The AI Journal

In 1944 the International Typographical Union held a leadership election in which the incumbent lost, as incumbents in that union had lost before and would lose again, to a candidate fielded by an organised opposition party operating openly inside the union’s own structure. Seymour Martin Lipset, Martin Trow and James Coleman spent much of the following decade asking why this single union, alone among the hundreds they might have studied, had sustained genuine, competitive two-party governance for as long as anyone could trace it. Their 1956 study, Union Democracy, is cited on every syllabus that mentions Robert Michels as the case that broke his “iron law of oligarchy” — the exception that showed a sweeping sociological generalisation could be tested and found wanting, like any other. And yet Michels’s law survived the finding with barely a dent. Seventy years later the law is still called iron in the textbooks, still invoked whenever a movement, party or platform produces an entrenched leadership, still treated as essentially confirmed despite its most famous counter-example. A finding that ought to have broken a sociological law instead left the law standing almost exactly where it was. That durability stops being a puzzle once the law’s actual construction is examined. Michels did not advance one causal claim. He advanced three, welded together under a single name, and the International Typographical Union tested only one of them.

Political Parties, Michels’s 1911 study of the German Social Democratic Party and its affiliated unions, gives three distinct reasons why democratic organisations turn into oligarchies. The first is technical: any organisation past a certain size needs specialised administrators, people who master the accounts, the correspondence and the negotiation with employers or rivals, and specialised knowledge, once concentrated in a few hands, does not redistribute itself back to a rotating membership without deliberate effort. The second concerns the leaders themselves: officers who hold office for years acquire, on Michels’s account, a settled taste for the prerogatives of office and come to defend their position for its own sake, independent of whatever technical function first justified it. The third concerns the followers: rank-and-file members, occupied by paid labour and untrained in administration, feel a gratitude toward those who spare them the burden of governing, and that gratitude hardens into a general reluctance to challenge incumbents at all. Michels braided these three mechanisms into a single argument and gave the braid one name. But a technical claim about the indispensability of expertise, a psychological claim about what tenure does to leaders, and a psychological claim about the passivity of the governed are three separate empirical hypotheses, each falsifiable independently of the other two. Nothing in Michels’s own text requires all three to hold in every organisation for oligarchy to set in.

C. W. Cassinelli, writing in 1953 with the explicit aim of stating Michels’s generalisation as precisely as its language would allow, found that precision required several separate propositions rather than one, and his article is still footnoted for the attempt without ever being credited with having shown that the law was not, formally, a single proposition to refine. The multiplicity was visible from the first serious effort at precision. It has simply never been treated as a reason to stop calling the result a law in the singular, and the singular name has done real work: it lets a defender of the theory treat any successful attack on one mechanism as an attack on an isolated detail rather than on the law itself.

That is exactly what happened to the International Typographical Union. What made the ITU different, on Lipset, Trow and Coleman’s own account, was that its members were skilled printers who had come up through a shared apprenticeship system, so that the technical expertise Michels assumed only officers could accumulate was already distributed across the entire membership before anyone took office. There was no exclusive body of specialised knowledge for leadership to monopolise, because ordinary compositors already possessed it. The occupational community that resulted, with its own clubs, its own newspaper and its own social life independent of the union hierarchy, sustained an opposition party that could challenge incumbents on equal technical footing indefinitely. What the ITU case refutes, in other words, is the first mechanism, and only the first: the claim that oligarchy follows from an unavoidable concentration of technical expertise. It says nothing at all about whether officers who do entrench become psychologically attached to their position, and nothing about whether members who lack the ITU’s peculiar craft culture default to passive gratitude. A defender of Michels can concede the whole of Union Democracy’s finding and lose nothing, because two of the law’s three components were never on trial.

Philip Selznick’s 1949 study of the Tennessee Valley Authority describes a different pathway to a similar-looking outcome: an organisation’s founding goals are captured and redirected by the very local elites it coopted to secure their cooperation, a mechanism with no necessary connection to either the technical indispensability of expertise or the psychological transformation of long-tenured officers. Selznick’s study is cited constantly in organisational sociology and rarely alongside Michels, as though goal displacement through cooptation and oligarchic entrenchment through leadership psychology were unrelated phenomena rather than two of several distinct routes to outcomes Michels had already gestured at without separating. Darcy Leach’s 2005 survey of the oligarchy literature comes as close as the field has come to naming the problem outright, arguing that decades of empirical work have applied the single word “oligarchy” to structurally unlike phenomena — formal leadership monopoly, informal elite influence exercised outside any formal position, professionalised staff simply outlasting an amateur membership — without agreeing on which is being measured in any given study. Leach’s complaint amounts to an argument for exactly the disaggregation proposed here, though it stops short of tracing the confusion back to Michels’s original bundle. The confusion it documents is that bundle’s direct legacy: one word doing the work of at least three unexamined claims.

The strongest objection to treating the three mechanisms as separable is that Michels never presented them as competing hypotheses to be tested one at a time. He presented them as mutually reinforcing stages of a single developmental process, in which technical specialisation creates the conditions for psychological entrenchment, and mass passivity supplies the followers who let it happen unchallenged. On this reading, disaggregating the law does not refine it; it dismantles an integrated system into fragments Michels never intended to stand alone, and the ITU’s escape from one fragment says nothing about the interlocking whole, which was always meant to require all three mechanisms working together rather than any one of them in isolation.

The objection narrows the claim rather than defeating it. Granting that Michels imagined the three mechanisms as jointly reinforcing in the fully developed oligarchic case, an interlocking system is not thereby immune to being tested at its joints. If the mechanisms genuinely reinforce one another, removing one load-bearing member should weaken, even where it does not by itself collapse, the resulting structure — and that is what happened at the ITU: printers whose shared craft neutralised the technical-indispensability mechanism did not merely produce a marginally less oligarchic union; they sustained, across decades, the case Lipset, Trow and Coleman treated as the clearest instance of stable, competitive union governance they had found in the American labour movement. A theory whose defenders can concede that one load-bearing mechanism failed outright in a given case, while insisting the whole edifice still stands because the remaining two mechanisms were never actually tested there, is a theory that has been protected from falsification by the very ambiguity in its own architecture that this argument identifies. Naming the joints does not weaken Michels’s account of how oligarchy typically forms. It only removes the cover the unnamed bundle has given the theory against the ordinary business of being wrong.