Probability Games, Workshops, and the Social World of International Science Communication

,

AU G U S T 2 0 2 2 E1748 P robabilistic thinking underpins a wide range of scientific claims from precipitation (Joslyn et al. 2009) to wildfires (Dacre et al. 2018) to oceanic conditions (Alves et al. 2013).We need probabilistic information when the variability of what we predict or explain cannot be described fully through mechanistic models-when uncertainty is a fundamental part of what we seek to understand.But while probabilistic thinking is a valuable resource, in practice, people differ in how they respond to uncertainty (Henrich and McElreath 2002).These differences are especially evident when contrasting highly different social groups like foreign scientists and smallholder farmers (Akay et al. 2012).Despite the substantial work devoted to more effectively communicating forecasts to smallholder farmers, the inability to adequately or appropriately communicate probabilistic information remains a major challenge for informed decision-making (Gigerenzer et al. 2005;Hartmann et al. 2002;Millner and Washington 2011;Orlove and Tosteson 1999;Pennesi 2007;Roncoli 2006).This article shares how one diverse group of researchers at first failed, and then ultimately succeeded in communicating probabilistic information of seasonal climate and crop yields to farmers, extension workers, and local officials in four farming communities in East Africa.More broadly, we show how researchers learned to harness the power of social ties to infuse our relationships with stakeholders with warmth, open-mindedness, and understanding to ultimately strengthen the communication of scientific knowledge in all directions.
When our early efforts to communicate probabilistic information through the forecast bulletins failed, we drew on the structural holes in our international, interdisciplinary scientific collaboration-"people connected across groups [who are] more familiar with alternative ways of thinking and behaving" (Burt 2004, 349-350)-to communicate across the cultural divide between scientists and our public.In our collaboration, conational social scientists who shared language and history with our community members and language and training with our engineers, rose to the challenge of communicating across groups especially effectively.
What does this contribute to research on forecast communication?Our work starts from the participatory best practices promoted in prior studies-coproducing knowledge with local stakeholders, using culturally grounded practices.But what about coproduction and using culturally grounded practices produces these good outcomes?Previous studies highlight two productive mechanisms: information sharing and trust in social ties (Özer et al. 2011;Gbangou et al. 2020).Our work draws insights from the sociology of emotions (Bericat 2016) and research on embeddedness and social ties (Granovetter 1985;Burt 2004) to more effectively capture a third mechanism that helps to produce successful science communication: these practices can also influence people's feelings toward a person, experience, or resource.Recognizing "the social nature of human emotions, and the emotional nature of social phenomena" (Bericat 2016), we can study experiences that often feel personal to the researcher-emotions, friendships, guesting norms-as social phenomena, identify patterns and connect them to aspects of the broader social world like interactions (Hochschild 2002), organizations and social networks (Parker and Hackett 2012), or even the environment (Norgaard and Reed 2017).
AFFILIATIONS: Amdework Atsbeha and Negatu-Addis Ababa University, Addis Ababa, Ethiopia; Holzer, Anagnostou, and Kirksey-University of Connecticut, Storrs, Connecticut; Block and Alexander-University of Wisconsin-Madison, Madison, Wisconsin We focus our analysis on an unglamorous institutional workhorse of international development and scientific enterprises: the stakeholder workshop.Workshops are regularly taken for granted, mentioned offhandedly in the reports and articles that focus instead on outcomes ("several stakeholder workshops were organized to disseminate such and such").The sterile descriptions belie the social reality that we live-the tension or excitement or boredom, the nerve-racking wait for the participants to show up, the technical disasters narrowly averted (or not), the thrill of understanding.Yet these workshops are an important, understudied part of the social organization of science.We build on the small, but interesting body of research on workshops, which has documented some persistent challenges and best practices: the most effective workshops are interactive, use problem-based learning, and contain concise messaging supplemented by well-designed materials (Lickiss and Cumiskey 2019;Racovita et al. 2013;Stewart et al. 2018;Colle et al. 2021), but even the most well-designed workshop can still fall prey to the inequalities in the social world (Colle et al. 2021;O'Connell et al. 2020).In Ethiopia, workshops have largely been used to compile and disseminate findings to stakeholders rather in the science communication approach (Awulachew et al. 2009;Tadesse et al. 2015;Worku et al. 2012), providing farmers with forecast information, assessing traditional forecasting techniques or validate information gathered from community members (Iticha and Husen 2019;Balehegn et al. 2019;Bekele 2020).
Thus, in this paper, we ask, How can researchers recognize, strengthen, and draw upon the social connections between workshop participants and between workshop participants and researchers to foster the good humor, trust, and commitment necessary for effective workshops?We present a political-institutional approach to science that mobilizes the social relationships between people working as scientists and those using scientific innovations and share one major success that came from our efforts: we learned one way to effectively communicate probabilistic information in seasonal forecasts of weather and crop yields with farmers, extension workers, and water managers.
Below, we briefly introduce the study site, data and methods, and intervention, and then share how we learned the lessons, beginning with confusion and ending with success.

Study site, data, methods, and intervention
The project explored farming communities and water management in the Blue Nile basin, Amhara region, Ethiopia (Fig. 1).As an interdisciplinary social science and engineering study, the project examined the process of scientific collaboration not just the outcomes of forecast communication.Drawing on Burawoy's (1998) extended case study approach, we used the forecast bulletins as an intervention that "by mutual reaction" allowed us to "discover the properties of the social order" (p.14), treating mistakes and misdirections as part of the social process to be studied (cf.Vaughan 1996;Beaulieu 2010).
Study site.Across the basin, rain-fed agriculture remains the dominant sector, so seasonal precipitation has a large impact on food security and economic well-being.The Blue Nile basin receives plentiful annual precipitation (on average more than 1,000 mm yr −1 near our study site), but precipitation is strongly seasonal and mostly falls during the June-September kiremt season (Yates and Strzepek 1998).In addition, the basin experiences both temporal and spatial interannual variations which may result in proximal areas experiencing dissimilar climate conditions and precipitation totals (Conway 2000(Conway , 2005;;Segele and Lamb 2005).Ethiopia's interannual variability in local precipitation ranks as one of the highest globally (World Bank 2006) stressing local communities through droughts or floods, which have led to continued land degradation and economic and food insecurity (World Bank 2010).These seasonal and interannual variations directly impact rain-fed agricultural livelihoods, and influence reservoir and groundwater availability for Unauthenticated | Downloaded 06/28/24 05:09 PM UTC irrigated agriculture during the dry season, resulting in substantial impacts to agricultural planning and production.
While the Blue Nile basin has the physical resources to drive regional economic growth through irrigated agriculture and hydropower development, its vulnerability to exceptional hydrologic variability and sensitivity to regional and global climate change have limited this development (World Bank 2010).Further, the Blue Nile basin contributes over 70% of the Nile flow, so water management decisions deeply influence all of East Africa (Conway 2005).Hydrologic extremes are also expected to increase under climate change (Soliman et al. 2009;Roth et al. 2018;Lala et al. 2020).In combination with rapid population growth and increased land use pressures, this may further deteriorate living conditions in the region (Patz et al. 2005;NMA 2007).Therefore, these communities need strategies to adapt to climate variability to foster their food, water, and economic security.
The four farming communities in the study are predominately composed of smallholder subsistence farms, which represent 95% of agricultural production in Ethiopia and 85% of all employment (FAO 2020).Two of the communities rely on rainfed agriculture-Rim (also spelled Reem) and Dangishta-while the other two communities include substantial irrigation-Kudmi (Koga irrigation scheme) and Gayta (surface irrigation systems).The four communities are largely representative of the Amhara region in being agriculture dependent and homogeneous in religion and ethnic composition.
Methods.We conducted 14 months of ethnographic fieldwork on science, technology, and agriculture in the four farming communities dividing fieldwork over four agricultural seasons (kiremt wet season 2019; bega dry season 2019; kiremt 2020; bega 2020), three preliminary field visits to establish and maintain ties to communities (July 2015; June-July 2017; January 2019), and three months of preseason fieldwork to develop a baseline before our interventions (February-July 2019).We primarily use field notes from kiremt 2019 for this article, but the broader study also includes subsequent field notes from 2019 to 2020, 1,856 household surveys, semistructured interviews with farmers and water managers, and dozens of informational interviews with farmers, extension experts, and other related actors.Conversations and interviews were conducted in Amharic by the researchers and translated into English in the field notes and transcripts.We analyzed the dynamics of science communication surrounding the forecast bulletins by first open coding field notes on interactions directly tied to the development of the bulletin and the workshops, systematically identifying and labeling patterns with limited presuppositions.Probabilistic thinking emerged as a key theme in this initial analysis, which we used as the central focus for a second focused coding.To capture systematically the issues surrounding probabilistic thinking and the workshops, we coded field notes to identify for each workshop: 1) what (mis)understanding of the forecast bulletins emerged, changed, and ultimately became ascendant; 2) what patterns emerged in the interactions at the workshops.We identified key patterns tied to social connectedness, understanding, and emotions, and drew on the broader research data to add depth to these findings as needed.Throughout this analysis, we approached falsification as the process of strengthening, eroding, or correcting prior expectations.
The intervention.For the kiremt wet season, the focus of this article, we developed localscale prediction models of hydro-agricultural-climatic variables that farmers, extension workers, and other stakeholders told us had value for them: season onset, total precipitation, soil moisture, crop yields, and irrigation storage volumes.We developed statistical models to predict season onset and total precipitation, using local and global climate patterns that are known to influence moisture transport into our study region (Lala et al. 2021;Alexander et al. 2019).The forecasts are available at 0.25° spatial and 6-hourly temporal resolution, covering the entire upper Blue Nile basin.These predictions are combined with seasonal forecasts from the National Centers for Environmental Prediction (NCEP) Climate Forecast System version 2 (CFSv2) (Saha et al. 2014).These forecasts that are initialized at 0000, 0600, 1200, and 1800 UTC each day and go out to 7 months, are inputs to a Coupled Routing and Excess Storage land surface model (Lazin et al. 2020) to provide a prediction of soil moisture conditions at the beginning of the season.Next, the expected wet-season precipitation and preseason soil moisture were used in a decision support system for agrotechnology transfer (DSSAT) model to translate expected moisture into predictions of maize and teff crop yields, since impact to crops (and food security) is the primary consequence of climate variability for farmers (Yang et al. 2020;Jones et al. 2003).The forecasts are available at the same spatial and temporal resolution as the input data.In short, our team developed and evaluated season-ahead climate forecasts to provide indications of expected season conditions prior to planting with the intention of alleviating some of the vulnerability of communities in our study to the differing amounts and timing of precipitation.In this study we used forecasts from the first 15 days of April each year to forecast the wet season precipitation and crop yield using a chain of hydrological (Alexander et al. 2020) and analog based approaches to issue probabilistic forecasts on seasonal precipitation amount, lake inflows and crop yields (Alexander et al. 2020) for selected crop types.The forecasts are available at 0.25° spatial resolution.To assess accuracy, we did a comparison against soil moisture citizen science data that were available at two of the farming sites.The comparison showed that our predictions are mostly within the CSI wetness indications.
We collected data on participants' understandings of the forecast bulletins at three stages of the intervention.First, prior to the workshop intervention, we used informational interviews and focus groups composed of extension experts or farmers to assess understanding.We used a simple oral question-and-answer approach: we gave respondents a forecast bulletin, allowed time for them to review, and then in Amharic, we asked for each section, What do you think [box 1] is saying?We wrote down their answers, and then assessed them for accuracy.
At the workshop during the intervention, we evaluated understanding at two additional points: before and after the probability games.We divided participants into groups, distributed bulletins, and asked them to review the bulletins so they could tell us what they meant.We collected ethnographic observations of a random sample of the small group discussions.Then we went from person to person asking them to explain what a section of the bulletins was saying.The one limitation of this oral technique was that a person asked later could have a correct answer, but be unintentionally convinced that they are wrong by the answers of their peers.But the evaluation needed to be oral because of the limited literacy rate, and that scenario was not highly likely based on the preintervention informational interviews.The postintervention evaluation was integrated into the probability games themselves to match the pedagogical practices common to the area: Ezana Amdework posed a question, waited for a volunteer to answer, and if no volunteer came forth, asked a specific person to answer, and then moved onto the next person with a related, but slightly different question.We observed two indicators of understanding, a direct indicator that we developed deliberately, and a second, subtle indicator that emerged in the analysis of field notes.The first indicator was whether they answered the question correctly.The second more subtle indicator was a shift in the word choice the person used to answer the question.We would then finish the exercise with a collective question-and-answer to reinforce the message.
We recruited participants in a purposive sample of community leaders, woreda and kebele extension experts, and influential farmers following best practices in agricultural development and technology adoption in Ethiopia.The ethnographers drew on their observations as well as district level agriculture experts to select workshop participants, who were then invited by the district level expert.The majority of the participants were influential farmers, "big men" involved in multiple community organizations or respected elders and a few influential women farmers.In addition to the farmers, we included a sizeable contingent of agricultural extension experts from the kebele and a few from the district.The limitations that stemmed from using this model of agricultural development in research is that participants were disproportionately people with power and influence in community.Demographically, the majority were older, male, and landowners, so we are unable to speak to the wider applicability of our model to more disadvantaged groups.Other observations from the fieldwork suggest stronger skepticism about technology and more fatalism about the weather that would limit direct forecast adoption among the very poor (e.g., sharecroppers).But poorer farmers do tend to make key agricultural decisions (e.g., planting in anticipation of the onset of rain) by following what successful neighboring farmers are doing, so that hierarchal model of agricultural development remains relevant.

Results and discussion
Creating visuals to communicate localized forecasts.As the time to communicate the forecasts approached, we began to develop visual representations of these probabilistic forecasts-forecast bulletins.The kiremt bulletin (Fig. 2), for example, communicated expected rainfall and soil moisture and described how these climate conditions may affect crop yields probabilistically, reflecting a possible range of outcomes.As discussed in the methodology section, a series of seasonal weather and crop yield forecasting dynamic and statistical models were executed in the first 15 days of April each year to issue probabilistic forecasts on the wet season onset, seasonal precipitation amount, and lake inflows, and crop yields for selected crop types.
We carried the visions of our forecast scientists and our community members back and forth (and back and forth) for about 5 months, drafting and redrafting the bulletins based on conversations with people likely to use the bulletins.We altered visuals, added text, then removed them again, added new text, changed colors and shapes.We settled on visually communicating the probabilistic nature of the results for the season onset using a multicolored calendar and for the total rainfall using a pie chart.Finally, when we had done as much as we could, the rest of the team left sociologist, Ezana, and engineer, Sarah Alexander, sitting under the canopy in a small hotel for one last slog at translation.This was the English version that we developed for Kudmi, one of the four communities (the climatic conditions varied across the four study areas, so we tailored the bulletins for each community).
The problem: Conceptual distance between localized forecasts for scientists and localized forecasts for the public.Our first training workshop was not a success.Despite our attempts to refine visuals and text based on user feedback, our field observations from the workshop and focus group interviews with respondents showed clearly that our community members did not understand the probabilistic nature of the information.We could not rely solely on the printed bulletin to communicate probabilistic information effectively.We feared that our community members would ultimately find the probabilistic information unintelligible, a challenge that others had encountered before us (Gigerenzer et al. 2005;Hartmann et al. 2002;Millner and Washington 2011;Orlove and Tosteson 1999;Roncoli 2006).We had already incorporated best practices in forecast communication, tying the current forecast to Unauthenticated | Downloaded 06/28/24 05:09 PM UTC a comparable recent year instead of the ambiguous "normal" or "rainy" categories, including crop yield forecasts not just rainfall, incorporating complementary indigenous forecasting knowledge, presenting highly localized, microscale forecast (Pennesi 2007).Yet we had no success.How could people use our seasonal forecasts if they could not incorporate probabilistic thinking into their decision-making?For the forecast scientists on our team, probabilistic thinking was vital.The forecasts were based on a combination of mechanistic and analog methods applied on different crop types based on a 35-yr atmospheric reanalysis dataset in the upper Blue Nile basin (Lala et al. 2021).To share insights from forecast models as though they were certain-rains will start in the last week of May-would be wholly ineligible, more than just inaccurate, a falsehood.Between our forecast scientists and our community members, we faced what we call the problem of incommensurability (Kuhn 1962).We lacked a common language and method for seeing the weather, leaving us unable to communicate across the divide.Or so we thought.
Fortunately, our collaboration was not just a research and development project-it was a graduate education project, too.Through coursework, graduate brown bags, fieldwork, and recruitment, we fostered researchers who could serve as structural holes in international scientific collaborations, "people connected across groups [who are] more familiar with alternative ways of thinking and behaving" (Burt 2004, 349-350).The challenge for scientists, like all social groups, is that knowledge and behaviors are much more homogeneous within a group than between groups.In our collaboration, conational social scientists who shared language and history with our community members and language and training with our engineers, rose to the challenge of communicating across groups especially effectively.But ultimately, we all learned the difference between the dissemination model of science and the open, back-and-forth of effective scientific communication (NASEM 2017).
Lesson 1: Harness structural holes in interdisciplinary, international scientific collaborations to find hidden opportunities in the cultural context.
Sociologists, Ezana and Elizabeth Holzer, sat in the Bahir Dar University office of Ethiopian engineering colleague Mamaru Moges talking through the unambiguous lack of understanding that we had witnessed.Then we had a thought: Could we make an exercise about probability that scaffolded our probabilistic information to everyday understandings of clouds-a key part of indigenous forecasting techniques-and more systematically build on existing experiences and knowledge?Other researchers have had at least some modest successes in using serious games and simulations to promote learning about physical scientific processes such as climate variability and change (Flood et al. 2018) or behavioral patterns to promote adaptive and actionable responses (Suarez et al. 2014).Luseno et al. (2003), for example, asked pastoralists to distribute stones into piles that represented possible rainfall amounts and found that most did exhibit probabilistic thinking.Other studies reported less positive results on the extent to which decision-makers understood or incorporated insights from such games and simulations into their own decisions (Patt and Gwata 2002;Roncoli 2006;Lemos and Dilling 2007;Unganai et al. 2013;Gunda et al. 2017).Games and simulations have the potential to promote learning, understanding, and uptake of probabilistic information, but ultimately, they must be contextualized to the decision-maker's frame of reference (de Suarez et al. 2012).
For farmers, analyzing clouds was a crucial part of the local context of weather forecasting.Leveraging this existing knowledge, we built new training workshops.At the workshop, Ezana would hold up a picture of a cloud and ask, what is the chance that it will rain with these clouds?To tie the probabilistic information to decision-making, Ezana would then ask community members what they would do if they saw these clouds using a range of specific activities that were important to them as concrete prompts.The cloud presentation always Unauthenticated | Downloaded 06/28/24 05:09 PM UTC prompted spirited debate.Below, we share field notes from several workshops, and then broaden the discussion to the lessons we learned from these experiments.
Reimagining stakeholder workshops: Playing probability games with neighbors.We arrived at the Rim kebele office on a cloudy Thursday morning to be greeted by three of the extension workers and the kebele 1 chairman.Our household surveys had shown that local extension workers and influential farmers called "model farmers" were the most trusted sources of forecast-based guidance, a finding that was consistent with other development projects in Ethiopia, though not without sociopolitical downsides (see Lefort 2012;Elias et al. 2013).
We waited for the workshop to begin, sitting in the dark meeting hall.How atomized and anonymous participants can appear, running a workshop as an outsider!But in truth, the participants were all enmeshed in social ties.Tigist, 2 the veteran extension worker, sat next to her colleague Fentahun, the head of the kebele agriculture office.On the same bench, another extension worker, who was also a local priest (in Amharic, Qes), sat next to Qes Teshager, another local priest and a farmer who served on the kebele administration and security committee.Not long after, Masresha joined this bench.Masresha, a respected farmer, who was a member of the regional council, a shopkeeper, and on the committee overseeing a community-led road construction project had come late.Sisay, our contact at the woreda office, had had to call Masresha to get him to show up.They were quite close-Sisay fondly recounted eating at the elder's home when he had first started working as a young field supervisor at the woreda agriculture office.So he called Masresha, and after a warm exchange of greetings, Sisay gently chastised him for being late.On the next bench sat Ambachew, a young farmer married to Qes Teshager's sister, sat with Bekalu and Zelalem.Despite his relative youth, Ambachew knows many other farmers from his leadership roles: deputy chair of the local party organization, head of one of the development teams at his kebele, and a former two-term member of the leadership of the local multipurpose cooperative.Bekalu, who sat beside him, was the kebele chair, and on his other side sat Bekalu's brother Zelalem, the current chair of the cooperative.Like Masresha, Zelalem is a model farmer and one of the "big men" leading an important civic road construction project, and as such works closely with both the kebele chair and our woreda key informant.
But why share with our readers in such painstaking detail the lives of a small group of workshop participants they will never meet?The reason for it is that the key to the success of the probability games lies not just in the cultural relevance of the games themselves-a point that has been well substantiated in other research as well-but in our social relationships with workshop participants.We delivered these games and examples during the workshops in a manner attuned to the social world of our community members-warmhearted, egalitarian, and anchored in the fundamental fact of social connections: between us and the community members and between the community members themselves.
When it came time for the cloud activity, Ezana held up the first picture, which shows a mostly blue sky with white clouds.He asked, "Will it rain?"Everyone agreed that this cloud did not carry rain.Then Ezana tied this assessment of the probability of rain to a concrete example of decision-making.He asked Qes Teshager if the first cloud warranted moving inside clothes hung to dry outside.(It did not.) Then Ezana held up the second picture, which showed a dark cloudy sky, with a narrow patch of light at the center.He asked Masresha, who was a shopkeeper as well as a farmer, if he would move his goods inside if he sees such a cloud.(He would.)Then Ezana turned to Bekalu and asked if he was threshing his finger millet and saw clouds like those in the second Unauthenticated | Downloaded 06/28/24 05:09 PM UTC picture, would he move his grain inside?Bekalu replied that he would try to finish threshing quickly and move his grain inside as it is most likely to rain.
The third picture caused the most animated discussion.The picture showed fierce clouds low in the sky.While we had chosen it as a "definitely rain" cloud, a sizeable contingent replied that, in their experience, such clouds were unlikely to bring rain as they were often accompanied by strong winds that would blow away the clouds.Others held our interpretation.Ezana asked Tigist if these clouds necessitated moving inside grains spread to dry outside.She explained that it would be difficult to decide because, as her neighbors had pointed out, strong winds often blow away such clouds.But, she said, if she was going away from the house, she would ask one of her neighbors to watch the grain and move it inside if it did start to rain.Then Zelalem elaborated that if the day is hot and there are no winds, clouds like those in the third picture are very likely to bring rain.Ezana asked him, "What if the day is hot but windy?" Zelalem replied that it would mean less chance of rain.
By the end of the exercise, our participants were answering clearly and decisively, agreeing with or debating each other's decisions.That people could disagree about how to interpret the clouds helped us drive home the probabilistic nature of weather forecasting-ours as well as traditional techniques.The discussion clarified the matter.It showed the first clear understanding of our probabilistic weather forecasts that we had witnessed not just among the extension experts, but the farmers, too.
Buoyed by our success with the cloud activity, we moved onto other scaffolding exercises.We made probability games from the probabilities of livestock giving birth, a blindfolded person picking a pen from among straw, catching a bajaj (a little three-wheeled taxi), the price of cattle rising or falling under various scenarios, and livestock being swept away by a flood in the local river.The examples shared two key characteristics-they were probabilistic events and they occurred in the everyday lives of the community members.Below, we share two more field notes from our training workshops to exemplify the social processes that underlies our political-institutional approach to probabilistic communication.
The 1-in-10 chance: Fishing a pen From the straw while blindFolded.On the day of the training workshop, Ezana and Holzer arrived in Kudmi and stood under the giant warka tree, as we had done a dozen times or more before.Earlier that week, Ezana had asked one of our collaborators, the woreda official, Sisay to arrange for us to run a training workshop to share our seasonal forecasts.Together, Ezana and Sisay identified three influential farmers who were likely to actually use the forecast and share their knowledge with others.Sisay proposed that he would have the farmers invited through the local community manager, and he volunteered to invite all the agricultural extension workers in the community and even the district extension department head.
Our workshop participants sat forward in the hard-backed benches, legs crossed, extension workers holding their usual small notebooks.We had spent months in Kudmi, and this was no longer an anonymous group, who we would treat as disconnected strangers.There was Ayele, sitting beside his colleague Dejen, who was Sisay's distant cousin on his mother's side.Sisay sat beside Dejen with Berhanu on his left-Berhanu was a relative on Sisay's father's side.On the second bench, Lakachew, the district extension department head, sat next to Gashaw, the local agriculture office focal person, who was his direct subordinate.Next sat Chalachew, who was a great uncle to Dejen, and a former long-time head of the farmers' union at the district.Qes Arega, a respected priest, father confessor to many in the community, and former local chairperson, sat next, quietly writing in his notebook.Later, we learned that Qes Arega was Yayeh's brother; Yayeh was the head of one of the irrigation users' associations, and hence, a colleague to Berhanu, head of the second irrigation users' association in the community.
Last sat Elias, an extension agent in Kudmi, who was a substitute for another extension agent who had gone away to school at the beginning of the year.
Embedded in the mundane details of names and ties lies the basic insight of our connection to the workshop participants and our recognition of their relationships with each other.Neighbors, colleagues, relatives-in all likelihood, if they had not known each other before the workshop, at least they knew a zemed, a catchall term for anyone who is not an immediate family member, but some kind of relative, including distant and marital relatives.
After the welcome, Ezana picked some straws from the ground and took a handful of pens from Holzer.He placed 10 straw sticks and pens in different combinations.First, he placed one pen and nine sticks, and asked what the chances are of picking up the pen with your eyes closed.Several people murmured tinish idil (in English, unlikely).A participant said there was a 1-in-10 chance of getting the pen.Ezana then asked a participant to come try for the pen, and the man came to the front.
Cover his eyes or he will peek, his neighbor shouted, teasingly.Among the rest, laughter.Ezana jokingly with exaggerated care blindfolded his hapless volunteer with his shawl, spun him around and guided him down to reach the pen, if he could.(He could not.)His neighbors hooted at him as he made his way back to his seat.Ezana changed the ratio of sticks to pens, prodded them to tell the new odds, and let them good-naturedly volunteer one by one with lower-medium probability, medium, higher-medium, and finally, a high probability to capture a pen.They cheered each other's good (or bad) fortune, while Ezana alternated between teasing and drawing the connections back to the forecast.That they fully grasped the concept of probability, was evident by their correct answers to Ezana's subsequent questions, that they enjoyed each other's company, in their laughter, nodding, and joking.
We were not a longstanding part of the web of group affiliation that linked our participants.But our approach to workshopping allowed us to be engidoch (in English, guests), to tap into the rich social ties to harness the good humor, goodwill, and commitment that is hard to muster when teaching a group of strangers-or when teaching a group of people treated as unconnected nameless "workshop participants" and so unintentionally encouraged to act as though they were strangers.
Lesson 2: Mobilize local social networks to harness the good humor, trust, and commitment of workshop participants.
Using analogies For the probability oF a one-time event: when does a cow give birth?Farmers often face a large gap between the information they need and the seasonal forecast information available (Hansen et al. 2011).But as part of the back and forth between our forecast scientists and community members, carefully cultivated by our sociologists over the years, we had managed to develop a prediction for a piece of information much sought after by farmers: the onset of rains.But our stakeholders struggled to read the calendar that we had made to convey this much desired information.Even the extension experts falsely believed that the calendar told them the amount of rain expected for any single day rather than the probability that the rains would begin in that date range.The failure of our audience to understand-our failure to effectively communicate these ideas-made us fear that our understanding of the probabilistic onset of the rainy season was simply too far removed from those of our stakeholders.
The onset of rain represented a particular type of probabilistic information-the probability of a one-time event.Community members dealt with the probability of a one-time event regularly.It was just that they did not tend to think of onset as a probabilistic one-time event and were not accustomed to our visual representations.So after considerable time brainstorming, Ezana came up with two examples-livestock birth and payouts from an iqub, an informal credit and saving practice-and we crafted probability exercises anchored in these aspects of everyday life.Below, we will share how we used the example of a cow giving birth.
We arrived for our workshop at the large public meeting hall in Dangishta.Sunlight streamed through the cracks in the mud-plastered wall, in gaps between the wall and the tin roof, and through the wide doorway, brightening the room enough for everyone to see our bulletins.Our workshop participants sat in the front rows on rough plank benches affixed to the ground or in wooden chairs that must have been taken from the schoolhouse for some other, more crowded public event.Seleshi, the crop expert, was seated up front with us while Gebre, the head of the kebele agricultural office, stood to lead the discussion.We were practicing handing over the training activities to our collaborators at the local agricultural extension office.On the right side of the hall, Mulugata, the kebele administrator sat next to Wondossen, the youth league representative, and Anteneh, an influential farmer.On the left, Asrat, another influential farmer, who previously participated in the first workshop in Dangila, was seated between the only two women in the group, both influential farmers: Sinedu and Kelem.Sinedu and Asrat knew each other well, having both been part of another science, technology, and development project, one run by Bahir Dar University that provided solar water pumps to farmers. 3 Asrat had served as the focal person for the project, following up on the work done by Sinedu and the other project participants.
The kebele agricultural office head, Gebre began to explain the forecast.He let everyone take some time in small groups to review the bulletins themselves.He asked for a volunteer to explain the onset of rain calendar.Unsurprising, Asrat spoke first, and he described it as amount of rain, not probability of rain.Mulugata and Anteneh and others followed with similar interpretations.Even our stakeholders who had taken part in our early Dangila training-our first, unsuccessful workshop-misunderstood the probabilistic information.We needed probability training, but fortunately we were prepared this time.
Ezana stepped forward to model this part of the training for Gebre.He asked the participants, When does a cow give birth?Several participants answered over each other: When the cow is 9 months pregnant.He prodded: But what if the cow is 8 months and 2 weeks pregnant?One of the farmers replied, Yes, still possible, that does not happen as often, but it happens some of the time.Then Ezana said, What about when the cow is 9 months and 2 weeks pregnant?Several people replied yes, and someone elaborated: that is common some of the time.He asked, But what about 9 months and 3 weeks?(Oh yes, that is possible, but that happens only if it is too late-very rare.)Ezana said, So what you have told me is that when the cow is 9 months, she has a very high probability of giving birth; when she is 9 months and 2 weeks, there is a moderate probability of giving birth, but as for giving birth at 10 months that is a low probability."That is right," people said, nodding.
Then he shifted the conversation back to the forecast bulletin, connecting the analogy and probability of onset."So when will the onset of rain start according to the scientists?"Ezana asked, holding up the bulletin.Is it Ginbot 4 15-24?Moderate probability, several participants replied.So when is the highest probability of the onset of rain for the season?The participants reply in unison, demonstrating clear understanding of the probabilistic forecast that they had misattributed earlier that morning: Ginbot 25-Sene 2. The exercise had worked.
Once again, Ezana drew it all together scaffolding the probabilistic dimensions of onset with the probabilistic dimensions of calving.He said, Remember the probability of a cow not giving birth up to the tenth month is highly unlikely.Similarly, there is a low probability that the onset of rain will be at the end of the month of Sene.What we have talked about (pointing to the colors on the calendar) is the probability of onset of rain and not the amount of rain.What is the highest probability of onset?Once again, a chorus of correct answers-the 3 For details on this other science and development project, which focused on solar pumps and other solar technologies, see www.agrilinks.org/post/solar-powered-irrigation-could-boost-climateresilience-millions. 4 The Ethiopian calendar has 13 months and starts in a different year than the Gregorian calendar that we use.
community members had connected their already existing understanding of probabilistic information to our forecasts.Our forecasts had become intelligible.
Lesson 3: Use analogies to help community members connect existing understandings of probabilistic information to forecasts.

Concluding thoughts
Through trial and error and by harnessing conational social scientists as structural holes in international interdisciplinary research, we created exercises that scaffolded our probabilistic forecast information to everyday understandings of clouds, farming, and other commonplaces, systematically built on the existing knowledge of our community members.The games and examples that we developed all shared two key characteristics: they were probabilistic events, and they occurred in the everyday lives of the community members.The games were simple, relevant culturally, and easily replicated or tweaked to fit different populations.But to identify what probabilistic events community members experienced regularly, we had to be grounded in the same social world.Our political-institutional approach to workshopping mobilized local networks to make ourselves engidoch-guests-to tap into the rich social ties to harness the good humor, goodwill, and commitment that people who live and work together forge over time.
Good humor, goodwill, and the resulting commitment are not random or incidental to the scientific process, but social phenomena that can be studied and cultivated like any other scientific practice or resource.Research has shown that a participatory, coproduction approach has positive outcomes, but a gap remains in understanding the process by which this approach produces the positive outcomes.In addition to the already recognized mechanisms of information sharing and trust building through social networks, our work has shown a third mechanism through which coproduction and participatory approaches can affect social action: by influencing participants' feelings toward a person, experience, or resource.
The lessons that we took from this research project stemmed not just from the probability games, but also from how we came to this solution-that, too, is part of the sociological study of science communication.One contribution of this paper is to clarify the role of certain knowledge brokers that occupy structural holes in interdisciplinary, international science collaborations and, ultimately, allowed us to capture a different approach to solving problems of communicating probability.Research has shown that working across disciplines have produce strong results; likewise, research has shown that international collaborations that incorporate local scholars work better.But a gap remains in identifying how international, interdisciplinary outcomes produce these good outcomes.We have identified one characteristic of interdisciplinary, international research networks: the potential for these networks to develop structural holes that link different flows of information to produce novel solutions.
A pressing area of future study remains in handover exercises that would allow local extension workers to run probability games for each other and for farmers.We have piloted two handover exercises, but have not yet been able to overcome their tendency toward highly formal, one-sided statements that do little to evoke the warm, good humor and trust that we consider fundamental to the probability games.A second area for future research is to more directly study the strategy's dynamic with disadvantaged community members.Our forecast communication model does not challenge the current best practices for agricultural and technological communication in the area, a model that reinforces existing hierarchies.

Fig. 1 .
Fig. 1.Map showing the Blue Nile (Abay) basin, Ethiopia, where our four study communities are located.

1
We treat the kebele, the smallest administrative unit, as the community.Kebeles are grouped into woredas, which is usually translated as "district."Woredas are organized into zones; zones are organized into regions.Regions are comparable to what we call states in the United States. 2 Names of community members are pseudonyms.