You are on page 1of 15

Journal of Economic Behavior & Organization 106 (2014) 352–366

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization
journal homepage: www.elsevier.com/locate/jebo

Strategic complexity and cooperation:
An experimental study夽
Matthew T. Jones ∗
Federal Trade Commission

a r t i c l e

i n f o

Article history:
Received 5 November 2013
Received in revised form 8 July 2014
Accepted 14 July 2014
Available online 22 July 2014
JEL classification:
C93
C73
D03
Keywords:
Complexity
Prisoner’s dilemma
Repeated game
Bounded rationality
Finite automata

a b s t r a c t
This study investigates whether cooperation in an indefinitely repeated prisoner’s dilemma
is sensitive to the complexity of cooperative strategies. An experimental design which
allows manipulations of the complexity of these strategies by making either the cooperate
action or the defect action state-dependent is used. Subjects are found to be less likely to use
a cooperative strategy and more likely to use a simpler selfish strategy when the complexity of cooperative strategies is increased. The robustness of this effect is supported by the
finding that cooperation falls even when the defect action is made state-dependent, which
increases the complexity of punishment-enforced cooperative strategies. A link between
subjects’ standardized test scores and the likelihood of cooperating is found, indicating that
greater cognitive ability makes subjects more likely to use complex strategies.
Published by Elsevier B.V.

1. Introduction
To implement a strategy in a repeated game, a player must process and respond to information she receives from her
environment such as the behavior of opponents, the state of nature, etc. Intuitively, one can say that the complexity of
a repeated game strategy depends on the amount of information that must be processed to implement it. For example,
consider a repeated oligopoly pricing game in which firms set a price in each stage after receiving information about demand
conditions and the prices set by rivals. To use a competitive pricing strategy, a firm sets its price equal to a constant marginal
cost in each stage. To use a collusive pricing strategy, a firm sets its price conditional on the demand state as well as the
prices set by rival firms. Hence, the collusive pricing strategy can be called more complex because implementing it involves
processing more information. If there are costs associated with this information processing in the form of management
compensation, operating costs, etc., they can affect the firm’s pricing strategy choice and make a relatively complex collusive
strategy less likely to be used. Similarly, cognitive costs associated with information processing may influence repeated game
strategy choice on the individual level, yielding important consequences for cooperation and efficiency.

夽 This work is supported by the NSF under Grant No. SES-1121085. Any opinions, findings and conclusions or recommendations expressed are those of
the author and do not necessarily reflect the views of the Federal Trade Commission or the NSF.
∗ Correspondence to: 600 Pennsylvania Avenue, NW, Mail Drop HQ-238, Washington, DC, United States. Tel.: +1 202 326 3539.
E-mail address: mjones1@ftc.gov
http://dx.doi.org/10.1016/j.jebo.2014.07.005
0167-2681/Published by Elsevier B.V.

Rubinstein (1986) shows that incorporating strategic complexity costs into the preferences of players in the infinitely repeated prisoner’s dilemma causes the efficient cooperative equilibrium to unravel. 1993) and public goods games (Palfrey and Rosenthal. which indicate a positive relationship between cognitive ability and cooperation. Accounting for strategic complexity can also have important implications in the study of market games. with the position of the dominated action varied between tables. 2.M. (2011) find that cooperation in this game depends on the payoffs and continuation probability. the complexity of strategic implementation is increased through random switching between permutations of a three-by-three version of the prisoner’s dilemma within each repeated game. the dominated action is permuted with the defect action instead of the cooperate action so that defecting requires subjects to account for random switching between payoff tables. this treatment also reduces cooperation compared to the baseline. while Dal Bo and Frechette (2011a) find that subgame perfection and risk dominance are necessary but not sufficient conditions for cooperation. Others have studied cooperation in related environments. one of the two payoff tables is drawn randomly and publicly announced to apply in that stage. Both of these manipulations increase the complexity of cooperative equilibrium strategies. Gale and Sabourian (2005) consider a market game with a finite number of sellers. which normally has both competitive and non-competitive equilibria. and Section 4 reports the experimental results. Because cooperating requires subjects to account for random switching between tables in order to choose the correct action. Feinberg and Husted. This feature of the design can be viewed as increasing complexity by requiring subjects to condition their action choices on observable changes in the state of nature in order to use certain types of strategies. The robustness of this effect is supported by the results of another treatment which increases the complexity of cooperative strategies in a different way. Relative to the baseline. Section 3 defines the research questions. In this treatment. Cooperation in the indefinitely repeated prisoner’s dilemma has been the subject of many experimental studies. Section 5 concludes. In this experiment. These results provide evidence that cognitive costs associated with strategic complexity can have an impact on cooperation and efficiency. such as indefinitely repeated oligopoly games (Holt. cooperative strategies are more complex in this treatment than in a baseline treatment in which the positions of the cooperate..T. Hence. Subjects are paid their cumulative earnings from both phases at a conversion rate of $0. and subjects with greater cognitive ability should be more able to bear the cognitive cost associated with this complexity. 2008). and Duffy and Ochs (2009) provide evidence that the indefinitely repeated prisoner’s dilemma fosters cooperation because it allows players to use punishment-enforced cooperative strategies. This relationship is consistent with the idea that cognitive costs of strategic complexity affect strategy choice because cooperative strategies are generally more complex than playing selfishly. but to my knowledge this is the first study to find such a relationship at the individual level in a repeated prisoner’s dilemma. and Blonski et al. and both make subjects less likely to use a cooperative strategy and more likely to use a simpler selfish strategy. Each session of these treatments is broken into two phases. cognitive costs associated with implementing complex strategies may discourage cooperation and harm efficiency. Though the manipulation affects cooperation in a less obvious way.1 but to my knowledge none has studied how strategy choice in this game might be affected by limitations on strategic complexity. Before each stage of a repeated game. the positions of the cooperate and dominated actions are permuted between the two payoff tables. . in repeated games where efficiency depends on players adopting relatively complex cooperative strategies rather than simple selfish strategies. 1994) as well as prisoner’s dilemmas with costly punishment (Dreber et al. In this paper. I find that increasing strategic complexity in this way reduces cooperation. 2009) and noisy implementation of intended actions (Fudenberg et al. as subjects have a greater tendency to adopt a simple selfish strategy in this treatment than in the baseline. and show that only the competitive equilibria remain with strategic complexity costs. Murnighan and Roth (1983). 1985. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 353 The theoretical literature suggests that strategic complexity is a practical equilibrium selection criterion in repeated games with both cooperative and selfish equilibria. but to my knowledge a theoretical model accounting for strategic complexity has not heretofore been tested empirically or experimentally. I present the results of an experiment designed to test how behavior in an indefinitely repeated prisoner’s dilemma depends on the complexity of available strategies.. I investigate this question using a design which allows manipulations of the implementation complexity of strategies by making either the cooperate or defect action state-dependent. The paper proceeds as follows. imperfect monitoring (Aoyagi and Frechette. Dal Bo (2005). The idea that cognitive costs of strategic complexity affect cooperation is further supported by data on subjects’ American College Test (ACT) and Scholastic Aptitude Test (SAT) scores. plus a 1 Roth and Murnighan (1978). this treatment increases the implementation complexity of strategies that support cooperation through the threat of punishment. defect and dominated actions are the same in both tables.004 per Experimental Currency Unit (ECU). Camera and Casari (2009). Each treatment employs two payoff tables with a strictly dominated action choice added to the cooperate and defect actions. In one treatment. Fershtman and Kalai (1993) show that collusion in a multi-market duopoly may be unsustainable when strategic complexity is bounded. A correlation between average SAT scores in the subject pool and aggregate cooperation was found by Jones (2008) in a metastudy of prisoner’s dilemma experiments. Section 2 describes the experimental design. Experimental design The experiment includes three main treatments. These results demonstrate that limitations on strategic complexity can have important consequences. 2012).

6 The experimental software is programmed in zTree (Fishbacher. which is always strictly dominated by the defect action and at least weakly dominated by the cooperate action. Payoff tables. while in all stages after the first the default is the action choice entered in the previous stage. for a total time limit of 60 seconds per stage. Of the two payoff tables in use in a particular treatment. 1. The focus of the paper is on the first phase of these sessions. In addition. 3 In the experimental instructions (see Section D of the Online Appendix). Each stage of a repeated game is broken into three parts. See Figs.354 M. in which subjects play one repeated game at a time. but drawn independently in each session. The third screen in stage t reports the action selected by the opponent in stage t along with the player’s payoff in that stage. to the standard 2-by-2 prisoner’s dilemma I add a third action choice (1 in table Z. and the chosen table is publicly announced before the stage begins. The payoff table that applies in stage t has been announced previously. The number of stages in each repeated game was drawn randomly prior to the first session. as in practice complexity costs are 2 In the second phase. subjects are told which of the two payoff tables applies in the first stage. but it is not shown at this time. Subjects are matched with the same opponent for the duration of each repeated game. and 2 in tables X and Y). D. which was randomly determined as described to subjects.2 Subjects participate in a series of seven indefinitely repeated prisoner’s dilemma games. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Fig. 2007). 4 Dominance is weak in table X and strict in tables Y. subjects are told that the number of stages (or “periods” as they are called in the instructions) in each round is random. so cooperate is action 1.3 Each treatment uses two of the four payoff tables shown in Fig. one is randomly chosen to apply in each stage of a repeated game. This feature is included so that time constraints in the experiment are binding. and the dominated action is action 2 in both tables of this treatment.6 Suppose the current stage is stage t. 6. Subjects can see both possible payoff tables for that treatment on their instructions at all times. 2 or 3) in stage t. 1. 3-by-3 versions of the prisoner’s dilemma.5 Before the beginning of each repeated game. with an 80% continuation probability in each stage.T. and the same round lengths are used in each session. Treatment D-SWITCH uses Y and Z . which are symmetric. Discussion of this treatment is also relegated to Section A of the Online Appendix. and matches are determined randomly and independently of matches in previous games. but which of the two payoff tables applies in stage t is not visible on their computer screens after it is announced. 8. with an 80% continuation probability in each stage of a repeated game. 7 and 4 in each session. This part of the experiment is discussed in Section A of the Online Appendix. Treatment C-SWITCH uses Y and Z. 2. which correspond to three screens that appear on subjects’ computer monitors. each session begins with a 5-stage trial game which does not count towards payment. The first screen asks for an action choice (1. A default choice is automatically entered if subjects fail to enter a choice before time expires in each stage. Z and Z . . subjects play multiple repeated games simultaneously. defect is action 3. For reasons explained in Section 3.4 Treatment BASE uses tables X and Y. is 4. 1. In the first stage of a repeated game the default choice is the defect action. The second screen in stage t announces which of the two payoff tables will apply in stage t + 1 if the game continues beyond stage t (whether it will continue or not is not yet revealed). 3 in table Z . 5 The random sequence of payoff tables is not the same in all sessions. so defect is 3 in both tables but the positions of the cooperate and dominated actions are permuted between 1 and 2 between the two tables in this treatment. so cooperate is 1 in both tables but the positions of the defect and dominated actions are permuted between 2 and 3 between the two tables in this treatment.1–D.3 in Section D of the Online Appendix for screenshots of the three stage screens. Each of these screens is viewable for 20 seconds. Sessions of a fourth treatment were also conducted to control for order effects on behavior in the first and second phases. participation fee of $6. The sequence of repeated game lengths.

where 1 is the initial state. an initial state. (1. See Johnson (2006a) and Salant (2011) for applications of finite automata in models of boundedly rational individual choice. The efficient cooperative equilibrium unravels when lexicographic complexity costs are added. (1. in which the transition function takes as its domain the cross-product of states and only the opponent’s observed action rather than the actions of both players. nor does it address the complexity associated with decisions under uncertainty because it is defined in a complete information environment.2% of choices in SWITCH-C. who show that the number of states in the minimal finite automaton implementing a strategy is equal to the number of different subgame strategies the original strategy induces.10 The simple measure of strategic complexity provided by this model is the minimal number of states that can be contained in Q such that the automaton can implement the strategy. the default action was triggered because time expired for only a small number of action choices: 1. mapping from states and the opponent’s last observed action. a behavior function. compared to strategies that condition action choices on the opponent’s action or payoff table announcement.11 The finite automata definition of strategic complexity has been used by Rubinstein (1986) and Abreu and Rubinstein (1988) to show that the set of equilibria of the infinitely repeated prisoner’s dilemma is drastically reduced when complexity costs are added to players’ preferences. and 3. but they are not instructed to do so. Q. Since Tit-for-Tat has two states and Always Defect only one. The automaton implementing Always Defect in this game is represented by the set of states. Instead. the default action choice feature of the experimental design (see Section 2) reflects the finite automaton remaining in the same state when the strategy it represents calls for no change in the action choice between stages. further increasing information processing requirements. This concept of complexity does not capture the complexity associated with computing optimal actions and strategies.8 The payoff table that applies in stage t + 1 is revealed before the opponent’s action in stage t so that strategies which condition on opponent behavior require additional contingent planning. Its effect is to reinforce the relative ease of implementing simple constant-action strategies which require no information processing during a repeated game. Another interpretation of this definition is due to Kalai and Stanford (1988). Increasing the amount of information that must be processed means that the strategy-implementing automaton must contain more states to keep track of its environment and follow a given plan of how to react to incoming information. which can affect the strategies players adopt. q0 ∈ Q. See Hopcroft and Ullman (1979) for a standard reference on automata theory. (2) = D. An automaton implementing a cooperative equilibrium strategy is more complex than one implementing a strategy of Always Defect because more states are needed for an automaton to enforce cooperation by monitoring the opponent and punishing defection. where 1 is the initial state. D} where C is cooperate and D is defect.4% of choices in SWITCH-D. These studies demonstrate that the standard folk theorem results may not be robust to strategic complexity costs. the behavior function.M. C) = 1. but these are not incorporated into the data. Tit-for-Tat is a more complex strategy. D) = 2. The automaton implementing Tit-for-Tat in this game is represented by the set of states.9 This approach measures the implementation complexity of a repeated game strategy by the number of states in the minimal finite automaton which implements the strategy. which require information processing in each stage. it measures the complexity of information processing or monitoring required to implement a strategy. A repeated game strategy is modeled in terms of finite automata as follows. and must enter a new action choice only when the automaton representing their strategy transitions between states. . 7 Overall. and the state transition function defined by (1. For example. 2}.T. Q = {1. s ∈ S. D) = (1. (1) = D. Because the payoff table that applies in stage t + 1 is announced before that stage begins and not shown when stage t + 1 choices are entered. and the distance between this outcome and the most efficient achievable outcome under this concept increases as the discount factor falls. A. 9 This approach was first suggested by Aumann (1981). 10 This type of finite automaton. I also motivate my analysis of how subjects’ ACT and SAT scores are related to their propensity to use cooperative strategies.  : Q → A. mapping from states into the set of possible actions. into states. The concept of strategic complexity I adopt in designing and analyzing this experiment is the theory of games played by finite state automata (see Chatterjee and Sabourian. A strategy consists of a finite number of machine states. subjects may passively continue to take the same action. 3.  : Q × S → Q. C) = 1. Q = {1}. I explain the theory behind the hypothesized effects of treatments C-SWITCH and D-SWITCH. consider a strategy of Tit-for-Tat played in a standard prisoner’s dilemma in which the set of possible actions is A = S = {C.7 The above cycle repeats for each stage of a repeated game. and the state transition function. or the fineness of the partition of the game history required to implement a strategy which conditions on that history. 11 In terms of the finite automaton model. 3. D) = (2. is known as a Moore machine. BASE. which increase the complexity of cooperative strategies compared to the control treatment. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 355 always linked somehow or another to time constraints. Research questions This experiment is designed to determine whether the level of cooperation in an indefinitely repeated prisoner’s dilemma is sensitive to the complexity of strategies that sustain cooperation. strategies that condition action choices on the payoff table announcement require additional information processing. 8 Roughly 50–60% of subjects in each treatment took notes.6% of choices in BASE. C) = (2. and a state transition function. After the first stage of a repeated game. Subjects are allowed to take notes on paper to help them recall information. the behavior function defined by (1) = C. In this section. 2009 for a recent survey).

Rather than showing which equilibria are robust to complexity costs and which unravel. The design of this experiment and the timing with which information is given to subjects make it possible to manipulate the complexity of cooperative equilibrium strategies. There is a more general sense in which these treatments increase the cognitive load or amount of information processing required of subjects. 14 The concept of risk dominance applies to this game after elimination of the dominated action. They show that in a majority-rule division game.’s (2012) study of strategic commitment in a repeated duopoly game.T. which is estimated to explain a significant proportion of the experimental data. or 3) that is the output of the behavior function when taking that state as the input. which are labeled by the action choice (1. Hence. Regarding complexity measures. which the number of states in an automaton may not fully represent. he finds that a cost associated with the number of states in the automaton harms cooperation. The tables differ only in the payoff if the action profile (2. but their effect on subjects may not be captured exactly by this model. Volij (2002) shows that with lexicographic complexity costs in an evolutionary setting. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Fig. Nodes represent the states of the automaton. which collapses the payoff matrix to 2 × 2. as well as Selfish Tit-for-Tat (STFT). the cooperate action is 1. and these strategies are risk-dominant compared to a strategy of Always Defect (AD). Always Defect (AD). Banks and Sundaram (1990) propose that the number of state transitions in an automaton should be considered as well as the number of states. the strategies players choose to be implemented by machines are relatively simple and frequently unconditional on the opponent’s actions. such that it is not expected to affect strategy choice. with Grim Trigger as the most successful automaton strategy.13 Though the finite automata model is a popular approach to strategic complexity around which my experimental design is built. 2 shows the directed graph representations of the minimal finite automata that implement selected strategies in BASE. 13 A related experimental work is Embrey et al. 2. as measured by the number of states in their minimal finite automata representations. Though the complexity of players’ strategies is not the focus of that study.12 highlighting the need for empirical evidence to inform further research on the implications of strategic complexity costs. Directed graphs of selected automaton strategies in BASE. Others have used the finite automata model to refine equilibrium concepts or to provide a more nuanced measure of strategic complexity. or 3) or the payoff table announcement (X or Y) as inputs and mapping to the output state. Johnson (2006b) defines the complexity of a strategy in terms of the algebraic properties of its minimal automaton representation. Fig. but action 2 is strictly dominated by defect and at least weakly dominated by cooperate in both tables. where all possible divisions can be subgame perfect equilibria. in terms of finite automata. In both of its payoff tables (X and Y). that with the simplest equilibrium strategies is the one selected. . Grim Trigger (GT) and Tit-for-Tat (TFT). Though it may not be completely accurate. 2. BASE includes two possible payoff tables and an announcement of which applies before each stage of a repeated game. In particular. In this treatment the payoff table manipulation is designed to be innocuous. the only stochastically stable automaton is the one-state automaton which always defects. Edges represent the state transition function taking the current state and either the opponent’s action (1. an objective measure of complexity is needed to formalize comparisons between treatments and provide testable predictions. it should not be taken too literally. symmetric and intuitive. The evolutionary fitness of repeated game strategies played by finite automata has also been studied analytically and computationally with mixed results.356 M. the parameters of the game are set to encourage a moderate level of baseline cooperation. Cooper (1996) finds that a folk theorem result is restored with a different definition of evolutionary stability and finite complexity costs. In contrast with these results. simple strategies commonly observed in repeated prisoner’s dilemmas have the same complexity in this treatment as in the two-by-two prisoner’s dilemma. which would reflect more precisely the amount of monitoring necessary in implementing a strategy. a simulation by Linster (1992) using a different algorithm and smaller strategy space shows convergence towards cooperation. Baron and Kalai (1993) use a finite automata approach to provide an equilibrium selection criterion in repeated games: when multiple equilibria exist.15 12 Binmore and Samuelson (1992) consider an evolutionary model of the indefinitely repeated prisoner’s dilemma played by finite automata with lexicographic complexity costs and find that the equilibrium automata reach the efficient cooperative equilibrium. The circled node is the initial state. Lipman and Srivastava (1990) measure complexity by a strategy’s sensitivity to small perturbations in the history of play. so the action to take in each stage for a given strategy is the same for both tables.14 To maintain the overall structure of the experiment across treatments and provide a valid control. To allow a meaningful comparison between treatment and control. Cooperation through the use of Grim Trigger (GT) and Tit-for-Tat (TFT) strategies is supportable as an equilibrium. Players in this experiment are given the opportunity to construct Moore machines to implement strategies for them. Finite automata can be thought of as a metaphor for the number of different “states of mind” that are possible when implementing a particular repeated game strategy. but a cost associated with the number of state transitions does not. and the finite automata model is the most appropriate model available. 2. This estimation is explained in detail in Section 4. the dominated action is 2 and the defect action is 3. A simulation by Ho (1996) finds that convergence towards cooperation depends critically on the specification of complexity costs. In contrast. The treatments described below increase the complexity of cooperative strategies in terms of finite automata. 15 The four strategies selected include three common equilibrium strategies.2.2) is selected. the equilibrium selected by this criterion is efficient. State transitions can occur when the payoff table for the next stage is announced or when the opponent’s action in the current stage is revealed.

M. 2. the additional information processing required to cooperate in this setting is non-trivial. Because the payoff table that applies in a given stage does not appear on the screen at the time when the action choice is entered. dependent? Compared to the baseline. 2 states). If one thinks of payoff table switching as random changes in the state of nature. Again. Because the cooperate action is conditional on the payoff table announced in a given stage. In other words. compared to the control. while the defect action is not because it is action 3 in both tables. a greater amount of information processing. choosing the correct defect action requires players to account for which payoff table is announced to apply in that stage. 3. enforcing cooperation through punishment requires contingent planning and. the number of states is not as important as the idea that more information processing is required to play cooperatively in C-SWITCH than to play cooperatively in BASE or selfishly in either treatment. The simple selfish strategy. 3 shows the minimal finite automata that implement selected strategies in C-SWITCH. or 3) that is the output of the behavior function when taking that state as the input. 2. while the cooperate action is not because it is action 1 in both tables. Edges represent the state transition function taking the current state and either the opponent’s action (1.T. are subjects less likely to adopt cooperative strategies when the cooperate action is state-dependent? In C-SWITCH. additional automaton states are needed to implement strategies that sustain cooperation. Nodes represent the states of the automaton. Using such a strategy in D-SWITCH requires a player to account for which payoff table will apply in the next stage in case the opponent defects in the current stage. Hence. this treatment may also reduce cooperation because subjects expect that implementing cooperative strategies will be prohibitively costly to opponents. the model should be viewed metaphorically rather than literally. One can say that the defect action is state-dependent in this treatment. but the complexity of the cooperative strategies increases by a greater margin (4 states vs. Because defecting involves choosing conditional on the payoff table announcement in this treatment. Question 2. one can say that the cooperate action is state-dependent in this treatment. requires only a 1-state automaton in both C-SWITCH and BASE. Fig. are subjects less likely to adopt cooperative strategies when the defect action is state- In D-SWITCH. a model in which the likelihood of choosing a strategy is negatively related to its complexity would predict less cooperation in D-SWITCH than in BASE. AD. Question 1. thus triggering punishment in the next stage. The finite automata model formalizes this idea. Because next stage’s payoff table is announced before the opponent’s action in the current stage is revealed. Though the finite automata model provides crisp comparisons between strategies by the number of states. a model of convex complexity costs would predict less cooperation in this treatment than in the control. the simplest selfish strategy of AD is more complex in D-SWITCH (2 states) than in BASE (1 state). These indirect effects would result from the same increase in implementation complexity measured by the finite automata model. State transitions can occur when the payoff table for the next stage is announced or when the opponent’s action in the current stage is revealed. Fig. Compared to the baseline. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 357 Fig. any model in which the likelihood that a strategy is adopted is negatively related to its complexity (as measured by finite automata) would predict less cooperation in C-SWITCH than in BASE. In addition to the direct effect on a subjects’ strategy choice due to cognitive costs. whereas they can be implemented by a 2-state automaton in BASE. Directed graphs of selected automaton strategies in C-SWITCH. Hence. This manipulation may affect the level of cooperation because it increases the implementation complexity of punishment-enforced cooperative strategies. These strategies require a finite automata with at least 3 states in C-SWITCH. Hence. or 3) or the payoff table announcement (Y or Z) as inputs and mapping to the output state. the defect action is action 3 in table Y and action 2 in table Z . the cooperate action is action 1 in table Y and action 2 in table Z. the exact predictions of the finite automata model are a formalization of the more important main idea: that this . or that they will be more likely to make mistakes due to the additional information processing. The circled node is the initial state. unless only the relative complexity of available strategies matters. Hence. 4 shows the minimal finite automata that implement selected strategies in D-SWITCH. which are the strategies that support cooperation in equilibrium. choosing the correct cooperate action in a given stage requires players to account for which payoff table is announced to apply in that stage. which can be thought of as the action of a third player (nature). which are labeled by the action choice (1.

with the most cooperation in BASE and the least in D-SWITCH. See Benjamin and Shapiro (2005). or 3) that is the output of the behavior function when taking that state as the input.. Branas-Garza et al. 2011). 16 Grim Trigger requires 4 states in D-SWITCH because the payoff table announcement for the next stage occurs before the opponent’s action in the current stage is revealed. . They also find that such subjects are more cooperative in a trust game similar to a one-shot prisoner’s dilemma and that they have more accurate beliefs about how others behave. (2007).18 and it also allows me to test a corollary to the idea that complexity costs influence strategy choice. with the highest prevalence in BASE and lowest in D-SWITCH. (2009) find that experimental subjects with higher cognitive ability are more patient and also more willing to take calculated risks.. Though the mechanism for these results is not precisely identified. Burks et al. 2. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Fig. (2013) find that subjects with high SAT scores in their endogenous-timing investment experiment are more likely to respond as predicted to informational externalities. In contrast.. to the extent that subjects are inclined to use GT subject to complexity costs. and Jones (2013) for other examples. However. and 4 states in D-SWITCH. which has a minimal finite automaton representation with 2 states in BASE. 2008. Edges represent the state transition function taking the current state and either the opponent’s action (1. In a meta-analysis of prisoner’s dilemma experiments. Ivanov et al. Burnham et al. Jones (2008) finds a positive relationship between aggregate cooperation and the average SAT score of the institution from which subjects are drawn. Agranov et al. which are labeled by the action choice (1. the player must remember whether to choose action 2 or 3 in stage t + 1. Hence. 2. Introducing additional tasks that are cognitively costly in prisoner’s dilemma experiments has been found to reduce cooperation. treatment increases the information processing required to play cooperatively. 3 states in C-SWITCH. When the opponent is observed to have defected in stage t. 4. Specifically.358 M. which may reduce cooperation relative to the baseline due to the associated cognitive costs. to my knowledge a relationship between individual-level cognitive ability and cooperation in repeated games has yet to be established in the literature. Cognitive ability is also found to affect strategies in public good games (Millet and Dewitte. Stevens et al. The circled node is the initial state. 17 This experiment is relevant to the literature documenting a broader relationship between cognitive ability and strategic behavior in games. Is there a relationship between subjects’ cognitive ability and their propensity to adopt a cooperative strategy? Extant experimental evidence suggests that individuals with higher cognitive ability are more likely to cooperate in a prisoner’s dilemma. This analysis contributes to the broader literature on individual characteristics that are associated with cooperation in repeated games. (2009). Question 3. These differences in the prevalence of GT may in turn lead to differences in cooperation rates between treatments. (2012) find little evidence of a relationship between strategic sophistication in other types of games and scores in several tests of cognitive ability. 2007) and the Traveler’s Dilemma (Branas-Garza et al. where lower guesses reflect greater strategic sophistication. two tendencies that support cooperation.T. 2012). Nodes represent the states of the automaton.16 Therefore. cooperation has been shown to fall when subjects are required to complete an unrelated memory task while playing (Milinski and Wedekind. they are consistent with the idea that artificially increasing the cognitive burden on subjects reduces their propensity to adopt the relatively complex strategies that support cooperation in a repeated game. we would expect a consistent ranking of treatments by the prevalence of this strategy. 18 This is one of only a few studies in the experimental economics literature to use verified ACT or SAT scores (as opposed to self-reported scores) as a measure of cognitive ability. Directed graphs of selected automaton strategies in D-SWITCH.17 I test for a relationship between cognitive ability and cooperation in this experiment by obtaining subjects’ consent to request their ACT and SAT scores from the Ohio State University registrar’s office. 1998. State transitions can occur when the payoff table for the next stage is announced or when the opponent’s action in the current stage is revealed. (2008) to be strongly correlated with general intelligence.. 2011). Casari et al. The three treatments in the experiment can be ranked by the complexity of the GT strategy. or 3) or the payoff table announcement (Y or Z) as inputs and mapping to the output state. two states that output cooperate are needed to “remember” the payoff table announcement and respond optimally in case the opponent defects. and when they must play a different repeated game simultaneously with the prisoner’s dilemma (Bednar et al. (2009) and Ivanov et al. (2011b). when they must rely on memory to track the actions of multiple opponents in a random sequence (Winkler et al. and Gill and Prowse (2012) find a negative relationship between cognitive ability and guesses in a p-beauty contest. Georganas et al. These test scores have been shown by Frey and Detterman (2004) and Koenig et al. (2011). Duffy and Smith. 2011a).

and cognitive costs associated with this complexity may discourage some individuals from cooperating. Hence. Stage outcomes and aggregate cooperation by treatment. I assess the statistical significance of differences in aggregate cooperation between treatments using a probit regression with an indicator variable for one of the treatments and standard errors clustered at the session level to account for arbitrary correlations between action choices in a session. this result would be consistent with the idea that strategic complexity is an important factor in strategy choice. 4. I also study the cooperation rate in the first stages of repeated games only.3% in BASE and 3. In a repeated prisoner’s dilemma. making them more likely to use complex strategies.T. as measured by the frequency of choices to cooperate among all action choices. **. but a 50% first stage cooperation rate. 5 for all three treatments.01 level. 5.2% in both C-SWITCH and D-SWITCH. This would lead to a 10% overall cooperation rate. 22 I also conduct probit regressions with random effects at the subject level to check whether observed treatment effects are robust to a more structured correlation between observations of the same subject. one player plays TFT and the other plays AD. They find that the guesses of more intelligent subjects are less influenced by the immediately previous signal and more influenced by earlier signals than their less intelligent counterparts. which indicates players’ intentions at the start of a repeated game before they observe and react to actions of their opponents. with two sessions and 34 subjects per treatment. reducing their prediction error. 5 also reports the overall cooperation and first stage cooperation rates in each treatment. 4. These results suggest a correlation between intelligence and the ability to use complex strategies in this repeated game against a computer. *. Difference from BASE significant at: ***. Strategies that sustain cooperation in equilibrium are more complex to implement than a simple selfish strategy. discussed in Section A of the Online Appendix. the correlation between cognitive ability and the ability to handle strategic complexity would lead to a higher frequency of cooperation among more intelligent subjects. 20 . Sessions lasted between 60 and 90 minutes with average earnings of $18. other things equal. Results Sessions were conducted at the Ohio State University Experimental Economics Lab in the winter and spring of 2011.1. Suppose that in a repeated game that lasts 5 stages. Though I cannot rule out other possible reasons for a positive relationship between cognitive ability and cooperation.1 in Section C of the Online Appendix. Aggregate cooperation I first study aggregate cooperation. other things equal.21 Fig. One would expect more cognitively able individuals to have a greater capacity to bear additional cognitive costs. Evidence of such a relationship is found in an experiment by Hawes et al.M. Compared to the baseline.85. A simple example illustrates this point.20 The distributions of stage-game outcomes on the payoff table are shown in Fig. A total of 102 subjects participated in the three main treatments of the experiment over six sessions. 21 The frequency of choosing the dominated action is 1.1 level. indicating that subjects quite accurately implement repeated game strategies that choose actions conditional on the payoff table announcement in C-SWITCH and D-SWITCH. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 359 Fig. were also conducted with a total of 34 subjects. The results of these regressions are reported in Table C. and thus indicates what type of strategy they adopt. less cooperation is observed in the first stage of a repeated game when the cooperate action is state-dependent. the first stage cooperation rate more accurately reflects the fact that 50% of players adopted a cooperative strategy at the start of the game.05 level.19 Subjects were recruited via email invitations sent randomly to students in a large database of Ohio State undergraduates of all majors. 19 Two sessions of a fourth treatment. both for all stages and for first stage outcomes only.22 Result 1. (2012) in which subjects are rewarded for recognizing patterns in computer-generated signals and correctly guessing the next signal.

which is significantly greater (at the 0. To the extent that subjects are inclined to 23 The first stage cooperation rate is similar to the 61.24 Fig.4% and 40. Both D-SWITCH and C-SWITCH increase the implementation complexity of cooperative strategies. The first stage cooperation rate in the last repeated game is 55.25 Result 2. 55.3% in BASE. Compared to the rate of cooperation observed in the first stages of BASE repeated games (55. where the cooperate action is state-dependent.1 in Section C of the Online Appendix) also reveal significant treatment effects. Compared to D-SWITCH. See Section A of the Online Appendix for a discussion of learning in phase II of the experiment. after subjects have experienced six other repeated games against randomly matched opponents.3%) than in BASE (40. Cooperation by round.23 I observe significantly less first stage cooperation in C-SWITCH (41. 6.6%). but the result is not significant for all rounds when standard errors are clustered to account for arbitrary correlations at the session-level.6%).27 This difference in overall cooperation rates is consistent with the difference between treatments in the complexity of the GT strategy.1% rate observed by Dal Bo and Frechette (2011a) in their treatment with parameters most similar to mine according to the cooperation indices of Murnighan and Roth (1983). which is greater than the 43. respectively (significant at the 0.1 in Section C of the Online Appendix) this treatment effect is significant for both first stage cooperation and overall cooperation.360 M. indicating that cooperation is more likely to be sustained over the course of the average repeated game in this treatment. 26 Probit regressions with random effects at the subject level (see Table C.4% vs. There is no clear trend in aggregate cooperation rates over this series of repeated games. Compared to the baseline.5% vs. 24 According to probit regressions with random effects at the subject level (see Table C. results of the experiment support the hypothesis that the complexity of the available strategies in a repeated game has a significant impact on which strategies players adopt. so learning does not appear have a marked effect on behavior in these games.T. . Compared to BASE. 40. Hence. as indicated by first stage actions.5%). both manipulations significantly reduce the frequency with which subjects adopt cooperative strategies. Because first stage actions signal players’ intentions before they begin reacting to their opponents. 6 shows levels of overall and first stage cooperation over the seven repeated games in a session in each of the three treatments.5%) in D-SWITCH. Treatment differences in the last repeated game alone are consistent with the results across all repeated games. less cooperation is observed when the defect action is state-dependent. 27 Neither of these differences is statistically significant according to probit regressions with standard errors clustered at the session-level.9% in BASE. 25 The overall rate of cooperation in the last repeated game is 49. respectively. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Fig. I observe significantly less overall cooperation (29.1% rates in C-SWITCH and D-SWITCH. significance of the effect is robust to within-subject correlation of choices. which requires 3 automaton states in C-SWITCH and 4 in D-SWITCH. Hence.4% rates in C-SWITCH and D-SWITCH.01 level. I also observe less overall cooperation in C-SWITCH (35. but using different manipulations. These pictures demonstrate that the complexity treatment effects on first stage actions are more pronounced and persistent over the series of games than the effects on all action choices. where multiple games are played simultaneously. but the difference in first stage cooperation is not significant. I observe slightly less first stage cooperation but more overall cooperation in C-SWITCH. It indicates that increasing the complexity of cooperative strategies through a statedependent cooperate action makes subjects less likely to use cooperative strategies.6%) and first stage cooperation (42. Nevertheless. According to a subject-level random effects regression. this result reveals a significant treatment effect on the type of strategies players adopt. but this difference is not statistically significant with standard errors clustered at the session-level.1 level) than both the 47. the difference between overall cooperation rates is statistically significant at the 0.26 This result indicates that increasing the complexity of punishment-enforced cooperative strategies through a state-dependent defect action makes subjects less likely to adopt these strategies.1% and 44.05 level for D-SWITCH only). This point is supported by differences between treatments in cooperation rates observed in the last repeated game.

36 (0.6% in the first stage of repeated games and increases to 48.. In one of the two C-SWITCH sessions.E. only three of the twenty candidate strategies are estimated to explain a significant proportion of individual treatment data: AD.16* 0.08) 0.  is the error term.18*** (0.08) (0.01 level.11) (0. (2012) in analyzing their experiment on prisoner’s dilemmas with exogenously imposed noisy implementation of intended actions. In Section 4.M.1.16* 0. The 28 29 See Section B of the Online Appendix for details of the first iteration and a description of the 20 candidate strategies in Table B. GT and STFT.29*** 0.E.07) Log-likelihood Gamma 0.48*** 0. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 361 Table 1 Maximum likelihood estimates of strategy prevalence: 4 candidate strategies. which repeats the same procedure using a set of only four candidate strategies: the three strategies that were significantly prevalent in the first iteration. However.06) 0.09) (0. In each stage of a repeated game. where K is the set of candidate strategies s1 . and  is the variance of the error.62*** (0.29 Therefore. and the log-likelihood function is maximized to estimate the proportion of the data explained by each candidate strategy. The resulting log-likelihood function has the form R T k k K p(s )pi (s )).48*** 527.5% rate in the BASE treatment.E. subjects in this session of C-SWITCH are much more likely to stabilize on cooperation over the course of a repeated game than in other sessions.10) (0.41** 0. (2012) to infer strategies from observed actions in their repeated prisoner’s dilemma experiments. with the addition of TFT due to its exceptional general popularity in prisoner’s dilemma experiments. Camera et al. Subjects’ underlying strategies can be estimated from their observed actions by a maximum likelihood technique developed by El-Gamal and Grether (1995) and extended to repeated game applications by Engle-Warnick and Slonim (2006) and Engle-Warnick et al. implement a GT strategy subject to complexity costs. I estimate how the prevalence of this and other strategies varies between treatments. In stage t of repeated game r. . I assume that subject i who uses strategy sk cooperates if the indicator function yirt (sk ) = 1{sirt (sk ) +  irt ≥ 0} takes a value of 1 and defects otherwise. the cooperation rate is 47. Results of the first iteration are reported in full in Table B.T. indicating that subjects in this session remain less likely to initially adopt a cooperative strategy than those in the control setting.16 413. I use two iterations of this technique to analyze the behavior of subjects in the experiment. A similar technique has been used by Aoyagi and Frechette (2009). *** . Estimate S.16) (0.09) (0. Each subject is assumed to use the same strategy in each repeated game. Dal Bo and Frechette (2011b). Strategy inference This study is concerned with the importance of complexity in strategy choice. Dal Bo and Frechette (2011a). The likelihood function of strategy sk for subject i has the logistic form  pi (sk ) =  {1/[1 + exp(−sirt (sk )/)]}  yirt {1/[1 + exp(sirt (sk )/)]} 1−yirt . average cooperation does not show the tendency to decline between the first stage and subsequent stages of repeated games as it does in all other sessions of the experiment. The maximum likelihood technique works as follows. . I conduct a second iteration. (2012).7% in stages after the first. Tit-for-Tat (TFT) Grim Trigger (GT) Always Defect (AD) Selfish Tit-for-Tat (STFT) 0.09) (0.27 524.62*** (0. we would expect to see greater overall cooperation in C-SWITCH. . Estimate S.08) 0. there is another explanation which at least partially explains the aggregate differences in behavior between C-SWITCH and D-SWITCH. sK and p(sk ) is the proportion of the data explained entire sequence of actions across repeated games is observed for each subject. . 4. Aside from differences in strategic complexity.26*** 0. Treatment BASE C-SWITCH D-SWITCH Estimate S. there is some probability that a subject deviates from the action prescribed by the chosen strategy. the first stage cooperation rate in this session is less than the 55.04) Significance level of Wald test for difference from zero: * . and Fudenberg et al. The first iteration uses a set of twenty candidate strategies. so aggregate results do not tell the whole story.14) 0. which are the same as those used by Fudenberg et al. Hence. The estimated coefficients of the second iteration and their bootstrapped standard errors are reported in Table 1. (2007).28 In this first iteration of estimates.12* (0. In this session.2.05 level.3.41*** 0.23** 0. where sirt (sk ) is the action prescribed by strategy sk (1 for cooperate and −1 for defect) given the history of repeated game r up to stage t. I ln( by sk .09) (0.11 0.2 in Section B of the Online Appendix. ** .09) (0.20 (0.1 level. This technique measures the proportion of each subject’s observed actions that can be explained by candidate repeated game strategies and estimates the prevalence of each candidate strategy by maximizing a log-likelihood function summing across all subjects and strategies.

E.231) (0. 4. BASE Estimate S. Six subjects were transfer students who reported neither test score.0019 (0.065 .401 0.550 0.006) (0.038) (0.E.0062 1024 (0.377*** . 32 See http://professionals. ** Significant at 0.0092) 0. Subjects are less likely to use a Grim Trigger strategy as the complexity of this strategy increases between treatments.048*** −0. I use two specifications: one with indicator variables for whether a subject has a test score in the top 5% of all test-takers or below the top 20% of all test-takers.362 M. with separate regressions for each treatment. This effect on strategies.294 0. Standard errors clustered at session level. variation in the estimated prevalence of GT between treatments is consistent with a ranking of treatments by the complexity of this strategy. Furthermore. I test for a relationship between cognitive ability and cooperation using probit regressions with the choice to cooperate as the dependent variable and regressors related to the subject’s ACT or SAT-ACT concordance score. where its minimal finite automaton has 3 states. the sets of coefficients estimated in each treatment are not jointly significantly different between treatments. Though GT’s estimated prevalence in D-SWITCH is not significantly different from its estimated prevalence in the other treatments.007) (.0084 (0. Analysis of ACT/SAT scores In this section. C-SWITCH Estimate S.011) (0.134** .082) (. SAT scores were obtained for 34 of the remaining 40 subjects.083 .0067*** (0. As the estimated prevalence of GT declines across treatments.0065) 217 0. and SAT-ACT concordance scores were used for these subjects. The estimated prevalence of GT is significantly greater than zero in both of these treatments.1 level). the estimated prevalence of AD increases.1 level) in BASE.pdf for SAT-ACT concordance tables. GT is most complex in D-SWITCH.193) (0. but differences in the prevalence of these strategies between treatments are not statistically significant.067) (.0070) 224 Excludes subjects with no ACT or SAT scores reported.107 −0. *** Significant at 0.05 level. 31 . Other reasonable score cutoffs do not produce symmetric tails.079*** −.E.442 0.062) (. ACT scores were obtained for 62 of the 102 subjects who participated in the three main treatments of the experiment.212*** .T.046) (0. These categories were selected because they represent roughly symmetric tails of the distribution of ACT scores for subjects in this experiment.32 For subjects reporting a test score. D-SWITCH Estimate S. where its minimal finite automaton has 4 states.369 0. According to a Wald test for joint differences. AD is significantly prevalent in all three treatments. Summary statistics are reported in Table C. by treatment.359*** −0.01 level.021) (.3. its point estimate is smallest in this treatment and not significantly different from zero. where its minimal finite automaton has 2 states. 30 I also conduct Wald tests for differences of individual coefficients from zero and differences in individual coefficients between treatments.0099*** 1056 992 (0. Interestingly. the estimated prevalence of the simple AD strategy is 22 percentage points greater in D-SWITCH than in BASE (marginally significant) and 15 percentage points greater in C-SWITCH than in BASE (not significant).055) (.1 level. Subjects are more likely to use a simple selfish strategy of Always Defect in treatments with increased complexity of cooperative strategies.209*** −0. Result 3. TFT is significantly prevalent in all three treatments (at the 0. These estimates are consistent with the prediction that subjects are more likely to adopt a simple selfish strategy when the complexity of implementing cooperative strategies increases.0026) 231 0.0009) 0. * Significant at 0.com/profdownload/act-sat-concordance-tables. I find some interesting differences in individual coefficients between treatments.30 However. than in C-SWITCH.023 0.2 in Section C of the Online Appendix.0067*** (0. The estimated prevalence of GT is significantly greater (at the 0.collegeboard.057 −0.429 0.0021) 1st Stage Mean ACT Top 5% ACT <Top 20% ACT Percentile (continuous) Observations 0. provides an explanation for the aggregate differences in overall cooperation between BASE and C-SWITCH and between C-SWITCH and D-SWITCH. I use ACT percentile as a measure of cognitive ability because ACT scores are based on a rank-order scale and not an additive scale. which is consistent with a model of costly strategic complexity. suggesting that as the complexity of GT increases subjects switch from this strategy to the simpler AD strategy. All Stages Mean ACT Top 5% ACT <Top 20% ACT Percentile (continuous) Observations 0. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Table 2 Marginal effects of test scores on cooperation. Table 2 shows the results of probit regressions with the choice to cooperate in any stage of a repeated game and in the first stage only as the dependent variables. while STFT is significant in BASE and D-SWITCH only.31 and one with the ACT percentile as a continuous variable.

038) (0.E.301* −0. Opp. Rd.018) 0. Coeff.089 −0.116) (0.188*** −0.033 −0. S.015 (0. these regressions fail to show strong evidence of a correlation between cognitive ability and cooperation in treatments that increase complexity of cooperative strategies relative to the baseline.019*** 0.138) (0.020 −0.019 0.009 −0.005) (0.065) −0.260*** −0.1 level. I conduct a series of four additional probit regressions which pool data from all three treatments and include a variety of controls.263*** 0.077 −0.134 −0.124) (0. Rd. S.071) 576 Excludes subjects with no ACT or SAT scores reported and results of first repeated game (used as regressor). Standard errors clustered at session level.368*** 0. and history of play on cooperation.M.045) (0.125** (0. Rd. Specification (1) includes only indicator variables for the C-SWITCH and D-SWITCH treatments.072) (0.032) (0.248*** 0.124** −0. 1 Opp.019 0. In order to isolate the relationship between test scores and cooperation and test for differences in this relationship between treatments.025) (0. test scores. Results of the specification using ACT percentile as a continuous variable are similar: ACT percentile has a significantly positive effect on cooperation overall and in the first stage. In the regressions using all stages of C-SWITCH. this evidence is consistent with the idea that strategy choice is influenced by cognitive costs of strategic complexity.-1st Stage. 1 Observations (2) (3) (4) Coeff. Prev.019*** 0.-1st Stage.074) (0.037) (0.087*** 0.242*** 0. Rd. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 363 Table 3 Marginal effects of treatments. Specification (2) adds a set of regressors that measure the effect of the history of play on the likelihood of cooperation.047) (0. while the coefficient on ACT percentile as a continuous measure is negative (but not significant).058) (0. # of Stages-Prev.029 576 (0.-1st Stage. both regressions indicate that having a test score in the top 5% of all test-takers increases the likelihood of cooperation significantly compared to those with a score in the top 20% but not the top 5%. as well as the number of rounds in the previous repeated game.-1st Stage.043) (0.027) (0. 1 Observations 1st stage C-SWITCH D-SWITCH ACT Top 5% ACT Top 5% × C-SWITCH ACT Top 5% × D-SWITCH ACT <Top 20% ACT <Top 20% × C-SWITCH ACT <Top 20% × D-SWITCH Coop.013) (0. Rd. For the treatments which increase the complexity of cooperative strategies beyond the baseline.033 (0.024 (0.126** −0.033) (0. Coop.115** 0.-1st Stage.161) (0.016) −0.027) (0.031) (0. including the subject’s action and the action of the subject’s opponent (1 if cooperate and 0 otherwise) in the first stage of the previous repeated game and the first stage of the first repeated game.160) (0.064) (0.107*** (0.083*** 0.079) 0.053) (0.084** −0.057 0.099) −0.085** −0.129) (0.218*** 0. Prev. Coop.057) (0.042) −0.032) 2688 −0.072) (0. Coop. Coeff.385*** −0.111) (0. Hence. Rd.043) (0.049 0.075) (0. Coeff. but only some of these are statistically significant.E.E.040 0.028) (0.276* −0. the omitted category.-1st Stage.043 −0.137* (0. Prev.E.079*** (0.258*** 0. *** Significant at 0.T. S.047) (0. Coop.039) (0. Result 4.022) (0.518*** 0.062) (0.103) (0.344*** 0.106) (0. # of Stages-Prev. the results are weaker and sometimes inconsistent with the above hypothesis.109*** −0.349*** −0.013) (0.043) (0.05 level.114* 0. 1 Opp. Variable (1) All stages C-SWITCH D-SWITCH ACT Top 5% ACT Top 5% × C-SWITCH ACT Top 5% × D-SWITCH ACT <Top 20% ACT <Top 20% × C-SWITCH ACT <Top 20% × D-SWITCH Coop.01 level.145) (0.004 −0.046 0.101 0.069** 0. two estimates are inconsistent with the expected relationship: the estimated effect of having a score below the top 20% is positive and significant.054) (0.-1st Stage.105*** −0.038) (0. I also find that having a test score below the top 20% decreases the likelihood of cooperation compared to the omitted category in this treatment.240*** 0.124*** 0.103) (0. Rd.335*** −0. Opp.041) (0.143*** 2688 (0.085 −0. Results of these regressions are reported in Table 3.049) 0. there is a positive relationship between ACT percentile and cooperation. ** Significant at 0. Specification (3) includes indicator variables for the complexity treatments and having an ACT score in the top 5% or below the top 20%. Hence. as well as terms for interactions between . Prev. Because cooperative strategies are relatively complex in this environment and players with high cognitive ability should be more able to bear the cognitive costs of using complex strategies. S. −0.132) (0. Rd.031) (0. Coop. Most of the estimated coefficients for C-SWITCH and D-SWITCH have the expected sign. For the BASE data.049) (0. In the baseline setting.006) (0.086) (0.180* 576 2688 (0.-1st Stage.047 −0. * Significant at 0.265*** 0. Coop. ACT scores provide strong evidence of a positive relationship between cognitive ability and cooperation in the standard repeated prisoner’s dilemma environment.022) 576 2688 −0. Rd. Rd.045) (0.495*** 0.107 −0.006) (0.

in all sessions of C-SWITCH and D-SWITCH the first stage cooperation rate is lower than in either session of BASE. Second. while Fig. this effect on cooperation does not confound the complexity interpretation of between-treatment differences.T.34 Result 5. even when controlling for ACT scores. Compared to these sessions. First. Together. which is observed on the aggregate level by Jones (2008) and on the individual level in this experiment. The likely reason for this difference between C-SWITCH and D-SWITCH is the greater tendency of subjects to stabilize on cooperation in one of the C-SWITCH sessions (discussed in detail below and in Section 4. The important implication of this result is that additional complexity attenuates the relationship between cognitive ability and cooperation in the standard prisoner’s dilemma. a breakdown of ACT scores and cooperation rates by session (see Table C. which is also the session with the lowest proportion of subjects with an ACT score in the top 5%. in most instances the C-SWITCH and D-SWITCH treatments interact significantly with the test score indicator variables with the opposite sign of the main test score effect. but lower cooperation rates prevail in the D-SWITCH session. having an ACT score above the top 5% significantly increases the propensity to cooperate. Specification (4) includes all of the regressors from (2) and (3). In terms of the minimal finite automata that implement these strategies. the D-SWITCH treatment effect remains significant. The number of stages in the previous repeated game is estimated to have a small but significant positive effect on the propensity to cooperate. The relationship between ACT scores and cooperation found in the baseline setting is mitigated in treatments that increase the complexity of cooperative strategies. In contrast. in the baseline setting the estimated relationship between cognitive ability and cooperation is robust to controlling for the history of play. Making the cooperate action state-dependent has a consistent. Because they are used as explanatory variables in some specifications. 5 because these regressions exclude the results of the first repeated game (used as a regressor). the data do not permit a precise explanation for the difference in the effect of ACT scores on cooperation across treatments. One interpretation would be that individuals with high cognitive ability are more likely to use cooperative strategies in a standard prisoner’s dilemma environment. Therefore. statistically significant. However. their greater tendency to cooperate may lead to differences between treatments independent of any complexity effect. Hence. While I cannot completely rule out this alternative explanation. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 the treatment and test score variables. though less statistically robust. All four of these specifications are estimated separately for the full data and for first stage actions only. However.3 in Section C of the Online Appendix) does not support this explanation. 33 Results of specification (1) differ somewhat from the overall treatment effects presented in Fig. which is not explained by test scores.33 The effect in D-SWITCH remains significant when additional regressors are included in specification (2). These results suggest that the relationship between the scores and cooperation is attenuated in the complexity treatments. and having a score below the top 20% significantly decreases the propensity to cooperate. Results of specification (1) confirm that the average effects of both complexity treatments on cooperation are negative and. but it disappears when they require automata of 3 or more states. session 2 of C-SWITCH has a higher proportion of subjects with ACT scores in the top 5% but lower cooperation rates. the behavior of the subject’s opponent in these past stages does not have a significant effect on the propensity to cooperate. In both specifications (3) and (4). Specifications (2) and (4) reveal that the propensity to cooperate in a given stage is correlated to a large and significant extent with whether the subject cooperated in the first stages of the previous repeated game and the first repeated game. the relationship between cognitive ability and cooperation appears to hold when cooperative strategies are implemented by 2-state automata. Of course. although the effect in C-SWITCH does not. The session with the highest overall cooperation rate is C-SWITCH session 1. indicating that subjects in the treatment sessions are less likely to adopt cooperative strategies than those in the control. Because the complexity treatment effects are attenuated when ACT scores are included in these regressions. for first stage actions. effect.364 M.1). as are the choices of subjects with no test scores reported. an alternative interpretation of the main experimental results seems plausible: if the proportion of subjects with high cognitive ability varies between sessions of the experiment. Because the experiment was primarily designed to seek evidence of a general relationship between strategic complexity and cooperation. . this objective measure should not be taken too literally. Hence. Session 2 of BASE and session 1 of D-SWITCH have similar ACT score distributions. Deeper understanding of what drives this result is an avenue for future research. 5 summarizes all action choices. choices in the first repeated game are excluded from the set of observations. increasing the complexity of cooperative strategies by making the defect action state-dependent has a robust cooperation-reducing effect. though I cannot rule out that differences in cognitive ability may be partially responsible for differences in cooperation rates between sessions. the complexity treatments appear to have a clearer and more consistent effect on strategy choice. but further increasing the implementation complexity of these strategies eliminates the advantage for these individuals. several features of the data are inconsistent with this hypothesis. 34 Because the same sequence of repeated game lengths is used in all sessions of this experiment. these results indicate that a subject’s choice to use a cooperative or a selfish strategy tends to be relatively stable across repeated games.

Prior research has established that additional cognitive costs such as external memory tasks (Milinski and Wedekind.. Brandon Restrepo. A natural extension would attempt to describe in more detail the relationship between cooperation and strategic complexity using similar manipulations to iteratively increase the complexity of cooperative strategies until cooperation stops.T. Analysis of subjects’ ACT and SAT scores reveals evidence of a link between cooperation and cognitive ability. This experimental design is grounded in an existing theoretical framework describing a potential mechanism for the above results: cognitive costs of information processing associated with strategic complexity. 2 (2). The process of choice in guessing games (Working paper). and the Kent State University business school. This manipulation increases the information processing necessary to implement strategies supporting cooperation. The structure of Nash equilibrium in repeated games with finite automata. On a practical level. but to my knowledge this study is the first to identify a link between cognitive ability and cooperation at the individual level in the repeated game. 2014. Zurich. C. Acknowledgements The author would like to thank Dan Levin and James Peck for valuable guidance and support. and complexity. anonymous referees. and seminar participants at the 2011 ESA International Meeting. G. John Kagel.. 1135–1165. Lixin Ye. 2011. Econometrica 56 (6). J. Banks.. 2012) can affect behavior in a repeated prisoner’s dilemma. This work benefited from comments by Asen Ivanov.1016/j. Vienna. For example. also deserves investigation. finite automata. 2011) and playing other games simultaneously (Bednar et al. This experiment contributes to the literature on how cognitive costs and abilities affect behavior in games by adapting a well-developed model of boundedly rational strategy selection to a familiar experimental setting.. Conclusion In this paper. R. this relationship supports the idea that strategy choice is influenced by cognitive costs of strategic complexity. M. A correlation between average SAT scores in the subject pool and aggregate cooperation levels was previously reported by Jones (2008) in a metastudy of prisoner’s dilemma experiments.J. Behav. Appendix A.S. Duffy and Smith. (Ed. 1990. Caplin. A. Finally. Mannheim. consider collusion in a duopoly with a fluctuating but publicly observable demand state. the results found using data on subjects’ ACT and SAT scores highlight the value of collecting such data in experimental research and point to another line of future work exploring in more depth the link between cognitive ability and cooperation found in this study. 97–117. Johnson. Ehud Kalai.. These results suggest several possible lines of future experimental research. Tergiman.). The present work addresses the importance of implementation complexity in strategy choice. M. Games Econ. 1998. and they may be particularly relevant to some specific applications.doi. Sundaram.J. Sustaining cooperation in the complex world in which we live often requires individuals to condition their actions not only on the behavior of others.07.. Survey of repeated games. 2009. Collusion as public monitoring becomes noisy: experimental evidence. Aumann. public goods or network formation games). A similar design could also be used to study the importance of strategic complexity in other games (for instance.005. The results suggest that the cognitive cost of implementation complexity can influence strategy choice and ultimately the efficiency of outcomes. Rubinstein.. Mark R. The experimental design simulates this source of complexity and shows that it can have an impact on cooperation. Frechette. experimental evidence such as this may help to improve the applicability of game-theoretic predictions to real world problems. sustaining collusion is less likely than in an environment with relatively constant demand. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 365 5.jebo. J. Econ. The model provides testable predictions which this experiment confirms with results that resemble those of other studies. the 2011 PEA Conference.K. .. where only the collusive price (if the competitive price is determined by a constant cost) or only the competitive price (if the collusive price is a constant “focal point” price) depends on demand fluctuations.. but the importance of a different but equally important type of complexity. Repeated games. but also on the observable state of nature. but the mechanism for this effect remains unknown. In: Aumann. computational complexity. Agranov. participants of the theory/experimental reading group at Ohio State. Supplementary data Supplementary data associated with this article can be found. A. Essays in Game Theory and Mathematical Economics in Honor of Oskar Morganstern. 144 (3).M. 1981. I study whether cooperation in the indefinitely repeated prisoner’s dilemma is sensitive to cognitive costs associated with strategic complexity. suggesting that the complexity of repeated game strategies carries information processing costs that interact with other sources of cognitive burden in influencing strategy choice. Bibliographisches Institute. Results indicate that increasing the complexity of cooperative strategies makes subjects less likely to adopt them. References Abreu. 1988. Strategic complexity in this game is increased through random switching between permutations of the payoff table during repeated games. D. R. Results of this experiment suggest that in either case.. at http://dx.org/10. Aoyagi. Theor. 1259–1281. The effect appears robust because cooperation is reduced regardless of whether the cooperate action or the defect action is manipulated to increase the complexity of cooperative strategies.. Because cooperative strategies are relatively complex and players with high cognitive ability should be more able to bear the cognitive costs of implementing complex strategies. R. in the online version.

Kalai. Conflict Resol.G...J.C. Rudolph.N. T. Finite automata play the repeated prisoner’s dilemma. 1484–1507. Bednar.. Cesarini. 1995.. Econ. 97 (4).. 1994.. Acad. J. G. J. Millet. strategic behavior. D. 2008. Econ. Supergames played by finite automata with finite costs of complexity in an evolutionary setting.. T.L.K..M. Blonski.. Wallace. 75 (3). J. In defense of DEFECT. J.H. 95 (5).. 99 (3). Sheremeta. and ability effects in common value auction experiments. Levin.. M. Engle-Warnick. Rev. Behav. E. Kagel. J.. K. M. A. 2009. Lipman.. Organ.. Rubinstein. J. Palfrey. 189–198. L.R. Proc. A. D. Econ. Sabourian. Finite automata play repeated prisoner’s dilemma with information processing costs. Grether. 2012. Games Econ. 1978. Nagel. Am. Theor. 2012. Simultaneous decision-making in competitive and cooperative environments. J. Forgetting constrains the emergence of cooperative decision strategies.. 2004. 233–245.. Procedural analysis of choice rules with applications to bounded rationality. 25–36.. Sheremeta. D. M.. J. 57 (2). 2010. M. vol.. 15 (6). Savikhin. G. 1993.. Are smarter groups more cooperative? Evidence from prisoner’s dilemma experiments.. Econ. Selection bias. 12–31. 58 (4). Econ. J.. T. Behav. Rieskamp.... Miller. 2006.: Microecon. Cognitive effort in the beauty contest game (Working paper). Complexity and competition. 1137–1145. Natl. Stat. 2008. Chatterjee. D. 2011. A. McCausland. J. 2012.. J.M. Does Cognitive Ability Reduce Psychological Bias? (Working paper). On the usefulness of memory skills in social interactions: modifying the iterated prisoner’s dilemma. Am. Detterman.. 2013. 2007. H. 2008. Ivanov. 785–812. M. T. Res. Roth. Burnham. Husted. M.A... Fershtman.. S. Branas-Garza. Natl. J. Econ. MA. Slow to anger and fast to forget: leniency and forgiveness in an uncertain world. L.R. An experimental test if the consistent-conjectures hypothesis. Econ. K.. Srivastava. Theor. Camera. Econ. Rev. Theor. A. Ham. K.M. 2007. Econ.C. Burks... Koenig. 83–96.. Cognitive skills affect economic preferences. Fishbacher. Johannesson.K. 411–429. J.R. 2010. M. Introduction to Automata Theory.. Econ. Assoc. 1996. Travelers’ types. M. 545–565. Southern Econ. 2011.A. Intelligence Moderates Responses to Gains and Losses by Enriching Task Representations (Working paper). Am.. E. Jones / Journal of Economic Behavior & Organization 106 (2014) 352–366 Baron.. 95 (23). R. Engle-Warnick. The evolution of cooperation in repeated games: experimental evidence. G. 1992.A. Murnighan. 164–192.. M. 2012. 397–410. Complexity considerations and market behavior. 570–586. Bigoni. Guth... Econ. Dal Bo. 68. Econ. M. Page. J. D. Prowse. Exp. z-Tree: Zurich Toolbox for ready-made economic experiments.. J..K. In: Encyclopedia of Complexity and System Science. O. 314–325. 2009. Languages. 2011... El-Gamal.. Psychol. R. Complexities and Aggregations (Working paper).. In: Meyers. J. (Ed.. Georganas. 1993. G. C. Front. Algebraic Complexity of Strategy-implementing Semiautomata for Repeated-Play Games (Working paper). Econ. Informational requirements and strategic complexity in repeated games.. Eur. 1983.G. Slonim. 54 (3). Hernan.. Dal Bo.. Springer.R. R. 273–290.E. Behav.. 2009. Gale. Sci. Behavioral spillovers in coordination games. Nowak. 41 (2).. J. J.. Scholastic assessment or g? The relationship between the SAT and general cognitive ability. Ind. Pers. J.... K. Goette. J. 72 (1). 90. L. Gangadharan.X. G. 1992..N. 102 (2). D. J.. 2012. 1998. An experimental test of discount-rate effects on collusive behaviour in duopoly markets. Higher cognitive ability is associated with lower entries in a p-beauty contest. Econ.. Mengel. Rev. 348–351. 51 (2). P.H.. A. 1986. H. Weber. J. Games Econ.). M. Binmore. Econ. 309–321. Nature 452. Am. Joseph.. J. 2011a. Organ. 45 (6). 1993... Cooperation spillovers and price competition in experimental markets. 2005. S. Hauk. Johnson. P. Rey-Biel. and insight: an experimental study of a small-market investment game with common and private values. Kirchkamp. Travelers’ types. 1591–1604... 106 (19)... Games Econ. A.. 1–12. 28 (3). 489–497.. J. T. Inferring repeated-game strategies from actions: evidence from trust game experiments. ACT and general cognitive ability. Inq. D. 724–748. S.. 2012. Cooperation under the shadow of the future: experimental evidence from infinitely repeated games. Detterman... D.. cooperation and coordination: an experimental study. 25–34.. V. . 1996. 1311–1323.M.. Rosenthal. Casari. Savikhin. L.. Rand J. 74 (1). Embrey. Fudenberg. 68 (1). D. 279–300.P. 1988. 1959–2003. Y. 10 (2).. Strategic commitment and cooperation in experimental games of strategic complements and substitutes (Working paper). Murnighan. 720–749. Rev. 171–178. 24 (2). Samuelson. 2011. S. 2007. 207–229. A.R. 2009. Am. J. W. 224–235.. 171–175. Y. Linster.. Cooperative strategies in anonymous economies: an experiment. Sci. “Nobody goes there anymore – it’s too crowded:” Level-k Thinking in the Restaurant Game (Working paper). Casari. J. J. 2013. M.K. H. On the Persistence of Strategic Sophistication (Working paper)... T. Hawes..G. Duffy. Salant. 2007. U. Winner’s don’t punish. Frey... Hindsight. Intelligence 36 (2). 87. 39 (2). Camera. S. R. Econ.. Repeated play.. S. Psychol.P. 316–326..A.. 373–378. Econ. Cooper. D. C. 13755–13758. 1979.. Working memory constrains human cooperation in the prisoner’s dilemma. Levin. Rev. Jones. Chen. 2009. S. 1278–1304. Frechette.D. Testing Forbearance Experimentally: Duopolistic Competition of Conglomerate Firms (Working paper). Am.A. J. Fudenberg. F.. Behav. U. Choice of partners in multiple two-person prisoner’s dilemma games: an experimental study. J. O. Cognitive Load in the Multi-player Prisoner’s Dilemma Game: Are There Brains in Games? (Working paper). Dewitte. DeYoung. 101 (2). T. Shapiro. 290–301. Casari.. P. Am. K. Ho. 52 (3). Peck. Ockenfels.E.T. Organ. The simplest equilibrium of a majority rule division game. and job attachment. 266–275.. 2005. 2012. Schwalbach. 603–632. K. 75 (2). Espinosa. Am. D. J. Rand. 2003... Jones. P. Finite rationality and interpersonal complexity in repeated games. Rand. Behav. J. Rev. Feinberg. S..-H. A. J. Proc. 2006a. I. 3 (3). Inq. Hager. Ochs. Econ. D. Cognitive ability and learning to play equilibrium: a level-k analysis (Working paper).. Stanford. Econ. 39 (1). 61 (3).. J. E. G.. U. Milinski. P. Strategy Choice in the Infinitely Repeated Prisoner’s Dilemma (Working paper). Behav. Stevens. 1990. W. Expecting continued play in prisoner’s dilemma games. A. Rev.. J.L. 2006b.... 51 (3). Stud. Evolutionary stability in the infinitely repeated prisoners’ dilemma played by two-state Moore machines. Sci.. R. Econometrica 73 (3). A. E. 2005. Peck. R. Multiple prisoner’s dilemma games with(out) an outside option: an experimental study. Organ. Econ. Theor. Addison-Wesley. Rev. B. Econometrica 56 (2). 2009. T. 173–207. Duffy. Behavioral biases in endogenous-timing herding games: an experimental study.. Carpenter. Econ. 41 (2). B.J. Rev. 153–160.. D.. Decis. 2002..J. Volstorf. 2011a. Peeters. Econ. Schooler. Roth. Sabourian.. Branas-Garza. 99 (4). Econ. 2011b. Rev. Smith.. D. 2 (3).. W. P.. Johnson. Dreber. The Ghost in the Machine: Inferring Machine-Based Strategies from Observed Behavior (Working paper). 56 (2). R. R. and Computation.. M.. Winkler. Benjamin. Are people Bayesian? Uncovering behavioral strategies.. 2001..M. Altruistic behavior as a costly signal of general intelligence.. Theor.. J. Econ. Dal Bo.G. Gray. Hopcroft. Control 20. Ullman. Behav. Behavioral spillovers and cognitive load in multiple games: an experimental study.. M.. 101 (1). D. Ivanov. J. Equilibrium selection in the repeated prisoner’s dilemma: axiomatic approach and experimental evidence. M. Econ. 66 (2). 153–160. Am. 2012.E.. A. J. Game theory and strategic complexity.. Cooperative behavior and the frequency of social interaction. 739–769.. New York. Behav. 375–384. Conflict Resol.. Am.. Garcia-Muno...J. Kalai. 770–793.R. D.. U. Equilibrium behavior and repeated play of the prisoner’s dilemma.. 278–305. Theor. B. Dynam. Frechette.. J. J. Math. Wedekind..A.. 7745–7750. Kalai. Cason. 2012. Econ. demographic effects. Games Econ. P.C. Rustichini.G.. 17 (2). A. 1985. 2011b. C... E. foresight. 27 (2). 2008. J. Psychol. Acad. C.. 880–903. 979–1005. Gill..A.. Volij. A. Evolutionary stability in repeated games played by finite automata. Liu. Conflict Resol. Behav.J. Hauk. P. 61 (2).. Economic Choice Semiautomata: Structure. Cason.. Frey. P. M. R. Holt. Dreber. Reading. Spagnolo. 1. Games Econ. Cooperation among strangers under the shadow of the future. 1715–1730. Healy. I.366 M. Lichtenstein. 78 (1–2).V. Rustichini.