For some time, I've been making rather a lot of the 'all social arrangements are coercive' line, particularly as a stick to beat libertarians with, with what's probably the clearest exposition here. Partly because of this comment, which points to what I must admit is something of a standing problem with the claim - that the generally pejorative nature of coercion doesn't seem to necessarily apply to social arrangements simply in virtue of being social arrangements - I'd like to give something of a genealogy of the idea, at least so far as I am concerned, which hopefully should illuminate what motivates it.
The idea comes from one way of running a critique of one of the dominant developments of post-Rawlsian political theory, luck egalitarianism, which holds that unequal outcomes are justified so long as they originate from choice, but that equality is required of outcomes which do not have their causal roots in choice, or, to make it clear why it is so named, that outcomes associated with brute luck must be equalized. Now, there are obviously a number of different ways of cashing out precisely what is meant by choice and brute luck, but the core of any view claiming to be luck egalitarian is the thought that when an agent is responsible for a particular outcome, interference with that outcome would be wrong, whereas when an outcome occurs through no fault of the agent, equalization is mandated. There may well be very serious problems in finding adequate accounts of responsibility, as the rather extensive literatures on both the freedom of the will and luck egalitarianism itself suggest, to provide anything like a sharp line between brute luck and choice that the luck egalitarians need, but I'm fairly sure that that is not the most serious problem with luck egalitarianism.
The fatal difficulty for luck egalitarianism is that it refuses to take its concern with responsibility anything like seriously enough, since in any society containing more than one individual, let alone incredibly complex ones containing millions of people, the idea that anyone is responsible for whatever costs happen to attach to any particular action is under exceptional high levels of strain. On any reasonable account of brute luck, it must surely be brute luck that the employment market values someone's labour at the level that it does, or even that there is something that we can sensibly call an employment market, since both depend on the actions of any number of others, over whom any one individual rarely has more than tangential control. If luck egalitarians are genuinely concerned with responsibility, they are stymied, because in saying that agents should bear the costs and gain the benefits that they are responsible for, they forget the surely incontestable fact that, whilst agents are often responsible for acts which have costs and benefits attached, they are not individually responsible for the particular costs and benefits that are attached. This stymies them, because it prevents them from taking a position on the legitimacy of any outcomes: if the actual value of an action is always a matter of brute luck, then the luck egalitarian claim that outcomes that are a result of brute luck are illegitimate and should be equalized does no work, because any outcome, including equalization, would be the result of brute luck.
To illustrate by example, think of a state which institutes luck egalitarianism, and compensates those who suffer and taxes those who gain as the result of brute luck. Presumably, anyone who disagreed with this situation would be entitled to complain to the relevant authorities that it was, for them, a matter of brute luck that they live in a state which has decided that it should adopt luck egalitarianism as a distributive principle. Say, moved by the force of this complaint, the authorities altered their policy to whatever distributive principle or principles that the complainant preferred. Yet, then, equally, wouldn't anybody who for some reason disapproved of that principle or principles be able to make the same complaint? No any one distribution can be justified by luck egalitarianism, because living under any one distributional principle is itself a matter of brute luck, something which agents are not generally responsible for, requiring redress.
This is where the 'all social arrangements are coercive' line comes in. The thought that no distribution is something which any one person can be responsible for, given the complexity of social interaction, leads, I think, relatively quickly, when coupled with the thought that others having control over the costs and benefits associated with a particular action is coercive, to the thought that all social arrangements are, simply because they are social arrangements, coercive. As I said in reply to Ben here, that doesn't imply an all-things-considered moral judgement, but it does mean that justification of them, to all of those who live under them, is required.