Influence of human capital on the trial and error learning process in a common... 47 Economía Agraria y Recursos Naturales. ISSN: 1578-0732. e-ISSN: 2174-7350. Vol. 21,2. (2021). pp. 47-77 Influence of human capital on the trial and error learning process in a common pool resource (CPR) game Influence of human capital on the trial and error learning process in a common... Daniel Guerreroa, Jordi Rosellb, José Santiago Arroyoc Guerrero, D.; Rosell, J.; Arroyo, J. S. ABSTRACT: This paper presents a study regarding the behavior of Pacific-Colombian fishers in a Common Pool Resource game. Results show that decision-making depends on human capital accumulation and the learning process. Specifically, through trial and error, those players with more human capital adjust their decisions on the basis of a cooperative-collusive solution by following the feedback of their own most successful strategies in past rounds. Notably, fishers with the higher levels of formal schooling tend to harvest less because they have a better understanding of dilemma-type games and the higher benefits involved when they cooperate. Influencia del capital humano en el proceso de aprendizaje de prueba y error en un juego de recursos de uso común (CPR) RESUMEN: Este estudio presenta resultados sobre el comportamiento de los pescadores del Pacífico colombiano en un juego de recursos de uso común. Los resultados muestran que la toma de decisiones depende del capital humano y del proceso de aprendizaje. A través de prueba y error, aquellos jugadores con más capital humano ajustan sus decisiones hacia una solución cooperativa-colusoria siguiendo la retroalimentación de sus propias estrategias más exitosas en rondas pasadas. Particularmente, los pescadores con niveles mayores de educación formal tienden a extraer menos unidades debido a mejor comprensión de los juegos tipo dilema social y los mayores beneficios obtenidos al cooperar. KEYWORDS / PALABRAS CLAVE: Colombian Pacific coast, Common-Pool Resources, Non- cooperative games, Trial and Error Learning Process / Costa Pacífico colombiana, Recursos de uso común, Juegos no cooperativos, Aprendizaje de prueba y error. JEL classification / Clasificación JEL: I12, C72, D83. DOI: https://doi.org/10.7201/earn.2021.02.03 a Faculty of Economics, Management and Accounting. Unidad Central del Valle, Colombia. E-mail: d.guererob@uceva.edu.co. b Department of Applied Economics. Universitat Autònoma de Barcelona, Spain. E-mail: Jordi.rosell@uab.cat. c Faculty of Business Management. Universidad del Valle, Colombia. E-mail: Santiago.arroyo@correounivalle.edu.co. Acknowledgement: This research was conducted within the project BIODIVERSITY CONSERVATION OF UTRIA NATIONAL PARK THROUGH THE SUSTAINABLE MANAGEMENT OF BUFFER ZONES, ex- ecuted by MarViva in cooperation with Word Wildlife Fund (WWF Colombia), Natural Heritage (Fund for Biodiversity and Protected Areas) and the United States Agency for International Development (USAID). The project aims at defining activities for artisanal fishing management agreements in the Gulf of Tribuga, Colombia. Results of this research contribute to the design of programs for sustainable fishing in the area and for strengthen- ing governance in local communities. Cite as: Guerrero, D., Rosell, J. & Arroyo, J.S. (2021). “Influence of human capital on the trial and error learning process in a common pool resource (CPR) game”. Economía Agraria y Recursos Naturales, 21(2), 47-77. https:// doi.org/10.7201/earn.2021.02.03 Corresponding Author: Daniel Guerrero. Received in July 2021. Accepted in October 2021. 48 Guerrero, D.; Rosell, J.; Arroyo, J. S. 1. Introduction The ZEPA (acronym in Spanish for Exclusive Areas for Artisanal Fishing) and the DRMI (acronym in Spanish for and Regional Districts for Integrated Manage- ment) are successful cases of territorial organization on the Northern Pacific coast of Colombia and examples to guide the management of fisheries resources towards sus- tainability (Vieira et al., 2016). The ZEPA declared 2.5 nautical miles permanently on July 2013 (MinAgricultura, 2013) and the DRMI established 55,974 marine hectares in October 2017 (MinAgricultura, 2017). These two areas that connect to each other, form a conjoint territory for exclusive use of artisanal fishers. These designations have provided benefits to the fisher communities by mitigat- ing the conflicts between artisanal and industrial fisheries, engaging at least 700 small scale fishers and reducing the pressure from industrial boats (Ramírez-Luna, 2013). Their derived benefits range from the increase in capture per unit of effort (López- Angarita et al., 2018); extra premiums paid by consumers for responsibly caught fish and directly bought from the territory (Sáenz-Pacheco, 2014). Consequently, higher revenues are derived from a moralistic supply chain based on selective and diversi- fied fishing and non-destructive gears (Cobos et al., 2016; Satizábal, 2018; Satizábal & Batterbury, 2018); food security (Ramírez-Luna, 2013); and biological services due to the nursing of several species (Navia et al., 2010). Notwithstanding, these designations have brought struggles among fishers, jeopardizing the effectiveness of their implementation (Hernández & Díaz, 2012; Ramirez-Luna, 2013). These struggles arise from the distortion of the social interac- tions and place-based social control mechanisms (Satizábal, 2018), and the unaware- ness of fishers in regard to the depletion and obligations of the resources and their environmental balance (Diaz & Caro, 2016). In order to propose solutions to these struggles under the experimental economics approach, several experiments have been run in Colombia following the framework proposed by Ostrom et al. (1994). These studies have provided findings regarding how the perceived inequality in the group constrains the effectiveness of communica- tion (Cardenas, 2000); how the modest enforcement of the cooperative equilibrium outperform face-to-face communications (Cardenas, 2003); and how the combina- tion of internal communication and external non-coercive intervention leads to better results in terms of community organization (Moreno-Sánchez & Maldonado, 2010). However, the direct effect of human capital and education in solving Common Pool Resources (CPR) dilemmas has not been conclusive and it is still not clear whether education promotes competitive or cooperative behavior. Particularly, some studies indicate that the education, both formal and informal, turns out to be an ef- fective management tool in marine protected areas. It increases community aware- ness, attitudes, behaviors, and perceptions (Adler, 1996); and despite total education costs that continually increases, it remains, in terms of money and effort, lower than the costs of enforcement (Alcock, 1991). Essentially, high school graduates should understand the significance of the ocean in the earth system and how the ocean and humans are interconnected (Plankis & Marrero, 2010) and be able to make informed Influence of human capital on the trial and error learning process in a common... 49 and responsible decisions concerning marine resources (Shoedinger et al., 2006); nevertheless, field experiments show diverging results. On the one hand, some field-lab studies show that more schooled players are open to accepting external interventions and rules aimed at promoting sustainable use of resources (Moreno-Sánchez & Maldonado, 2010). Additionally, players with higher levels of human capital and cognitive skills are willing to contribute more to collec- tive actions (Arroyo-Mina et al., 2016; Cardenas et al., 2009; Brañas et al., 2009), and are less inclined to free-ride (Brañas et al., 2009). On the other hand, education can also promote competitive behavior when the more schooled players take advantage of less schooled ones. In this case, when the latter ones are unsure about their own decisions, they use the others as a reference leading to a competitive equilibrium (Velez et al., 2009). Results from filed experiments often indicate that, under some circumstances, external motivations drive the strategic behavior of those more educated players towards cooperative outcomes; however, we notice that none of them explain how this behavior is internally driven by the available information those players have, and how they use it as feedback to look for better outcomes. Due to this gap in the behavioral and experimental economics literature, we attempt to explain how higher levels of schooling drive the individual behavior to a cooperative-collusive solution through a Trial and Error Learning Process (Huck et al., 2000; 2004a). Our purpose is to contribute to this discussion by explaining how education, human capital and cog- nitive skill formation are influential in driving strategic behavior. By applying CPR games over 10 rounds to a set of villagers in the Pacific coast of Colombia, we find that players with higher schooling levels lead their actions towards higher payoffs. These games were carried out with Pacific-Colombian fishers inhabiting a remote territory where there is low schooling and a illiteracy rate of 15.2 % of the population (DANE, 2005). The remaining part of the paper is structured as follows: Section 2 presents a brief discussion on the influence of human capital on strategic behavior. Section 3 presents the theoretical model for analyzing the behavior of fishers. Section 4 presents the game design, the procedures during the experiments and reports on the results ob- tained. Section 5 provides final comments and a conclusion. 2. Human capital and strategic behavior 1 Despite the lack of evidence on how human capital drives decisions in CPR games, the role of cognitive skills in decision making has been studied intensively under the context of repeated Prisoner’s Dilemma (PD) games. In this context players with higher human capital are better at understanding repeated PD games (Brosig, 1 As a global consensus, years of formal education and schooling are considered a measure of human capital, although this may not necessarily translate into marketable skills (Becker, 1964); besides, formal education enhances the human capital stock by improving the cognitive skills for problem solving (Bowen, 1977; Pallas, 2000). 50 Guerrero, D.; Rosell, J.; Arroyo, J. S. 2002; Burks et al., 2003; Schramm, 1998; Tan & Zizzo, 2008) and better able to adapt their behavior, even when they do not know each other and have played only once (Boone et al., 2002; De Jong, 2015). Additionally, the evidence from experiments using university students as subjects present interesting findings. These studies argue that players with higher cognitive skills are more patient, therefore have higher savings rates (Jones, 2008) in both the short and long-run. In addition, these skills are associated with higher social aware- ness, better prediction of consequences of their actions, and performance of calcula- tions for reducing the number of errors2. These skills seem to elicit a greater tendency to be cooperative in a strategic setting, and players with these skills appear to cooper- ate more often when playing repeated PD games (Burks et al., 2009). Aside from this, human capital drives strategies to cooperative-collusive solution behavior in contradiction to the rationality proposed in game theory (De Jong, 2015). Interestingly, these findings make clear that the more educated players are not neces- sarily more altruistic; instead, they behave strategically in order to control their con- text and to obtain higher outcomes. In other words, in a PD game, players with more schooling play more cooperatively on average because it is in their self-interest, and in the long run, cooperation provides higher payoffs (Boone et al., 2002). Now, considering that previous experiences in Colombia are not conclusive on how education drives strategic behavior in a CPR game, we analyze how villagers with different levels of schooling make decisions. In order to do this, we designed a CPR game in which the payoff structure simulated a dilemma between the coop- erative-collusive and competitive solutions. In this game, we analyzed how players adjusted their decisions according to their levels of schooling and learning processes. 3. Theoretical Framework: Learning Process A learning process is defined as a searching among strategies and payoffs (Huck et al., 2000; 2004a). This search is initiated by an aspiration-induced mechanism which motivates a player to search for a new decision if current payoff is below an initial aspiration level; then changing their strategy to another with relatively higher payoffs (Karandikar et al., 1998; Vainstein et al., 2007). After several repetitions, players learn how to direct their actions towards an overall higher payoff (Huck et al., 1999; 2004b; Selten & Buchta, 1998). 3.1. Learning process based on the experience of others Although learning process has not received much attention under a CPR game context, Huck et al. (1999) define two types of learning process under a Cournot’s Oligopoly game context, which provides repeated dilemmas in decision-making similar to CPR’s framework. These learning processes are based on the experience of 2 Kosmidis (2018) indicates that formal schooling trains cognitive skills and strategies; thereby, there are some procedures that unschooled individuals are not able to manage. Influence of human capital on the trial and error learning process in a common... 51 others. The first one is defined as Best Reply, where players choose in every period a best reply from their rivals’ total output in the last period. To play best replies requires knowing the rivals’ profits; and in an infinite-period game, Best Reply converges to a static Nash equilibrium. The second one is defined as Imitate the Best, where every player chooses the strategy that received the highest payoff in the last period; mostly, when players behave according to this dynamic they attain the Walrasian equilibrium in the long run. Imitating the best requires knowing and listing each period’s quan- tities and the profits of each player. Accordingly, players learn from rivals if their beliefs and rewards are available (Offerman & Sonnemans, 1998). As long as players have more information about their rivals’ outcomes, they become more competitive (Huck et al., 1999). For analyzing these processes based on others’ experiences, we assume that the benefit of a fisher is defined by a constant and symmetric benefit from external ac- tivities to fishing w. A fishing harvest function g(xi)=axi-bxi2. The strict concavity of g(xi) indicates diminishing marginal private returns to harvesting. This harvesting is sold at price p=1. For simplicity we assume zero costs at first stage. Finally, we have the externality cost of the aggregate harvest φ∑xj caused by n fishers. The total pri- vate benefit is expressed in Equation 13: = + − − ∑ [1] Now, when optimizing the former expression, we obtain: / = −2 − =0 and the resulting competitive harvest: = ( − )/2 [2] This expression indicates that the individual optimal harvest does not depend on the rivals’ harvest, and the individual decision is driven only by their own benefit maximization. Now, if we assume all fishers harvest the same amount by imitating each other’s harvest, we replace the expression of the aggregate harvest φ∑xj for φnx. Then, obtaining an identical benefit function for every fisher: = + − − when optimizing it we obtain: / = −2 − =0 and the resulting solution for imitation: =( − )/2 [3] 3 The theoretical model implemented is adapted from Cardenas (2000). 52 Guerrero, D.; Rosell, J.; Arroyo, J. S. Note that the former expression requires that all fishers have the same benefit structure, i.e. they are symmetrical. Thus, to behave under a learning process based on others’ experiences it requires symmetrical benefit functions. Additionally, we can observe that the more fishers, the less they can harvest. Now, by introducing a cost ci > cj we obtain the following expression: = + − − − ∑ for xi : =( − − )/2 and for xj : =( − − )/2b Under these circumstances we obtain xi < xj; thus, imitation is not suitable nor is further cooperation-collusion a feasible solution. According to the literature, human behavior has evolved to consider a neighbors’ experience as good as a player’s own and players tend to imitate the behavior of those they can observe (Eshel et al., 1998); although conditioned to homogeneity in the structure of the neighborhood and the local interaction of players in it. Only then, it is more likely for a player to learn from each other’s experience (Kirchkamp & Nagel, 2007). This argument allows us to understand that, although players may interact in a framework of incomplete information about the types of rivals, they can be consistent with rational behavior (Kreps et al., 1982); Now, we know the imitation process is not always suitable; because of this, Huck et al. (1999) indicates that there are other possible learning processes when there is a lack of coordination during a game. This process, defined as Trial and Error, does not require players to know the information about their rivals’ actions and payoffs and it is based on their own experiences during the game. In the next section we explain what is a Trial and Error learning process and the implication of following it during a game. 3.2. Trial and error and learning based on own experience The basic argument of this learning process is that players would not repeat a mistake. If players experience a decrease in payoffs in the last round due to say an increase in quantity, then they would not increase the quantity again, and vice versa. Thereof, players’ strategies can oscillate between two extreme values led by their own feedback; then, it is feasible that players’ decisions are led to a cooperative- collusive outcome (Huck et al., 2000; 2004a). Moreover, players can exhibit this behavior even in an environment dominated by defectors (Eshel et al., 1998; Helbing & Yu, 2009). Influence of human capital on the trial and error learning process in a common... 53 This learning is mostly an intuitive process performed by the player. Here, in the absence of information about their rivals’ costs or benefits, players can conclude where to find the best decisions to make. The expression (4) formalizes the idea that if a direction was successful in the past round, it is repeated in the current round; when a direction fails, a reverse action is taken in the current round; if the change in payoffs or the change in quantities were zero in the previous round, the current quantity remains the same (Huck et al., 2000). Mainly, players who behave according to trial and error learning are aware of feed- back from past rounds. The discrete-time version supposes that players behave according to the following rule: = + ( − )∗ ( − ) [4] where δ > 0 is some arbitrarily small size step and sign (∆) is defined as: −1 Δ < 0 (Δ) = 0 Δ=0 [5] 1 Δ>0 It is plausible that during the learning process players make mistakes at executing their strategies and systematically try out different actions. Then, they do not always coordinate decisions to attain cooperation-collusion. Furthermore, yielding cooperative-collusive solution requires that all subjects play according to the rule of Trial and Error. If the case any subject fails to comply, there is not convergence to the theoretical prediction. Hence, aggregate quantities can fall far from the cooperative-collusive solution (Huck et al., 2000). We propose that during these games, players behave according to the Trial and Error learning process thus, due to this dynamic, it does not require any cognitive effort from players, nor information about rivals’ actions, nor the payoff function of the game. Therefore, it is applicable to repeat dilemma-type games like CPR because strategy sets can be ordered, players cannot observe the individual actions and out- comes of their rivals, and they have no coordination or punitive menace (Huck et al., 2000; 2004a). 4. Game design and performance The design of this game follows the one proposed by Cardenas (2000). We modi- fied it slightly to allow the possibility to apply it to groups of 4 and 5 players. The objective of this modification is to permit all arriving participants to play the game once it is being performed in the field lab. Thus, we avoid the exclusion of any fisher in the communities and minimize any ethical problems in future activities. 54 Guerrero, D.; Rosell, J.; Arroyo, J. S. In this game, a group of players perform a 10-round game simulating a social dilemma. Here, each player individually and secretly decides how much to harvest from a fishing bank, simultaneously interacting with the aggregate harvest of co- players or rivals. Moreover, the player only knows the rivals’ harvest at the end of the round, when the assistant researcher discloses the aggregation. In the end, each player’s payoff depends on a combination between their individual and rivals’ aggre- gate decisions. This procedure is repeated during the 10 rounds. To summarize: The aggregate harvest of the group is public information while individual harvest, payoff and rivals’ harvest is private information. To calculate the payoff structure, we follow expression (1) and replace w = 1530, a = 130, b = 5 and φ = 50. We use these parameters to ensure a strictly positive har- vest and to calibrate the payoffs to local daily earnings (see Appendix A). Then, we substitute the parameters in expression (3), individual harvest ranges 1 ≤ xi ≤ 8. Ag- gregate harvest for 4-player groups ranges 4 ≤ ∑xj ≤ 32; for 5-player groups ranges 5 ≤ ∑xj ≤ 40. The social equilibrium, i.e. cooperative-collusive solution for 4-players group is obtained when (xi, ∑xj–xi) = (1,3); for 5-players groups when (xi, ∑xj–xi) = (1,4). The Nash equilibrium, i.e. the competitive solution for 4-player group is obtained when (xi, ∑xj–xi) = (8, 24); for 5-player groups when (xi, ∑xj –xi) = (8, 32). The payoffs range between a minimum of 50 points (corresponding to a harvest xi = 1, when rivals’ harvest is ∑xj–xi = 32; i.e. a private social harvest against a competi- tive harvest from rivals) and a maximum of 1730 points (corresponding to a harvest of xi=8, when the rivals’ harvest is ∑xj–xi = 3 or ∑xj–xi = 4; i.e. a free-riders’ outcome). In order to avoid ethical problems with payoffs of different-participant groups, we ensure that both types of group had the same range of payoff (see appendix A). With this function, we generate a social dilemma regarding the payoffs structure; thus cre- ating a conflict between individual and rival’s payoffs. 4.1. Sampling These games were performed in 8 communities located on the Northern Pacific coast of Colombia: Nuqui, Pangui, Coqui, Jovi, Termales, Arusi, Tribuga and Juru- bira. For obtaining the sample, we applied two methods. The first one was to publish on local billboards located in community councils. The second one was word-of- mouth, to reach isolated communities. The latter is very effective for disclosing in- formation among the Colombian villages because this method is a traditional form of communication medium. According to DANE (2005), 29.8 % of population live in Nuqui, the head of the municipality. The remaining 70.2 % are distributed in the rural villages. In the sample, we obtained 154 participants. The largest attendance was in Nuqui, with 40 participants. This implies that we obtained 26 % participants from Nuqui, and 74 % from rural villages. We conducted only one daily session per community and due to a weekend in the middle, the entire fieldwork lasted 10 days. The average number of Influence of human capital on the trial and error learning process in a common... 55 participants per session was 19.87 and the session lasted around 2 hours, accounting also for the time devoted to the final payments and a semi-structured survey con- ducted at the end of each game. In this survey we obtained information from 79 players who attended up to any level of high schooling. We defined these as HS players4. We also obtained informa- tion from 75 who had only attended up to any level of elementary school (Table 1). We defined these as ES players. Additionally, we collected information on age, gen- der, experience in fishing and civil status (proportion of married fishers as a proxy of head of the household). Participants have received earnings from the fishing activity at least once in their lives. TABLE 1 Socioeconomic information Schooling Variable Average Std. Dev. Min Max Age 38.68 (14.07) 15 74 Gender (male) 89 % 1 (0.32) - - HS Years of experience 21.47 (14.75) 1 65 Civil status (married) 65 % 3 (0.48) - - Age 47.39 (14.47) 15 79 Gender (male) 91 % 2 (0.29) - - ES Years of experience 28.75 (16.27) 1 69 Civil Status (married) 75 % 4 (0.44) - - 12 Proportion of male fisher are generally higher in these communities. 34 Proportion of fishers who are married. Source: Own elaboration. In each session, groups of 4 and 5 players with different levels of schooling were formed in circles turning their back to each other, assisted by a researcher who forbade communication among them. We obtained 34 groups in total, 16 with 4 participants and 18 with 5 participants. The players were distributed randomly in the groups. We obtained 17 groups (50 %) dominated by HS players; i.e. at least 3 of them were HS players. Before starting the game, the context and the rules to follow were presented to players. For ensuring understanding of the game, three previous rounds were per- formed before the real game started; thereof, any feedback and extra explanations for better understanding were provided. Additionally, we informed players how many periods the game lasts for ensuring common knowledge of duration. During these games, anonymity was enforced and every player was assigned a code, with which they were identified for final payment at the end of the game. 4 In Colombia, elementary school comprises 5 years of schooling; high school takes 6 more years of schooling; and a technician degree is an undergraduate program that takes up to three years of schooling; a bachelor degree takes five years to obtain. 56 Guerrero, D.; Rosell, J.; Arroyo, J. S. 4.2. Descriptive analysis For understanding the behavior of players, we analyze independently the group of HS players from the ES5 one (Table 2). The overall average harvest is 4.55 units. Discriminating by level of schooling, the ES average harvest is higher than HS. This difference of 0.27 units is significant at the 5 % level (Mann-Whitney test: z = 2.44; p-value = 0.015). TABLE 2 Harvest behavior during the game Variable 1 Average St. Dev. Min Max xi Overall 4.55 (2.19) 1 8 xi HS 4.42 (2.19) 1 8 xi ES 4.69 (2.18) 1 8 ∑xj 21.06 (5.06) 8 36 ∑xj–xi 16.51 (4.55) 4 31 πi HS 2 1,018.54 (243.09) $100 $1,700 πi ES 3 1,020.08 (239.16) $200 $1,730 1 Average calculated by round. 23 Exchange rate USD$ 1 = COP$3,123. Source: Own elaboration. Now, by analyzing players’ behavior by round, we observe that ES players harvest 4.2 units on average in first round and exhibit a tendency to increase their harvest as long the game is carried out. Their average harvest in the last round is 4.72. On the other hand, HS players harvest 4.6 units on average in the first round and exhibit a slight tendency to decrease their harvest during the experiment. Moreover, they finish with 4.2 units of harvest on average in round 10 (Figure 1). 5 We analyzed the levels of schooling for each group in order to identify if there were a real difference in hu- man capital accumulation. For ES players, the average schooling was 2.82 years; for HS was 9.42 years. Mann- Whitney test: z = –34.31; p-value = 0.00. Influence of human capital on the trial and error learning process in a common... 57 FIGURE 1 The average harvest of ES and HS players Source: Own elaboration. By analyzing the feedback processing, we can address some differences between behaviors of the two types of players. ES players tend to increase their harvest de- spite payoffs feedback and move either the same or contrary directions. Additionally, HS players tend to increase their harvest when feedback is positive, whereas they decrease their harvest when feedback is negative, creating a monotonic path between quantities and payoffs; this means that HS players seem to reverse their decision when being competitive does not generate rewards (Figure 2). 58 Guerrero, D.; Rosell, J.; Arroyo, J. S. FIGURE 2 Feedback processing Source: Own elaboration. 4.3. Statistical analysis Firstly, for understanding the behavior and learning process of players during the game, we analyze how players harvest xi according different demographic variables. Following expression (4) from the theoretical model, we estimate ordinary least squares (OLS), panel regression controlling for random effects (RE)6, and panel- corrected standard errors (PCSE) models. It is necessary to specify that PCSE, estimated according to serial correlation, is explicitly addressed in the theoretical and experimental analysis. This estimation allows for fitting correlation and unequal- variance error. Then, we enhance the efficiency of the parameter in the presence of autocorrelation AR(1) within panels, cross-sectional correlation, and heteroskedastic- ity across panels. By estimating this model with asymptotic standard errors corrected for correlation between panels, allows for relaxing the restriction that requires at least as many period observations as there are panels (Beck & Katz 1995). The individual harvest xit is estimated as the dependent variable; the first lag of individual harvest xit–1 and the function quantities-payoffs fxpi that measure feedback 6 Since the fixed effects model only accounts for within-subject variation and does not contemplate time- invariant variables, we propose random effects and panel-corrected standard error models. This latter model is more suitable for this case study, since the variable dHS, which is the main object of analysis, is time-invariant and we want to know its effect on the individual decision of players. Influence of human capital on the trial and error learning process in a common... 59 are estimated as independent variables; additionally, we use a function of the lag of rivals’ harvest fzj in order to identify whether players follow some sort of imitation from rivals’ former strategies. This function takes value 1 if the harvest increased in the previous round, -1 if it decreased and 0 it remained constant. We propose using this function instead of the aggregate amount because for a player it is easier to iden- tify if the harvest increases, decreases or is stable rather than 28 possible strategies the rivals could choose. Additionally, we introduce a control variable domHS, which takes the value of 1 if the group where a player performs is dominated by HS players; 0 otherwise. These variables are estimated interacting with a dummy variable dHS that takes value 1 if player is HS and 0 otherwise. These interactions are performed for discriminating behavior according to levels of schooling. Finally, we estimate the effect of control variables such as gender s, which is a dummy variable that takes value 1 when the player is a man and 0 when player is a woman7; the age of player a; years of experience in fishing exp; civil status cs as a proxy for head of household, which is a dummy variable taking the value of 1 when a player is married or in con- sensual union and 0 when is single or divorced; and the variable g is given to control the size of the group. This variable takes value of 1 if group is 5-player and 0 when it is 4-player (Table 4)8. TABLE 4 ES vs. HS. Learning processes Variable 1 Schooling OLS OLS (n.c.)2 RE RE (n.c.)3 PCSE PCSE (n.c.)4 xit–1 ES 0.13*** 0.13*** 0.13** 0.13** 0.1*** 0.11*** (0.03) (0.03) (0.05) (0.05) (0.04) (0.04) HS 0.14*** 0.14*** 0.14** 0.14** 0.08** 0.08** (0.03) (0.03) (0.05) (0.05) (0.03) (0.03) fxpi ES 0.14 0.14 0.14 0.14 0.08 0.07 (0.09) (0.09) (0.10) (0.10) (0.08) (0.08) HS 0.17* 0.17* 0.17* 0.17* 0.18** 0.18** (0.09) (0.09) (0.10) (0.10) (0.09) (0.09) fzj ES 0.13 0.13 0.13 0.13 0.13 0.13 (0.09) (0.09) (0.10) (0.10) (0.09) (0.09) HS -0.15* -0.15* -0.15* -0.15* -0.15* -0.15* (0.09) (0.09) (0.08) (0.08) (0.09) (0.09) domHS ES 0.22 0.20 0.22 0.20 0.21 0.19 (0.20) (0.20) (0.23) (0.22) (0.19) (0.19) 7 Interactions between s and dHS are not estimated because we obtained information for only 7 ES and 9 HS women players. Gender discussions of the activity are beyond the scope of this research. 8 We also estimated a Heckman’s Model (see Appendix B) to correct the possible selection bias. Since the In- verse Mills Ratio is not significant, we assume that models presented in Table 4 represent the behavior of players during the experiment. 60 Guerrero, D.; Rosell, J.; Arroyo, J. S. Variable 1 Schooling OLS OLS (n.c.)2 RE RE (n.c.)3 PCSE PCSE (n.c.)4 HS -0.32* -0.35* -0.32 -0.35 -0.31* -0.34* (0.19) (0.19) (0.27) (0.26) (0.19) (0.18) s 0.47** 0.45** 0.47 0.45 0.46** 0.44** (0.22) (0.21) (0.32) (0.29) (0.22) (0.20) a 0.00 - 0.00 - 0.00 - (0.01) - (0.01) - (0.01) - ex 0.00 - 0.00 - 0.00 - (0.019) - (0.01) - (0.01) - cs -0.12 - -0.12 - -0.12 - (0.14) - (0.19) - (0.13) - g -0.33** -0.32** -0.33** -0.32** -0.32** -0.31** (0.14) (0.14) (0.18) (0.18) (0.14) (0.14) Constant 3.77** 3.83*** 3.77*** 3.83*** 3.67*** 3.73*** (0.34) (0.25) (0.49) (0.39) (0.34) (0.26) 1 Dependent variable xit. 234 These models are estimated without non-significant controls to verified robustness. Significance: *10 %; **5 %; *** 1 %. Source: Own elaboration. Results show that harvest decisions in round t depend on decisions made in the past round t–1 for both levels of schooling in every model estimated. By including AR(1) correlation in PCSE, the current harvest decision exhibits a slight decrease in dependence on lagged decisions. Estimations on feedback present interesting results. On the one hand for ES play- ers, fxpi exhibit no statistical relation to harvest decision xit in any model estimated, and coefficients vary slightly. On the other hand for HS players, fxpi exhibit a sta- tistically significant relation indicating that their successful decisions made in the past round are repeated in the current round. These estimations also indicate that HS players identify and rely on feedback from past rounds (Archibald & Elliot, 1989). Accordingly, HS players move forwards on a monotonic path, evidencing that play- ers conclude in which direction better decisions can be made. Then, in the absence of information about rivals’ feedback, the best decision to make is in the direction that was successful in the last period. Thereby, all moving players who adopt this behav- ior improve their profit building in a monotone-payoff path, even though others may experience losses. This evidence provides arguments to accept the hypothesis that HS players behave strategically following a trial and error learning process. Analyzing the function of the lag of rivals’ harvest fzj we can conclude that ES players seem not to follow their rivals’ experiences; nevertheless, HS players tend to decrease their harvest when their rivals increased in past rounds and vice versa. How- Influence of human capital on the trial and error learning process in a common... 61 ever, oligopolistic and CPR games are not ideal to assess imitation learning because its large strategy space. The main reason is because players can choose many differ- ent strategies among a large number of possible quantities. Often, players will choose new quantities that have not been tried before (Kirchkamp & Nagel, 2007). Regarding variable domHS, we find a slight negative and significant relation of HS harvest when interacting in groups dominated by HS. Thereby, the context-based imitation learning process that prompts other agents to become cooperative-collusive can be observed over HS players. This conclusion aligns with similar arguments pro- posed by Eshel et al. (1998) and Kirchkamp & Nagel (2007). Then, a neighborhood dominated by HS players tends to reach cooperation-collusion during these games. Nevertheless, ES players show no context-based imitation. Accordingly, variable s indicates that women are less competitive than men (Cro- son & Gneezy, 2009). However, remaining control variables like age, experience and civil status seem not to have statistical significance in relation to harvest during the performances of the game. Estimation regarding the size of the group g aligns with the theoretical prediction of expression (3). This expression indicates that when groups are smaller, they can optimally harvest more. This estimation is significant in every model and indicates that groups of 4 participants harvest more compared to groups of 5 participants. This is a rational behavior motivated by the maximization of the benefits function. The Constant in these models provides a broad estimation of the players’ harvest starting point. This estimation is not too intuitive and requires an extra explanation. Assuming that in some rounds player’s feedback is equal to zero, meaning that player did not find a way to improve their profit, the player is motivated to alternate between moving up or down (Huck et al., 2004a), starting again looking for the strategies which provide profit improvement. This starting point ranges from 3.73 to 3.83 de- pending on the estimation. 4.4. Decision-making and payoffs In order to analyze the incentives of players during the game, we run a regression over of payoffs πit obtained during the game9. The independent variables are the indi- vidual harvest xit; rivals’ aggregate harvest zjt, i.e. the total amount of rivals’ harvest instead of the function. Here we use the total amount because the payoffs are defined by the ordered pair made between the individual and aggregate harvest, and not by the simple function. Finally, we estimate the influence of the environment domHS. These variables are estimated interacting with level of schooling dHS. Finally, we control for gender s and group size g (Table 5). 9 Since the fixed effects model only accounts for within-subject variation and does not contemplate time- invariant variables, we propose random effects and panel-corrected standard error models. This latter model is more suitable for this case study, since the variable dHS, which is the main object of analysis, is time-invariant and we want to know its effect on the individual decision of players. 62 Guerrero, D.; Rosell, J.; Arroyo, J. S. TABLE 5 Payoffs Variable1 Schooling OLS OLS (n.c.)2 RE RE (n.c.)3 PCSE PCSE (n.c.)4 xit ES 36.51*** 36.42*** 36.95*** 36.94*** 36.51*** 36.42*** (1.60) (1.60) (1.44) (1.43) (1.66) (1.65) HS 35.87*** 35.74*** 35.22*** 35.21*** 35.87*** 35.74*** (1.58) (1.57) (1.54) (1.53) (1.52) (1.51) zjt ES –42.36*** –42.36*** –46.20*** –46.19*** –42.36*** –42.36*** (0.78) (0.78) (0.91) (0.91) (0.78) (0.78) HS –43.72*** –43.77*** –47.13*** –47.14*** –43.72*** –43.77*** (0.76) (0.76) (0.64) (0.64) (0.73) (0.72) domHS e ES~domHS 1.124 –0.256 –6.370 –7.498 1.124 –0.256 (7.94) (7.87) (14.57) (14.11) (7.12) (7.03) HS~domES 37.19* 38.84* 32.47* 33.08* 37.19* 38.84* (22.38) (22.35) (19.71) (19.91) (20.81) (20.75) HS~domHS 10.04 10.42 2.256 1.892 10.04 10.42 (21.49) (21.49) (18.02) (18.16) (20.44) (20.43) s –10.59 - -8.639 - -10.59 - (8.172) - (17.64) - (8.860) - g –71.76*** –70.53*** –64.34*** –63.34*** –71.76*** –70.53*** (5.67) (5.59) (11.75) (11.39) (5.19) (5.14) Constant 1,608*** 1,598*** 1,667*** 1,659*** 1,608*** 1,598*** (17.35) (15.75) (19.10) (12.45) (16.51) (14.57) 1 Dependent variable: Payoff by round πit. 234 These models are estimated without non-significant controls to verified robustness. e Baseline: ES dominated by ES players, ES~domES. Significance: *10 %; **5 %; *** 1 %. Source: Own elaboration. Estimations show that harvests of both levels of schooling are positively related to payoffs; i.e. payoffs increase when players increase harvests despite their level of schooling. These coefficients are statistically equal (2(1) = 0.36; p-value = 0.547); thereof marginally, ES and HS players receive the same payoffs. These estimations also indicate that when rivals’ aggregate harvests increase, individual payoffs de- crease for both ES and HS players. This effect is statistically the same for both levels of schooling (2(1) = 0.01; p-value = 0.918). From estimations on domHS, we observe that the environment yields differences in payoffs during the game. Having ES interacting in an environment dominated by ES as baseline, we observe that when HS players interact in an environment dominated by ES players (HS~domES), they receive higher payoffs than ES players Influence of human capital on the trial and error learning process in a common... 63 despite the environment in which they interact (HS~domES = ES~domHS: 2(1) = 14.58; p-value = 0.00). When HS players interact in an environment dominated by HS players, their payoffs tend to be also higher; however, none of these latter coef- ficients are statistically significant in any model estimated. These results indicate that by harvesting closer to cooperative-collusive solution, HS players can obtain higher payoffs and revenues, even if they face defection from rivals. Smaller groups, that harvest higher amounts (see Table 4) can obtain higher pay- off. This is because the payoff structure allows smaller groups to reach the optimum at a higher level of harvest with higher revenues (see expressions (1) and (3)). From this situation, we can conclude that despite the benefits range is the same for either 4-player or 5-player groups, smaller groups obtain higher revenues when playing this dilemma-type game. However, this particular case should be explored more deeply in further studies. Finally, Constants in these models provide insights into average payoffs per round of a female fisher with elemental schooling, who interacts in an environment domi- nated by ES players and plays against 3 rivals. These values range from $1,598 ≤ π ≤ $1,667 depending on the estimation. 4.5. Discussions 4.5.1. ES players’ behavior It is necessary to emphasize that the human brain has not evolved specifically for developing core skills like reading, writing or making calculations; these are very complex skills created by contemporary human culture. Therefore, our brains have to make an extra effort using abilities that have evolved for different purposes. Thus, to use cognitive skills has become a great accomplishment, which requires polish- ing many functions including basic visual searching, phonological awareness and working memory (Huettig et al., 2018). These advantages lead literates, frequently correlated with schooled individuals, to outperform illiterates, even in simple visual searching tasks (Malik-Moraleda et al., 2018). For these reasons, ES players are not presumed to be bad learners or to commit errors; besides, statistical evidence raises the hypothesis that the chosen action is the best possible considering their ability to process feedback. Thus, first changing ex- pectations would make it difficult to have accurate perceptions of their alternatives. Hence, we presume that ES players act looking for some payoff aspiration level, behaving as competitors instead of as colluders. This behavior incentivizes players to harvest, subject to small random perturbations; these influence players to switch strategies despite the fact that payoffs obtained were lower than before (Karandikar et al., 1998; Young, 2009). 64 Guerrero, D.; Rosell, J.; Arroyo, J. S. 4.5.2. HS players’ behavior According to Huck et al. (2000, 2004a), it is feasible that individual harvesting tends towards cooperation-collusion despite the aggregate harvest in the group that was not even close to the collusive outcome, i.e. interacting in a group dominated by defectors (Helbing & Yu, 2009). Furthermore, a unique competitive solution will not be stable under the trial and error approach because there are movements away from the equilibrium that are profitable for all players, and somehow in repeated dilemma- type games, players reduce their probability of defection. This will increase players’ expected payoff. Unfortunately evidence fails to determine when cooperative-collusive behavior appears. Nevertheless, findings indicate that infinite rounds are not required to make cooperation possible, and often, a few rounds seem sufficient (Andreoni & Miller, 1993; Kreps et al., 1982). However, it is important to note that these findings state that cooperative-collusive behavior occurs when players interact in fixed groups during the entire experiment (Huck et al., 2001; Huck et al., 2004b). Also, they are required to understand what is a joint-profit maximizing solution and how to interpret their rivals’ decisions (Rassenti et al., 2000). Besides, since players were not con- strained to adjust their harvest by one unit, it is possible that cooperative-collusive individual harvest appears suddenly during the early rounds. 4.5.3. Human Capital and Cognitive Skills Even when human capital and cognitive skills are important characteristics for understanding the behavior of agents and their ability to learn when facing different economic situations, they are generally treated as endowments and approximations of the individual’s welfare. Notwithstanding, learning is optimal only if feedback is optimal, though this condition is strict and infrequently satisfied (Archibald & El- liot, 1989). This situation leads behavior to a non-unique equilibrium set sensitive to variations in initial expectations. Feedback is indisputably incomplete because players are unable to discover the consequences of non-chosen strategies. Hereby, it is important to consider that play- ers with higher cognitive skills and human capital make these decisions because of a better understanding of the information provided, and better processing of the feed- back from their own game interactions. This behavior proceeds from the advantage that schooling brings to more educated players at information processing, allowing maximization of their well-being through problem solving and better adaptation to new situations with different expectations (Bowen, 1977; Pallas, 2000; Pallas & Jen- nings, 2009). However, confidence in judgment due to cognitive skills is not a regular econo- mistic attribute in homo economicus (Archibald & Elliot, 1989). Therefore, it raises the need for correlating trial and error of well-informed rational agents with hu- man capital, cognitive skills, and different preferences. This also raises the need for creating a new theoretical approach for explaining economic behavior, containing Influence of human capital on the trial and error learning process in a common... 65 explanations on the implications of the aforementioned characteristics in successful decision-making, risk aversion, and inter-temporal allocation of resources; as well as entrepreneurship, longevity, health and better judgment in their actions and conse- quences. 4.5.4. What about the territorial context? The need for complying with FAO’s Code of Conduct for Responsible Fisher- ies (FAO, 1995) led to the creation of the Law 13 in Colombia, in 1990. Based on this law, ZEPA and DRMI were established to ensure the management of fishery resources towards sustainability (Vieira et al., 2016). These designations, formally defined as Territorial Use Rights for Fisheries (TURF’s), are site-specific management arrangements attached to territorial culture. TURF-based management systems provide relative controls, preventing rent dissipa- tion from economic waste due to excess of effort caused by applying higher levels of capital and labor (Christy, 1982). Accordingly, some studies have found that TURF- based management coordinates fishing effort for avoiding inefficient allocation of fishing effort (Cancino et al., 2007; Gaspart & Seki, 2003); although the sustainabil- ity of fisheries is subject to on-going debate (Nguyen ThiQuynh et al., 2017). Notwithstanding, it is necessary that the government complies not only by declar- ing territories destined for exclusive sustainable exploitation in this region, but by ensuring a broader form of welfare for local communities. Accounting for a 15.2 % rate of illiteracy and the 41.4 % who only attended basic school, it is necessary that a more active presence of the government be felt; especially in those communities that require more investment in education and human development. In this way, it will be possible to encourage fishers to be more aware of the resources they exploit (Adler, 1996), and help them to understand how the ocean and humans are interconnected (Plankis & Marrero, 2010). Only by so doing, would coastal communities be able to make informed and responsible decisions concerning the ocean resources. 5. Conclusions and final remarks Results show that players with higher human capital exhibit a trial and error learning process moving towards higher payoffs under the context of a CPR game. According to Huck et al. (2000; 2004a) this behavior leads to cooperative-collusive solutions. These findings confirm the existence of individual learning based on feedback from past rounds (Huck et al., 2000; 2004a; Selten & Buchta, 1998). Par- ticularly, we can observe that players with higher levels of schooling, therefor with higher human capital, exhibit this learning process. Previous findings indicate that subjects with higher human capital are willing to contribute more to collective actions (Arroyo-Mina et al., 2016; Brañas et al., 2009; Cardenas et al., 2009), and are less inclined to free-ride (Brañas et al., 2009). Fur- thermore, players with higher human capital are better at understanding and adapting 66 Guerrero, D.; Rosell, J.; Arroyo, J. S. their behavior in a repeated PD game (Boone et al., 2002; Brosig, 2002; Burks et al., 2003; De Jong, 2015; Schramm, 1998; Tan & Zizzo, 2008), and on average they play more cooperatively considering that in the long run cooperation provides higher pay- offs (Boone et al., 2002). During this experiment, both ES and HS current harvests depend on decisions made in past rounds. However, those HS players with the higher human capital con- sider that the best decision to make is in the direction that was successful last time, relying on feedback and building a monotone-payoff path to cooperative-collusive solutions; this happens in the absence of coordination or information about their ri- vals’ decisions. HS players tend to harvest towards cooperative-collusive solutions and, when sur- rounded by homogenous players, they tend to yield cooperation. This behavior can be sustained even though they interact in a group dominated by defectors (Helbing & Yu, 2009). However, evidence fails to determine when cooperative-collusive behav- ior appears. Nevertheless, during the game players are allowed to adjust their harvest at their convenience, thus cooperative-collusive solution decisions can be made sud- denly during the first rounds. Estimations indicate that marginal payoffs are the same for both levels of school- ing. However, the environment in which they interact yields differences in their payoffs. A HS player obtains higher payoffs when interacting in an environment dominated by ES players. Moreover, a HS player tends to obtain higher payoffs when interacting in an environment dominated by HS players; however, these latter estimations are not significant. Therefore, overall higher payoffs motivate HS play- ers harvesting closer to cooperative-collusive solution where higher payoffs can be obtained, even when defection from rivals produces losses. The territorial context can help to explain the behavior of fishers regarding the coordination when approaching cooperative-collusive solutions. The institutional efforts in the region have promoted sustainable fishing for over 20 years and they could bring this experience to the field game. Particularly, players with higher human capital are the ones that exhibited evidence for using this as an advantage during the experiment. Considering this, it is necessary that the government takes an active role by investing more in education in the region. Finally, we are aware that the standard for any empirical study in the social sci- ences is random sampling and experimentation (Angrist & Pischke, 2008); a non- appropriate randomization could lead to truncated samples in which the values of the independent variable are unknown because the dependent variable is unobserved for part of the relevant population. However, after testing the possibility of selection bias, our results indicate that the proposed models are unbiased and efficient estima- tions on players’ behaviour. Influence of human capital on the trial and error learning process in a common... 67 References Adler, J. (1996). “Costs and Effectiveness of Education and Enforcement, Cairns Section of the Great Barrier Reef Marine Park”. Environmental Management, 20, 541-551. https://dx.doi.org/10.1007/BF01474654. Alcock, D. (1991). “Education and extension: Management’s best strategy”. Austra- lian Parks and Recreation, 27(1), 15-17. Andreoni, J. & Miller J.H. (1993). “Rational cooperation in the finitely repeated Prisoner’s Dilemma: Experimental Evidence”. The Economic Journal, 103(418), 570-585. https://dx.doi.org/10.2307/2234532. Angrist, J.D. & Pischke, J.S. (2008). Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton: Princeton University Press. Arroyo-Mina, J.S., Revollo-Fernández, D.A., Aguilar-Ibarra, A. & Georgantzis, N. (2016). “Economic behavior of fishers under climate-related uncertainty: Results from field experiments in Mexico and Colombia”. Fisheries Research, 183, 304- 317. https://dx.doi.org/10.1016/j.fishres.2016.05.020. Archibald, R.B. & Elliot, C.S. (1989). “Trial and Error Learning and Economic Mod- els”. Kyklos, 42, 39-59. Beck, N.L. & Katz, J.N. (1995). “What to do (and not to do) with time-series cross- section data”. American Political Science Review, 89(3), 634-647. https://dx.doi. org/10.2307/2082979. Becker, G. (1964). Human Capital. Chicago: The University of Chicago Press. Brañas P., Cardenas, J.C. & Rossi, M. (2009). Gender, Education, and Reciprocal Generosity: Evidence from 1,500 Experiment Subjects. ECINEQ Working Paper 2009, 128. Boone, C., De Brabander, B., Carree, M., De Jong, G., Van Olffen, W. & Van Wit- teloostuijn, A. (2002). “Locus of control and learning to cooperate in a prisoner’s dilemma game”. Personality and Individual Differences, 32(5), 929-946. https:// dx.doi.org/10.1016/S0191-8869(01)00100-3. Bowen, H.R. (1977). Investment in Learning: The Individual and Social Value of American Higher Education. San Francisco: Routledge. Brosig, J. (2002). “Identifying cooperative behavior: some experimental results in a prisoner’s dilemma game”. Journal of Economic Behavior and Organization, 47(3), 275-290. https://dx.doi.org/10.1016/S0167-2681(01)00211-6. Burks, S.V., Carpenter, J.P., Goette, L. & Rustichini, A. (2009). “Cognitive skills affect economic preferences, strategic behavior, and job attachment”. Proceed- ings of the National Academy of Sciences, 106(19), 7745-7750. https://dx.doi. org/10.1073/pnas.0812360106. Burks, S.V., Carpenter, J.P. & Verhoogen, E. (2003). “Playing both roles in the trust game”. Journal of Economic Behavior and Organization, 51(2), 195-216. https:// dx.doi.org/10.1016/S0167-2681(02)00093-8. 68 Guerrero, D.; Rosell, J.; Arroyo, J. S. Cancino, J.P., Uchida, H. & Wilen, J.D. (2007). “TURF’s and ITQs: Collective vs. In- dividual Decision Making”. Marine Resource Economics, 22(4) 391-406. https:// dx.doi.org/10.1086/mre.22.4.42629569. Cárdenas, J.C. (2000). “How do groups solve local commons dilemmas? Lessons from experimental economics in the field”. Environment, Development and Sus- tainability, 2, 305-322. https://dx.doi.org/10.1023/A:1011422313042. Cárdenas, J.C. (2003). “Real wealth and experimental cooperation: experiments in the field lab”. Journal of Development Economics, 70(2), 263-289. https://dx.doi. org/10.1016/S0304-3878(02)00098-6. Cárdenas, J.C., Chong, A. & Ñopo, H. (2009). “To what extent do Latin Americans trust, reciprocate, and cooperate? Comments evidence from experiments in six Latin American countries”. Economia, 9(2), 45-94. Certo, S.T., Busenbark, J.R., Woo, H.S., Semadeni, M. (2016). Sample selection bias and Heckman models in strategic management research. Strategic Management Journal, 37, 2639-2657. https://doi.org/10.1002/smj.2475. Christy, F. (1982). Territorial use rights in marine fisheries: definitions and condi- tions. FAO Fisheries Technical Paper No. 227. Rome: Food and Agriculture Orga- nization of the United Nations. Cobos, A.J., Due-as, J.D. & Velandia, M.C. (2016). “Aspectos economicos de la pesca artesanal”. In Diaz, J.M. Guillot, L. & Velandia, C.M. (Eds.): La pesca Artesanal en el Norte del Pacífico Colombiano: Un Horizonte Ambivalente (pp. 91-103). Bogota: Fundación MarViva. Croson, R. & Gneezy, U. (2009). “Gender differences in preferences”. Journal of Economic Literature, 47(2), 448-474. https://dx.doi.org/10.1257/jel.47.2.448. Deb, P. & Seck, P. (2009). Internal migration, selection bias and human development: Evidence from Indonesia and Mexico. Human Development Research Paper (HDRP) Series no. 31. De Jong, G. (2015). “The impact of social and human capital on individual coop- erative behaviour. Implications for international strategic alliances”. Critical Perspectives on International Business, 11(1), 4-29. https://dx.doi.org/10.1108/ cpoib-12-2012-0063. Departamento Administrativo Nacional de Estadística (DANE). (2005). Censo Gen- eral. Bogotá. Díaz, J.M. & Caro, N. (2016). “El mar como territorio y la pesca como actividad tradicional en el territorio Chocoano”. In Díaz, J.M., Guillot, L. & Velandia, M.E. (Eds.): La Pesca Artesanal en el Norte del Pacífico Colombiano: Un Horizonte Ambivalente (pp. 29-43). Bogota: Fundación MarViva. Eshel, I., Samuelson, L. & Shaked, A. (1998). “Altruists, egoists, and hooligans in a local interaction model”. American Economic Review, 88(1), 157-179. FAO. (1995). Code of Conduct for Responsible Fisheries. Rome: Food and Agricul- tural Organization of the United Nations. Influence of human capital on the trial and error learning process in a common... 69 Gaspart, F. & Seki, E. (2003). “Cooperation, status seeking and competitive behav- ior: Theory and evidence”. Journal of Economic Behavior and Organization, 51(1), 51-77. https://dx.doi.org/10.1016/S0167-2681(02)00139-7. Helbing, D. & Yu, W. (2009). “The outbreak of cooperation among success-driven individuals under noisy conditions”. Proceedings of National Academy of Sci- ences, 106(10), 3680-3865. https://dx.doi.org/10.1073/pnas.0811503106. Hernández, A. & Diaz, J.M. (2012). “La Zona Exclusiva de Pesca Artesanal en la costa norte del Pacífico colombiano: Escenario de conflictos socio-ambientales”. La Timonera, 17, 49-51. Huck, S., Müller, W. & Normann, H. (2001). “Stackelberg beats Cournot: On collu- sion and efficiency in experimental markets”. The Economic Journal, 111(474), 749-765. Huck, S., Normann, H.T. & Oechssler, J. (1999). “Learning in Cournot Oligopoly. An experiment”. The Economic Journal, 109 (454), 80-95. Huck, S., Normann, H.T. & Oechssler, J. (2000). Trial and error to collusion, the dis- crete case. Bonn Econ Discussion Papers, 6. Bonn: Graduate School of Econom- ics (BGSE), Universität Bonn. Huck, S., Normann, H.T. & Oechssler, J. (2004a). “Through trial and error to collu- sion”. International Economic Review, 45(1), 205-224. Huck, S., Normann, H.T. & Oechssler, J. (2004b). “Two are few and four are many: Number effects in experimental oligopolies”. Journal of Economic Behavior and Organization, 53(4), 435-446. https://dx.doi.org/10.1016/j.jebo.2002.10.002. Huettig, F., Kolinsky, R. & Lachmann, T. (2018). “The culturally co-opted brain: How literacy affects the human mind”. Language, Cognition and Neuroscience, 33(3), 275-277. https://dx.doi.org/10.1080/23273798.2018.1425803. Jones, G. (2008). “Are smarter groups more cooperative? Evidence from prisoner’s dilemma experiments, 1959-2003”. Journal of Economic Behavior and Organiza- tion, 68(3-4) 489-497. https://dx.doi.org/10.1016/j.jebo.2008.06.010. Karandikar, R., Mookherjee, D., Ray, D. & Vega-Redondo, F. (1998). “Evolving As- pirations and Cooperation”. Journal of Economic Theory, 80(2), 292-331. https:// dx.doi.org/10.1006/jeth.1997.2379. Kreps, D.M., Milgrom, P., Roberts, J. & Wilson, R. (1982). “Rational cooperation in the finitely repeated prisoners’ dilemma”. Journal of Economic Theory, 27(2), 245-52. https://dx.doi.org/10.1016/0022-0531(82)90029-1. Kirchkamp, O. & Nagel, R. (2007). “Naive learning and cooperation in network experiments”. Games and Economic Behavior, 58(2), 269-292. https://dx.doi. org/10.1016/j.geb.2006.04.002. Kosmidis, M.H. (2018). “Challenges in the neuropsychological assessment of illiter- ate older adults”. Language, Cognition and Neuroscience, 33(3), 373-386. https:// dx.doi.org/10.1080/23273798.2017.1379605. 70 Guerrero, D.; Rosell, J.; Arroyo, J. S. López-Angarita, J., Tilley, A., Díaz, J.M., Hawkins, J.P, Cagua, E.F & Roberts, C.M. (2018). “Winners and losers in area-based management of a small-scale fishery in the Colombian Pacific”. Frontiers in Marine Science, 53, 1-12. https://dx.doi. org/10.3389/fmars.2018.00023. Malik-Moraleda, S., Orihuela, K., Carreiras, M. & Duñabeita, J.A. (2018). “The consequences of literacy and schooling for parsing strings”. Language, Cognition and Neuroscience, 33(3), 293-299. https://dx.doi.org/10.1080/23273798.2017.13 13436. Moreno-Sanchez, R.P. & Maldonado, J.H. (2010). “Evaluating the role of co-man- agement in improving governance of marine protected areas: An experimental ap- proach in the Colombian Caribbean”. Ecological Economics, 69(15), 2557-2567. https://dx.doi.org/10.1016/j.ecolecon.2010.07.032. Ministerio de Agricultura y Desarrollo Rural, Colombia (MinAgricultura). (2013). Resolución 0899. Bogotá: Autoridad Nacional de Acuicultura y Pesca. Ministerio de Agricultura y Desarrollo Rural, Colombia (MinAgricultura). (2017). Resolución 2111. Bogotá, Colombia: Autoridad Nacional de Acuicultura y Pesca. Navia, A.F., Mejía-Falla, P.A., López-García, J., Mufioz, L.A. & Ramírez-Luna, V. (2010). Pesquería artesanal de la zona norte del Pacifico colombiano: Aportando herramientas para su administración. Fase 2. Technical Report. Colombia: Fun- dación SQUALUS. Nguyen ThiQuynh, C., Schilizzi, S., Hailu, A. & Iftekhar, S. (2017). “Territorial Use Rights for Fisheries (TURFs): State of the art and the road ahead”. Marine Policy, 75, 41-52. https://dx.doi.org/10.1016/j.marpol.2016.10.004. Offerman, T. & Sonnemans, J. (1998). “Learning by experience and learning by imitating successful others”. Journal of Economic Behavior and Organizations, 34(4), 559-575. https://dx.doi.org/10.1016/S0167-2681(97)00109-1. Ostrom, E., Gardner, R. & Walker, J.M. (1994). Rules, Games, and Common-Pool Resources. Michigan: Ann Arbor. Pallas, A.M. (2000). “The effects of schooling on individual lives”. In Hallinan, M.T. (Ed.): Handbook of the Sociology of Education (pp. 499-526). New York: Kluwer Academic/Plenum Publishers. Pallas, A.M. & Jennings, J.L. (2009). “Cumulative knowledge about cumulative ad- vantage”. Swiss Journal of Sociology, 35(2), 211-229. Plankis, B.J. & Marrero, M.E. (2010). “Recent ocean literacy research in United States public schools: Results and implications”. International Electronic Journal of Environmental Education, 1, 21-50. Ramírez-Luna, A.V. (2013). The Exclusive Fishing Zone for the Artisanal Fishery in Choco Colombia: Origins, Development, and Consequences for Artisanal Fish- eries and Food Security. Master Dissertation. Environmental Science Program. Memorial University of Newfoundland and Labrador. Influence of human capital on the trial and error learning process in a common... 71 Rassenti, S., Reynolds, S., Smith, V.L. & Szidarovszky, F. (2000). “Adaption and convergence of behavior in repeated experimental Cournot games”. Journal of Economic Behavior and Organization, 41(2), 117-146. https://dx.doi.org/10.1016/ S0167-2681(99)00090-6. Sáenz-Pacheco, G. (2014). Marviva, Pesca Responsable y Comunicación Para la Gobernanza en Bahía Solano y Nuquí (Pacífico Norte Colombiano). Barran- quilla, Colombia: Editorial Universidad del Norte. Satizábal, P. (2018). “The unintended consequences of ‘responsible fishing’ for small-scale fisheries: Lessons from the Pacific coast of Colombia”. Marine Policy, 89, 50-57. https://dx.doi.org/10.1016/j.marpol.2017.12.005. Satizábal, P. & Batterbury, S.P. (2018). “Fluid geographies: Marine territorialisation and the scaling up of local aquatic epistemologies on the Pacific coast of Colom- bia”. Transactions of the Institute of British Geographers, 43(1), 61-78.https:// dx.doi.org/10.1111/tran.12199. Schramm, A.J.H.C. (1998). What Have We Learned From Experiments? Gothenburg, Sweeden: Presidential Address, European Public Choice Society. Selten, R. & Buchta, J. (1998). “Experimental sealed bid first price auctions with directly observed bid functions”. In Budescu, D., Erev, I. & Zwick, R. (Eds.): Games and Human Behavior: Essays in Honor of Amnon Rapoport. Mahwak, USA: Lawrenz Erlaaum Association. Schoedinger, S., Cava, F. & Jewell, B. (2006). “The need for ocean literacy in the classroom: Part I”. The Science Teacher, 73(6), 44-47. Vainstein, M.H., Silva, A. & Arenzon, J. (2007). “Does mobility decrease co- operation?” Journal of Theoretical Biology, 244(4), 722-728. https://dx.doi. org/10.1016/j.jtbi.2006.09.012. Velez, M.A., Stranlund, J.K. & Murphy, J.J. (2009). “What motivates common pool resource users? Experimental evidence from the field”. Journal of Eco- nomic Behavior and Organization, 70(3), 485-497. https://dx.doi.org/10.1016/j. jebo.2008.02.008. Vieira, C.A., Díaz, M.C. & Díaz, J.M. (2016) “Ordenamiento y manejo pesquero en la costa norte del Pacífico colombiano”. In Diaz, J.M., Guillot, L. & Velandia, M.C. (Eds.): La pesca Artesanal en la Costa Norte del Pacífico Colombiano: Un Horizonte Ambivalente (pp. 45–57). Bogota: Fundación MarViva. Tan, J.H.W. & Zizzo, D.J. (2008). “Groups, cooperation and conflict in games”. The Journal of Socio-Economics, 37(1), 1-17. https://dx.doi.org/10.1016/j.so- cec.2006.12.023. Young, H.P. (2009).” Learning by trial and error”. Games and Economic Behavior, 65(2), 626-643. https://dx.doi.org/10.1016/j.geb.2008.02.011. Wooldridge, J.M. (2010). Econometric Analysis of Cross Section and Panel Data. Cambridge: MIT Press. 72 Guerrero, D.; Rosell, J.; Arroyo, J. S. Appendix A. Payoffs calculations Considering Equation (1), we assigned the following values w = 1525, a = 130, b = 5 and φ = 50. These parameters were chosen to obtain roughly equivalent outcomes to local daily earnings, as well as to ensure that these parameters are strictly positive. After optimizing and replacing parameters, we obtain the individual competitive harvest: = ( − )⁄2 = (130 − 50)⁄2 (5) = 8 Now, accounting for the social equilibrium solution, we replaced the parameters for a group of 5 players: =( − )⁄2 = 130 − 50(5) ⁄2 (5) = −12 For a group of 4 players, the parameters can be replaced as such: =( − )⁄2 = 130 − 50(4) ⁄2 (5) = −7 Because we cannot account for a socially optimal negative harvest during the ex- periment, we constrained xsoc =1. Therefore, the individual harvest ranges from 1 ≤ xi ≤ 8. Now, to calculate the payoffs, we replaced the parameters w = 1530, a = 130, b = 5 and φ = 50. We considered w as independent from the harvest and made an arithme- tic series with a marginal benefit function. In the case of 1 unit of private harvest, and 1 unit of harvest for each rival, the individual payoff is: = 1530 + ∑ ⁄ − = 1530 + ∑ ( −2 )− = 1530 + 130(130)(1) − 2(5)(1) − 50(4) = 1450 When the rivals’ harvest xj increases in 1 unit, the private benefit decreases in 50 points, which is the marginal cost of the externality φ. We used 4 rivals as a bench- mark and kept the same payoff structure for 3 rivals. We decided to use this approach to facilitate the sums and avoid quadratic calculations in the field due to the lack of access to calculators or any computational device. A3. Payoff structure According to the expression: = 1530 + ∑ (130 − 10 ) − 50 when a player increased their harvest xi by one unit (keeping the harvest of rivals xj constant), the payment increased; however, marginally decreasing. Now, when the rivals’ harvest xj increased by one unit (keeping the private harvest xi constant), the Influence of human capital on the trial and error learning process in a common... 73 individual payment decreased by 50 units, which was the cost of the externality. This decrease was constant. A.4. Payoffs Sheet for 4-player groups MY HARVEST 1 2 3 4 5 6 7 8 3 1,450 1,520 1,580 1,630 1,670 1,700 1,720 1,730 4 1,400 1,470 1,530 1,580 1,620 1,650 1,670 1,680 5 1,350 1,420 1,480 1,530 1,570 1,600 1,620 1,630 6 1,300 1,370 1,430 1,480 1,520 1,550 1,570 1,580 7 1,250 1,320 1,380 1,430 1,470 1,500 1,520 1,530 8 1,200 1,270 1,330 1,380 1,420 1,450 1,470 1,480 AGGREGATE HARVEST OF RIVALS 9 1,150 1,220 1,280 1,330 1,370 1,400 1,420 1,430 10 1,100 1,170 1,230 1,280 1,320 1,350 1,370 1,380 11 1,050 1,120 1,180 1,230 1,270 1,300 1,320 1,330 12 1,000 1,070 1,130 1,180 1,220 1,250 1,270 1,280 13 950 1,020 1,080 1,130 1,170 1,200 1,220 1,230 14 900 970 1,030 1,080 1,120 1,150 1,170 1,180 15 850 920 980 1,030 1,070 1,100 1,120 1,130 16 800 870 930 980 1,020 1,050 1,070 1,080 17 750 820 880 930 970 1,000 1,020 1,030 18 700 770 830 880 920 950 970 980 19 650 720 780 830 870 900 920 930 20 600 670 730 780 820 850 870 880 21 550 620 680 730 770 800 820 830 22 500 570 630 680 720 750 770 780 23 450 520 580 630 670 700 720 730 24 400 470 530 580 620 650 670 680 74 Guerrero, D.; Rosell, J.; Arroyo, J. S. A.4 Payoff Sheet for 5-player groups MY HARVEST 1 2 3 4 5 6 7 8 4 1,450 1,520 1,580 1,630 1,670 1,700 1,720 1,730 5 1,400 1,470 1,530 1,580 1,620 1,650 1,670 1,680 6 1,350 1,420 1,480 1,530 1,570 1,600 1,620 1,630 7 1,300 1,370 1,430 1,480 1,520 1,550 1,570 1,580 8 1,250 1,320 1,380 1,430 1,470 1,500 1,520 1,530 9 1,200 1,270 1,330 1,380 1,420 1,450 1,470 1,480 10 1,150 1,220 1,280 1,330 1,370 1,400 1,420 1,430 11 1,100 1,170 1,230 1,280 1,320 1,350 1,370 1,380 12 1,050 1,120 1,180 1,230 1,270 1,300 1,320 1,330 13 1,000 1,070 1,130 1,180 1,220 1,250 1,270 1,280 AGGREGATE HARVEST OF RIVALS 14 950 1,020 1,080 1,130 1,170 1,200 1,220 1,230 15 900 970 1,030 1,080 1,120 1,150 1,170 1,180 16 850 920 980 1,030 1,070 1,100 1,120 1,130 17 800 870 930 980 1,020 1,050 1,070 1,080 18 750 820 880 930 970 1,000 1,020 1,030 19 700 770 830 880 920 950 970 980 20 650 720 780 830 870 900 920 930 21 600 670 730 780 820 850 870 880 22 550 620 680 730 770 800 820 830 23 500 570 630 680 720 750 770 780 24 450 520 580 630 670 700 720 730 25 400 470 530 580 620 650 670 680 26 350 420 480 530 570 600 620 630 27 300 370 430 480 520 550 570 580 28 250 320 380 430 470 500 520 530 29 200 270 330 380 420 450 470 480 30 150 220 280 330 370 400 420 430 31 100 170 230 280 320 350 370 380 32 50 120 180 230 270 300 320 330 Influence of human capital on the trial and error learning process in a common... 75 Appendix B. Estimation of selection bias We made a quota sampling to try to ensure a rigorous procedure. Communities for the sample are Nuqui, which is the head of the municipality; Pangui, Coqui, Jovi, Termales, Arusi, Tribuga and Jurubira, which are corregimientos or rural villages that belong to the municipality. According to DANE (2005), 29.8 % of population live in Nuqui. The remaining 70.2 % are distributed in the rural villages. Among all of the communities, we obtained 159 participants. The largest attendance was in Nuqui, with 40 participants. This implies that we obtained 26.62 % participants from Nuqui, and 73.38 % from rural villages. A Heckman Model is estimated to correct the possible selection bias. We used the probability of inhabiting the head of municipality of Nuqui dN as the censored variable. As instrumental variables we selected age a, age squared a2. These variables were selected based on Deb & Seck (2009), who argue that age is a factor that deter- mines human migration from rural villages to urban centers in search of opportuni- ties and better income. In our case study, we can tell that Nuqui offers full school coverage, health centers and other sources of work than fishing. Finally, given the exclusion restrictions that require that variables of the first stage do not appear in the second stage (Certo et al., 2016), age is omitted (Table B1). TABLE B1 Players’ behavior: Heckman’s correction Variable 1 Schooling Heckman 2 xit–1 ES 0.231*** (0.0615) HS 0.0939 (0.0646) fxpi ES 0.151 (0.193) HS –0.103 (0.173) fzj ES 0.182 (0.187) HS –0.112 (0.166) domHS ES –0.0700 (0.390) HS 0.170 (0.349) s 0.400 (0.516) g –0.721* (0.407) Constant 2.828*** (1.011) 76 Guerrero, D.; Rosell, J.; Arroyo, J. S. Variable 1 Schooling Heckman 2 First stage 3 a 0.0759*** (0.0154) a2 –0.000554*** (0.000162) Constant –2.935*** (0.350)  0.624 (0.597) 1 Second stage dependent variable: Individual harvest xit. 2 Models estimated with two-stage least squares. 3 First stage dependent variable: Nuqui inhabitant dN. Significance: *10 %; **5 %; *** 1 %. The first stage exhibits that age and age squared are significant indicating that as players age increases, the probability to live in the head of the municipality increases; later this probability decreases after a turning point. The second stage shows some similarities with results presented in Table 4. These estimations share signs with variables such as xit–1, fzj, s and g. However, there are differences in the magnitudes and significance of their coefficients. For variables such as fxpi and domHS, we find significant differences in sign, magnitude and significance of coefficients. Due to this, we state 2 remarks: 1. The relation between players’ age and the probability of inhabiting Nuqui is quadratic. This means that players age is a predictor to explain players’ mobil- ity. This is largely due to the fact that these rural villages have school cover- age only up to 5 years of elementary school. Therefore, young people tend to travel to the head of the municipality to complete their high school studies. In general, these results show consistency with those shown by Deb & Seck (2009). 2. The coefficient that accompanies the inverse Mills ratio (IMR)  is not signif- icant. This indicates that this model does not contain an endogeneity problem due to selection bias (Certo et al., 2016; Wooldridge, 2010). We assume that models presented in Table 4 represent the behavior of the participants in the experiment more efficiently than a two-least square model. Influence of human capital on the trial and error learning process in a common... 77 Appendix C. Individual decision sheet Player No. Age Sex Years of schooling A: Individual B: Aggregate Harvest C (B–A): Aggregate Rounds D: Payoff Harvest of Group Harvest from Rivals Practice 1 Practice 2 Practice 3 1 2 3 ... 10 Total