Understanding consumer preferences within the food system A research study in the city of Cali, Colombia Version: 4th January 26th 2021 MSc thesis (ENR-80436) Environmental economics and Natural Resources (ENR) Catalina Mesa Álvarez ID 901120560080 Supervisor: Francisco Alpízar Rodriguez Contents Preface ...............................................................................................................................................4 Acknowledgements ............................................................................................................................5 Abstract ..............................................................................................................................................6 1. Introduction ............................................................................................................................7 1.1. Problem definition ...............................................................................................................7 1.2. State of the art ....................................................................................................................8 1.2.1. Literature review .............................................................................................................8 1.2.2. Knowledge gap .............................................................................................................. 11 1.3. Relevance of the research.................................................................................................. 11 1.3.1. Scientific relevance ........................................................................................................ 11 1.3.2. Social relevance ............................................................................................................. 12 1.4. The objective of the research ............................................................................................ 12 1.5. Research questions ............................................................................................................ 12 1.5.1. General Research Question ............................................................................................ 12 1.5.2. Specific Research Questions........................................................................................... 13 1.6. Hypotheses........................................................................................................................ 13 2. Materials & Methods ............................................................................................................ 14 2.1. Discrete Choice Experiment ............................................................................................... 14 2.2. Choice experiment design.................................................................................................. 14 Definition of attributes, attributes levels and customization ....................................................... 14 Experimental Design ................................................................................................................... 16 Experimental Context and survey development .......................................................................... 16 Sample and sampling strategy ..................................................................................................... 17 2.3. Online survey (LookApp) .................................................................................................... 18 2.4. Statistical analysis .............................................................................................................. 19 3. Results .................................................................................................................................. 20 3.1. General results .................................................................................................................. 20 3.2. Environmental, health & social attributes for food consumers ........................................... 21 3.3. Effect of social demographic factors in a more sustainable diet choice .............................. 22 3.4. Effect of non-demographic factors in a more sustainable diet choice ................................. 26 3.5. Effect of information level in a more sustainable diet choice ............................................. 27 4. Discussion ............................................................................................................................. 28 4.1. Findings versus literature review ....................................................................................... 28 4.2. Strengths of the study ....................................................................................................... 29 4.3. Weaknesses of the study ................................................................................................... 29 5. Conclusions & Recommendations.......................................................................................... 30 6. References ............................................................................................................................ 31 7. Annexes ................................................................................................................................ 34 Annex 1: Online survey template (general scenario) ...................................................................... 34 Annex 2: Online survey template (second scenario) ....................................................................... 34 Annex 3: Raw data results.............................................................................................................. 34 Annex 4: Stata Do file & Stata database ......................................................................................... 34 Annex 5: Stata screenshots models ................................................................................................ 34 Annex 6: MWTP calculations in Excel ............................................................................................. 47 Annex 7: LookApp invoice .............................................................................................................. 47 Preface “Understanding consumer preferences within the food system. A research study in the city of Cali, Colombia” is a master (MSc) thesis research whose objective is to study the preferences of consumers for food products with varying environmental, social and health attributes by using a discrete choice experiment. I estimate people’s Willingness To Pay (WTP) for alternative characteristics of food, and the socio-economic characteristics that determine that WTP . The thesis is a compulsory part of the Master in Environmental Sciences (MES) programme within the Environmental Economics and Natural Resources (ENR) chair group at Wageningen University & Research (WUR) in The Netherlands. The data collection process was funded by The International Center for Tropical Agriculture (CIAT) located in Colombia; therefore, all the legal rights of this research belong to both the University (i.e., WUR) and the research center (i.e., CIAT). Acknowledgements I would like to acknowledge to all the people who were part of this learning process and for all their support to accomplish this stage in my academic career. Thank you to my parents and friends back in Colombia who constantly showed interested in the research topic and helped me with their valuable inputs, as well as my friends and second family here in Wageningen. Special thanks to the CIAT who trusted in my work and skills to conduct this research considering the uncertain COVID-19 circumstances, the LookApp team for taking care of every detail of the survey and finally my supervisor, Francisco Alpizar, and the ENR chair group for all their helpful feedback. This study was funded by the Alliance Bioversity-CIAT; supported by the CGIAR Research Program on Water, Land and Ecosystems (WLE) and CGIAR Fund Donors (https://wle.cgiar.org/donors). In collaboration with the CGIAR Program on Climate Change, Agriculture and Food Security (CCAFS) its CGIAR Fund Donors (https://ccafs.cgiar.org/donors). The study was part of a MSc thesis in The Netherlands at Wageningen University & Research. Abstract Following the Food and Agriculture Organization of the United Nations (FAO) definition of a food system, this encompasses all the stages of keeping the human population fed. At a global level, the food system is responsible for around 19-29% of the world’s GHG emissions, and is the major cause of deforestation. It also accounts for 20% of energy consumption and it is the biggest consumer of freshwater, using 70% of available resources. This complex system covers multiple stakeholders, both from the supply and demand side. This study focuses on the demand side and the aim is to understand the consumer preferences when choosing food with varying environmental, social and health attributes by using Discrete Choice Experiments (DCE). Consumers value environmental, health and social attributes related with food differently. Agrochemicals (with an impact on both the health of the environment and the human population) along with environmental factors such as deforestation level are the most important attributes when choosing food products. Human health factors are also part of the preferences of consumers but in a less importance. (Non) Social demographic factors do have an effect on the consumers ‘interests and therefore in their MWTP for a more sustainable diet. Keywords: food system, consumer preferences, behavioural economics, Contingent Valuation (CV), Discrete Choice Experiment (DCE), Willingness To Pay, WTP. 1. Introduction 1.1. Problem definition If you out of curiosity look for the definition of energy, you can find something such as “the strength and vitality required for sustained physical or mental activity”1. Human beings are constantly using energy sources to move and do their activities. One of the most important energy sources comes from the daily food intake which is influenced by cultural, economic and social factors, and it depends on the interests, preferences and needs of each individual. Behind each every day meal there is a complex food system operating to produce it and bring it to our table. According to Mann, D. et al (2018), this food system at a global level is responsible for around 19-29% of the world’s GHG emissions, and it is the major cause of deforestation. Also, it accounts for 20% of energy consumption and it is the biggest consumer of freshwater, using 70% of available resources. Additionally, the current food system covers around 30% of all ice-free land in the world. Human population growth is one the main drivers that put the current system and its viability at risk. As specified by the United Nations, in the upcoming thirty years the global population will increase by 2 billion persons, reaching 9,7 billion in 2050 (UN, 2019). Under the current food system, this expected growth will bring increased pressure on Earth’s resources, threatening the viability of the coupled human-environment system. There is a need to rethink and adjust the current food system and our consumption patterns in order to match the planetary boundaries, and sustainably manage the available resources. We need a food system that can properly feed the population while taking into account both the human and environmental health. Thus, one of the most considerable challenges nowadays is to act upon the fast-changing demand for food from a bigger and more affluent population to its supply in the time that environmentally and socially sustainable methods are being used (Godfray et al., 2010). Following the FAO2’s definition(FAO, 2016), a food system encompasses all the stages of keeping us fed: growing, harvesting, packing, processing, transforming, marketing, consuming and disposing of food. Thus, it is evident that many stakeholders are involved throughout the entire system. Food supply chains upstream from farms, to the food industry in the midstream segments of processing and wholesale and in the downstream segment of retail, then on to final consumers (Reardon & Timmer, 2012). Within the supply-side, latest studies such as the research of Bojana Bajželj, demonstrates that current tendencies in yield improvement will not be enough to fulfil the food demand forecast in 2050, and therefore a further expansion of agricultural area will be needed (Bajželj, B et al., 2014). It is important to note that agriculture is the most important driver of biodiversity losses and a significant contributor to climate change and pollution; as a result of this, further expansion is not desirable. The commonly proposed alternative, which is intensification with increased resource use, has adverse effects as well. Given the objectionable consequences of the aforementioned solutions from the supply-side, we must do better in terms of our consumption choices; so, it is imperative to evaluate and boost the initiatives from the demand side; in other words, the actions coming from the final consumer. Food consumption choices can alter the current supply and lead to a shift. This decision may be affected both for reasons 1 Oxford Languages definitions. 2 FAO, an agency of the United Nations, stands for Food and Agriculture Organization. in the short term, such as income level and nutritional levels of the products, as well as reasons in the long term that involve a more environmentally sustainable diet as well as in strengthening levels of health of the human body. Considering the different options that consumers have when they are deciding what to eat, one of the greatest challenges is to incorporate the environmental factor into their decisions, through product alternatives that have the least environmental impact in terms of their form of production and distribution while the health requirements are fulfilled. The aim of this study is to understand the consumer preferences when choosing food. This research studies consumers responses in the city of Cali, which is the third-most populous city in Colombia with 2.4 million inhabitants, in relation to health, social and environmental factors and their willingness to pay for those attributes. Thus, the main goal is to evaluate the potential of the demand-side to take centre stage in the shift towards a more sustainable food system. With the outcomes of this research, there will be a better understanding of the consumer’s trade-offs between environmental, health and social attributes, thereby informing policy makers and food producers about their client’s interest and willingness to pay for more sustainable food. This work also aims to extend a current investigation lead by The International Center for Tropical Agriculture (CIAT)3 in relation to carbon footprint and climate risk of the most popular food products consumed in Cali, with the purpose of characterizing the current food consumption in terms of its environmental, social and health consequences. 1.2. State of the art 1.2.1. Literature review Several studies have been conducted on consumer diets and their impact on both health and the environment. It is important to highlight that several researchers have stressed the importance of understanding the preferences of consumers when buying their food, in order to create strategies from the demand-side that guarantee food availability in the upcoming years, while doing both a sustainable natural resources management and a transition to healthier diets. Tilman and Clark (2014) made a study at global level where they quantified the relationship among diet, environmental sustainability and human health using results for food Life Cycle Assessments (LCA). They have evaluated the potential of environmental impacts of the global dietary transition, and how to alleviate the consequent impacts of the diet-environment-health trilemma. They showed how Greenhouse Gases (GHG) emissions are highly dependent on the consumer’s diet and not necessarily a healthier diet minimizes the environmental impacts; thus, it is needed to seek for healthier diet with low GHG emissions. Along with the consequences of the diets we choose, this choices are influenced by culture, nutritional knowledge, price, availability, taste and convenience (Tilman and Clark, 2014). The authors’ conclusion, due to the great challenge of the global dietary transition, is to build joint substantial solutions together with nutritionists, agriculturists, public health professional, educator, policy makers and food industries. 3 Project’s name: Assessing the resilience and environmental sustainability of the Cali city region food system. It belongs to research program on Water, Land and Ecosystems (WLE). A scenario study of the global land system made in 2014 (Bajželj, B et al., 2014) stressed the idea that food waste and dietary change are the two most prominent demand-side measures and have been shown to have a large potential. The different results for the current trend scenarios and the future scenarios the authors used to do the forecast show that solely when the plan includes relevant elements of demand reductions, then it is plausible to avert an increase in agricultural expansion and therefore the related GHG emissions. These results also are supported by the recent report of the Intergovernmental Panel on Climate Change (IPCC) which stated a significant, but unknown potential for GHG emissions’ reduction in agriculture from demand-side measure, while improving the scenario for food security and environmental sustainability (Smith P. et al, 2018). Supporting the previous research, a study about the diet, health and environment trilemma in 2018 brings up the need to evaluate this issue in a local level, taking into account the social, economic and cultural values of each region. Beyond, they expressed there is a special attention in developing countries where the increased affluence and urbanization is seen as a cause of less healthy and less sustainable diets (Clark, Hill and Tilman, 2018). Globally, as possible results of the adoption of a healthier and more environmental friendly diet, the authors explained the world could see a diet- related GHG emissions reduction by approximately 30% to 60%, a drop of approximately 20% to 35% in cropland use, a considerable reduction of future threats to biodiversity and a lower use of fertilizer inputs. This quantifies the importance of working on how to make more sustainable the consumer’s food choices in order to guarantee enough and proper food in the following decades. Now, understanding that the final consumer is a key stakeholder in the search for more sustainable food systems, numerous studies have been carried out to understand how consumers make more frequent pro-environment, pro-health and pro-social welfare decisions, and what kind of stimuli make them more aware or willing to change their food preferences. A study made in Australia by Mann et al. (2018) strengthen the idea that food choices take an important part to the pressure that human population set on the environment. Contrary to what Tilman and Clark declared in 2014, they argued that diets with improved nutrition profiles lead to lower impact on the environment than the current less healthy diets followed in many Western countries. Whether there is a correlation between health and environment within the food diets’ perspective, the goal is to engage people to make both healthy and environmental friendly decisions towards a more sustainable consumption (Mann, D., et al., 2018). Coming back to the understanding of consumer preferences, the Kollmuss and Ageyman’s model has been used in multiple researches as a guideline to explain consumer’s engagement in pro- environmental behaviors (Kollmuss and Agyeman, 2002). The pro-environmental behaviour is affected in this model by three main different variables: attitudes, knowledge level and perceived effectiveness or locus of control4. The study developed in Australia by Mann et al. (2018) used this model and the conclusions suggest more information, or an increase in the consumers’ level of knowledge, is required to help them towards the direction of pro-sustainable eating patterns. Nevertheless, knowledge is just one of a wide range of interventions needed to achieve this shift. Internal (e.g. perceived effectiveness) and external (food availability) barriers exist, so a combination of multiple interventions is mandatory. It is important to point out that, even though plenty theoretical frameworks have been created to explain the interaction between these variables (i.e. environmental knowledge, awareness, attitudes) 4 Consumer Perceived Effectiveness (CPE) refers to the person’s belief that his/her behavior can contribute to environmental preservation. and the real actions taken by the consumers, no definitive answers have been found (Kollmuss and Agyeman, 2002). What is commonly expressed within this field is that, even if today there is not an unique response to act upon more sustainable diets and changes in consumer preferences, a change in dietary behaviors in response to the interventions have been done is slow and it needs the participation of diverse stakeholders. This suggestion is bolstered by Godfray et al. in a study of the impact of meat consumption on health and the environment (2018). They said that social norms can change and therefore do change, but a coordination of efforts among civil society, health organizations and government is mandatory. Previous to this coordinated work, a high-quality understanding of the consumption patterns impacts to health and environment should be addressed and deployed to society for a suite of strategies to encourage change (Godfray et al., 2018). The gap between the attitudes and behavioral intention of consumers was analyzed by Vermeir and Verbeke in 2006. This study explores the decision-making process that includes the consumer’s social responsibility (e.g. social, environmental and health aspects) into the individuals needs and desires (Vermeir and Verbeke, 2006). Supported by statements made by the Food Standards Agency (FSA) in the United Kingdom, they expressed the daily consumption habits are driven by “convenience, habits, value of money, personal health concerns, hedonism and individual responses to social and institutional norms” and those practices are prone to be resistant to change. This change also depends on the type of consumers. Within this post-modern society, reflexivity plays a key role. Here, Giddens (1991) argues that the reflexive consumer is able to do an individual risk assessment of his/her own decisions and it is more aligned to cultural norms, and Dupuis (2000) claims that food is a distinct example for these type of consumers since food consumption involves the decision to either eat or not a specific product and let into his/her body. Another type of consumer is described in this study, the ethical consumer. This individual feels responsible for society in many ways such as environmental issues, animal welfare, human rights, labor working conditions, etc. That feeling influences his/her purchase behavior (Vermeir and Verbeke, 2006). As a general norm, the authors argued that the attitude-behavior gap applies to all consumers. This gap explains that considering just attitudes is frequently a bad predictor of the real behavioral intention, and characteristics as quality, price and brand familiarity are the main attributes to make a decision rather than ethical or reflexive decisions that are priorities for a minority (Weatherell et al., 2003). Thus, sustainable food products still belong to a very small niche of consumers. For the purpose of the study, a consumer behavior model was introduced by Jager (2000) as part of his doctorate thesis. There, the factors for the consumer behavior research towards sustainable food products involves three groups of attributes, beyond the socio-demographic aspects: personal values, needs and motivations (e.g. Involvement), information and knowledge (e.g. Level of whether certainty or uncertainty) and behavioral control (e.g. PCE and product availability). 1.2.2. Knowledge gap In regard to the current knowledge gap, two main sections were identified as a challenge for deepen research and therefore understanding of the problem: behavioural aspects and geographical context. Several studies have been conducted to understand how consumers decide what to buy on a daily basis. However, more evidence is crucial for measuring the effectiveness of miscellaneous interventions within the food field (Godfray et al., 2018) to evaluate if the potential and desired changes affect either the conscious, reflective decision- making system or non-conscious, automatic processes of the consumers. Also, the role of information and knowledge about food consumption impacts into the environment and health is not clearly defined under this context and it represents an opportunity for policymakers to integrate sustainability into official guidelines and therefore measure the adoption rates of healthier and more sustainable diets (Clark, Hill and Tilman, 2018). On the other hand, most of the existing literature presents a global panorama that, although it helps to have an overall understanding of the current situation, more is needed in local terms, taking into account the different context, norms and values of the society. For the case studies, all the revised papers are developed in a global north frame where product systems and consumer segment are substantially different compared to the global south territory. Hence, the consumer preferences and willingness to pay within a developing context (i.e. Latin America) is still not very known and therefore, it is neither to clear if consumers experience a greater preference for one of the sustainable attributes (i.e. environment, health, social impact) nor the extent and form of their perceived interrelationship (Charry, A. et al, 2019). Several opportunities have been identified. However, for the purpose of this research, the following topics will be addressed: to give more insights about the consumer preferences and willingness to pay within a developing country, in a local context which is the city of Cali; additionally, evaluate the impact of information and knowledge about food consumption through an extra scenario5 that let us know the effects in trade-offs among the different attributes. 1.3. Relevance of the research 1.3.1. Scientific relevance According to Mann, D. et al (2018), the food system at a global level is the responsible for around 19- 29% of the world’s GHG emissions, and it is the major cause of deforestation. Also, it accounts for 20% of energy consumption and it is the biggest available freshwater consumer with a 70%. Last but not least, the current food system covers around 30% of all ice-free land in the world. These figures give an understanding of the level of urgency that the situation warrants. It is relevant to continue researching, either to create solutions from the supply side and from the demand side, and thus guarantee the availability and quality of food for the existing and future population, opening up a dialogue among the different stakeholders to address the ongoing food system and spur more sustainable consumption patterns, by developing improved strategies and tools to move towards the desire conditions. 5 To see the different scenarios were used for the data collection, please see the materials & methods section. To conclude, it is important to understand that most of the studies done in this field are very recent, which represents a potential to continue to strengthen through research and new data the solutions that provide more effective results for the context and studied population. 1.3.2. Social relevance Detecting ways to raise the adoption of healthier and more sustainable diets is a challenge. Nowadays, human beings frequently prefer foods with elevated levels of fats, salt and sugar, and it is easy to find them into the commercial and processed foods, that are present in large quantities in stores; additionally, these kind of products are regularly linked with poor health and massive environmental impacts (Clark, Hill and Tilman, 2018). Within the Colombian context, a developing country where the food production companies see a market opportunity to sell their products, whether they are healthy or not, this study can be helpful to understand what are the drivers to promote a more sustainable diet that helps to improve the consumers’ and the environment welfare; thus, to strengthen the policy interventions and tackle the problematic understanding the different cultural, social and economic variables. A requirement for the design of policies that promote a healthier and more environmentally friendly diet must start from a better understanding of consumer preferences for healthy food and / or due to a lower environmental impact. In this case, a choice experiment allows us to ask people if they want a type of food with certain characteristics (i.e. sustainable, social and health) compared to another type of food at a different price. 1.4. The objective of the research The aim of this project is to study the preferences of consumers in the city of Cali for food products with varying environmental, social and health attributes by using choice experiments. The study will estimate people’s Willingness To Pay (WTP) for alternative characteristics of food, and the socio- economic characteristics that determine that WTP . 1.5. Research questions 1.5.1. General Research Question For the purpose of this research, the following General Research Question (GRQ) will be answered: What are the consumer preferences regarding food products with varied attributes in the city of Cali (Colombia) and their Willingness To Pay (WTP) for those characteristics? 1.5.2. Specific Research Questions To decide on the previous GRQ, the following Specific Research Questions (SRQ) will be used: SRQ 1: To what extent is the population of Cali willing to pay for food with improved environmental friendly characteristics? SRQ 2: To what extent is the population of Cali willing to pay for food with improved health characteristics in terms of both human and animal welfare? SRQ 3: To what extent is the population of Cali willing to pay for food with a social positive impact ? SRQ 4: Which characteristics of consumers make them more prone to a sustainable diet choice in terms of environment, health and social aspects? SRQ 5: Does the knowledge of environmental, social and health impacts of the current food system have an effect on the willingness to pay for a more sustainable diet? 1.6. Hypotheses The following are the hypotheses for this study to be tested: H16: Consumers value environmental, health and social attributes related with food differently. H27: Social demographic factors such as age, gender, education level, social strata and household size have an effect on the willingness to pay for a more sustainable diet. H3: Non demographic aspects have a different effect on the willingness to pay for a more sustainable diet, where: • Positive attitude towards buying sustainable food does not have an effect on the WTP. • Perceived effectiveness (CPE) does have an effect on the WTP. H48: Additional information given about the impacts on the food system in terms of the environment, the society and the health has a bigger effect on the WTP for a more sustainable diet. 6 In response to SRQ1, SRQ2 and SRQ3 7 In response to SRQ4. Same with H3 8 In response to SRQ5 2. Materials & Methods 2.1. Discrete Choice Experiment A Discrete Choice Experiment (DCE) is one of the most used techniques for eliciting preferences with the aim to comprehend consumer demand for goods and services, commonly not traded on a defined market, and where it is not plausible to use revealed preference data on the choices made by individuals (Mangham, Hanson & McPake, 2008). This kind of experiments demand respondents to declare their choice over diverse sets of hypothetical alternatives that evoke a real-world decision. For each alternative or scenario, there are several characteristics (e.g. attributes) who guide the individual to decide for a specific alternative. The theory behind DCE relies on the random utility theory, including the assumptions of economic rationality and utility maximization (Hall et al., 2004). The main assumption is the consumers will choose the alternative that returns the highest individual benefit (i.e. utility). Moreover, as Lancaster stated (1996, p.134), “the utility yielded by an alternative is assumed to depend on the utilities associated with its composing attribute and attribute levels”. Mathematically, 𝑌𝑖𝑞 = 𝑋𝑖𝛽𝑖 + 𝜀𝑖𝑞 Where Yiq represents the utility of the consumer q for the ith alternative; Xi, is a vector of attributes for the ith alternative that goes along with a set of weights (𝛽𝑖), establishing the relative contribution of each attribute to the benefit linked with the alternative. 𝜀𝑖𝑞 equals to the unobserved variation in the characteristics of different options and any measurement errors. Each alternative is composed by different attributes at different levels. The levels should indicate the range of situations that respondents might suppose to experience. By using this approach, then we are able to outline the significance of the attributes that are related with the good or service, and the degree to which respondents are willing to trade one attribute for another one (Drummond et al. 2005). Thus, a DCE presents a logical and direct task and one which more closely matches a real-world decision (Mangham et al., 2009). Since the aim of this study is to evaluate the food preferences of consumers by calculating their willingness to pay, this is the most adequate method to do it. 2.2. Choice experiment design9 Definition of attributes, attributes levels and customization To properly define the attributes and their levels, I conducted a literature review to understand the major impacts of the food system performing and its consequences on the environment, health and social level. In addition, I held several online meetings and focus groups with the research center (CIAT) to contextualize the chosen attributes to the reality of the city of Cali were made. For the purpose of this study, the chosen attributes are as follow: Deforestation linked to food production: The land use for agricultural and livestock purposes may have an impact on biodiversity levels (amount of fauna and flora), on levels of water availability for human 9 Section based on the proposed outline in Choice Experiments for Non-Market Valuation. Francisco Alpízar, Fredrik Carlsson and Peter Martinsson. Economic Issues, Vol.8 , Part 1, 2001 consumption, in deforestation (felling of trees) and, in the carbon dioxide (CO2) emissions associated with this deforestation. Producers can reduce that impact. Among the alternatives presented to the consumer, they will have the options of choosing a meal that is potentially linked either to a low, medium or high level of deforestation depending on the food and its form of production. Thus, this first attribute contains 3 levels with the values low, medium and high. Use of agrochemicals for food production: In agriculture, agrochemicals are used for greater productiveness and less diseases and pests in food. Many times these agrochemicals, whose use is normally unavoidable, have a negative impact at different levels. In the case of this attribute, the objective is to evaluate what is most important for the consumer at the environmental and / or human health level. There are 4 levels available for this attribute, which are the following: negative impact on nature at the local level (e.g. through pollution of rivers and land), negative impact on consumer health (e.g. through the levels of toxins in the human body due to the consumption of these foods), a negative impact at both levels (i.e. nature at the local level and the consumer health) and finally, without any negative impact since agrochemicals were not used for the production of the foods contained in the meal. Meal Nutritional balance: To maintain a healthy body, the consumption of macronutrients is essential. Macronutrients are those nutrients that supply human bodies with energy and allow them to function properly. A healthy diet should contain these macronutrients, which can be classified into three categories: protein (such as eggs, all kind of meat, dairy, vegetables), carbohydrates (such as rice, potato, banana, vegetables / fruits), and fats (such as oil, avocado, nuts and seeds). For this attribute, three different levels are presented concerning the amount of macronutrients contained in the meal. The first level corresponds to a meal that contains one out of the three necessary macronutrients; the second contains two out of the three macronutrients and the third and last contains three out of the three (i.e. all) necessary macronutrients. It is important to mention this attribute represents the health category without including other nutritional factors such as micronutrients, minerals and vitamins. In order to be as simple and clear as possible with the respondents Fair Trade: Fair trade refers to an agreement where small producers or peasants receive a fair price for the products they sell, allowing them decent living conditions where there is a guarantee and protection of human rights. Additionally, it is very likely that the marketing and distribution chains will be reduced to achieve these fair prices, making the purchasing process much more local and more direct between producer and final consumer. Although in Colombia the use of labels or certificates is not common or comparable to the current development in Europe, the implementation of peasant markets, where the producer sells directly to the final consumer, and the trust guarantee labels where traders commercialize only products from farms with decent working conditions and the payment to farmers is fair, it has been growing in lately in Colombia. In the case of this social attribute, two levels are considered for the study: the meal has a fair trade guarantee or it does not have it. Monetary value: The average price of a lunch in a restaurant at Cali is around 12,000 Colombian pesos (COP). Taking into account that almost half of the population in the city belongs to the socioeconomic strata 1 (low-low) and 2 (low-medium) and, besides Colombia is a country with a highly price-sensitive population, it is important to re-evaluate the average price found because it can be considered high for almost half of the population that, socioeconomically speaking, are below the middle stratum (i.e. strata 3). By being a population sensitive to the monetary value, more levels should be considered in this attribute and be more differentiated to evaluate a true trade-off or compensation when choosing different alternatives per choice set . Thus, 7 price levels will be evaluated with a minimum price of 8,000 COP up to a maximum price of 20,000 COP with differences of 2,000 COP between each level (8,000 COP, 10,000 COP, 12,000 COP, 14,000 COP, 16,000 COP, 18,000 COP and 20,000 COP). Experimental Design To obtain the optimal combinations of the attributes and their levels and, subsequently, combine these obtained profiles into the choice sets from a statistical perspective, a technique based on the D- optimal criteria for non-linear models was considered for the evaluation of the variables. The possible combinations in this study are calculated as follows: (3^2) 𝑥( 4^1) 𝑥 (2^1) 𝑥( 7^1) = 504 Where the base corresponds to the levels of the attributes and the exponents to the number of attributes with that number of levels. This calculation is done by alternative; thus, having two alternatives of meals in this study (i.e. Dish A and Dish B), 1008 combinations of attributes and their levels are possible. Respondents cannot be exposed to so many combinations, so the four principles identified by Huber and Zwerina (1996) were considered for an efficient design of choice experiments in a non-linear model: orthogonality, level balance, minimum superposition and balance in utility. Stata was used to select the efficient combinations by using the DCREATE command who integrates three of the four principles in its calculations to obtain the most efficient ones. Utility balance was not considered as I did not have priors on the targeted parameters of the utility function. Experimental Context and survey development With the aim to minimize the level of complexity of the choice experiment and therefore reduce the risk of obtaining information that does not have the optimal quality for the data analysis, priority was given to the five attributes mentioned in the first section when initially there were nine. According to Mazotta and Opaluch (1995), including more than four or five attributes on a card (choice set) can lead to a detriment in the quality of the information collected due to the level of complexity for the respondent, creating confusion at the moment of choosing the alternatives, affecting the variance in the results. The alternatives for each choice set are two: Dish A and Dish B. The approach chosen to present these alternatives is the generic one, since respondents are less inclined to choose only because of the particular title of an alternative and therefore, this approach makes it possible to evaluate the trade- off between attributes and to be able to calculate the Marginal Rate of Substitution between them, which is one of the objectives of this study. Among the alternatives, instead of having an opt-out scenario, I included the status quo levels for each of the attributes previously described. Each respondent had 6 choice sets or cards to answer and, per scenario, there were two sets of 6 cards each, seeking to include as many efficient options as possible in the study while not increasing the complexity in the execution of the experiment. Thus, a total of 4 sets of 6 cards each were considered for the collection of information; this means that, of the possible effective combinations, 24 of them were chosen randomly for the execution of this research. Both the scenarios formulated and the questionnaire with the different cards can be seen in Annexes 1 and 2, in their original language (i.e., Spanish). Sample and sampling strategy Taking into account the technological application used for the execution of the survey by the supplier, the first filter for the sample of the population in Cali was made with its database. This database has been created with all the people who voluntarily decide to download the application and participate in surveys or market research that are carried out by LookApp and who receive financial compensation for being part of this service. Within this database, a stratified random sampling strategy was used to select sampling points, using the national socioeconomic stratification system as the stratification variable. The main criterion for selecting the sample was the representativeness by socioeconomic stratum and by age, understanding that by default all the people belonging to the LookApp community are of legal age. According to the most recent population census carried out in Cali (year 2018) by the National Department of Statistics (DANE), the population of the city reaches almost 2.4 million inhabitants and has the following distribution by socioeconomic stratum: Table 1 Cali's population by socioeconomic level Socioeconomic Number of Population(*) Index level households 1 120.101 480.404 20,2 % 2 159.981 639.924 27,0 % 3 188.652 754.608 31,7 % 4 64.222 256.888 10,8 % 5 45.390 181.560 7,6 % 6 15.907 63.628 2,7 % Total 594.253 2.377.012 100 % (*) Estimated average of 4 persons/household Using the formula presented by Louviere et al. (2000) to calculate the minimum sample of respondents and, supported by the calculations made by LookApp with a margin of error (percentage of deviation) of 5.2% and a confidence level of 95% and considering the budget constraint for the data collection (see as a legal support Annex 7) , a representative sample of 350 people is chosen. This amount was surveyed under the first scenario or general scenario; additionally, 60 people were surveyed under scenario 2 or the one that has extra information as an exploratory study. This, in order to test the effect on final consumers' Willingness To Pay when they have greater knowledge or extra information about the (negative) impacts of the food they consume today on the global food system. A total of 4920 observations were collected, 12 observations per individual. This is explained by the importance for choice experiments to analyze not only the alternative chosen by the user (in this case 6 different times) but to include in the analysis those alternatives that were not chosen (the other 6 left alternatives). Both the people surveyed in the general scenario and in the additional scenario were selected randomly within the existing database but considering that they are representative (proportional) with the socioeconomic strata presented by DANE. The sample size is divided by socioeconomic level and scenarios as follows: Table 2 Sampling size per socioeconomic level and survey scenario Socioeconomic Scenario 1: Scenario 2: Extra Ix. Per level General information socioeconomic given level 1 70 8 19,0% 2 84 17 24,6% 3 110 21 32,0% 4 38 8 11,2% 5 26 5 7,6% 6 22 1 5,6% Total 350 60 100% 2.3. Online survey (LookApp) Several sections were included in the online survey presented to the respondents. These sections are explained below: First, an initial presentation and legal consent were showed to the respondent where he/she needed to confirm his/her availability to respond the survey. Follow by that, some initial questions were posted in order to know more about their preferences and opinions about environmental, health and social aspects. Questions such as perceived changes in the temperature, healthy food concept and opinions, fair trade definition and some of their regular activities were asked. The hypothetical scenario comes after this section. Here, a description about the context of eating outside home (i.e. in a restaurant) and the characteristics they can find in the restaurant’s menu are defined. There is a general overview of the attributes they will see in the next section along with the aim of the survey. In the case of the scenario 2 where extra information is given, this section covers the main current impacts in terms of the environment, the human health and the social aspects within the global food system. Then, the main section of the survey is covered by the choice experiment where 6 cards were shown to each respondent and they needed to choose the best option for them taking into account their utility level. A summary of the attributes and their levels can be found in the following table: Table 3 Choice experiment: Attributes and levels Category Attribute Levels Description of levels Environment Deforestation linked to Three (3) Low, medium, high food production Environment, health Use of agrochemicals Four (4) Impact on nature, impact on human for food production health, impact on both, impact on none (non-use of agrochemicals). Health Meal nutritional Three (3) One, two, there (out of three) balance Social Fair trade Two (2) Yes / No Economic Monetary value Seven (7) From 8.000 to 20.000 with differences of 2.000 COP (8.000, 10.000, 12.000, 14.000, 16.000, 18.000, 20.000 COP)10 Once the cards were finished, the following section refers to a context information and knowledge about the topic, by bringing the OMS definitions for food groups and their level of health; in terms of the environment, a series of statements were evaluated using a Likert Scale in order to measure their opinions and position about non-sociodemographic aspects that might have an impact on the WTP. The final section corresponds to the sociodemographic information of each individual. Further details about the survey can be found in Annexes 1 and 2. Moreover, an Excel file with the raw data collected from the survey is attached in the Annex 3. 2.4. Statistical analysis The version 13th of Stata was used to analyse the collected data Based on the purpose of this research, a Choice Model (CM) was chosen for the results interpretation. The outcome of a CM are choices selected by a respondent (e.g. user) from a set of possible alternatives. In this case, the modelling is for discrete choices given that each individual selects a single alternative in each choice set (StataCorp, 2019). A mix logit model, also known as a mixed multinomial logit model or random-parameter logit model was used to study the collected data from the surveys. This type of model uses random coefficients to model the correlation of choices across alternatives, which relaxes the property of Independence of Irrelevant Alternatives (IIA). The model allows to include fixed coefficients as well. Here, the variable of price is managed as an independent variable with non-random coefficient while the rest of attributes from the choice experiments are managed as an independent variables with random coefficients. The reason of having a fixed parameter for the “price” attribute relies on the reason to ensure the estimated coefficient will always be negative. For valuation purposes, a reduction in the income resulting from goods with higher prices should lead to a reduction in the utility level of the individual. As a random parameter formulation using either a logistic or normal distribution goes from minus to plus infinity, it might be possible that the coefficient will change signs for some respondents and this is not ideal for the study; thus, it should be fixed (i.e., non-random). It is important to mention that the choice made by the respondent is the dependent variable. In order to treat the variables as 10 Currency exchange, 1 € = 4.177,20 COP (January 15th 2020) discrete and not continuous, the model starts with “xi”, indicating to Stata the type of variables included within it. Due to the decision of use generic alternatives for this research, all the individual-specific variables were including in the model using the conditional if. When a variable is not alternative-specific, such as the attributes used in the CE, but individual-specific, such as the sociodemographic variables or statements that do not change over each alternative but over each individual, the model does not run with these variables and series of conditional were included to evaluate the WTP based on the chosen condition. Once the mixlogit model was developed, the use of the command nlcom was used to calculate the Marginal Willingness To Pay (MWTP) under the scenario of the model. Nlcom stands for nonlinear combination of estimators and it shows the relation between the attribute coefficient with the price attribute coefficient. It computes point estimates, standard errors, test statistics, significance levels and confidence internvals for (possibly) nonlinear combinations of parameter estimates after any Stata estimation command (e.g. mixlogit) (StataCorp, 2013). A detailed process with the Stata coding, the complete variables dictionary and executed codes can be seen in the Do-File (Annex 4). In addition to this Annex 4, the Annex 5 comprises a series of screenshots with all the models considered in this research. The variables that are statistically significant are those whose p values are between 0% and 10% (i.e. 0.000 < (P>z) > 0.100). In relation to the coefficients, the initial consideration to be analysed from them are the different signs. When the coefficients are negative, then those decrease the probability of choosing the alternative. Conversely, positive coefficients mean those attributes increases the probability of choosing the alternative. What it is usually expected is to have a negative coefficient for the cost or price attribute, since individuals prefer to expend less money or to have a better cost/benefit ratio due to their utility level that decreases with the money expenditure and due to a budget constraint. 3. Results 3.1. General results The consumers’ preferences for food in the city of Cali vary depending on the attributes and their (non)sociodemographic characteristics. Thus, the MWTP varies over these different categories. To answer the general question of the research, the subsequent paragraphs are answering the results from the Specific Research Questions (SRQ). For further details about the Stata results, please see Annex 5. Regarding the initial three SRQ, annex 5.1 comprises the general results from the mixlogit model. As it is shown, the p values from the different attributes of the choice experiment are statistically significant for the model11 (Mean section) but the result of the attribute agrochhum (p value of 0,962). In this case, this attribute should be excluded from the model because it is highly insignificant to the model and it does not help to understand the decision made by the respondent. Regarding the others, those attributes help to understand the decision made by the respondents. The mixlogit model shows a negative coefficient for the price, which complies with the economic theory. 11 Statistically significant with a maximum value of 10%. This means the results from the sample can be extrapolated to the total universe, or explain the behavior from the population; thus, the study is significant to make inferences about the population in Cali. With the attributes statistically significant for the model, it is feasible to calculate to what extent the population of Cali is willing to pay for food with improved environmental friendly, human health and social positive impact characteristics. Because the coefficient of the attribute “price”, although is different from zero, is not salient enough (-0,0000279126), the Willingness To Pay calculated using the nlcom command is very high and therefore unreliable (e.g., respondents willing to pay almost 50% extra of the total price of the dish for just one of the attributes). Unfortunately for this research, this result inhibits the interpretation of the WTP estimates in absolute terms, as they are all overestimations of the true WTP. Instead, we shall limit the discussion of these results to ordinal comparisons (i.e., ranking preferences) rather than cardinal (considering both the amount or value, and ranking or order ). A comparison among coefficients is correct because the nlcom produces a ratio between coefficients that takes out the inherent scale from each regression. As a result, the unit of measurement is the same for all the results and it is feasible to compare these ratios and identify an order (i.e., ordinal analysis). Regarding the (non)sociodemographic factors, there are noticeable differences when the model is calculated under each situation; in other words, using the conditional if to evaluate every (non)sociodemographic factor in the subsamples, the attributes are valued differently and there are cases when the attribute price is not statistically significant for the subsample, which disables the MWTP calculation. Nonetheless, it is possible to determine certain preferences from the consumers in relation to the dish attributes. Further details can be found in Chapter 3.3. and the Annexes from A5.3 to A5.24. As an exploratory activity, a subsample of N=60 respondents received extra information about the current impacts in the global food system in terms of environmental, social and human health related factors. Price does not play a role when people know about the current impacts and the relevance of the remaining attributes change compared to the general model. Since the MWTP is not possible to be calculated, then the hypothesis cannot be solved but differences can be identified. In the Annexes A5.25 and A5.26 are the results from Stata. 3.2. Environmental, health & social attributes for food consumers agrochnone def_low fairtrade def_medium nut3 nut2 agrochenv Figure 1 Marginal Willingness To Pay for the general mixlogit model. From figure 1, the defined environmental characteristics are in green; for the case of human health, the blue attributes represent the category. A particular attribute that combines both human health and environment characteristics (i.e. agrochnone) is coloured in yellow. Last but not least, social categories are coloured in orange. Since the p value for the attribute agrochhum is not statistically significant, then the MWTP is not calculated. These results show how much the consumers are willing to pay for an extra unit of the attribute. The base is the initial level of each attribute. People valued environmental, human health and social characteristics related with food differently. Within the environmental category, the lower the deforestation level of the dish, the higher the preference of the consumer for choosing that dish. People prefer dishes with low deforestation level than dishes linked with medium deforestation levels. In the case of agrochemicals, the base is the agrochemicals used for food production that have an impact on both: human health and the environment. In the case of agrochenv, consumers prefer to consume food with this attribute instead of having agrochemicals in their dishes impacting both the health and the nature; so, it does cancel any effect on the human health, remaining the part of the environment. That is the reason why this is considered as part of the health characteristics and coloured in blue since consumers thought about human health over the environment. When the idea is to remove completely the effect of agrochemicals (by not using them) and therefore suppressing the effect on the environment and the human health, the preference is higher, so much that it turns to be the attribute with the highest preference for the respondents when choosing between the different alternatives. Now, in relation to the other human health attributes, the amount of macronutrients in the dish is important for the consumers. The more nutrients a dish has, the higher the preference for the consumers to choosing that option. Finally, fair trade (i.e. fairtrade) is the top 3 characteristic among the respondents, representing the social factor within this choice experiment. As it is seen, consumers are willing to pay for more sustainable dishes, but they valued every attribute and therefore, every category12, differently. It can be deduced that individuals might have a ranking of importance based on the MWTP results, where the removal of agrochemicals for the food production is the most valuable attribute, followed by the idea of having a low level of deforestation instead of a high one linked to the food and, within the top three, the dish with a guarantee of Fairtrade for its production. Attributes related with human health are the least preferred when deciding for food products. 3.3. Effect of social demographic factors in a more sustainable diet choice Six sociodemographic factors were included in the online survey in order to determine if they have some effect on the choice made by respondents. These factors are as follows: gender, socioeconomic level (i.e. stratum), education level, ethnicity, household composition and age. Considering gender, there are two extra models as follow: mixlogit model with only female observations and mixlogit model with only male observations. Due to the P value of the price attribute in the “only male” model (P value = 0,75), the MWTP calculation is not coherent to do it because of the weak results of the estimation. In this case, a comparison between the general model and the “only female” model in relative terms is made. As explained at the beginning of this chapter, although the 12 Environmental, social and human health actual number in absolute terms for MWTP does not make sense, it is possible to compare female respondents to the general population in terms of preferences order. Figure 2 shows women rank in a similar way their preferences for the attributes compared to the total population. Two remarks are important for this sociodemographic factor: first, all MWTP are lower for female than the total sample. So, even when only women are also willing to pay for more sustainable dishes, higher values can be seen when modelling for all the observations. Second, fair trade is less important to only women and deforestation, both levels, are in the top three of their preferences along with the agrochnone attribute which continues being the most preferred attribute. agrochenv nut2 nut3 def_medium fairtrade def_low agrochnone Female General Figure 2 Marginal Willingness To Pay (MWTP) for female gender compared to the MWTP of the general model Continuing with the socioeconomic level, a MWTP’s comparison between all the variables is not possible because of the P values results from all the sectors are statistically insignificant (P value low level = 0,538, P value medium level = 0,140 and P value high level = 0,402). However, a marginal effects interpretation is possible with those attributes statistically significant for the model. Table 4 reflects the variables considered for the marginal effects interpretation. Table 4 Relevant variables depending on the socioeconomic level Socioeconomic level Statistically significant variables Low (1 & 2) def_medium, def_low, fairtrade, nut3, agrochenv, agrochnone Medium (3 & 4) def_medium, def_low, fairtrade, nut2, nut3, agrochenv, agrochnone High (5 & 6) def_medium, def_low For the respondents who belong to the low socioeconomic level, a change in the deforestation level from high to medium increases the probability to choose a dish in 45%. When the change goes from high to a low level of deforestation, then the probability of choosing a dish increases and it equals 81%. A dish with a fairtrade guarantee increases the probability to choose it in 47%. Another relevant attribute is agrochnone for the low socioeconomic level respondents. Compared with a dish using agrochemicals with an impact on both, the environment and the human health, a dish without any agrochemical increases the probability of choosing that option in 79,58%. Regarding the respondents within the medium socioeconomic level, agrochnone increases the probability of choosing a dish in 97% when the food in the dish is made with agrochemicals that affect both the environment and the human health. Fairtrade also plays an important role within this level. By having a dish with a fairtrade guarantee increases the probability of choosing a dish in 62%. To finalize with the socioeconomic level, for the consumers with a high socioeconomic level deforestation is the only attribute relevant when deciding for a dish. People were asked about their education level. For the purpose of this research and looking to have more significant subsamples, three categories were made: people with studies until primary school and high school belong to the category of school. For the case of technical/technological, it remains the same; and, bachelor and postgraduates studies, the category created was University. The P value from consumers with university studies is statistically significant (P value = 0,079) and the MWTP can be seen in Figure 3. The other two categories present not significant P values so a calculation is not possible (P value school = 0,267 and P value technical/technological = 0,613). In the case of university, deforestation gets the highest MWTP when the level is low; after a low deforestation, omitting agrochemicals is the second most valued attribute and near to it is the nutritional balance when the dish, instead of having 1 macronutrient, has all of them. In comparison to the general model, the MWTP of the human health attributes (i.e., nut2 and nut3) is higher for people with university studies than all the respondents. Even when agrochnone is not the most important attribute for them, it is still part of the top 3. For the remaining attributes, the MWTP are higher in the general model than this specific one. General University Figure 3 Marginal Willingness To Pay (MWTP) for consumers with university studies compared to the MWTP of the general model Next, the ethnicity was taken into account. Three different ethnic groups are present in the city of Cali: Black, afro-colombian, raizal and palenquera group as the main one; then, there are indigenous people and gypsy in a less percentage. All of these three ethnicities were grouped into the category “ethnic”. The rest of the population do not belong to an ethnic group. For this particular factor, the price attribute is statistically significant (P value = 0,069) just for those who do not belong to an ethnic group. Special attention is made in the case of the ethnic group, since most of the attributes are not statistically significant, except for those related with agrochemicals. agrochenv nut2 nut3 No ethnic def_medium General fairtrade def_low agrochnone Figure 4 Marginal Willingness To Pay (MWTP) for consumers who do not belong to any ethnicity compared to the MWTP of the general model Figure 4 displays for almost all of the attributes the MWTP is higher in the general model compared to the model without observations of respondents who belong to an ethnical group but nut2. In other words, it is most important for this group a dish with two macronutrients instead of one than for all the respondents. Meals without the use of agrochemicals, along with a low level deforestation and a fair trade guarantee are the most preferred characteristics when choosing the dishes for this subgroup. Now, in relation to the household composition, three categories are analyzed: people living alone, people living with adults and people living with sensitive people which means people under legal age (< 18 years old) and elderly people (> 65 years old). Due to a small subsample of people living alone, a model is not possible to run. Regarding the other two groups, the price attribute for those living in a sensitive household is highly insignificant to make the choice (P value = 0,65). Thus, the MWTP is just relevant to be calculated, in relative terms, for the people living with adults only. Figure 5 shows a comparison of these results and the general model. As it is seen, the ranking of preferences for the people living with adults is similar to the female group, where agrochnone is the most relevant characteristic followed by a low and medium deforestation levels. The general model shows higher MWTP than the model of household composition of just adult population. General hh_adults Figure 5 Marginal Willingness To Pay (MWTP) for consumers who are living with adults compared to the MWTP of the general model For the last socioeconomic factor of this study, the age was another factor with interesting results. The sample was divided into four different age groups: the young people (i.e., from 18 to 24 years old); then, young adults (i.e., from 25 to 34 years old). The biggest group is for the adults (i.e., from 35 to 54 years old). The last group corresponds to the old adults, over 55 years old. None of the models based on age have statistically significant the price variable (P value young people = 0,899, P value young adults = 0,271, P value adults = 0,249, P value old adults not possible to calculate due to the very small sample); therefore, a MWTP calculation is not considered and the age is not relevant when deciding for an alternative. 3.4. Effect of non-demographic factors in a more sustainable diet choice Two non-demographic factors were asked in the survey using the Likert Scale methodology: positive attitude and Consumer Perceived Effectiveness (CPE). Up to now, those factors are two of the most mentioned ones in the developed behavioural economics literature. In the case of positive attitude, the following statement was evaluated using a 5-point scale, going from totally disagree to totally agree; “If a food product seems to be more sustainable and environmental friendly, I will consider in buying it instead of other because it is better”. The analysis made considers the people with a positive attitude when they were agree or totally agree on the statement. The hypothesis for this non-socio demographic factor states there is not an effect on the MWTP. Nevertheless, figure 6 shows people with positive attitude have a preference for more sustainable food, even when their MWTP are lower compared to the general model. The environmental factors (i.e., def_low, def_medium) are the two most preferred attributes for choosing a dish and the human health related attributes (i.e., nut2, nut3, agrochenv) are the least preferred ones. General Possitive attitude Figure 6 Marginal Willingness To Pay’s comparison between the general mixlogit model and people with positive attitudeAbout the CPE, this was the statement used with the same 5-point scale as the positive attitude factor: “My behaviour and actions when consuming and buying food can definitely contribute to the care and improvement of the environment”. For this factor, the initial assumption is that it does have an effect on the WTP. Unfortunately, even when the P values of the price variable are statistically significant (P value CPE+ = 0,004 and P value CPE-=0,018), the coefficient is positive for the subsample with a disagreeability on the CPE. agrochnone def_low fairtrade def_medium nut3 nut2 agrochenv General Possitive attitude CPE + Figure 7 Marginal Willingness To Pay’s comparison between the general mixlogit model and people with positive attitude and positive CPE 3.5. Effect of information level in a more sustainable diet choice Two different scenarios were used with the sample. The base scenario (N=350 respondents) did not include additional information about the current impacts on the environment, the human health and the social aspects in the food system. On the contrary, 60 respondents received extra information with relevant examples of the impacts with the idea to evaluate if the knowledge, in this case the extra information given, does have an effect on the MWTP. Unfortunately, a conclusion is not possible to make in this case, due to a lack of significance of the P values when doing the mixlogit model (P value extra info = 0,611 and P value without extra info = 0,404). 4. Discussion 4.1. Findings versus literature review Consumer behavior can be understood as a decision on how to distribute available resources (i.e., money, effort, time) in consumption-related products. It covers what, where, when, why and how often they buy the products, as well as their own evaluation once they bought it and use it, and the way to dispose it (Schiffman & Kanuk, 2010). According to the existing literature, a growing concern regarding the environmental issues and topics related with the human health has been generating the emergence of a new consumer niche in the market, where they are demanding new types of products and, in most of the cases, these demands are reflected in a premium price for more sustainable products (Finisterra, Barata & Leal, 2009). This behaviour has been strengthened worldwide due to an income level increase and also because of a rising demand for these type of products (Salgado, Subirá & Beltrán, 2009). Taking into account the results from the first hypothesis, it is clear the consumers are willing to pay for a more environmental friendly, healthier and with a positive social impact dish. Indeed, their behavior shows a trend for preferring more sustainable meals. However, the ranking of preferences sometimes differ from the literature, due to an initial understanding that health attributes are often the most valued characteristics rather than the environmental and the social ones. Health attributes are classified in this study in the last part of the consumer ranking, while the environment and the social impact are leading it. This finding contradict the literature, and further research is recommended in order to understand if the trend is common in other communities or contexts within Colombia or the region. Regarding the socio-demographic factors, a series of studies have carried out showing interesting results. Regarding age, there is not a consensus about the effect on food consumption with environmental friendly attributes. Some researchers suggest there is a positive relationship between WTP and age (Padel and Foster, 2005; Roitner-Schobesbrger et al., 2008 cited by Shafie and Rennie, 2012). On the other hand, different studies show a negative correlation between both variables; thus, the younger the person, the more prone is to buy sustainable products (Yiridoe, Bontiankomah and Martin, 2005). In the case of this study, a conclusion cannot be made about age since the factor did not play a role (i.e., it was not statistically significant) when choosing the different alternatives. In the case of education level and socioeconomic level (linked with the income level), the research does not show any particular result compared to the general model and the absolute MWTP does not make sense due to the very small price coefficient. Nonetheless, and based on previous studies, education level and income level are factors that have been increasing attention when analyzing the consumer behavior and how they are buying products with a lower environmental impact. Consumers with a higher education level and therefore a better and broader access to information, are expected to act in a more conscious way respect to the environment (Finisterra et al., 2009). According to Chekimaa et al. (2016), the factors that push the purchase of more sustainable products are much more in the case of individuals with a high education level. This is confirmed by several studies where a positive relation between education variable and environmental friendly attitudes are tested and confirmed (e.g., Jaramillo, Vargas and Guerrero, 2015; Roberts, 1996;Zimmer, 1994). When considering the income level, there is a positive effect on the consumption of products with environmental characteristics. This is supported by the statement that an individual with a higher income level can overcome easily the premium price linked to these type of food products (Straughan and Roberts, 1999; Villanueva, Vargas López and Guerrero, 2015). However, the income level tend to be correlated with the amount of these goods rather than the willingness to really access to them; thus, households with a high income level do not necessary show a higher purchase level (Fotopoulos and Krystallis, 2002). Finally, there are some studies with the aim to validate the relation between the amount of children in the household and the environmental behavior, showing a positive relationship between them (Fotopoulos and Krystallis, 2002; Thompson and Kidwell, 1998). Grunert (1991) found out that household size has a positive effect on environmental awareness and positive attitude towards environmental quality. Unfortunately, this MSc thesis do not support the existing literature and a bigger sample should be considered in order to obtain more reliable conclusions. 4.2. Strengths of the study In terms of the methodology used, including the experimental design and the statistical analysis, this study presents a strong statistical model (i.e., mixlogit) when conducting a choice experiment research. The mixlogit model is one of the most advanced and well developed models to analize consumers’ choices. In this regard, it is positive to have worked with this model and include a very detail questionnaire who allows the CIAT and future studies to have a solid database to propose strategies to the policy makers in order to propel a more sustainable food system, in this case, from a demand point of view. Now, when contemplating the knowledge gap, an reduction on it has been made in the case of having now more insights about the consumer preferences and willingness to pay within a developing country as well as having information of a local context study rather than global or regional data. On the contrary, in the case of evaluating the impact of knowledge about food consumption impacts when choosing for the alternatives, this gap could not be minimized with this study and it is part of the improvements for further researches. 4.3. Weaknesses of the study When focusing on the methodology used, one of the biggest opportunities of this study relies on the lack of time and space due to the external circumstances of having face to face focus groups prior the launching of the questionnaire to define attributes and levels, particularly with the price attribute. The coefficient of this specific attribute was not salient enough to the respondent choice. This could mean many things, but most importantly it shows that price concerns were not central to the decision to choose one or the other dish. This finding hints to the saliency of the environmental, health and social dimensions faced by the respondents. If this study would have the possibility of having a bigger sample size (i.e., bigger budget) and the feasibility of collecting the data in the field (i.e., not online), the lack of saliency for the consumers would have noticed. Unfortunately, the COVID-19 conditions also did not allow us to conduct the survey face to face; instead, an online survey was developed and some of the behaviors and way to respond of the consumers are missing for the purpose of this investigation. So, the way they interpreted the questions and possible biases due to the “social desirability” or the “right answer” when they were answering (i.e., answer socially accepted instead of their own opinion or preference) affect the confidence level when performing online surveys. 5. Conclusions & Recommendations The main objective was to study the preferences of the consumers in the city of Cali for food products regarding several environmental ,social and health characteristics by using choice experiments. Consumers value environmental, health and social attributes related with food differently. Agrochemicals (with an impact on both the health of the environment and the human population) along with environmental factors such as deforestation level are the most important attributes when choosing food products. Human health factors are also part of the preferences of consumers but in a less importance. Social demographic factors such as age, gender, education level, social level (i.e., stratum) and household size do have an effect on the consumers ‘interests and therefore their MWTP for a more sustainable diet. Even when an absolute analysis is not possible due to the coefficient of the price attribute, a ranking of preferences is identified depending on the different subgroups according to these factors. In the case of non-demographic aspects (i.e., positive attitude and CPE), both concepts do have an effect on the MWTP, showing that people with a positive attitude have a higher MWTP compared to the people with a positive Consumer Perceived Effectiveness. To conclude with the initial hypotheses, a conclusion of the impact of additional information given about the impacts in the food system in terms of the environment, the society and the health is not possible to do due to the weak results of the price coefficient when performing the mixlogit model for this subsample. This study brings new knowledge in terms of choice experiment case studies under the developing country and local context level and further studies with a significant sample should be considered to understand the role of knowledge and extra information on the consumers’ purchase decisions in relation to food products. Further studies should considered, along with a bigger sample size, the idea to elaborate surveys face to face, an increase the amount of focus groups when evaluating the price ranges. Thus, an absolute terms analysis could be made. By knowing the ranking of preferences by sociodemographic factor, or considering what is more relevant for consumers in terms of environmental, social and health aspects, policy makers can develop better and more accurate communication strategies to boost the needed change in the current food system and pave the way to a more sustainable food system. 6. References Anja Kollmuss & Julian Agyeman (2002) Mind the Gap: Why do people act environmentally and what are the barriers to pro-environmental behavior?, Environmental Education Research, 8:3, 239-260, DOI: 10.1080/13504620220145401 Bajželj, B et al. (2014). Importance of food-demand management for climate mitigation. Nature Climate Change, 4(10):924-929. Charry, A., Narjes, M., Enciso, K. et al. Sustainable intensification of beef production in Colombia— Chances for product differentiation and price premiums. Agric Econ 7, 22 (2019). https://doi.org/10.1186/s40100-019-0143-7 Chekimaa, B., Wafaa, S. A. W., Igaua, O. A., Chekimab, S., & Sondoh, S. L. (2016). Examining green consumerism motivational drivers: does premium price and demographics matter to green purchasing? Journal of Cleaner Production, 12(4), 3436–3450. Clark, M., Hill, J. and Tilman, D., 2018. The Diet, Health, and Environment Trilemma. Annual Review of Environment and Resources, 43(1), pp.109-134. Clark, M., Hill, J. and Tilman, D., 2018. The Diet, Health, and Environment Trilemma. Annual Review of Environment and Resources, 43(1), pp.109-134. Drummond MF, Sculpher MJ, Torrance GW, O’Brien BJ, Stoddart GL. 2005. Methods for the economic evaluation of health care programmes. Third edition. Oxford: Oxford University Press Dupuis, E. (2000), ‘‘Not In My Body: rBGH and the Rise of Organic Milk,’’ Agriculture and Human Values 17(3), pp. 285–295. FAO. (2016). The future food system: the world on one plate? | FAO. Retrieved 6 January 2021, from http://www.fao.org/cfs/home/blog/blog-articles/article/en/c/448182/ Finisterra, A., Barata, M. L., & Leal, W. (2009). Identifying the green consumer : A segmentation study. Journal of Targeting, Measurement and Analysis for Marketing (2009), 17(1), 17–25. http://doi.org/10.1057/jt.2008.28 Fotopoulos, C., & Krystallis, A. (2002). Purchasing Motives and Profile of the Greek Organic Consumer: A Countrywide Survey. British Food Journal, 104(9), 730–765. http://doi.org/10.1108/00070700210443110 Giddens, A. (1991), Modernity and Self-identity: Self and Society in the Late Modern Age, Palo Alto: Stanford University Press. Godfray, H. C. J., et al. (2018). Meat consumption, health, and the environment. Science.361(6399) Godfray, H. C. J., et al. (2018). Meat consumption, health, and the environment. Science.361(6399) Godfray, H., Beddington, J., Crute, I., Haddad, L., Lawrence, D., Muir, J., Pretty, J., Robinson, S., Thomas, S. and Toulmin, C., 2010. Food Security: The Challenge of Feeding 9 Billion People. Science, 327(5967), pp.812-818. Grunert, S. C. (1991). Everybody Seems Concerned About the Environment: But Is This Concern Reflected in (Danish) Consumers’ Food Choice? European Advances in Consumer Research, 1, 428– 433. Retrieved from http://acrwebsite.org/volumes/11481/volumes/e01/E-01 Hall J, Viney R, Haas M, Louviere JJ. 2004. Using stated preference discrete choice modelling to evaluate health care programs. Journal of Business Research 57: 1026–32. Jager, W. (2000), Modelling consumer behavior, PhD thesis, University of Groningen, Groningen. Jaramillo, J. L., Vargas, S., & Guerrero, J. de D. (2015). Preferencias de consumidores y disponibilidad a pagar por atributos de calidad en carne de conejo orgánico. Revista Mexicana De Ciencias Pecuarias, 6(2), 221–232. Lancaster K. 1966. A new approach to consumer theory. The Journal of Political Economy 74: 132–157. Lindsay J Mangham, Kara Hanson, Barbara McPake, How to do (or not to do) … Designing a discrete choice experiment for application in a low-income country, Health Policy and Planning, Volume 24, Issue 2, March 2009, Pages 151–158, https://doi.org/10.1093/heapol/czn047 Louviere J, Hensher DA, Swait JD. Stated choice methods: analysis and application. Cambridge: Cambridge University Press; 2000. Mangham, L., Hanson, K., & McPake, B. (2008). How to do (or not to do) … Designing a discrete choice experiment for application in a low-income country. Health Policy And Planning, 24(2), 151-158. doi: 10.1093/heapol/czn047 Mann, D., et al. (2018). Australian consumers’ views towards an environmentally sustainable eating pattern. Public Health Nutrition, 21(14), 2714-2722. Mann, D., et al. (2018). Australian consumers’ views towards an environmentally sustainable eating pattern. Public Health Nutrition, 21(14), 2714-2722. Mazzotta, M., & Opaluch, J. (1995). Decision Making When Choices Are Complex: A Test of Heiner's Hypothesis. Land Economics, 71(4), 500-515. doi:10.2307/3146714 Reardon, T., & Timmer, C. (2012). The Economics of the Food System Revolution. Annual Review Of Resource Economics, 4(1), 225-264. doi: 10.1146/annurev.resource.050708.144147 Roberts, J. A. (1996). Green Consumers in the 1990s: Profile and Implications for Advertising. Journal of Business Research, 36(3), 217–231. Retrieved from http://www.sciencedirect.com/science/article/pii/0148-2963(95)00150-6 Salgado, L., Subirá, M. E., & Beltrán, L. F. (2009). Consumo orgánico y conciencia ambiental de los consumidores. Revista Latinoamericana de Economia, 140(157), 190–199. Schiffman, L. G., & Kanuk, L. L. (2010). Comportamiento del Consumidor 10ed. (G. Dominguez, Ed.). Mexico. Shafie, F. A., & Rennie, D. (2012). Consumer Perceptions Towards Organic Food. Procedia - Social and Behavioral Sciences, 49, 360–367. http://doi.org/10.1016/j.sbspro.2012.07.034 Smith P., M. Bustamante, H. Ahammad, H. Clark, H. Dong, E.A. Elsiddig, H. Haberl, R. Harper, J. House, M. Jafari, O. Masera, C. Mbow, N.H. Ravindranath, C.W. Rice, C. Robledo Abad, A. Romanovskaya, F. Sperling, and F. Tubiello, 2014: Agriculture, Forestry and Other Land Use (AFOLU). In: Climate Change 2014: Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Edenhofer, O., R. Pichs-Madruga, Y. Sokona, E. Farahani, S. Kadner, K. Seyboth, A. Adler, I. Baum, S. Brunner, P. Eickemeier, B. Kriemann, J. Savolainen, S. Schlömer, C. von Stechow, T. Zwickel and J.C. Minx (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA. StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP. StataCorp. 2019. Stata: Release 16. Statistical Software. College Station, TX: StataCorp LLC. Straughan, R. D., & Roberts, J. A. (1999). Environmental segmentation alternatives: a look at green consumer behavior in the new millennium. Journal of Consumer Marketing, 16(6), 558–575. http://doi.org/http://dx.doi.org/10.1108/07363769910297506 Thompson, G., & Kidwell, J. (1998). Explaining the Choice of Organic Produce: Cosmetic Defects, Prices, and Consumer Preferences. American Journal of Agricultural Economics, 80(2), 227–287. http://doi.org/10.2307/1244500 Tilman, D. and Clark, M., 2014. Global diets link environmental sustainability and human health. Nature, 515(7528), pp.518-522. UN, 2019. Peace, Dignity And Equality On A Healthy Planet. [online] United Nations. Available at: [Accessed 6 April 2020]. Vermeir, I. and Verbeke, W., 2006. Sustainable Food Consumption: Exploring the Consumer “Attitude – Behavioral Intention” Gap. Journal of Agricultural and Environmental Ethics, 19(2), pp.169-194. Villanuevaa, J. L., Vargas López, Samuel Guerrero, J. de D. (2015). Behavior of consumers and willingness to pay for quality attributes of organic meat rabbit. Revista Mexicana De Ciencias Pecuarias, 6(1), 221–232. Weatherell, C., A. Tregear, and J. Allinson (2003), ‘‘In Search of the Concerned Consumer: UK Public Perceptions of Food, Farming and Buying Local,’’ Journal of Rural Studies 19(2), pp. 233–244. Yiridoe, E. K., Bontiankomah, S., & Martin, R. C. (2005). Comparison of consumer perceptions and preference toward organic versus conventionally produced foods : A review and update of the literature. Renewable Agriculture and Food Systems, 20(4), 193–205. http://doi.org/10.1079/RAF2005113 Zimmer., M. R. (1994). Green issues: Dimensions of environmental concern. Journal of Business Research, 30(1), 63–74. Retrieved from http://dx.doi.org/10.1016/0148-2963(94)90069-8 7. Annexes Annex 1: Online survey template (general scenario) 1. Online survey_General scenar io case.docx Annex 2: Online survey template (second scenario) 2. Online survey_Second scenar io case.docx Annex 3: Raw data results 3. Food choices Cali STATA.xlsx Annex 4: Stata Do file & Stata database Annex 5: Stata screenshots models A5.1 Mixlogit model for both scenarios (Sample N=410). General case . xi: mixlogit choice price, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1540.6777 (not concave) Iteration 1: log likelihood = -1535.8751 (not concave) Iteration 2: log likelihood = -1534.1575 Iteration 3: log likelihood = -1521.5118 Iteration 4: log likelihood = -1520.0024 Iteration 5: log likelihood = -1519.9343 Iteration 6: log likelihood = -1519.934 Iteration 7: log likelihood = -1519.934 Mixed logit model Number of obs = 4920 LR chi2(8) = 44.44 Log likelihood = -1519.934 Prob > chi2 = 0.0000 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0279176 .0162301 -1.72 0.085 -.059728 .0038927 def_medium .5199561 .1505636 3.45 0.001 .2248569 .8150553 def_low .7948398 .1405763 5.65 0.000 .5193153 1.070364 fairtrade .523194 .0778615 6.72 0.000 .3705882 .6757998 nut2 .2444896 .1068176 2.29 0.022 .035131 .4538483 nut3 .386995 .0957979 4.04 0.000 .1992346 .5747554 agrochhum -.0041142 .0866775 -0.05 0.962 -.173999 .1657705 agrochenv .2322625 .0964599 2.41 0.016 .0432046 .4213203 agrochnone .8200429 .1353858 6.06 0.000 .5546915 1.085394 SD def_medium -.0188374 .2229673 -0.08 0.933 -.4558453 .4181706 def_low .7489196 .1323445 5.66 0.000 .4895291 1.00831 fairtrade .520105 .1056495 4.92 0.000 .3130358 .7271743 nut2 -.5694404 .1343332 -4.24 0.000 -.8327285 -.3061522 nut3 -.2859985 .2283372 -1.25 0.210 -.7335312 .1615342 agrochhum -.0048955 .1789061 -0.03 0.978 -.355545 .345754 agrochenv .1445494 .2197746 0.66 0.511 -.2862009 .5752996 agrochnone .8691337 .1670506 5.20 0.000 .5417205 1.196547 A5.2 WTP’s calculation using the nlcom command with the results from A5.1 . nlcom -(_b[def_medium]/_b[price]) _nl_1: -(_b[def_medium]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 18.62464 8.634962 2.16 0.031 1.700425 35.54886 . nlcom -(_b[def_low]/_b[price]) _nl_1: -(_b[def_low]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 28.47088 13.69314 2.08 0.038 1.63282 55.30894 . nlcom -(_b[fairtrade]/_b[price]) _nl_1: -(_b[fairtrade]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 18.74062 10.65682 1.76 0.079 -2.146366 39.62761 . nlcom -(_b[nut2]/_b[price]) _nl_1: -(_b[nut2]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 8.757531 6.146001 1.42 0.154 -3.28841 20.80347 . nlcom -(_b[nut3]/_b[price]) _nl_1: -(_b[nut3]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 13.86202 8.562773 1.62 0.105 -2.920704 30.64475 . nlcom -(_b[agrochhum]/_b[price]) _nl_1: -(_b[agrochhum]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 -.1473707 3.128988 -0.05 0.962 -6.280075 5.985334 . nlcom -(_b[agrochenv]/_b[price]) _nl_1: -(_b[agrochenv]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 8.319558 5.174093 1.61 0.108 -1.821477 18.46059 . nlcom -(_b[agrochnone]/_b[price]) _nl_1: -(_b[agrochnone]/_b[price]) choice Coef. Std. Err. z P>|z| [95% Conf. Interval] _nl_1 29.37364 15.50233 1.89 0.058 -1.010371 59.75766 A5.3 Model when the gender is male . xi: mixlogit choice price if gend==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -756.42631 (not concave) Iteration 1: log likelihood = -755.99094 (not concave) Iteration 2: log likelihood = -754.61275 (not concave) Iteration 3: log likelihood = -751.24901 Iteration 4: log likelihood = -747.96164 Iteration 5: log likelihood = -743.09542 Iteration 6: log likelihood = -742.6545 Iteration 7: log likelihood = -742.65185 Iteration 8: log likelihood = -742.65185 Mixed logit model Number of obs = 2388 LR chi2(8) = 29.49 Log likelihood = -742.65185 Prob > chi2 = 0.0003 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0075629 .0237423 -0.32 0.750 -.054097 .0389712 def_medium .4033097 .2241792 1.80 0.072 -.0360735 .842693 def_low .6547298 .2115545 3.09 0.002 .2400906 1.069369 fairtrade .5194323 .1156613 4.49 0.000 .2927403 .7461242 nut2 .2241918 .1591691 1.41 0.159 -.0877739 .5361574 nut3 .2789992 .14335 1.95 0.052 -.0019617 .5599602 agrochhum -.1355763 .1280258 -1.06 0.290 -.3865023 .1153496 agrochenv .2267351 .1427119 1.59 0.112 -.0529751 .5064452 agrochnone .692287 .2041713 3.39 0.001 .2921186 1.092455 SD def_medium .0853465 .4301774 0.20 0.843 -.7577858 .9284789 def_low .7805857 .1868661 4.18 0.000 .4143349 1.146837 fairtrade .6351979 .1490049 4.26 0.000 .3431536 .9272422 nut2 .6370156 .1938962 3.29 0.001 .256986 1.017045 nut3 -.345004 .3406849 -1.01 0.311 -1.012734 .3227261 agrochhum .0438645 .2531137 0.17 0.862 -.4522292 .5399582 agrochenv -.1477062 .2447192 -0.60 0.546 -.627347 .3319346 agrochnone 1.052555 .2457104 4.28 0.000 .5709712 1.534138 A5.4 Model when the gender is female . xi: mixlogit choice price if gend==0, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -781.25204 (not concave) Iteration 1: log likelihood = -780.65498 (not concave) Iteration 2: log likelihood = -780.14994 (not concave) Iteration 3: log likelihood = -779.44622 (not concave) Iteration 4: log likelihood = -778.33439 (not concave) Iteration 5: log likelihood = -776.7688 Iteration 6: log likelihood = -774.42352 Iteration 7: log likelihood = -774.30462 Iteration 8: log likelihood = -774.30364 Iteration 9: log likelihood = -774.30364 Mixed logit model Number of obs = 2532 LR chi2(8) = 14.57 Log likelihood = -774.30364 Prob > chi2 = 0.0681 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0413909 .0219765 -1.88 0.060 -.0844639 .0016822 def_medium .621123 .204942 3.03 0.002 .219444 1.022802 def_low .889311 .1897857 4.69 0.000 .5173379 1.261284 fairtrade .5097833 .1030618 4.95 0.000 .3077859 .7117806 nut2 .2591809 .1437418 1.80 0.071 -.022548 .5409097 nut3 .4821266 .1285788 3.75 0.000 .2301167 .7341365 agrochhum .1148943 .1185573 0.97 0.332 -.1174737 .3472623 agrochenv .2533655 .1360599 1.86 0.063 -.0133071 .520038 agrochnone .939978 .1724825 5.45 0.000 .6019185 1.278038 SD def_medium -.0201413 .2720915 -0.07 0.941 -.5534308 .5131482 def_low .7484396 .1801876 4.15 0.000 .3952783 1.101601 fairtrade .4060894 .1654978 2.45 0.014 .0817197 .7304591 nut2 .4966302 .1922294 2.58 0.010 .1198675 .8733929 nut3 .1461543 .3640129 0.40 0.688 -.5672979 .8596066 agrochhum .0228544 .2243389 0.10 0.919 -.4168418 .4625505 agrochenv .3989732 .1961138 2.03 0.042 .0145972 .7833492 agrochnone -.4730144 .3119921 -1.52 0.129 -1.084508 .1384789 A5.5 Model when there is a low socioeconomic level (1 &2) . xi: mixlogit choice price if estrato_bajo==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -674.11932 (not concave) Iteration 1: log likelihood = -672.87673 (not concave) Iteration 2: log likelihood = -672.5165 (not concave) Iteration 3: log likelihood = -667.6263 Iteration 4: log likelihood = -667.24678 Iteration 5: log likelihood = -665.19792 Iteration 6: log likelihood = -665.18083 Iteration 7: log likelihood = -665.18082 Mixed logit model Number of obs = 2148 LR chi2(8) = 18.42 Log likelihood = -665.18082 Prob > chi2 = 0.0183 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0148494 .0241398 -0.62 0.538 -.0621626 .0324638 def_medium .4562395 .2298775 1.98 0.047 .0056878 .9067912 def_low .8081416 .2191204 3.69 0.000 .3786735 1.23761 fairtrade .4677136 .1144443 4.09 0.000 .2434068 .6920203 nut2 .1389009 .1582653 0.88 0.380 -.1712934 .4490953 nut3 .2601248 .1560359 1.67 0.095 -.0456998 .5659495 agrochhum -.0887862 .1313624 -0.68 0.499 -.3462517 .1686794 agrochenv .2979373 .1482716 2.01 0.044 .0073304 .5885443 agrochnone .7958319 .1911099 4.16 0.000 .4212634 1.1704 SD def_medium .1162838 .3258662 0.36 0.721 -.5224022 .7549698 def_low .88213 .1927012 4.58 0.000 .5044426 1.259817 fairtrade .5035574 .1635381 3.08 0.002 .1830285 .8240862 nut2 .4848088 .2197623 2.21 0.027 .0540827 .915535 nut3 -.4408247 .3572719 -1.23 0.217 -1.141065 .2594154 agrochhum -.0167603 .1707033 -0.10 0.922 -.3513326 .3178121 agrochenv .1093528 .2621476 0.42 0.677 -.4044471 .6231527 agrochnone .4098711 .3539913 1.16 0.247 -.2839392 1.103681 A5.6 Model when there is a medium socioeconomic level (3&4) . xi: mixlogit choice price if estrato_medio==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -657.34496 (not concave) Iteration 1: log likelihood = -657.20076 (not concave) Iteration 2: log likelihood = -655.38154 (not concave) Iteration 3: log likelihood = -654.66007 Iteration 4: log likelihood = -654.54273 Iteration 5: log likelihood = -649.4921 Iteration 6: log likelihood = -649.35456 Iteration 7: log likelihood = -649.17306 Iteration 8: log likelihood = -649.17135 Iteration 9: log likelihood = -649.17134 Mixed logit model Number of obs = 2124 LR chi2(8) = 16.18 Log likelihood = -649.17134 Prob > chi2 = 0.0398 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0364485 .0247186 -1.47 0.140 -.0848962 .0119991 def_medium .55777 .2256298 2.47 0.013 .1155437 .9999962 def_low .792738 .2071928 3.83 0.000 .3866475 1.198829 fairtrade .6198796 .1297837 4.78 0.000 .3655084 .8742509 nut2 .3637499 .1635285 2.22 0.026 .0432399 .6842598 nut3 .4910164 .1458967 3.37 0.001 .2050642 .7769686 agrochhum .103631 .1332999 0.78 0.437 -.1576319 .364894 agrochenv .2635537 .1447128 1.82 0.069 -.0200782 .5471856 agrochnone .9733847 .2159984 4.51 0.000 .5500356 1.396734 SD def_medium -.4127984 .2196487 -1.88 0.060 -.843302 .0177052 def_low .5464417 .2636804 2.07 0.038 .0296377 1.063246 fairtrade .6382716 .1701808 3.75 0.000 .3047233 .9718199 nut2 -.4850797 .2075551 -2.34 0.019 -.8918803 -.0782792 nut3 -.4297766 .318746 -1.35 0.178 -1.054507 .1949541 agrochhum -.2774558 .3160334 -0.88 0.380 -.8968699 .3419583 agrochenv -.066872 .3177527 -0.21 0.833 -.6896559 .5559118 agrochnone .9885531 .2795114 3.54 0.000 .4407209 1.536385 A5.7 Model when there is a high socioeconomic level (5&6) . . xi: mixlogit choice price if estrato_alto==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -204.57935 (not concave) Iteration 1: log likelihood = -202.35207 (not concave) Iteration 2: log likelihood = -201.73209 (not concave) Iteration 3: log likelihood = -201.3235 Iteration 4: log likelihood = -200.04367 Iteration 5: log likelihood = -198.94601 Iteration 6: log likelihood = -198.4637 Iteration 7: log likelihood = -198.44492 Iteration 8: log likelihood = -198.4449 Mixed logit model Number of obs = 648 LR chi2(8) = 12.66 Log likelihood = -198.4449 Prob > chi2 = 0.1240 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0497069 .0593463 -0.84 0.402 -.1660235 .0666096 def_medium 1.204452 .5956621 2.02 0.043 .0369752 2.371928 def_low 1.310737 .5464488 2.40 0.016 .2397175 2.381758 fairtrade .3028638 .2544005 1.19 0.234 -.1957521 .8014797 nut2 .0573265 .3872302 0.15 0.882 -.7016308 .8162838 nut3 .4758488 .3190752 1.49 0.136 -.1495272 1.101225 agrochhum -.1101511 .32419 -0.34 0.734 -.7455519 .5252497 agrochenv .1111798 .3257729 0.34 0.733 -.5273234 .749683 agrochnone .4893332 .416423 1.18 0.240 -.326841 1.305507 SD def_medium -.0416539 .3114179 -0.13 0.894 -.6520217 .5687139 def_low .9348828 .4003385 2.34 0.020 .1502338 1.719532 fairtrade .7623639 .3167192 2.41 0.016 .1416056 1.383122 nut2 .9722569 .3498861 2.78 0.005 .2864927 1.658021 nut3 .0554818 .3376365 0.16 0.869 -.6062735 .7172371 agrochhum .8665889 .4184646 2.07 0.038 .0464134 1.686764 agrochenv -.4955459 .4049978 -1.22 0.221 -1.289327 .2982352 agrochnone -.9655786 .497392 -1.94 0.052 -1.940449 .0092918 A5.8 Model when the individual belongs to an ethnic group . xi: mixlogit choice price if ethnic==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -187.31028 (not concave) Iteration 1: log likelihood = -184.98206 (not concave) Iteration 2: log likelihood = -184.59275 (not concave) Iteration 3: log likelihood = -184.41782 Iteration 4: log likelihood = -183.48233 Iteration 5: log likelihood = -182.71368 Iteration 6: log likelihood = -182.67887 Iteration 7: log likelihood = -182.67868 Iteration 8: log likelihood = -182.67868 Mixed logit model Number of obs = 600 LR chi2(8) = 9.58 Log likelihood = -182.67868 Prob > chi2 = 0.2958 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .0123285 .0750192 0.16 0.869 -.1347065 .1593634 def_medium .7838761 .7758327 1.01 0.312 -.7367281 2.30448 def_low .9598204 .7428137 1.29 0.196 -.4960677 2.415708 fairtrade .2736656 .3195519 0.86 0.392 -.3526447 .8999759 nut2 -.059679 .4418284 -0.14 0.893 -.9256467 .8062887 nut3 .3166628 .3882015 0.82 0.415 -.4441981 1.077524 agrochhum -.0982188 .2889151 -0.34 0.734 -.664482 .4680443 agrochenv .7532885 .4423967 1.70 0.089 -.1137931 1.62037 agrochnone 1.279122 .636444 2.01 0.044 .0317148 2.526529 SD def_medium -.39086 .5577658 -0.70 0.483 -1.484061 .7023409 def_low 1.017951 .5254027 1.94 0.053 -.0118197 2.047721 fairtrade .6481826 .3769119 1.72 0.085 -.0905512 1.386916 nut2 .3370288 .4203003 0.80 0.423 -.4867445 1.160802 nut3 1.334427 .5747251 2.32 0.020 .207986 2.460867 agrochhum .3706508 .3438523 1.08 0.281 -.3032873 1.044589 agrochenv .7262702 .4085737 1.78 0.075 -.0745196 1.52706 agrochnone 1.433006 .7456704 1.92 0.055 -.0284807 2.894494 A5.9 Model when the individual does not belong to an ethnic group . xi: mixlogit choice price if ethnic==0, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1350.6439 (not concave) Iteration 1: log likelihood = -1339.1707 Iteration 2: log likelihood = -1332.7091 Iteration 3: log likelihood = -1332.4734 Iteration 4: log likelihood = -1332.4697 Iteration 5: log likelihood = -1332.4697 Mixed logit model Number of obs = 4320 LR chi2(8) = 38.32 Log likelihood = -1332.4697 Prob > chi2 = 0.0000 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.031078 .0170833 -1.82 0.069 -.0645606 .0024046 def_medium .5205353 .1579801 3.29 0.001 .2109001 .8301706 def_low .8116109 .1478326 5.49 0.000 .5218644 1.101357 fairtrade .5487555 .0841011 6.52 0.000 .3839204 .7135906 nut2 .2862145 .1151554 2.49 0.013 .0605139 .511915 nut3 .4019954 .1028874 3.91 0.000 .2003399 .6036509 agrochhum .0250107 .0938967 0.27 0.790 -.1590235 .2090449 agrochenv .2244832 .1032852 2.17 0.030 .022048 .4269185 agrochnone .8150401 .1397882 5.83 0.000 .5410604 1.08902 SD def_medium .116612 .2636501 0.44 0.658 -.4001327 .6333567 def_low .7641193 .1403925 5.44 0.000 .4889551 1.039284 fairtrade .5677026 .1133286 5.01 0.000 .3455826 .7898226 nut2 .6109701 .1430001 4.27 0.000 .3306951 .8912452 nut3 .1856107 .324702 0.57 0.568 -.4507936 .822015 agrochhum .1340785 .1979874 0.68 0.498 -.2539697 .5221267 agrochenv -.1985338 .1883151 -1.05 0.292 -.5676246 .1705571 agrochnone .7463447 .1897761 3.93 0.000 .3743905 1.118299 A5.10 Model when the individual did studies until high school . do "C:\Users\mesa001\AppData\Local\Temp\STD00000000.tmp" . xi: mixlogit choice price if educacion_colegio==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -550.3118 (not concave) Iteration 1: log likelihood = -548.65712 Iteration 2: log likelihood = -546.23793 Iteration 3: log likelihood = -545.2087 Iteration 4: log likelihood = -545.18408 Iteration 5: log likelihood = -545.18407 Mixed logit model Number of obs = 1740 LR chi2(8) = 10.34 Log likelihood = -545.18407 Prob > chi2 = 0.2418 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0300471 .0270654 -1.11 0.267 -.0830944 .0230001 def_medium .4776414 .254293 1.88 0.060 -.0207637 .9760465 def_low .5863634 .2306435 2.54 0.011 .1343104 1.038416 fairtrade .6209266 .1316661 4.72 0.000 .3628658 .8789875 nut2 -.0042451 .1686395 -0.03 0.980 -.3347725 .3262823 nut3 -.0104587 .1516298 -0.07 0.945 -.3076476 .2867302 agrochhum -.3298232 .1471487 -2.24 0.025 -.6182294 -.041417 agrochenv .1429687 .1593755 0.90 0.370 -.1694015 .4553388 agrochnone .6198299 .2056195 3.01 0.003 .216823 1.022837 SD def_medium .0437146 .381508 0.11 0.909 -.7040274 .7914566 def_low .5745647 .2396145 2.40 0.016 .104929 1.0442 fairtrade .6689303 .1646898 4.06 0.000 .3461442 .9917164 nut2 .5093552 .2460644 2.07 0.038 .0270779 .9916326 nut3 .0034883 .3358129 0.01 0.992 -.6546929 .6616694 agrochhum -.0117569 .1976593 -0.06 0.953 -.399162 .3756482 agrochenv -.0095814 .3481605 -0.03 0.978 -.6919635 .6728007 agrochnone .3302766 .377104 0.88 0.381 -.4088336 1.069387 A5.11 Model when the individual did technical/technological studies . do "C:\Users\mesa001\AppData\Local\Temp\STD00000000.tmp" . xi: mixlogit choice price if education_level==3, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -430.09947 (not concave) Iteration 1: log likelihood = -429.65269 (not concave) Iteration 2: log likelihood = -428.57932 (not concave) Iteration 3: log likelihood = -426.60868 (not concave) Iteration 4: log likelihood = -423.93891 (not concave) Iteration 5: log likelihood = -421.96653 Iteration 6: log likelihood = -421.1262 Iteration 7: log likelihood = -421.1137 Iteration 8: log likelihood = -421.11369 Mixed logit model Number of obs = 1404 LR chi2(8) = 18.32 Log likelihood = -421.11369 Prob > chi2 = 0.0189 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .0170328 .0336953 0.51 0.613 -.0490087 .0830743 def_medium .4019605 .3082999 1.30 0.192 -.2022962 1.006217 def_low .5978361 .2848272 2.10 0.036 .0395851 1.156087 fairtrade .5595216 .1639762 3.41 0.001 .2381342 .8809091 nut2 .3790187 .2185954 1.73 0.083 -.0494205 .8074579 nut3 .3825301 .2036157 1.88 0.060 -.0165493 .7816094 agrochhum .114488 .1688557 0.68 0.498 -.2164631 .4454391 agrochenv .2940268 .1954717 1.50 0.133 -.0890907 .6771444 agrochnone 1.044407 .3009721 3.47 0.001 .4545131 1.634302 SD def_medium .258792 .3832348 0.68 0.499 -.4923344 1.009918 def_low 1.014958 .3108017 3.27 0.001 .4057977 1.624118 fairtrade .4307248 .2314489 1.86 0.063 -.0229068 .8843563 nut2 .6112978 .28384 2.15 0.031 .0549816 1.167614 nut3 .7722336 .2684862 2.88 0.004 .2460104 1.298457 agrochhum -.0238177 .2843616 -0.08 0.933 -.5811563 .5335208 agrochenv -.3734078 .3101916 -1.20 0.229 -.9813721 .2345565 agrochnone -1.089723 .3443912 -3.16 0.002 -1.764717 -.4147283 A5.12 Model when the individual did University studies . xi: mixlogit choice price if educacion_universidad==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -541.99793 (not concave) Iteration 1: log likelihood = -540.60921 (not concave) Iteration 2: log likelihood = -540.36488 Iteration 3: log likelihood = -539.23931 Iteration 4: log likelihood = -537.00555 (not concave) Iteration 5: log likelihood = -536.44961 (not concave) Iteration 6: log likelihood = -535.57373 Iteration 7: log likelihood = -532.05661 Iteration 8: log likelihood = -531.55564 Iteration 9: log likelihood = -531.55285 Iteration 10: log likelihood = -531.55285 Mixed logit model Number of obs = 1776 LR chi2(8) = 22.08 Log likelihood = -531.55285 Prob > chi2 = 0.0048 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.049974 .0284576 -1.76 0.079 -.10575 .0058019 def_medium .6182357 .2703686 2.29 0.022 .088323 1.148148 def_low 1.177705 .2556688 4.61 0.000 .6766039 1.678807 fairtrade .4634819 .1430115 3.24 0.001 .1831845 .7437792 nut2 .4657145 .2016194 2.31 0.021 .0705478 .8608812 nut3 .8755373 .1789777 4.89 0.000 .5247475 1.226327 agrochhum .24719 .1614934 1.53 0.126 -.0693313 .5637112 agrochenv .2951542 .168346 1.75 0.080 -.0347979 .6251063 agrochnone .9818973 .247267 3.97 0.000 .4972629 1.466532 SD def_medium .2219276 .3203352 0.69 0.488 -.4059179 .849773 def_low .6806951 .2673038 2.55 0.011 .1567892 1.204601 fairtrade .6349408 .1737095 3.66 0.000 .2944764 .9754051 nut2 .6917203 .2620976 2.64 0.008 .1780185 1.205422 nut3 -.2518845 .375041 -0.67 0.502 -.9869514 .4831824 agrochhum -.5565355 .3029529 -1.84 0.066 -1.150312 .0372413 agrochenv -.1272595 .3116768 -0.41 0.683 -.7381347 .4836157 agrochnone -1.048961 .267222 -3.93 0.000 -1.572707 -.5252159 A5.13 Model when the household is composed by sensitive people (i.e. under legal age, elderly people) . . xi: mixlogit choice price if hh_sensitive==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -336.38889 (not concave) Iteration 1: log likelihood = -332.1272 Iteration 2: log likelihood = -330.40034 Iteration 3: log likelihood = -329.11931 Iteration 4: log likelihood = -329.03386 Iteration 5: log likelihood = -329.03337 Iteration 6: log likelihood = -329.03337 Mixed logit model Number of obs = 1140 LR chi2(8) = 14.66 Log likelihood = -329.03337 Prob > chi2 = 0.0662 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .0191991 .0423503 0.45 0.650 -.063806 .1022042 def_medium .3038804 .3836832 0.79 0.428 -.4481249 1.055886 def_low .5742175 .3587673 1.60 0.109 -.1289535 1.277388 fairtrade .9222754 .2269075 4.06 0.000 .4775449 1.367006 nut2 .3828221 .2724688 1.41 0.160 -.1512068 .9168511 nut3 .2493805 .2498017 1.00 0.318 -.240222 .7389829 agrochhum -.0807918 .2079716 -0.39 0.698 -.4884088 .3268251 agrochenv .3218979 .2414895 1.33 0.183 -.1514127 .7952086 agrochnone 1.200935 .3778818 3.18 0.001 .4603005 1.94157 SD def_medium -.6049704 .4303483 -1.41 0.160 -1.448437 .2384967 def_low 1.269582 .4537264 2.80 0.005 .3802951 2.15887 fairtrade .8753994 .3096486 2.83 0.005 .2684993 1.482299 nut2 .6373726 .2948759 2.16 0.031 .0594265 1.215319 nut3 -.5972451 .4028978 -1.48 0.138 -1.38691 .1924201 agrochhum -.1708425 .3446317 -0.50 0.620 -.8463081 .5046231 agrochenv -.2014183 .3870204 -0.52 0.603 -.9599643 .5571277 agrochnone 1.387284 .4825717 2.87 0.004 .4414607 2.333107 A5.14 Model when the household is composed by adults . xi: mixlogit choice price if hh_adults==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1072.4458 (not concave) Iteration 1: log likelihood = -1068.8056 (not concave) Iteration 2: log likelihood = -1067.6941 Iteration 3: log likelihood = -1062.8918 Iteration 4: log likelihood = -1062.7902 Iteration 5: log likelihood = -1062.7668 Iteration 6: log likelihood = -1062.7667 Iteration 7: log likelihood = -1062.7667 Mixed logit model Number of obs = 3348 LR chi2(8) = 19.80 Log likelihood = -1062.7667 Prob > chi2 = 0.0111 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0320579 .0183222 -1.75 0.080 -.0679688 .003853 def_medium .4783011 .1743258 2.74 0.006 .1366288 .8199734 def_low .7133973 .1625324 4.39 0.000 .3948396 1.031955 fairtrade .4166775 .0872619 4.78 0.000 .2456473 .5877077 nut2 .2093761 .1226004 1.71 0.088 -.0309163 .4496685 nut3 .375832 .109967 3.42 0.001 .1603006 .5913634 agrochhum .0280226 .1029005 0.27 0.785 -.1736586 .2297038 agrochenv .2105736 .114467 1.84 0.066 -.0137775 .4349247 agrochnone .7660858 .1519184 5.04 0.000 .4683312 1.06384 SD def_medium -.0072336 .292115 -0.02 0.980 -.5797684 .5653012 def_low .6681912 .1517458 4.40 0.000 .3707749 .9656075 fairtrade .433893 .1307934 3.32 0.001 .1775427 .6902433 nut2 .4059738 .1715772 2.37 0.018 .0696887 .7422589 nut3 .1963912 .4049212 0.49 0.628 -.5972398 .9900222 agrochhum .1591488 .179459 0.89 0.375 -.1925843 .5108819 agrochenv -.3031061 .2000202 -1.52 0.130 -.6951385 .0889264 agrochnone .7250259 .2103471 3.45 0.001 .3127531 1.137299 A5.15 Model when individuals are age_young . xi: mixlogit choice price if age_young==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -557.32251 (not concave) Iteration 1: log likelihood = -555.74954 (not concave) Iteration 2: log likelihood = -551.31528 (not concave) Iteration 3: log likelihood = -548.84162 Iteration 4: log likelihood = -546.62395 Iteration 5: log likelihood = -543.6859 Iteration 6: log likelihood = -543.55026 Iteration 7: log likelihood = -543.54999 Iteration 8: log likelihood = -543.54999 Mixed logit model Number of obs = 1752 LR chi2(8) = 29.29 Log likelihood = -543.54999 Prob > chi2 = 0.0003 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .0038163 .0300969 0.13 0.899 -.0551725 .0628051 def_medium .1003457 .3038213 0.33 0.741 -.495133 .6958244 def_low .5042802 .2663415 1.89 0.058 -.0177394 1.0263 fairtrade .5072827 .1479968 3.43 0.001 .2172142 .7973511 nut2 .4502304 .2141863 2.10 0.036 .030433 .8700278 nut3 .5838569 .1818626 3.21 0.001 .2274128 .940301 agrochhum .0347817 .1561649 0.22 0.824 -.2712959 .3408593 agrochenv .0545381 .1767732 0.31 0.758 -.2919309 .4010071 agrochnone .5696758 .2312121 2.46 0.014 .1165084 1.022843 SD def_medium .1283288 .3570933 0.36 0.719 -.5715613 .8282189 def_low .976636 .213122 4.58 0.000 .5589246 1.394347 fairtrade .6031984 .1583181 3.81 0.000 .2929006 .9134961 nut2 .9092157 .211306 4.30 0.000 .4950635 1.323368 nut3 .3017071 .3884039 0.78 0.437 -.4595505 1.062965 agrochhum -.0571563 .2660395 -0.21 0.830 -.5785841 .4642714 agrochenv -.1447532 .28372 -0.51 0.610 -.7008343 .4113278 agrochnone .6146264 .2720805 2.26 0.024 .0813584 1.147894 A5.16 Model when individuals are age_young_adult . . xi: mixlogit choice price if age_adult_young==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -533.59364 (not concave) Iteration 1: log likelihood = -527.01549 (not concave) Iteration 2: log likelihood = -524.55385 Iteration 3: log likelihood = -519.64504 Iteration 4: log likelihood = -519.40554 Iteration 5: log likelihood = -519.40246 Iteration 6: log likelihood = -519.40246 Mixed logit model Number of obs = 1776 LR chi2(8) = 28.73 Log likelihood = -519.40246 Prob > chi2 = 0.0004 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0327847 .0297901 -1.10 0.271 -.0911722 .0256029 def_medium .785537 .2654461 2.96 0.003 .2652723 1.305802 def_low 1.192022 .2746303 4.34 0.000 .6537563 1.730287 fairtrade .5990353 .1514576 3.96 0.000 .302184 .8958867 nut2 .2956545 .1913273 1.55 0.122 -.0793401 .670649 nut3 .5848367 .1913541 3.06 0.002 .2097894 .9598839 agrochhum .2095669 .1656839 1.26 0.206 -.1151675 .5343013 agrochenv .5371255 .1784512 3.01 0.003 .1873675 .8868834 agrochnone 1.031705 .259945 3.97 0.000 .522222 1.541188 SD def_medium -.142348 .3253379 -0.44 0.662 -.7799985 .4953025 def_low 1.068882 .2423643 4.41 0.000 .5938564 1.543907 fairtrade .8044951 .2086424 3.86 0.000 .3955634 1.213427 nut2 .2031205 .2778098 0.73 0.465 -.3413767 .7476177 nut3 -.5609099 .2867038 -1.96 0.050 -1.122839 .0010193 agrochhum .5823505 .2782433 2.09 0.036 .0370037 1.127697 agrochenv .1640454 .3240507 0.51 0.613 -.4710824 .7991732 agrochnone 1.485225 .3510769 4.23 0.000 .7971265 2.173323 A5.17 Model when individuals are age_adults . xi: mixlogit choice price if age_adults==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -405.08424 (not concave) Iteration 1: log likelihood = -404.76448 (not concave) Iteration 2: log likelihood = -404.66435 (not concave) Iteration 3: log likelihood = -403.97254 Iteration 4: log likelihood = -403.24834 Iteration 5: log likelihood = -403.24304 Iteration 6: log likelihood = -403.24304 Mixed logit model Number of obs = 1296 LR chi2(8) = 3.61 Log likelihood = -403.24304 Prob > chi2 = 0.8905 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0348323 .030218 -1.15 0.249 -.0940585 .0243939 def_medium .6391528 .27636 2.31 0.021 .0974972 1.180808 def_low .6560042 .2564597 2.56 0.011 .1533524 1.158656 fairtrade .5360733 .1346182 3.98 0.000 .2722265 .7999202 nut2 -.0363269 .1872071 -0.19 0.846 -.4032461 .3305923 nut3 .0180756 .1731939 0.10 0.917 -.3213781 .3575293 agrochhum -.3290541 .1670171 -1.97 0.049 -.6564015 -.0017066 agrochenv .173968 .1813309 0.96 0.337 -.1814341 .52937 agrochnone .7593922 .2378605 3.19 0.001 .2931942 1.22559 SD def_medium -.0350856 .3306674 -0.11 0.915 -.6831817 .6130106 def_low -.3033151 .3956446 -0.77 0.443 -1.078764 .4721339 fairtrade .2937004 .2634956 1.11 0.265 -.2227415 .8101423 nut2 .5460872 .2625698 2.08 0.038 .0314599 1.060714 nut3 -.0534978 .5190219 -0.10 0.918 -1.070762 .9637664 agrochhum .0332955 .2026248 0.16 0.869 -.3638418 .4304329 agrochenv .4616405 .2870606 1.61 0.108 -.100988 1.024269 agrochnone .2190192 1.02648 0.21 0.831 -1.792845 2.230883 A5.18 Model when individuals buy their food in low cost modern markets . xi: mixlogit choice price if lowcost_market==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -421.69186 (not concave) Iteration 1: log likelihood = -421.39813 (not concave) Iteration 2: log likelihood = -419.50826 (not concave) Iteration 3: log likelihood = -417.22252 Iteration 4: log likelihood = -417.03972 Iteration 5: log likelihood = -414.91863 Iteration 6: log likelihood = -414.61469 Iteration 7: log likelihood = -414.61325 Iteration 8: log likelihood = -414.61325 Mixed logit model Number of obs = 1344 LR chi2(8) = 14.13 Log likelihood = -414.61325 Prob > chi2 = 0.0785 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0618889 .0352173 -1.76 0.079 -.1309134 .0071357 def_medium .4763981 .3346516 1.42 0.155 -.1795069 1.132303 def_low .7355134 .3047671 2.41 0.016 .1381808 1.332846 fairtrade .6835387 .1882062 3.63 0.000 .3146614 1.052416 nut2 .6846022 .2598679 2.63 0.008 .1752704 1.193934 nut3 .8213825 .2629727 3.12 0.002 .3059655 1.3368 agrochhum .1060808 .1793546 0.59 0.554 -.2454478 .4576094 agrochenv .1961042 .2032509 0.96 0.335 -.2022602 .5944687 agrochnone .7866374 .2996189 2.63 0.009 .1993951 1.37388 SD def_medium .6094187 .2822909 2.16 0.031 .0561388 1.162699 def_low 1.044406 .3078523 3.39 0.001 .4410268 1.647786 fairtrade .7115477 .2682666 2.65 0.008 .1857549 1.237341 nut2 -.6643799 .2876583 -2.31 0.021 -1.22818 -.1005799 nut3 .9538915 .3555841 2.68 0.007 .2569595 1.650823 agrochhum -.1179783 .4977114 -0.24 0.813 -1.093475 .8575181 agrochenv -.4144454 .3267931 -1.27 0.205 -1.054948 .2260573 agrochnone -1.080148 .3773543 -2.86 0.004 -1.819749 -.3405472 A5.19 Model when individuals buy their food in traditional markets . . xi: mixlogit choice price if traditional_market==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -522.21439 (not concave) Iteration 1: log likelihood = -521.43731 (not concave) Iteration 2: log likelihood = -519.97744 (not concave) Iteration 3: log likelihood = -518.14717 Iteration 4: log likelihood = -517.00206 Iteration 5: log likelihood = -516.94299 Iteration 6: log likelihood = -516.90864 Iteration 7: log likelihood = -516.9085 Iteration 8: log likelihood = -516.9085 Mixed logit model Number of obs = 1632 LR chi2(8) = 10.47 Log likelihood = -516.9085 Prob > chi2 = 0.2339 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0198115 .0254561 -0.78 0.436 -.0697047 .0300816 def_medium .3722039 .2531836 1.47 0.142 -.1240267 .8684346 def_low .7198551 .2271782 3.17 0.002 .2745941 1.165116 fairtrade .5316924 .1335864 3.98 0.000 .2698679 .793517 nut2 .1598916 .1837694 0.87 0.384 -.2002898 .520073 nut3 .145006 .1633366 0.89 0.375 -.1751278 .4651398 agrochhum -.1195171 .1497666 -0.80 0.425 -.4130542 .1740201 agrochenv .2243541 .1674707 1.34 0.180 -.1038824 .5525906 agrochnone .6070526 .2027193 2.99 0.003 .2097301 1.004375 SD def_medium -.2195457 .4742083 -0.46 0.643 -1.148977 .7098855 def_low .7802563 .2399386 3.25 0.001 .3099853 1.250527 fairtrade .6135776 .169467 3.62 0.000 .2814284 .9457269 nut2 -.4436271 .2481235 -1.79 0.074 -.9299402 .0426861 nut3 -.0167057 .4836197 -0.03 0.972 -.9645829 .9311716 agrochhum .0555229 .206497 0.27 0.788 -.3492039 .4602496 agrochenv -.0254758 .2381103 -0.11 0.915 -.4921634 .4412117 agrochnone -.495791 .297707 -1.67 0.096 -1.079286 .0877039 A5.20 Model when individuals buy their food in the modern market . . xi: mixlogit choice price if modern_market==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -586.24332 (not concave) Iteration 1: log likelihood = -580.06696 (not concave) Iteration 2: log likelihood = -579.09224 (not concave) Iteration 3: log likelihood = -578.4936 Iteration 4: log likelihood = -575.59087 Iteration 5: log likelihood = -575.29088 Iteration 6: log likelihood = -574.49057 Iteration 7: log likelihood = -574.48658 Iteration 8: log likelihood = -574.48658 Mixed logit model Number of obs = 1944 LR chi2(8) = 26.06 Log likelihood = -574.48658 Prob > chi2 = 0.0010 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0070066 .0299418 -0.23 0.815 -.0656915 .0516783 def_medium .7211141 .2692409 2.68 0.007 .1934117 1.248817 def_low .9841584 .2609238 3.77 0.000 .4727571 1.49556 fairtrade .4774414 .1325348 3.60 0.000 .217678 .7372048 nut2 .0945206 .1791243 0.53 0.598 -.2565565 .4455978 nut3 .4607419 .1587567 2.90 0.004 .1495844 .7718993 agrochhum -.0035023 .1472213 -0.02 0.981 -.2920508 .2850461 agrochenv .3029915 .1762574 1.72 0.086 -.0424667 .6484496 agrochnone 1.211799 .2562392 4.73 0.000 .709579 1.714018 SD def_medium -.0764415 .2591721 -0.29 0.768 -.5844094 .4315264 def_low .6704697 .2476829 2.71 0.007 .18502 1.155919 fairtrade .6446346 .1728002 3.73 0.000 .3059525 .9833168 nut2 .6362471 .2477 2.57 0.010 .150764 1.12173 nut3 -.3136298 .296982 -1.06 0.291 -.8957038 .2684443 agrochhum .1540664 .5203289 0.30 0.767 -.8657595 1.173892 agrochenv .6155362 .2944084 2.09 0.037 .0385065 1.192566 agrochnone .9389233 .3269421 2.87 0.004 .2981285 1.579718 A5.21 Model when there is a positive relationship with the CPE variable . xi: mixlogit choice price if cpe>=4, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1056.6428 (not concave) Iteration 1: log likelihood = -1054.0212 (not concave) Iteration 2: log likelihood = -1049.6658 (not concave) Iteration 3: log likelihood = -1044.2946 Iteration 4: log likelihood = -1039.4224 Iteration 5: log likelihood = -1038.8035 Iteration 6: log likelihood = -1034.6241 Iteration 7: log likelihood = -1034.5206 Iteration 8: log likelihood = -1034.5205 Iteration 9: log likelihood = -1034.5205 Mixed logit model Number of obs = 3516 LR chi2(8) = 43.96 Log likelihood = -1034.5205 Prob > chi2 = 0.0000 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.0637121 .022362 -2.85 0.004 -.1075408 -.0198833 def_medium .9631253 .2062371 4.67 0.000 .5589079 1.367343 def_low 1.353041 .208532 6.49 0.000 .9443262 1.761757 fairtrade .6293458 .1060282 5.94 0.000 .4215344 .8371572 nut2 .3147167 .1372416 2.29 0.022 .045728 .5837053 nut3 .4027978 .1275392 3.16 0.002 .1528255 .65277 agrochhum -.0390557 .1085628 -0.36 0.719 -.2518348 .1737235 agrochenv .3705685 .1234392 3.00 0.003 .1286322 .6125049 agrochnone 1.142973 .1956425 5.84 0.000 .7595209 1.526426 SD def_medium .1162079 .2381243 0.49 0.626 -.3505071 .5829229 def_low .9569029 .1668098 5.74 0.000 .6299616 1.283844 fairtrade .5603544 .133583 4.19 0.000 .2985365 .8221723 nut2 .7373147 .1730596 4.26 0.000 .3981241 1.076505 nut3 .5827911 .2561825 2.27 0.023 .0806826 1.0849 agrochhum -.1235251 .2115415 -0.58 0.559 -.5381389 .2910887 agrochenv -.1142527 .2201998 -0.52 0.604 -.5458363 .3173309 agrochnone -1.185642 .2214633 -5.35 0.000 -1.619702 -.7515816 A5.22 Model when there is a negative relationship with the CPE variable . xi: mixlogit choice price if cpe<=2, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -85.403263 (not concave) Iteration 1: log likelihood = -83.579529 (not concave) Iteration 2: log likelihood = -83.390641 Iteration 3: log likelihood = -82.715839 Iteration 4: log likelihood = -82.456895 Iteration 5: log likelihood = -82.383681 Iteration 6: log likelihood = -82.382316 Iteration 7: log likelihood = -82.382315 Mixed logit model Number of obs = 276 LR chi2(8) = 6.17 Log likelihood = -82.382315 Prob > chi2 = 0.6277 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .2140877 .0905772 2.36 0.018 .0365597 .3916157 def_medium -1.102957 .8043496 -1.37 0.170 -2.679454 .4735388 def_low -1.187078 .7296186 -1.63 0.104 -2.617104 .2429483 fairtrade .1245991 .3944725 0.32 0.752 -.6485527 .897751 nut2 .2751587 .5703728 0.48 0.630 -.8427513 1.393069 nut3 -.0468302 .407657 -0.11 0.909 -.8458232 .7521628 agrochhum .157589 .4112756 0.38 0.702 -.6484963 .9636744 agrochenv -.0584743 .5387199 -0.11 0.914 -1.114346 .9973973 agrochnone 1.065155 .7213748 1.48 0.140 -.3487131 2.479024 SD def_medium .1507386 .6373427 0.24 0.813 -1.09843 1.399907 def_low .1041591 .6055814 0.17 0.863 -1.082759 1.291077 fairtrade .8410873 .5173562 1.63 0.104 -.1729123 1.855087 nut2 .9359124 .7857969 1.19 0.234 -.6042212 2.476046 nut3 -.0183845 .5927931 -0.03 0.975 -1.180238 1.143469 agrochhum -.6194235 .5797352 -1.07 0.285 -1.755684 .5168366 agrochenv 1.022534 .8211023 1.25 0.213 -.5867973 2.631864 agrochnone 1.106028 .7886387 1.40 0.161 -.4396756 2.651731 A5.23 Model when there is a positive relationship with the positive attitude variable . xi: mixlogit choice price if possitiveattitude>=4, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1047.072 (not concave) Iteration 1: log likelihood = -1038.3758 (not concave) Iteration 2: log likelihood = -1035.1548 Iteration 3: log likelihood = -1029.7233 Iteration 4: log likelihood = -1029.571 Iteration 5: log likelihood = -1029.5706 Iteration 6: log likelihood = -1029.5706 Mixed logit model Number of obs = 3528 LR chi2(8) = 36.14 Log likelihood = -1029.5706 Prob > chi2 = 0.0000 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.050993 .0211802 -2.41 0.016 -.0925054 -.0094806 def_medium .8966624 .1942426 4.62 0.000 .5159539 1.277371 def_low 1.195448 .1872752 6.38 0.000 .8283956 1.562501 fairtrade .6482441 .1016967 6.37 0.000 .4489222 .8475659 nut2 .3188892 .131014 2.43 0.015 .0621064 .575672 nut3 .4721136 .1268375 3.72 0.000 .2235166 .7207106 agrochhum .0023737 .1055471 0.02 0.982 -.2044947 .2092422 agrochenv .2929972 .1192203 2.46 0.014 .0593297 .5266646 agrochnone 1.05698 .1825259 5.79 0.000 .6992361 1.414724 SD def_medium -.0548974 .2219482 -0.25 0.805 -.4899079 .3801132 def_low .8427871 .1750523 4.81 0.000 .4996909 1.185883 fairtrade .5881808 .1366294 4.30 0.000 .3203921 .8559696 nut2 .566106 .1879393 3.01 0.003 .1977517 .9344602 nut3 .6136839 .2819765 2.18 0.030 .0610202 1.166348 agrochhum -.0408121 .225156 -0.18 0.856 -.4821098 .4004855 agrochenv -.0784936 .2715355 -0.29 0.773 -.6106934 .4537062 agrochnone 1.03877 .2056158 5.05 0.000 .6357705 1.441769 A5.24 Model when there is a negative relationship with the positive attitude variable . xi: mixlogit choice price if possitiveattitude<=2, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -176.10419 Iteration 1: log likelihood = -176.05549 Iteration 2: log likelihood = -175.38764 Iteration 3: log likelihood = -175.38272 Iteration 4: log likelihood = -175.36738 Iteration 5: log likelihood = -175.36707 Iteration 6: log likelihood = -175.36707 Mixed logit model Number of obs = 516 LR chi2(8) = 0.51 Log likelihood = -175.36707 Prob > chi2 = 0.9999 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .0223556 .038892 0.57 0.565 -.0538713 .0985826 def_medium .2382925 .3972249 0.60 0.549 -.540254 1.016839 def_low .0873551 .3143489 0.28 0.781 -.5287574 .7034677 fairtrade .028119 .1802781 0.16 0.876 -.3252197 .3814576 nut2 -.4288699 .2924326 -1.47 0.142 -1.002027 .1442874 nut3 -.2324165 .246792 -0.94 0.346 -.71612 .2512871 agrochhum -.2112066 .233325 -0.91 0.365 -.6685151 .246102 agrochenv -.32896 .2537768 -1.30 0.195 -.8263533 .1684334 agrochnone .0308065 .2944414 0.10 0.917 -.546288 .607901 SD def_medium .0058657 .2717429 0.02 0.983 -.5267407 .538472 def_low .009377 .3113018 0.03 0.976 -.6007634 .6195174 fairtrade .1377561 .4671511 0.29 0.768 -.7778431 1.053355 nut2 -.1488789 .8075308 -0.18 0.854 -1.73161 1.433852 nut3 -.0094649 .233186 -0.04 0.968 -.466501 .4475712 agrochhum .048253 .3064051 0.16 0.875 -.5522899 .6487959 agrochenv -.2099344 .4599442 -0.46 0.648 -1.111409 .6915398 agrochnone -.1244412 .3104526 -0.40 0.689 -.7329172 .4840348 A5.25 Model for the second scenario (extra information given) . xi: mixlogit choice price if extrainfo==1, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -178.24351 (not concave) Iteration 1: log likelihood = -178.20933 (not concave) Iteration 2: log likelihood = -173.77633 Iteration 3: log likelihood = -172.19398 Iteration 4: log likelihood = -168.48003 Iteration 5: log likelihood = -168.34134 Iteration 6: log likelihood = -167.81948 Iteration 7: log likelihood = -167.81159 Iteration 8: log likelihood = -167.81158 Mixed logit model Number of obs = 720 LR chi2(8) = 20.81 Log likelihood = -167.81158 Prob > chi2 = 0.0077 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price .047863 .094175 0.51 0.611 -.1367167 .2324427 def_medium 1.265424 .4020548 3.15 0.002 .4774113 2.053437 def_low 1.414502 .6611427 2.14 0.032 .118686 2.710318 fairtrade 1.076443 .4727796 2.28 0.023 .1498119 2.003074 nut2 .0637496 .7993674 0.08 0.936 -1.502982 1.630481 nut3 .0416505 .6712671 0.06 0.951 -1.274009 1.35731 agrochhum .6277449 .6838315 0.92 0.359 -.7125401 1.96803 agrochenv .8296728 .6576001 1.26 0.207 -.4591997 2.118545 agrochnone 1.450726 .5462095 2.66 0.008 .3801748 2.521277 SD def_medium -.7092878 .6047834 -1.17 0.241 -1.894641 .4760658 def_low .5145151 .6923191 0.74 0.457 -.8424054 1.871436 fairtrade 1.776636 .6058347 2.93 0.003 .5892219 2.96405 nut2 -.6825834 .5072618 -1.35 0.178 -1.676798 .3116314 nut3 .1942164 .5181706 0.37 0.708 -.8213792 1.209812 agrochhum -.6361534 .73084 -0.87 0.384 -2.068574 .7962667 agrochenv -.5361353 .4643443 -1.15 0.248 -1.446233 .3739627 agrochnone 1.234886 .7235508 1.71 0.088 -.1832471 2.65302 A5.26 Model for the general scenario (no extra information given) . xi: mixlogit choice price if extrainfo==0, rand (def_medium def_low fairtrade nut2 nut3 agrochhum agrochenv agrochnone) gr(choiceset) id(user) Iteration 0: log likelihood = -1346.7954 (not concave) Iteration 1: log likelihood = -1345.9004 (not concave) Iteration 2: log likelihood = -1342.0672 Iteration 3: log likelihood = -1333.6424 (not concave) Iteration 4: log likelihood = -1332.3974 Iteration 5: log likelihood = -1331.968 Iteration 6: log likelihood = -1331.7483 Iteration 7: log likelihood = -1331.748 Iteration 8: log likelihood = -1331.748 Mixed logit model Number of obs = 4200 LR chi2(8) = 31.89 Log likelihood = -1331.748 Prob > chi2 = 0.0001 choice Coef. Std. Err. z P>|z| [95% Conf. Interval] Mean price -.017379 .0208371 -0.83 0.404 -.0582189 .023461 def_medium .395519 .2520624 1.57 0.117 -.0985141 .8895522 def_low .6828112 .2042876 3.34 0.001 .2824148 1.083208 fairtrade .383936 .1019044 3.77 0.000 .1842071 .5836649 nut2 .2082679 .1559942 1.34 0.182 -.0974752 .514011 nut3 .4859843 .1107285 4.39 0.000 .2689605 .7030081 agrochhum -.0052209 .0950507 -0.05 0.956 -.1915169 .181075 agrochenv .202706 .1173202 1.73 0.084 -.0272375 .4326494 agrochnone .6898215 .1728354 3.99 0.000 .3510702 1.028573 SD def_medium .0037136 .177565 0.02 0.983 -.3443073 .3517346 def_low .6421241 .1551405 4.14 0.000 .3380542 .9461939 fairtrade -.3072504 .1438715 -2.14 0.033 -.5892333 -.0252674 nut2 .5246986 .1510349 3.47 0.001 .2286757 .8207214 nut3 -.5014497 .2000088 -2.51 0.012 -.8934597 -.1094397 agrochhum -.0178736 .1757713 -0.10 0.919 -.3623791 .3266318 agrochenv .0356817 .1912296 0.19 0.852 -.3391215 .4104848 agrochnone .9019246 .1685747 5.35 0.000 .5715242 1.232325 Annex 6: MWTP calculations in Excel 6. MWTP Calculations v3.0.xlsx Annex 7: LookApp invoice 7. LookApp invoice.pdf