Published in Theory and Decision 32 (1992), 165-202

(Note: This is a manuscript version. Please quote only the published version.)

 

Raimo Tuomela

ON THE STRUCTURAL ASPECTS OF COLLECTIVE ACTION AND FREE-RIDING

 

I WHAT IS THE PROBLEM OF COLLECTIVE ACTION?

1. One of the main aims of this paper is to study the possibilities for free-riding type of behavior in various kinds of many-person interaction situations. In particular it will be of interest to see what kinds of game-theoretic structures, defined in terms of the participants' outcome-preferences, can be involved in cases of free-riding. I shall also be interested in the related problem or dilemma of collective action in a somewhat broader sense. By the dilemma of collective action I mean, generally speaking, the conflict between individual and collective rationality and the conflict between corresponding actions, in the sense it has been discussed in recent literature. Typically (although not invariably) collective action problems and free-rider problems coexist.
Let me start my discussion by considering what Elster (1985) has to say about the subject. First, the notion of collective action itself should be characterized. Elster defines it as follows (p. 137): "By collective action I mean the choice by all or most individuals of the course of action that, when chosen by all or most individuals, leads to the collectively best outcome." While this characterization is informative in the present context, I think that it is not appropriate as a general characterization. It may provide a sufficient condition, but it fails as a necessary condition. One reason for this is that there may not be a single collectively best outcome at all. Instead, I suggest we follow common sense and take collective action simply to be action by a collection or group of people, where these people (or at least many of them) act with the aim of achieving a common end or goal (this notion understood very broadly so as to include e.g. following norms, practices, and customs). We also require of a situation of collective action that the participants have several (or at least two) possible courses of action open to them.
Elster's above definition of collective action goes in terms of the collectively best outcome or goal. As a general account this seems too strong (compare with my above definition), but in the context of the problem of collective action and the free-rider problem we need a notion like this. I prefer to call Elster's above notion collectively or socially best collective action. Suppose now that we can make sense of the notion of individually best action. The problem of collective action can then be taken in a preliminary way to be a dilemma or conflict between collectively and individually best action, where the action required for achieving the collectively best outcome or goal is different from (and in conflict with) the action required for achieving the individually best outcome - provided the notions of collectively and individually best outcome are applicable to the context in question. Or, as we may also put it, means-end rational action realizing what is collectively best (or, more broadly, what is collectively good) is in conflict with means-end rational action realizing individual best (or good). Given this general preliminary characterization let me now consider our problem in some more detail and in part by reference to how the problem has been analyzed in the literature. Before that I would like to point out that the problem of collective action is closely related to, although not equivalent with, the free-rider problem, which will be the main concern of the present paper. My approach to the free-rider problem will be presented in Section II. The rest of the paper is devoted to a discussion of structurally different many-person games, mainly three-person games, with special reference to the free-rider problem.
2. Elster (1985), p. 139, following Schelling (1978), gives two different analyses of the notion of a collective action problem: the strong sense and the weak sense of the notion. Consider first the strong sense, which is defined by the following two clauses for a standard two-choice situation, with cooperation or contribution, C, in the production of the public good (or whatever job is at stake) and defection, D, as the choice-alternatives: 1) Each individual derives greater benefits under conditions of universal cooperation than he does under conditions of universal noncooperation; 2) each derives more benefits if he abstains from cooperation, regardless of what others do. Prisoner's dilemma is of course a typical case of a situation which should satisfy an analysis of the problem of collective action. Concentrating on the two person case here, we recall that in a Prisoner's dilemma (PD) the row player has the following preference ordering of the outcomes: DC, CC, DD, CD; and the column player has symmetric preferences. (Note that I shall in this paper formulate preference orderings in the above fashion, letting the comma - or in some cases semicolon - represent the better-than-or-equal-to -relation.) Now (assuming that the players indeed choose over outcomes rather than e.g. disjunctive bundles of them) each player gains more from cooperation than from cooperation whether or not the other player cooperates.
However, as Elster himself notes, this definition is too strong in many cases. For instance, it excludes pivotal cases of voting - cases in which an individual is in a decisive position (in which his vote will decide an issue) and in which it accordingly is rational for him to vote (see Elster, 1985, p. 139). There are also many other cases of the collective action problem which fail to satisfy Elster's strong definition. Thus certain situations which have the structure of a Chicken game, Assurance game, or the Battle of the Sexes, for instance, qualify, as also Elster notices and as we shall soon see.
Elster's weak sense of a collective action problem is defined by means of two clauses the first of which coincides with clause 1) of the definition of the strong sense. Clause 2) says that cooperation is individually unstable and individually inaccessible. Individual instability means that each individual has an incentive to defect from a situation of universal cooperation. Individual inaccessibility means that an individual has no incentive to take the first step away from a situation of universal noncooperation. Now Elster points out that there are, however, cases of collective action in which cooperation is either individually unstable or individually inaccessible but not both - for example, Chicken and Assurance games - but which nevertheless present collective action problems (though less serious ones in the case of Assurance games). An example of an individually inaccessible but stable case is given by Elster's (1985) example about erosion occurring in a fictional village - I shall call this the erosion case of the first kind. It goes as follows. On each plot being cultivated at a riverbank village the erosion can be stopped if and only if trees are planted on it and on both adjoining plots. (Planting qualifies as cooperation or contribution here.) This gives a game of Assurance.1 An example of an accessible but individually unstable case is given by Chicken - illustrated by the second type of erosion example. In it the assumption is made that erosion will occur if and only if trees are cut down on the individual plot and both the adjoining plots. (Here non-cutting represents contribution.)
Also Taylor (1987) discusses the characterization of collective action dilemmas. According to his preliminary characterization the defining characteristic of a collective action problem is roughly that rational egoists are unlikely to succeed in cooperating to promote their common interests (Taylor, 1987, p. 3). But, as I shall argue, the egoism-altruism distinction as well as other similar psychological characteristics are actually irrelevant to the problem of collective action. In his discussion Taylor correctly points out that the category of collective action problems includes many (but not all) public good problems, where we take a good or service to be a public good (if society-wide) or collective good (if relative to a specific collective, not necessarily a society) if it is in some degree indivisible and nonexcludable in the collective in question. Indivisibility (or jointness of supply) relative to a collective involves that the good, once produced, is available to any member of the collective in principle. Nonexcludability means the impossibility (or at least prohibitive costliness) of preventing individual consumption of the good, once produced. (But note that a collective action problem and a free-rider problem can arise when there is nonexcludability but not indivisibility - cf. user-sensitive recreation areas open to everybody.)
Taylor thinks that the following, final definition covers all cases of collective action problems: A collective action problem exists where rational individual action can lead to a strictly Pareto-inferior outcome, i.e., an outcome which is strictly less preferred by every individual than at least one other outcome (see, Taylor, 1987, p. 19). This definition, while on the right track, relies on an unclear notion of rational action. As Taylor points out, almost anything can be rationalized in this context, and this makes his analysis problematic.
I shall not here discuss rationalization generally but only some examples. In the case of the Prisoner's dilemma (PD) obviously the dominance of D over C (irrespective of what the others are believed to be doing) serves to rationalize it. Next consider the more interesting case of Chicken (CG) and let the two choice alternatives again be C (typically: cooperation) and D (typically: defection). What is being rationalized in the first place is (conditional) actions - or strategies - but not individual joint outcomes. For instance, even defection and the all-defection outcome in CG can be rationalized. Defection (D) can be rationalized as follows. If a player believes (with objectively good or bad reasons) that the other player (or, in the many-person case at least K-1 others - see below) will do C, then it is rational (in a utility-maximizing sense) for him to do D. This rationalizes D and derivatively the outcomes DC and DD. Of course, if DD results, that is not what our agent expected. But it is still a consequence of a rational action, even if an unexpected one. Taylor speaks of rationalizing outcomes, and this must be understood in the just mentioned derivative sense, I think. (It would of course not be rational in any sense that all the players intentionally jointly did D and arrived at the mutual defection outcome. This is not the way to rationalize the mutual defection outcome.) Note also the following: it is enough for outcome-rationalization that rational individual action in the case of some (e.g. a few or, perhaps, a great minority, as the case may be) individuals leads to a collective outcome different from the socially optimal one. This possibility arises in the case of collective action (or collective action situation) where the participants do not have symmetric payoffs.
Note that Taylor's present account does not speak about egoism (or altruism). It also had better not to do so, because we can have collective action problems also in a world of altruists. For instance, complete altruists could clearly face a Prisoner's dilemma. Accordingly we should not always take the choice alternatives C and D to represent cooperation and defection (these notions understood in their normal, full-blooded sense). For taken in this full sense it seems that we build in altruism and egoism to some extent, at least in a behavioral sense. But that is more than we want here. Accordingly, I want to emphasize here that we can have collective action dilemmas to which the altruism-egoism dichotomy (as well as the cooperation-defection interpretation of C and D) is not directly relevant (see below).
As noted, the basic problem with Taylor's analysis is that, while it works well in many contexts, there are situations in which it is not clear what individually rational action amounts to. For instance, Taylor excludes Assurance games (cf. Section II below) from the class of collective action problems, because of its several equilibrium outcomes the Pareto-preferred one is strictly preferred by everyone. He accordingly claims that individually rational action does not lead to a Pareto-inferior outcome. But I would like to claim that in some contexts also other actions in an Assurance game (AG) can be regarded as rational - possibly depending on the common history and the resulting beliefs of the participants. It should be kept in mind that even if we pertain to standard game-theoretic characterizations of rationality, individual rationality can be characterized in many different ways: in terms of dominance, the maximin criterion, maximization of expected utility, and so on. (In Section II I will present a two-person AG in which defection is the individually rational alternative in the sense of the maximin criterion.) It is worth noticing that in some game situations different background beliefs may affect the subjective probabilities in such a way that the expected utility criterion may give a different result than acting according to plain preference or maximin. Indeed, accepting the expected utility criterion as a possible criterion of rationality not only takes us out from purely structural considerations (viz. interactive preference-dependencies) but leads to a more subjectivistic notion of collective action problems (at least as long as subjective probabilities are involved). A purely subjective notion of a collective action problem would make it wholly dependent on what the players regard as individually and collectively rational in the situation in question. (I shall here resist the temptation to go far in the subjective direction.)2
In accordance with what was said above it seems reasonable to characterize collective action problems or dilemmas as conflicts between collectively and individually rational action (my phrasing is geared to player-symmetric situations):

(CAD ) Let S be an interpersonal outcome structure (viz. a game-theoretic structure) or a token of such a structure. Then S involves a collective action problem (or dilemma ) if and only if S involves a conflict between collectively rational and individually rational action, viz. if and only if there are actions or strategies C and D such that C is a collectively rational action (e.g. in the sense of leading to a Pareto-preferred outcome when chosen by all or most) and D is an individually rational action (e.g. in the sense of individual expected utility-maximization) and C and D are incompatible actions or strategies.

Structures are taken to be types here. Not all tokens of such a type, e.g. a CG-structure, need involve a collective action dilemma - this may depend on the participants' beliefs, as seen. Therefore S must be allowed to be only a token. It is worth noting that in interpersonal structures in which the participants have symmetric preference orderings - this is the case to which the phrasing in (CAD ) alludes to in the first place - D is assumed to be an individually rational action in the case of all participants. In other, asymmetric cases it suffices that D be individually rational in the case of at least some (and preferably most) participants; D must be better than C at least for those individuals. (The criterion of Pareto-preference can normally be used for defining collective rationality. It involves the assumption that the outcomes be comparable and that the strategies or actions deterministically lead to the outcomes in question; the strategies are to applied without knowledge of other players' choices. It is not necessary for our philosophical purposes to discuss these technical qualifications.)
When applying (CAD ) to particular cases it may be necessary occasionally (e.g. in the context of some Assurance games) to rely in part on the players' conceptions of what is individually and collectively rational and evaluate the applicability of the analysans of (CAD ) on that basis. But with that qualification (and small concession to "subjectivism" and understood in view of our preceding discussion, (CAD ) will work as expected at least in the context this paper, as we will mostly operate with specific numeric examples which do not pose problems. Notice that we may have conflicts between two individually rational actions which are rational in different sense, and we may have rational collective actions which are analogously in conflict (viz. there is in this case no unique collectively rational action alternative but two or more collectively rational actions which are in conflict; cf. simple coordination problems). Such cases are not (at least directly) at stake when we are discussing collective action problems.



II FREE-RIDING DEFINED

1. In this section I will present an analysis of the notion of free-riding related to a collective action X. X can be a joint action performed together by some members of the collective or it can be an action performed by a collective. (An action performed by a collective involves a joint action by the members or representatives of that collective bringing about that action of the collective; see Tuomela, 1989b, and 1991b. Even if free-riding is perhaps most naturally discussed in the context of actions assumed to be performable together, we need not assume quite that much here. What we need is that the action X is in the present context assumed to produce a nonexcludable public good in joint supply when at least K agents (out of N) participate. We shall also need to assume that there some sort of obligation for the agents to participate in the production of the good. (If X is a joint action that we jointly plan or "we-intend" to perform together the required kind of obligation, indeed a joint obligation, will exist; cf. Tuomela, 1991a, 1991b, Chapter 3.3) When the public good has been produced, we say that X has been performed successfully. We take the action alternatives to be C (doing one's part of X) and D (defecting from doing one's part of X).
There is an extensive literature on free-riding. I shall not survey it here (see e.g. Hardin, 1982). My aims will be twofold. First, I will present an analysis of what is basically involved in the free-riding problem. Secondly, I will discuss several kinds of many-person situations (actually three-person ones) in which free-riding can take place, and I shall characterize these situations in terms of the various kinds of interpersonal or social control involved (cf. Tuomela, 1985). This section will be devoted to a characterization of the key features of free-riding, and my account will emphasize the role of the potential free-rider's intention to free-ride (and I shall below draw from Tuomela and Miller, 1991).
Before presenting my analysis of free-riding it is appropriate to say a few words about the production function involved in the provision of the public good in question. Our analysis is meant to be compatible with any kind of production function - as long as it allows the good to be produced for at least some argument values. The production function will be of the general form x=f(y,z,...) where x represents the public good in question and y the number of cooperative action tokens. The other arguments (z,...) can deal with the qualitative features of the situation, taking care that the right kind of variety of actions is present, if the production of the public good is based on a division of tasks and labor. For simplicity's sake, we shall below suppress these other variables and deal with a simple function of the kind x=f(y), assuming the variable y to be of the right kind so that it measures the number of actions, of the right kind, needed for the production of any amount of the good x in the first place and, in the second place, for the production of an increment in the variable x. The only assumption about the form we shall really need in this paper is that there is a number, say K, which is an allowed value of the variable y such that f(K)=x(0), where x(0) represents a token of the minimally existing quantity of the public good X. f may be a monotonously growing (discrete or continuous) function of the number of cooperative action tokens or it may be S-shaped so that f(K+N)>f(K) for some N but yet f(K+N+P)<f(K+N), for some positive number P; and so on; the possibilities are many and various, as has been documented in the literature (cf. Hardin, 1982, Hampton, 1987).
Fortunately, we do not here have to rely much on any particular assumptions about the form of the production function. It is possible to relax even the idealized assumption that K participants are needed for the production of a minimal amount of public good. We can do with the vaguer condition that there will be sufficiently many contributors to provide at least some amount of the public good.
Let me now propose an analysis of an agent's intention to free-ride. This analysis will of course also clarify the notion of free-riding, for a person's intention to free-ride will under favorable conditions (including that his relevant beliefs mentioned in the analysans, are true and that he does not change his mind) lead to his free-riding intentionally. Intentional free-riding may also come about due to a person's conditional intention to free-ride, provided he appropriately deconditionalizes the intention. (I shall also comment on such conditional intentions below.) Arguably, these are the only ways intentional free-riding can come about (and in saying this I include also so-called action-intentions or "endeavorings" among the intentions just mentioned). Following Tuomela and Miller (1991), my analysis of the notion of an intention to free-ride in the case of full-fledged and adequately informed members of G can be rendered as follows (I phrase it to fit a good X which exists to a maximal degree once it has been produced):

(FR ) A member A of a collective G intends to free-ride relative to a public good produced by a joint action X if and only if
1) A intends to defect (viz. not to contribute or do his part of X).
2) A has a belief to the effect that the joint action opportunities for the performance of X will obtain, especially that at least K members (or a sufficient number of members required for the provision of the public good produced by the performance of X) contribute (or do their parts).
3) A believes that he ought to participate in the production of X and that there is (or will be) a mutual belief among the full-fledged and adequately informed members of G to the effect that the joint action opportunities for the performance of X will obtain, and to the effect that each full-fledged and adequately informed member ought to contribute.
4) A believes that he will gain more from defection (doing D) than from contribution (doing C) if at least K agents contribute, viz. if at least K agents out of N do C, K being the minimal number of agents capable of jointly performing X.
5) A believes that the outcome resulting from all the agents contributing (doing C), is better than the outcome when all defect (do D).
6) A believes that his defection involves a cost (possibly nil) to the contributing members of G.

Of our clauses 1) gives the basic intention in question here. An intention to defect is an intention to omit doing one's share in a situation where the provision of a public (or, as I shall synonymously say, collective) good is at stake. It is, however, a weaker intention than the intention to free-ride. Clauses 2) and 3) are conceptual presuppositions of an intention to defect, while 4), 5), and 6) are conceptual presuppositions of an intention to free-ride but not of an intention to defect. As I have in part stipulated the difference between an intention to defect and an intention to free-ride to be this, the previous claim does not seem to need further defense. That something like clauses 2) and 3) are indeed needed as presuppositions should be clear from the idea that we are concerned with defection in the case of a situation of the provision of a public good. Clause 3) requires that A believes that he is expected to, or that he should, contribute to X. This is an obvious idea based on a group obligation stemming from his group's, G's, commitment (intention) to produce the public good related to X. Clause 3) also says that he believes that there is at least some degree of consensus among the members of G that the relevant members of G ought to contribute. (Note that the intentions to defect and the intention to free-ride must be distinguished from the intention to choose the alternative D. The last mentioned intention does not have to satisfy any of the clauses 2)-6).)
4) states a central idea involved in free-riding: Once the nonexcludable good has been provided a free-rider can enjoy it without participating in its production costs. (Recall that we are speaking of gains as net gains, viz. gross gains less costs, in this paper.) As the first clause formulates a quite obvious requirement, widely accepted in the literature on the topic, it does not seem to need extra defense here. Let me note that when I later in the paper speak of a free-rider effect I mean just the condition here formulated by 4).
Clause 5) formulates the idea that - to put it somewhat misleadingly - joint action is preferable to acting alone, and serves to define a sense of collective rationality. This clause can be explicated in different ways. The standard way, to be accepted in this paper at least for typical cases, is the distributive sense called Pareto optimality. Considering a two-choice situation with C and D as the alternatives, Pareto optimality means that the outcome resulting from everyone's doing C is preferred by everyone to the outcome resulting from everyone's doing D. But I would like to emphasize that this formulation is somewhat idealized. This is because the production function might be so shaped that for some value of K it might be the case that f(K) > f(K+R), where R is any positive number. In this kind of case of "crowding" we should replace 5) by
5') A believes that the outcome resulting from all those agents' contribution (doing C) who are ought (or are normatively expected) to contribute in G is better than the outcome when they all defect (do D).
The notion of someone's being normatively expected to contribute will depend on the believed shape of the production function (and will typically be connected to the accepted division of labor in G). In spite of 5') being more realistic than 5) I will below continue to use the latter in our discussion, to keep matters simple.
Clause 6) introduces the idea that free-riding may be costly to others. It need not be so, but on the other hand it can even be so costly as to bring some of the cooperators below the universal defection line (cf. Pettit's (1986) distinction between free-riders and foul dealers and see also Tuomela (1988)). We can in fact speak of free-riders in three senses. First there are the free-riders in the literal sense. They are ones who do not (at least ideally) impose any costs to the contributing members of G. Secondly we have free-riders who do impose some cost on them; but the cost is "tolerable" in the sense of leaving some gain to the contributors as compared with the case where no good or no amount of good is produced. This second kind of free-riders might be called parasites. And thirdly we have foul-dealers.
Note that (FR ) involves only a subjective notion of the free-riding situation in that clauses 2)-6) only concern the agent's beliefs, which are not required to be true. But if they are true then (FR ) of course serves to characterize the intention to free-ride in an objective social free-riding situation, and an action satisfying this intention accordingly will be a free-riding action.4
We may also discuss conditional intentions to free-ride. In such cases the intention in the first clause is some contingent condition, e.g. that at least K other members contribute. Let us now consider the mentioned situation: the condition that at least K other members contribute may be i) sufficient, ii) necessary and sufficient, iii) necessary, or iv) it, together with some other conditions - left implicit in clause 4) - is necessary and sufficient. Consider i), the case where A has the intention to defect if at least K others contribute. This allows for the possibility that A contributes if fewer than K others participate (and in particular in the critical case where K-1 others contribute), and A may then also contribute on the basis of his we-intention to do X. (At least it is rational for A to form the we-intention to cooperate, for in this case cooperation simply is more valuable to him than defection.) Case i) represents a typical case of free-riding. With fewer than K other participants, the agent may often decide on the basis of the costs of participation and the expected gains of participation whether he will participate, and the present possibility leaves open the results of such deliberation.
The case ii) where A intends to defect if and only if at least K other members participate is a strong one. It requires A to participate when fewer than K others participate. If intentional participation in a collective or joint action indeed requires the we-intention to perform that joint action, as we have required, we have here the somewhat special combination of we-intention to cooperate (when fewer than K other persons participate) and defect (when at least K other persons participate). In the third case iii) no commitment to defection follows from the fulfillment of the condition, viz. that A defect only if at least K others contribute. This case involves that A is committed to participation if fewer than K other persons participate and he is needed (this is the case if K-1 others participate - and in our present setup only then). If at least K others participate he can decide whether to cooperate or to defect; in the case of a rational agent that clearly depends on the production function of the public good in question and of course on which particular values of its argument the agent will at that moment have to consider. In any case, this kind of case seems typical in real life. Also case iv) seems rather typical - in this case A still requires that some further conditions hold before forming the intention to defect.
Do the above cases i)-iv) all qualify for representing free-riding as we normally understand this notion? The answer is positive in the case of i) and ii), for in their case the agent satisfying our analysans of (FR ) will indeed under normal circumstances intentionally free-ride (precisely with the free-rider's reason of gaining something due to what the others do) if the collective good is provided by the others. But also iii) and iv) seem acceptable, keeping in mind that we are dealing with a disposition to free-ride. In fact iv) is not problematic here at all. Case iii) is problematic and indeed unacceptable only if A believes that there cannot be any further conditions, which, when added to the present one would make him free-ride. For then we presumably would not say that A is disposed to free-ride - and it is just this disposition we are trying to analyze here.
2. Free-riding situations can be illustrated in part in terms of game-theoretic structures. The "classical" account of free-riding, to my knowledge originating from Olson (1965) and Hardin (1971), connects free-riding exclusively to Prisoner's dilemmas. But, as various authors (Taylor and Ward, 1982, Taylor, 1987, Hampton, 1987, and Tuomela, 1988) have shown, there are actually several game-theoretic structures that are relevant for studies of free-riding. In this paper I shall continue this line of thought and argue that depending on the situation at hand, elements of conflict (with exchange as its special case) and coordination will both be present to varying degrees. I shall start by discussing some simple game-theoretic situations in which free-riding can be exemplified. In this section my numeric examples will concern only two-person two-choice situations. Later in the paper I will take up numeric examples also of three-person games as well as consider some additional questions related to many-person situations.
The aim of this section is to show that conditional intentions can be connected to various games in the context of free-riding. The game of Chicken (CG) is of special interest here, and I shall consider a numeric example of two-person Chicken. Many-person games are trickier to analyze and I shall discuss them later in the paper. I shall here start with the single-shot case and only later comment on the dynamic case. For the time being the reader can think of a single-shot game as representing the structure of a situation related to a particular argument value of the production function f.
Let us now consider the following numeric example of Chicken, where the ranking of the outcomes is given for our reference point player 1 (the row player):

C

D

C

3, 3

2, 4

D

4, 2

1, 1

(CG): DC, CC, CD, DD


This game is player-symmetric - as are most of the games to be studied in this paper. The notation DC, CC, CD, DD simply means that player 1 ranks the outcome DC (resulting from his defection and the other's cooperation or contribution) as the best - all things considered, while CC is the second best, CD the third best, and the mutual defection outcome DD ranks as the worst; equal preferences are allowed by our ranking relation. (Symmetrically, player 2's ranking is CD, CC, DC, DD.) There is a free-rider effect in CG. By a free-rider effect I mean that a player profits more from defection than from cooperation given that at least K-1 other players participate (in the two-person case of course K=2). It is quite clear that there is a collective action dilemma in the case of CG. For, first, the universal cooperation outcome is Pareto optimal relative to the universal defection outcome. Secondly, it can be rational for a player to defect and more rational to defect than to cooperate. For instance, defection may maximize a player's expected utility. To take an example, let player 1's subjective probability for 2's choosing C be 0.8. Then the expected utility of C for 1 becomes 2.8 and that for D becomes 3.4. (A still simpler possibility is that 1 flatly believes that 2 will choose C; player 1 might have the well-grounded belief that 2 is a Kantian unconditional cooperator.)
In CG precommitment plays a crucial role. And precommitment here means forming an (absolute or conditional) intention to defect (rather than to cooperate). Here the social outcome or the public good must be brought about or obtained. One participant may be able to do it. Or in some cases all the participants must be involved. In this kind of situation it is typically reasonable for a person to form a precommitment (intention) to defect. This precommitment may be absolute or conditional. It is rational for a person to form an absolute commitment or intention if the person believes that his announcing this precommitment will with a high probability guarantee that the others will take care of the job, viz. provide the good by performing the joint action in question. Alternatively this intention to defect may be conditional, the condition here being something thought relevant to the production of the good. (This game obviously can represent free-riding; cf. our above remarks and (FR ).)
This kind of conditional intention to defect may, however, be combined with a conditional we-intention to cooperate . That is, a player (participant) can be in reserve: he will do C or participate in the provision of the public good if the others are not going to do it (see Tuomela and Miller, 1991, for a discussion of the notion of standing in reserve). In view of the low payoff (relative to the other outcomes) of the all-defect outcome it is of course rational to form this conditional we-intention. (In some cases the discussed kind of conditional intention to do C can be only a mere personal intention to do it, perhaps when an agent thinks that it is likely - even if perhaps far from sure - that the others or at least K-1 of them will cooperate.) Note, too, that an agent may first think that his announcing the precommitment to defect will make it practically certain that the others will provide the good, and that he as a consequence genuinely forms the absolute intention to defect. But this is compatible with his later changing his estimate of the likelihood of the good being provided. He may come to think that it is at best probable that it will be provided, and accordingly he may change his absolute intention to defect to a conditional one, with some relevant condition.
There are also other games besides Chicken which are centrally relevant to our analysis of conditional we-intentions, standing in reserve, and to free-riding. Such a game is most notably Prisoner's dilemma, but (at least in some limiting cases) also the Assurance game, Imitation game, Battle of the Sexes, as well as various turn-taking games are relevant. Below I give numeric examples of these games, all of which have at least some relevance to free-riding. The following game is the Prisoner's Dilemma:

C

D

C

3, 3

1, 4

D

4, 1

2, 2

(PD): DC, CC, DD, CD

Clause 1) of (FR ) can here be taken to be satisfied in the case of rational players in the case of one-shot games, for defection is 1's sole equilibrium strategy, and thus it is rational for player 1 to form not only the conditional intention to defect but in this case even the absolute intention to defect. The presuppositional clauses obviously can be satisfied in PD. Conditions 4) and 5) of (FR ) are clearly satisfied (taking K=2). Also clause 6) is obviously satisfied, as player 2 will lose 2 units by player 1's (viz. row-player's) defection. PD obviously then can represent a case where a person intends to free-ride, and it clearly involves a collective action dilemma. It worth noting, however, that in the case of PD and CG one cannot - except in their limiting cases giving the second player the same payoff in the outcomes CC and DC - be the kind of free-rider that imposes no costs on the other.
Next follow Assurance game (AG) and Imitation game (IG), which also have some relevance to the issues discussed:

C

D

C

4, 4

1, 3

D

3, 1

2, 2

(AG): CC, DC, DD, CD

C

D

C

4, 4

1, 2

D

2, 1

3, 3

(IG): CC, DD, DC, CD

In AG D is a maximin strategy. Furthermore, D can maximize a player's expected utility. Let player 1's subjective probability for 2's choice of C be relatively low, for instance 0.3. Then the expected utility of C for 1 will be 1.9 and that of D 2.3. Thus D is a rational choice also on this ground. (Changing the utility of DC for 1 from 3 to 4 and that of CD for 2 similarly would make this new AG a borderline game and give it a zero rather than negative free-rider effect in the defined sense. Then, if 'more' in clause 4) of (FR ) is taken to allow for zero increase, (FR ) would be satisfied.) So we see that while the above two-person AG does not strictly speaking involve the free-rider problem, it still involves a collective action dilemma. The same general conclusion holds for IG. D is a maximin strategy and under some conditions it analogously with AG has D has a strategy maximizing expected utility. (As we shall see, in the many-person case both AG and IG can involve a free-rider effect, depending on how exactly the generalization to the many-person situation is made.)
Consider still the following two-person Threat Game (TG):

C

D

C

3, 3

2, 3

D

2, 1

2, 2

(TG): CC, DC, CD, DD (player 1)

CD, CC, DD, DC (player 2)

In this asymmetric game there is no free-rider effect, but the game still poses a collective action dilemma (recall my comments related to (CAD )). The last mentioned claim is true because in the case of player 2 D is the dominant strategy. Thus rational utility maximizing action on the part of both players leads to the CD outcome rather than the Pareto optimal cooperative outcome CC. In our example CC is collectively somewhat better than CD. Thus even if we changed our criterion (CAD ) to allow that different strategies be allowed for individually rational or best ones in the case of differen agents, still the cooperative outcome would be better in some sense. (Note that if we made such a change, then we would have to account for the fact that both CC and CD are Pareto optimal - thus CC would have to be regarded as collectively better in another sense.)
Next we shall consider coordination games (or games with a high coordination component). Here we must interpret the situation differently. Above we have considered conjunctive collective actions. In them each participant or (or at least K participants) must contribute before the collective action comes about. Now we must interpret the collective action disjunctively and give up the interpretation of C and D as contribution and defection, respectively. To see what this amounts to consider the following simple Turn-taking game:

C

D

C

0, 0

0, 2

D

2, 0

0, 0

(TT): DC, CD, CC, DD (player 1)

CD, DC, DD, DD (player 2)

Here the outcomes DC and CD can be taken to make the collective action come about. We can say that the result of the collective action consists of the disjunction DCvCD. As each player only can make one choice at a time this disjunction will be an exclusive one. The turn-taking aspect then comes into picture if both players want to have some gain from their action, if e.g. an even distribution is wanted. Then the players obviously have to alternate between DC and CD and in this sense coordinate their actions. Note that we do have a collective action dilemma in the defined sense here, for D is the dominant strategy in the case of both players, and using it would result in the outcome DD and thus to no collective action in the meant sense. In the two-person case there is no free-rider effect, but in the many-person case even that becomes a possibility if only a subset of the players is needed for the collective action to come about.
Finally consider the following Battle of the Sexes game, which is primarily a coordination type of game (preference-ranking given for player 1):

C

D

C

2, 2

3, 4

D

4, 3

1, 1

(BS): DC, CD, CC, DD

We may interpret this game in two ways. First, the above disjunctive interpretation can be used by defining DCvCD to be the disjunctive result of the collective action in question. As in the previous game CD and DC are both equilibrium points, and there is no other way to come to a solution than by some kind of agreement to select one of them (and stay with it) or then to alternate. Note that there is a collective action dilemma here. If, for instance, player 1 believes that 2 will (at least with a high probability) choose C he will choose D, and vice versa, and DD will result.
(There is also another relevant interpretation of BS available here. Suppose only one of the players can represent (or volunteer for) the dyad. How is the representative to be chosen? The players must obviously coordinate either on CD or DC, if D means representing (or, more broadly, actively contributing) and C not doing it. We shall later comment on this aspect of collective action dilemmas when discussing many-person games in detail.)
As discussed in Tuomela (1984), pp. 141-144, there are many other kinds of joint and collective actions besides conjunctive and disjunctive ones. Serial collective actions, where the participants must do their shares in a certain order for the collective action to come about, represent one such type. They seem to require the kind of multi-play treatment (e.g. in terms of an iterated Prisoner's dilemma game) to be commented on below.
Let us still return to our (FR ). We may relax it by liberalizing the first condition so that not every participant but perhaps instead most of them satisfy its requirement 1). Then various nonsymmetric games also qualify for characterizing the structure of the free-riding situation. One such game is the following: Agent A1 plays Chicken with the utilities mentioned in our earlier Chicken game, and A2 has the following preference ranking: CC, DC, DD, CD.
To conclude, let me finally comment briefly on the dynamic case. We have been assuming up to now that each participant acts only once, viz. produces one appropriate action token reflected in the value of the argument y in the production function; or at least we have assumed that he behaves as if he acted only once (even if in fact he may have produced several action tokens at different times). Depending on what the production function is like at a given argument value, it may be represented by different games for different such values (see Hampton, 1987, for a discussion). One typical dynamic situation is to have a long succession of Prisoner's dilemmas - all values of y may even correspond to such cases. Or we may have Chicken games for some values of y and Prisoner's dilemmas for other values and no dilemmas for some values (e.g. when there are exactly K-1 other contributors in the case of a particular value of y, viz. when the value of y is K-1). It all depends on the production function in question what kind of situations must be faced.
The best hope for finding rational cooperative solutions to social action dilemmas of course relates to the case of iterated games. For then various psychological "continuity" factors, such as reputation and trust, can be argued to affect the strategic situation. Even without such factors solutions have been found - see e.g. Taylor (1987) (also cf. Axelrod, 1984). For instance, Taylor (1987) shows that under certain specific (and, indeed, quite strict) conditions it is rational (in the utility maximization sense) for agents (even egoistic agents) constantly to cooperate. (In fact, he speaks for the importance of conditional cooperation strategies, which indeed seem closely related to conditional we-intentions.) Tit-for-tat is one such strategy, which under certains conditions will result in an equilibrium, indeed a coordination equilibrium (one in which no one benefits from unilaterally changing strategy). But in real life the conditions required may be hard to satisfy: the solutions are in many ways problematic. In fact here the situation is often plagued by the problem that there are too many cooperative or partly cooperative equilibrium-solutions and the problem that too much problem-solving capacity and rationality is required from the agents.
While I shall not in this paper seriously discuss how to solve the free-rider problem(s) or the collective action dilemma, let me here point out that the feedback based interaction possibilities between the agents should be utilized whenever feasible. At least in small groups this kind of interaction seems central. Nevertheless studies such as Taylor's otherwise first-rate investigations fail to do that. The players are treated by him as independent except that the discount parameter defined over the trials reflects some of the agents' expectations about each other's behavior (but that is not feedback-based interaction.) A proper discussion of the dynamic case will be left for another context.


III COORDINATION VERSUS CONFLICT IN SELECTION SITUATIONS

When a collective performs an action, say X, this involves that there are some operative members of the collective who jointly perform something which brings about (the result event of) X (see Tuomela, 1989b). But how are the operative members to be selected in a rational way? And how to allocate shares or parts to them? How to get them to perform their parts? These are three interesting problems worth investigating.
In the case of formal collectives such as organizations there are typically rules which make it clear who can act for the collective. But in the case of many informal groups, at least, it may be a problem who can and will act on behalf of the collective. To be sure, many collectives (e.g. crowds) act in virtue of all their members suitably acting, but here we will not be interested in those cases. Rather we will be concerned with cases like this: 1) Our informal running team will receive a trophy after having won a race. But only one of us is allowed to represent the team. Everyone of us would like to be the person to receive the trophy on behalf of the team. Who will be the one to do it? 2) Our army patrol needs one or two men to check the wires when crossing the enemy line. We all dislike the job. How shall we choose the man to do the job, supposing that none of us has the power to order anyone for the task?
Another example (related to case 2)) is Hume's famous meadow draining case:
"Two neighbours may agree to drain a meadow, which they possess in common; because 'tis easy for them to know each other's mind; and each must perceive, that the immediate consequence of his failing in his part, is, the abandoning of the whole project. But tis' very difficult, and indeed impossible, that a thousand persons shou'd agree in any such action; it being difficult for them to concert so complicated a design, and still more difficult for them to execute it; while each seeks a pretext to free himself of the trouble and expense, and wou'd lay the whole burden on others." (Hume, 1965, p. 538)

Here the meadow has got to be drained, and at least two of us, let us suppose, are needed for the job. Who will do it? Let us now discuss this problem (and related ones) involving both a coordination and a conflict (or exchange) aspect. In the case of wanted jobs the members of the collective face a coordination problem. Often this problem seems to be of the Battle of the Sexes (BS) type, occasionally it may have the structure of an Imitation game (IG), an Assurance game (AG), or a Turn-taking game (TT) - and this list could be extended to cover other games involving a coordination problem. But in the case of unwanted jobs there is conflict involved, and the game of Chicken is often appropriate for representing the structure of the situation. Also PD seems possible in the case of unwanted jobs. In the case of CG our agents regard it as crucial that the job will be done, while in the case of PD that is less important than to avoid being a sucker (the sole contributor - or, more generally, one of few contributors). (Conflict here involves that the players want the others to act against their - the latters' - preferences.) Considering first, for simplicity, the two-person case, let me present the preference orderings for the Battle of the Sexes (BS), Imitation game (IG), Assurance game (AG), Chicken (CG) as well as the Prisoner's Dilemma (PD) - in the case of player 1 (the games are player-symmetric):

IG: CC, DD, DC, CD
AG: CC, DC, DD, CD
BS: DC, CD, CC, DD
TT: DC, CD, CC, DD
CG: DC, CC, CD, DD
PD: DC, CC, DD, CD

There is a central difference between Chicken, which is a game of conflict, and BS, which is a game of coordination with some conflict of interest. Chicken games have only Nash equilibria, while BS games have coordination equilibria.5 IG seems to be pertinent when the players consider doing something together to be important, while what is being done together is not so important. AG can come about when an agent agrees to participate as long as the other one will - but dislikes doing the job alone. TT is often at stake in the case of disjunctive collective tasks.
Consider now the meadow draining situation (and analogous cases). In the two-person case we get the result that we are dealing with CG, supposing - at least seemingly contra Hume - that even one player alone can drain the meadow. Draining the meadow is a cost for someone and thus CC represents a cheaper alternative for player 1 than CD. Consider next the general many-person case, illustrated, for simplicity, on the basis of the three-person case. Suppose thus that there are three participants and that, for a change, at least two persons are needed for the job. Here we get for player 1:DCC, CCC, CCD, CDC, DCD, DDC, CDD, DDD
(Strictly speaking, we only have DCC, CCC, CCD=CDC, ...., DDD, allowing for the other triplets to come in any order.) This is a CG (and so is the case where one alone is taken to be able to drain the meadow, as the reader can easily check). However, note that if I count my being the only one to contribute as a very costly thing (assuming that two persons are needed for the job), DDD goes before CDD and we get not quite a Prisoner's Dilemma but what has been called a PD-structured situation (Ullmann-Margalit, 1977, p. 23). (This situation is not a full-blown PD because each player still prefers contributing to not doing so if exactly one of the other two participants contributes.)6
The selection of operative members accordingly is sometimes a CG (or even PD-resembling) problem and sometimes a coordination problem such as BS. When specialization and differentiation of tasks are centrally involved we typically have a coordination type of problem, e.g. BS or perhaps turn-taking. Consider playing a violin sonata with piano accompaniment (in the case of a specific violinist and an accompanying pianist). This can be treated as a BS problem. But if there are several violinists and pianists we must first choose one of each, and this often (to be precise, when the job is wanted) involves another coordination type of problem.
An interesting feature about selection situations is that there is a different kind of problem when the job is wanted as compared with a case where it is hated. Let us consider a case where the job is wanted and where there is a surplus of candidates for it. We consider the problem of selecting one representative (or more representatives if you like, but here we shall for simplicity only consider the case of selecting one). We let there be three players who want to elect a king or leader or simply a person to represent the collective (or whatever similar task you can think of). Supposing that everyone wants the job, we face a coordination problem which is of the BS type. For the preference ordering can be taken to be (in the case of player 1): CDD, DCD, DDC, ... (the other alternatives are all bad). This game is a BS game, but let me emphasize that C now is interpreted as the "good" action relative to the other players' doing D. In this sense we have now reversed the roles of C and D here, but this does not affect the basic nature of the game. The three outcomes mentioned above will be equilibrium points (in the sense that it pays for no one to switch to another alternative given that the other two stick to their present ones).
Let us now change the situation so that if the players agree on a specific triplet, they may form a triumvirate, viz. select CCC. Now CCC will be fourth in their preference ranking and it will also be an equilibrium (for a change by one player would give a triplet with two C's which is worse). Put more exactly we now have the following rankings for the players:

1: CDD, DCD, DDC, CCC, DCC, CCD, CDC, DDD
2: DCD, CDD, DDC, CCC, CDC, CCD, DCC, DDD
3: DDC, CDD, DCD, CCC, CCD, CDC, DCC, DDD

(The ranking obviously goes from left to right, the leftmost outcome being the most preferred.) Of course, the order of the fifth, sixth, and seventh triplet in these rankings could be different, too. To get a numeric example, we assign the following numbers to our triplets: CDD=<4,3,3>, DCD=<3,4,3>, DDC=<3,3,4,>, CCC=<2,2,2>, DCC=<1,1,1>, CCD=<1,1,1>, CDC=<1,1,1>, DDD=<1,1,1>. A still more realistic possibility would be to think that the triumvirate-alternative CCC is the second best so that e.g. player 1's ranking becomes: CDD (4,2,2), CCC (3,3,3), DCD (2,4,2), DDC (2,2,4), DCC (1,1,1), CCD (1,1,1), CDC (1,1,1), DDD (1,1,1).

However, if nobody wants to be the king (or the representative or whatever operative member) the players face a CG or a PD or something like that. For then we have the same situation as with meadow draining. Accordingly, we have found something informative to say about the selection of representatives or - more generally - operative members both when the job is wanted and when it is not wanted.
As I have above discussed the selection problem in the rather general case of three-person games, it is easy to do it for the general N-person case (N>3). I shall not bore the reader with that nor with the equally obvious generalization of the number of operative members exceed one agent. But let me here emphasize one feature of N-person games. In case of step goods which require at least K persons to participate there will be a selection problem involved for an agent when N is larger than K. If N=K, our reference point agent believes that he must participate, given that the K-1 others participate. Or more generally, whenever our agent thinks that he is needed to produce the good, there will be no selection problem (or surplus problem).
Let me conclude this section by a couple of general remarks. We have discussed the selection of producers of a good and noted that it may involve not only a coordination problem but conflict (e.g. in the sense of CG), too. This is contrary to what Hampton (1987) at least seems to think. She conceptualizes the problem of the provision of a public good as follows: first there is the coordination problem of selecting the producers of the good, and after that there is the problem of how to get the selected producers to act - and here CG may be appropriate. But why should we analyze it so? In view of what was said above, it seems that we should be prepared to face all kinds of conflict/coordination combinations in selecting the producers; and sometimes the selection of the producers may motivationally involve or determine that they indeed will produce, so that we only need to solve one game. On the other hand, as said in the beginning of this section, we may also have the three problems of 1) selecting the operative members (using our earlier terminology), 2) allocating tasks to them and 3) getting them to perform. In solving problems 1) and 2) we may deal with situation structures exemplified by e.g. BS, AG, CG, PD - or there may be a simple unproblematic situation involving no problem of strategic interaction at all. (Hampton, 1987, only speaks of BS, AG, and CG but not PD in this connection.) In the case of problem 3) more conflict may be involved, but all of the mentioned games still represent possibilities also in this case. All kinds of combinations of coordination and conflict seem possible in principle, and, it seems, we cannot a priori exclude any interesting types of structures.


IV THREE-PERSON GAMES WITH A COLLECTIVE ACTION PROBLEM

1. Two-person games can be used to give guidelines as to how to characterize many-person games. Thus, in the case of the Prisoner's dilemma we go about as follows. As is usual, we accept from the two-person case that dominance of defection and Pareto-optimality of the joint cooperative outcome are properties to be preserved (cf. Pettit, 1986). We thus get in the case of n agents with the two choice-alternatives C and D the following partial ranking for agent i: C1,...,Di,...,CN; ... C1,...,Ci,...,CN; ... D1,...,Di,...,DN; ... D1,...,Ci,...,DN. Altogether we have M=2N choice-alternatives, and here we fix the order of four of them and make the first one of the aforementioned alternatives the first also in the total ranking and the last one of them the last in the total ranking. The rest of the 2N-4 alternatives (for instance, C1,D2,...,Di,...,Cn) go somewhere in between the first and the last alternative. In the case of Chicken we similarly can get on the basis of the two-person game the following partial ranking for i: C1,...,Di,...,CN; ... C1,...,Ci,...CN; ... D1,...,Ci,...,DN; ... D1,...,Di,...,DN.
The three-person case can of course be taken to represent the general N-person case in many contexts (except that three persons do not amount to more than a small group, not e.g. a mass or crowd) and as the number of rankings grows rapidly with growing N I will here concentrate on three-person structures. (As far as I know there is no previous systematic work available on the three-person case - and note, too, that the technique to be described of course applies to any finite number N.) I will present a methodology or technique one may use in the study of three-person structures. I will be less concerned with the strategic and other features of the resulting structures. For lack of space, my general remarks on three-person games concentrate on CG (and PD); but other relevant games (such as AG and BS) can be discussed quite analogously. The reader may be reminded here that by their very characterization CG and PD - except for some borderline cases of these games - involve a free-rider effect of some kind and a collective action dilemma.
Let me now proceed to a discussion of three-person Chicken. (What I will say applies by rather direct analogy to PD, too.) We must here obviously deal with 23 =8 possible choice-combinations here. We know that DDD ranks lowest and CCC somewhere between the highest and the second lowest alternatives. For instance, DCC will be preferred by player 1 to CCC; and he will also prefer CCC to CDD, and CDD to DDD. But there are a great many total rankings compatible with this. In fact, given certain plausible alternative assumptions about the provision of public goods, there are 20 possible CG-structures in the context of the provision of public goods. (Altogether, if we - as is natural - glue CDD and DDD together and let nothing go in between there will be 86 possible CG-structures satisfying merely the constraints on many-person Chicken games stated above.) Let me now present the details.
We must assume in all Chicken games that two persons be able to provide the good in question - or at least to provide a substantial amount of it which makes the third person's free-riding possible (so that it pays for him to defect rather than cooperate). I shall divide the Chicken games into three categories depending in part on how much 1 person and how much 2 persons can contribute toward the provision of the good. To characterize the first class, suppose first that one contributor can provide enough of the good for it to become free-ridable. In this category of Chicken games the first person ranks highest the alternatives DDC, DCD, and DCC, in some order. These give us 3! (=6) permutations. Thus above CCC we have those alternatives in which the first person defects and one or two of the other players contribute. After CCC we have those combinations in which the first person's cooperation is combined with one other person's contribution, viz. the alternatives CDC and CCD, or with the other two persons' defection, viz. CDD. We get 3! permutations, and altogether there are then 36 Chicken games of this kind. Let us in addition make the natural assumption that CDC and CCD give more of the good than what the first person alone can provide. We should then put these two combinations above CDD. Given this, we have in our first category 12 Chicken games satisfying the assumptions made.
In the second category we have the possibilities in which, again, one contributor can provide a sufficient, free-ridable amount of the good. In addition we assume in this class of games that a person does not much mind if one other person defects but does mind being himself the sole contributor. Here we put below CCC only CDD and DDD. Thus DDC, DCD, DCC, CDC, and CCD go above CCC in the ranking. There are 24 possibilities here. Of them 4 are structures in which the first person values free-riding more than contribution, and we will include only those.
Finally we have as our third category the class of cases where two persons' contribution is required for a free-ridable amount of the good. Therefore only DCC goes above CCC. Furthermore, we assume that two persons will provide more of the good than one person will (so that CCD and CDC rank above DDC and DCD); and CDD comes last except for DDD. There are four structures of this kind. Thus, we have in our three categories altogether 20 structures of CG which are compatible with the assumptions made. One can of course make many other kinds of typologies as well, depending on what assumes of the situation. I shall not here go into a deeper discussion.
Three-person PD can be handled completely analogously. The only essential difference of course is that DDD always comes after CCC so that we only have to consider the combinations which go above CCC and those which go below DDD. In partial contrast to CG we may want to consider cases in which some alternatives go in between DDD and CDD (the lowest ranking alternative). Over and above the possibilities we discussed in the case of CG we should then include these additional preference structures.
2. I shall below consider some numerical examples of three-person Imitation game (IG), Assurance game (AG), Chicken game (CG), Battle of the Sexes (BS), and the Prisoner's dilemma (PD). (These examples are from Tuomela, 1989a, from a somewhat different context.) No kind of exhaustiveness for my selections will be assumed or claimed, but the point is in any case to indicate how these game-theoretic structures can exemplify collective action dilemmas and free-riderism.
In the case of IG the preferences in the two-person case with two alternatives C and D are as follows. Player 1: CC, DD, DC, CD; player 2: CC, DD, CD, DC. In the three-person case we can get the following ordering (for player 1, and symmetrically for the others) illustrated by two numeric examples within parentheses: CCC (4,4,4; 4,4,4), DDD (3,3,3; 3,3,3), DCC (3,1,1; 1,1,1) CDD (1,1,1; 1,1,1), DCD (1,1,1; 1,1,1), DDC (1,1,1; 1,1,1), CDC (1,3,1; 1,1,1), CCD (1,1,3; 1,1,1). I shall call the first of these examples IG1 and the second IG2. In IG1 it is assumed that the sole defector gains and that there thus is a free-rider effect. In IG2 there is no free-rider effect. (Actually IG2 is a very special case of an Imitation game because it is really a coordination game: the three agents must all try to do the same thing (C or D), preferably C; otherwise they get only 1 utile.)
In AG the preferences in the two-person case are: CC, DC, DD, CD for player 1 and symmetrically for player 2, viz. CC, CD, DD, DC. The three-person case becomes the following, illustrated by two numeric examples : CCC (5,5,5; 5,5,5), DCC (4,3,3; 3,4,4), CDC (3,4,3; 4,3,4), CCD (3,3,4; 4,4,3), DDC (2,2,0; 2,2,0), DCD (2,0,2; 2,0,2), DDD (1,1,1; 1,1,1), CDD (0,2,2; 0,2,2). Here we are assuming that two cooperators can provide the good (although not as fully as three can) and give a better result for everyone than does one sole cooperator, who still can provide some of the good (although it yields a negative net benefit to him). In AG1 there is a free-rider effect, but there is none in AG2. There is one in AG1 because each person prefers the outcome in which the other two contribute but he does not to the two outcomes in which he and exactly one of the other two participants contribute. However, all three participants prefer the outcome where they all three contribute to any of the other possible outcomes, including those in which they do not - but the other two do - contribute; and, furthermore, each prefers to contribute if exactly one of the other two contributes. (One might say that free-riding is not "individually accessible" in AG1.)
Consider now CG. We recall that in the two-person case we get for both players the ranking: DC, CC, CD, DD (player 1), and CD, CC, DC, DD in the case of player 2. With specific reference to the meadow draining example, we can consider, in the three-person-case, the following for player 1 (and symmetrically for the others), using now three slightly different numeric illustrations: DCC (4,2,2; 4,2,2; 4,2,2),DCD (3,1,3; 4,1,4; 3,1,3), DDC (3,3,1; 4,4,1; 3,3,1), CCC (3,3,3; 3,3,3; 2,2,2;), CCD (2,2,4; 2,2,4; 2,2,4), CDC (2,4,2; 2,4,2; 2,4,2), CDD (1,3,3; 1,4,4; 1,3,3), DDD (0,0,0; 0,0,0; 0,0,0). considering the meadow draining example, in the first illustration I have assumed that the sole free-rider gets 4 utiles. Doing one third of the job gives 3 utiles, one half 2 utiles and doing it alone gives 1 utile. In the first numeric case I have assumed that a sole agent is not able alone to do the job to the same degree or as well as jointly with one or two other agents, viz. that two agents are needed for doing the job properly. In the second case, in contrast, one agent is assumed to be able to do the job fully satisfactorily. I have also presented as my third case a case of CG. In it the third participant brings about a slight crowding effect (cf. CCC=(2,2,2)).
Next consider the Battle of the Sexes. In the two-person case the ranking for player 1 is: DC, CD, CC, DD and for player 2 CD, DC, CC, DD. The idea in the three-person case is to think of the selection of a leader or something like that. I have assigned the numbers in two different ways to get for player 1 (and symmetrically for the others) the following: CDD (4,3,3; 4,2,2), DCD (3,4,3; 2,4,2), DDC (3,3,4; 2,2,4), CCC (2,2,2; 3,3,3), DCC (1,1,1; 1,1,1), CCD (1,1,1; 1,1,1), CDC (1,1,1; 1,1,1), DDD (1,1,1; 1,1,1). My numeric examples, BS1 and BS2, correspond to the kind of situations discussed in the previous section in the case of selecting a leader or representative.
We recall that in the case of the two-person Prisoner's dilemma we have the ranking DC,CC,DD,CD, in the case of player 1 and CD, CC, DD, DC in the case of player 2. In the three-person case I give as my illustration the following ranking, with three different numeric assignments: DCC (4,2,2; 4,2,2; 5,2,2), CCC (3,3,3; 3,3,3; 4,4,4), DCD (2,1,2; 3,1,3; 3,1,3), DDC (2,2,1; 3,3,1; 3,3,1), CCD (2,2,4; 2,2,4; 3,3,5), CDC (2,4,2; 2,4,2; 3,5,3), DDD (2,2,2; 2,2,2; 2,2,2), CDD (1,2,2; 1,3,3; 1,3,3).
Next I present an example of a collective action dilemma which is not, however, a free-rider dilemma. This is a borderline game between PD and AG, and I will call it A/P (or AP). In the two-person case the following illustration can be given:

C

D

C

3, 3

0, 3

C

3, 0

2, 2 (A/P)

Here clearly is no free-rider effect. Nevertheless, CC is Pareto-optimal but D is a maximin and a weakly dominating strategy in the case of both players. In the three-person case we can analogously have the following: CCC (3,3,3; 3,3,3);DCC (3,1,1; 3,3,3); DDD (2,2,2; 2,2,2); DDC (2,2,0; 2,2,0); DCD (2,0,2; 2,0,2); CCD (1,1,3; 3,3,3); CDC (1,3,1; 3,3,3); CDD (0,2,2; 0,2,2). C might here be e.g. mowing a lawn. It is assumed in the first of my numeric examples (AP1) that one cooperator cannot alone achieve anything, two can bring about that the defector goes up a bit, while it is quite costly for them. (The defector also pays for the noise brought about by the mowing.) Three cooperators can bring about the good in a satisfactory sense. In the second numeric example (AP2) a defector does not even bring about a cost to the cooperators.
At this point we may note that not only are there collective action dilemmas without a free-rider effect but there are also situations with a free-rider effect which do not satisfy our criterion of a collective action dilemma. Suppose we are dealing with a case of many-person AG whose players regard C as the individually rational (or most rational) alternative (and are not maximin-players who find defection as rational in the sense of security-maximizing.) Then we may concur and apply our characterization (CAD ) of collective action problems into this situation in accordance with this and regard this particular AG as a game without a collective action problem. However, this AG can have a free-rider effect if the cooperative outcome is assumed to be achievable by a majority of players. Then the rest of the players can free-ride on this: for this minority of players defection would in that situation be the utility-maximizing action-strategy.
Next consider Elster's (1985) erosion examples discussed in Section I. The first case can be concisely stated as follows (with C=non-cutting):
I Erosion occurs if and only if and to the extent that one cuts his trees and also one's both neighbors cut their trees.
This case represents Chicken (ERC), as can easily be seen by stating the preferences of the symmetrically placed participants. I also present a numeric example below:
DCC (4,2,2); CCC (3,3,3); DCD (2,1,2); DDC (2,2,1); CDC (2,4,2); CCD (2,2,4); CDD (1,2,2); DDD (0,0,0).
The second case of erosion is defined concisely as follows (assuming now that C=planting):
II Erosion will be stopped if and only if and to the extent that one plants and one's neighbors plant.
We get an AG, for which I give two numeric illustrations (called EA1 and EA2): CCC (3,3,3; 4,4,4); DDD (2,2,2; 2,2,2); DDC (2,2,1; 2,2,1); DCD (2,1,2; 2,1,2); DCC (2,1,1; 2,1,1); CDC (1,2,1; 1,2,1); CCD (1,1,2; 1,1,2); CDD (1,2,2; 1,2,2). As you can see, the only difference between EA1 and EA2 concerns CCC: in EA2 all persons' cooperation is regarded as more valuable than in EA1.
Next we consider what "Compleat Cooperators" would have as their rankings in a cooperative three-person game (COO). Such a cooperator places cooperation above anything else: CCC (4,4,4); CCD (3,3,2); CDC (3,2,3); CDD (2,1,1); DCC (2,3,3); DCD (1,2,1); DDC (1,1,2); DDD (0,0,0).
We recall from the previous section that at least disjunctive tasks require turn-taking. Here is a Turn-taking game (TT) with one numeric illustration: CCC (4,4,4); DDD (3,3,3); DCC (3,2,2); CCD (2,2,3); CDC (2,3,2); DDC (0,0,0); DCD (0,0,0); CDD (0,0,0).
Finally consider a simple nonsymmetric threat game. Obviously there are a great many different kinds of threat games, and here I can only take up one example, called TG. In the two-person case it goes as follows: CC (3,2), DC (2,1), CD (2,3), DD (1,2) for player 1; and CD, DD, CC, DC for player 2. My illustration of this game in the three-person case goes by assuming that there is one threatener (player 1 as in the two-person case) while players 2 and 3 are have symmetric utilities. We get for player 1: CCC (3,3,3); CCD (2,3,4); CDC (2,4,3); DCC (2,1,1); CDD (2,4,4); DCD (2,1,2), DDC (2,2,1), DDD (2,2,2). The preference orderings of the other two players can be read off from the numerical example, and I shall not bother to write them out more explicitly here.7


V CONCLUDING REMARKS

In the beginning of this paper I defined both collective action dilemmas (or situations) and free-rider situations. Much of the paper was devoted to an examination of what kind of preference structures can be involved in collective action situations and, especially, in free-rider situations. Detailed studies of both two-person and three-person cases were presented. Obviously the three-person cases are the more interesting of these, for most of the relevant applications concern many-person cases, and three-person cases represent, if not all, at least most of the central problems that can be involved in many-person free-rider situations. In fact the present study is the first comprehensive and detailed study of three-person cases - as far as I know.8
In the present paper no attempt was made to give "solutions" to free-rider problems in the sense of giving recommendations or prescriptions for avoiding free-riderism. Instead the focus was on clarifying the variety of situations and their features that are structurally relevant to free-rider problems.
The results of this paper are perhaps not as clear-cut and informative as one might have hoped for; but no doubt this in part reflects that the phenomena studied indeed are varied and many-sided. It should also be noted that as our numeric investigations were rather preliminary in their nature, future research may of course help to clarify the issues and perhaps bring new features into focus.
Let me summarize some of the theses found to be true or at least well supported by the research reported in this paper:
1) Situations involving collective action problems (viz. a conflict between collectively and merely personally rational action) can be exemplified by a variety of game-theoretic structures (or strategic interaction structures).
2) Free-rider situations can be exemplified by many kinds of game-theoretic structures (e.g. PD, CG, BS, AG, TT, and so on). Nevertheless, there is no good reason to think that some kind of strict definition (viz. necessary and sufficient conditions) of a free-rider situation can be given in game-theoretic terms (e.g. as some kind of disjunction or other conglomeration of game-structures).
3) There are collective action situations which do not involve a free-rider problem or even a free-rider effect (cf. our AP2).
4) There are collective action situations with a free-rider effect which do not, however, involve a collective action problem (in the sense of our (CAD )).
5) While it is fruitful in this kind of study to make an analytic distinction between selection (or coordination) problems and problems of conflict (or exchange), our investigations showed that typical cases are not pure ones but that various mixtures of these elements can be (and often are) present.9
6) Our investigations also showed that, contrary to what the existing literature on the topic suggests, it is not only conjunctive collective tasks that are relevant but also disjunctive tasks and perhaps many other kinds as well (cf. Tuomela, 1984, Chapter 5 for the various possibilities).



NOTES

1) Elster's claim that the present case of erosion is individually inaccessible is, however, problematic. Consider a two-person situation. Suppose one of the agents goes out and starts planting trees. At least if the preferences are mutually known to the players - as is assumed in game theory - the other player will have an incentive to join in (see Section II on two-person Assurance game). But this also means that the first player originally can have reasoned so and thus have had an incentive to start planting; and thus the cooperative outcome can be argued to be individually accessible, after all. Many-person situations, although more complex, can be handled analogously.

2) I have above assumed that the notion of preference is clear enough not to cause problems for a characterization of rational action in an objective or intersubjective sense. But this is also a problematic assumption if we want to get hold not only of means-end rational action but also end-rationality ("value-rationality"). Even if we would take preferences as preferences all-things-considered in the context in question, we should then presumably be concerned with "real" preferences, viz. preferences based on "real" as opposed to "artificial" needs and/or desires. Furthermore, the subjective beliefs should accordingly in such a more refined analysis be required to be suitably "rational" or "optimal", too. Rational acting could then be taken to amount to - or at least involve -acting with a reason based on real needs and relevant optimal beliefs. But these notions are all very problematic and I must here leave these questions untouched and operate on a less refined level. Indeed it suffices for us here to be concerned only with means-end rational action.

3) The notions of we-intention and conditional we-intention relevant to this paper can be analyzed as follows (see Tuomela, 1991a, for discussion and clarification):

(WI ) A member Ai of a collective G we-intends to do X if and only if
(i) Ai intends to do his part of X (as his part of X);
(ii) Ai has a belief to the effect that the joint action opportunities for an intentional performance of X will obtain, especially that a right number of the full-fledged and adequately informed members of G, as required for the performance of X, will (or at least probably will) do their parts of X, which will under normal conditions result in an intentional joint performance of X by the participants;
(iii) Ai believes that there is (or will be) a mutual belief among the participating members of G (or at least among those participants who do their parts of X intentionally as their parts of X) to the effect that the joint action opportunities for an intentional performance of X will obtain.

(CWI ) A member Ai of a collective G conditionally we-intends to do X if and only if
(i) there is a condition Ci such that Ai intends to do his part of X, given that Ci will obtain;
(ii) Ai has a belief to the effect that in Ci the joint action opportunities (except possibly those related to the participants' specific action conditions) for an intentional performance of X will obtain, especially that a right number of the full-fledged and adequately informed members of G, as required for the performance of X, will (or at least probably will) conditionally do their parts of X, which will under normal conditions result in an intentional joint performance of X by the participants, given that every other participant's condition is satisfied and believed by the participant to be satisfied.
(Conditional acting here means that, for j=1,2,..., there are conditions Cj such that Aj will do his part Xj given that Cj obtains, where Cj is the condition for Aj's action to do Xj. The conditions C1 ,C2 ,...,Cm are believed by Ai to be jointly realizable.
4) Let me characterize the notion of an objective free-iding situation somewhat more precisely. Admitting that the fine details might be put differently, I here make the following simple proposal:
(FRS ) S is a situation of free-riding in a collective G relative to a public good produced by a joint action X if and only if
1) the joint action opportunities for the performance of X obtain, especially it is the case that the contribution (doing their parts) of K members leads to the provision of the public good related to X);
2) there is a mutual belief among the full-fledged and adequately informed members of G to the effect that the joint action opportunities for the performance of X, and to the effect that each full-fledged and adequately informed member ought to contribute,
3) everyone in G will gain more from defection (doing D) than from contribution (doing C) if at least K agents contribute, viz. if at least K agents out of N do C, K being the minimal number of agents capable of jointly performing X.
4) the outcome resulting from all the agents contributing (doing C), is better than the outcome when all defect (do D).
5) anybody's defection involves a cost (possibly nil) to the contributing members of G.

5) A Nash equilibrium is an equilibrium in pure strategies in the maximin sense. For instance, in the case of two-person Chicken CC is such a Nash equilibrium, for C maximizes the minimal possible gain that a player can get in the case of both players. An equilibrium is an outcome such that no one can gain by alone acting otherwise. An equilibrium is a coordination equilibrium in case no one can gain if any one agent alone acted otherwise, either the agent himself or someone else.

6) Hampton (1987) claims that Hume's meadow draining example represents BS. This cannot be right. We have claimed that depending on one's special assumptions it can be made a CG or a PD-like situation. On p. 252 of her paper she explicitly assumes that DDD cannot be ranked lowest. Thus BS becomes an excluded possibility.

7) Summarywise, the numeric examples dealt with above and in the numeric computations of Tuomela (1989a) can be rendered as follows (omitting the commas in the triplets):
CCC, CCD, CDC, CDD, DCC, DCD, DDC, DDD
IG1: 444, 113, 131, 111, 311, 111, 111, 333
IG2: 444, 111, 111, 111, 111, 111, 111, 333
AG1: 555, 334, 343, 022, 433, 202, 220, 111AG2: 555, 444, 444, 022, 444, 202, 220, 111
CG1: 333, 224, 242, 133, 422, 313, 331, 000
CG2: 333, 224, 242, 144, 422, 414, 441, 000
CG3: 222, 224, 242, 133, 422, 313, 331, 000
BS1: 222, 111, 111, 433, 111, 343, 334, 111
BS2: 333, 111, 111, 422, 111, 242, 224, 111
PD1: 333, 224, 242, 122, 422, 212, 221, 222
PD2: 333, 224, 242, 133, 422, 313, 331, 222
PD3: 444, 335, 353, 133, 533, 313, 331, 222
AP1: 333, 113, 131, 022, 311, 202, 220, 222
AP2: 333, 333, 333, 022, 333, 202, 220, 222
ERC: 333, 224, 242, 122, 422, 212, 221, 000
EA1: 333, 112, 121, 122, 211, 212, 221, 222
EA2: 444, 112, 121, 122, 211, 212, 221, 222
COO: 444, 332, 323, 211, 233, 121, 112, 000
TT : 444, 223, 232, 000, 322, 000, 000, 333
TG : 333, 234, 243, 244, 211, 212, 221, 222


8) I have shown in Tuomela (1989a) that my earlier (1985) theory of the components of social control (improving and extending Kelley's and Thibaut's (1978) important work) can be usefully applied to the study of free-riderism and other problems related to collective action.

9) The mentioned features relate to the absolute control over the other participants (this can represent conflict and exchange) or to the conditional control of some kind between them (selection, coordination), and it should not be forgotten that also the agents' reflexive control (viz. their control over their own utilities) can be relevant. I discuss these in Tuomela (1989a), where I also emphasize the general point that the components of social control give a fruitful analytic tool for investigating free-rider situations and other collective action situations.


REFERENCES

Axelrod, R. (1984), The Evolution of Cooperation , Basic Books, New York
Elster, J. (1985), 'Rationality, Morality, and Collective Action', Ethics 96 , 136-155
Hampton, J. (1987), 'Free-Rider Problems in the Production of Collective Goods', Economics and Philosophy 3 , 245-273
Hardin, R. (1971), 'Collective Action as an Agreeable N-Prisoner's Dilemma', Behavioral Science 16 , 472-481Hardin, R. (1982), Collective Action , The Johns Hopkins Press for Resources for the Future, Baltimore
Hume, D. (1965), A Treatise of Human Nature , L.A. Selby-Bigge (ed.), The Clarendon Press, Oxford
Kelley, H. and Thibaut, J. (1978), Interpersonal Relations , Wiley, New York
Olson, M. (1965), The Logic of Collective Action , Harvard UP, Cambridge, Mass.
Pettit, P. (1986), 'Free Riding and Foul Dealing', The Journal of Philosophy LXXXIII , 361-379
Schelling, T. (1978), 'Hockey Helmets, Daylight Saving, and Other Binary Choices', in Schelling, T., Micromotives and Macrobehavior , Norton, New York, 211-243
Taylor, M. (1987), The Possibility of Cooperation , Cambridge UP, Cambridge
Taylor, M. and Ward, H. (1982), 'Chickens, Whales, and Lumpy Goods: Alternative Models of Public Goods Provision', Political Studies 30 , 350-370
Tuomela, R. (1984), A Theory of Social Action , Reidel, Dordrecht and Boston
Tuomela, R. (1985), 'The Components of Social Control', Quality and Quantity 19 , 1-51
Tuomela, R. (1988), 'Free-Riding and the Prisoner's Dilemma', The Journal of Philosophy LXXXV , 421-427
Tuomela, R. (1989a), 'Collective Action, Free-riders, and Interpersonal Control, Reports from the Department of Philosophy , University of Helsinki, n:o 2
Tuomela, R. (1989b), 'Actions by Collectives', Philosophical Perspectives 3 , 471-496
Tuomela, R. (1991a), 'We Will Do It: An Analysis of Group-Intentions', forthcoming in Philosophy and Phenomenological Research
Tuomela, R. 1991b, The Importance of Us: A Philosophical Study of Basic Social Notions , in preparation
Tuomela, R. and Miller, K. (1991), 'Free-Riding, Being in Reserve, and Conditional Intentions', forthcoming in Erkenntnis
Ullmann-Margalit, E., 1977, The Emergence of Norms , Oxford University Press, Oxford



Department of Philosophy
University of Helsinki
Unioninkatu 40 B
00170 Helsinki
Finland