Skip to content
Publicly Available Published by De Gruyter July 12, 2016

Simple Unawareness in Dynamic Psychological Games

  • Carsten S. Nielsen and Alexander Sebald EMAIL logo

Abstract

Building on Battigalli and Dufwenberg (2009)’s framework of dynamic psychological games and the progress in the modeling of dynamic unawareness by Heifetz, Meier, and Schipper (2013a) we model and analyze the impact of asymmetric awareness in the strategic interaction of players motivated by reciprocity and guilt. Specifically we characterize extensive-form games with psychological payoffs and simple unawareness, define extensive-form rationalizability and, using this, show that unawareness has a pervasive impact on the strategic interaction of psychologically motivated players. Intuitively, unawareness influences players’ beliefs concerning, for example, the intentions and expectations of others which in turn impacts their behavior.

JEL Classification: C72; C73; D80

1 Introduction

Recent lab and field evidence suggests that people not only care about the monetary consequences of their actions, but that their behavior is also driven by psychological payoffs (for example, Fehr, Kirchsteiger, and Riedl 1993; Charness and Dufwenberg 2006; Falk, Fehr, and Fischbacher 2008; Bellemare, Sebald, and Strobel 2011). Two prominent examples of psychological payoffs in the hitherto existing literature are reciprocity (for example, Rabin 1993; Dufwenberg and Kirchsteiger 2004; Falk and Fischbacher 2006) and guilt aversion (for example, Charness and Dufwenberg 2006; Battigalli and Dufwenberg 2007). Departing from the strictly consequentialist tradition in economics, Geanakoplos, Pearce, and Stacchetti (1989) and Battigalli and Dufwenberg (2009) present general frameworks for analyzing the strategic interaction of players with psychological payoffs: “psychological games”. Roughly speaking, psychological games are games in which players’ preferences depend upon players’ beliefs about the strategies that are being played, players’ beliefs about the beliefs of others about the strategies that are being played, and so on.

A widely unspoken assumption that is underlying game-theoretic analyses, and therefore also the analyses of psychological games, is that players are aware of all facts characterizing the strategic environment they are in. However, in many real life situations this is not the case. People often have asymmetric awareness levels concerning their own as well as the feasible choices of others although they are part of the same strategic environment. People are frequently surprised in the sense that they become aware of new strategic alternatives by observing actions they had previously been unaware of. In recent years different models of unawareness have been proposed showing the importance of unawareness for individual decision making problems as well as the strategic interaction of players in standard (non-psychological) games (for example, Fagin and Halpern 1988; Dekel, Lipman, and Rustichini 1998; Modica and Rustichini 1999; Halpern 2001; Heifetz, Meier, and Schipper 2006, 2013a, 2013b; Halpern and Rêgo 2006, 2008, 2009; Heifetz, Meier, and Schipper 2008; Li 2009; Grant and Quiggin 2013).

However, it is not only in standard games that unawareness is important. We show in our analysis that unawareness has a profound and distinct impact on the strategic interaction of players in psychological games. To see this consider the following intuitive example: Imagine two friends, Ann and Bob. Assume it is Bob’s birthday, he is planning a party and would be very happy, if Ann could come. Unfortunately Bob’s birthday coincides with the date of Ann’s final exam at university. She can either decide to take the exam the morning after Bob’s party or two weeks later at a second date. Ann is certain that Bob would feel let down, if she were to cancel his party without having a very good excuse. Quite intuitively, although Ann would really like to get over her exam as soon as possible, she might anticipate feeling guilty from letting down Bob if she canceled his party to take the exam the following morning. As a consequence, Ann might choose the second date to avoid letting Bob down. In contrast, consider now the following variant of the same example: Ann knows that Bob is unaware of the second date. In this situation Ann might choose to take the exam on the first date and not feel guilty. Since Bob is unaware of the second date and the final exam is a good excuse, he does not expect Ann to come. Ann knows this and, hence, does not feel guilty as Bob is not let down. In fact, if she were certain that Bob would never become aware of the second date, she probably had an emotional incentive to leave him unaware in order not to raise his expectations. That is, she had an incentive not to make him aware of the fact that she actually has the time to come to his party, but just wants to get over her exam. Interestingly, if Ann were only interested in her own payoff in this strategic situation with unawareness, she would not care whether Bob is or will become aware of the second date. She would simply not attend his party irrespective of Bob’s awareness. Only her belief-dependent feeling of guilt towards Bob creates the strong emotional incentive to leave him unaware.

Bob’s unawareness concerning Ann’s ability to come to his party and, connectedly, Ann’s incentive not to tell him about the second date intuitively highlight the focus of our analysis. We analyze the influence and importance of unawareness concerning feasible paths of play for the strategic interaction of players in psychological games. To simplify the analysis we concentrate on two-player strategic environments with simple unawareness. More specifically, building on Battigalli and Dufwenberg (2009)’s framework of dynamic psychological games and the recent progress in the modeling of (Heifetz, Meier, and Schipper 2006, 2008, 2013a, 2013b), we define a two-player model in which players are motivated by psychological payoffs and one player is potentially unaware of certain feasible paths of play. More specifically, in our two-player setting with simple unawareness, one player is initially aware of all paths of play, whereas the other player is initially unaware of some paths of play. We assume that the aware player is aware of the unaware player’s unawareness, but the unaware player is not. We refrain from moves of chance implying that players’ information sets are singletons. We restrict ourselves in this way to clearly investigate and highlight the pervasive role of asymmetric awareness on the strategic interaction of players motivated by belief-dependent preferences. Extensions to broader settings are of course feasible, and will definitely allow for the analysis of a lot of interesting applications, but we leave it for future research to explore these directions. Using our framework we provide different examples highlighting the role of unawareness in the strategic interaction of players motivated by reciprocity à la Dufwenberg and Kirchsteiger (2004) and guilt aversion à Battigalli and Dufwenberg (2007).We limit ourselves to two-player environments and simple asymmetric awareness scenarios in order to intuitively introduce our model and clearly uncover the role of unawareness without burdening the analysis with technical issues arising in strategic environments allowing for more players and more complex unawareness.

Our examples demonstrate that the strategic behavior of players motivated by psychological payoffs crucially depends on their awareness concerning the strategic environment they are in, their perception concerning the awareness of others, their perception concerning the perception of others, and so on – a fact that implies both an opportunity as well as a challenge to analyses empirically investigating the strength and nature of psychological payoffs. On the one hand, in line with experimental evidence suggesting that people are more prone to selfish choices if they believe that others will remain unaware of them (for example, Dana, Cain, and Dawes 2006, Dana, Weber, and Kuang 2007; Broberg, Ellingsen, and Johannesson 2007; Andreoni and Bernheim 2009; Lazear, Malmendier, and Weber 2012),our examples show that redchanging the awareness of players that are motivated by psychological payoffs leads to intuitive and testable predictions distinct from predictions based on consequentialist preferences like selfishness and inequality aversion (Fehr and Schmidt 1999). On the other hand, it poses a challenge for experimental investigations in relatively uncontrolled environments like the field or the Internet. As also seen in our introductory example, not controlling for Ann’s perception concerning Bob’s awareness might lead to wrong inferences concerning Ann’s inclination to feel guilty towards Bob. Furthermore, our examples reveal that over and above the actual choices that are made, managing other people’s awareness levels has to be understood as an integral and important part of any strategic interaction. By managing other’s awareness, we influence the others’ expectations and perceptions concerning our intentions, which in turn influences their behavior.

We start out by formulating a model concentrating on two-player extensive-forms with complete information, observable actions and no chance moves. To allow for unawareness we use a standard extensive-form representing the game tree and a subtree thereof, and define extensive-forms with simple unawareness with the help of singleton information sets. These singleton information sets describes at each decision node in the game tree, and copy thereof in the subtree, the frame of mind of a player. Our two-player extensive-forms are in essence a special case of Heifetz, Meier, and Schipper (2013a)’s generalized extensive-forms, and therefore embeddable in their setting.Of course, our extensive-form with unawareness is not typically common knowledge among players, and therefore should be interpreted from the modeler’s point of view. In fact, any game that does not explicitly distinguish between the players’ description of the strategic environment and the modeler’s will fail to capture (see Dekel, Lipman, and Rustichini 1998).

Having defined our class of two-player extensive-forms with unawareness, we formally characterize psychological payoffs in our setting. In synthesis, we define a player’s strategies and conditional beliefs about the other player’s pure strategies (first-order beliefs), beliefs about the other player’s beliefs (second-order beliefs), and so on. The infinite hierarchy of conditional beliefs that we define takes player’s awareness, players’s perception regarding the other’s awareness, and so forth, into account and is used for the general specification of our psychological payoffs and, hence, the characterization of our class of dynamic psychological games with simple unawareness. As mentioned above, specific types of psychological payoffs that can be embedded in our model are among others reciprocity and guilt aversion.

Dufwenberg and Kirchsteiger (2004), Battigalli and Dufwenberg (2007) and Sebald (2010) propose sequential equilibrium as a solution concept for their psychological games. However, assuming equilibrium play is very demanding in strategic environments involving unawareness. The implicit assumption made when imposing sequential equilibrium on strategic settings with unawareness is that if a player becomes aware of more during the game, he will compute new equilibrium beliefs not rationalizing, for example, why the other player made him aware. Sequential equilibrium only requires a player to reason about the other player’s future behavior. For this reason, we impose extensive-form rationalizability (Pearce 1984), which embodies forward induction, as a solution concept for our psychological games with simple unawareness. Extensive-form rationalizability implies, that along each feasible path of play, every active player is always certain that the other player sequential best responds, certain that the other player is certain that he sequential best responds, and so on. If a player finds himself at some information set, where the other player’s strategies that could lead to that information set are inconsistent with the players previous certainty in the other player’s best response, then the player seeks a best rationalization which could have led to that information set (Battigalli 1997; Battigalli and Siniscalchi 2002). That is, if the player is “surprised” by the other player’s unexpected action, and cannot use Bayesian updating, then he forms new beliefs that justify this observed inconsistency. In its simplest form, forward-induction reasoning involves the assumption that, upon observing an unexpected (but undominated) action of the other player, a player maintains the working hypothesis that the latter is a sequential best response. The best rationalization principle captures precisely this type of argument.

After having defined our model, the solution concept and two prominent notions of psychological payoffs, reciprocity and guilt aversion, we describe two examples to highlight the role of unawareness in the interaction of players motivated by reciprocity and guilt aversion. First, we consider a version of the sequential prisoners dilemma also analyzed by Dufwenberg and Kirchsteiger (2004) featuring a reciprocal second mover, Bob, who is unaware that the first mover, Ann, can defect. [1] Different to Dufwenberg and Kirchsteiger (2004)’s analysis assuming full awareness, it is shown that as long as Bob is unaware of the fact that Ann could have defected, he defects independent of his sensitivity to reciprocity – even when Ann chooses to cooperate. The way he perceives Ann’s kindness does not only depend on what she does, but also on what Bob thinks she could have done given his awareness of the strategic situation. Ann anticipates this and defects as long as she cannot cooperate and simultaneously make Bob aware of the fact that she could have defected. As a second example, we formally revisit the example of Ann not wanting to come to Bob’s party because of the exam, and consider how her aversion to guilt affects her choice. If Ann knows that Bob will not be “let down” if he is unaware of the second exam date, then Ann does not feel any guilt towards Bob if she chooses not to come to the party. Because Bob is unaware of the second exam date he foresees that Ann will write the exam on the first date, and thus, does not expect Ann to come. Both examples highlight that unawareness in the interaction of players with psychological payoffs leads to very intuitive behavioral predictions distinct from predictions using non-psychological preferences or no unawareness. Furthermore, it becomes evident that managing others’ awareness levels is an important and integral part of strategic interactions of players motivated by psychological payoffs.

The organization of the paper is as follows: In Section 2 we introduce dynamic games with simple unawareness. Following this, in Section 3, we define the hierarchies of beliefs and psychological payoffs. Section 4 contains the definition of our solution concept: extensive-form rationalizability. In Section 5 we give two examples of how our model can be applied. Finally, Section 6 concludes.

2 Dynamic Games with Simple Unawareness

A finite extensive-form game with singleton information sets and no chance moves, called the simple unawareness game, is played by two players, player i and some other player j, who move one at a time and have possibly different views on the feasible paths of play. One player is initially aware of all paths of play, whereas the other player is initially unaware of some paths of play. The aware player knows that the other player is unaware, while the unaware player is unaware of his own unawareness and thinks that the other player is aware of the same as he. The game specifies material payoffs for each player at each terminal node. These payoffs describe the material consequences of the players’ actions, not their preferences. The players’ psychological payoffs will be introduced in Section 3.

The simple unawareness game we consider adapts Heifetz, Meier, and Schipper (2013a)’s generalization of the standard extensive-form game (Fudenberg and Tirole 1991, chapter 3.3). By simple unawareness we mean that our game is restricted to unawareness with just two trees – a game tree specifying all physical paths of play and a subtree thereof. Both will be described in detail below. Our game allows for a parsimonious analysis of the applications we consider. More general applications and extensions are certainly very interesting (for example, delusion or awareness of unawareness) but are left for future research in order to not burden our analysis with additional notational complexity.

Game tree. The physical paths of play are given by a finite set T of nodes together with a binary relation on T that represents precedence. The binary relation must be a partial order, and (T,) must form an arborescence: the relation totally orders the predecessors of each member of T. The order of play thus constitute a game tree that begins at an initial node with no predecessor and then proceeds along some path from node to an immediate successor, terminating when a node with no successor is reached. The various paths give the various possible orders of play. Let Z denote the set of terminal nodes, and let X=TZ be the set of decision nodes.

Moves in the game tree. To represent the choices available to players at decision nodes, we have a finite set A of actions and a function ψ that labels each non-initial decision node x with the last action taken to reach it. We require that ψ be one-to-one on the set of immediate successors of each decision node x, so that different successors correspond to different actions. Let A(x) denote the set of feasible actions at x. Actions are labeled so that A(x)A(x)= for xx. To represent the rules for determining who moves at a decision node, we have a function ι:X{i,j} that assigns to each decision node the player whose turn it is.

Subtree. A subtree is defined by a subset of nodes TT for which (T,) is also an arborescence – perhaps starting at a different initial node. To ensure that the subtree is associated with well-defined payoffs to the players, we impose that all terminal nodes in the subtree are also in Z. Decision nodes that appear in the subtree, also appear in the game tree. We will need to explicitly differentiate these decision nodes and define them as distinct elements. To this effect, we label by y the copy of the decision node x, whenever the copy of x is part of the subtree T. Let Y be the set of copies of decision nodes in the subtree. The subtree together with the structure introduced below is intended to model the subjective partial view of the player.

Moves in the subtree. To ensure consistency between moves in the game tree and subtree, we impose that feasible actions at the copy y of decision node x are given by a non-empty subset A(y)A(x) for which the properties of the function ψ are not violated. This implies that if the action aA(x) leads from x to successor x in the game tree T, then a also leads from the copy y to y (the copy of x in the subtree T) whenever both copies appear in the subtree. We also require that the same player moves at decision node x and copy y, so that there is no disagreement about which player that has to move.

To exemplify the implications of the above definitions, consider the the extensive-form underlying the sequential prisoners dilemma also analyzed by Dufwenberg and Kirchsteiger (2004):

Consider first the game tree T in Figure 1. At the initial node x0, Ann is active and can corporate (the action C) or defect (the action D). At nodes x1 and x2, Bob is active and can corporate (the action c) or defect (the action d). Now consider the subtree T. The copy of decision node x2 is not a part of the subtree, which implies that at the copy y0 of the initial node x0, Ann only has one feasible action C. Bob, on the other hand, can still choose c or d when he is active at copy y1 of decision node x1.

Figure 1: A game tree and a subtree thereof.
Figure 1:

A game tree and a subtree thereof.

Generic decision nodes. A generic decision node n is an element of the disjoint union N=XY. By generic we mean that n can describe both decision nodes xX as well as the copies yY.

Singleton information sets. The information players have when choosing their actions is the most subtle part of our simple unawareness games. In the simple unawareness game players know all previous moves, but a player’s frame of mind may not allow him to be aware of all paths of play. The information possessed by player i is represented by singleton information sets hi(n)Hi for all nN. The singleton information set hi(n) defines the frame of mind of player i by identifying the paths of play the player conceives possible at n. Unlike a standard information set, the generic decision node n does not need to be contained in hi(n). For example, at decision node x in the game tree T it might be that hi(x) is in the subtree T. When the singleton information set hi(n) is in the subtree T, player i is unaware of all paths of play not described by the subtree.

In games with psychological payoffs, it is important to represent players’ information also at nodes where they are not active. The information structure Hi of player i thus contains, as a subset IiHi, the information structure of active player i. That is, the subset Ii contains as elements singleton information sets hi(n) of player ι(n)=i.

Let T and T′′ be two generic trees that each can be either the subtree T or the game tree T. Each singleton information set hi(n)Hi has the following static properties that parallel those in Heifetz, Meier, and Schipper (2013a, 59–60) (see also Figure 2) [2]:

U0 Confined awareness: if nT′′ then hi(n)T with TT′′.

U1 Generalized reflexivity: for TT, xT, and hi(x)T, if T contains a copy y of x then yhi(x).

U2 Introspection: if nhi(n) then hi(n)=hi(n).

U3 The subtree preserve awareness: for TT, xT, and xhi(x), if T contains a copy y of x then yhi(y).

Figure 2: Examples that agree and disagree with properties U0-U3.
Figure 2:

Examples that agree and disagree with properties U0-U3.

Properties U0-U3 are all static properties, as a dynamic property we impose perfect recall. We require that if generic decision node n′′hi(n) and if n is a predecessor of n, then there is a predecessor nˆ of n′′ in the same tree as n′′ for which hi(nˆ)=hi(n), and that the action taken at n along the path to n is the same as the action taken at nˆ along the path to n′′. Intuitively, the generic decision nodes n and n′′ are distinguished by information player i does not have, so he cannot have had it when he was at information set hi(n); n and n′′ must be consistent with the same action at hi(n), since the player remembers his action there. Perfect recall together with confined awareness implies that the player cannot become unaware along a path of play. Suppose that yhi(n), hi(n)T and generic decision node n is a predecessor of n. For y there is – by perfect recall – a copy yT that is a predecessor of y, such that hi(y)=hi(n). By confined awareness it must be that hi(n)T. Thus, player i could not have been aware of more at n.

In games with psychological payoffs, where players somehow care about the beliefs or intentions of others, it is also important to be explicit about what a player is aware of and what he – given this awareness – thinks the other player is aware of. To be explicit about this, we use composite singleton information sets hji(n)=hjhi(n). [3] Finally, each player’s singleton information set is assumed a primitive of the game. When there is no need to be explicit it thus makes sense to write hi instead of hi(n), use the notation ι(hi) and A(hi) instead of ι(n) and A(n), and write hji instead of hji(n).

Figure 3 displays the extensive-form underlying the sequential prisoners dilemma considered in Figure 1 now with singleton information sets that are consistent with properties U0-U3 and perfect recall added. In the figure, singleton information sets are shown as arrows. The “solid arrows” indicate Ann’s singleton information sets, while the “broken arrows” indicate Bob’s singleton information sets. For the sake of simplicity, we omit Bob’s redundant singleton information sets at y0 and y1. Ann is aware when active. When choosing an action at singleton information set hA(x0), she considers the physical paths of play. At hA(y0) the action C should be interpreted as the action she would have taken had she only been aware of feasible path of play in the subtree. Moreover, she thinks that Bob is unaware since hBA(x0)T. When Bob is active at hB(x1), he is unaware of the physical path where Ann could have chosen D. He thinks that Ann is also unaware since hAB(x1)T. If Ann chooses D, then Bob is aware and active at hB(x2). At this singleton information set Bob will be surprised. He realizes that had he observed Ann choosing C, then he would not have suspected that she could have chosen anything other than that action.

Figure 3: A game tree and a subtree thereof with singleton information sets.
Figure 3:

A game tree and a subtree thereof with singleton information sets.

Pure strategies. Let Ai=hiHiAi(hi) be the set of all actions for player i. A pure strategy for aware player i is a map si:IiAi, with si(hi)A(hi) for hiIi. A pure strategy for player i thus specifies an action at each of the singleton information sets at which the player is active. Player i’s set of pure strategies, Si, is simply the space of all such si. Since each of these pure strategies is a map from singleton information sets to some action, we can write Si as the Cartesian product of the action sets at each hi:

Si=ΠhiIiA(hi).

Because the aware player i is aware of all of his singleton information sets, his set of pure strategies is equal to Si.

Remember, in our simple unawareness game only the game tree T represents the physical paths of play. The subtree T represents the restricted subjective view of the feasible paths in the mind of an unaware player, or the view of the feasible paths that an aware player assign to the unaware player, and so on. Moreover, as the game evolves a player may become aware of paths of which he was unaware earlier. A strategy can thus not in the simple unawareness game be conceived as an ex ante plan of actions. Instead, it should be interpret as a list of answers to the question “what would player i do if hi were the singleton information set he considers as possible?”

For example, in Figure 3 we can identify aware Ann’s strategies at the singleton information set hA(x0) with the actions C and D – the actions she actually takes. Aware Ann’s set of strategies is thus SA={CC,DC}, where the first index refers to the action taken at hA(x0) and the second index to the action taken at hA(y0). Following Ann’s action D, at singleton information set hB(x2), Bob is aware, not only of the fact that he can take actions c and d, but also that he could have taken the actions c and d at hB(x1) following Ann’s action C. We can thus identify Bob’s strategies by SB={cc,cd,dc,dd}.

Denote by S=Si×Sj the set of strategy pairs. The path caused by a strategy pair sS yields a terminal node denoted z(s)Z. Strategy si reaches singleton information set hi if the path induced by sS reaches hi. Otherwise, we say that hi is excluded by the strategy si. The set of player i’s strategies that reaches hi is denoted Si(hi). We will sometimes also write Sj(hi), meaning player j’s strategies that reach hi.

Unaware player i is unaware of singleton information sets in the game tree. For strategy siSi, we denote by siT the strategy of unaware player i induced by si. Strategy si induces strategy siT if si(hi)=siT(hi) for every hiT. For example, in Figure 3 the induced strategy of Ann’s strategy si=(CC) is sAT=(C), while the induced strategy of Bob’s strategy sBT=(cd) is sBT=(c). Let SiT be the set of player i’s induced strategies, and SiT(hi) denote the set of strategies that reaches hiT. The path caused by the induced strategy pairs yields a terminal node z(sT)Z. If RiSi is some set of strategies of player i, denote by RiT the set of strategies induced by Ri in the subtree T. In the rest of the paper we will use the notation siλ(hi)Siλ(hi), with index function λ(hi)={} if hiT and λ(hi)={T} if hiT, to define player i’s strategies at a singleton information set hi.

It is important to understand that Bob at hB(x2) (after he has become aware) is not deluded to think that the strategic interaction at hB(x1) is described by paths of play in the subtree, nor does he think that Ann was ever unaware. Rather, Bob interprets Ann’s induced strategy sAT={C} as describing the action Ann would have taken, had she decide to keep him unaware. For any solution concept we need to analyse what, for example, Ann thinks of Bob’s actions in the subtree, such actions are determined by Bob’s induced strategy. This is why Ann’s strategy also determines her action at hA(y0) in the subtree, even though she will never actually only consider this action as possible given that she is aware of all feasible paths of play.

Material payoffs. Above we defined the strategic interaction in an extensive form with simple unawareness. To obtain a dynamic game with simple unawareness we add a specification of the players’ material payoffs assigned to the terminal nodes. Because each terminal node zZ completely determines a path through the game tree, we can assign to player i material payoffs using functions πi:ZR.

3 Beliefs and Psychological Payoffs

Extensive games with simple unawareness, as described above, assumed that payoffs depend only on induced paths of play. This in not sufficient for describing the motivations and choices of players who care about, for example, guilt aversion and reciprocity. Psychological games, on the other hand, allow payoffs to depend directly on beliefs (about beliefs), and via such beliefs capture, for example, emotions like reciprocity and guilt.

Beliefs. Conditional on each singleton information set hiHi, player i holds an updated, or revised, belief αi(|hi)Δ(Sjλ(hi)(hi));

αi=(αi(|hi))hiHiΠhiHiΔSjλ(hi)(hi)

is the system of first-order beliefs of player i.

For example, at hB(x0) in Figure 3, Bob is certain that Ann’s strategy is sAT={C}. If Bob subsequently finds himself at hB(x2), then his belief will change, so that he now is certain that Ann’s strategy is sA={CD}. Bob thus becomes aware that Ann could have kept him unaware by choosing strategy sA={CC}, but instead chose to make him aware by choosing action D.

At hi player i also holds a second-order belief βi(hji) about the first-order belief system αj of player j, a third-order belief about the second-order beliefs, and so on. For the purpose of this paper, we may assume that higher-order beliefs are degenerate point beliefs. Thus, with a slight abuse of notation we identify βi(hji) with a particular first-order belief system αj. A similar notational convention applies to other higher-order beliefs. Let the sequence μi=(αi,βi,) denote player i’s hierarchy of beliefs, and Mi the (compact) set of such hierarchies. In our applications we consider beliefs at most of the second order.

At hA(x0) in Figure 3, Ann holds a first-order belief αA(|hA(x0))Δ({cc,cd,dc,dd}). She thinks that Bob is unaware, and her belief about Bob’s first-order belief about her strategy must reflect this. Ann thus needs to consider Bob’s frame of mind. That is why her second-order belief βA(hBA(x0)) is conditioned on the composite singleton information set, implying that she is certain that Bob is certain that her strategy is sAT={C}.

Players should not change their beliefs unless the play reaches a singleton information set which falsifies it. We therefore assume that player i’s hierarchy of beliefs Mi are consistent such that: there is at least one strategy of player j in the support of αi(|hi) at some hi, and that beliefs must satisfy Bayes’ rule and common knowledge of Bayes’ rule whenever possible. Consistency of the updating system requires that αi(|hi), βi(hji), and so on, at hi are consistent with hi being reached and that no beliefs are abandoned unless falsified. Thus, when Bob in Figure 3 finds himself at hB(x2), he must change his beliefs such that they are consistent with being aware.

Psychological payoffs. Section 2 defines dynamic games with simple unawareness. To obtain a dynamic psychological game with simple unawareness we extent payoffs to include beliefs. Specifically, we assign to player i psychological payoffs using functions ui:Z×MiR. The psychological payoffs will be obtained from the material payoff functions πi:ZR. We exemplify this definition by considering two prominent functional forms capturing simple guilt aversion and reciprocity, respectively.

Simple guilt aversion. Simple guilt aversion as in Battigalli and Dufwenberg (2007) implies that player i judges the initial expectations of player j concerning his material payoff and feels guilty whenever he does not live up to these expectations. Given his strategy sjλ(hj(x0)) and the first-order belief system αj, player j forms an initial expectation about his material payoffs πj:

Esjλ(hj(x0)),αj[πj|hj(x0)]=siλ(hj(x0))Siλ(hj(x0))αjsiλ(hj(x0))|hj(x0)πjzsjλ(hj(x0)),siλ(hj(x0)).

For a given terminal node z the function

Djz,sjλ(hj(x0)),αj=max0,Esjλ(hj(x0)),αj[πj|hj(x0)]πj(z)

measures how much player j is “let down”. Player i does not know player j’s strategy and first-order beliefs, but holds a belief about these. Denote player i’s belief about player j’s “let down” by Dij(z,sjλ(hji(x0)),βi), where βi is player i’s second-order belief system. Given this, player i is motivated by simply guilt aversion if his psychological payoffs are represented by:

(1)ui(z,μi)=πi(z)θijDijz,sjλ(hji(x0)),βi

where θij reflect player i’s sensitivity to guilt. Guilt averse player i senses a psychological cost connected to his feeling of guilt in case he does not live up to his belief about player j’s initial expectation and takes this into account when deciding on his optimal behavior. Players’ feelings of guilt thus depend on their awareness, and their beliefs about the awareness of the other player.

Reciprocity. Different from guilt aversion, reciprocity as in Dufwenberg and Kirchsteiger (2004) assumes that players judge the kindness of others. Whenever player i judges player j to be kind, he reciprocates by being kind himself. Whenever player i judges player j to be unkind, he acts unkindly in return. More formally, given a singleton information set hj, a strategy sjλ(hj) that reaches hj and the first-order belief system αj, player j forms an expectation about player i’s material payoff πi:

Esjλ(hj),αj[πi|hj]=siλ(hj)Siλ(hj)αjsiλ(hj)|hjπizsiλ(hj),sjλ(hj)

Player j’s kindness towards player i is described by how much player j expects to give player i relative to some equitable payoff πie(hj) at singleton information set hj:

Kj(hj)=Esjλ(hj),αj[πi|hj]πie(hj).

The equitable payoff is the threshold or neutral payoff above (below) which player j treats player i kindly (unkindly). In other words, if Kj()>0, then player j treats player i kindly. Conversely, if Kj()<0, then player j treats player i unkindly. Let the equitable payoff be

πie(hj)=12×maxsiλ(hj)Siλ(hj)Esjλ(hj),αj[πi|hj]+minsiλ(hj)Siλ(hj)Esjλ(hj),αj[πi|hj].

The equitable payoff is the average player j is able to give to player i in material terms based on his awareness and first-order belief. As in the case of guilt aversion, player i does not know strategy sj and first-order beliefs αj, but holds a first- and second-order belief about them. Denote player i’s judgement of player j’s kindness at information set hi by Kji(hji)=Esjλ(hji),βi[πi|hji]πie(hji), where βi is player i’s second-order belief system. We say player i is motivated by reciprocity if he has belief-dependent preferences represented by a psychological payoff function of the form:

(2)ui(z,μi|hji)=πi(z)+Yi×Kji(hji)×πj(z),

where Yi>0 is player i’s sensitivity to reciprocity. Whenever player i perceives player j to be kind, player i is motivated to also maximize player j’s material payoff. In case player i judges player j to be unkind, player i is motivated to reduce player j’s material payoff. This definition of reciprocity implies that if player i is unaware of paths of play, he judges the kindness of player j based on the paths that he is aware of.

4 Extensive-Form Rationalizability

We adapt to the present framework the extensive-form rationalizability concept of Pearce (1984). To do so, we first extend the notion of sequential rationality and then redefine the best-rationalization principle as outlined by Battigalli (1997) and Battigalli and Siniscalchi (2002).

Sequential rationality. Our basic behavioral assumption is that player i chooses and carries out a strategy si that reaches singleton information set hi and is optimal given his hierarchy of beliefs μi, conditional upon any singleton information set consistent with si. It is thus not required that a strategy specifies behavior at singleton information sets that cannot be reached by si.

Fix a singleton information set hi. Player i’s expectation of the psychological payoff ui, given si and μi is

Esi,μi[ui|hi]=sjλ(hi)Sjλ(hi)αisjλ(hi)|hi×uizsiλ(hi),sjλ(hi),μi.

Given a hierarchy of beliefs μiMi, strategy si is a sequential best response to μi if for all hiIi:

siargmaxsiSi(hi)Esi,μi[ui|hi].

For any hierarchy of beliefs μi, let BRi(μi) denote the set of strategies si that are sequential best responses to μi. The set of best responses thus consists of strategies si of player i that, for a given μi, are undominated at every singleton information set hiHi, given his awareness at hi.

Clearly, BRi is nonempty valued. The first-order belief αi(|hi) is continuous. Since ui is also continuous, we have that αi(|hi)×ui(,) is continuous, which implies that BRi is an upper hemicontinuous correspondence.

To see how the extension of sequential rationality works, consider Bob in Figure 3. Bob’s strategy sB={cd}, for example, is sequential rational if it is undominated by his other strategies sˆB{cc,dc,dd} at every hBHB. Bob is unaware when he is active at singleton information set hB(x1). Here his strategy sB={cd} is undominated if his induced strategy sBT={c} gives him an expected psychological payoff that is equal or higher than his induced strategy sˆBT={d}, given Ann’s induced strategy sAT={C} that reach hB(x1). If Bob is active at hB(x2), then he has become aware. Here his strategy sB={cd} is undominated if it gives him an expected psychological payoff that is equal or higher than his strategies sˆB{cc,dc,dd}, given Ann’s strategy sA={CD} that reach hB(x2). Thus, Bob’s strategy sB={cd} is sequential rational if, and only if, it is undominated at both hB(x1) and hB(x2).

Best-rationalization principle. Our redefinition of the best-rationalization principle requires that players’ beliefs conditional upon observing singleton information set hi be consistent with the highest degree of strategic sophistication of the other player. In the following we clarify what we mean by strategic sophistication in terms of dynamic psychological games with unawareness. Consider the following extensive-form rationalization procedure for player i:

Mi[1]=Mi,Ri[1]=siSisuchthatthereexistsahierarchyofbeliefsμiMi[1]forwhichsiBR(μi),Mi[k]=μiMi[k1]suchthatforallsingletoninformationsetshiHi,ifRj[k1](hi),thenαi(Rjλ(hi)[k1]|hi)=1,Ri[k]=siSisuchthatthereexistsahierarchyofbeliefsμiMi[k]forwhichsiBR(μi).

Let Ri[]=k0Ri[k]. Player i’s strategies in Ri[] are said to be extensive-form (correlated) rationalizable in a dynamic psychological game with simple unawareness.

Intuitively our definition of extensive-form rationalizability starts at the level of strategic thinking of player i, whose step 1 rationalizable strategies are sequential best responses to some nonrestricted hierarchy of beliefs. Strategies that are not sequential best responses to any nonrestricted hierarchy of beliefs are not step 1 rationalizable.

Next, player i restricts his hierarchies of beliefs to those for which he, at each singleton information set hi, is certain (given his awareness) of those step 1 rationalizable strategies of player j that reach hi. Player i then chooses step 2 rationalizable strategies that are sequential best responses to these restricted hierarchies of beliefs. Strategies that are not sequential best responses to these restricted hierarchies of beliefs, on player j’s step 1 rationalizable strategies, are not step 2 rationalizable.

Furthermore, player i must restrict his restricted hierarchies of beliefs to those for which he, at each singleton information set hi, is certain (given his awareness) of those step 2 rationalizable strategies of player j that reach hi. Player i then chooses step 3 rationalizable strategies that are sequential best responses to these twice restricted hierarchies of beliefs. Strategies that are not sequential best responses to these twice restricted hierarchies of beliefs, on player j’s step 2 rationalizable strategies, are not step 3 rationalizable, and so on.

The sequence {Ri[k]:k0} is a weakly decreasing, that is, Ri[k+1]Ri[k] for all k. Since Ri is finite the sequence converges in countably many steps. The limit is given by the first integer K such that Ri[K]=Ri[K+1]. Consider the limit set Ri[K] of player i. The sequence Rj[0],Rj[1],,Rj[K1] represents a hierarchy of increasingly strong hypotheses of player i about the behavior of player j. When player i implements a strategy siRi[K], he always optimize accordingly. At the beginning of the game, it is the common knowledge that all players update and behave in this way.

The set Mk (for k>1) implies that along each feasible path of play, at a singleton information set an active player is certain that the other player sequential best responds, certain that the other player is certain he sequential best responds, and so on. If a player finds himself at a succeeding singleton information set, where the other player’s strategies that could lead to that singleton information set are inconsistent with the player’s previous certainty in the other player’s best response, then the player seeks a best rationalization that could have led to that singleton information set. That is, if the player is “surprised” by the other player’s unexpected action, and cannot use Bayesian updating, then he forms new beliefs that justify this observed inconsistency. In its simplest form, forward-induction reasoning involves the assumption that, upon observing an unexpected (but undominated) action of the other player, a player maintains the working hypothesis that the latter is a sequential rational. The best-rationalization principle captures precisely this type of argument.

Forward-induction reasoning thus implies that at hB(x0) and onwards, unaware Bob is certain that Ann’s sequential best response is sAT={C}. However, if singleton information set hB(x2) is reached and Bob becomes aware, then he is certain given his newly found awareness that Ann’s action sA={CD} is a sequential best response to some hierarchy of beliefs of hers. At hB(x2), Bob has no choice but to revert to being certain that Ann would not choose the strategy sA={CC} rationally, and excludes Ann’s hierarchies of beliefs for which sA={CC} is a sequential best response.

5 Two Examples

In the following we present two examples to highlight the impact and importance of simple unawareness in dynamic psychological games. In particular, our examples demonstrate that the strategic behavior of players motivated by psychological payoffs can be effected by unawareness even in situations where it would not have mattered had the player been selfish. In these situations the aware player often has a subtle incentive to manage the awareness of the unaware player by disclosing paths of play if possible. In the first example, we analyze a sequential prisoners dilemma featuring unawareness and reciprocity. Second, we re-visit our introductory Ann/Bob-example featuring guilt aversion. A full description of the strategic interaction with all possible awareness levels is beyond the scope of this paper. Therefore, we limit the analysis to specific awareness scenarios.

Example 1 (reciprocity). Consider the the extensive-form underlying the sequential prisonors dilemma with unawareness depicted in Figure 3 now with Ann’s and Bob’s material payoffs added.

Remember, at the initial node x0, Ann is active and can corporate (the action C) or defect (the action D). At nodes x1 and x2, Bob is active and can corporate (the action c) or defect (the action d). In the strategic setting depicted in Figure 4, Ann is initially aware of everything, whereas Bob is initially unaware. Ann knows this, and knows that he only becomes aware of everything if she chooses D. Also, Ann knows that Bob perceives her to be unaware. The “solid arrows” indicate Ann’s singleton information sets, while the “broken arrows” indicate Bob’s singleton information sets. For the sake of simplicity, we omit Bob’s redundant singleton information sets at y0 and y1.

Figure 4: Sequential prisoners dilemma game with unawareness.
Figure 4:

Sequential prisoners dilemma game with unawareness.

We assume that Bob is motivated by reciprocity and Ann is selfish. Bob’s psychological payoff is thus described by eq. [2], while Ann only cares about her material payoff πA(z). Ann’s optimal strategy depends on her first-order belief about Bob strategy (conditional on her behavior). Her strategy sA={CC} is a sequential best response as long as her expected material payoff from strategy sA={CC} exceeds her expected material payoff from strategy sA={CD}. Her expected material payoff from strategy sA={CC} is:

E{CC},αA[πA|hA(x0)]=αA({c}|hA(x0))1+(1αA({c}|hA(x0)))1,

where αA({c}|hA(x0)):=sB{cc,cd}αA(sB|hA(x0)) is a shorthand notation for Ann’s first-order belief about the strategies of Bob that select the action c at the singelton information set hB(x1). Ann’s expected material payoff from following strategy sA={CD} is:

E{CD},αA[πA|hA(x0)]=αA({c}|hA(x0))2,

where αA({c}|hA(x0)):=sB{cc,dc}αA(sB|hA(x0)) is an akin shorthand notation. The step 1 rationalizable strategies of Ann are thus

RA[1]={sA:αA({c}|hA(x0))αA({c}|hA(x0))12sA={CC},
otherwisesA={CD}}.

Bob, on the other hand, is initially passive and only becomes active at singleton information sets hB(x1) and hB(x2). Remember that at singleton information set hB(x1), Bob is unaware and holds a first-order belief αB(sAT|hB(x1)) and a second-order point belief βB(hB(x1)) concerning Ann’s first-order belief. Given his unawareness at hB(x1), Bob thinks that Ann’s only action is C. Independent of his second-order belief and his sensitivity to reciprocity YB he thus judges Ann as neither kind nor unkind (because KBAB(hB(x1))=0). Consequently, whenever Bob finds himself at hB(x1) he will simply maximize his own material payoff by choosing d.

At singleton information set hB(x2), Bob is aware of everything and knows that he could have earned a material payoff of 2 had Ann initially chosen to keep him unaware by choosing C. Bob’s judgment of Ann’s intention towards him at hB(x2) is thus

E{CD},βB[πB|hB(x2)]=βB({c}|hB(x2))1,

where βB({c}|hB(x2)):=sB{cc,dc}βB(hB(x2)) is a shorthand notation for Bob’s second-order belief about the likelihood with which Ann believes he chooses c at information set hB(x2). Clearly, 2>E{CD},βB[πB|hB(x2)] independent of second-order belief βB(hB(x2)). Hence, Bob judges Ann as unkind when finding himself at hB(x2). The step 1 rationalizable strategies of Bob are thus

RB[1]={sB:forallβB,YBsB={dd}}.

Although Bob is motivated by reciprocity as defined in eq. [2], his behavior in our sequential prisoners dilemma with unawareness is independent of his (second-order) beliefs and independent of his sensitivity to reciprocity YB.

Bob’s set of step 1 rationalizable strategies is a singleton set RB[1]={dd}. Ann is thus at all her singelton information sets certain that Bob follows strategy sB={dd}. That is, Ann’s hierarchy of beliefs μAMA[2] are all such that αA({dd}|)=1. Being certain that Bob chooses d no matter what, Ann’s sequential best response strategy must also select D as an action (since 0>1). The step 2 rationalizable strategies of Ann are thus

RA[2]={sA:forallμAMA[2]sA={CD}}.
Ann anticipates that given his awareness, Bob judges her as unkind and chooses d independent of what she does. Consequently, since Ann is only interested in her own material payoff, she chooses D herself to get a material payoff of 0 instead of 1.

To study the impact of unawareness, we compare this scenario to the rationalizable solution of the sequential prisoners dilemma with reciprocity and full awareness (see Figure 5).

Figure 5: Sequential prisoners dilemma with full awareness.
Figure 5:

Sequential prisoners dilemma with full awareness.

Now Ann chooses to C in the initial singleton information set hA(x0) as long as she believes sufficiently strongly that Bob will choose c. Her expected payoff from choosing either sA={CC} or sA={CD} at hA(x0) is the same as before, and her step 1 rationalizable strategies are again:

RA[1]={sA:αA({c}|hA(x0))αA({c}|hA(x0))12sA={CC},otherwisesA={CD}}.
Bob’s optimal behavior at hB(x2) remains the same as before. That is, Bob chooses d out of material and reciprocal reasons. However, Bob’s optimal behavior at hB(x1) now depends on his sensitivity to reciprocity YB. Let Bob’s (second-order) belief about Ann’s belief concerning the likelihood with which he chooses c at hB(x1), be βB({c}|hB(x1)):=sB{cc,cd}βB(hB(x1)). The step 1 rationalizable strategies of Bob are thus:
(3)RB[1]={sB:βB({c}|hB(x1))21YBsB={cd},otherwisesB={dd}}.

The lower Bob’s second-order belief is, the kinder he perceives Ann’s strategy sA={CC} (because KBAB=112βB({c}|hB(x1))), which provides him with a payoff which is higher than if she had chosen strategy sA={CD}. At hB(x1), Bob never actually thinks Ann is unkind. The question at this augmented history simply is whether he thinks she is kind enough, given his sensitivity to reciprocity YB, such that he prefers to reciprocate her kindness.

If Bob’s sensitivity to reciprocity is low (YB12), such that he for sure chooses d if Ann chooses C, then Ann will choose strategy sA={CD} as this provides her with a higher expected material payoff. Conversely, if Bob’s sensitivity to reciprocity is high (YB1), Ann is certain that Bob chooses c if she chooses C. Given this, she chooses strategy sA={CC} as this provides her with a higher expected material payoff. Notice, if Bob is sensitive enough to Ann’s kindness, then he chooses c at hB(x1) independent of his second-order belief. Given this Ann also chooses C, something she would not do were she sure that Bob would be unaware. Based on RB[1], Ann’s beliefs are

MA[2]={μA:forallYB1αA({c}|hA(x0))=1,forallYB<12αA({c}|hA(x0))=0}.

At step 2 reasoning, Ann is certain that a very sensitive Bob will always choose c if she also chooses C, whereas a very insensitive Bob will always choose d no matter what. For intermediate sensitivity levels, 12YB<1, her beliefs are MA[2]=MA[1]. Ann’s step 2 rationalizable strategies are thus:

RA[2]={sA:forYB1sA={CC},for12YB<1,γA({c}|hA(n0))21YBsA={CC},otherwisesA={CD}}.

where γA denotes Ann’s (third-order) point belief about Bob’s second-order belief. Based on RA[2], Bob’s step 3 beliefs are

MB[3]={μB:forYB1βB({c}|hb(x1))=1,forYB<12βB({c}|hB(x1))=0,otherwiseδB({c}|hB(x1))21YB,βB({c}|hB(x1))=1}.

where δB denotes Bob’s (fourth-order) belief about Ann’s third-order beliefs. MB[3] implies that Bob believes that Ann expects him to choose sB={cd} whenever he finds himself at singleton information set hB(x1). Bob’s step 3 rationalizable strategies are thus:

RB[3]={sB:forYB1sB={cd},otherwiseYB<1sB={dd}}.

If Bob is sensitive enough to reciprocity (YB1), he always chooses c if Ann chooses C, and chooses d if Ann chooses D. If he is insensitive enough to reciprocity (YB<12), then he always chooses d no matter what Ann does. However, if his sensitivity to reciprocity is 12YB<1 and he finds himself at hB(x1), then he is certain that Ann believes that he will choose c (because βB({c}|hB(x1))=1). That is, 1>21YB and he defects although he believes that Ann believes that he will choose c. Finally, Ann’s step 4 rationalizable strategies are:

RA[4]={sA:forallYB1sA={CC},forallYB<1sB={CD}}.

With full awareness, if Bob is sufficiently sensitive to reciprocity (YB1), then Ann chooses C since this induces Bob to choose c. This stands in contrast to the result of the first awareness scenario in which Bob was unaware of Ann’s possibility to defect even after her choice C. This example highlights that, although reciprocity is only based on first- and second-order beliefs, the recursive nature of extensive-form rationalizability requires the specification of higher (potentially infinite) orders of beliefs.

In synthesis, although Bob’s sensitivity to reciprocity might be very high, his behavior in the first scenario stands in contrast to the result in the second senario. With full awareness Bob’s behavior following Ann’s action C depends on Bob’s sensitivity to reciprocity. For sufficiently high levels of sensitivity Bob reciprocates by choosing c. With unawareness Bob’s behavior is independent of his sensitivity to reciprocity. Bob simply defects as he perceives Ann’s action as unkind no matter what she does. As a consequence, also Ann’s behavior is qualified – she defects as well. Interestingly, Bob behaves as if he is selfish, although he is not. It is only his subjective perception concerning the strategic environment which drives his behavior.

It is at the intersection of these two scenarios that the implications of unawareness for the behavior of people motivated by psychological payoffs become most visible. If Bob were only interested in his own material payoff, his behavior in the two scenarios would be the same. Most importantly, being only interested in his material payoff Bob would choose d following Ann’s decision to choose C independent of whether there are different awareness levels (as in the first scenario) or not (as in the second scenario). It is only his psychological payoff which explains the above-described difference in behavior. Ann thus has a strong incentive to make Bob aware and choose C. As long as Bob is unaware that Ann could have defected, he will not perceive her corporation as kind. However, were Ann able to make Bob aware, then she (as well as Bob) would have been better off. This shows, similar to our example in the introduction, how the opportunity to disclose paths of play is important when players are motivated by psychological payoffs.

Example 2 (guilt aversion): Consider again the example from the introduction featuring Bob’s birthday party and Ann’s final exam at university. Assume Bob would like to organize a party and would be happy if Ann could come. Bob gets a material payoff of 2 from not organizing the party (denoted NP), he gets a material payoff of 3 from organizing the party (denoted P) if Ann can join in, and a material payoff of 1 if she cannot make it. Thus, he would rather not organize the party if Ann cannot make it. Furthermore, assume that Bob is unaware of the second date, and hence, he is certain that she cannot make it.

Suppose that Ann knows what Bob is aware of, thus she knows that he is certain that she has an exam the day after his birthday party and cannot come, but in addition she also knows that she could actually write the final exam 2 months later at a second date. Denote her decision to write the exam on the first date (the day after the party) by No 1 and the second date by No 2. She can only attend his party if she decides to write the exam on date No 2, if either Bob does not organize a party (NP) or Ann chooses No 1, she writes the exam on the first date. Assume Ann’s material payoff from not going to the party and writing the exam on the second date is 4, whereas her material payoff from going to the party and writing the exam on the second date is 2. Figure 6 depicts this simple unawareness game. Assume Ann is motivated by simple guilt aversion as described by eq. [1] with θ>2, and that Bob is only interested in his own material payoff. The “solid arrows” now indicate Bob’s singleton information sets, while the “broken arrows” indicate Ann’s singleton information sets. Again we omit redundant singleton information sets.

Figure 6: Example 2 with guilt aversion.
Figure 6:

Example 2 with guilt aversion.

Before looking at this unawareness game consider – as a benchmark – a situation of full awareness. With full awareness the rationalizable solution is that Bob chooses P and Ann chooses No 2. The reason is the following: Bob only chooses P if it implies an expected material payoff which is higher than 2. That is,

RB[1]={sB:αB(No2|hB(x0))12sB={P},otherwisesB={NP}}.

If Ann finds herself at singleton information set hA(x1) she has to believe that Bob’s belief regarding the likelihood that she chooses No 2 is αB(No2|hB(x1))12. That is, Ann’s second-order belief is βA(No2|hA(x1))12. Ann will choose to postpone the exam and attend the party (No 2) as long as this gives her a higher expected psychological payoff. Since θ>2 it holds that:

RA[1]={sA:βA(No2|hA(x1))12sA={No2}}.

Since Bob can be certain that Ann will choose No 2 if he chooses P this is the only rationalizable outcome.

Consider now the case with unawareness on Bob’s side. Bob would only choose to organize a party if he sufficiently believed that Ann could come. Because Bob is unaware of the second exam date, he is certain that he will get a material payoff of 1 if he organizes the party and 2 if he does not. Bob’s only rationalizable strategy is to not organize the party (that is, choose NP). Ann knows that unaware Bob is not “let down” if she chooses No 1 at hA(x1) following Bob’s decision to organize the party, that is αB(No2|hB(x0))=0 which implies βA(No2|hA(x1))=0 so that DAB=0. Ann is therefore certain that she will get a psychological payoff of 4 if she takes the exam on the date No.  1. If Ann takes the exam on the second date then she is certain that she will not feel guilt, however she will only get a psychological payoff of 2. Ann’s only rationalizable strategy is to take the exam on the first date (that is, choose No 1).

Like the previous example featuring reciprocity, also this example with guilt aversion demonstrates the impact of unawareness on players motivated by psychological payoffs and highlights the subtle incentive to disclose paths of play to the unaware player. In this game Ann has no incentive to make Bob aware of the second date as long as she is sure that he will never become aware of it because this leaves her with a payoff of 4 instead of 2. Interestingly, were Ann only interested in his own material payoff, she would not be concerned about Bob being or not being aware of her action No 2. Ann would in any case choose to write the exam on date No 1 irrespective of Bob’s awareness.

6 Conclusion

We have analyzed the influence and importance of simple unawareness concerning feasible paths of play for the strategic interaction of players in dynamic psychological games, and defined a two-player model in which players are motivated by psychological payoffs and simple unawareness of certain feasible paths of play. Using this model we provide different examples highlighting the role of unawareness in the strategic interaction of players motivated by reciprocity à la Dufwenberg and Kirchsteiger (2004) and guilt aversion à la Battigalli and Dufwenberg (2007). Our examples demonstrate that the strategic behavior of players motivated by belief-dependent preferences crucially depends on their awareness concerning the strategic environment they are in, their perception concerning the awareness of others, their perception concerning the perception of others etc. In other words, unawareness has a profound and intuitive impact on the strategic interaction of players motivated by psychological payoffs – a fact that creates both an opportunity as well as a challenge to empirically investigations analyzing the strength and nature of psychological payoffs. Concentrating on two-player environments and simple awareness scenarios obviously puts limits to the strategic situations that can be analyzed with our model. Nevertheless our simple model has allowed us to uncover intriguing effects. More general strategic environments with more complex awareness scenarios are left for future research.

Acknowledgements

We would like to thank the editor, Burkhard Schipper, and an anonymous referee for their very helpful comments. We are also grateful to Geir Asheim, Pierpaolo Battigalli, Mie la Cour Sonne, Martin Dufwenberg, Aviad Heifetz, Georg Kirchsteiger, Peter Norman Sø rensen, Lars Peter Ø sterdal and participants at the EDGE Jamboree in Dublin, the DGPE Workshop in Copenhagen, the Econometric Society’s European Winter Meeting in Rome, the Royal Economic Society’s Postgraduate Meeting in London, the Seminar in Microeconomics in Lund, the CNEE Workshop in Copenhagen and the CES seminar in Leuven for helpful comments and suggestions. Both authors gratefully acknowledge the financial support from the Danish Council for Independent Research in Social Sciences (Grant ID: DFF-4003-00032). All errors are our own.

References

Andreoni, J., and B. Bernheim. 2009. “Social Image and the 50–50 Norm: A Theoretical and Experimental Analysis of Audience Effects.” Econometrica 77 (5):1607–36.10.3982/ECTA7384Search in Google Scholar

Battigalli, P. 1997. “On Rationalizability in Extensive Games.” Journal of Economic Theory 74 (1):40–61.10.1006/jeth.1996.2252Search in Google Scholar

Battigalli, P., and M. Dufwenberg. 2007. “Guilt in Games.” The American Economic Review, Papers and Proceedings 97 (2):170–6.10.1257/aer.97.2.170Search in Google Scholar

Battigalli, P., and M. Dufwenberg. 2009. “Dynamic Psychological Games.” Journal of Economic Theory 144:1–35.10.1016/j.jet.2008.01.004Search in Google Scholar

Battigalli, P., and M. Siniscalchi. 2002. “Strong Belief and Forward Induction Reasoning.” Journal of Economic Theory 106 (2):356–91.10.1006/jeth.2001.2942Search in Google Scholar

Bellemare, C., A. Sebald, and M. Strobel. 2011. “Measuring the Willingness to Pay to Avoid Guilt: Estimation Using Equilibrium and Stated Belief Models.” Journal of Applied Econometrics 26 (3):437–53.10.1002/jae.1227Search in Google Scholar

Broberg, T., T. Ellingsen, and M. Johannesson. 2007. “Is Generosity Involuntary?.” Economics Letters 94 (1):32–7.10.1016/j.econlet.2006.07.006Search in Google Scholar

Charness, G., and M. Dufwenberg. 2006. “Promises and Partnership.” Econometrica 74 (6):1579–601.10.1111/j.1468-0262.2006.00719.xSearch in Google Scholar

Dana, J., D. Cain, and R. Dawes. 2006. “What You Don’t Know Won’t Hurt Me: Costly (but Quiet) Exit in Dictator Games.” Organizational Behavior and Human Decision Processes 100 (2):193–201.10.1037/e722842011-057Search in Google Scholar

Dana, J., R. Weber, and J. Kuang. 2007. “Exploiting Moral Wiggle Room: Experiments Demonstrating an Illusory Preference for Fairness.” Economic Theory 33 (1):67–80.10.1007/s00199-006-0153-zSearch in Google Scholar

Dekel, E., B. Lipman, and A. Rustichini. 1998. “Standard State-Space Models Preclude Unawareness.” Econometrica 66:159–73.10.2307/2998545Search in Google Scholar

Dufwenberg, M., and G. Kirchsteiger. 2004. “A Theory of Sequential Reciprocity.” Games and Economic Behavior 47 (2):268–98.10.1016/j.geb.2003.06.003Search in Google Scholar

Fagin, R., and J. Halpern. 1988. “Beliefs, Awareness, and Limited Reasoning.” Artificial Intelligence 34:39–76.10.1016/0004-3702(87)90003-8Search in Google Scholar

Falk, A., E. Fehr, and U. Fischbacher. 2008. “Testing Theories of Fairness–Intentions Matter.” Games and Economic Behavior 62 (1):287–303.10.1016/j.geb.2007.06.001Search in Google Scholar

Falk, A., and U. Fischbacher. 2006. “A Theory of Reciprocity.” Games and Economic Behavior 54 (2):293–315.10.1016/j.geb.2005.03.001Search in Google Scholar

Fehr, E., G. Kirchsteiger, and A. Riedl. 1993. “Does Fairness Prevent Market Clearing? an Experimental Investigation.” The Quarterly Journal of Economics 108 (2):437–59.10.2307/2118338Search in Google Scholar

Fehr, E., and K. Schmidt. 1999. “A Theory of Fairness, Competition, and Cooperation.” The Quarterly Journal of Economics 114 (3):817.10.2307/j.ctvcm4j8j.14Search in Google Scholar

Feinberg, Y. 2004. “Subjective Reasoning Games with Unawareness.” Stanford University Research Paper (1875).10.2139/ssrn.624672Search in Google Scholar

Fudenberg, D., and J. Tirole. 1991. Game Theory. Cambridge: MIT Press.Search in Google Scholar

Geanakoplos, J., D. Pearce, and E. Stacchetti. 1989. “Psychological Games and Sequential Rationality.” Games and Economic Behavior 1 (1):60–79.10.1016/0899-8256(89)90005-5Search in Google Scholar

Grant, S., and J. Quiggin. 2013. “Inductive Reasoning About Unawareness.” Economic Theory 54 (3):717–55.10.1007/s00199-012-0734-ySearch in Google Scholar

Halpern, J. 2001. “Alternative Semantics for Unawareness.” Games and Economic Behavior 37:321–39.10.1006/game.2000.0832Search in Google Scholar

Halpern, J., and L. Rêgo. 2006. “Extensive Games with Possibly Unaware Players.” In: Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multiagent Systems, ACM, pp. 744–751.Search in Google Scholar

Halpern, J., and L. Rêgo. 2008. “Interactive Unawareness Revisited.” Games and Economic Behavior 62:323–262.10.1016/j.geb.2007.01.012Search in Google Scholar

Halpern, J., and L. Rêgo. 2009. “Reasoning About Knowledge of Unawareness.” Games and Economic Behavior 67 (2):503–25.10.1016/j.geb.2009.02.001Search in Google Scholar

Heifetz, A., M. Meier, and B. Schipper. 2006. “Interactive Unawareness.” Journal of Economic Theory 130:78–94.10.1016/j.jet.2005.02.007Search in Google Scholar

Heifetz, A., M. Meier, and B. Schipper. 2008. “A Cannonical Model for Interactive Unawareness.” Games and Economic Behavior 62:304–24.10.1016/j.geb.2007.07.003Search in Google Scholar

Heifetz, A., M. Meier, and B. Schipper. 2013a. “Dynamic Unawareness and Rationalizable Behavior.” Games and Economic Behavior 81:50–68.10.1016/j.geb.2013.04.003Search in Google Scholar

Heifetz, A., M. Meier, and B. Schipper. 2013b. “Unawareness, Beliefs, and Speculative Trade.” Games and Economic Behavior 77:100–21.10.1016/j.geb.2012.09.003Search in Google Scholar

Lazear, E., U. Malmendier, and R. Weber. 2012. “Sorting in Experiments with Application to Social Preferences.” American Economic Journal: Applied Economics 4 (1):136–63.10.1257/app.4.1.136Search in Google Scholar

Li, J. 2009. “Information Structures with Unawareness.” Journal of Economic Theory 144 (3):977–93.10.1016/j.jet.2008.10.001Search in Google Scholar

Modica, S., and A. Rustichini. 1999. “Unawareness and Partitional Information Structures.” Games and Economic Behavior 27:265–98.10.1006/game.1998.0666Search in Google Scholar

Pearce, D. 1984. “Rationalizable Strategic Behavior and the Problem of Perfection.” Econometrica 52 (4):1029–50.10.2307/1911197Search in Google Scholar

Rabin, M. 1993. “Incorporating Fairness Into Game Theory and Economics.” The American Economic Review 83 (5):1281–302.10.2307/j.ctvcm4j8j.15Search in Google Scholar

Sebald, A. 2010. “Attribution and Reciprocity.” Games and Economic Behavior 68 (1):339–52.10.1016/j.geb.2009.07.004Search in Google Scholar

Published Online: 2016-7-12
Published in Print: 2017-1-1

©2017 by De Gruyter

Downloaded on 28.3.2024 from https://www.degruyter.com/document/doi/10.1515/bejte-2015-0011/html
Scroll to top button