Hemp Growing

It is also possible that gender norms themselves slowly respond to female employment, in which case the longer term impact on violence could differ from the deleterious effect observed here. Rather than suggesting that female employment should not be encouraged, the evidence presented here indicates that economic theory, domestic violence policy and female employment programs should take the costs to men of violations of traditional gender roles seriously – insofar as such violations prove costly for women.Farming is inherently knowledge intensive. This knowledge base is multi-faceted and context specific, and often informed by scientists, researchers, policymakers, government, extension agents as well as by farmers. While farmer knowledge is a critical component of this knowledge base, in the United States farmer knowledge has been widely underappreciated . Long considered “informal” knowledge, farmer knowledge is generally not regarded as scientifically valid and therefore infrequently recorded, whether formally or informally . Since the 1950s, due to an increase in knowledge standardization within production agriculture combined with widespread deskilling among farmers and farm workers, farmer knowledge has become increasingly undervalued . However, farmers who practice alternative agriculture often amass an incredible wealth and depth of knowledge that integrates multiple ways of knowing and reflects diverse knowledge systems for thinking about evidence; perhaps most importantly, farmer knowledge is based in practice . If current trends in consolidation of land ownership, dry rack cannabis chemical-based intensification of agriculture, and standardization of farmer knowledge continue, local farmer knowledge may be endangered or permanently lost .

Before this occurs, it is essential that we elevate the critical role of farmer knowledge and: 1) understand the key features of farmer knowledge; 2) understand the substance of farmer knowledge; and 3) systematically document farmer knowledge in specific local contexts. Understanding the substance of farmer knowledge serves as a first step to conserve this essential knowledge base in practice; however, it is equally critical to document the particularities of farmer expertise in local contexts to provide essential knowledge for other contemporaneous and future generations of farmers, scientists, and policymakers alike. Moving forward, there is therefore a need to elevate the importance and value of farmer knowledge across multiple disciplines such that farmer knowledge is considered “expert” knowledge throughout alternative agriculture . While other studies attempt to integrate the artificial binary between “formal” and “informal,” or “expert” and “non-expert” knowledge and view the two forms of knowledge as complementary , in this paper we maintain that farmer knowledge is scientifically valid, expert knowledge and therefore warrants formal, standalone documentation within the scientific literature .While it is true that the terms “traditional,” “folk,” and/or “indigenous” knowledge are applied in certain contexts, in this paper, the term “local knowledge” is most appropriate , as farmer participants were all white and all either first- or second generation settlers on unceded Patwin-speaking Wintun Nation tribal lands in Yolo County, CA. To frame this paper, we apply Agrawal’s definition of local knowledge as knowledge that is “integrally linked with the lives of people, always produced in dynamic interactions among humans and between humans and nature, and constantly changing.” This definition of local knowledge recognizes the key elements of local knowledge: 1) It is produced by people and among people; 2) It is always produced in relationship with nature; and 3) It is a dynamic process.

More broadly defined, local knowledge involves dynamic processes and complex systems of experiences, practices, and skills developed and sustained by people in their environmental and socioeconomic realties . Further, local knowledge may develop even within one or two generations of place-based experience . In the US, there exists a handful of studies documenting rural local knowledge and rancher local knowledge . Very few studies explicitly examine local knowledge in the context of alternative agricultural or organic systems, referred to as “farmer knowledge” in the literature. This type of knowledge is a subset of local knowledge that enables knowledge holders to farm alternatively in their specifical local contexts. To date, most formal studies on farmer knowledge tend to focus on farmer decision making as it relates to the adoption of new practices . Few studies exist at the intersection of local knowledge, alternative agriculture, and soil management.To consider this gap, we focus this study on a significant epicenter for alternative agriculture in the United States: Yolo County, California, which represents unceded Patwin-speaking Wintun Nation tribal lands. This region in northern California is unique in that it is among the handful of places in the country that emerged as a catalyst and knowledge hub for the organic agriculture movement and where a large concentration of high value, innovative organic production farms continue to thrive today. Due to a unique set of historical and ecological circumstances, the region experienced an influx of organic farmers beginning in the 1970s . During this decade, Yolo County—in combination with Santa Cruz, CA—became a significant node in the organic movement. Its emergence as a significant node was in part due to Yolo County’s proximity to the San Francisco Bay Area and the University of California, Davis—which provided key institutional support—and also partially due to the existence of largely prime agricultural lands combined with a temperate climate ideal for growing year-round.

As a result, Yolo County became one of a few of places where regulations for organic production first evolved and experimentation with organic farming first emerged . Following the farm financial crisis of the 1980s, land prices in the County sharply dropped ; this economic window provided an opportunity for a new generation of farmers to insert a more ecologically-minded approach to farming. Many of these farmers arrived to Yolo County relatively new to farming —often young, educated white urbanites with a desire to farm alternatively to the industrial agribusinesses that had historically dominated the landscape of Yolo County since the early 1900s . When these so-called “back-to-the-land” farmers arrived, many were particularly interested in soil fertility—a conscious effort to avoid “mining the soil” and address ongoing issues with soil degradation in agriculture . While initially these back-to-the-landers lacked historically- and ecologically specific knowledge of the lands they cultivated , over the last three decades or more, it is highly probable that they have individually amassed a wealth of local, place-based knowledge of their specific management contexts and soil landscapes . In this sense, farmer knowledge of soil management presents a particularly salient entry point for further examination in the context of Yolo County specifically. How did these particular farmers address the challenge of soil management in their region? What have they individually and collectively learned about soil management, in theory and in practice? Such questions are particularly important to consider given that—from a pedological and agricultural perspective—soils are heterogenous across landscapes. For example, even at the scale of a single field, roll bench differences in microenvironments, management histories, inherent soil characteristics, and time of year can all dramatically influence how a particular field can be most effectively managed. Addressing this challenge in soil management and understanding the nuances of soil management are fundamental to organic systems—where deep place-based knowledge of soil landscapes is the basis for building and sustaining healthy soils on-farm—and more broadly, resilient agriculture. Yet, farmer knowledge of soil management is still generally under-researched, particularly in the United States and particularly among organic farmers. Though documentation of farmer knowledge of soil management in alternative agriculture exists, most studies focus within the “development” context . Similarly, research on indigenous knowledge of soil is frequently approached from an ethnopedological or traditional ecological knowledge perspective , and lacks focus on production and/or organic agriculture. To date, farmer knowledge of local soil landscapes and related soil management practices remains entirely undocumented in Yolo County. Yet, the unique historical and ecological context makes farmer knowledge of soil health and soil management in this region especially important to document; this knowledge is potentially foundational as organic farmers adapt their farming approaches and management in the face of increasing social, economic, and environmental uncertainties.Though many organic farmers in Yolo County are informed by principles of alternative agriculture when managing their soils, it is less clear how these farmers have translated their ethos into practice and the substance of the soil management practices applied. To address this gap, we examined local farmer ethos and practical knowledge of soil management in this region. Our objectives were to: 1) understand how farmers acquire local knowledge of their soils; 2) document what organic farmers know about their soils; and 3) determine how these farmers translate this local knowledge into specific management practices related to soil health and on farm resilience.This research is informed by a Farmer First approach, which recognizes farmers as experts and crucial partners in researching and innovating solutions for resilient, alternative agriculture . The Farmer First approach recognizes multiple knowledge forms and challenges the standard “information transfer” pipeline model that is often applied in research and extension contexts . We used an open-ended, qualitative approach that relied on in-depth and in-person interviews to study farmer knowledge. Such methods are complementary to surveys that use quantitative methods for capturing a large sample of responses . Because they are more open-ended, qualitative approaches allow for more unanticipated directions ; however, as Scoones and Thompson point out, removing local knowledge from its local context and attempting to fit it into the constrictive framework of Western scientific rationality is likely to lead to significant errors in interpretation, assimilation, and application.

While interviews are not able to capture the quantity of farmer input that surveys do, in-depth interviews allow researchers to access a deeper knowledge base that has inherent value—despite limitations in scalability and/or transferability—as participants respond in their own words, using their own categorization, and perceived associations . Such in-depth interviews are therefore essential to research on farmer knowledge and local knowledge .In-person interviews were conducted in the winter, between December 2019 – February 2020; three interviews were conducted in December 2020. We used a two-tiered interview process, where we scheduled an initial field visit and then returned for an in-depth, semi-structured interview. The purpose of the preliminary field visit was to help establish rapport and increase the amount and depth of knowledge farmers shared during the semi-structured interviews. The initial field visit typically lasted one hour and was completed with all thirteen participants. Farmers were asked to walk through their farm and talk more generally about their fields, their management practices, and their understanding of the term “soil health.” The field interview also provided an opportunity for open dialogue with farmers regarding management practices and local knowledge . Because local knowledge is often tacit, the field component was beneficial to connect knowledge shared to specific fields and specific practices. After the initial field visits, all 13 farmers were contacted to participate in a follow up visit to their farm that consisted of a semi-structured interview followed by a brief survey. These in-depth interviews allowed us to ask the same questions of each farmer so that comparisons between interviews could be made. To develop interview questions for the semistructured interviews , we established initial topics such as the farmer’s background, farm history, general farm management and soil management approaches. We consulted with two organic farmers to develop final interview questions. The final format of the semi-structured interviews was designed to encourage deep knowledge sharing. For example, the interview questions were structured such that questions revisited topics to allow interviewees to expand on and deepen their answer with each subsequent version of the question. Certain questions attempted to understand farmer perspectives from multiple angles and avoided scientific jargon or frameworks whenever possible. Most questions promoted open-ended responses to elicit the full range of possible responses from farmers. In the interviews, we posed questions about the history and background of the participant and their farm operation, how participants learned to farm, and to describe this process of learning in their own words, as well as details about their general management approaches. Farmers were encouraged to share specific stories and observations that related to specific questions. Next, we asked a detailed set of questions about their soil management practices, including specific questions about soil quality and soil fertility on their farm. In this context, soil quality focused on ecological aspects of their soil’s ability to perform key functions for their farm operation ; while soil fertility centered on agronomic aspects of their soils’ ability to sustain nutrients necessary for production agriculture .

Under team pay, biased upstream workers are unable to increase the relative pay of favored downstream workers by distorting relative supply. As a result, horizontal misallocation of flowers was eliminated. Total output in teams in which the two processors were of different ethnic groups therefore increased, the introduction of team pay returning the difference in output between such teams and homogeneous teams to pre-conflict levels. Overall output also increased, even though the results indicate that team pay led processors to freeride on each others’ effort. This paper’s results indicate that, if taste for discrimination is high enough, firms are forced to adopt “second best” policies to limit the distortions caused by such discrimination. But entirely removing workers’ incentives for discrimination is difficult. At the plant, team pay had little effect on the degree of discrimination in teams that were ethnically differentiated vertically rather than horizontally, as also predicted by the model. The obvious “solution” to discrimination – segregating workers – may be undesirable for reasons unrelated to productivity in the short term. The extent and multiplier effects of taste-based misallocation also depend on a number of other factors, such as pay systems, the structure of production, and the “geographical” distribution of ethnic groups in the productive system, however. More speculatively, cannabis grow equipment it is possible that such factors respond endogenously to ethnic diversity. Social segregation is commonly observed in diverse societies but likely becomes harder to achieve as urbanization brings larger groups of workers together. The linkages and specialization required in industrialized production are rarely observed in the most ethnically diverse countries.

My findings also suggest that the economic costs of ethnic diversity vary with the political environment. Relatively brief episodes of ethnic conflict can have a long-lasting impact on economically distortionary attitudes: I find no decay in discrimination in the nine months after conflict ended. Multiple equilibria may thus exist if the occurence of conflict itself depends on attitudes towards non-coethnics, some diverse societies being characterized by tolerance and little conflict and others by ethnic biases and frequent conflict.The quotes above illustrate a prevailing view among policymakers which sees the creation of job opportunities for parents – especially mothers – as a quintessential tool for improving the lives of children in poor countries. The view appears to be based in large part on extrapolation of findings from studies of the effects in the household of increases in unearned income. Relying on such extrapolation may be adequate if the dominant household models – in which children typically appear only as an expenditure category for the decision-making parents 1 – provide an accurate picture of a poor country household. If instead there is substitution between parents’ and childrens’ time use, then employment may be a fundamentally different “treatment” than pure income transfers due to its implications for the employed parent’s time use. In that case the lack of causal evidence on the consequences for children of parent’s employment is a problematic gap in the literature on poor countries. Taking advantage of a field experiment that randomized long-term job offers this paper presents direct evidence on the impact of a parent’s employment on children’s lives. Five Ethiopian flower farms agreed to allocate fall 2008 job offers through a lottery system.

The experiment was “natural” in the sense that parents sought employment in the exact same way they would have done in the absence of the research team. Because households thus themselves determined if the mother or the father applied, I analyze the two sub-samples separately. The farms were willing to randomize job offers because open positions attracted large numbers of mostly inexperienced applicants and screening was difficult. Before the lottery took place, enumerators surveyed acceptable applicants. Winners and losers were re-surveyed five to seven months after employment commenced. The randomization was effectively stratified on gender. The main results are as follows. As daughters take over house-work left undone when a mother gets employed, their school-time falls by 24 percent per week. Daughters’ time use is unaffected by father’s employment. An increase in sons’ school time of about ten percent when a mother or a father gets employed appears to be due to higher household income; sons’ house-work time is unaffected by parents’ employment. After documenting the impact of parents’ employment on childrens’ time use, I present a simple collective framework in which each parent attaches weight to daughters’ well-being and daughters derive utility from going to school, but only females can do house-work . The framework highlights the variables upon which heterogeneity in the response to mother’s employment is likely to depend if the primary underlying force is time use substitution between mothers and daughters.

Testing the framework’s predictions, I find that, the higher the proportion of daughters – a variable that is shown to be exogenous in the sample studied – the less negative the impact of mother’s employment on a given daughter’s school-time, the greater the weight attached to daughters’ well-being, the less negative the impact of mother’s employment on a daughter’s school-time, and the greater the initial bargaining power of the mother, the greater the reduction in daughters’ school-time when mothers get employed. Daughters themselves appear to have little influence over the change in their time use when mothers get employed. Interestingly, selection into mother’s versus father’s employment appears to depend on the same covariates, providing further evidence of the importance of female house-work substitution. These results have important implications for the design of employment programs and for how selection into parent’s employment and its effects in the household should be modeled. If full-time school enrollment is not universal, explicitly accounting for children’s time use is important. In situations where the house-work necessary to run a household is time consuming, the substitutability between parents’ and children’s effort introduces a potential trade-off between parents’ and children’s preferences. If house-work is effectively gender specific, then the conventional wisdom – that economically empowering mothers is of greater benefit to daughters than empowering fathers – is not necessarily the full story when it comes to parent’s employment, even if mothers weigh daughters’ well-being more than fathers do. The reason is that mothers may face a trade-off between own and daughters’ time use that fathers do not. If female participation in the market economy over time influences the norms governing the division of labor in the household, then the longer-term effects of mother’s employment may differ from those observed here, but such norms are likely slow to change. This paper builds on and extends the overlapping literatures on adult employment, vertical grow rack child labor and schooling, and intra-household decision-making in poor countries. Causal evidence on the effects in the household of long-term parental employment in poor countries is to my knowledge largely absent, credible exogenous variation in employment rarely being available. Indirect inference – for example on the basis of findings from studies of unearned income – has been attempted, but there are good reasons to study parent’s employment directly. Beyond the implied time use reconfiguration, employment may for example affect the two parents’ relative bargaining power differently than government transfers or income from other sources do. This paper presents the first experimental evidence on the effects in the household of a parent’s long-term employment. Children’s time use is one of the primary determinants of human capital accumulation and child well-being. The degree of substitutability between parents’ and children’s time use is therefore important. Several existing studies find correlations between a mother’s employment status and children’s time use in poor countries . Doran convincingly shows that adults in Mexico work more when children work less due to an exogenous increase in time spent in school.

But his focus is on paid child labor; though understudied in the literature due to a lack of data child house-work is much more common than paid work in most of the developing world, and the effect of parents’ time use on children’s time use is typically of greater relevance for policy than the converse. Gender specificity of house-work in combination with the typically greater time requirements of “female” responsibilities may be a particularly important though often overlooked form of son favoritism, especially because child labor and schooling are negatively related . I take advantage of an exogenous increase in mother’s and father’s work hours to provide causal evidence on time use substitution between mothers, fathers, daughters and sons. Bar and Basu argue that an inverted-U relationship can arise because of missing labor markets for children: the results in this paper suggest that missing labor markets for adults can also lead to a range in which child labor may appear to be increasing in parents’ income. As formal employment opportunities arise for mothers, daughters may be forced to take over house-work. The preferences of children and parents are not perfectly aligned, even if parents are partially altruistic. An important question is how much influence children have over their own lives: the review in Edmonds argues that our almost complete lack of knowledge about parent-child agency and who makes child time use decisions is the most pressing issue in the literature on child labor. This paper’s results indicate that the reconfiguration of a daughter’s time that occurs when a mother gets employed in rural Ethiopia is decided by parents, primarily mothers, while daughters themselves have little influence over the change in their time use. The paper is organized as follows. In section 2, I present the setting and the experiment. The reduced form time use estimates are in section 3. In section 4, I present a simple theoretical framework of household work and schooling decisions that illustrates the forces that underlie the results in section 3, and derive auxiliary predictions. The predictions are tested in sections 5. Section 6 provides further evidence on how time use decisions are made and section 7 analyzes selection into employment.Growth in the commercial floriculture sector in Ethiopia has been explosive in recent years, fueled in part by government incentives and in part by the abundant availability of cheap land and labor in rural areas. In 2008, 81 flower farms employed around 50,000 unskilled workers. Most flower farm workers work in greenhouses, growing and harvesting flowers, or in “pack houses”, packaging flowers and preparing them for shipping. Over 70 percent of flower farm workers are women . Hiring on Ethiopian flower farms typically takes place in October and November, before the main growing and harvesting season. The supervisors on five flower farms agreed to randomize job offers during fall 2008 because of an unusual situation in the labor market for flower farm workers at the time. Because comparable jobs were seldom available in the areas suitable for flower growing, applicants almost always outnumbered the positions to be filled by large margins. Ethiopian flower farms – still getting to grips with cost components significantly larger than labor and with little ability to predict the productivity of the mostly uneducated, illiterate and inexperienced applicants – did not prioritize optimization of the unskilled workforce . Because supervisors were already allocating job offers relatively arbitrarily when approached by the research team, explicit randomization was a modest procedural change. When Ethiopian flower farms hire, word is typically spread in nearby villages. Job-seekers arrive at the farm on announced “hiring days”. At the participating farms, supervisors first excluded any unacceptable applicants. A team of enumerators then carried out the baseline survey with the remaining applicants. Finally, the names of the number of female and male workers to be hired were drawn randomly from a hat. The full sample thus consists of 527 households in which at least one spouse applied to a flower farm job and was deemed acceptable for hiring. There are 346 women in the sample and 188 men: in almost all cases one of two spouses applied. We attempted to re-interview everyone in the treatment and control groups 5 – 7 months after employment commenced. Because few farms were hiring workers in the season that followed the randomization, only 6 re-interviewed individuals in the control group had managed to obtain employment. Careful tracking procedures led to a re-interview rate of 88 percent and no statistically significant differential attrition. Almost all the job-seekers are parents: the focus here is on the effects of a parent’s employment for children in the household.

The output gap between vertically mixed and homogeneous teams points to vertical discrimination: it appears that upstream workers are willing to accept lower own pay in order to lower the pay of non-coethnic co-workers. About 86 percent of the output gap between horizontally mixed and homogeneous teams is due to vertical misallocation and 14 percent due to horizontal misallocation. Because Kikuyu and Luo workers are of similar productivity on average, horizontal misallocation has little impact on total output. But the distribution of output across downstream workers is affected: in horizontally mixed teams, processors of the supplier’s ethnic group earn 27 percent more than processors of the other ethnic group. In the second main result of the paper, I find that the output gap between homogeneous and diverse teams nearly doubled when conflict between the Kikuyu and Luo political blocs began in early 2008. The reason appears to an increase in workers’ taste for ethnicdiscrimination. I estimate a decrease of approximately 35 percent in the utility-weight of non-coethnic co-workers when conflict began, through a reduced form approach. As also predicted by the model, there was a small but significant increase in the output of processors of the supplier’s ethnic group in horizontally mixed teams in early 2008. A back-of-the envelope calculation suggest that the decrease in productivity in mixed teams may have cost the farm half a million dollars in annual profit, had it not responded. It is clear from these results that the economic costs of ethnic diversity vary with the political environment. In the third main result of the paper, I find that the introduction of team pay for processors six weeks into the conflict period led to an increase in output in horizontally mixed team, cannabis grow racks returning the difference in output between homogeneous and horizontally mixed teams to pre-conflict levels.

The increase was likely due to a reduction in horizontal misallocation: a 32 percent output gap between coethnic and non-coethnic processors in horizontally mixed teams was eliminated when team pay was introduced, as predicted by the model. As a result, overall output increased, even though there was a modest decrease in output in homogeneous and vertically mixed teams. These results indicate that that firms are forced to adopt “second best” policies to limit the distortionary effects of ethnic diversity in the workforce when taste for discrimination is high enough. Figure 2 illustrates the evolution of output in teams of different ethnicity configurations during each of the three sample periods observed. This paper’s findings have important implications for theory and policy. Distortionary, taste-based discrimination in production appears to be the primary explanation behind my results. Theories of non-taste-based ethnic diversity effects are unlikely to simultaneously explain a differential fall in mixed teams’ output during conflict and equalization of downstream workers’ output under team pay. Distinguishing between different channels through which ethnic diversity may affect productivity is important. Higher output in homogeneous teams may be efficient if due to technological differences across diverse and homogeneous teams. But discriminatory preferences should lead to distortionary misallocation of resources in most joint production situations in which individuals influence the output and income of others. Interacting economically with individuals of other ethnic backgrounds is hard to avoid when urbanization and economic modernization brings larger groups of workers together, and large multiplier effects are associated with misallocation of intermediate goods . The contribution of taste-based discrimination in production to the lower incomes observed in diverse countries may thus be sizable.

The findings of this paper also suggest that relatively brief episodes of conflict can have a long-lasting impact on distortionary attitudes towards individuals of other groups. I find no reversion in ethnic discrimination in the nine months after conflict ended. It appears that the economic costs of ethnic diversity vary with the political environment because social preferences are affected by conflict, forcing firms to adjust their policies in conflictual environments. Entirely removing incentives to discriminate through contractual design is difficult, however. At the plant, biased upstream workers continued to derive less benefit from flowers supplied to pairs of processors that included non-coethnics under team pay. As a consequence, it appears, output in vertically mixed teams was 15 percent lower than in homogeneous teams after team pay was introduced.This paper contributes to and ties together several areas of research. Its results are to my knowledge the first to carefully identify and explain a negative effect of ethnic diversity on productivity in the private sector, perhaps because well-measured, micro-level output data from poor countries is rarely available. By showing that a taste for ethnic discrimination can lower output by leading to misallocation of intermediate goods, I also contribute to the literature on workplace favoritism initiated by Becker and the recent literature on social preferences at work . The difference between the findings of Bandiera, Barankay, and Rasul in the U.K. and my findings in Kenya are particularly interesting. The authors find that “upstream” supervisors at a fruit farm in the U.K., in their allocation of own effort and in their assignment of “downstream” workers to rows with different amounts of fruit, discriminate against workers to whom they are not socially connected only when doing so is costless to the supervisor. In contrast, this paper documents an upstream willingness to pay to lower the incomes of non-favored downstream workers, to my knowledge the first paper to do so in data on consequential choices made every day.

Ethnic antagonism may be of greater importance to workers in Kenya than social connections are to workers in the U.K. Burgess, Jedwab, Miguel, Morjaria, and i Miquel and La Ferrara show that Africans belonging to a different ethnic group than “upstream” decisionmakers have less access to economic resources in other contexts, suggesting that distortionary discrimination may be a common phenomenon in Africa. If individuals have discriminatory preferences, output is likely to be lower in diverse production units in most production situations in which co-workers affect each other’s income. I begin to address how the productivity effects of ethnic diversity are likely to vary across time and space by studying how workplace discrimination responds to increased ethnic conflict in society, and how firms respond to distortionary discrimination. I follow an innovative paper by Krueger and Mas in exploring worker behavior during conflict, but my focus is on a poor country characterized by frequent, ethnic tensions. I follow Ksoll, Macchiavello, and Morjaria in studying Kenyan flower farms during the political crisis of 2008, but focus on the effect of conflict on distortionary attitudes towards non-coethnics. As such, this paper also adds to an emerging literature investigating how social preferences are shaped . How firms respond to distortions due to ethnic diversity and how to optimally organize production in the presence of discriminatory attitudes is an exciting venue for future research. Prendergast and Topel provides a theoretical analysis of the influence of favoritism on optimal compensation and extent of authority for managers. In studying the motivation behind the introduction of team pay at the plant, this paper is particularly re-lated to La Ferrara who shows that ethnically diverse cooperatives are more likely to adopt group-pay. I also investigate why the plant chose not to segregate Kikuyu and Luo workers. Finally, there are interesting connections between this paper’s results on within-firm misallocation and the literature in macroeconomics on across-firm misallocation of capital and intermediate goods in poor countries . First, some of the distortionary policies studied by macroeconomists may exist in part as a means for politicians to skew the distribution of resources towards their own ethnic groups and thus ultimately arise from biased preferences upstream. Second, firms whose output suffers from internal misallocation due to ethnic diversity distortions may survive due to macro-level misallocation of capital. Jones points out that to understand development we need to understand both why misallocation occurs and the intermediate goods and linkages through which its effects are amplified. The paper is organized as follows. In section 2, I describe the setting and the organization of production at the plant, cannabis drying racks outline the data used, and test for systematic assignment to teams. The model of upstream discrimination is presented in section 3, and its predictions for the three sample periods observed tested in section 4. Section 5 explores the extent to which other ethnic diversity mechanisms may explain my results. Section 6 investigates the response of distortionary attitudes towards non-coethnics to conflict in more depth, and section 7 the plant’s response to discrimination. Ethnic divisions have influenced Kenyan society and politics since independence and contributed to periodical violence. The country’s biggest tribe, the Kikuyu, was favored by Kenya’s British colonizers, a fact that has had long-lasting influence on tribal relations. The Kikuyu has also been the most economically successful and politically influential tribe during most periods of the post-independence era.

Although the relationships between different tribes have varied over time, the other major tribes have typically defined themselves politically in opposition to the Kikuyu. In recent years the opposition has been led by the second biggest – and persistently politically active – tribe, the Luo. Most Kenyan tribes have aligned themselves with one of the two associated camps. I therefore categorize workers according to the tribal coalition to which their tribe is seen to belong – the “Kikuyu” and the “Luo” .3 An interesting case study in the context of ethnic divisions is Kenya’s vibrant floriculture sector, which brings together large numbers of workers of different backgrounds. A rapid expansion of the sector began in the 1980s; Kenya is now the third-largest exporter of flowers in the world and supplies approximately 31 percent of flowers imported into Europe . Around 50,000 Kenyans are employed in floriculture, and 500,000 in associated industries. Flower farms are part of the fastest growing sub-sector of the Kenyan economy . Production takes place on large farms that typically sell their product through auctions in The Netherlands. Most flower farm employees work either in greenhouses or packing plants . On some farms, including the one I focus on, workers reside on farm property in gated communities. Such farms essentially constitute a miniature society – complete with schools, health clinics and other amenities – in which groups of individuals from different ethnic backgrounds live and work together. Flower farm jobs are considered relatively desirable.The sample farm primarily produces roses. Plant workers are roughly equally divided across three halls. Packing takes place in three-person teams, as depicted in figure 1a. One upstream “supplier” supplies two downstream “processors” working on separate tables. The supplier brings flowers arriving from the greenhouses to her worktable and throws out poor quality flowers. She then sorts flowers of different lengths/types into piles that are placed on the worktable of one of the processors. The processors remove leaves, cut flowers down to the right size, and finally create bunches that are labeled with the worker’s ID number. Nearly all workers are observed in both positions . My primary data source is records of daily processor output from 2007 and 2008. There are 924 packing plant workers in total. The quantities produced were recorded on paper by the farm for remuneration purposes and subsequently converted to electronic format by the research team. A survey provides additional information about workers’ experience, ethnicity, birthplace and other background information. Summary statistics are in table 1. 59 percent of workers are female and 46 percent Kikuyu. The average worker is 35 years old and has five years of tenure at the factory. These figures are similar for Kikuyu and Luo workers. On average, workers are observed working for 22 days followed by two leave days. When a worker takes leave, another worker returning from leave joins the two remaining workers. Teams are observed for 10 consecutive days on average, but because there is substantial variation in the length of individual work spells, the same is true for team spells. The length of work spells is statistically unrelated to characteristics of workers and teams. 28280 different teams are observed during the sample period. Individual workers are observed on 90 different teams on average. Suppliers are paid a piece rate w per rose finalized by the processors supplied throughout the sample period. In 2007, the first year of the sample period, each rose finalized by a processor earned her a piece rate 2w. In February 2008 the factory began paying the two processors based on their combined output, which led to a change in suppliers’ incentives that I exploit in section 4.

While all organic matter is originally derived from plant tissues, animals , and animal manures are a secondary and valuable source of organic matter. The decomposed remains of microorganisms can contribute up to 20% of the total organic matter content of biologically active soils. Green manures, crop residues and weeds, as well as intentional grass/legume cover crops incorporated into soils on a regular basis serve as fundamental building blocks of organic matter and plant nutrition . Organic matter is a major force in the formation and stabilization of granular or crumb structure of soil aggregates . When organic matter is added to a soil via cultivation, the plant residues cement or bind soil particles together as a result of gels, gums, and glues that are byproducts of decomposition. Mycelial strands or webs of fungi also bind soil particles together.The Parisian market gardens for which the practice was originally named were small plots of land that were deeply and attentively cultivated by French gardeners, or “maraîchers.” The “marais” system, as it is known in French, was formed in part as a response to the increasing urbanization of Paris, the attendant increase in the cost of urban land, and the ready availability of horse manure as a fertility source. English master gardener Alan Chadwick popularized both the term and the gardening method in the U.S. when he introduced them at UC Santa Cruz’s Student Garden Project in 1967, and they have served as the theoretical foundation supporting the cultivation methods used at the UCSC Farm & Garden ever since. But as Chadwick was quick to point out, cannabis indoor growing other societies were using similar practices far earlier than the Parisian market gardeners. He acknowledged the influence of early Chinese, Greek, and Roman agriculture specifically, on the development of the French-intensive method.

The concept of small farms dedicated to intensive cultivation of the land, improved soil fertility, water conservation, and closed-loop systems was a feature common to many early civilizations and, in fact, characterizes the majority of agriculture today in developing countries where these techniques have been passed down to successive generations. Of the world’s 525 million farms, approximately 85% are fewer than 4 acres in size, tended to mostly by poor farmers in China, India, and Africa,1 where methods often reflect the same philosophies of stewardship and cultivation that inform the French intensive method we use today. And in much of the developing world, locally adapted traditions continue to shape the way agriculture is practiced. This supplement examines some of the methods used by farmers around the world, past and present, reflecting the principles on which the French-intensive method is based.In Japan, compost production has been tied to small-scale farming for centuries. Farmers harvested herbaceous growth from nearby hillsides as a source of compost material. Compost houses were built and filled with this herbage, manure, and soil daily until piles reached five feet high. Water was constantly added to ensure saturation. Once the designated height was reached farmers let the piles sit five weeks in summer and seven weeks in winter before turning them to the other side of the house. The compost was then applied to dryland cereal crops in spring. A study conducted in the early 20th century found that nitrogen, phosphorus, and potassium were replenished by this composting system nearly at the level lost through harvest.2Vegetable growers on California’s Central Coast rely on draw bar-pulled offset wheel discs, often with a ring roller run behind, as a primary means of tillage. The disc/ring roller combination quickly and efficiently mixes crop residue with the soil and effectively knocks down and incorporates weeds that have emerged in non-cropped open fields.

One advantage of discing is the speed at which ground can be covered. For example, an 8-foot wide offset disc running at the optimum 4 miles per hour can easily cover an acre in 20 minutes. However, for small farms, discing with a drawbar-pulled offset wheel disc can be challenging, as it is difficult to disc small plots of ground effectively. Discs don’t corner well and require significant space to turn. They move soil around in a field , necessitating either the use of a land plane or wheel scraper to cut high spots and fill low spots left by the disc. In small plot situations where cover crop residue is not excessive, flail-mowed cover crop residue can be effectively incorporated with a three-point rototiller behind a small tractor. Multiple passes with a three point off-set or tandem disc will also work to incorporate moderate amounts of cover crop residue, but these implements tend to be light and will require more passes compared to the heavier offset wheel discs. Notched blades on a three-point disc will greatly enhance its ability to work deeper and cut through heavy residue. One of the most important considerations when pulling a draw bar-hitched offset wheel disc is the need to disc in a “pattern” . Drawbar-pulled offset discs can only be turned to the left. Turning the disc to the right when it is soil engaged will result in serious damage to the disc frame or the discs themselves. This is very important to understand. While running through the field the front gang of discs throws soil to the right and the second gang throws the soil back to the left . The back gang leaves what is referred to as a “dead furrow.” The second pass of the disc will cover this dead furrow and leave a new one on the right side of the disc. Follow the last pass on the right side to minimize the number of dead furrows in the field. As an example, when discing a one-acre plot, it is important to disc in a pattern that allows you to make left turns only and follow your last pass on the right side.

If done correctly the field would have a dead furrow down the middle and along each of the two sides. It is always advantageous to cross disc either on a diagonal or perpendicular to your last pass. It is not uncommon to disc a field multiple times to get the desired mixing and tilth. Optimum soil moisture for discing is just moist enough that you don’t raise a huge dust cloud. On soils prone to compaction, discing when soil moisture is too high can be extremely detrimental to soil tilth—the soil should never be so wet that the soil sticks to the discs. Note that there are many different configurations for offset discs but they all have two disc gangs that are diagonally opposed. Many of the small offset discs are set up on a three-point hitch, but their light weight limits their usefulness.On most soil types, deeper tillage is best done in the fall at the end of the cropping season when the soil is relatively dry prior to the onset of fall rains so that the compacted layers “fracture” effectively. Deep ripping is critical on some soils to improve water infiltration and break up compacted layers formed from prior shallower tillage. Ripping is usually followed by discing to break up clods brought to the surface from the ripping. A ring roller running behind the disc should be used to break clods and push small clods into the soil so that they will more easily take in and hold moisture from irrigation or rainfall. Ripping typically involves two passes, with the second pass done diagonally to the first pass. This second pass allows the ripper to penetrate deeper, allows for a more complete “shatter” of the sub soil, cannabis grow kit and is much easier on the tractor operator than a perpendicular pass since the tractor will rock from side to side rather that slam up and down as it goes over the soil indents left from the first pass. Ripping is a slow, energy intensive and tedious task and requires excellent traction.Once you know the center-to-center spacing of your rear tires then all bed forming implements must be set to this spacing. All other “in-row” implements used following the initial bed shaping must also match this spacing . Most small farms are thus either on a 48” or 60” bed spacing. These spacings dictate either a single 48”- or 60”-wide bed, or two 24”- or 30”-wide beds . Another critical consideration when setting up bed spacings is tire width. If possible, tires should not be any wider than 12 inches for vegetable farming if you plan on getting into the field to perform in row operations after beds have been formed. Wider rear tires will take up critical production space. Also note that most lower-horsepower 4-wheel drive tractors have poor clearance for “in-row” work. The basic limitation with poor clearance is that final cultivations on taller crops must be done much earlier, since crop height will dictate timing. This limitation could impact weed management options.The greenhouses, growing containers, and growing media needed to grow healthy transplants are not only costly, adding to the already high initial capital investment required to begin a farming operation, but also use large quantities of non-renewable resources. As input costs and impacts continue to rise worldwide, farmers need to find alternative sources of energy and inputs to support their plant’s growing needs. Although many of the costs related to farming that make it financially risky are fixed or inelastic, meaning they are difficult to change , there are some that can be minimized. Without easy access to government-subsidized credit, it is essential that organic farmers minimize costs wherever possible to make their operation economically viable. Likewise, in urban areas where fixed costs may be even higher and access to raw materials and farmer know-how is limited, low-cost alternatives to traditional greenhouse propagation that include doit-yourself options can mean the difference between success and struggle, and often provide more environmentally sustainable and socially just solutions.

Here are a few options for greenhouse propagation that reduce the costs, and in turn the barriers, to starting a farm or market garden.Seed saving not only reduces the cost of propagation, it provides adaptive on-farm benefits and preserves genetic diversity. Saving seed also embodies the philosophy of sustainability that guides agroecological farming. Seed costs, while not the largest operating expense on a farm, can be significant, especially when the cost of cover crop seed is factored in. Additionally, there is a price differential between conventional and organic seed—and organic seed for a number of varieties isn’t always available, even from commercial organic seed companies. Seed saving requires some botany and ecology knowledge to preserve varietal integrity. It also requires additional in-ground time commitment for most crops as well as the labor to harvest, process/ clean saved seed. As discussed in Supplement 1 in Unit 1.4, by saving seed you can select for plants adapted to local climate and soil features, and maintains genetic diversity in an era when genetic engineering and hybrid technology threaten crop diversity worldwide. By saving seed, farmers can lower overall operating costs as well as supply the farm with its own organic, locally adapted seed. Seed saving can be a central part of developing a closed-loop system, minimizing external dependence and enhancing the process of community seed sovereignty. These benefits and challenges should be carefully weighed against the cost and convenience of buying seed from existing sources.The greenhouse is by far the largest propagation related investment for a farmer. Most commercial greenhouses are expensive to buy or have built, and often maximize only the sun’s light energy while relying on fossil fuels in the form of electrically powered vents, fans, lights, heating tables, and thermostats to moderate heat. Passive solar greenhouses, on the other hand, are designed to maximize use of the sun’s light and heat energy with little to no reliance on other sources of energy to control temperature or air circulation. Passive solar heating relies on maximizing sunlight during the day and then storing the trapped heat overnight using a thermal mass, usually large drums of water, blocks of stone, or gravel beds, within the greenhouse. Besides their use of “free” energy from the sun, passive solar greenhouses are relatively inexpensive to build when compared to commercial greenhouses and can be built by someone without extensive construction experience. Building a greenhouse independently not only reduces one of the few variable capital costs in starting a farm, but also allows the farmer to customize the design for her/his specific location, climate, and production goals.

Direct observation of feed bunk behaviors would ultimately be needed to positively determine the underlying mechanism for this behavioral pattern among this subset of animals.In order to account for this informational redundancy, age in days at start of trial and the 95th quantile of daily milk yield were here scaled to uniform variance and jointly encoded using the ecodePlot utility . Visualizations of this encoding can be found in Supplemental Materials. Bivariate tree tests, however, returned no significant associations between this age-yield encoding and any of the tensor plot encodings of daily time budgets. This result differed slightly from analyses of overall time budgets, wherein encodings that accounted for heterogeneity in sensor error and plasticity in daily response also returned no significant association, but encodings that accounted only for sensor noise found that heifers were significantly under represented among the most moderate time time budgets . Collectively, these results suggest that plasticity in daily time budgets might be obscuring links to age and productivity, but that this relationship cannot be brought into resolutions by accounting for systematic responses to environmental factors alone. This could occur if there were heterogeneity in the lag time between an environmental stressor and the behavioral response, or if there are more complex interactions occurring between transient environmental variables and more persistent internal biological states. If this in fact the case, then this result may suggest that an encoding strategy capturing only the variability in the response, trim tray screens and not systematic fluctuations, may in fact be a more effective means of pinning down this bivariate association.

Visualizations of these associations are provided in Supplemental Materials. As with overall time budgets, these resultsrevealed that cows with no health complications were over represented among the most moderate time budgets while sick cows were over represented among cows with low time spent eating. Overall, however, these associations were stronger with encodings of overall time budgets than with daily time budgets. Bivariate associations were also evaluated against health records disaggregated into five broad categories: healthy, mammary infection and injury, hoof and leg lameness, infections of the reproductive tract, and digestive and metabolic diseases. Significant associations were recovered for all three dissimilarity estimators used to encode daily time budgets. Visualizations of these associations can again be found in Supplemental Materials. Again, these results mirror the bivariate patterns recovered for overall time budgets, with no clear evidence that temporal dynamics notably modified these associations. Figure 4 displays the strongest association recovered for the ensemble-weighted encoding, wherein the optimal metaparameter combination for the tensor mechanics encoding employed no re-weighting along the temporal axis. As with overall time budgets, mammary infections were over-represented among cows with consistently extremely high time spent eating, while digestive and metabolic diseases were over-represented among low eating time budgets . Infections of the reproductive tract were also again over-represented among the most moderate time budgets, but in this encoding with daily time budgets, it is more visually apparent that these same time budgets are also the most variable over time. As these observations occurred fairly early in the lactation period, it seems unlikely that temporary perturbations in behavioral patterns attributable to estrus behaviors could fully explain so much variability in eating times .Finally, significant bivariate associations were recovered between patterns in milking order and encodings of daily time budget using all three dissimilarity estimators.

Visualizations of these bivariate patterns are provided for all encodings in Supplemental Materials. As a whole, the bivariate associations recovered against encodings created using overall time budget were stronger than for encodings using daily time budgets . Results for the unweighted Euclidean distance and the KLD distance largely mirrored patterns recovered using overall time budgets, with extremely low eating times being under represented among cows entering at the end of the queue and over represented among animals just ahead of these caboose animals. Visualizations of the bivariate association recovered against the ensemble-weighted Euclidean distance encoding are provided in Figure 5. Here the pattern between time budgets and cows that enter nearer the rear of the milking queue are brought into slightly higher resolution. This visualization shows that it is cows that consistently enter at the end of the queue, but not the last handful of cows, that are over represented among time budgets with moderate-to-high time spent eating. Time budgets with consistently low time spent eating, on the other hand, are significantly over represented among cows that routinely jumped between the middle and rear of the milking queue. Collectively, these results confirm that a later milking position does not preclude a cow from investing a significant amount of time at the feed bunk, though it should be noted that these results may not extrapolate to larger milking groups with longer waiting times in the milk parlor . As systematic temporal patterns in daily time budgets do not, however, appear to play a significant role in further distinguishing these bivariate patterns, future work might consider if these asynchronous perturbations in queuing patterns and time budgets might be more directly linked within individual, and consider latent biological causes of these isolated behavioral responses.Drawing complete and holistic quantitative inferences is always challenging when working with multivariate data streams. When a temporal component is also present, information compression via data aggregation can simplify the analytical pipeline, but a considerable amount of behavioral complexity can be lost.

In analyses of Cow Manager rear tag accelerometer records available in the Organilac dataset, tensor mechanics analyses revealed that cows in this experimental herd were remarkably consistent in their overall time budgets. Never the less, these algorithms were still able to capture subtle shifts in the tradeoffs between the eating and nonactive axes and the eating and highly active axes as cows progressed in their lactations, even though such temporal patterns were only found amongst a subset of the herd. While the resulting encodings did not differ substantially from overall time budgets, these analyses helped to better visualize differences between animals in the relative plasticity of their behavioral responses. While tensor mechanics algorithms are designed to extract systematic changes over time by leveraging collective responses of animals to a shared environment, these results have served to highlight the need to look beyond the mean value and more fully explore the richness of PLF datasets, which can provide repeated measurements on individuals at a scale never before possible .Future work should consider how to further leverage these approaches to not only explore patters in such plastic responses that can be extracted by leveraging group level information, but also variability attributable only to the individual, and subsequently any complex interactions that may occur between these systematic and fleeting perturbations in behavioral responses. Rice is the most widely consumed staple food for a large part of the human population, especially in Asia, trimming tray with screen providing more than one fifth of the calories consumed by humans worldwide . In many Asian countries, rice accounts for more than 70% of human caloric intake. China is the world’s largest rice producer, accounting for 30% of the total world production, followed by India , Indonesia , and Bangladesh . The traditional method of rice cultivation in the world used to be transplanting rice , which ensured a steady yield during the long history of mankind . However, in all climatic zones, human labor represents more than 50% of the cost of TPR farming , followed by the cost of other inputs such as water and fertilizers . With the development of rice cultivation science and the requirements of different climatic zones, many other methods of rice cultivation gradually emerged, such as dry or wet direct rice seeding . Direct-seeded rice , which is cultivated by directly broadcasting seeds onto the topsoil of paddy fields without needing to raise and transplant seedlings, provides an opportunity to save both labor and time .

This farming method also enables earlier crop establishment, providing an opportunity to make better use of early season rainfall, while increasing crop intensification in some rice-based systems . Moreover, the development of early-maturing varieties and improved nutrient management techniques, along with increased availability of chemical weed control methods, have encouraged many farmers in Asia to switch from transplanted to direct-seeded rice culture . During the last two decades, the change in the method of crop establishment from manual transplanting of seedlings to directseeding has occurred in many Asian countries in response to rising production costs, especially those of labor and water . In the Taihu Lake Basin of east China, many farmers have accepted the cultivation of DSR, although the average yield of DSR is not as stable and is still slightly lower than that of TPR . The area with DSR cultivation has rapidly increased and has already exceeded 50% of the total farmland in many TLB regions . In accordance with this development trend, the direct-seeding approach will likely continue to remain popular in the TLB. Water flow in paddy fields with cultivated rice involves the interaction of very complex processes, and their observation and evaluation under field conditions is relatively difficult, costly, and time consuming. Therefore, a large number of scientists increasingly use computer models to study the complex processes in the soil and to provide management and planning guidance. Hydrus-1D and Hydrus are numerical models that have often been used by many researchers to simulate water flow in agricultural fields with different crops and various irrigation schemes , including TPR fields . However, the Hydrus-1D model has not yet been used for simulating water flow in DSR fields. Compared to traditional TPR, DSR requires different water management, which provides rice with a different growth environment, particularly during its seedling stage . During the first two weeks after seeding, rather than being flooded as with TPR, the top soil only needs to keep sufficient moisture to allow for seed germination . As a result, the root mass of DSR is distributed shallower than that of traditional TPR, which consequently produces different vertical profiles of the water content. Furthermore, compared to TPR, DSR prefers an alternative drying and wetting soil environment during the middle late season when multiple smaller irrigations can benefit both the plant growth and deeper root growth . This water management produces distinctly different characteristics of the water flow regime and water losses from DSR fields compared to TPR fields. In this study, field observations in a DSR field in the TLB during two consecutive rice-growing seasons are evaluated using Hydrus-1D, and the main characteristics of the water flow regime and water losses are discussed.The agricultural land in the Taihu Lake Basin is used for very intensive production of the rice crop. The basin area, which is located south of the Yangtze River, is approximately 36,900 km2, with rice fields accounting for about 34.8% of this area. Rotations of rice with either wheat or rape are the most popular cultivating modes in this region. The basin has a subtropical monsoon climate with average annual rainfall of 1181 mm, 60% of which occurs from May through September. The annual PAN evaporation from the water surface is approximately 822 mm, and the average annual air temperature is 15–17 ◦C. The study site is in the Dangyang region , upstream of Taihu Lake . The dominant soil type in this region is classified as a hydromorphic paddy soil, and the parent material is a lacustrine deposit. The physical and chemical properties of the soil at the site are listed in Table 1.In our experiments, the variety of rice used for the DSR cultivation was Wuxiangjing 14 , a type of Japonica rice thatis predominantly cultivated in the Dangyang region. The observations were carried out in the same field during two growth seasons. After the mechanical field preparation, the seeds were evenly broadcasted by hand on the soil surface at 75 kg/ha, without prior soaking, on June 8th in 2008 and June 11th in 2009. After seeding, the fields were irrigated until the surface soil was saturated . The harvest dates were on November 1st in 2008 and November 5th in 2009. The total growing periods during these two years were thus 147 and 149 days, respectively. The water management in the DSR field followed instructions from the local Agricultural Technical Guidance Station and drew from the farmers’ own experience.

If instead some cows were less consistent in their time budgets across days, then the sampling error imposed by the subsampling routine would be greater, resulting in a larger ensemble variance estimate. Thus, we would expect a stronger penalty to be applied to cows whose demonstrated greater plasticity in their behavioral response to both transient and persistent changes in the production environment. For small datasets with limited number of replications, the number of subsamples could be set quite close to the size of the complete sample, and would thus emulate a jackknife approach to variance estimation . For larger datasets, however, the subsample size could be set smaller to make the resulting ensemble variance estimates progressively more sensitive to the uncertainty in the underlying behavioral signal. To evaluate the empirical performance of these dissimilarity estimators, distance matrices were calculated for the 177 cows with complete CowManager time budget records. Euclidean distance and KL Distance were calculated using base R utilities, with speed up options utilizing the Rfast package . An ensemble-weighted dissimilarity matrix was first calculated using simulated values accounting only for measurement error using the most conservative joint Dirichlet-Multinomial sampling scheme, hereafter referred to as Noise-Penalized Distance. A second ensemble-weighted dissimilarity matrix was then calculated using the same sampling scheme for measurement noise but aggregated over a 14-day subsample to account for behavioral plasticity in daily time budgets,hereafter referred to as Plasticity-Penalized Distance.

The LIT package provides users a clustering visualization utility, weed trimming trays which converts dissimilarity matrices into a dendrogram using the hclust utility in base R with default Ward D2 linkage and the generates heatmap visualizations of the resulting clustering results using the pheatmap package . HHaWPaSV ZHUH JHQHUaWHG RQ a JULG RI cOXVWHU YaOXHV IURP N = 1«10 IRU HacK RI WKH IRXU dissimilarity estimators, with complete results provided in Supplemental materials, and the results for k = 10 clusters compared in Figure 2. The LIT package also provides users a plotting utility to visually contrast the broader patterns between behavioral encodings. Outputs from the clustering utility are passed in to create a contingency matrix generated using ggplot2 with cells colored by their corresponding cell count . The heatmap visualizations for each encoding are then added to the row and column margins of the contingency matrix using the ggpubr package , and arranged such that each row cluster in either heatmap matches the order of the contingency matrix reading either up-down or left-to-right, allowing for direct and detailed visual comparison of the discretized behavioral patterns. Comparisons between the Noise-Penalized and Plasticity-Penalized encodings are provided in Figure 3.An optimal encoding strategy seeks to minimize the loss of relevant information by retaining as much of the underlying deterministic signal as possible while hemorrhaging only noise . In a hierarchical clustering framework, this is achieved by pruning thedendrogram built from the dissimilarity matrix where the branches cease to represent differences in the underlying signal. Standard pruning strategies typically either 1) to allow the user to provide a dissimilarity cutoff, below which value all further branches are grouped into the same bin, or 2) allow users to extract the first K branches of the tree .

As with the default Euclidean distance dissimilarity estimator, this approach may be appropriate for datasets with relatively homogenous variance structures. For data drawn from intrinsically heterogenous distributions, however, the branch lengths cannot be directly compared across the domain of support, making globally-defined pruning rules a suboptimal strategy for analysis of time budget data. More fundamentally, a homogenous pruning strategy may be too simplistic for many PLF sensor datasets, for which the underlying signal often represents a complex composite of behavioral mechanisms that operate at multiple scales. While some environmental factors might be expected to have an impact on cattle behaviors that is fairly uniform across the herd, other factors might elicit responses that differ in magnitude for different subgroups within the larger populations, or even become isolated within smaller social cliques. For example, we might expect the number of times cows are moved each day for milking will place similar constraints on the time left to lie down across all animals, but overstocking with respect to stall spaces might have a much larger magnitude of impact on the lying patterns of subordinate heifers than the more dominant older cows . In such a complex system, we would expect the heterogeneity imposed by the underlying biological signal to differ in scale across the dataset. Subsequently, in attempting to employ a global cutoff decision to encodeinformation for such a dataset, we would always be faced with the difficult decision to either ignore the subtler behavioral patterns present in some branches of the tree, or else allow noise to contaminate our encoding of other branches with intrinsically coarser behavioral patterns.

While all the components that contribute to the signal in a complex livestock system might be difficult to anticipate a priori, we propose that a more dynamic pruning algorithm might still be achieved by again employing flexible simulation-based approaches to emulate the comparably simpler sources of uncertainty. If each branch of the dendrogram is viewed as a pairwise contrast between two groups of animals, then we need only to determine whether the bifurcation under inspection represents a difference in the underlying signal can be reliably distinguished from noise. By implementing such a branch-level test recursively, we can gradually work our way down the tree with locally-defined pruning decisions. To evaluate the reliability of the behavioral signal encoded at each bifurcation of the tree , our branch test utility utilizes two mimicries. The first set of simulations are generated under the alternative hypothesis that assumes a branch contains an underlying deterministic signal that is only partially obscured by stochastic noise. Here then we can simply repurpose the ensemble of simulated data sets used previously to calculate the ensemble weighted dissimilarity metrics by mimicking the uncertainty in the observed data. As the null implies that animals demonstrate equivalent patterns of behavior within the resolution of the sample, this mimicry can be generated quite efficiently using a standard bootstrapping routine , wherein time budgets simulated under the alternative are unconditionally resampled from amongst all animals in a given branch. HClustering is then performed independently on each data mimicry in either ensemble and the first k branches are extracted to create an ensemble of discrete encodings. Under the alternative hypothesis, a strong signal should produce a robust tree structure such that, even after the addition of simulated noise, the resulting encoding would still closely mirror that of the original observed data. As the stochastic component of a dataset becomes stronger relative to the signal, these bifurcation points will become progressively less stable and the subsequent encodings less reliably aligned with the original data. When the signal falls below the resolution of the data, the tree structures of the simulated data would then seldom match that of the original data, and so would become poorly distinguished from encodings generated under the null with no signal component. We propose that mutual information, which can be calculated without any additional distributional assumptions, can be used to quantify the similarity between of the observed data and each mimicked dataset, and subsequently used to determine if simulations under the alternative are distinguishable . Equipped with an appropriate encoding to discretely represent the heterogeneity in overall time budgets within this herd, trimming trays for weed and provided the encoding of longitudinal patterns in parlor entry position from previous work with this data set, a potential question to ask would be: There are a number of nonparametric and parametric techniques available to evaluate the overall strength of association between two discrete variables by evaluating the distribution of animals in the joint encoding .

There is, however, perhaps greater practical utility in characterizing low and high points within the joint encodings, which would provide more detailed insights into the tradeoffs between specific behavioral patterns recovered from the data streams in these distinct farm contexts. Towards this end, information theory offers a more comprehensive approach to decomposing the stochasticity within discretely encoded variables, and thus may provide a more holistic approach to evaluating both the global and local features of a joint encoding, while employing few structural assumptions . First, to evaluate the strength of overall relationship between two discretized behavioral responses, the LIT package provides users a permutation-based bivariate testing utility thatuses the mutual information estimator to quantify the amount information entropy that is redundant between the two encodings . We can anticipate, however, that the efficacy of this test in recovering significant relationships between the underlying biological signals will be affected by the resolutions of the encodings. Suppose that a single latent biological factor impacts the behavioral responses collected by both PLF data streams, creating informational redundancy between the two encodings. If we cut the trees above the intrinsic magnitude of its impact on a given behavior, its influence may be overlooked and mutual information under-estimated. On the other hand, if we prune the tree far below the magnitude of its impact, our inferences can lose power as bins sizes in the joint encoding become progressively smaller, weakening the empirical estimation of the joint probability distribution and thereby increasing estimation error in the MI estimator. The resolution of our encodings must, therefore, be optimized to match the dynamics of the system, or a false negative result may be returned. To further complicate matters, however, we cannot necessarily assume that the magnitude of impact of a given latent factor will be uniform across behaviors, nor should we expect in a complex farm environment that behaviors will be influenced by a single latent factor. To overcome this logistical challenge without falling back on dubious a priori assumptions, the LIT package implements mutual information-based permutation tests on a grid, varying the cluster resolutions across both behavioral axes . Under the null hypothesis that no significant bivariate relationship exists between data streams, cow ID labels are randomly permuted within each tree, preserving the marginal distribution of the data along each axis but destroying any latent bivariate relationships. These permuted trees are then cut and the mutual information of the joint encoding estimated for each combination of cluster counts on the grid. A p-value is then generated by comparing the observed MI value of the joint encoding at each grid point against the corresponding distribution of MI values simulated under the null. Just as a scientist varies the focus of a microscope to bring microbes of different size into resolution, we can expect that geometric features of the joint probability distribution imposed by latent deterministic variables that vary in scale of impact will come into and fall out resolution as these meta-parameters are varied across the grid of cluster counts. To help the user visually identify where such features have come into resolution, the LIT package also returns a heatmap visualization of the observed MI value for each grid point that is centered and scaled relative to the distribution of MI values under the null. For behavioral measurements subject to the influence of multiple biological and environmental factors operating simultaneously, this exhaustive approach to parameterization enables users not only to build a more complete picture of a complex behavioral system, but may also provide insights into the hierarchy of these behavioral responses. Unfortunately, as the resolution of the encodings is increased, MI estimates not only become less precise, but they may also become less accurate. Bias is introduced when empirical estimates of the joint probability distribution become so granular that regions with low but nonzero probabilities go unsampled. These zero-count bins cause the total entropy calculated from the empirical joint probability distribution to be under-estimated, which in turn causes the relative amount of redundant information to be over-estimated. While the magnitude of this bias is partially dependent on the total sample size, it is also contingent on the structure of the joint probability distribution itself, namely the number of low-probability cells. Given that the joint probability distribution under the null, which is randomly permuted to intentionally remove any nonrandom features in the sample, can be expected to have a more uniform distribution of probability than the observed dataset, we can anticipate that the magnitude of the bias may differ between these two distributions as the sample becomes more granular, preventing MI estimates from being directly comparable.

After centering and scaling cow attribute variables, linear fixed effects were added for cow age , calving date , and peak milk yield . Interaction effects were created for each combination of these linear terms, and a categorical effect added for the control and treatment groups of the fat supplementation trial. Models were generated for both the complete dataset and the subset of animals with no recorded health events, which consisted of 160 and 104 cows respectively after removing animals with incomplete attribute records. While UML insights served to improve the specification of model variance structures within animal, the validity of statistical insights made at the between-animal level is still contingent upon the correct estimation of model degrees of freedom. A fundamental assumption of frequentist tests is that observations must be independently sampled. When observations are not independent, the effective degrees of freedom present in the model may be lower than the nominal value. This causes the model to be overconfident in its estimation of error terms, increasing the risk of a false positive result. Non-independence due to repeated sampling has here been accounted for by fitting arandom effect for each cow, but non-independence between animals has not been accommodated. The results of the diffusion map and data mechanics visualizations did not recover overwhelming evidence of coordinated movements between animals through the queue, which would have signified non-independence due to social cohesion ; however, mobile racking we both visualized via data mechanics and know intuitively that in this physically constrained system any cow moving forward in the queue must be countered with other cows being forced backwards and vice versa.

If this effect extends beyond isolated fluctuations in daily formation of the queue, then the presence of some animals in the herd might systematically dampen or even completely prevent other animals from demonstrating behavioral patterns that they would otherwise display independently or in another herd with a different social composition . This would not only serve to confound the behavioral mechanisms at play, but such cows whose behaviors are suppressed by their herd mates cannot be said to be contributing fully to the model, potentially reducing the effective sample size. This could allow sampling fluctuations to produce misleading statistical inferences, even in this large sample of animals . UML algorithms cannot recover information about behaviors that were never expressed, and so are also not immune to the biasing effects of non-independence between animals. These tools can, however, provide model-free tests of association that may serve as a sanity check for statistical inferences when degrees of freedom may be uncertain. We explore this option here by again combining modern clustering tools with a flexible information theoretic approach to pattern detection . First, independent clustering tress were used to subdivide the herd based on queuing records and each of the cow attributed variable. The resulting categorical variables were then used to form contingency tables between queue subgroups and each of the candidate predictor variables. If no relationship existed between these two axes, then a cow belonging to a given row category based on queue records would be just as likely to belong to any of the column categories based on cow attribute and vice versa.

If instead an underlying biological mechanism was present linking these axes, then cows within a range of cow attribute values would be spread unevenly among queue subgroups. Such heterogeneity in cell counts was quantified by calculating a weighted mutual conditional entropy value across first the rows and then the columns of the contingency table and averaging the results, which reflected the amount of mutual information shared between the two variables. To determine if the observed MCE value was significantly smaller than would be expected from random fluctuations in the sample, row and column classifiers were randomly permuted across cows to remove any underlying bivariate relationship and MCE recalculated. This randomization procedure was repeated over 2000 iterations, and the observed entropy value compared to the resulting empirical CDF to produce a p-value for the significance of the bivariate association. Mutual conditional entropy tests were performed for all significant or marginally significant linear effects for both regression models. As this herd was also fitted with ear tag accelerometers, it is here also possible to explore relationships between queue position and behavioral patterns displayed between milkings. Due to the size of these datasets, however, this small step beyond the bounds of the existing literature constitutes a considerable leap in statistical complexity within a linear modeling framework. A multivariate mixed model that considers all observations from either dataset would exceed the capacity of many solvers . A simpler approach to exploring this relationship might therefore be to compress the information available in parlor entry records into a grouping variable and then attempt to identify differences in the various home pen behaviors across the resulting subsections of the herd.

We implement this strategy here by using the nlme package to fit linear mixed models, with cow fit as a random intercept, against each of the five behaviors recorded by the CowManager platform and also average body temperature . To avoid the risk of anomalous behaviors that might skew model inferences, only cows with no recorded health events were used. Hour of the day was fit as a categorical variable to capture cyclical patterns. Days on trial was also fit as a categorical fixed effect to allow for non-smooth longitudinal changes in behaviors due weather and also the shift to pasture. Finally, queue groups were determined by arbitrarily dividing the herd into quartiles based on median entry position. The resulting categorical variable was then fit as both a main effect and an interaction effect against both cyclic and longitudinal time effects. Due to the size of the model, temporal correlation and heterogeneous variance models both exceeded the capacity of this package to converge. Comparisons of the cyclic and longitudinal trends in behavioral patterns between queue groups were made using the plotting utility available in the emmeans package , with the complete results provided in the supplemental materials. While linear models provide an expedient means to statistically evaluate targeted experimental hypotheses, the more open-ended approach to knowledge discovery provided by UML algorithms may offer an advantage in exploratory data analysis problems such as this. We explore the utility of this alternative strategy here by again employing a mutual conditional entropy test to identify significant associations between these two behavioral axes . The flexibility of hierarchical clustering tools allows this technique to be directly extended from the previous section, which compared repeated measures of queue position against a univariate covariate, to accommodate both high dimensional datasets. For each parameter recorded by the CowManager platform, this model free test of association was performed on the complete sensor record, on subsets of the records corresponding to each of the three lounging periods , vertical grow system and finally on a subset of the records where observations from all three lounging periods had been aggregated. As in the previous section, the number of clusters used to discretize queue and sensor data were evaluated on a grid, here from tree depths 2-10. To characterize the divergent behavioral patterns across queue groups identified by significant tests of association, tube plots were created by plotting each within day subgroup median on a circular grid and then stacking rings to form a tube using the 3D plotting tools in the plotly package .Looking first at the entropy calculations for each segment of the queue visualized in Figure 1, it is clear that all parlor entry positions are not stochastically equivalent. The same animals are seen consistently at the very front and back of the queue, such that the resulting entropy values are far lower than would be seen with a purely random queueing process. Moving towards the middle of the queue, however, there is progressively less consistency in the animals present across milkings, such that the observed entropy values approach a random process. Looking next at the stochasticity demonstrated by each individual cow in Figure 2, we see there is again a clear gradient.

Cows with median entry quantiles at the front and rear of the herd again show far greater consistency in their entry positions. As their median quantile position moves towards the center of the herd they become more variable intheir entry positions over the observation window. This gradient is seen using both entropy and variance as estimators of stochasticity, but is more visually distinct using entropy estimates. While discretizing an intrinsically continuous parameter results in a loss of information, we see here that this sacrifice has excluded extraneous noise in the system to bring the underlying stochastic pattern into clearer resolution. This data thus highlights the potential upside of amending entropy estimates to the traditional cadre of summary statistics, particularly when working with outcome variables that are prone to extreme or anomalous values.In examining the results of the permutation tests, nearly all animals demonstrated significantly less stochasticity in their entry positions at the standard 0.05 significance level as compared with a completely randomized queueing process. Only 3 cows out of 114 overlapped with the empirical distribution of entropy estimates under a randomized queueing pattern, and only 1 cow overlapped when variance was used as the estimator of stochasticity. This suggests that nearly all animals in the herd might contribute some information about the underlying nonrandom patterns in queue formation to subsequent analyses; however, the amount of information they contribute may not be equal, as there is considerable heterogeneity between cows. Of greater concern, this heterogeneity is systematic, as there are no cows showing high consistency in entry quantile in the center of the queue. If this pattern is not driven by variability in the underlying predictors of queue position, but instead reflects either an underlying behavioral mechanism or something even more fundamental to this system such as the inherent domain constraint , this could lead to inaccurate statistical inferences. Toavoid such risks, these simple visualizations provide clear evidence that a nontrivial variance model should be incorporated into the model specification phase to accommodate the heterogeneous variance structures in this dataset. Finally, the insights gleaned from these entropy-based visualization techniques agree well with the prior literature. Previous studies have repeatedly determined milk order records to be significantly more consistent than would be expected from a random queuing process using an array of correlation and regression-based approaches . Fewer papers, however, have explored differences in the consistency of entry positions between animals. Gadbury observed that only a subset of his herd seemed to demonstrate clear preferences for parlor entry positions. Such preferences do not appear to have been constrained to the front or back of the queue, however, as Gadbury also reported animals with a preference for the middle of the queue. In a more recent analysis with large commercial herds, however, Beggs et al. reported a nearly identical parabolic relationship between mean entry quantile and variance. With clear and consistent evidence of nonrandom patterns having been recovered from this dataset, further investigation of the behavioral mechanisms that might give rise to such heterogeneity in milk order records was clearly warranted.Visual inspection of the scree plot produced from PCA analysis revealed only one significant dimension was recovered from the original 80-dimensional dataset. To visualize the resulting projections, the first two principal components were plotted . In two dimensions points also appeared randomly scattered with no clear clustering. Thus, the PCA results revealed no compelling visual evidence of social cohesion. As this was the only significant dimension, this may suggest that a linear model to predict variations in central moment would be a reasonable representation of this dataset. This feature of the dataset was not, however, self-evident in the geometric relationships between data points revealed by the PCA projection, and thus might have been overlooked without specification of color encoding by median entry quantile value a priori.Evaluation of eigenvalues returned by the diffusion map embedding identified five significant dimensions. The 3D visualizations of these axes in Figure 3:B and provided in supplemental materials revealed quite clearly the underlying linear geometry of this dataset. Color encodings showed that the relative positions of animals along this narrow geometric band were determined by median entry quantile, further reinforcing that central moment was the most defining feature of this dataset.

We therefore examine this cycle not only as a means to save water, but ask if and how it can enhance the viability of nonindustrial farming operations as the food system adapts to restricted water availability. We consider the relevant policy recommendations outlined in Blesh et al.’s analysis of how institutional pathways can act synergistically with farmer networks to enable agricultural diversification , asking which have the potential to point future dry farming towards scaling size vs scope.To better situate these policy options in the local context, we first look to the outcomes of institutional intervention in organic strawberry production in a very similar region on the Central Coast, and consider the analogous options for dry farm tomatoes. Similar to dry farm tomatoes, organic strawberry production was launched into the spotlight by government-mandated input curtailments . For strawberries, the development of an organic strawberry production system also coincided with the adoption of an organic certification process by the US Department of Agriculture. Growing public interest in organic strawberries and the methyl bromide ban led to the rapid expansion of industrial-scale organic strawberry production– blatantly scaling size of production . As production increased, organic strawberry markets saturated and prices crashed, leaving an economic landscape where only the largest operations could remain viable selling strawberries at market prices . At this point, agroecological growers had to redouble their efforts to target local consumers with direct marketing strategies, weed drying rack as the organic label no longer added the necessary value to profitably sell their product.

In an analogous case for dry farm tomatoes, it is easy to see the immediate appeal of establishing a “dry farm” label that can incorporate the social value added to dry farm tomatoes into the price of the product without relying on consumers trusting and paying a premium based solely on higher qualities. However, by divorcing dry farm practices from quality premiums and trusting relationships with customers, a dry farm label would make it much easier for large-scale growers to enter the dry farm market. These larger operations–which may struggle to produce high quality fruits or maintain direct relationships with customers but can still decrease water usage enough to produce a certified dry farm tomato–could easily grow dry farm produce at large enough scales to edge smaller growers out of the label. As has been seen in the organic program, industrial growers could also lobby for an official relaxation–a literal watering down–of label standards . This sidestep of the dry farm practices described in the above interviews would not only further advantage large scale farmers, but would also undermine the very water savings that they are meant to encourage.Larger scale growers may also be favored when farmers are paid to implement specific practices. Administrative costs involved in enrolling in payment-for-practice programs can be a cumbersome barrier to entry, while low payouts at small scales dissuade small farmers who implement the practice from enrolling . These patterns are currently seen in programs offering cost shares for cover cropping, where farm size is significantly larger for participants than non-participants .Given farmers’ interest and current experimentation with dry farming non-tomato vegetables, expanding the set of crops that can be dry farmed and adapted to local conditions is a clear target for future policies. Support for research and participatory breeding programs/variety evaluation could spur development of locally-adapted dry farm varietals.

By compensating farmers for experimentation with diversified dry farm rotations and development of locally adapted varietals, policymakers can also absorb some of the risk inherent to on-farm experimentation and encourage innovation on the farms that are most familiar with the practice, while simultaneously lowering barriers for farmers new to the practice. To create a policy environment where experimentation feels more accessible to farmers, minimum lease terms could be set for farmland, allowing farmers to feel more secure in investing in localized practices .Priority could also be given to creating programs that connect farmers–particularly new farmers and those who hold underrepresented identities–to available farmland. Without the burden of securing water access, lands that would otherwise be impossible to farm with summer crops could become arable, particularly in conjunction with the concurrent support of the other policies discussed here. Though many areas will still require some access to water to successfully dry farm , crops’ need for water coincides with points in the season when surface water is most available , making areas with inconsistent water access over the course of the season likely candidates for dry farm success. Priority might initially be given to areas shown as suitable on the map, but as new and locally adapted crop varieties emerge, access may also extend.What is the difference between data and information? Although these two terms are often used interchangeably in biology, entire sub-fields of data science are dedicated to solving this riddle . While mathematicians have developed formal definitions for these terms , we will attempt to illustrate the conceptual differences with a simple example. Suppose a graduate student wants to analyze the gait dynamics of a dairy cow using a 120 Hz leg-mounted accelerometer.

In developing their data collection protocol, they surmise that if one accelerometer would be good, then two accelerometers must be even better. For their pilot study, they strap two sensors onto the same leg of their favorite cow and record her walking normally down the alley way in a straight line. If both sensors were calibrated and synchronized properly, then these sensors might produce data that looks something like the plot in Figure 1A. From this graph, can we learn anything about this cow from the blue sensor that we could not have inferred from the red sensors ? Since these sensors are quite precise, the two signals are virtually identical. Thus, even though this graduate student has doubled the number of data points recorded by adding a second sensor, they have not really collected any additional information about the gait dynamics of this cow.As this example has hopefully highlighted, data can be considered a physical resource ± something that can be measured in bytes and memory space. Information, on the other hand, should be regarded as a more nebulous unit of measure that represents how much can be learned from a dataset. In experimental settings, collecting measurements is typically costly, so sampling strategies are often developed to minimize redundancy between data points, such that data and information are often functionally equivalent terms . With sensor technologies, however, sampling density is often dictated by the hardware, which can result in a considerable amount of redundancy between data points . With such datasets, information compression strategies are often implemented in an effort to reduce the size of a dataset without losing any potentially useful information about the system. Returning to our previous example, suppose the graduate realizes they have not requisitioned enough hard drives to store data from both accelerometers. They decide to average the output of the two sensors at each time point, as illustrated in Figure 1B. This simple information compression step reduces the size of their dataset by half, but no details about the signal have been sacrificed. In the struggle that ensues to attach the accelerometers, one sensor is accidentally strapped on upside down. As a result, the two signals become inverted, as illustrated in Figure 1C. Subsequently, if the previously validated rolling average filter were applied to this new dataset, mobile rack the signals would cancel out, leaving only the measurement noise as shown in Figure 1D. Thus, this example serves to illustrate just how easily biologically relevant patterns can be lost when the assumptions employed in an information compression strategy do not match the data to which it is applied. As the range of Precision Livestock Farming technologies that are available to farmers to herds continues to expand , sodoes the need for robust algorithms that are capable distilling the large quantities of data that these systems produce down to useable knowledge . In data science, there LV ³QR fUee OXQcK´ ±no single algorithm that can be expected to perform optimally in all scenarios . In this paper, however, we will demonstrate how model-free unsupervised machine learning approaches can be used to recover and encode complex behavioral patterns from datasets generated outside of controlled experimental contexts, where an appropriate model may not be immediately obvious.

We will then explore how model-free information theoretic approaches may be used to recover complex nonlinear associations between farm records and multi-sensor systems in order to provide more comprehensive ethological insights.Animal scientists are seldom forced to work with raw accelerometer data, as in the previous example. Therefore, to further illustrate the fundamental differences between model-dependent and model-free approaches to information compression and knowledge discovery, lets consider a more practical example. Suppose a farmer has a group of 100 cows that are currently overstocked at 200% stocking density in a free stall barn. Concerned they might fail the outcome measures on their upcoming welfare audit, the farmer contracts a consultant to analyze data from the leg mounted accelerometer system that the farm uses in its estrus detection program.The farmer would like to know if there is any clear evidence of animals with compromised welfare in 60 days of archival records? They begin by implementing standard exploratory data analysis techniques, and produce the plot in Figure 2, wherein each dot represents the proportion of time a given cow is recorded lying down on a given observation day, with a LOESS curve fitted to reflect the mean lying time for the entire herd at each time point. From this visualization they glean that there is a considerable amount of variation between cows, and that there certainly are animals on any given day that are not spending a sufficient amount of time lying down . There is not, however,any clear evidence in this graph that would indicate that lying time patterns are changing at the herd level over time – quite the opposite actually, the herd mean is incredibly stable over the observation interval.Based on this visualization, the consultant constructs a mixed effect model with ³cow´ as the random effect, to account for repeated measurements and avoid pseudo-replication. Since the data appears stationary, they add ³day of observation´ as a simple continuous fixed effect. Finally, based on their preexisting knowledge of factors that may affect lying time, they add a categorical fixed effect to distinguish between ³heifers´ and ³cows´ within the herd. The results of this modelare provided in full in Supplemental Materials. As anticipated from the EDA visualizations, the fixed effect for day is neither biologically nor statistically significant . The consultant is, however, a bit surprised to find that the difference in average lying time between heifers and cows is also neither statistically or biologically significant . Thus, the consultant determines that the expected proportion of time spent lying for any animal within this herd for any given day within the observation window is 40% . In appraising the variance estimates for the random effects in this model, the consultant also sees that between-cow variance is considerably smaller than the within-cow error term , from which they conclude that there is also no evidence of consistent individual differences in lying time within this herd, and so these results also do not indicate that any individuals are consistently above or below the expected lying time for this herd. Subsequently, they might confirm that, while cows might not get to lie down an adequate amount of time every day, and that the average lying time could certainly be improved, there is also no evidence of cows in this herd that are consistently, and critically, under-rested. Let¶ s now suppose that the milk buyer is dubious of this result, and so they send the data to a new consultant who specializes in welfare audits. They ask the farmer some additional questions about this herd before constructing their own model and learn that the start of grazing season fell somewhere in the middle of this observation window, at which point cows had free access to pasture from their free stall barn. This consultant, suspecting that heifers and cows might react differently to this pasture access, develops a similar model to the first consultant, except this time an interaction term is included to allow the temporal dynamics to differ between these subgroups. With this small modification to the fixed effects matrix, this model now tells an entirely different story.

The experiment was conducted on seven certified organic dry farm tomato fields in Santa Cruz and San Mateo counties in California during the 2021 growing season. Five blocks were established on each field over the course of a full growing season , for a total of 70 experimental plots. These fields are managed by six farms; one farm contributed two fields at two separate sites. Each farmer continued to manage their field for the duration of the experiment according to their typical practices. Each dry farm crop was preceded by a crop in the winter prior to the experiment, either in the form of a cover crop , or continuous winter production . All fields were disked prior to planting, and two fields additionally ripped down to 60-90cm. Each field’s plant and bed spacing, plant date, and tomato variety are listed in Table 1, along with amendments added to the soil. Fields also varied in their rotational history . The mapped soil series, measured texture, and soil pH are listed in Table 3. From March 2 to October 27 there were 15 rain events greater than 1 mm recorded at the De Laveaga CIMIS weather station , none of which occurred between the months of May and October . Monthly weather data is summarized in Table 4. A nested experimental design was used to account for management and biophysical differences across fields. Ten plots were established at each field site within three days of tomato transplant. Each plot contained 12 plants, and plots were divided across two beds with a buffer row between . Plots were randomly selected to be inoculated in the first experimental row and then paired with a counterpart in the second experimental row that received the opposite inoculation condition to achieve a randomized complete block design with five blocks per field. Here we refer to a pair of inoculated and control plots as a block. There were three non-inoculated buffer plants between each plot and at least twenty buffer plants at the start and end of each experimental row.

This inoculum has been shown to impact crop physiology and improve plant water status in various field and greenhouse applications. Each of the 12 plants in plots in the inoculation condition received 0.2 g of inoculum, rack heavy duty which was mixed with 40 mL of water and then poured at the base of the plant within three days of transplanting, as per manufacturer instructions.Harvests began when farmers indicated that they were beginning to harvest the portion of their field that included the experimental plots. Each field was harvested once per week from its start date to its end date, with the exception of Farm 5, which was harvested twice per week, in accordance with farmer desires. All red tomatoes were harvested from each plot and sorted into marketable, blossom end rot, sunburnt, or “other unmarketable” fruits and then weighed. Harvests stopped when there were no remaining tomatoes in the field or when farmers decided to terminate the field. Fruit size and quality were assessed on the third, sixth, and ninth week of harvest at a given field. Ten representative marketable tomatoes were taken from each plot, weighed, dried at 70 degrees C and then weighed again to establish the percent dry weight . PDW was used as a proxy for fruit quality, with fruits with a lower water content increasing fruit quality up to a certain point. Extension research has linked dry farm fruit quality with lower fruit water content, as opposed to specific compounds that are elevated in dry farm tomatoes, and we expect PDW to correlate highly with the concentration of flavors previously found to create dry farm fruits’ superior quality. After eliciting quality categorization from farmers in the study, we determined that fruit quality increases up to a PDW of 8%, peaks between 8 and 12%, and falls above 12%.

Soil samples were taken three times over the course of the field season: once at transplant , once mid-season , and once during harvest . Each time samples were taken from four depths at each plot. Samples were homogenized and a subsample was immediately put on ice for transport to the lab. Each sample was then divided into fresh soil , dried at 60 degrees C , and dried at 105 degrees C . Ammonium and nitrate levels were measured after using 2M KCl to extract samples from transplant , midseason , and harvest samples using colorimetry35,36. As soil pH was close to neutral, Olsen P37 was used to measure plant-available phosphate on samples from transplant and midseason . Gravimetric water content was assessed for all samples. Samples from transplant were composited by depth at each field, and texture was assessed using a modified pipette method. At transplant, a soil core was taken with a bucket auger down to one meter from a central plot in each field and used to calculate bulk density at each depth increment. We then took a weighted average of GWC at each plot to calculate available water using bulk density and a pedotransfer function based on soil texture. Potentially leachable soil nitrate levels were calculated for each field using nitrate concentrations from the top 15cm at the harvest sampling event, which occurred within the first three weeks of harvest. Though the plants continued to grow for the duration of the harvest, it is unlikely that nitrate from the top 15cm were used due to the soil’s low water content, and no precipitation orirrigation occurred for the duration of harvest. Bulk density in the top 15cm was assumed to be 1.2 g soil/cm3 as experimental bulk density was measured with 1m of soil and likely overestimated the bulk density at the surface of the soil.Soil subsamples taken from 0-15cm and 30-60cm at midseason were set aside for DNA analysis. In addition to the experimental plots, samples were also taken from both depths at the nearest irrigated crop production areas and non-cultivated soils, such as hedgerows, field sides, etc. . Gloves were worn while taking these samples and the auger was cleaned thoroughly with a wire brush between each sample.

Roots were also collected from one plant per plot and were dug out using a trowel from the top 15 cm of soil. These samples were stored on-site in an ice-filled cooler and transferred to a -80 degree C freezer immediately upon returning to the lab . Roots were later washed in PBS Buffer/Tween20 and ground using liquid N.The ITS2 rRNA region was selected for amplification and fungal community analysis. This region has been successfully utilized in recent AMF community studies. Though AMF-specific primers exist , we chose the more general ITS2 fungal primers for several key reasons. First, in the field, SSU primers detect more taxa in nonGlomeraceae families but give lower resolution in the Glomeraceae family. Because the four species in our inoculant are in the Glomeraceae family and this family is dominant in agricultural systems and clay soils, we prioritized species resolution in Glomeraceae over other families. More broadly, the higher variability in the ITS2 region can lead to more unassigned taxa, but does not run as much of a risk that distinct taxa will be lumped together. Third, and of particular importance in our root samples, these primers are better able to select for fungal over plant material than other ITS primer options. Finally, ITS2 allowed us to also examine the broader fungal community in our samples, whereas SSU and LSU options are AMF specific and cannot be used to characterize other fungi.Qiime2 was used for all bioinformatics. Reads without a primer were discarded, and primer/adapter sequences were trimmed off reads using cutadapt. Samples were denoised with DADA2, and taxonomy was assigned using the UNITE version 9 dynamic classifier for all eukaryotes. Taxa outside of the fungal kingdom were removed from all samples and SRS normalization was used to reduce each sample to 7190 reads. 7190 was chosen as a cutoff due to a natural break where no samples fell between 4000 and 7190 reads. Because depths below 4000 retained less than 90% of sample richness, 7190 was chosen, cannabis drying system retaining over 95% of richness. The 22 samples out of 301 samples that fell below this cutoff were discarded. These samples included all 5 blanks, 3 samples from field 1A , 4 samples from field 1B , 2 samples from field 2 , 4 samples from field , and 4 samples from field 4 .In addition to the variables of interest, each model had a random effect of field and block within field. Yields were modeled using the total marketable fruit weight harvested from each plot at each harvest point, while BER was modeled using the proportion of fruits that were classified as non-marketable due to BER from each plot at each harvest point. Yield models and BER models treated weekly harvests as repeated measures, adding random effects of plot within block and harvest number. For hurdle models, random effects were treated as correlated between the conditional and hurdle portions of the model. Because PDW was measured at three time points, the initial PDW model treated the time points as a repeated measure and added a random effect of plot within block. However, given the nonlinear relationship between PDW and fruit quality described by farmers, further models used only PDW at the 6th harvest when fruit quality was at its peak and therefore did not include any repeated measures.The initial model for each outcome variable included plant spacing and PC1 for soil texture , along with PC1 for GWC and PCs 1 and 2 for nutrients at all four depths , as well as the interaction between texture and GWC. In this initial model, only one depth showed a statistically clear relationship with each outcome variable .

To improve model interpretability, we then replaced the two PC’s from the depth of interest with the scaled transplant values of nitrate, ammonium and phosphate at that depth, also adding the ratio of nitrate to ammonium and an ammonium-squared term to allow for non-linearities in outcome response to nitrogen levels. Because all nutrient variables had variance inflation factors over 5 in this model , we dropped nutrient PC’s for each depth that was not of interest, leaving only the transplant nutrient values at the depth of interest in the model. All nutrient VIF values were below 5 in the resulting model. Reported models were run using unscaled nutrient values for ease of interpretation. Transplant nutrient levels were used rather than midseason/harvest both because they are the most relevant to farmer management and because their interpretation is more clear than later timepoints, when low levels can either indicate lower initial nutrient levels, or that plants have more thoroughly depleted those nutrients.Two fungal community descriptors were calculated for each soil depth and root fungal community: the Shannon index and the count of OTUs in the class Sordariomycetes, which was identified as an indicator of dry farm soils . Counts were scaled, and both community descriptors were added to the final model described in the “Variable selection” section to determine the impact of fungal community structure while controlling for water, nutrients, and texture. Because the metrics between roots and the two depths of soil fungal communities were highly correlated, three separate models were run: one with both fungal community metrics from 0-15cm, one with metrics from 30-60 cm, and one with root community metrics.A PERMANOVA using Bray distances showed statistically clear differences in fungal community composition in irrigated, dry farm, and non-cultivated bulk soils as well as communities at 0-15cm and 30-60cm when stratifying by field and controlling for water, texture and their interaction, which also significantly differentiated between communities . Though dry farm, non-cultivated and irrigated soils each had more unique taxa than taxa shared with another location, dry farm and non-cultivated soils each had nearly twice as many unique taxa as taxa shared with a single other location, while irrigated soils had more taxa shared with dry farm soils than unique taxa . Abundance analysis showed that there were 466 taxa that significantly discriminated between the three soil locations. We then set the LDA threshold to 3.75 to highlight only the most stark differences, resulting in 13 discriminative taxa . All of the taxa identified as being enriched in dry farm soils were sub-taxa of Sordariomycetes, a fungal class that is highly variable in terms of morphology and function.

An interesting idea is that farmers themselves have a better idea of how they will benefit from the new technology. And this should be reflected in their willingness-to-pay to experiment with the new technology . Experimentation has shown that incentivizing contact farmers with the payment of bonuses proportional to their success with diffusion of the innovation in their community may be necessary and can be effective . It remains however unclear whether this recommendation is scalable. Experiments include sending SMS reminders to farmers on agronomic practices and using SMS to make recommendations customized to farmers’ own idiosyncratic soil conditions . In some specific cases, there may be simple techniques that can substantially improve farmers’ ability to obtain information through self-experimentation. An interesting experiment provides farmers with leaf color charts that guide them in their fertilizer doses decisions . The extension models described above rely on what can be called a “push approach”, whereby contact farmers are expected to pass information to others in their social networks. There is also potential for a “pull approach” to the diffusion of information through social learning. In this case, information is broadly broad casted in the community that something is to be learned from informed farmers. This “buzzing” can motivate farmers to seek information from informed farmers that may be inside or outside their normal social networks. Dar et al. thus show that “buzzing” through visible demonstration plots that use a counterfactual can be just as effective as seeding information through central farmers .

Under this approach, farmers know that there is something to be learned and they know with whom to engage in conversation to be informed. Banerjee et al. found a similar result in helping people understand the demonetization process in India: inform selected individuals in the community and widely tell the community who they are to induce information-seeking conversations. The advantage is lower cost and ability to reach less privileged farmers who do not belong to well informed social networks. In general, drying rack weed results show that information remains a serious constraint on SHF modernization. Extension services are even more under-funded than Research-and-Development and in need of new approaches that may work. Social learning can be made more efficient by a better choice of contact farmers and giving them high-powered incentives to diffuse information. Social learning can be reversed from push to pull for broader impact. Using IT services such as Digital Green and digital platforms offer interesting new options . Innovative approaches in addressing the information problem are in need of conceptualization and experimentation, with significant opportunities to make a large difference on SHF modernization.Incentives to adopt require good access to well performing markets. In contrast, SHF typically face poor infrastructure and high transaction costs, limited access to information on prices, lack of competition on local markets, and problems with quality recognition for inputs and outputs. Distance to market and poor infrastructure are major contributors to higher input prices and lower product prices for net sellers, which in turn act as a tax that discourages the adoption of innovations. Aggarwal et al. thus show that distance to market is equivalent to a 6% advalorem tax per kilometer for villages in Tanzania. Reducing travel cost by 50%, which is said to be equivalent to paving rural roads, would in this case increase local maize prices and double fertilizer adoption.

Improved infrastructure may however increase or decrease the prices of local crops depending on the competitiveness of local goods with those from further away. For Sierra Leone, Casaburi, Glennester, and Suri show that improving rural roads lowers prices on rural markets, benefiting consumers and hurting producers. They find that only when cell phone services help traders reach markets further away does improved infrastructure raise local prices, benefiting producers. For teff in Ethiopia, Vendercasteelen et al. show that proximity to cities increases the price received by farmers as well as the use of fertilizer and improved seed, resulting in higher yields, especially proximity to primary as opposed to secondary cities. Improved price information can also have mixed effects on farmer welfare. The role of IT services in reducing search costs, lowering local price volatility, raising average producer prices, and lowering consumer prices was observed by Jensen for fish in India and by Aker for grains in Niger. Better information on prices was also observed by Svensson and Yanagizawa in Uganda to have a positive effect on the level of prices received by farmers. However, better price information may have no effect on prices received by farmers if they have no option to sell on these markets and no bargaining power with local merchants . Local markets may also not be competitive, with the possibility of extensive collusion among merchants as shown by Bergquist in Kenya using an experiment exogenously varying the number of merchants on local markets with no consequent impact on price. This finding is however not supported in an extensive literature review by Dillon and Dambro who found that local markets in Sub-Saharan Africa overall tend to be competitive.Finally, quality recognition is a major issue on local markets, even though urban consumers may be willing to pay a price premium for higher quality, particularly for higher phytosanitary standards.

Quality recognition via third party certification, as for onions in Senegal, resulted in higher prices for good quality produce and created incentives for farmers to adopt quality enhancing technology . On input markets, Bold et al. find that there is extensive cheating on the quality of fertilizers, contributing to low adoption in Uganda. An experiment by Hasanain, Khan, and Rezaee shows that quality recognition in services markets via IT ratings can lead to improved veterinary services for artificial insemination in Pakistan. This was due to increased veterinarian effort once success rates were known to cattle owners. Lack of quality recognition for domestic production is a major issue for the competitiveness of SHFs with imported food on urban markets. It creates an increasing disconnection between what farmers produce and what urban households consume. The large number of SHFs and aggregation of production by traders high in the value chain prevents creating incentives for farmers to increase the quality of what they produce. This contributes to rising dependency of urban consumers on imported foods and low prices for domestic producers, discouraging technological upgrading. The frequent poor performance of markets due to high transaction costs, partial transmission of information, frequent lack of competitiveness, and lack of quality recognition remains a major obstacle to profitability and hence to technological upgrading. Addressing these constraints requires not only institutional innovations, but also costly public investments in infrastructure and marketing facilities.The supply-side approach to constraint removal and value chain development has helped identify a large number of technological and institutional innovations with potential to enhance adoption. In spite of this, technology adoption and modernization has been modest. As shown by continuously rising cereal yield gaps, and in spite of many local success stories, a global Green Revolution for Africa is still in the waiting. A major difficulty for technology adoption under rainfed conditions is heterogeneity of conditions. At the household level, this applies to three dimensions: farmer circumstances, farmer objectives, and farmer capacity. If these dimensions are immutable or too costly to change, technological innovations must be customized to fit these dimensions. Farmers’ circumstances such as agro-ecological conditions vary widely over short distances and across years in particular regarding rainfall patterns and soil fertility . For Zambia, Burke et al. show that only 8% of farmers can profit from basal chemical fertilizer applications due to lack of a complementary factor, in this case lime to achieve the desirable level of soil acidity.

In Western Kenya, Marenya and Barrett find that only 55% of plots can profitably use chemical fertilizers due to lack of a complementary factor, racking system in this case soil organic matter as measured by carbon content. Barghava et al. similarly find that there is complementarity between soil organic carbon and modern inputs. For adoption to go beyond farmers with complementary factors in place, technological innovations must either be customized to fit heterogeneous contextual conditions, or complementary factors must be delivered jointly with the technological innovation. Farmers’ objectives are different from breeders’ who typically focus on maximum yields in experimental plots with highly favorable controlled conditions . Farmers maximize profit or utility weighting return and risk. They may also have labor calendar objectives such as labor-saving at peak periods and labor-smoothing in the rest of the year. Labor constraints on farming may come from involvement in rural non-farm economy activities and seasonal migration, requiring to fit farming systems to accommodate complementarities between on- and off-farm labor engagements, including a gender division of tasks. The household will have nutritional objectives if part of the harvest is home consumed, and diversity of diets matter for the choice of farming systems. These specific objectives must feed into the design of new customized technological innovations. Farmers’ capacity may be improved through the acquisition of information and skills, but other dimensions of capacity are fixed factors to which technological innovations must adapt. T.W. Schultz and Foster and Rosenzweig famously showed that farmers’ education matters for technology adoption. Low skills may reduce the capacity and the speed of learning . Again, limits on capacity must be taken into account on the supply side of technology if it cannot be addressed as a demand-side constraint that can be relaxed. Technology must be kept relatively simple to use. An example is SwarnaSub1 that requires the same agronomic practices as the widely used Swarna rice variety. Another is the leaf color chart to adjust the quantity and timing of fertilizer applications. It is thus possible that available technology is not adapted to the circumstances and demands of a majority of farmers. Either it has to be adapted to the lack of key complementary factors, or the complementary factors have to be jointly delivered as a technological package. Unless this is done, lack of technological upgrading for a majority of farmers may not be an adoption issue but a supply-side issue concerning the availability of technologies that are profitable and adoptable by a majority of farmers. Lack of investment in Research-and-Development to address the specificity and heterogeneity of SubSaharan conditions noted above adds credibility to this interpretation. This is documented by Pardey et al. who shows that there is both under-investment in agricultural research in Sub-Saharan Africa as revealed by an estimated average internal rate of return of 42% for 25 countries over the 1975-2014 period, well in excess the expected return on public investment, and a continuing deterioration of the situation. Goyal and Nash document a net decapitalization of agriculture Research-and-Development capacity in Sub-Saharan Africa over the last decade. Conclusion is that an approach to using Agriculture for Development that seeks to remove constraints on adoption of existing technology from the supply side is essential, but likely to hit a low ceiling due to heterogeneity of conditions, lack of complementary factors, and diverse farmers’ objectives and capacity. Lack of quality recognition also creates increasing disconnectedness between what farmers produce and what urban consumers demand. A complementary approach to address these issues is development of inclusive value chains and constraint removal starting instead from the demand side.A demand-side approach consists in creating incentives for SHF to modernize through their participation in vertically coordinated value chains that provide links to markets for products with a profitable effective demand, while at the same time potentially offering solutions to market and institutional failures. The advantage of a demand-side approach is that it does not predetermine the solution to adoption but seeks instead broad complementarities in the ways of achieving modernization that are specific to the agent in question. Referring to the Byerlee and Haggblade classification mentioned above, we include under the category of vertically coordinated value chains, resource providing contracts and the more complex multi-stakeholder structure. Which elements are included in contracts and the specific structure and institutional form that vertical coordination will take is endogenous, depending on the particular needs of producers, the end buyers, and the context of market failures and institutional deficits.A key element of modern value chains that can result in SHF modernization is implementation of resource-providing contracts .