The Azteca Chess experience : learning how to share concepts of ecological complexity with small coffee farmers

Small-scale coffee farmers understand certain complex ecological processes, and successfully navigate some of the challenges emerging from the ecological complexity on their farms. It is generally thought that scientific knowledge is able to complement farmers’ knowledge. However, for this collaboration to be fruitful, the gap between the knowledge frameworks of both farmers and scientists will need to be closed. We report on the learning results of 14 workshops held in Chiapas, Mexico during 2015 in which 117 small-scale coffee farmers of all genders (30% women) and ages who had little schooling were exposed by researchers to a natural history narrative, a multispecies network representation, a board game, and a series of graphical quizzes, all related to a nine-species complex ecological network with potential for autonomous control of the ongoing and devastating coffee rust epidemic that was affecting them. Farmers’ retention and understanding of direct and indirect bilateral interactions among organisms was assessed with different methods to elucidate the effect of adding Azteca Chess gaming sessions to a detailed and very graphical lecture. Evaluation methods that were better adapted to farmers’ conditions improved learning scores and showed statistically significant age effect (players older than 40 had lower retention scores) and gaming effect (lower retention of interactions included in the lecture but not in the game). The combination of lecture and game sessions helped participants better understand cascades of trait-mediated interactions. Participants’ debriefings confirmed qualitatively that they learned that beneficial organisms and interactions occur on their farms, and that gaming was enjoyable, motivating, and critical to grasp complex interactions. Many of the farmers concluded that the outcome of these interactions is not unique and not always in favor of rust control but is context dependent. Many concluded that there are feasible things they can do on their farms, derived from what they learned, to favor potential autonomous pest control.


INTRODUCTION
Small-scale shade-coffee producers and community ecologists studying coffee farms share a familiarity with the ecological complexity of diverse agroecosystems.The convergence and complementarity of farmer and academic "ecological knowledge frameworks" is an active field of inquiry with potential to inform research agendas and production processes in landscapes composed of small coffee farms (Vandermeer and Perfecto 2013).In the past three decades, there has been an acknowledgment that the standard top-down "transfer of technology" approach to agricultural research is frequently inadequate, and a transition to participatory-action research became common (Kindon et al. 2007, Blackstock et al. 2010).One of the most conspicuous of these programs is based on the idea of "farmer-to-farmer" knowledge-sharing (Holt-Giménez 2006), evidenced in the experiences of Farmer Field Schools (Damtie et al. 2011).Such knowledge-sharing implicitly and sometimes explicitly deals with complex social and ecological networks.However, most frequently, the ecological complexity has been reduced to simple and generalized recommendations-for example, integrated pest management techniques or nature-friendly and organic production.For coffee farmers, this translates into farmers exchanging knowledge concerning direct applications of materials or techniques aimed at very specific consequences-soil amendments, shade management, and judicious use of herbicides, fungicides, and insecticides (Damtie et al. 2011).
Farmer-to-farmer and farmer-to-scientist experiences in coffee farms have frequently focused on biological pest control (Jarquin et al. 2006).Most cases thus far explored have involved a single pest and one or two (sometimes exogenous) natural enemies (e.g., Gómez et al. 2012).Results of producing and liberating natural enemies into the field, commonly referred to as "classical biological control," have been modest because of ecological and logistic reasons but also, we argue, due to a knowledge and communication gap between scientists and farmers (Segura et al. 2004, Jarquin et al. 2005).The challenges of building common ecological knowledge increase significantly when dealing with more sophisticated programs of pest management (e.g., Lewis et al. 1997), such as so called "autonomous pest control" (Vandermeer et al. 2010).By their very nature, these more sophisticated programs acknowledge the existence of complex interactions involving many species.In their management practices, experienced farmers deal effectively with certain complex ecological processes, but most pay little attention to, or have little knowledge of, the behavior of the many small inconspicuous organisms that may be key to the operation of autonomous pest control, let alone those forces that indirectly, but significantly, relate to pest outbreaks (Perfecto and Vandermeer 2015).Conversely, ecologists committed to unraveling the details of pest management in coffee frequently lack the knowledge and tools to develop effective ways of incorporating their frameworks and findings into farmer-to-farmer and other participatory experiences in ways that help farmers (a) continuously update their management strategies based on a deep understanding of the specific case, and (b) develop better insights about the ecological complexity of other social-ecological processes that they must navigate (García-Barrios et al. 2016).
We report on the results of a series of workshops held in the Sierra Madre de Chiapas, Mexico during 2015 in which small-scale coffee farmers of all genders (30% women) and ages who had little schooling were exposed to (a) a natural history narrative, (b) a multispecies network representation and graphical analysis, and (c) a board game, all related to a complex ecological network with potential for autonomous control of the ongoing and devastating coffee rust epidemics (Hemileia vastatrix) affecting them.Pre-and post-game graphical quizzes were administered to evaluate the learning process that occurred through these activities.The basic ecological interactions of the nine-species network have been extensively researched empirically and theoretically by Perfecto and Vandermeer, with collaboration from other scholars and farmers (Vandermeer et al. 2010, Perfecto et al. 2014) as part of a broader coffee agroecology project (Perfecto and Vandermeer 2015 ).The game Azteca Chess-designed by the authors and named after the key ant species Azteca sericeasur-has been successfully tested with students for playability, engagement, and learning (García-Barrios et al. 2016).
Our objectives are to (a) share our workshops and "gaming methodology" with small-scale farmers as a resource to be added to more comprehensive and long-term field-based educational processes, (b) explain why and how our network-learning evaluation tools had to be better adapted to farmers' conditions during the process, (c) describe (in their and our terms) what farmers learned from this experience, and (d) define to what extent the gaming sessions made a difference in being able to understand and remember the network's interactions and deduce some of its consequences for autonomous pest management.

The Azteca autonomous pest management network
We provide a stylized and brief description of the Azteca network.For more details and ecological depth, see Vandermeer et al. (2010), Perfecto andVandermeer (2015), andGarcía-Barrios et al. (2016).Fig. 1 depicts a network of empirically verified interactions occurring among organisms on shade-coffee bushes.This is the network that has the potential to aid in the control of the coffee rust fungus.The sessile coffee scale insect, Coccus viridis, serves as alternative prey/host to the white halo fungus Lecanicilium lecanii, and thus facilitates the latter's predatory action on the coffee rust.However, the scale is heavily consumed by the adults and larvae of a lady beetle, Azya orbigera, and by a group of parasitoid wasps (in Fig. 1 represented by the "scalekiller wasp").But a tree-nesting ant (Azteca sericeasur; Azteca hereafter) protects scales in exchange for the honey dew they extrude.Azteca patrols scale colonies and scares away all but the beetle's larva, which is protected from the ants by the waxy filaments that cover its body.The beetle larva is controlled by parasitoid wasps (mainly Homalotylus shaviakane, plus other wasps in the family Encyrtidae; in Fig. 1 represented by the "larvae-killer wasp").This complementary ant/wasp protection of the scale is compromised by (a) the ant itself, since the Azteca ant does not discriminate and also scares away larvae-killerwasps, and (b) the parasitoid fly, Pseudacteon lascinosus (in Fig. 1 represented by the "head-hunting phorid fly"), that attacks Azteca ants when they are moving and triggers a temporary halt in the patrolling activity of the ant as a defense mechanism.The reduced activity of the ant when the parasitoid flies are present provides a window of opportunity for the adult lady beetle to oviposit under the scales (and also eat scales), and for the scalekiller and larvae-killer wasps to approach their hosts.Thus, we see two main types of interaction: predatory (X consumes Y), and https://www.ecologyandsociety.org/vol22/iss2/art37/trait-mediated (Z's behavior modifies X's behavior, which changes X's predatory efficiency over Y).Various levels of interaction occur.In the first level, X harms Y by eating, parasitizing, or scaring it; in the second level, Z benefits Y by harming X, an enemy of Y; and finally, a higher level that is represented by the cascading effects from levels 1 and 2. The predatory and traitmediated interactions that cascade through different pathways from the fly to the coffee scale can ultimately lead to one of two attractors (local exuberant coffee scale colonies and local coffee scale extinction).The theoretical and empirical relations between coffee scale and rust dynamics are even more complex, and are explored in Vandermeer et al. (2014) and McCook and Vandermeer (2015).

Azteca Chess
The features and rules of this board game are thoroughly described in García- Barrios et al. (2016).We explore its use as a learning and communication tool between farmers and researchers.Nevertheless, in Appendix 1, we provide a graphic summary of the game and links to the manual, for the reader's convenience.In Fig. 2, we display the hexagonal-cell game board, which highly stylizes a transversal section of a coffee bush.It exposes the initial spatial display of tokens representing different organisms.

First set of workshops
Two sets of workshops with coffee farmers were organized: five in the Sepultura Reserve in April and nine in Sepultura and Tacaná reserves in July-August 2015 (Fig. 3).
LGB and JCM had conducted previous work in the region and were trusted by many of the participants.In the first workshops, family members of all genders and ages were invited by public broadcast to a lecture that addressed the rust problem, followed by Azteca Chess training, a game tournament with prizes, and a small dinner followed by a reflection session.With a limit of 16 players per workshop, a total of 72 people aged 12-70 and with a deficient rural primary school level attended and played, and 67 completed all quizzes.Workshops lasted 4 hours.
During a 45-minute PowerPoint-supported lecture, we: 1. presented the basics of the rust's biology and data on regional damage caused to coffee producers in the region; 2. explained and practiced with farmers a "happy-face grammar" to establish the notion and graphical representation of first-and second-level ecological interactions.For this purpose, we used the simplest network: the rabbit harms [angry face-first level] a crop; the fox harms [angry face-first level] the rabbit; the fox benefits [happy face-second level] the crop; and 3. constructed step-by-step-and thoroughly explained-the natural history and network representation of each element in Fig. 1.At intervals, volunteer participants were asked to come up to the stage and use the grammar to follow the interaction paths down to the scale and rust to deduce if a given organism had the potential to indirectly benefit or harm the farmer (Fig. 4).The lecture's goal was to familiarize farmers with the diversity of organisms involved, their elaborate behaviors and choices, their types and levels of interactions, and their ultimate effect on scales and, potentially, on rust.Immediately after the lecture, participants were asked without previous notice to answer a graphical, semi-open quiz "A" in 7 minutes.The same procedure was repeated after the gaming sessions (approximately 3 hours later).Quiz A (Figs. 5 and 6) explored mainly retention (memorization) and understanding, the first two of Bloom's taxonomy of learning levels (Anderson   et al. 2001).Players were asked to connect beneficial and harmful faces departing from each organism to one or more organisms affected by such behaviors.To obtain retention scores per player, we counted only matches with 14 of the 18 bilateral interactions presented in Fig. 1.The 14 were selected to make the average interaction levels of the nine "in the game" interactions and the five "not in the game" interactions as similar as possible.Comparing pre-and post-gaming quiz scores does not immediately yield the gaming effect because the difference in scores results from the lecture's learning effect minus the forgetting effect plus the positive effect of quizzes on countering the forgetting effect (Roediger and Karpicke 2006) plus the gaming effect itself.Comparison of pre-versus post-gaming quiz scores only allows knowing if all post-lecture activities were able to counter or even overcompensate the lecture-forgetting effect.We will call it the post-lecture effect.To isolate the gaming effect, two performance scores (0-10 scale) were calculated per player: one for the nine "in the game" interactions and another for the five "not in the game" interactions.This was done separately for the pre-and post-game quiz A. Fifty-six of the respondents were younger than 41, and 11 were between 41 and 68 years of age.We noticed that the second age group more commonly struggled to follow the workshop activities and to do the quizzes, so in all statistical analysis, age group was considered as an additional explanatory factor.The pooled set of "in the game" and "not in the game" scores obtained in the pre-and post-game A quizzes was subject to a three-way Univariate General Linear Model analysis; i.e., a UGLIM (pre/post quiz x "in the game?" x age group).Additionally, paired t tests were performed to compare pre-versus post-game sets of scores for each age group.All tests were performed in SPSS (Version 16, 2007, SPSS Inc., Chicago, Illinois, USA).https://www.ecologyandsociety.org/vol22/iss2/art37/Participants were trained in Azteca Chess by LGB through a 20minute lecture and a step-by-step hands-on demonstration, followed by personalized clarifications during the game by JCM and LGB.Each participant played three times (Fig. 7).Post-game collective reflections were conducted by JCM.They were not tape recorded; only notes were made on paper.

Second set of workshops
A different set of 50 players from the Sepultura and Tacaná subregions participated in these workshops, again conducted by LGB and JCM.Interesting findings, inconclusive results, and perceived weakness of the learning evaluation process motivated six changes in our learning evaluation tools, which were decided by all authors: First, the Azteca network and game-training lectures-and the game itself-were unaltered, but the nine workshop groups were limited to six players each to allow people to perform three supervised post-game quizzes through question and answer (Q&A) dialogues with a researcher.Groups were formed by open invitation made by a local contact-farmer in each community.
Participants were informed they were considered to be working collaborators who were exploring a learning method, so each received a 1-day rural salary of US$10.Fifty participants completed all activities.Half were aged 12-40 and half were 41-68, again with a deficient primary school level as average schooling.Participants played three times in rotation with others.Workshops lasted 4-5 hours.
Second, before starting the workshop, each player was interviewed briefly by JCM about his/her knowledge of species dwelling on coffee bushes, their interactions, and their effect on pest problems.
Third, the semi-open "A" quiz was substituted with a closed "B" quiz (Figs. 8 and 9) that was meant to capture in a step-by-step, unambiguous, and less overwhelming way the same information, as well as to familiarize the participant with responding quizzes.The player held the graphical quiz, and the researcher asked them to decide, for each of 16 bilateral interactions, (a) if it was included in the game, (b) if species X benefited or harmed species Y, and how (to distinguish knowing from guessing).Researchers were extremely careful not to hint, approve, or disapprove responses.Two performance scores (0-10 scale) were calculated per player: one for the 10 "in the game" interactions and another for the six "not in the game" interactions.The pooled set of scores was subject to a two-way UGLIM (age group x "in the game?") to reveal the game effect.To test if this quiz "B" procedure was better able to capture the possible gaming effect than was quiz "A", we pooled the two workshop set data and performed a three-way UGLIM, comparing scores from post-game quizzes "A" and "B".
Additionally, the frequency of players correctly answering each of the 16 bilateral interactions was calculated and arc-sin transformed.A linear regression model with dummy variables was used to explore how such frequencies were affected by age group (0 = young; 1 = older), "in the game?" (0 = no; 1 = yes), level (1, 2), and type of interaction (predatory = 0; trait-mediated = 1).
Fourth, to explore if players recognized trait-mediated interactions in the game, the researcher offered the player a board with tokens and asked them to make the proper moves to show (a) how an ant scares away a beetle, (b) how an ant scares away a wasp, (c) how a fly favors a beetle by killing a contiguous ant, and (d) how a fly's mere presence deters ants from approaching a beetle.All except (c) are trait-mediated interactions.
Fifth, high-level cascades along the Azteca network were explored qualitatively and briefly with farmers during the lecture (e.g., the fly harms the ant that harms the beetle that harms the scale that benefits the fungus that harms the rust).We developed a postgame hands-on quiz (the cascade quiz) (Fig. 10) in which six medium-sized cards of the sequence (fly-ant-larvae-killer wasplarval beetle-scale-rust) were presented to the player as an initial condition.We then changed the current state of the network by increasing the fly's population (the size of the card) and asked the player to propagate the consequences on other species' populations by changing, if necessary, the size of the subsequent cards (smaller and bigger card were available).Players were asked the reason for each change they made to discriminate correct answers from guessing.We registered how many steps the player propagated correctly along the sequence and built the frequency distribution of correct steps for all players.Before taking the cascade quiz, the farmer practiced with a similar story using foxrabbit-plant cards.As with the Lotka-Volterra dynamical model, the Azteca network exhibits complicated oscillatory behaviors due to its many predator-prey interactions, which were not discussed with the farmers.In this exercise, we obviously avoided such complications and focused on a very stylized form of cascading quantitative effects in a single time step.
Sixth, after players were personally interviewed for quizzes, we held a collective reflection session about the workshop, the game, and the practical implications of the experience.Unlike the first set of workshops, all reflection sessions were recorded and transcribed.Thirty-five farmer quotes were selected for this paper.

First set of workshops
The five tournament sessions were successful in that all 72 participants engaged in the lecture and game, and 67 completed both "A" quizzes and participated in reflection sessions.
The average pre-and post-game quiz scores, which measured retention of bilateral interactions, were both low (around 50% correct answers).Table 1 shows there was a significant age group effect (younger did better than older) but no gaming effect, as measured both by the pre-versus post-game score comparison and the "in" or "not in" the game comparison.The younger age group performed slightly better than the older group.For the younger group paired t tests, "in the game" post-game scores were marginally higher (5.73; p = 0.09) than "in the game" pre-game scores (5.32).This first set of gaming experiences revealed an important age group effect, but only slight and statistically marginal post-lecture effects and gaming effects that could not counter the effect of partially forgetting the lecture details after 3 hours.Participants were surprised to know how many small organism live and actually interact on a coffee bush, were interested in the fact that these animals act pro or against coffee rust control both directly and indirectly, and were amazed and amused by some of the insects' elaborate behaviors.Overall, only one-fifth of participants recalled and/or could graphically represent 60% or more of the direct and indirect relations in the Azteca network.Yet in all debriefings and reflection sessions, players commented positively on more general aspects of the learning experience.They concluded that insecticides should be applied carefully or not at all to avoid killing beneficial insects.Finally, they considered that learning the game was challenging at first, but later it was fun and rewarding, and some noted that the game complemented the lecture or was absolutely necessary to understand the interactions.

Second set of workshops
A different group of 50 farmers from the Tacaná and Sepultura subregions participated in this second set of workshops.While more than 80% of players could list, before the workshop, five or more vertebrates and invertebrates that could be found on their coffee bushes, 75% were unaware of some of the interactions, and the same proportion did not know that some interactions can control pests.Interestingly, unawareness was somewhat lower in the younger group.Thirteen percent of young players and twenty percent of older players spontaneously said that most species found on coffee bushes are pests.The average player had 73% correct answers in the single quiz applied post-game (i.e., average score = 7.3).Sixty percent of players got a score between 8 and 10 (i.e., good to excellent).Table 2 shows significant age group and gaming effects on quiz "B" performance.Younger players got higher scores than older players (8.1 versus 6.5; p < 0.0001), and "in the game" scores were higher than "not in the game" scores (8.4 versus 6.1; p < 0.0001) Table 3 compares the post-game quiz scores of the two sets of players that were exposed to quiz "A" and quiz "B", respectively.It confirms significant age group effects across both workshop sets and a significant effect of the second set of workshops on capturing the gaming effect (see the "workshop x gaming" interaction effect).Thus, with the new and unambiguous quiz "B" format, filled through a step-by-step Q&A interaction between player and researcher, the effect of the game and of age group on the player's capacity to retain and understand network interactions became starkly apparent.Fig. 11 graphically summarizes the various statistical trends presented, and the positive effect of the quiz "B" procedure.Figs. 12 and 13 show, for the younger and the older group, respectively, the frequency of players who recalled each of the 16 bilateral interactions in the lecture.According to the linear regression model (Table 4), the frequency of correct answers to a given interaction was significantly increased by being an "in the game" interaction, was reduced by being the answer of an older player, and was not influenced by the interaction's level (1 or 2)  and type (trait-mediated or predatory).In other words, the frequency of correct answers did not change significantly by being a higher level interaction or a trait-mediated interaction.Most players did not have trouble distinguishing in the quiz which bilateral interactions were included in the game (92% of answers were correct).Table 5 shows that almost every young person could translate the most relevant trait-mediated level 1 interactions into token moves on the board, while only two-thirds could do so for the trait-mediated level 2 interactions.Older players followed the same pattern but at lower frequencies.
After the game, 64% of players were able to elucidate the cascading effects of a change (sudden fly increase) on a six-level interaction pathway, while 25% did not understand the exercise and/or could not go beyond [more fly -> less ant].Performance was not significantly related to age group.
In Appendix 2, we present 35 selected farmer quotes on the different topics addressed during post-game reflection sessions.
Here, we reproduce five: Ecology and Society 22(2): 37 https://www.ecologyandsociety.org/vol22/iss2/art37/ 1. "With the game I understood a bit better, because with words only, I get drowsy-that's the truth-every time I attend a workshop.In the game, as I look at the board and move the tokens, and as I see how the good and the bad animals eat each other, I really get to understand which helps me and which doesn't." 2. "On the wall we only see [PowerPoint] figures.But once in the game, it's as if we were seeing it in reality: each animal did his act, defending his life, giving life to others he helps.It benefits our minds because we have to think, we have to analyze what we are about to do; it clears our mind because in our mind everything is so tangled, but once our mind can focus on how we are going to deal with the situation…so yes, it was great!" 3. [The youngest player 12 years old; already a farmer]: "The fly helps the beetle by killing the ant.The fly also helps the beetle by hovering above the ants so that they can't approach; this way of helping the beetle is more difficult to perceive.The beetle is freed from the ant and now has no problem to eat the scales, and without these, the rust can better reproduce.Now, the wasp kills the larva beetle so the rust won't reproduce, but if ants become abundant again, they frighten the wasp and it flees from the coffee bush." 4.
[Researcher]: "Which outcome do you expect more frequently on a coffee bush: the scale population flourish and rust is somewhat controlled; scale population remains low; scale population is eliminated and rust flourishes."Answers: (1) "It depends on what animals are there: for example, if the fly is absent the beetle is busted; it depends."(2) "When there are more ants and wasps, scales increase, but ants are in danger also, so sometimes they win and sometimes they lose.Sometimes they are more abundant and sometimes less." 5. "We are very aggressive with ants and all animals, but today we are learning about them, and I will take to my family the message that dear God sent us today through you.You came today to awaken our belief that our coffee farms are not 100% lost to rust: We have defenders, ants that are struggling for us.We are very rude with them, but starting today, we are going to give ants a little bit of freedom; we won't mess with them, just let them be there."

DISCUSSION AND CONCLUSIONS
Farmer Field Schools (FFS) and other "farmer to farmer" learning experiences with scientists' involvement have contributed to strengthening the capacity of legions of rural people to develop and share with others control strategies that seek to promote comprehensive knowledge and adaptive management.Yet, results are mixed, such learning is not easy to evaluate, and many challenges remain (Henk van den Berg: FFS evaluation 2004 report for the Global IPM Facility).The situation described is similar for small-scale coffee farm experiences (Damtie et al. 2011).In Sierra Madre de Chiapas coffee farms, important efforts have been made by researchers to (1) identify the relevant insects and fungus (e.g., Barrera 2008), (2) explore pest versus natural enemy bilateral interactions (e.g., Gómez et al. 2012, Jackson et al. 2012), and (3) describe farmers' knowledge and develop coffee FFS and participatory biological control programs (e.g., Segura et al. 2004, Jarquín et al. 2006, Barrera 2008).The ongoing rust epidemics are exhibiting processes that invite the actors involved to gain an even broader and shared understanding of the complexity of autonomous pest control, emerging from cascading, multispecies trait-mediated interactions.
The coffee rust pandemic that started in 2012 in the Sierra Madre de Chiapas has been devastating for coffee farmers.Many resorted to heavy use of fungicides during the acute phase and are partially or totally substituting varieties of Coffea arabica, which are rust susceptible, high quality, and shade-tolerant, with varieties that are rust resistant but reportedly of lower quality.Some are https://www.ecologyandsociety.org/vol22/iss2/art37/planting the resistant (thus far) species Coffea canephora (also known as robusta), well known to be of low quality.Others are waiting for the epidemics to pass and are substituting dead coffee plants with the same or less susceptible C. arabica varieties (Valencia, personal communication).While short-term responses alleviate the crises generated by the rust epidemic, in the long term it is strategic to pay more attention to complex networks of species that interact with each other and keep pests under control.It is important to convey that autonomous pest management is not a simple recipe or a "magic bullet" but rather a complex, contextdependent process (Lewis et al. 1997, Vandermeer et al. 2014) that can be embraced and explored adaptively through long-term building of a collective agroecological culture among the different actors involved.
Small-scale coffee farmers have sophisticated ecological knowledge about many processes occurring in their farms (Vandermeer andPerfecto 2013, Valencia et al. 2015), but as the literature reports and our preworkshop survey confirms, they know their shade-coffee farms thrive with life but pay little attention to many organisms unless their harm is significant (Segura 2004, Jarquín 2005, 2006, López-del-Toro et al. 2009); they are rarely aware that some pesky organisms (e.g., ants and scales) and their inconspicuous ecological associates indirectly exert potential autonomous control over rust and other coffee pests (Perfecto and Vandermeer 2015).Furthermore, they generally lack a framework for learning about subtle ecological processes that would improve pest control at broad spatial and temporal scales (Rebaudo and Dangles 2015).
In our case, a first small step in providing such a framework to farmers was to explore with them the learning effects of combining ( 1 Results suggested that either the learning evaluation tool was not adequate or that forgetting the interactions as presented in the lecture was not prevented by the subsequent activities.We observed that the open-ended quiz "A" for recalling and reconstructing bilateral interactions was prone to different interpretations, and/or it created challenges that not all players could deal with when left alone to work with this tool.We decided to modify, adapt, and expand our evaluation procedures to avoid overwhelming farmers and to probe and better understand their learning. In the resultant, second set of workshops, quiz "B" addressed explicitly and systematically the interactions to be recalled.Scores improved significantly for both age groups, and the effect of the game and of age group on the player's capacity to retain and understand network interactions became starkly apparent.Regression results suggest that the workshop and game might have countered the difficulty of grasping and retaining level 2 and trait-mediated interactions.Most players did not have trouble distinguishing which bilateral interactions were included in the game, and almost every young player could translate the most relevant trait-mediated level 1 interactions into token moves on the board, while only two-thirds could do so for the trait-mediated level 2 interaction.Older players followed the same pattern but at lower frequencies.After the game, two-thirds of players (both younger and older) were able to elucidate the cascading effects of a population change.
Farmers' comments during reflection sessions confirmed qualitatively that participants learned that potentially beneficial organisms and interactions occur on their farms, and that gaming was enjoyable, motivating, and critical to grasp complex interactions.Many of the farmers concluded that the outcome of these interactions is not unique and not always in favor of rust control, but is context dependent.Most saw that there are feasible actions derived from what was learned (tolerate ants and keep the trees they use to nest, tolerate scales, reduce pesticides, pay more attention to small organisms and their behaviors, etc.).Farmers also gave researchers insights on how they learn; how they https://www.ecologyandsociety.org/vol22/iss2/art37/sometimes struggle with lectures, gaming rules, and quizzes; how more practice and time could allow them to master the topic; and how field visits would consolidate the learning process.Overall, the effects of learning and evaluation tools displayed in the second set of workshops show that a significant proportion of small-scale coffee farmers were capable of dealing with a complex ecological interaction network, and deriving general lessons, changes in attitudes, and potential actions.The general learning experiences will probably persist in participants' minds, even as the details might fade away.Any actions that farmers might take as a result of this experience are not part of our study's framework, but the literature reports a significant effect of coffee farmer learning on their subsequent actions (Damtie 2011).To better define these actions and their actual pest control capacity, some farmers requested future discussion and work in the field about specific conditions and managements that could foster a significant effect of white halo fungus over coffee rust in their farms.As stated in the Introduction, these are context-dependent, open questions which need-and offer the opportunity for-collaborative onfarm research.Diniz et al. (2015) report that it is unusual and difficult to involve farmers in multilevel interaction network analyses, and Mani et al. (2013) discusses how financial worries partially impair poor farmer's capacities for such complex cognitive task.Therefore, it is encouraging that younger farmer participants performed very well with the B quiz, and as well as outstanding urban students did in a previous set of workshops using the A quiz (García- Barrios et al. 2016).Azteca workshops need to be further adapted, and complemented with field visits, more so for older participants who struggled with this learning approach.
We are confident that farmers and their allies will become interested in Azteca workshops, both for understanding this specific network but mainly for a better appreciation of the complexity of agroecosytem ecological networks (Benitez et al. 2014, Perfecto andVandermeer 2015).We expect others to significantly contribute to further adapting these workshops and to incorporating them into broader participatory research and learning experiences in small-scale coffee farmer territories.The kind of abilities Azteca workshops seek to promote in small-scale farmers (observing subtle elements and processes, conceiving and integrating their networked interactions, and mobilizing the latter through game simulations) might be a stepping stone toward even more ambitious goals such as empowered and effective small-scale farmers' participation in multiactor social-ecological analysis and decision-making processes (e.g., Etienne et al. 2011, d'Aquino and Bah 2013, Diniz et al. 2015).
If the knowledge framework gap between ecologists and farmers regarding complex agroecological issues is to be reduced in both directions, farmers' interests and capacities to better understand the ecology of their farms should not be preconceived, overestimated, or underestimated.As researchers, we need to go further in our dialogue with farmers, be sensitive to cultural differences in dealing with complex processes (Strohschneider 2002), and learn how to facilitate learning in ways that empower small-scale coffee farmers, both to understand and take action in their own fields and to allow them to fully participate in both mainstream and critical multiactor deliberations and decisions about complex social-ecological processes that strongly affect them (García- Barrios et al. 2015).
Responses to this article can be read online at: http://www.ecologyandsociety.org/issues/responses.php/9184

Fig. 1 .
Fig. 1.Each organism projects an action (face) through a line directed toward another organism(s) or interaction(s) which it affects.An angry face signifies harm.Black solid lines mean harm by consumption.The black dashed line means the scale facilitates harm done by the white halo fungus to the rust fungus.The behavior of the ant that changes the behavior of predators is represented by dashed blue lines (trait-mediated interactions) projected upon predation interactions.The behavior of the fly that changes the behavior of the ant is represented by dashed red lines projected upon blue lines.Happy faces with gray solid lines represent the capacity of the scale to provide honey dew to the ant, and of the adult lady beetle to oviposit and produce its larvae.Species marked with an asterisk are included in the workshop lectures but not in the game.(Figure modified after Fig.5.14 in Perfecto and Vandermeer 2015.)

Fig. 2 .Fig. 3 .
Fig. 2. Azteca Chess board game with initial display of tokens.Pink circles are exclusively for the phorid fly to move clockwise in each round.

Fig. 4 .
Fig. 4. Participants explain how interactions cascade down the network (e.g., the fly harms the ant that harms the beetle that harms the scale…).

Fig. 5 .
Fig. 5. Quiz A. Network elements of Fig. 1 are redisplayed in a circle.The quiz was handed out without the red and green lines in this figure.Players were asked to connect, with lines, the beneficial and harmful faces departing from each organism to one or more organisms affected by such behaviors.For analysis, we counted only matches with those connections represented in the figure, and which correspond with 14 of the 18 lines in Fig.1(i.e., the presence of these lines in the player's quiz was considered correct; their absence incorrect).

Fig. 8 .
Fig. 8. Quiz B. In each of the 16 rows, participants circle the correct face, and later cross the square if the interaction is in the game.

Fig. 9 .
Fig. 9.One of the researchers (LGB) assisting a farmer who is filling in quiz B, through a Q&A dialogue.

Fig. 10 .
Fig. 10.The cascade quiz.The initial condition, the change in the fly population, and the correct dynamic consequence.See text for further explanation.

Fig. 11 .
Fig. 11.Comparison of learning performance measurements (post-game quiz "A" and "B" scores) as affected by workshop set and its quiz type, by age group, and by gaming effect.For each of the eight combinations of these three variables, we provide the box and whiskers distribution and, in bold, the average score.Open circles are statistically atypical data.Asterisks are statistically very atypical data.

Fig. 12 .
Fig. 12. Frequency of 25 younger participants (%) who correctly identified each of the 16 interactions in quiz B. The network is the same as in Fig. 1 but has a different spatial arrangement to reduce number crowding as much as possible.Black lines = first-level interaction, predatory effect; gray lines = first-level interaction, beneficial effect; dashed black line = second-level interaction.The coffee-scale facilitates rust predation by the white halo fungus; dashed blue lines = traitmediated, first-level interactions through which the ant harms two wasps and the adult beetle (as predator and as egg layer).Dashed red lines = trait-mediated, second-level interactions through which the fly reverts the ant's potential harm on the wasps and adult beetle.The three species marked with an asterisk are not included in the game.Number colors: black = the interaction is in the game; green = the interaction is not in the game.

Fig. 13 .
Fig. 13.Frequency of 25 older participants (%) who correctly identified each of the 16 interactions in quiz B. Symbols as in Fig. 12. Larger bold number = frequency was lower than in the younger group.

Table 1 .
Variables explaining quiz "A" scores.Univariate General Linear Model.Note: Dependent variable: retention scores obtained by players for interactions in and not in the game, before and after playing Azteca Chess (pre-and post-game "A" quizzes applied during the first set of workshops).Age groups: scores of younger and older players.In the game?: score for interactions in the game vs. score interactions not in the game.Pre vs. post: pre-game quiz "A" total score vs. post-game quiz "A" total score.

Table 2 .
Variables explaining quiz "B" scores.Univariate General Linear Model.Note: Dependent variable: retention scores obtained by players for interactions in and not in the game (postgame "B" quiz applied during second set of workshops).

Table 3 .
Variables explaining quizzes "A" and "B" scores.Univariate General Linear Model.Note: Dependent variable: retention scores obtained by players in post-game "A" and "B" quizzes applied during first and second set of workshops.

Table 4 .
Frequency of players who correctly identified each of 16 bilateral interactions, as a function of age group, interaction in or not in the game, and level and type of interaction, for the second set of workshops.

Table 5 .
Frequency of younger and older players who could translate interactions into moves on the board.Younger = 12-40 years old; Older = 41-68 years old.