Synthesis

INTRODUCTION

The sustainable transformation of the agricultural sector is incentivized by a wide range of environmental policy instruments aimed at reducing ecologically harmful practices and preserving multifunctional landscapes (van Zanten et al. 2014, Hasler et al. 2022). To ensure that agricultural policies are designed in line with these objectives, they must be critically evaluated. Especially for ex-ante policy evaluation, model-based analyses are well suited. They allow for a systematic assessment of policies under changing environmental, economic, or institutional conditions (Baumgärtner et al. 2008, Grimm et al. 2020). In particular, models can consider direct and indirect drivers of a system, such as future climate and market conditions, as well as different policy regimes and management options (Bruch and Atwell 2015, Holtz et al. 2015). By providing socioeconomic and environmental outputs for different future scenarios, models can help evaluate the effectiveness of different policy designs in achieving sustainable and economically viable outcomes under changing conditions.

Because agricultural policies usually target the farm level (Langrell et al. 2013, Kremmydas et al. 2018), it is particularly valuable to use models for policy analysis that allow a representation of individual farmers—the crucial actors in land use systems (Nolan et al. 2009, Reidsma et al. 2018). Agent-based models (ABMs) are particularly suitable to serve this purpose. They can capture heterogeneity in terms of decision-making, available resources, and environmental conditions (Parker et al. 2003, Matthews et al. 2007), and make it possible to distinguish farmer types (Arneth et al. 2014) as well as to explore the role of interactions among farmers. Thus, for example, the conditions under which farmers decide to adopt sustainable practices can be explicitly represented. Moreover, when biophysical aspects are taken into account, feedbacks between the behavior of individual farmers and impacts on the environment can be disentangled (An 2012, Schlüter et al. 2012, Robinson et al. 2018).

Yet, although the explicit representation of human behavior is one of the strengths of agent-based models, it is also one of its greatest challenges (Elsawah et al. 2020, An et al. 2021). Part of the criticism of land use ABMs is that simulated decision-making is often based on ad hoc assumptions that are not supported by empirical evidence or grounded in behavioral theories (Crooks et al. 2008, Groeneveld et al. 2017). Furthermore, some ABMs in the land use context have been criticized for omitting elements that may be crucial for farm management decisions, such as heterogeneity among farmers and social interactions (Huber et al. 2018). However, in order for ABMs to provide policy makers with the information necessary to achieve the desired policy design, an adequate representation of farmers’ decision-making is crucial.

Although recently several approaches to formalize theories of human decision-making for ABMs of social-ecological systems have been developed (Schlüter et al. 2017, Schwarz et al. 2020, Constantino et al. 2021), most social science theories only consider certain aspects of decision-making and fall short when it comes to accounting for the multiple influences that farmers face (Meyfroidt 2013). Therefore, even well-formalized theories might not represent farmers’ decision-making processes in their entirety. In addition, in order to be applied to real-world situations, theory-driven models need an empirical reference, which requires that primary data collection be specifically designed to cover all aspects relevant to a theory. It can also be difficult to formalize into behavioral model rules empirical observations derived without a well-formalized underlying theory. Here, a broad range of complex interplaying factors need to be captured in a limited number of survey questions, which later have to be simplified into clear cause-effect relationships.

Depending on the scope of application, both explicitly data-driven and theory-driven ABMs are relevant (Taghikhah et al. 2021). However, all applications that aim to inform ABMs with empirical information about agents’ behavior require a structured process of translating these data into model rules. There is a wide range of approaches to empirically support ABMs for land use research, including surveys, unstructured or semi-structured interviews, ethnographic methods (such as participant observation), focus groups, field and laboratory experiments, participatory modeling, GIS and remote sensing, or even the use of mobile phone and social media data (Janssen and Ostrom 2006, Robinson et al. 2007, Bell 2017, Smith and Rand 2018). Here, we focus on qualitative and quantitative primary data collected in interviews and surveys that explicitly address farmers’ behavior and the reasons farmers provide for their behavior. Moreover, we narrow our scope to how empirical data can be used to determine the model rules related to human decision-making in an ABM. We thus omit a number of different purposes for using empirical data, such as validation or stakeholder discussions (Boero and Squazzoni 2005, Polhill et al. 2010, Smajgl et al. 2011, Smajgl and Barreteau 2017). All of these approaches have their justification. In particular, participatory modeling has proven to be effective in the context of modeling land management (see examples in Voinov et al. 2016). However, these methods also have their individual challenges, some of which vary widely (see, e.g., Voinov and Bousquet 2010, Voinov et al. 2016, Gray et al. 2018, or Sterling et al. 2019 for experiences related to participatory modeling), and thus require separate treatments. With interviews and surveys to inform farmer decision-making, we only focus on a small part of the field of empirically supported ABMs, but throughout the paper we provide perspectives on where this approach reaches its limits and how these can be overcome by alternative methods.

Although existing frameworks mainly deal with guiding the parameterization of human behavior (Smajgl et al. 2011) and describing the formalization (e.g., ODD+D, Müller et al. 2013), there are few guidelines on how to translate primary data into behavioral model rules. To contribute to a better representation of farmers’ behavior in land use models, the main objective of this article is to structure the process of using primary data to inform the representation of human decision-making in ABMs. We believe this can help realize the full potential of ABMs and, ultimately, assist policy makers in identifying appropriate interventions for transformation of agriculture toward sustainability. We divide the process in three steps:

1. Selecting approach: We discuss which primary data collection options are appropriate for inferring specific aspects of farmers’ behavior. Specifically, we distinguish the potential of inductive and deductive approaches and show which approach is best suited to obtain which type of information. In this context, we also distinguish the extent to which an approach is based on behavioral theories and how the data are obtained and analyzed.
2. Choosing key elements: We summarize aspects that need to be captured in empirical data to inform behavioral model rules in an ABM. In particular, we discuss how heterogeneity, temporal processes, and interactions can be captured with primary data.
3. Translating findings: We compile approaches for translating empirical findings into formalized decision rules. Here, we observe that it is often unclear how empirical evidence is accounted for in the model and highlight how to increase the transparency of the formalization.

The remainder of the paper is organized along these three guiding steps. For all three steps, we discuss challenges and trade-offs arising in the use of various approaches. We underpin each formalization step with model examples from the literature. A summary of the guiding steps and the aspects they cover is provided in Figure 1. In addition, we draw on our own experience in developing an empirically grounded ABM that includes a decision-making framework for the adoption of agri-environmental schemes in Europe. We present how we addressed the guiding steps and highlight challenges we faced during the formalization process. We conclude with a discussion on the potential of using primary data for ABMs in the context of farmers’ behavior, its limitations, and how these can be overcome by applying other approaches to include empirical knowledge into ABMs.

FORMALIZING HUMAN BEHAVIOR IN ABMS USING PRIMARY DATA

Formalizing human behavior in ABMs using primary data spans multiple steps from designing data collection to translating empirical observations into behavioral model rules. In order to structure this process, we loosely follow a framework developed by Schwarz et al. 2020 on formalizing theories, which summarizes key steps to consider when including theories of human decision-making in ABMs. The main focus of this framework is on selecting and formalizing a theory. Furthermore, it deals with the translation of formalization into code and the documentation of the model development. For models based on primary data, we focus on the first two aspects because we perceive parallels with the translation to code and the model documentation when using primary data to inform an ABM. Adapted to the empirical context, we end up with three main topics that are crucial for using primary data in ABMs: (1) how to select an appropriate empirical method to collect primary data?, (2) which elements to cover in primary data collection to inform an ABM?, and (3) how to formalize decision rules in ABMs based on empirical data? For each of the three aspects, we provide a general conceptual overview of the ways in which the problem can be addressed. In addition, we present literature examples to show the wide range of possible realizations of empirically based behavioral model rules in ABMs and highlight strengths and limitations of the different approaches. To identify suitable examples from the ABM literature, we performed a Web of Science Core Collection TOPIC search using the search string (agricult* OR farm*) AND “agent-based model*” AND (interview* OR survey*). All articles indexed by 7 August 2023 were included, which resulted in a total of 73 publications. We selected those studies in which (a) it was made clear how the decision rules were derived from interview or survey data, and which (b) provided sufficient information on how the data were collected and translated into behavioral model rules. This resulted in 13 articles. In addition, we screened the literature included in a recent review of the representation of decision-making in European agricultural ABMs (Huber et al. 2018) and added three studies that fulfill the selection criteria but were not found through the database search (Millington et al. 2008, Valbuena et al. 2010a, Acosta et al. 2014).

Approaches for primary data collection to inform farmers’ behavior in an ABM

To be able to draw the desired information from empirical observations, it is crucial to design data collection among farmers, stakeholders, or other relevant informants in an appropriate way. We structure the process behind the selection of a suitable strategy for the design of empirical data collection into three dimensions: (1) the reasoning behind the interview and survey questions that are posed, (2) the extent to which behavioral theories are behind the questions, and (3) the methods used to elicit answers to the questions.

For the first dimension, we differentiate between inductive and deductive approaches. In inductive data collection, no prior assumptions about farmers’ behavior are included in the formulation of the questions but determinants for behavior or specific behavioral rules are derived from primary data. A common approach for formalizing logical relationships on the basis of empirical observations is the “grounded theory” (Strauss and Corbin 1997) that involves constructing hypotheses and theories through a systematic analysis of qualitative data. It is particularly suitable when targeting a new field of research, addressing a region where no data collection on a specific topic has been conducted yet, or when it is unclear if existing theories or assumptions fit to the specific case and therefore should not dominate data collection. In deductive data collection, interview or survey questions are based on prior information, usually from empirical or theoretical literature. In this case, the behavioral factors of interest are known, and data collection is used to quantify the importance of these factors or to better understand their relationships with each other.

A priori insights about farmers’ behavior may be derived from previous surveys or discussions with stakeholders, a literature review, or behavioral theories. This leads to our second dimension where we distinguish the extent to which interview or survey questions are driven by theory. In particular, we distinguish two stylized categories that reflect the origin of the prior information: theory-driven and data-driven. By data-driven we mean the absence of an explicit reference to an established and recognized theory, although even in this case a behavioral theory may implicitly influence the selection of factors that are considered relevant. In fact, it is fair to assume that every researcher working on human behavior has implicit theories/models of how people behave. However, as long as they remain implicit, it is effectively not existent for the outside observer (e.g., reader of a scientific publication). We apply this differentiation only to deductive approaches, because data collection with inductive approaches is by definition free of prior assumptions and therefore always falls into the “data-driven” category. With theory-driven data collection, we refer to studies that explicitly assume that existing behavioral concepts and theories, such as expected utility theory (von Neumann and Morgenstern 1947, as applied by Bocquého et al. 2013), theory of planned behavior (Ajzen 1991, as applied by Borges and Oude Lansink 2016, Despotović et al. 2019, and Bonke and Mussof 2020), or prospect theory (Kahneman and Tversky 1979, as applied by Coelho et al. 2012), provide an appropriate description of farmers’ decision behavior. However, it can be difficult to capture all components relevant to a specific situation from empirical data with one specific theory. Depending on the decision context, different influence factors may be of importance (Dessart et al. 2019, Epanchin-Niell et al. 2022). Uncertainty and risk play a role, for example, when an outcome of a decision cannot be entirely influenced (e.g., when a result-based payment scheme is evaluated before subsidies are granted). Societal influences are particularly relevant in observable actions, such as spraying pesticides or planting flower strips. Thus, a data-driven approach based on empirical observations or literature might be more appropriate to guide the development of suitable survey questions. Alternatively, structured frameworks combining elements from different theories (e.g., Klöckner 2013, Epanchin-Niell et al. 2022) can be useful in such situations.

The last decision that has to be made with regard to data collection is that of the method with which the questions are asked. Here, we distinguish between qualitative and quantitative approaches. In principle, data from open-ended questions can also be analyzed by means of quantitative methods, such as natural language processing or topic modeling. However, in most cases, quantitative methods are used to analyze more structured data (from questionnaires, etc.), so we simplify by distinguishing “qualitative” and “quantitative data collection.” Qualitative data collection refers to many forms of collecting non-numerical (text, audio, video) data focusing on in-depth understanding of a case and its context. This allows for generating hypotheses, building new theories and concepts, or reconstructing social action. One method popular in the context of empirical ABMs are unstructured or semi-structured interviews (Kruse 2014, Przyborski and Wohlrab-Sahr 2014). These are mostly conducted in direct contact between interviewer and respondent (either face-to-face or via phone/virtually). This approach is often used to get general insights into specific decisions without limiting the range of relevant factors a priori. In the context of farmers’ behavior, this can, for example, include reasons for application of specific environmentally friendly practices.

Quantitative data collection involves structured questionnaires with mostly closed-ended questions that allow a statistical analysis (in some cases open-ended questions are coded to be translated to quantitative values). Often, this includes data collection on independent (e.g., socio-demographic data, attitudes) and dependent variables (e.g., adoption of specific practices). With sufficiently large sample sizes, this can be used to statistically determine behavioral rules (e.g., via regression analysis). Alternatively, it is possible to ask respondents to rate the importance of specific reasonings for a certain action. One could ask about the role of certain influencing factors, such as the economy or the environment, or preformulate possible behaviors and analyze farmers’ preferences. This could include specific rules such as “If insect x is spotted in my region, I apply insecticide y” or “I always use insecticide z.” In combination with the collection of socio-demographic data, these data can be analyzed to derive different types of agents, e.g., farmer types (Bartkowski et al. 2022). Again, the differentiation of qualitative and quantitative data collection is only relevant for deductive approaches. Inductive collection of primary data is always qualitative because without any prior knowledge predefined questions to obtain quantitative data are not reasonable. (However, administrative data in quantitative format not collected by researchers themselves might theoretically be used for inductive explorations of behavior.)

To provide an overview of possible approaches to data collection, we summarize the three dimensions in a decision tree (Fig. 2). Before discussing the resulting paths of possible combinations, we want to point out that the classification into binary categories is a simplification that does not cover the entire spectrum of data collection approaches. We acknowledge that some methods cannot be clearly assigned to either the qualitative or quantitative category (e.g., Q methodology explicitly combines both approaches with participants having to rank statements and explain their ordering, and the results being evaluated in a factor analysis), but we believe that even the simple classification can be helpful in distinguishing different ways of data collection and their strengths and limitations.

Out of the possible eight combinations of the three binary dimensions, we highlight three paths for data collection for ABMs. As argued before, we omit theory-driven and quantitative approaches for inductive study approaches because these do not fit with the general definition of inductive data collection. In addition, we did not find studies that combine deductive reasoning for survey design with qualitative data collection when the study is meant to inform an ABM. The few studies we are aware of that use the approach deductive–theory-driven–qualitative (e.g., Riley 2016 using Bourdieu’s notion of hysteresis to investigate farmers’ agri-environmental actions and identities) are not conducted in context of ABMs. For the combination deductive–data-driven–qualitative we have not found any examples. When empirical data are collected for ABMs, a possible reason why these combinations of approaches are not used could be that quantitative information is needed to parameterize the behavioral model rules. Qualitative approaches might therefore rather be applied in inductive settings where no prior information about driving factors for behavior is available.

The remaining three paths are (1) inductive–data-driven–qualitative, (2) deductive–theory-driven–quantitative, and (3) deductive–data-driven–quantitative. To underline the potential of these different approaches, we summarize their strengths and limitations and their importance in the modeling cycle (Schmolke et al. 2010). We add literature examples from studies where farmers’ decision-making in ABMs is based on empirical data (Table 1). We do not claim to provide an exhaustive overview of all existing studies. Rather, we aim to highlight the diverse range of how empirical information is used to derive respective model rules and parameterize farmers’ behavior. Among the selected studies, we find five studies that follow path 1, four path 2, and four path 3. The remaining three studies pursue a combined approach of path 1 and path 3.

The main strength of the first path (inductive–data-driven–qualitative) compared to the other two is its potential to uncover unexpected reasons or determinants for certain behavior. The approach is often characterized by small sample sizes because interviews conducted face-to-face require far more time compared to online or postal surveys (or even structured surveys conducted face-to-face). For studies in our literature sample, the number of respondents ranged from five local stakeholders (Millington et al. 2008) to interviews with 53 farmers and 14 key informants (Zagaria et al. 2021). Acosta et al. 2014 conducted face-to-face interviews with 28 farmers, which represent more than 90% of the total farmers in the region and therefore provide a good overview of the case study. Although small sample sizes are often sufficient in qualitative social science once a saturation of knowledge and a general understanding of the system is achieved (Roller and Lavrakas 2015), it is also possible to instead use participatory modeling with focus group discussions or workshops that involve not only farmers but also experts, such as agricultural policy makers, administrators, or advisors. This approach has been widely used in the context of land use management (e.g., Gaube et al. 2009 or Bakker et al. 2015). However, one should also keep in mind that farmers’ (and in particular other actors’ beliefs about farmers’) behavior may be biased or even plainly wrong (see Brown et al. 2021, Gütschow et al. 2021). In addition, parameterizing the model is difficult when quantitative data are lacking. Data collected through the first path can therefore mainly contribute to a general system understanding and model conceptualization (Yang and Gilbert 2008).

The scopes of the second (deductive–theory-driven–quantitative) and third (deductive–data-driven–quantitative) paths both mainly serve model parameterization. With those approaches, researchers have the opportunity to reach many respondents and to provide a broader and potentially representative picture of the surveyed population. Valbuena et al. 2010a, for example, based their behavioral model rules on a postal survey with 495 Dutch farmers (Jongeneel et al. 2008). Kaufmann et al. 2009 included their question block on farmers’ behavior in a larger survey with farm managers and were therefore able to interview 170 farmers in Latvia and 187 farmers in Estonia. However, to conduct such a deductive study, additional information must be available in advance. For data-driven approaches (path 3) this can, for example, be a general understanding of agronomic and economic processes or an existing conceptual framework of farmers’ behavior refined to the specific case study (Valbuena et al. 2010a, 2010b). If the approach is theory-driven (path 2), the empirical information must in addition fit into a theoretical framework that forms the basis for the questionnaire and the formalization into behavioral model rules. Kaufmann et al. 2009 argue that they have chosen the theory of planned behavior to model conversion to organic farming because of its conceptual parsimony and successful empirical applications. Liu et al. 2021 formalized their behavioral model rules on the belief-desire-intention theory to include agents’ knowledge about the farm and the environment, their farming goals as well as their actual technology adoption. Although this approach provides the possibility to parameterize predefined behavioral rules or the importance of different behavioral options, there is limited potential to detect unexpected behavior (e.g., by giving the possibility to indicate “other reasons” in an open-ended way to otherwise fixed response options or by providing a feedback format at the end of the survey).

In order to precisely tailor the empirical basis to one’s own case study and collecting quantitative information to parameterize the empirical observations, a combination of an inductive approach (path 1) and a deductive approach (path 3) is possible. In our literature sample, this approach was performed through a combination of semi-structured expert interviews to identify issues regarding hedgerow management (Tieskens et al. 2017), cultivation patterns (Yao et al. 2018), or participation in biogas production (Burg et al. 2021), and a survey to parameterize action rules for farmers in the model. Although such a combined approach can provide detailed insights for one’s research question, it also comes with challenges that must be considered when planning a project. First, it can be challenging to translate the interpreted findings from the qualitative part into meaningful quantitative questions. Moreover, observations from the inductive data collection do not necessarily need to be supported by the quantitative survey, which makes it difficult to justify the underlying model concept. Apart from that, the combination of approaches requires a lot of expertise, as the team needs to be familiar with qualitative social research and quantitative methods from economics or psychology in addition to ABM expertise (with additional knowledge on ecological aspects required in the land use context). Combining this experience and conducting the two-step approach takes a considerable amount of time, which is hardly available in typical projects with a duration of a few years.

Key elements of primary data needed to inform an ABM

To adequately inform the representation of farmers’ decision-making in an ABM with primary data, certain aspects of farm and farmer characteristics have to be addressed when collecting empirical information. Inspired by the elements considered in the MoHuB framework, which was developed for mapping behavioral theories in models of social-ecological systems (Schlüter et al. 2017), we distinguish three main categories: (1) heterogeneity between farmers, (2) temporal aspects, and (3) interactions (for a list of examples of survey components that fall into the different categories see Table 2).

Heterogeneity between farmers

It cannot be assumed that all farmers behave in the same way; on the contrary, a high degree of heterogeneity must be taken into account (Malek and Verburg 2020, Bartkowski et al. 2022). In fact, heterogeneity of actors is one of the main reasons for applying ABM (Railsback and Grimm 2012) rather than, e.g., conventional bioeconomic models based on a representative agent approach. During empirical data collection for ABMs, this heterogeneity among respondents needs to be addressed. On the one hand, in many empirical studies heterogeneity is primarily derived from farm and farmer characteristics. In this case, socio-demographic variables, such as age, gender, education, or the availability of a successor, and farm properties, such as location, area, or specialization, can be sampled. On the other hand, heterogeneity may also be characterized by strictly behavioral attributes. These can include attitudes toward past changes in land use, expected impacts of climate change, or the importance of landscape functions or environmentally friendly farming practices. Similarly, perceptions of current or previous situations as well as aspirational thresholds can be used to distinguish among farmer groups. Existing studies cover, among others, questions on barriers or enablers of future adaptation strategies as well as past drivers of change in the agricultural sector. Heterogeneity in attitudes and perceptions can be assessed in quantitative studies by using Likert scales, where farmers are asked to rate their agreement with particular statements or the importance of certain measures. To ease comparability between studies, we suggest using standardized frameworks, especially when sampling attitudes and perceptions instead of inventing new questions for each study. Attitudes can, for example, be classified according to the Environmental Attitudes Inventory (Milfont and Duckitt 2010) or the New Ecological Paradigm (Dunlap et al. 2000). Risk preferences should ideally be investigated by different elicitation methods, such as lottery decisions (e.g., following the initial setup by Holt and Laury 2002) or domain-specific self-assessment (see Finger et al. 2022 for an example on Swiss fruit producers). To analyze the heterogeneity among farmers, deriving farmer typologies is a promising approach (see Bartkowski et al. 2022 for a summary of existing studies on farmer typologies and methods that are used to synthesize empirical data). In ABMs, such typologies can be used to represent diversity between farmers while keeping a meaningful number of different behavioral types (Arneth et al. 2014).

Temporal processes

Decision-making processes often do not consist of single choices but represent a series of consecutive choices. In addition, behavior may change over time, especially in response to evolving biophysical and socioeconomic conditions. To model the decision-making of farmers in ABMs, it is therefore necessary to have knowledge about the temporal processes behind the behavior. With respect to data that are meant to inform an ABM, we reflect this in three main aspects: (1) development over time, (2) the sequential aspect of decision-making, and (3) scheduling of the behavior of several agents.

First, we argue that it is important to have a dynamic view on a system and not assume that decisions remain fixed over time. Especially if behavior is conditioned by previous experiences or other developments such as extreme events, it needs to be considered that agents might learn or adapt and actions might change. This has to be reflected in interviews and surveys, for example, by asking for past actions in response to exceptional situations and the likelihood for future implementation of adaptations given a potential increase of extreme events (Zagaria et al. 2021). In these questions, however, it is difficult to distinguish sudden from gradual changes in behavior. Such different rationale may be easier to uncover in unstructured or semi-structured interviews with inductive reasoning (path 1), as farmers may inherently distinguish long-term from short-term responses. In addition, it is worth investigating if there are specific triggers for some behavior. Possible questions in this regard can be targeted at behavior in an exceptionally warm, dry, or wet year and reasons for this behavior. In the context of pesticide use in vineyards, Chen et al. 2022 asked, for example, for the amount and number of sprayings in average years compared to years with exceptionally high and low use of pesticides as well as for a description of the climatic conditions of those exceptional years. However, dynamic responses are highly dependent on the farmer’s perceptions of the social and biophysical environment as well as the evaluation of the perceived state of the system that make it difficult to elucidate specific moments when someone actually reconsiders his or her behavior (Sutherland et al. 2012). Alternatively (though very resource-intensively), one could use a longitudinal approach, such as, e.g., Riley 2016 who interviewed farmers twice across multiple years to see changes in their attitudes.

Second, the sequential aspect of decision-making has to be considered, i.e., that a decision is a stepwise process ranging from problem detection to implementation (Öhlmér et al. 1998). Here, the crucial aspects that need to be derived from empirical data are the phases of the specific process and the importance of each element. Existing theories, such as theory of planned behavior, inherently include some kind of scheduling of processes. Although this theory has been included into a more general concept of farmers’ action space (Gütschow et al. 2021) that allows the formalization of opportunities and constraints at specific steps in the decision-making process, for some processes it might not be possible to be formalized in such frameworks. In that case, unstructured or semi-structured interviews could be particularly helpful to get an understanding of the sequential manner behind how actions are performed. In addition, it needs to be taken into account that farmers take several related but disconnected decisions (e.g., investment decision, decision what to crop, where to allocate which crop, etc.), i.e., the sequential aspects might need to be disentangled for several actions relevant for the specific context.

The last aspect concerning temporal processes refers to the multitude of actors in a system. This is particularly important when actions may be constrained by previous actions of other actors because of the limited availability of resources, such as when making decisions about using common property or applying for subsidies with limited resources (see, e.g., Johnson and Salemi 2022 for a discussion on conception of time to capture resource competition in an agent-based model on firewood harvesting). Here, empirical data collection should clarify whether there are hierarchies among actors that might influence the order of decision-making, or whether resource allocation follows a first-come, first-served basis. Hierarchies can be revealed indirectly in interviews or surveys by asking about preconditions that must be met before an action can be performed. Beyond interviews or surveys, such constraints can also be exposed in role-playing games, where the order of actions can be inferred directly from observing the behavior of the players.

Interactions

A third crucial aspect of ABMs is interaction among agents (Railsback and Grimm 2012). Although farms may be influenced by other farms via the environment (e.g., pesticide usage in one field affects organic farming in other fields) or agricultural trade (Dou et al. 2019), with respect to social interaction we here primarily focus on the effect of perceptions of what others do, as well as on the impact of perceived societal expectations on one’s own behavior (Burton et al. 2004, Dessart et al. 2019). Although social contracts through membership in farming cooperatives and access to advisory services can be relatively easily accessed through interviews and surveys, peer orientation can be difficult to derive (Bartkowski and Bartke 2018). One approach to eliciting these influences in interviews is to ask directly whether farmers consult other farmers (Mathijs 2003) or to have farmers assess the importance of friends’ or neighbors’ behavior on their own behavior (Caffaro et al. 2019). However, it is unclear how reliable responses to such specific questions are. Farmers may not want to admit (or may not even be aware) that they are influenced by the actions of others because they want to present themselves as an independent company. Additionally, there may be collaborations that are not consciously perceived as such because they have always been established, for example, when farmers share equipment and automatically engage in exchanges about land use strategies. Such tacit social influences might not surface through explicit questions in questionnaires or surveys and other ways of obtaining the information may be required. Kreft et al. 2021 applied a touchscreen-based data collection where farmers were asked to draw lines between persons with whom they regularly exchange about specific agricultural issues and rank the influence of others. Furthermore, in terms of social influence, secondary data may be suited to infer imitation or cooperation. Here, administrative data including location or remote sensing data could be helpful to infer similar behavior of proximate farms. Vroege et al. 2020 show, for example, that there are spatial spillover effects with respect to diversification using census data of farms in the Netherlands (see there also for an overview of literature regarding spillover effects in agricultural decision-making). Cooperative farming strategies can also be analyzed through serious games (see Janssen et al. 2023 for an overview on collective action games). Ryschawy et al. 2022, for example, applied a participatory approach to generate scenarios in the context of crop-livestock integration among farms. Rommel et al. 2022 evaluated the willingness to cooperate using a public goods game.

Translation of primary data into decision rules

In the following, we compile approaches to translating empirical findings into formalized decision rules based on our selection of literature. Although this overview does not provide exhaustive guidelines of how behavioral model rules can be derived from primary data, it can nevertheless provide illustrations of how empirical data can be used in the process of formalization.

In the literature examples we collected, we find two main approaches of how primary data are used to derive behavioral model rules: qualitative translations of empirical findings and formalizations built on a framework based on existing theories, meta-theories, or other conceptualizations. For the first approach, behavioral model rules are derived ad hoc on the basis of empirical findings, i.e., empirical findings are formulated into a heuristic framework that is qualitatively translated into behavioral model rules, yet without providing explicit reasons for the interpretation beyond own observations and a general understanding of agronomic and economic processes. As this approach is based on qualitative input data, it is closely linked to the approach taken in path 1 (i.e., inductive reasoning to collect qualitative data without reference to a specific theory anchored in the literature). This also reflects that in our literature sample, all examples that derive behavioral model rules in this way follow path 1 to collect empirical data (Millington et al. 2008, Tieskens et al. 2017, Yao et al. 2018, Burg et al. 2021). For models that base the derivation of behavioral model rules on a conceptual framework, we observe different approaches related to the paths that were used for data collection. For models where empirical data are obtained through deductive reasoning with quantitative approaches (paths 2 and 3), behavioral model rules are based on the framework that was used to structure the questionnaire for data collection. For example, in our sample that is based on a theoretical framework (path 2), farmers’ behavior is formalized in a way to match the theory of planned behavior (Kaufmann et al. 2009, Pouladi et al. 2019, Noeldeke et al. 2022) or the belief-desire-intention model (Liu et al. 2021). Kaufmann et al. 2009, for example, calculate the intention to switch between conventional and organic farming as a weighted sum of attitude, subjective norm, and perceived behavioral control. Pouladi et al. 2019 and Noeldeke et al. 2022 perform a statistical analysis using a structural equation model to derive the relationship among structural parameters of the theory of planned behavior. Liu et al. 2021 weighted agents’ belief, desire and intention according to the statistical analysis of farmer characteristics, risk attitude, and environmental awareness. For the four example models following path 3 (i.e., quantitative data collection based on deductive reasoning without an explicit theory behind), behavioral model rules are the translation of a conceptual framework based on a systematic review of empirical literature (Valbuena et al. 2010a, 2010b) or a statistical analysis of the collected data using Bayesian belief networks (Sun and Müller 2013) or the association rule (Pouladi et al. 2020) to derive patterns of farmer behavior. In cases where data collection was not based on assumptions for behavior (path 1; i.e., no theory or framework was used to develop the interview questions), we find that four of the five models were afterwards tested for compatibility with an existing framework that comprises the basis for the behavioral model rules. Zagaria et al. 2021 apply an extensive qualitative content analysis of the transcribed interview data and structure their empirical data using the MoHuB framework (Schlüter et al. 2017). This approach helps them to link their findings on the decision process to different behavioral theories, which then build the basis for the formalized model rules. Acosta et al. 2014, Mertens et al. 2018, and Pacilly et al. 2019 argue that the elements found in the analysis of their empirical data (i.e., repetition, deliberation, imitation, social comparison) are well aligned with the concept of CONSUMAT approach (Jager 2000, Jager and Janssen 2012), which is based on different psychological theories and incorporates components such as uncertainty, satisfaction behavior, habits, and influence of others. Therefore, they base their behavioral model rules on this meta-theoretical framework.

We also find that primary data are often used to derive farmer typologies. For different farmer types, either different behavioral rules or different parameterizations are assumed (Rounsevell et al. 2012, Huber et al. 2018). If agents vary in their decision rules, behavioral paths that farmers can follow are defined by their characteristics. In our sample, this approach was taken by two studies that collected primary data in a qualitative way (path 1). Millington et al. 2008, for example, derived two types of farmers (commercial and traditional), who have different options for land management. Acosta et al. 2014 formalize farmers’ behavior in a decision tree where farmer characteristics, such as farm size, education, or income, determine which type of land management is performed. In models where farmer types follow the same behavioral rules but are parameterized differently, only their probabilities for certain behaviors differ. For such probabilistic approaches, quantitative data need to be collected. Valbuena et al. 2010a used this to determine decisions on expansion and diversification. Farmer types in Tieskens et al. 2017 differ in their likelihood to expand or reduce their farm area as well as the probability to participate in agri-environmental schemes at the beginning or during the course of the simulation and to continue participation in agri-environmental schemes. Kaufmann et al. 2009 clustered farmers by their attitude, subjective norm, perceived behavioral control, and the respective levels of uncertainty as derived from the empirical data. For each farmer type, the mean value of these characteristics was used to parameterize the behavioral model rules in the ABM.

For the derivation of farmer types, lack of transparency in reporting methods and data has been criticized (Bartkowski et al. 2022). In our sample, we found that formalization of behavioral model rules was described in a particularly non-transparent manner. Especially for models that are not built on a predefined framework (path 1), it often remains unclear how empirical evidence is accounted for in the model. In models where the decision rules are based on theories or conceptual frameworks, the formalization of the behavioral model rules is often clearer because the basic assumptions are known. Here, it is particularly necessary to argue why certain assumptions are supported by the (subsequently collected) empirical data. Especially if a rather abstract theory is used as a basis, an explanation is needed of how the theory is interpreted and how the empirical observations are linked to the underlying theoretical framework (Muelder and Filatova 2018). This involves documenting the model formalization following a protocol such as the ODD+D (Müller et al. 2013) or a documentation with a particular focus on data use such as ODD+2D (Laatabi et al. 2018) or the RAT-RS reporting standard (Achter et al. 2022). However, these protocols mostly only include the description of the final model version but seldomly an explanation how this version was reached. For a fully transparent description of model development, it is necessary to explain how the formalization was derived, especially if there is no existing (theoretical) framework underlying the chosen representation of human behavior, and being realistic about what empirical data can provide and what their limitations are (Edmonds 2015). If assumptions are made ad hoc and are not based on empirical data (e.g., because relevant data are difficult to obtain), this does not necessarily reduce the quality of the model. In particular, models that are not based on empirical data may even be more flexible in terms of including possible influences than empirically driven models and might therefore show a larger range of (potential) outcomes. In addition, care must be taken to ensure that the consideration of all available data does not lead to the model becoming too complex, which can lead to a lack of understanding, especially if the model is used for policy advice (Sun et al. 2016, Zellner et al. 2022). In any case, when including both empirical input and assumptions, it must be made clear which behavioral model rules are based on empirical data and which are not. Furthermore, sensitivity analysis is particularly important for the ad hoc parts of the model to illustrate the implications of certain choices (for suitable sensitivity analyses for ABMs see Ligmann-Zielinska et al. 2014, ten Broeke et al. 2016, or Troost et al. 2023). A thorough sensitivity analysis can provide a deep understanding of the potential solution space resulting from different input parameters. Policy analyses can benefit from these detailed results as they can reveal unexpected outcomes (e.g., extreme cases) that may not be contained in empirical data. In addition, a documentation of alternative model variants that were considered but not chosen for the final model version is helpful to follow the complete process of model development. By documenting all steps, the iterative character of model development becomes clear. No model is built all at once, but it always includes step-by-step development of system understanding, which is influenced by empirical observations and (theoretical) assumptions (more details on the gradual model development are included in the concept of the modeling cycle, see e.g., Augusiak et al. 2014 for ecological models). This documentation can be included in a modeling notebook such as TRACE (Schmolke et al. 2010, Grimm et al. 2014, Ayllón et al. 2021). Finally, we strongly advocate that the questionnaire or semi-structured guiding questions used to collect the empirical data are made public. This is the only way to explicitly understand which questions provided which model-relevant information.

EXEMPLARY FORMALIZATION OF FARMERS’ BEHAVIOR ON THE ADOPTION OF AGRI-ENVIRONMENTAL SCHEMES

To illustrate the process of model formalization based on primary data and outline possible difficulties, we draw on our own experience in the BESTMAP project (Ziv et al. 2020) where we developed an empirically grounded ABM that includes a decision-making framework for the adoption of agri-environmental schemes (AES) in Europe. AES are voluntary programs offered as part of the European Union’s Common Agricultural Policy to provide incentives for environmentally friendly farming practices (European Union 2013a). To systematically test the impact of different policy designs on the adoption rate among farmers and the resulting spatial allocation of AES, we designed an ABM where decisions of individual farmers on four selected schemes are explicitly included. In the following, we describe our approach of data collection, key elements included in the model, model formalization, and parameterization, and highlight the lessons we learned from our approach. Our own experiences are not intended to serve as a particular example of best practice. Rather, we would like to show one potential approach but also highlight shortcomings and difficulties we have encountered. At best, this will help to ensure that certain aspects are taken into account from the outset in future projects, which could lead to behavioral model rules being based on empirical data to an even greater extent than in our case.

Approach: semi-structured interviews and follow-up survey

To formalize and parameterize behavioral model rules, we applied a two-step approach. First, we conducted semi-structured interviews (path 1) that we used to derive behavioral model rules. In a second step, we performed an online survey to parameterize the model (path 3). When designing the interview campaign, we applied an inductive approach with no prior assumptions on the adoption of agri-environmental schemes. Nevertheless, any study (even if semi-structured) needs a starting point to guide the interview into the desired direction. In our case, we first collected possible influence factors in an extensive literature search to capture the most important processes of the decision-making on AES. We considered reviews that specifically focus on AES (Lastra-Bravo et al. 2015, Brown et al. 2021) and others that give a general overview on factors affecting the adoption of sustainable farming practices (Dessart et al. 2019) and agricultural soil management (Bartkowski and Bartke 2018). Based on the literature search, we derived themes and questions that provided the basis for an interview campaign with farmers in five case study (CS) regions across Europe (Czech Republic, Germany, Serbia, Spain, and the United Kingdom). We chose semi-structured face-to-face interviews as a method to identify potential key factors that influence the decision-making process. Data collection consisted of two parts: (1) a qualitative interview, based on an interview protocol, covering open questions on the farmer’s background, attitudes toward farming, reflection on ecological aspects, and especially the motivation to apply, or not apply, for AES; and (2) a standardized questionnaire, focusing on background information on the farm, information on environmentally sustainable practices, concrete experiences with two selected AES most common in the respective CS, motivation to apply for AES, and opinions on the EU’s Common Agricultural Policy in general. Whereas the interviews enabled an in-depth dialogue about farmers’ decisions on AES and their practical experiences and needs, the focus of the questionnaire was on closed questions, mostly using quantitative scales and answer categories. Across all case studies, a total of 124 interviews were conducted. The identification of overarching and more CS-specific factors was based on qualitative content analysis of transcripts (Schreier 2012) by using a coding frame informed both by our literature search and a first set of interviews, CS summary reports provided by the interview teams, as well as statistical exploration of the questionnaire data (see Wittstock et al. 2022 for an analysis of the results of the German case study and Bartkowski et al. 2023 for a case study comparison). In order to be able to evaluate the relative importance of all factors, local case study experts with necessary context knowledge about the local circumstances of farmers’ decision-making on AES were consulted (see Table 3 for a summary of the importance of all factors). A key observation from the interviews was that farmers face a sequence of decision-making elements for AES participation. Based on the empirical findings, Wittstock et al. 2022 derived a heuristic framework in which the decision-making process is divided into three elements: (1) a farm-specific decision-making context, (2) a leeway in decision-making that is restricted by AES administration in a given funding period, and (3) more specific decision-influencing factors that influence if and how a particular plot is devoted to a particular scheme.

Despite the quantitative component, our interview campaign provided limited information to parameterize model rules and infer differences in behavior between farmer types. Therefore, we decided to conduct a follow-up online survey consisting of mostly closed-ended questions and a discrete choice experiment (DCE) (see Appendix 1 for the complete questionnaire). With the survey, we covered in particular farm characteristics (including questions on the specialization of the farm), personal values (covering questions on attitudes toward the environment, societal influence, and exchange with other farmers), and socio-demographic questions. In addition, we included experience with existing AES and asked for reasons why farmers did or did not participate in specific schemes. In the DCE, respondents had to choose between four alternative AES and a “no scheme” option where farmers would not get any funding for agri-environmental practices. In addition to the offered payment level, we chose contract length, bureaucratic effort, and advisory support as key attributes to be varied between the schemes, i.e., those aspects mentioned as most important with respect to contract design in the interviews (cf. Table 3). By using a combination of the results of the DCE and the survey questions, our aim was to derive the expected payment level (“willingness to accept”) for each AES and farmer type.

Key elements: decision factors included in the ABM

For a description of the main aspects included in the ABM, we follow the key elements of primary data for informing ABMs, namely (1) heterogeneity between farmers with a focus on farmer types; (2) temporal aspects, in particular the sequential decision-making; and (3) interaction between farmers, in our case specifically social influence. Because of missing data that could also not be collected in the survey, we could not, however, include all aspects considered relevant in the framework developed in Wittstock et al. 2022 that forms the basis for our formalization (the third column in Table 3 indicates if the aspects were included; see footnotes for justifications).

Heterogeneity: farm types

From the interviews, we learned that a good fit with established farm practices is one of the key factors in whether or not a farm participates in AES. Within the project, a generalized typology of farming systems (farming system archetypes) that are assumed to have similar responses to policy change has been developed (Langerwisch et al. 2021). With respect to farm practices, farming system archetypes are distinguished by their farm specialization (general cropping, horticulture, permanent crops, and grazing livestock, or “mixed” if not at least 2/3 of the total farm area is dedicated to one of the corresponding land use types). Farm size was found to mainly positively influence the adoption of AES (Paulus et al. 2022, Wittstock et al. 2022). The farming system archetypes are therefore further divided by economic size, i.e., the economic output calculated using crop specific standard output coefficients (EUROSTAT 2017) multiplied by the field size (four groups: < 2000 EUR, small, medium, and large). We also leave the possibility to differentiate between organic and conventional farms to account for perceptions about the environment. However, because some AES cannot be combined with funding for organic farming, organic farming sometimes reduces the participation in AES. We therefore did not include the effect of organic farming on AES adoption in the farm types so far.

Temporal processes: sequential decision-making

We account for the sequential process of decision-making as described in the heuristic framework developed in Wittstock et al. 2022 as key concept for the decision-making process. Additional aspects with a temporal component were found to be past experience with AES as well as the duration of AES contracts. Both are included in the behavioral model rules. The duration of tenure contracts as further temporal aspect would, however, need to be available on a field-specific basis to accurately account for which fields are suitable for AES and which not. This factor is not captured in the data because of privacy issues.

Interaction: social influence

In the interviews, interaction was mentioned with respect to three instances. First, tenant-owner relationship was highlighted as a limiting factor because tenants might not be in accordance with adoption of AES on their fields. However, information on tenure would need to be available for all farmers and could not be extrapolated from a limited sample. Secondly, farmers indirectly interact with others in ways that might externally influence AES outcomes, such as dog walkers who do not respect the presence of the schemes (Wittstock et al. 2022). However, because this interaction is mostly relevant for results-based schemes that are not covered with the ABM, it is not included. Finally, although social influence was not reported as being important for the decision on AES, we decided to include the factor given our skepticism about the explanatory power of the interviews in that context.

Translation into decision rules: formalization of farmers’ behavior

With respect to the specific conceptualization of the model, we found that no behavioral theory or framework (such as CONSUMAT) includes all factors that were considered important for the decision to adopt AES in the heuristic framework concluded from our interviews. Therefore, we decided to derive our own formalization based on the heuristic framework that can be adapted to peculiarities in the different case studies, e.g., by allowing us to switch on or off some components that are more or less important in some case studies (Table 4 for an overview of the resulting decision-making framework).

In their framework, Wittstock et al. 2022 first account for the decision-making context in which farmers consider AES. This first step does not focus on economic considerations but rather includes preceding considerations and restrictions before taking the actual decision on AES adoption. Wittstock et al. 2022 argue that farmers have legal and contractual land use commitments, make non-AES related land use decisions, and are restricted by workflows corresponding to their farm type and size. Because our model focuses on the decision of AES adoption and therefore does not cover the entire farm management, we do not explicitly model these decisions. Instead, we decided to summarize this into an identity-driven decision at farm level on whether the farmer is open to considering the adoption of a specific AES. In order to nevertheless take into account the diversity of influencing factors, we consider openness to be composed of several aspects. First, we assume that farmers have an intrinsic motivation to participate in AES. Second, we consider that farmers’ perceptions about the relevance of AES might change over time depending on their own experience and external circumstances. Although perceptions about the environment and related changing climatic conditions were not found to influence the decision-making with respect to the adoption of AES, in the interviews it was observed that farmers who previously adopted a scheme are likely to continue with that practice in subsequent funding periods. However, some also reported negative experiences that caused them to refrain from participating. We account for these observations in a probability with higher chances of being open for participation with previous participation. The last aspect that was reported to decrease the general willingness to adopt AES was missing knowledge, e.g., regarding administrative issues or implementation. With regard to this, we assume that farmers with higher knowledge are more likely to consider AES. Although perceived corruption may also discourage farmers from participating, this factor was mentioned in only a few case studies and therefore was not considered in the overall decision framework. We assume that knowledge can be generated through advisory support or by exchange in social networks. We summarized these aspects in a stepwise process to derive openness based on probabilities for being open due to the different influencing factors (intrinsic openness, prior experience, advisory support, experience in social network). Using a probabilistic approach, we accounted for unexpected behavior not captured in the model rules. Especially because we could not explicitly include the decision-making steps expressed in the interviews, this provides the required flexibility.

According to Wittstock et al. 2022, more practical aspects like availability of funding for specific AES or the eligibility of certain areas for specific AES in the computer-based AES management system further restrict the participation in AES (referred to as “leeway in decision-making” in the heuristic framework). Although we do not have precise data on the availability of schemes in the application software, we nevertheless included that farmers consider if they have suitable land for the respective schemes. This comprises checking if they have the required land type for a specific scheme (e.g., arable or grassland) and land that is not occupied by other funding schemes, such as ecological focus areas that are not compatible with AES. We also leave the possibility open of restricting the selection of suitable fields according to a minimum field size, which is sometimes given as a contract detail. In this decision step, we closely adhere to the heuristic framework in terms of restricting the decision-making to external constraints.

For the final step, Wittstock et al. 2022 conclude that a range of factors influence the actual decision of participating in AES (namely they distinguish economic, routine-related, biophysical, geographic, legal, ownership-related, policy-related, and administrative factors). Again, we were not able to explicitly include all of these aspects. Instead, we focus on the particularly high importance of economic factors for the decision on AES adoption (Bartkowski et al. 2023). For the ABM, we therefore base the last decision step on an economic decision. To account for individual circumstances on a farm, we assume an individual expected payment for each AES and farm type. Specifically, we assume that farmers might be willing to accept a lower compensation if the circumstances for a scheme better fit their expectations (for example, with respect to the administrative effort for applying and maintaining a scheme). We assume that a scheme is adopted if the payment offered as defined in the policy design is at least as high as the amount expected. With respect to the final decision on where to implement a scheme, we assume that biophysical characteristics such as soil quality or distance from the farm are the primary factors, given the importance of biophysical factors in the interviews.

Challenges

Although in principle the two-step approach has shown to be a valuable combination of inductive data collection to get a general model understanding and formalize behavioral model rules (path 1) and a deductive approach with which the developed decision framework can be further specified and parameterized (path 3), we have encountered difficulties in particular with representing farmer heterogeneity. First, the empirically derived heuristic framework (Wittstock et al. 2022) included a broad range of factors influencing the behavior on AES adoption. Many of these were related to available funding for AES on particular plots, but data for the underlying computer-based management system are not publicly available and could therefore not be included in the ABM. Similarly, tenure-related aspects and legal constraints were mentioned as critical to AES participation, but information on tenure agreements is not made public and could also not be asked for in the survey because this information needs to be available for all fields. Here we had to make a reasonable decision on which factors to omit. We solved the problem by keeping the general structure of the heuristic framework, i.e. a three-stage decision process, and including only those factors for which data were available or could in principle be collected in the follow-up survey. Because we omitted a number of factors with this approach, we applied a probabilistic decision rule to allow for behaviors not explicitly covered in the model rules.

With respect to the follow-up survey, we decided to design the questionnaire as an online survey to reach as many farmers as possible and distributed the link through farmer associations and contacting farmers directly. In addition, we incentivized participation in the survey with the opportunity to take part in a raffle. However, with 381 completed responses across case studies (sample sizes ranging from 69 for the Czech Republic to 131 for Serbia; for Spain we received only 22 completed responses, which we decided not to evaluate because of lack of representativeness), we still had to deal with a low sample size. With between 50% and 77% of respondents quitting the survey before reaching the end, we furthermore had a rather large drop-out rate. Many farmers reported that the survey was too long and some participants perceived the DCE as too stylized, which might have led to some not finishing the survey and in particular the DCE. Because a statistically sound analysis of a DCE requires a substantial number of responses (especially if more detailed analyses, such as latent class analysis, are to be conducted to uncover hidden clusters among farmers), this limits the practical use of the DCE for parameterization. Furthermore, there are no alternative statistical approaches to meaningfully analyze the DCE part of the survey. These difficulties prevented us from realizing the full potential of a combined interview and follow-up survey approach. Accounting for these difficulties would have meant repeating the survey in a different way (e.g., collecting data that provide reliable results with smaller sample sizes) or involving stakeholders in a way that was not originally planned. Unfortunately, given the tight time frame and limited financial resources of scientific projects, we were unable to apply these steps to our own research. As a result, we had to resort to alternative approaches, such as pattern-oriented modeling (Grimm et al. 2005) and sensitivity analyses especially with respect to parameterizing the expected payment level in the third step of the decision-making framework. This leads to analyses that are less empirically based and therefore have less potential to support policy. We are still able to show a range of possible outcomes, but the results are subject to a greater uncertainty than those which are parameterized with more robust quantitative empirical data.

DISCUSSION

Underpinning farmers’ behavior in agent-based models with empirical knowledge is increasingly important especially when these models are used for assessing agricultural policies (Kremmydas et al. 2018). However, although substantial amounts of empirical data are collected among farmers, it is often unclear how the data can be used in models. Based on insights from literature examples, we presented three pathways of how primary data can be collected among farmers and three main types of information that need to be covered to derive behavioral rules for an ABM. Below, we discuss the specific challenges related to these aspects with a particular focus on problems we faced in developing a model for farmers’ decision-making in the adoption of agri-environmental measures.

In terms of the three paths, we conclude that there is not one path that works best in every situation. Depending on the problem and the specific context, each approach has its advantages but also comes with a number of challenges. Path 1 (i.e., a qualitative approach not based on an established theory) is adaptable to the specific needs of a research question and therefore well suited to exploring unfamiliar contexts, but it requires a lot of assumptions in the formalization process. Nevertheless, such an approach has the potential to well represent hidden processes that might get lost in other approaches. With respect to model-based assessment of agricultural policies, it may therefore be able to capture a wide range of mechanisms and thus highlight diverse policy implications. In many cases, however, models that are solely based on primary data collected following path 1 are parameterized by assumptions or their parameters are varied in sensitivity analyses. In both cases, this increases the uncertainty of the results compared to a parameterization with quantitative data. Models following this path can therefore mainly show the possible space of outcomes after the introduction of a policy. A theory-driven approach where quantitative data are collected (path 2) is the least ad hoc and most transparent approach, but, because of the framework of the theory, it is also the least flexible approach. Yet, the immediate transparency that one expects in a theory-guided study only occurs if the theoretical concept is adequately considered in data collection (see Sok et al. 2020 for an overview of the use of the theory of planned behavior in empirical studies in agriculture and the difficulties that arise when the theory is only loosely followed). Moreover, when using theory-based approaches for policy evaluation, it might be more difficult to convince stakeholders of the potential of such a formalized approach, which may appear overly scientific (in the sense of being detached from reality), particularly in the context of farmer behavior. A quantitative approach that is not based on theory (path 3) can provide insightful contributions as a stand-alone approach if a rich literature on (related) contexts is available from which factors and relationships (i.e., implicitly an underlying framework) can be derived. In the absence of such literature, formalization assumptions drawn from this approach can quickly appear ad hoc. In such cases, transparency about the choices made in the design of both the data collection and, ultimately, the model is particularly important. This also needs to be considered when applying this approach for policy assessment. Here, in particular, a combination of path 1 and path 3 can help overcome the challenges of the individual approaches and provide a solid basis for evaluating agricultural policies.

With respect to the main elements that need to be collected with primary data to support ABM construction, we find that heterogeneity between farmers can be derived rather easily because many approaches are available to obtain socio-demographic data, farm characteristics, as well as characteristics such as values or perceptions that might differ between farmers. Especially for the latter two types of data, we argue that the use of standardized survey instruments and constructs can help make different studies comparable. Information about the social environment and interactions among farmers might be more difficult to identify using primary data and complementary indirect analyses may be required. Even more difficult is the collection of information about temporal processes and their translation to a dynamic ABM. In real life, many processes are not as clear-cut as they need to be for a questionnaire. Here, concepts such as the consumer journey mapping (Rosenbaum et al. 2017) where stages of decision-making are disentangled might help to structure the process. However, this would add additional layers of complexity to an already extensive amount of data required.

For our own implementation, we decided to use a combination of an inductive qualitative interview campaign and a follow-up survey. Although this allowed us to tailor the questionnaire specifically to our needs, we also encountered difficulties. In particular, we had to deal with an aspect that is often problematic in surveys among farmers: small sample sizes and moderate representativeness. Acquiring sufficient participants and covering a representative share is particularly problematic in surveys for ABMs, which are usually designed for spatially limited case studies where the total population of farmers is not too large. In addition, we faced high drop-out rates in the survey. To avoid this, a trade-off between the high level of detail needed for parameterizing the ABM and cognitively demanding surveys (such as in our case the DCE where farmers repeatedly had to decide on similar situations) with high risk of drop out or unreliable responses must be found. Since farmers are short on time (especially in certain seasons) and filling out extensive questionnaires can take a long time, this might cause them not participating in the survey, abandoning it before it is completed or, which may be the worst possibility, filling it quickly in an inconsistent and unreflected manner. Therefore, it is important to prioritize the importance of getting specific information and to consider alternative data sources. To limit the number of farmer surveys, surveys can be supplemented with expert knowledge gained through focus group discussions or workshops. Such participatory modeling approaches comprise their own field of research and have been successfully applied to a range of case studies (see, e.g., Voinov et al. 2018 for an overview of methods). For relationships that are assumed to be less decisive, it might also be appropriate to assume ad hoc rules rather than include additional questions in a questionnaire (and thus making it longer). This does not need to affect the validity of the model as long as it is explained explicitly and the influence of the model rule is tested (e.g., in a sensitivity analysis).

With an increasing amount of studies targeting farmers and given that many questions in surveys are similar (especially regarding farm characteristics and socio-demographic properties), it should be considered whether results from previous surveys can be used or surveys on different topics can be combined. To achieve this goal, it is important to make the survey results publicly available to enable sharing of existing data across research groups (ensuring the necessary data protection). Similar to re-using survey data, also applying the same model concepts to several case studies reduces the effort for collecting empirical data. Although data would still be needed to parameterize the model, it would be assumed that the behavioral model rules remain structurally the same as, for instance, done by Chen et al. 2022 for European viticultural practices. In a similar vein, Murray-Rust et al. 2014 developed the agent-based modeling framework Aporia with the aim to reduce the difficulty with creating land use models. Going even further, Huber et al. 2021 developed the FARMIND framework that allows integrating different bioeconomic farm models in the same generic behavioral ABM (drawing upon the CONSUMAT framework). ABMs that cover the entire farm sector instead of focusing on specific decisions (such as the adoption of AES) can be more easily adapted to address different research questions. Examples in this regard are the AgriPoliS model, which has been calibrated with empirical data for different regions (Brady et al. 2009, Hristov et al. 2020) or the ALUAM-AB model that was developed to map land use changes influenced by market and policy dynamics and has, for example, been applied in the context of payment for environmental services (Huber et al. 2013), land abandonment and reforestation (Brändle et al. 2015), and resilience of social-ecological systems to global change (Grêt-Regamey et al. 2019). However, to develop such a comprehensive ABM, a large team with dedicated long-term funding is needed. Current funding schemes often do not offer such opportunities in a structured way.

In addition, secondary data (i.e., data processed from primary data) can serve as a valuable source of information, complementing primary data collection. Examples of potential secondary data sources for agricultural ABMs are the European Union’s farm accountancy data network (FADN) (European Union 2013b) (to be extended to a farm sustainability data network, FSDN) or the Integrated Administration and Control System (IACS) (European Union 2013c). Whereas the former records agricultural income and farm activity in a representative but aggregated manner, the latter contains farm types, received subsidies, livestock and spatial information on parcels such as size and agricultural use in anonymized form. Zimmermann et al. 2015, for example, used FADN data to parameterize farmers in an ABM of the Swiss agricultural sector. Paulus et al. 2022 analyzed the influence of farm structure and landscape characteristics on the adoption of agri-environmental schemes using spatially explicit LPIS data of the Mulde region in Germany. The identified regression models describe farmers’ behavior in terms of the likelihood of a farm adopting AES, the selection of fields to which these schemes are applied, and the area that farmers consider to put under AES, and can be translated into behavioral model rules of an ABM. However, limitations for using secondary data are that reasons for behavior are not explicitly covered and representativeness is not given with respect to all relevant determinants (e.g., FADN data are representative with respect to FADN region, economic size and type of farming but not necessarily with respect to other dimensions [Neuenfeldt and Gocht 2014]).

CONCLUSION

To effectively use agent-based modeling to support agricultural policy, the appropriate incorporation of farmer behavior is crucial. Although it is generally acknowledged that model assumptions must be empirically supported to adequately represent realistic relationships, how empirical observations are translated into behavioral model rules is often not explicitly described. To structure the process of formalizing farmers’ behavior in ABMs using primary data, we divided it into three main steps, for each of which we discussed different approaches using literature examples. In particular, we focused on the selection of an appropriate empirical method, the design of the study to best inform an ABM, and the actual formalization of the empirical data into behavioral model rules. We highlighted the strengths and weaknesses of different strategies for collecting empirical data, particularly for use in ABMs. In addition, we elaborated that empirical data must at least cover heterogeneity among farmers, temporal aspects, and interaction in order to adequately formalize farmers’ decision-making in an ABM with primary data. Although it would be desirable to formulate exhaustive guidelines for translating empirical findings into decision rules in order to make the formalization of behavioral model rules more systematic, we see no way to do so at present given the wide variety of empirical contexts. Nevertheless, we take a first step in this direction by highlighting both successful examples and shortcomings in communicating formalization. In this way, we provide an overview from which readers can learn and draw inspirations from own efforts. If future formalizations of human behavior take into account the elements we consider important, they could also form the basis for guidelines that we have not yet been able to derive from the current state of the literature. By incorporating our own experience with a model on agri-environmental schemes in Europe, we were also able to point out particular difficulties that can arise on the way to formalizing a model based on empirical observations. We hope that our methodological contribution will help to strengthen the empirical embedding of farmers’ behavior into ABMs. This is ultimately critical to using ABMs as a reliable tool for effective policy advice in the agricultural sector.

LITERATURE CITED

Achter, S., M. Borit, E. Chattoe-Brown, and P-O. Siebers. 2022. RAT-RS: a reporting standard for improving the documentation of data use in agent-based modelling. International Journal of Social Research Methodology 25(4):517-540. https://doi.org/10.1080/13645579.2022.2049511

Acosta, L. A., M. D. A. Rounsevell, M. Bakker, A. V. Doorn, M. Gómez-Delgado, and M. Delgado. 2014. An agent-based assessment of land use and ecosystem changes in traditional agricultural landscape of Portugal. Intelligent Information Management 6:55-80. https://doi.org/10.4236/iim.2014.62008

Ajzen, I. 1991. The theory of planned behavior. Organizational Behavior and Human Decision Processes 50(2):179-211. https://doi.org/10.1016/0749-5978(91)90020-T

An, L. 2012. Modeling human decisions in coupled human and natural systems: review of agent-based models. Ecological Modelling 229:25-36. https://doi.org/10.1016/j.ecolmodel.2011.07.010

An, L., V. Grimm, A. Sullivan, B. L. Turner II, N. Malleson, A. Heppenstall, C. Vincenot, D. Robinson, X. Ye, J. Liu, E. Lindkvist, and W. Tang. 2021. Challenges, tasks, and opportunities in modeling agent-based complex systems. Ecological Modelling 457:109685. https://doi.org/10.1016/j.ecolmodel.2021.109685

Arneth, A., C. Brown, and M. D. A. Rounsevell. 2014. Global models of human decision-making for land-based mitigation and adaptation assessment. Nature Climate Change 4:550-557. https://doi.org/10.1038/nclimate2250

Augusiak, J., P. J. Van den Brink, and V. Grimm. Merging validation and evaluation of ecological models to ‛evaludation’: a review of terminology and a practical approach. 2014. Ecological Modelling 280:117-128. https://doi.org/10.1016/j.ecolmodel.2013.11.009

Ayllón, D., S. F. Railsback, C. Gallagher, J. Augusiak, H. Baveco, U. Berger, S. Charles, R. Martin, A. Focks, N. Galic, et al. 2021. Keeping modelling notebooks with TRACE: good for you and good for environmental research and management support. Environmental Modelling & Software 136:104932. https://doi.org/10.1016/j.envsoft.2020.104932

Bakker, M. M., S. J. Alam, J. van Dijk, and M. D. A. Rounsevell. 2015. Land-use change arising from rural land exchange: an agent-based simulation model. Landscape Ecology 30:273-286. https://doi.org/10.1007/s10980-014-0116-x

Bartkowski, B., and S. Bartke. 2018. Leverage points for governing agricultural soils: a review of empirical studies of European farmers’ decision-making. Sustainability 10(9):3179. https://doi.org/10.3390/su10093179

Bartkowski, B., C. Schüßler, and B. Müller. 2022. Typologies of European farmers: approaches, methods and research gaps. Regional Environmental Change 22:43. https://doi.org/10.1007/s10113-022-01899-y

Bartkowski, B., M. Beckmann, M. Bednář, S. Biffi, C. Domingo-Marimon, M. Mesaroš, C. Schüßler, B. Šarapatka, S. Tarčak, T. Václavík, G. Ziv, and F. Wittstock. 2023. Adoption and potential of agri-environmental schemes in Europe: cross-regional evidence from interviews with farmers. People and Nature 5:1610-1621. https://doi.org/10.1002/pan3.10526

Baumgärtner, S., C. Becker, K. Frank, B. Müller, and M. Quaas. 2008. Relating the philosophy and practice of ecological economics: the role of concepts, models, and case studies in inter- and transdisciplinary sustainability research. Ecological Economics 67(3):384-393. https://doi.org/10.1016/j.ecolecon.2008.07.018

Bell, A. R. 2017. Informing decisions in agent-based models – a mobile update. Environmental Modelling & Software 93:310-321. https://doi.org/10.1016/j.envsoft.2017.03.028

Bocquého, G., F. Jacquet, and A. Reynaud. 2013. Expected utility or prospect theory maximisers? Assessing farmers’ risk behaviour from field-experiment data. European Review of Agricultural Economics 41(1):135-172. https://doi.org/10.1093/erae/jbt006

Boero, R., and F. Squazzoni. 2005. Does empirical embeddedness matter? Methodological issues on agent-based models for analytical social science. Journal of Artificial Societies and Social Simulation 8(4):6.

Bonke, V., O. and Musshoff. 2020. Understanding German farmer’s intention to adopt mixed cropping using the theory of planned behavior. Agronomy for Sustainable Development 40:48. https://doi.org/10.1007/s13593-020-00653-0

Borges, J. A. R., and A. G. J. M. Oude Lansink. 2016. Identifying psychological factors that determine cattle farmers’ intention to use improved natural grassland. Journal of Environmental Psychology 45:89-96. https://doi.org/10.1016/j.jenvp.2015.12.001

Brady, M., K. Kellermann, C. Sahrbacher, and L. Jelinek. 2009. Impacts of decoupled agricultural support on farm structure, biodiversity and landscape mosaic: some EU results. Journal of Agricultural Economics 60(3):563-585. https://doi.org/10.1111/j.1477-9552.2009.00216.x

Brändle, J., G. Langendijk, S. Peter, S. Brunner, and R. Huber. 2015. Sensitivity analysis of a land-use change model with and without agents to assess land abandonment and long-term re-forestation in a Swiss mountain region. Land 4(2):475-512. https://doi.org/10.3390/land4020475

Brown, C., E. Kovács, I. Herzon, S. Villamayor-Tomas, A. Albizua, A. Galanaki, I. Grammatikopoulou, D. McCracken, J. A. Olsson, and Y. Zinngrebe. 2021. Simplistic understandings of farmer motivations could undermine the environmental potential of the common agricultural policy. Land Use Policy 101:105136. https://doi.org/10.1016/j.landusepol.2020.105136

Bruch, E., and J. Atwell. 2015. Agent-based models in empirical social research. Sociological Methods & Research 44(2):186-221. https://doi.org/10.1177/0049124113506405

Burg, V., K. G. Troitzsch, D. Akyol, U. Baier, S. Hellweg, and O. Thees. 2021. Farmer’s willingness to adopt private and collective biogas facilities: an agent-based modeling approach. Resources, Conservation and Recycling 167:105400. https://doi.org/10.1016/j.resconrec.2021.105400

Burton, R. J. F. 2004. Seeing through the ‛good farmer’s’ eyes: towards developing an understanding of the social symbolic value of ‛productivist’ behaviour. Sociologia Ruralis 44(2):195-215. https://doi.org/10.1111/j.1467-9523.2004.00270.x

Caffaro, F., M. Roccato, M. Micheletti Cremasco, and E. Cavallo. 2019. An ergonomic approach to sustainable development: the role of information environment and social-psychological variables in the adoption of agri-environmental innovations. Sustainable Development 27(6):1049-1062. https://doi.org/10.1002/sd.1956

Chen, Y., R. A. Herrera, E. Benitez, C. Hoffmann, S. Möth, D. Paredes, E. Plaas, D. Popescu, S. Rascher, A. Rusch, et al. 2022. Winegrowers’ decision-making: a pan-European perspective on pesticide use and inter-row management. Journal of Rural Studies 94:37-53. https://doi.org/10.1016/j.jrurstud.2022.05.021

Coelho, L. A. G., C. M. P. Pires, A. T. Dionísio, and A. J. da C. Serrão. 2012. The impact of CAP policy in farmer’s behavior – a modeling approach using the Cumulative Prospect Theory. Journal of Policy Modeling 34(1):81-98. https://doi.org/10.1016/j.jpolmod.2011.03.009

Constantino, S. M., M. Schlüter, E. U. Weber, and N. Wijermans. 2021. Cognition and behavior in context: a framework and theories to explain natural resource use decisions in social-ecological systems. Sustainability Science 16:1651-1671. https://doi.org/10.1007/s11625-021-00989-w

Crooks, A., C. Castle, and M. Batty. 2008. Key challenges in agent-based modelling for geo-spatial simulation. Computers, Environment and Urban Systems 32(6):417-430. https://doi.org/10.1016/j.compenvurbsys.2008.09.004

Despotović, J., V. Rodić, and F. Caracciolo. 2019. Factors affecting farmers’ adoption of integrated pest management in Serbia: an application of the theory of planned behavior. Journal of Cleaner Production 228:1196-1205. https://doi.org/10.1016/j.jclepro.2019.04.149

Dessart, F. J., J. Barreiro-Hurlé, and R. van Bavel. 2019. Behavioural factors affecting the adoption of sustainable farming practices: a policy-oriented review. European Review of Agricultural Economics 46(3):417-471. https://doi.org/10.1093/erae/jbz019

Dou, Y., J. D. A. Millington, R. F. B. Da Silva, P. McCord, A. Viña, Q. Song, Q. Yu, W. Wu, M. Batistella, E. Moran, and J. Liu. 2019. Land-use changes across distant places: design of a telecoupled agent-based model. Journal of Land Use Science 14(3):191-209. https://doi.org/10.1080/1747423X.2019.1687769

Dunlap, R. E., K. D. Van Liere, A. G. Mertig, and R. E. Jones. 2000. New trends in measuring environmental attitudes: measuring endorsement of the new ecological paradigm: a revised NEP scale. Journal of Social Issues 56:425-442. https://doi.org/10.1111/0022-4537.00176

Edmonds, B. 2015. Using qualitative evidence to inform the specification of agent-based models. Journal of Artificial Societies and Social Simulation 18(1):18. https://doi.org/10.18564/jasss.2762

Elsawah, S., T. Filatova, A. J. Jakeman, A. J. Kettner, M. L. Zellner, I. N. Athanasiadis, S. H. Hamilton, R. L. Axtell, D. G. Brown, J. M. Gilligan, et al. 2020. Eight grand challenges in socio-environmental systems modeling. Socio-Environmental Systems Modelling 2:16226. https://doi.org/10.18174/sesmo.2020a16226

Epanchin-Niell, R. S., D. B. Jackson-Smith, R. S. Wilson, M. Ashenfarb, A. A. Dayer, V. Hillis, G. D. Iacona, E. M. Markowitz, S. T. Marquart-Pyatt, and T. Treakle. 2022. Private land conservation decision-making: an integrative social science model. Journal of Environmental Management 302:113961. https://doi.org/10.1016/j.jenvman.2021.113961

European Union. 2013a. Regulation (EU) No 1305/2013 of the European Parliament and of the Council of 17 December 2013 on support for rural development by the European Agricultural Fund for Rural Development (EAFRD) and repealing Council Regulation (EC) No 1698/2005.

European Union. 2013b. Regulation (EU) No 1318/2013 of the European Parliament and of the Council of 22 October 2013 amending Council Regulation (EC) No 1217/2009 setting up a network for the collection of accountancy data on the incomes and business operation of agricultural holdings in the European Community.

European Union. 2013c. Regulation (EU) No 1306/2013 of the European Parliament and of the Council of 17 December 2013 on the financing, management and monitoring of the common agricultural policy and repealing Council Regulations (EEC) No 352/78, (EC) No 165/94, (EC) No 2799/98, (EC) No 814/2000, (EC) No 1290/2005 and (EC) No 485/2008.

EUROSTAT. 2017. Additional data. Agriculture. Standard output coefficients. https://ec.europa.eu/eurostat/web/agriculture/database/additional-data

Finger, R., D. Wüpper, and C. McCallum. 2022. The (in)stability of farmer risk preferences. Journal of Agricultural Economics 74(1):155-167. https://doi.org/10.1111/1477-9552.12496

Gaube, V., C. Kaiser, M. Wildenberg, H. Adensam, P. Fleissner, J. Kobler, J. Lutz, A. Schaumberger, J. Schaumberger, B. Smetschka, et al. 2009. Combining agent-based and stock-flow modelling approaches in a participative analysis of the integrated land system in Reichraming, Austria. Landscape Ecology 24:1149-1165. https://doi.org/10.1007/s10980-009-9356-6

Gray, S., A. Voinov, M. Paolisso, R. Jordan, T. BenDor, P. Bommel, P. Glynn, B. Hedelin, K. Hubacek, J. Introne, et al. 2018. Purpose, processes, partnerships, and products: four Ps to advance participatory socio-environmental modeling. Ecological Applications 28:46-61. https://doi.org/10.1002/eap.1627

Grêt-Regamey, A., S. H. Huber, and R. Huber. 2019. Actors’ diversity and the resilience of social-ecological systems to global change. Nature Sustainability 2:290-297. https://doi.org/10.1038/s41893-019-0236-z

Grimm, V., J. Augusiak, A. Focks, B. M. Frank, F. Gabsi, A. S. A. Johnston, C. Liu, B. T. Martin, M. Meli, V. Radchuk, P. Thorbek, and S. F. Railsback. 2014. Towards better modelling and decision support: documenting model development, testing, and analysis using TRACE. Ecological Modelling 280:129-139. https://doi.org/10.1016/j.ecolmodel.2014.01.018

Grimm, V., A. S. A. Johnston, H.-H. Thulke, V. E. Forbes, and P. Thorbek. 2020. Three questions to ask before using model outputs for decision support. Nature Communications 11:4959. https://doi.org/10.1038/s41467-020-17785-2

Grimm, V., E. Revilla, U. Berger, F. Jeltsch, W. M. Mooij, S. F. Railsback, H.-H. Thulke, J. Weiner, T. Wiegand, and D. L. DeAngelis. 2005. Pattern-oriented modeling of agent-based complex systems: lessons from ecology. Science 310(5750):987-991. https://doi.org/10.1126/science.1116681

Groeneveld, J., B. Müller, C. M. Buchmann, G. Dressler, C. Guo, N. Hase, F. Hoffmann, F. John, C. Klassert, T. Lauf, et al. 2017. Theoretical foundations of human decision-making in agent-based land use models – a review. Environmental Modelling & Software 87:39-48. https://doi.org/10.1016/j.envsoft.2016.10.008

Gütschow, M., B. Bartkowski, and M. R. Felipe-Lucia. 2021. Farmers’ action space to adopt sustainable practices: a study of arable farming in Saxony. Regional Environmental Change 21:103. https://doi.org/10.1007/s10113-021-01848-1

Hasler, B., M. Termansen, H. Ø. Nielsen, C. Daugbjerg, S. Wunder, and U. Latacz-Lohmann. 2022. European agri-environmental policy: evolution, effectiveness, and challenges. Review of Environmental Economics and Policy 16(1):105-125. https://doi.org/10.1086/718212

Holt, C. A., and S. K. Laury. 2002. Risk aversion and incentive effects. American Economic Review 92:1644-1655. https://doi.org/10.1257/000282802762024700

Holtz, G., F. Alkemade, F. de Haan, J. Köhler, E. Trutnevyte, T. Luthe, J. Halbe, G. Papachristos, E. Chappin, J. Kwakkel, and S. Ruutu. 2015. Prospects of modelling societal transitions: position paper of an emerging community. Environmental Innovation and Societal Transitions 17:41-58. https://doi.org/10.1016/j.eist.2015.05.006

Hristov, J., Y. Clough, U. Sahlin, H. G. Smith, M. Stjernman, O. Olsson, A. Sahrbacher, and M. V. Brady. 2020. Impacts of the EU’s common agricultural policy “greening” reform on agricultural development, biodiversity, and ecosystem services. Applied Economic Perspectives and Policy 42(4):716-738. https://doi.org/10.1002/aepp.13037

Huber, R., M. Bakker, A. Balmann, T. Berger, M. Bithell, C. Brown, A. Grêt-Regamey, H. Xiong, Q. B. Le, G. Mack, et al. 2018. Representation of decision-making in European agricultural agent-based models. Agricultural Systems 167:143-160. https://doi.org/10.1016/j.agsy.2018.09.007

Huber, R., S. Briner, A. Peringer, S. Lauber, R. Seidl, A. Widmer, F. Gillet, A. Buttler, Q. B. Le, and C. Hirschi. 2013. Modeling social-ecological feedback effects in the implementation of payments for environmental services in pasture-woodlands. Ecology and Society 18(2):41. https://doi.org/10.5751/ES-05487-180241

Huber, R., H. Xiong, K. Keller, and R. Finger. 2021. Bridging behavioural factors and standard bio-economic modelling in an agent-based modelling framework. Journal of Agricultural Economics 73(1):35-63. https://doi.org/10.1111/1477-9552.12447

Jager, W., and M. Janssen. 2012. An updated conceptual framework for integrated modeling of human decision making: the Consumat II. Paper for workshop complexity in the real world, ECCS. 5-6 September. Brussels, Belgium.

Jager, W., and M. A. Janssen, H. J. M. De Vries, J. De Greef, C. A. J. Vlek. 2000. Behaviour in commons dilemmas: Homo economicus and Homo psychologicus in an ecological-economic model. Ecological Economics 35(3):357-379. https://doi.org/10.1016/S0921-8009(00)00220-2

Janssen, M. A., and E. Ostrom. 2006. Empirically based, agent-based models. Ecology and Society 11(2):37. https://doi.org/10.5751/ES-01861-110237

Janssen, M. A., T. Falk, R. Meinzen-Dick, and B. Vollan. 2023. Using games for social learning to promote self-governance. Current Opinion in Environmental Sustainability 62:101289. https://doi.org/10.1016/j.cosust.2023.101289

Johnson, J. A., and C. Salemi. 2022. Agents on a landscape: simulating spatial and temporal interactions in economic and ecological systems. Frontiers in Ecology and Evolution 10:845435. https://doi.org/10.3389/fevo.2022.845435

Jongeneel, R. A., N. B. P. Polman, and L. H. G. Slangen. 2008. Why are Dutch farmers going multifunctional? Land Use Policy 25(1):81-94. https://doi.org/10.1016/j.landusepol.2007.03.001

Kahneman, D., and A. Tversky. 1979. Prospect theory: an analysis of decision under risk. Econometrica 47(2):263-292. https://doi.org/10.2307/1914185

Kaufmann, P., S. Stagl, and D. W. Franks. 2009. Simulating the diffusion of organic farming practices in two New EU Member States. Ecological Economics 68(10):2580-2593. https://doi.org/10.1016/j.ecolecon.2009.04.001

Klöckner, C. A. 2013. A comprehensive model of the psychology of environmental behaviour—a meta-analysis. Global Environmental Change 23:1028-1038. https://doi.org/10.1016/j.gloenvcha.2013.05.014

Kreft, C. S., M. Angst, R. Huber, and R. Finger. 2021. Social network data of Swiss farmers related to agricultural climate change mitigation. Data in Brief 35:106898. https://doi.org/10.1016/j.dib.2021.106898

Kremmydas, D., I. N. Athanasiadis, and S. Rozakis. 2018. A review of agent based modeling for agricultural policy evaluation. Agricultural Systems 164:95-106. https://doi.org/10.1016/j.agsy.2018.03.010

Kruse, J. 2014. Qualitative Interviewforschung: ein integrativer Ansatz, Grundlagentexte Methoden. Beltz Juventa, Weinheim; Basel.

Laatabi, A., N. Marilleau, T. Nguyen-Huu, H. Hbid, and M. Ait Babram. 2018. ODD+2D: an ODD based protocol for mapping data to empirical ABMs. Journal of Artificial Societies and Social Simulation 21(2):9. https://doi.org/10.18564/jasss.3646

Langerwisch, F., T. Václavík, G. Ziv, J. Gunning, A. Gosal, A. Paulus, S. Brdar, P. Lugonja, S. Stojković, M. Bednář, et al. 2021. Farming system archetypes for each CS. Deliverable D3.5 EU Horizon 2020 BESTMAP Project, Grant agreement No. 817501.

Langrell, S., P. Ciaian, M. Espinosa Goded, S. Gomez Y Paloma, T. Heckelei, P. Sckokai, K. Elouhichi, A. Thomas, and T. Vard. 2013. Farm level modelling of CAP: a methodological review. JRC Scientific and Policy Report.

Lastra-Bravo, X. B., C. Hubbard, G. Garrod, and A. Tolón-Becerra. 2015. What drives farmers’ participation in EU agri-environmental schemes?: Results from a qualitative meta-analysis. Environmental Science & Policy 54:1-9. https://doi.org/10.1016/j.envsci.2015.06.002

Ligmann-Zielinska, A., D. B. Kramer, K. S. Cheruvelil, and P. A. Soranno. Using uncertainty and sensitivity analyses in socioecological agent-based models to improve their analytical performance and policy relevance. 2014. PLoS ONE 9(10):e109779. https://doi.org/10.1371/journal.pone.0109779

Liu, H. B., M. Y. Wu, X. Liu, J. Gao, X. Luo, and Y. Wu. 2021. Simulation of policy tools’ effects on farmers’ adoption of conservation tillage technology: an empirical analysis in China. Land 10(10):1075. https://doi.org/10.3390/land10101075

Malek, Ž., and P. H. Verburg. 2020. Mapping global patterns of land use decision-making. Global Environmental Change 65:102170. https://doi.org/10.1016/j.gloenvcha.2020.102170

Mathijs, E. 2003. Social capital and farmers’ willingness to adopt countryside stewardship schemes. Outlook on Agriculture 32(1):13-16. https://doi.org/10.5367/000000003101294217

Matthews, R. B., N. G. Gilbert, A. Roach, J. G. Polhill, and N. M. Gotts. 2007. Agent-based land-use models: a review of applications. Landscape Ecology 22:1447-1459. https://doi.org/10.1007/s10980-007-9135-1

Mertens, A., J. V. Meensel, L. Willem, L. Lauwers, and J. Buysse. 2018. Ensuring continuous feedstock supply in agricultural residue value chains: a complex interplay of five influencing factors. Biomass and Bioenergy 109:209-220. https://doi.org/10.1016/j.biombioe.2017.12.024

Meyfroidt, P. 2013. Environmental cognitions, land change, and social-ecological feedbacks: an overview. Journal of Land Use Science 8(3):341-367. https://doi.org/10.1080/1747423X.2012.667452

Milfont, T. L., and J. Duckitt. 2010. The environmental attitudes inventory: a valid and reliable measure to assess the structure of environmental attitudes. Journal of Environmental Psychology 30(1):80-94. https://doi.org/10.1016/j.jenvp.2009.09.001

Millington, J., R. Romero-Calcerrada, J. Wainwright, and G. Perry. 2008. An agent-based model of Mediterranean agricultural land-use/cover change for examining wildfire risk. Journal of Artificial Societies and Social Simulation 11(4):4.

Muelder, H., and T. Filatova. 2018. One theory - many formalizations: testing different code implementations of the theory of planned behaviour in energy agent-based models. Journal of Artificial Societies and Social Simulation 21(4):5. https://doi.org/10.18564/jasss.3855

Müller, B., F. Bohn, G. Dreßler, J. Groeneveld, C. Klassert, R. Martin, M. Schlüter, J. Schulze, H. Weise, and N. Schwarz. 2013. Describing human decisions in agent-based models - ODD + D, an extension of the ODD protocol. Environmental Modelling & Software 48:37-48. https://doi.org/10.1016/j.envsoft.2013.06.003

Murray-Rust, D., D. T. Robinson, E. Guillem, E. Karali, and M. Rounsevell. 2014. An open framework for agent based modelling of agricultural land use change. Environmental Modelling & Software 61:19-38. https://doi.org/10.1016/j.envsoft.2014.06.027

Neuenfeldt, S., and A. Gocht. 2014. A handbook on the use of FADN database in programming models. Thünen Working Paper 35.

Noeldeke, B., E. Winter, and E. B. Ntawuhiganayo. 2022. Representing human decision-making in agent-based simulation models: Agroforestry adoption in rural Rwanda. Ecological Economics 200:107529. https://doi.org/10.1016/j.ecolecon.2022.107529

Nolan, J., D. Parker, G. C. Van Kooten, and T. Berger. 2009. An Overview of Computational Modeling in Agricultural and Resource Economics. Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie 57:417-429. https://doi.org/10.1111/j.1744-7976.2009.01163.x

ÖhlméYr, B., K. Olson, and B. Brehmer. 1998. Understanding farmers’ decision making processes and improving managerial assistance. Agricultural Economics 18:273-290. https://doi.org/10.1111/j.1574-0862.1998.tb00505.x

Pacilly, F. C. A., G. J. Hofstede, E. T. L. van Bueren, and J. C. J. Groot. 2019. Analysing social-ecological interactions in disease control: an agent-based model on farmers’ decision making and potato late blight dynamics. Environmental Modelling & Software 119:354-373. https://doi.org/10.1016/j.envsoft.2019.06.016

Parker, D. C., S. M. Manson, M. A. Janssen, M. J. Hoffmann, and P. Deadman. 2003. Multi-agent systems for the simulation of land-use and land-cover change: a review. Annals of the Association of American Geographers 93(2):314-337. https://doi.org/10.1111/1467-8306.9302004

Paulus, A., N. Hagemann, M. C. Baaken, S. Roilo, V. Alarcón-Segura, A. F. Cord, and M. Beckmann. 2022. Landscape context and farm characteristics are key to farmers’ adoption of agri-environmental schemes. Land Use Policy 121:106320. https://doi.org/10.1016/j.landusepol.2022.106320

Polhill, J. G., L.-A. Sutherland, and N. M. Gotts. 2010. Using qualitative evidence to enhance an agent-based modelling system for studying land use change. Journal of Artificial Societies and Social Simulation 13(2):10. https://doi.org/10.18564/jasss.1563

Pouladi, P., A. Afshar, M. H. Afshar, A. Molajou, and H. Farahmand. 2019. Agent-based socio-hydrological modeling for restoration of Urmia Lake: application of theory of planned behavior. Journal of Hydrology 576:736-748. https://doi.org/10.1016/j.jhydrol.2019.06.080

Pouladi, P., A. Afshar, A. Molajou, and M. H. Afshar. 2020. Socio-hydrological framework for investigating farmers’ activities affecting the shrinkage of Urmia Lake; hybrid data mining and agent-based modelling. Hydrological Sciences Journal 65(8):1249-1261. https://doi.org/10.1080/02626667.2020.1749763

Przyborski, A., and M. Wohlrab-Sahr. 2014. Forschungsdesigns für die qualitative Sozialforschung, Pages 117-133 in N. Baur and J. Blasius, editors. Handbuch Methoden Der empirischen Sozialforschung. Springer Fachmedien, Wiesbaden, Germany. https://doi.org/10.1007/978-3-531-18939-0_6

Railsback, S. F., and V. Grimm. 2012. Agent-based and individual-based modeling: a practical introduction. Princeton University Press, Princeton, New Jersey, USA.

Reidsma, P., S. Janssen, J. Jansen, and M. K. van Ittersum. 2018. On the development and use of farm models for policy impact assessment in the European Union – a review. Agricultural Systems 159:111-125. https://doi.org/10.1016/j.agsy.2017.10.012

Riley, M. 2016. How does longer term participation in agri-environment schemes [re]shape farmers’ environmental dispositions and identities? Land Use Policy 52:62-75. https://doi.org/10.1016/j.landusepol.2015.12.010

Robinson, D. T., D. G. Brown, D. C. Parker, P. Schreinemachers, M. A. Janssen, M. Huigen, H. Wittmer, N. Gotts, P. Promburom, E. Irwin, et al. 2007. Comparison of empirical methods for building agent-based models in land use science. Journal of Land Use Science 2(1):31-55. https://doi.org/10.1080/17474230701201349

Robinson, D. T., A. Di Vittorio, P. Alexander, A. Arneth, C. M. Barton, D. G. Brown, A. Kettner, C. Lemmen, B. C. O’Neill, M. Janssen, et al. 2018. Modelling feedbacks between human and natural processes in the land system. Earth System Dynamics 9(2):895-914. https://doi.org/10.5194/esd-9-895-2018

Roller, M. R., and P. J. Lavrakas. 2015. Applied qualitative research design: a total quality framework approach. Guilford, New York, New York, USA.

Rommel, J., C. Schulze, B. Matzdorf, J. Sagebiel, and V. Wechner. 2022. Learning about German farmers’ willingness to cooperate from public goods games and expert predictions. Q Open 3(3):qoac023. https://doi.org/10.1093/qopen/qoac023

Rosenbaum, M. S., M. L. Otalora, and G. C. Ramírez. 2017. How to create a realistic customer journey map. Business Horizons 60(1):143-150. https://doi.org/10.1016/j.bushor.2016.09.010

Rounsevell, M. D. A., D. T. Robinson, and D. Murray-Rust. 2012. From actors to agents in socio-ecological systems models. Philosophical Transactions of The Royal Society B: Biological Sciences 367:259-269. https://doi.org/10.1098/rstb.2011.0187

Ryschawy, J., M. Grillot, A. Charmeau, A. Pelletier, M. Moraine, and G. Martin. 2022. A participatory approach based on the serious game Dynamix to co-design scenarios of crop-livestock integration among farms. Agricultural Systems 201:103414. https://doi.org/10.1016/j.agsy.2022.103414

Schlüter, M., A. Baeza, G. Dressler, K. Frank, J. Groeneveld, W. Jager, M. A. Janssen, R. R. J. McAllister, B. Müller, K. Orach, et al. 2017. A framework for mapping and comparing behavioural theories in models of social-ecological systems. Ecological Economics 131:21-35. https://doi.org/10.1016/j.ecolecon.2016.08.008

Schlüter, M., R. R. J. McAllister, R. Arlinghaus, N. Bunnefeld, K. Eisenack, F. Hölker, E. J. Milner-Gulland, B. Müller, E. Nicholson, M. Quaas, and M. Stöven. 2012. New horizons for managing the environment: a review of coupled social-ecological systems modeling. Natural Resource Modeling 25(1):219-272. https://doi.org/10.1111/j.1939-7445.2011.00108.x

Schmolke, A., P. Thorbek, D. L. DeAngelis, and V. Grimm. 2010. Ecological models supporting environmental decision making: a strategy for the future. Trends in Ecology & Evolution 25(8):479-486. https://doi.org/10.1016/j.tree.2010.05.001

Schreier, M. 2012. Qualitative content analysis in practice. SAGE, Los Angeles, California, USA. https://doi.org/10.4135/9781529682571

Schwarz, N., G. Dressler, K. Frank, W. Jager, M. Janssen, B. Müller, M. Schlüter, N. Wijermans, and J. Groeneveld. 2020. Formalising theories of human decision-making for agent-based modelling of social-ecological systems: practical lessons learned and ways forward. Socio-Environmental Systems Modelling 2:16340. https://doi.org/10.18174/sesmo.2020a16340

Smajgl, A., and O. Barreteau. 2017. Framing options for characterising and parameterising human agents in empirical ABM. Environmental Modelling & Software 93:29-41. https://doi.org/10.1016/j.envsoft.2017.02.011

Smajgl, A., D. G. Brown, D. Valbuena, and M. G. A. Huigen. 2011. Empirical characterisation of agent behaviours in socio-ecological systems. Environmental Modelling & Software 26(7):837-844. https://doi.org/10.1016/j.envsoft.2011.02.011

Smith, E. B., and W. Rand. 2018. Simulating macro-level effects from micro-level observations. Management Science 64(11):5405-5421. https://doi.org/10.1287/mnsc.2017.2877

Sok, J., J. R. Borges, P. Schmidt, and I. Ajzen. 2020. Farmer behaviour as reasoned action: a critical review of research with the theory of planned behaviour. Journal of Agricultural Economics 72(2):388-412. https://doi.org/10.1111/1477-9552.12408

Sterling, E. J., M. Zellner, K. E. Jenni, K. Leong, P. D. Glynn, T. K. BenDor, P. Bommel, K. Hubacek, A. J. Jetter, R. Jordan, et al. 2019. Try, try again: lessons learned from success and failure in participatory modeling. Elementa: Science of the Anthropocene 7:9. https://doi.org/10.1525/elementa.347

Strauss, A., and J. M. Corbin. 1997. Grounded theory in practice. SAGE Publications, Thousand Oaks, California, USA.

Sun, Z., I. Lorscheid, J. D. Millington, S. Lauf, N. R. Magliocca, J. Groeneveld, S. Balbi, H. Nolzen, B. Müller, J. Schulze, and C. M. Buchmann. 2016. Simple or complicated agent-based models? A complicated issue. Environmental Modelling & Software 86:56-67. https://doi.org/10.1016/j.envsoft.2016.09.006

Sun, Z., and D. Müller. 2013. A framework for modeling payments for ecosystem services with agent-based models, Bayesian belief networks and opinion dynamics models. Environmental Modelling & Software 45:15-28. https://doi.org/10.1016/j.envsoft.2012.06.007

Sutherland, L.-A., R. J. F. Burton, J. Ingram, K. Blackstock, B. Slee, and N. Gotts. 2012. Triggering change: towards a conceptualisation of major change processes in farm decision-making. Journal of Environmental Management 104:142-151. https://doi.org/10.1016/j.jenvman.2012.03.013

Taghikhah, F., T. Filatova, and A. Voinov. 2021. Where does theory have it right? A comparison of theory-driven and empirical agent based models. Journal of Artificial Societies and Social Simulation 24(2):4. https://doi.org/10.18564/jasss.4573

ten Broeke, G., G. van Voorn, and A. Ligtenberg. Which sensitivity analysis method should I use for my agent-based model? 2016. Journal of Artificial Societies and Social Simulation 19(1):5. https://doi.org/10.18564/jasss.2857

Tieskens, K. F., B. J. Shaw, T. Haer, C. J. E. Schulp, and P. H. Verburg. 2017. Cultural landscapes of the future: using agent-based modeling to discuss and develop the use and management of the cultural landscape of South West Devon. Landscape Ecology 32:2113-2132. https://doi.org/10.1007/s10980-017-0502-2

Troost, C., R. Huber, A. R. Bell, H. van Delden, T. Filatova, Q. B. Le, M. Lippe, L. Niamir, J. G. Polhill, Z. Sun, and T. Berger. 2023. How to keep it adequate: a protocol for ensuring validity in agent-based simulation. Environmental Modelling & Software 159:105559. https://doi.org/10.1016/j.envsoft.2022.105559

Valbuena, D., A. K. Bregt, C. McAlpine, P. H. Verburg, and L. Seabrook. 2010b. An agent-based approach to explore the effect of voluntary mechanisms on land use change: a case in rural Queensland, Australia. Journal of Environmental Management 91(12):2615-2625. https://doi.org/10.1016/j.jenvman.2010.07.041

Valbuena, D., P. H. Verburg, A. Veldkamp, A. K. Bregt, and A. Ligtenberg. 2010a. Effects of farmers’ decisions on the landscape structure of a Dutch rural region: an agent-based approach. Landscape and Urban Planning 97(2):98-110. https://doi.org/10.1016/j.landurbplan.2010.05.001

van Zanten, B. T., P. H. Verburg, M. Espinosa, S. Gomez-y-Paloma, G. Galimberti, J. Kantelhardt, M. Kapfer, M. Lefebvre, R. Manrique, A. Piorr, et al. 2014. European agricultural landscapes, common agricultural policy and ecosystem services: a review. Agronomy for Sustainable Development 34:309-325. https://doi.org/10.1007/s13593-013-0183-4

Voinov, A., and F. Bousquet. 2010. Modelling with stakeholders. Environmental Modelling & Software 25(11):1268-1281. https://doi.org/10.1016/j.envsoft.2010.03.007

Voinov, A., N. Kolagani, M. K. McCall, P. D. Glynn, M. E. Kragt, F. O. Ostermann, S. A. Pierce, and P. Ramu. 2016. Modelling with stakeholders – next generation. Environmental Modelling & Software 77:196-220. https://doi.org/10.1016/j.envsoft.2015.11.016

Voinov, A., K. Jenni, S. Gray, N. Kolagani, P. D. Glynn, P. Bommel, C. Prell, M. Zellner, M. Paolisso, R. Jordan, et al. 2018. Tools and methods in participatory modeling: selecting the right tool for the job. Environmental Modelling & Software 109:232-255. https://doi.org/10.1016/j.envsoft.2018.08.028

von Neumann, J., and O. Morgenstern. 1947. Theory of games and economic behavior. Princeton University Press, Princeton, New Jersey, USA.

Vroege, W., M. Meraner, N. Polman, H. Storm, W. Heijman, and R. Finger. 2020. Beyond the single farm – a spatial econometric analysis of spill-overs in farm diversification in the Netherlands. Land Use Policy 99:105019. https://doi.org/10.1016/j.landusepol.2020.105019

Wittstock, F., A. Paulus, M. Beckmann, N. Hagemann, and M. C. Baaken. 2022. Understanding farmers’ decision-making on agri-environmental schemes: a case study from Saxony, Germany. Land Use Policy 122:106371 https://doi.org/10.1016/j.landusepol.2022.106371

Yang, L., and N. Gilbert. 2008. Getting away from numbers: using qualitative observation for agent-based modeling. Advances in Complex Systems 11(2):175-185. https://doi.org/10.1142/S0219525908001556

Yao, J. T., Z. Q. Duan, X. B. Kong, R. Lal, Y. J. Hu, Y. T. Zhang, S. Liu, and Q. Chu. 2018. An agent-based model to simulate the cultivation pattern change of farmer households in the North China Plain. Journal of Land Use Science 13(5):508-534. https://doi.org/10.1080/1747423X.2018.1499828

Zagaria, C., C. J. E. Schulp, M. Zavalloni, D. Viaggi, and P. H. Verburg. 2021. Modelling transformational adaptation to climate change among crop farming systems in Romagna, Italy. Agricultural Systems 188:103024. https://doi.org/10.1016/j.agsy.2020.103024

Zellner, M. L., D. Milz, L. Lyons, C. J. Hoch, and J. Radinsky. 2022. Finding the balance between simplicity and realism in participatory modeling for environmental planning. Environmental Modelling & Software 157:105481. https://doi.org/10.1016/j.envsoft.2022.105481

Zimmermann, A., A. Möhring, G. Mack, A. Ferjani, and S. Mann. 2015. Pathways to truth: comparing different upscaling options for an agent-based sector model. Journal of Artificial Societies and Social Simulation 18(4):11. https://doi.org/10.18564/jasss.2862

Ziv, G., M. Beckmann, J. Bullock, A. Cord, R. Delzeit, C. Domingo, G. Dreßler, N. Hagemann, J. Masó, B. Müller, et al. 2020. BESTMAP: behavioural, ecological and socio-economic tools for modelling agricultural policy. Research Ideas and Outcomes 6:e52052. https://doi.org/10.3897/rio.6.e52052