal arguments offer no (or very little) explanation. Second, (b) there is the notion of mechanism as a particular type of causal explanation. Here, mechanisms are “frequently occurring and easily recognizable causal patterns that are triggered under generally unknown conditions or with indeterminate consequences”Footnote ³ (emphasis in original).

At present, I am primarily interested in meaning (b). However, the two meanings are often difficult to distinguish because (b) sometimes fulfills the role of (a)—thus functioning as a subset of (a).

Let us probe the content of (b) more closely. For Elster, the question of mechanisms usually centers on human motivations, that is, emotions and/or cognitions. For example:

Elster claims to draw these mechanisms from Tocqueville's work, thus generating a degree of theoretical consistency in the latter that would otherwise be lacking. This claim I will leave in abeyance since our focus is on the more general subject of mechanisms and their role in contemporary social science.

Elster emphasizes the probabilistic nature of causal mechanisms—in contrast to deterministic causal laws (pp. 2, 184).Footnote ⁴ There are exceptions to every rule, contextual limits to where a rule is applicable, and causal heterogeneity even within the same general context (the same cause leading to different outcomes). Yet Elster is also at pains to emphasize that mechanismic generalizations strive for a lawlike status: They are would-be laws, since laws (insofar as they are true) are more useful in advancing the goals of science: “Explanation by laws is better—but also more difficult.”Footnote ⁵

To my way of thinking, the distinction between a law and a probabilistic statement is best understood as a matter of degree rather than a dichotomy. Surely, there is always some uncertainty about whether X causes Y. And if one buys the Heisenberg uncertainty principle, this uncertainty is with reference not just to our knowledge of an event but also to its ontological properties. Anyway, if the uncertainty of the universe—and, a fortiori, of the social-scientific universe—is accepted, then it makes little sense to dichotomize deterministic and probabilistic statements about the world. Rather, we should speak of degrees of uncertainty.

In any case, the three mechanisms sketched above—spillover, compensation, and satiation—fulfill Elster's bill. They are probabilistic rather than deterministic. They embody causal heterogeneity (the same causal factor may trigger different psychological mechanisms). Whether one or another mechanism comes into play is the product of numerous contextual factors. At the same time, they are presumably generalizable, perhaps even universal (in the loose form in which they are articulated here).

One can intuit several additional features characterizing Elster's concept of a mechanism. First, a mechanism is intended to explain not merely a class of events but also particular events, for example, the French Revolution or the advent of American democracy (two issues explored by Tocqueville). Mechanisms possess an immanence that many causal laws do not. They explain actors' motives. Second, mechanisms are not amenable to direct manipulation, even in principle. One cannot manipulate a motivation. Of course, one can manipulate the trigger for a motivation. But since the trigger is only probabilistically related to the mechanism, and only one potential cause of that mechanism, it must be regarded as a proxy for the phenomenon of interest. Third, human motivations are difficult to measure. Usually, we intuit them from patterns of behavior.

The latter two factors—non-manipulability and non-operationalizability—pose major problems for empirical investigation, limiting the falsifiability of Elster's approach (as acknowledged in Explaining Social Behavior).Footnote ⁶ Indeed, Elster's mechanisms invite a renewal of Karl Popper's attack on Freudianism (which Popper condemned as unfalsifiable and therefore unscientific), and would be banished from empirical work focused on causal effects.Footnote ⁷ Nonetheless, they are plausible. And they are not easily dispensed with. It is difficult to imagine social explanations without theories (at least implicit) of human motivation.

The Microfoundational Agenda. A final important characteristic of mechanisms is their microfoundational quality. This feature is embraced by Elster,Footnote ⁸ as well as by most other writers who have identified with the mechanismic agenda.Footnote ⁹ In this respect, Tocqueville is usefully contrasted with Marx, Durkheim, and other nineteenth-century founders of the macrofoundational approach to social science (p. 7). It is true that Tocqueville's works are littered with vast generalizations; but it is also true that, as Elster rightly points out, they are usually generalizations about individual-level behavior, with special reference to human motivations. Elster's choice of avatars is apt.

To be sure, Elster takes a less extreme view of the microfoundational enterprise than some others. He does not say, for example, that all macro-level phenomena can be explained by micro-level phenomena (the pure reductionist view). Indeed, he is careful to show in the conclusion that causal arrows go in both directions. However, the weight of the book, and of Elster's other work, is on microfoundations: explaining by scoping down, usually to the individual level.

This is a persuasive plan for theoretical reconstruction in the social sciences, as evidenced in recent work reviewed in the previously cited studies.Footnote ¹⁰ However, it is not without its difficulties. First, one must decide how “micro” an explanation needs to be in order to qualify as microfoundational—evidently, a matter of degrees. One interpretation suggests burrowing down at least one level of analysis lower than the initial covariational regularity.

Another interpretation is that one ought to burrow down to lower and lower levels of analysis until one reaches a level of explanation that is considered solid (“foundational”). Of course, reasonable people may disagree upon what sort of knowledge is sufficiently self-evident to be regarded as foundational. And it leaves open the possibility that no deeper level of analysis may be needed; the covariational result is interpretable without further burrowing.

Nowadays, macro-level explanation in the social sciences is usually understood to apply to large groups or organizations (e.g., communities, nation-states, governments), while micro-level explanation applies to individual-level behavior. Indeed, for most writers, the goal of microfoundational explanation is explicitly tied to explanations that reach down to the level of individual actors.Footnote ¹¹

Recent work in neurological science suggests that the behavior of humans may be disaggregated much further, for example, to cognitive activity as measured by brain scans.Footnote ¹² At present, however, it seems fair to equate microfoundational explanations with individual-level explanations, understood through folk or scientific models of human motivation.

It is worth noting that if the microfoundational dictum were to be strictly adhered to, a good deal of contemporary political science—including major portions of comparative politics and international relations—would be written out of the discipline. Consider development, the resource curse, the democratic peace, and geographic arguments. All these topics, as usually considered, lie far outside the bounds of microfoundationalism, as envisioned by Elster and others. The range of distal causal relationships that typically provide the focus for cross-national comparison as undertaken by qualitative comparative analysis,Footnote ¹³ the comparative method,Footnote ¹⁴ the comparative-historical method,Footnote ¹⁵ and cross-country regression analysisFootnote ¹⁶ would seem to be out of bounds for a social science constructed securely upon microfoundations.

Of course, one may hope that distal causal relationships could be disaggregated into proximal (microfoundational) causes, and explored piece by piece. Insofar as macro causal relations are built on micro causal relations, we may learn a lot by scoping down—for example, from democracy to its component parts, from structures to “cogs and wheels.”Footnote ¹⁷ Likewise, we ought to be able to aggregate up to macro levels. Elster proposes that “we can use atomic mechanisms as building blocks in more complex ‘molecular’ mechanisms.”Footnote ¹⁸ This ladder of dis/aggregation may also be understood as a macro–micro relationship across disciplines, with social sciences at the top (most macro), moving downward (more micro) through psychology, biology, chemistry, and physics.Footnote ¹⁹ Yet this presumes quite a lot. While it is true—more or less by definition—that macro-level phenomena are the product of micro-level phenomena, it is not clear that we will actually be able to reconstruct macro-level causal relationships by tracing their micro-level components. Nation-states are composed of citizens, just as a brick wall is composed of bricks, but we cannot observe a nation-state being constructed, person by person.

The microfoundational approach is basic to all physical sciences.Footnote ²⁰ One can scarcely imagine medical science without biology, or biology without chemistry and physics. Each builds on a lower-level foundation that makes sense of the causal relationships observed at a higher level—from animals to cells to molecules to atoms to quarks (and to what next we can only guess). The undoubted success of the natural sciences may partly account for the current appeal of the microfoundational project in the social sciences. However, an approach successful in analyzing nature may not be so helpful in analyzing decisional behavior. Here, molar causes are not always inferable from molecular activity.

For example, in making sense of patterns of interstate behavior, it is common to base theoretical expectations on the presumed interests of states, understood as unitary bodies.Footnote ²¹ But this theory has little or nothing to say about the individuals (citizens or subjects) living within those states, or the leaders who rule them. As such, it is not clear whether, or to what extent, the state-centered theories of international relations such as realism could be delineated in a microfoundational manner. This deficiency of realism has been extensively and persuasively critiqued.Footnote ²² Even so, with this theory as with others pertaining to the actions of large groups and organizations, the leap from individual to group behavior is a quantum leap. Different incentives, and therefore different theoretical frameworks, may apply. To say that a resolutely “macro” theory like realism should be cast away simply because its predictions cannot be process traced at the micro level would be to sweep away a good deal of the theoretical architecture of social science.

Now, let me critique my critique. I have approached Jon Elster's mechanismic project for social science as if it were a totalizing paradigm, a paradigm for what everybody in the business of analyzing decisional behavior should be doing. So viewed, it is problematic, for a variety of reasons that I have touched upon in the preceding pages. However, this is not an entirely fair reading of Elster's project—or of others in the mechanismic tradition. Indeed, if viewed more modestly—as a suggestion for how some of those engaged in the enterprise of social science should be crafting their work—it is an eminently reasonable set of proposals. Or, to put the matter another way, the mechanismic ideal is a very fine aspiration; we should all embrace it. But we must also recognize that there are situations where it is not achievable, or only partially achievable. There, a more traditional style of theorizing and investigating seems justified, where macro forces are grappled with, to the best of our abilities.

A Universal Toolbox of Explanations. The tantalizing prospect of a mechanism-centered social science, à la Jon Elster, is that we might arrive at a more or less comprehensive—and, at some abstract level, unified—set of mechanisms that could be applied to many settings (economic, sociological, political, and so on). That is, a universal toolbox of explanations.Footnote ²³

Elster's expansive oeuvre constitutes perhaps the most impressive effort of this nature to date. This oeuvre builds on, but also extends beyond, the framework established by game theory—which his books have done much to popularize, extend, and critique. However, Elster's agenda is limited in two respects. First, it is not unificationist. Rather, it is an attempt to assemble bits and pieces—“exportable causal mechanisms” (p. 9)—that seem to have broad import. Second, he is focused primarily on individual motives for action, rather than on broader incentive structures that lie outside the actor. “By and large,” he writes, “I believe that the subjective factor of choice has greater explanatory power than the objective factors of constraints and selection.”Footnote ²⁴

Within this purview, Elster's agenda is extraordinarily ambitious. What he envisions (but probably would not be so bold as to publicly announce) might well be labeled a science of human nature. This ambition lies at the heart of Alexis de Tocqueville and of most of his other books. With this in mind, he draws on literature from far-flung fields—anthropology, economics, history, philosophy, political science, psychology, sociology—as well as the classics of the Western canon. Indeed, it is the extraordinary range of his erudition that charms and enlightens (and stands in stark contrast to everyday social science, which tends to be narrowly pitched and highly technical), along with his aperçus, which are more than occasional.

Even so, Elster's work teases us with a promise that is not fulfilled. His work, like Tocqueville's, is selective rather than comprehensive, ad hoc rather than systematic. It is not clear that there is anything approaching a theory of human motivation lurking in the Elster corpus. Again, one must appreciate that in these features lies much of the enchantment of reading Democracy in America, as well as Alexis de Tocqueville. Elster is not attempting to write a textbook.Footnote ²⁵ Nonetheless, if the idea of a mechanismic toolbox is to prove its worth, the idea must be pushed further, and harder.

To start this very long conversation, let us consider a brief list of explanatory models that are commonly employed in the social sciences: (a) spatial models; (b) game theory (including bargaining, exchange, and coalition models); (c) transaction costs; (d) credible commitment; (e) collective action and coordination problems; (f) anarchic or hierarchical methods of organization; (g) delegation; (h) communication and control (Exit or Voice); (i) information and uncertainty; (j) evolution and adaptation; (k) threshold effects; (l) critical juncture/path dependence; (m) diffusion; (n) relational models (e.g., social capital, networks); (o) norms, reference groups, imitation; (p) learning, persuasion, deliberation; (q) institutionalization; (r) size; (s) organizational structure; (t) leadership; (u) levels of analysis; and (v) genetics.

We commonly invoke these mechanisms when attempting to explain social behavior. Certainly, they could be arranged in some sort of taxonomic order, rather than simply listed seriatim (as here). They might even be unified into a single framework. This constitutes a vital theoretical agenda for the social sciences. Of course, we cannot say for sure how achievable this agenda might be. It could be that plausible mechanisms will continue to multiply, rather than unify, as one looks closely at diverse contexts of human behavior. It could be that a unified field theory for social science looks a lot like the framework of social action devised by Talcott Parsons some years ago, that is, abstract to the point of disconnecting from reality.Footnote ²⁶

Even so, it seems worth the effort. With the exception of rational choice—a set of modeling strategies that Elster shows is useful but limited in its purview—we simply have not tried. Social science, and political science in particular, has focused obsessively on problems that are empirically tractable, while importing theoretical frameworks and specific mechanisms in a promiscuous fashion, without serious attempt at synthesis.Footnote ²⁷ Perhaps Elster's work will serve as the impetus—one might even say the progenitor—of that ambitious enterprise.