Combining research styles of the natural and social sciences in agricultural research

Abstract

The need for interdisciplinarity in agricultural and development-oriented research has become widely recognized. In this paper a framework is suggested to integrate research methods of the social and natural sciences. It is argued that the context–mechanism–outcome configuration, based on critical realism, allows a more comprehensive understanding of all candidate mechanisms that have a social, technical or socio-technical basis, related to a particular question. Candidate mechanisms are all possible mechanisms postulated to explain a particular phenomenon. Four research styles can be recognized in both the social and the natural sciences. These research styles help choosing the appropriate methods to test the various candidate mechanisms related to a single research question. Combining the context–mechanism–outcome configuration with the four research styles may reduce the chances of missing out important candidate mechanisms. In this way the proposed framework may help optimize the research set-up and methodology of an interdisciplinary research project. Understanding which disciplines and research styles to combine can also allow interdisciplinary research to go beyond triangulation, as it provides more clarity about the possibilities for tightly integrating research methods and/or different data sets. It is suggested that the absence of a clear methodology for interdisciplinary research holds the advantage that it helps building bridges and developing alternative paths in science.

Keywords:

• Interdisciplinarity
• Agriculture
• Triangulation
• Farming systems
• Research styles
• Development

1 Introduction

The need for interdisciplinarity to understand the complexity of socio-biological systems has become recognized across sociology, ecology and biology disciplines [1–3]. As agricultural systems are a complex interplay of technologies with social and biological factors, interdisciplinarity may give a better understanding why improved technologies, improved crop varieties, mechanization, and inorganic fertilizers and pesticides are often not adopted by farmers working under suboptimal conditions, particularly in Sub-Saharan Africa. A better understanding of more suitable technologies is crucial for solving problems such as eradicating poverty, improving food security and mitigating the effects of climate change.

Interdisciplinary approaches and methods are needed because agricultural technologies – such as crop varieties, hoes, and sickles – are shaped by both natural and socio-cultural factors. Some natural factors that shape a crop variety (particularly the diversity within that variety) include, for example, breeding system and number of seeds per flower or per panicle. Socio-cultural practices that relate to maintaining mixtures in seed (and hence shaping a variety) include conservation of other varieties, adaptation to uncertain conditions [4], prevention and tracing of theft (M.P. Temudo, personal communication), or as protection against witchcraft [5]. Moreover, natural and social factors often interact in shaping a technology. For example, the tool for harvesting rice in West Africa is shaped by plant architecture and by cultural beliefs. Harvesting rice with a sickle leads to higher levels of mixture in the seed than harvesting with a knife [4]. Short rice is harvested much more easily with a sickle, but cultural beliefs determine who is allowed to use a sickle. This example shows how natural and social factors may interact in very subtle ways within a single technology.

This means that in order to understand how a technology functions within a farming system we need to integrate research methods of the natural (biological) and the social sciences in an interdisciplinary¹ fashion. The strength of interdisciplinary research is that because a researcher integrates different disciplinary perspectives and methods within a single research project, he will be able to identify the subtle interactions between natural and social factors as described above. Interdisciplinary research means quickly and systematically shifting between various perspectives. An interdisciplinary approach is particularly useful to recognize subtle nuances and differences at a local level, whereas a multidisciplinary¹ approach may be more suitable to describe patterns at a more regional level. In addition, as it can be argued that farmers use a comprehensive perspective and language (i.e., concerned with natural and social properties and mechanisms at the same time) to understand, describe and improve their farm, an interdisciplinary research approach may help to better understand farmers’ motivations and actions [6]. Interdisciplinary approaches may identify the priorities of farmers better and therefore help the development of technologies – ideally in collaboration with farmers – that will be more readily adopted by farmers.

Problems related with interdisciplinary research are, amongst other ones, higher career risk, lower rewards and inadequate communication channels [7]. Compared with disciplinary research, interdisciplinary research is often considered too time-consuming, partly because few guidelines exist on doing interdisciplinary research [1]. Successful interdisciplinary research depends heavily on the willingness of researchers to be open to alternative concepts, definitions and languages [7,8]. A certain level of pragmatism and flexibility is required for conducting interdisciplinary research. When working in a team, sufficient time should be allocated for discussing and understanding each other's concepts and jargon. One or two workshops are not enough. It may happen that only half way a project the researchers realize they use a certain concept differently. For individual researchers this may be less of an issue, although it could happen that, for example, a PhD-student realizes at the end of his project that certain phrases are understood differently by the supervisors coming from different scientific backgrounds.

A question not well answered yet is how research methods of the natural and social sciences can actually be integrated in a research project [9], and how this can be done in a scientifically rigorous way. Answering this question is important to enable more scientists to engage in interdisciplinary research, and to facilitate communication of the results during conferences and through the publishing of interdisciplinary research papers. This paper aims to provide some guidelines for linking research methods of the natural and social sciences in interdisciplinary agricultural research.² These guidelines may also be useful for reflection. First, the existence of various views is described on the possibility of integrating natural and social science disciplines, including the merits of grounded theory, triangulation and system approaches. After that a framework is described to facilitate the integration of research methods of natural and social science disciplines.

2 Various views on integrating natural and social science disciplines

The natural and social sciences are often seen as two different worlds of practice with different paradigms and methods [9]. A common perspective is that the natural sciences use quantitative approaches and the social sciences use qualitative approaches, which by nature cannot be integrated [10]. The following statement by Patton [11] reflects this idea:

In short, the Verstehen approach assumes that the social sciences need methods different from those used in agricultural experimentation and natural sciences because human beings are different from plants [10].

The underlying idea is that quantitative research methods used by natural scientists would be objective, whereas qualitative research methods (such as participant observation and in-depth interviews with key stakeholders) used by social scientists are subjective in nature [12]. But it was only in the 1970s and 1980s that the belief developed that quantitative and qualitative methods are intrinsically linked to two different paradigms [13]. Many qualitative researchers believed that using numbers means believing in a single objective reality [14]. A common response by quantitative researchers to uncomfortable qualitative results is that these results are anecdotes and, hence, can be dismissed. The idea that natural and social sciences cannot be integrated does not die easily because discussions on interdisciplinary research run in parallel in the social and natural sciences and are scarcely connected [2]. The little interaction that exists between the social³ and natural sciences is because of boundaries being maintained through differences in concepts, jargon, social networks and literature [2]. Other boundaries are institutional culture and related disincentives to stimulate interdisciplinary research [12,15].

However, the idea that the natural and social sciences cannot be integrated is based on prejudice rather than on facts [9]. The natural and social sciences, and quantitative and qualitative methods, are not tightly linked to paradigms and epistemological preferences [9,14,16]. As to agriculture broadly two dominant scientific paradigms can be identified [17,18]. The paradigm related to conventional agriculture is related to the world view that science can solve all problems, using a top–down and reductionist approach rooted in experimental biology. The other paradigm, related to sustainable agriculture, is related to a world view that problems need to be solved at community level using a broad range of approaches, including non-reductionist and participatory approaches. Within this paradigm the wide range of existing methods are considered complementary to each other [17]. Instead of paradigms and disciplines acting as barriers for interdisciplinary research, the real boundaries are actively demarcated and defended by scientific communities, each having their own jargon and culture [12].

Although there may be differences in preferences, within the natural and social sciences both quantitative and qualitative approaches are used. For example, botany commonly uses qualitative methods, whereas neoclassical economics and social psychology mainly use quantitative methods. Experiments are not the sole domain of the natural sciences and economics, but are also used in sociology and related disciplines. For example, Humphreys et al. [19] used an experiment in a development-oriented research project to illustrate the significant role of leaders in decision-making processes. In their research on leadership effects they ensured solid statistical analysis, based on the principles written down by Fisher, using randomization and a large number of observations. In addition, both quantitative and qualitative research styles use the logic of deductions and hypotheses [20,21].

Two examples illustrate that social and natural scientists may deal with similar problems in terms of analysis. An example from the social sciences is attempting to measure the effect of advertising on people's behaviour. It may be possible to observe a change in behaviour, or not, but in neither case can one rule out the possibility that there is a causative effect between the advertisement and behavioural change, or not. Other ‘confounding’ factors may be involved causing, preventing or counterbalancing change. Another possibility is that the change may only take effect at a later stage. So apart from observing the behaviour, it is also important to understand how people used the advertising information.

Natural scientists may encounter similar problems when, for example, studying the effect of fertilizer on plant behaviour in different locations, or over different seasons. In one location, or season, the fertilizer may lead to increased plant height, whereas in another location, or season, no such change may be observed. The difference may not necessarily be related to the fertilizer applied. In one location, or season, the increased plant height may not be due to the fertilizer, and in the other location, or season, where plant height did not change, the effect of the fertilizer may have been neutralized by another factor. To understand such different reactions, the complex interactions between the plant, the fertilizer and other factors need to be studied.

This suggests that the types of research questions and methods of the social and the natural sciences seem not to differ much. Heilman [20] argues that pragmatism is important for combining research approaches and that some researchers have better skills in combining different research styles and methods than others. This argument is appealing given the diversity existing within the qualitative and quantitative approaches for both the natural and social sciences. Doing interdisciplinary research means learning a certain body of theory related to each discipline intended to be used in a research project. Some may struggle finding out which perspectives to integrate. Others may struggle to decide which research methods to use and how to integrate them. Below, grounded theory, triangulation and system approaches are discussed with a focus on their merit for integrating research methods of the social and natural sciences.

2.1 Grounded theory

Grounded theory may be a useful approach because it encourages developing theory in relation to a certain research problem based on empirical data obtained with both qualitative and quantitative methods [7,21] and to link inductive with deductive thinking [22]. Hence, grounded theory may be suitable to facilitate the integration of theoretical perspectives and research methods from different disciplines of the social and natural sciences. But how to do this is in a way to ensure scientific rigour is not yet well elaborated. Grounded theory has mainly been used in the social sciences, and is still developing as a research tradition [23].

2.2 Triangulation

Essentially, triangulation means using multiple research methods to answer a particular research question: the more methods, the more solid the answer obtained. Erzberger and Prein [24] suggest three possible outcomes of triangulation: (1) converging research findings that increase validity, (2) complementing findings that can generate a more comprehensive picture or a new comprehension, and (3) diverging results that lead to a falsification of previous theoretical assumptions. The third outcome can allow the initiation of new lines of thinking [16]. Triangulation is mentioned as a concept for linking quantitative and qualitative methods and data sets [24,25]. It is not specifically developed for integrating social and natural science research methods. Only limited guidelines are provided for the design, measurements, analysis, interpretation and integration of the research findings [25]. No general strategy exists for triangulating different research methods [24]. A few useful, though general, guidelines are that (1) the world should be viewed as a whole interactive system, (2) both objective and subjective data are legitimate data sets, and (3) both reductionist and holistic thinking need to be used in design and analysis [26]. Another guideline is that qualitative and quantitative datasets can only be linked in a sensible way based on theoretical assumptions or hypotheses that determine the focus of the research [24]. Kopinak [25] even goes a step further by saying that a single paradigm is needed for research using triangulation. This may not be a problem if we agree that agricultural research is dominated by only two paradigms (see Pretty [17] and Lyson [18]).

The most basic form of triangulation is that after conducting literature research a list of hypotheses is formulated and qualitative, observational research is conducted to refine the findings of the literature research (or vice versa) and to cross out a number of hypotheses [24]. This is followed by a third phase of quantitative research, through an experiment or a survey. A fourth phase of qualitative research may follow in which the understanding of the context is strengthened. The quantitative research allows some generalization and helps to avoid bias in the interpretation of the qualitative data, whereas the qualitative research helps to understand the quantitative data better and to contextualize them [16,27]. This, in fact, very much resembles the basic rules for conducting field trials: based on theory and a good description of the reality, hypotheses are formulated for which various parameters are measured and analysed using advanced statistics, after which hypotheses are rejected, approved or adapted, and a new cycle of research follows.

In some cases qualitative and quantitative data can be collected simultaneously through semi-structured interviews with closed-ended and open-ended questions [28]. Another form of triangulation is answering different sub-research questions, relating to the same research objective, using different research methods [29]. The research cycle described by Giller et al. [3], in which design is followed by description, explanation and exploration, is similar to the sociologic cycle described by Mann [27], but the research cycle described by Giller et al. [3] allows space for triangulating qualitative and quantitative data collected with social and natural science methods at the various stages of description and explanation. But using triangulation does not necessarily imply conducting interdisciplinary research if only at the end of a research cycle the outcomes of the various data sets are compared. Overall, the literature on triangulation does not provide many guidelines for integrating social and natural science research methods.

2.3 System approaches

An alternative interdisciplinary approach specifically for agricultural research can be found in a system approach. Because of the wholeness, also a requirement for triangulation, a system approach may have a certain level of inherent interdisciplinarity. Several system approaches oriented towards agriculture have their origin in the biological sciences and have incorporated social science elements over the past 20 years [30]. But according to Jansen [30] these system approaches ignore three elements essential to farming systems: power issues (inequality, gender, age, etc.), openness (the impossibility⁴ to define boundaries of a farming system) and stratification (a farming system being the result of a complex of diverse mechanisms of origin and the emergence of new properties that are the result of a complex interplay of various biological and social factors across strata). Jansen [30] suggests critical realism may be better equipped to deal with elements of power, openness and stratification because it aims at identifying and understanding underlying mechanisms.

3 A framework for integrating natural and social science disciplines

The above illustrates it is possible to combine research methods from the social and natural sciences. But it is not that much clear which research methods should be integrated to answer which research questions. Similar to triangulation, the type of interdisciplinarity (combination of different research methods) needed is context specific. Depending on the context, particular disciplines (and their methods and approaches) will be chosen and combined at various levels. For each research question a researcher needs to reflect which methods are relevant to use in a research approach integrating social and natural science research methods. The framework described below aims to provide some guidelines to better understand which research methods to use. The framework consists of two elements, the context–mechanism–outcome configuration [31] and the identification of research styles [32].

3.1 Context–mechanism–outcome configuration

Critical realism aims to avoid the traditional epistemological poles of positivism and relativism [31]. As such it meets the guidelines for triangulation described by Shih [26]. Critical realism distinguishes three levels of the world: the real, the actual and the empirical [33]. The real refers to what exists, both natural and social, and to the structures and powers of objects and institutions. The actual refers to what happens if and when those powers are activated. The empirical refers to the domain of experience, observations and measurements. The empirical refers to the real and the actual, although it is contingent whether we know the real and the actual [33]. This is important as it emphasizes not to take a significant association for granted as an explanation, but to look for less easily observable ‘hidden’ mechanisms that may offer a better explanation. A critical realist perspective allows studying all sorts of interactions between material and social elements in an agricultural system, using a stratified perspective.⁵ Using a critical realist perspective, Pawson and Tilley [31] developed the concept named context–mechanism–outcome configuration, in short CMO configuration. The aim of the concept is to better understand which mechanism given a particular context makes X to change into Y, leading to a new outcome (a changed context). Pawson and Tilley [31] use the formula or function: regularity = mechanism + context. In many cases several mechanisms are interconnected and make up a complex mechanism (Fig. 1): the more complex the mechanism, the more complex the function. This rather abstract description does not imply a linear view. In different contexts the same mechanism may lead to different outcomes and different mechanisms may lead to the same outcome [34]. The CMO configuration recognizes that a particular mechanism may be part of the context of another mechanism, which facilitates the understanding of complex interactions. Very importantly, the CMO configuration does not make a fundamental distinction between quantitative and qualitative approaches.

Fig. 1 Context–mechanism–outcome configuration: Given a certain context (C), a number of mechanisms (M_1,…,n) may cause X to change into Y, leading to a new outcome (which in turn is a changed context).

To find out which mechanisms may constitute regularity, alone or together, an essential exercise is the listing of so-called ‘candidate mechanisms’ (hypotheses), preferably as many as possible. The more elaborate the list, the smaller the chance that a potential candidate mechanism is overlooked. Candidate mechanisms are all possible mechanisms postulated to explain a particular phenomenon or regularity. In the case of agricultural technologies, some candidate mechanisms have a social character, other ones a biological character, and yet other ones have a combined social and biological background (complex mechanisms). For example, the design of a hoe used for ploughing rice fields in West Africa is shaped by natural factors (the water level in the field and possibly soil structure) and social factors (such as gender, culture and availability of iron). The selection of crop varieties is the result of the matching to agro-ecological and socio-cultural factors. Integrating insights from various natural and social sciences helps developing complex candidate mechanisms. System approaches may also be useful to identify complex candidate mechanisms. Technography, as described by Richards [35] and Jansen and Vellema [36], and developed to understand interactions between a technology and its wider agro-ecological and socio-political context, may provide additional candidate mechanisms. From this list of candidate mechanisms follows which research methods are needed to test each of these candidate mechanisms. Some candidate mechanisms may only emerge during fieldwork. Depending on the nature of such emerging candidate mechanisms it may be possible to test them in the same research project using the principles of grounded theory, such as comparative analysis, achieving much diversity in relevant categories, and comparing similarities and dissimilarities of those categories using qualitative and quantitative methods [21].

3.2 Research styles

This section describes how the research styles from the social and natural sciences can be integrated in a more thorough way.

3.2.1 Four research styles

Galmiche-Tejeda [6] and Bellamy [32] provide a matrix integrating four basic styles of social science research. On the x-axis are explanatory versus interpretative research methods and on the y-axis qualitative versus quantitative research methods (Fig. 2). Their model leads them to identify the following four research styles:

(A)	Developing candidate mechanisms (qualitative and explanatory): theory building. Based on theory the most plausible candidate mechanisms are identified.

(B)	Description by observation (qualitative and interpretative): rich and detailed description of contexts and outcomes. Research questions are answered with qualitative interviews, participant observations, etc. In some cases complex mechanisms can be described.

(C)	Statistic associations (quantitative and explanatory): experiments and modelling. The focus is on using advanced statistics for testing candidate mechanisms (hypotheses). Advanced statistics are particularly useful for testing complex mechanisms and detecting hidden mechanisms.

(D)	Description by measure (quantitative and interpretative): provides general trends and patterns, contexts and outcomes. Research questions are answered based on questionnaires and counting, using descriptive statistics, but also through the literature review of multiple case studies.

Fig. 2 Four basic styles of social science research.

Based on [6] and [33].

This model could be considered a further elaboration of the thinking on triangulation by Miles and Huberman [16], who mentioned the linking of quantitative surveys with qualitative fieldwork and quantitative experiments, Erzberger and Prein [24], who made suggestions how to link qualitative and quantitative research findings with theory, and Salomon [37], who, instead of making a distinction between qualitative and quantitative research approaches, suggested to make a distinction between studies of causal relations (experiments) and systemic studies of complex environments.

In principle, all four research styles can be triangulated (Fig. 2). These groupings do not have clear boundaries: grey zones exist between the four styles. At first thought there is one exception, i.e., that there is no grey zone between styles B (associated with, e.g., cultural anthropology) and C (associated with, e.g., neoclassical economics), as it is difficult to imagine how the qualitative and quantitative data can be integrated into a single analysis. However, a very useful integration of styles B and C is that anthropological insights (research style B) could be used to reach a better definition and more thorough understanding of the units and variables to be analysed using advanced statistics (research style C). An example comes from an experiment in which male and female farmers in two Gambian villages were asked to group rice and millet inflorescences. The comments and justifications made by farmers during the grouping exercise were very important for understanding the subsequent statistical analysis of the experiment [38]. Some examples of social science disciplines with a perceived bias towards one of the four research styles are shown in Fig. 3.

Fig. 3 Examples of social and natural science disciplines with a perceived bias towards a particular research style.

The research conducted by Humphreys et al. [19], in which the effect of leadership on the outcome of participatory decision processes was analysed statistically through the randomized assignment of facilitators, is an example of a sociology case study that is very suited for the use of advanced statistics (style C). The difference between research styles C (advanced statistics) and D (descriptive statistics) is greater than it seems. Although sometimes it may be possible to use advanced statistics on a dataset produced with methods belonging to research style D (a kind of hybrid C–D research method), in most cases the research methods of research styles C and D produce datasets designed to answer different research questions. For example, research style D is about quantitatively describing a farming system, market or ecosystem, research style C is geared towards understanding hidden relationships between various variables and testing mechanisms. Hence, in most cases the sampling methods will be different: a representative sampling method is related to research style D, whereas a purposive sampling method (an equal number of villages and farmers per variable to be studied) is more suitable when using research style C.

Certain research methods may not belong to a single research style. For example, semi-structured interviews can be considered to be a method intermediate between research styles B and D, as they allow some quantification of the collected data and at the same time provide more detailed information (see Hesse-Biber [28]). Literature review may in some cases be considered to belong to style A, for building a theoretical framework, whereas in other cases it belongs to research style D when many case studies are compared in order to identify patterns.

3.2.2 Applying the framework to the natural sciences

The scheme can also be applied to the natural sciences. Obviously, the basic style ‘B, construction of understanding’ does not apply to the study of plants, soil, molecules or other natural science objects. If replacing the phrase ‘interpretive nature’ with ‘descriptive nature’ and rephrasing the basic style B into ‘description by observation’, which is also part of style B in the social sciences, the scheme in Fig. 4 is applicable to the natural sciences. Note that a natural scientist may consider the term ‘interpretation’ more like a hypothetical explanation. He may use it when the results of an experiment are pointing to a certain explanation but are not sufficiently clear (not statistically significant) to call it an explanation. So there are some differences in jargon used by social and natural scientists we need to be aware of, which have developed over time because of limited interaction [2].

Fig. 4 Four basic research styles in the natural and social sciences, and humanities.

The main, very important, difference for the social and natural sciences in respect to research style B is that the social sciences study people, including their relationships with technology and the environment, whereas the natural sciences study predominantly study objects, plants and animals (but also physical aspects of humans). This means that the interaction between the researcher and the researched (persons or objects) is fundamentally different between the natural and social sciences. In the case of social science research a so-called double hermeneutics develops. A researcher tries to understand the world of the people he studies and the people being studied try to understand their world, including the researcher (and possibly modify their behaviour towards the researcher if deemed necessary). But in natural science there is only a single hermeneutics – objects, plants and most animals do not try to understand their world. It is also suggested that human societies are more dynamic than agro-ecological systems. And that it is easier to replicate an observation based study of an ecological system than of a social group. However, the dynamics of an ecological system should not be underestimated. Some examples of disciplines with a perceived bias towards a particular research style are shown in Fig. 3.

3.2.3 Incorporating the humanities into the framework

The scheme in Fig. 4 can also be applied for integrating social sciences with the humanities, or integrating both natural and social sciences with humanities. At first glance the focus of the humanities seems to be more on the qualitative research styles, possibly with an emphasis on theoretical issues (philosophy, law, etc.). But also in the humanities all four research styles are represented. A clear example of style C is glottochronology, which is an approach in linguistics that aims to understand relationships between languages, using advanced statistics in a similar way as in quantitative genetics. Another example is that of history, which often uses literature research (style D).

3.2.4 Linking research styles to the CMO configuration

The research styles can be linked with the CMO configuration as follows. Research style A is primarily about developing candidate mechanisms, research styles B and D are about describing the context, possible candidate outcomes and testing candidate mechanisms. Research style C can be mainly used for testing candidate mechanisms. In some cases research style C can also be used for testing candidate outcomes. This concurs with Miles and Huberman [16], who suggested that qualitative and quantitative approaches can be used for descriptive, exploratory, inductive purposes and for explanatory, hypothesis-testing purposes. Heilman [20] also argued that both qualitative and quantitative research styles use the logic of deduction and hypothesis.

All four research styles can be used to (in)validate candidate mechanisms, albeit in different ways. Following the supposed divide between quantitative hypothetico–deductive research and qualitative holistic–inductive research [10], it is proposed that research style A is linked to deduction, style D more to induction and styles B and C may be both linked to abduction⁶ but in different ways. Good abductions in style B depend on the skills and knowledge of the researcher being able to draw general conclusions based on qualitative data collected in a few case studies. Good abductions in style C are based on probabilities and inherent uncertainties in combination with the skills of a researcher to interpret such statistics. If indeed research styles B and C both produce abductions, but derived in different ways and based on different skills, this may partly explain the tension between researchers using these styles.

Qualitative approaches (style B) are open to new information, and to theory generation, whereas style C tends to emphasize theory and hypothesis testing [28]. A disadvantage of research style C is that the research area and scope need to be clearly demarcated. It allows less flexibility during data collection than research style B. It is common for style B to define new categories, variables, and hypotheses during the course of the research. Realizing these differences is very important as they are probably at the heart of the tension between styles B and C. Integrating these two styles implies a continuous tension during the research. The new exciting categories defined using research style B may not be analysed using advanced statistics (style C) because earlier collected quantitative material (both socio-economic and molecular data) may not allow the testing of new questions and hypotheses in relation to these new categories. One may say that each research style runs at its own speed. Recognizing this difference may be very important to understand and to combat the prejudice scientists have about the gap between the natural and social sciences. An additional explanation for the tension between the two styles could be a difference in ‘mental models’ of two communities of researchers, in that style C is oriented towards variables and correlations, and style B to persons, events and related processes [14].

But this does not necessarily mean that, at least in the context of agriculture (see Pretty [17]), research styles B and C are associated with different research questions, paradigms, and thus inherent in different world views, as suggested by Salomon [37]. For example, a study on farmer management of rice diversity by Nuijten et al. [39], described below in greater detail, integrated research styles B and C within the context of one paradigm. Amongst other, research style B was used to trace the origin of a new rice type and research style C was used to describe its genetic relationship with the existing rice species. In cases where there may be differences in paradigms, it is also feasible, as Heilman [20] suggested, finding a middle ground using pragmatism. In such a case, negotiating a middle ground may lead to the discovery of new research methods, a new theoretical perspective, and perhaps a shift in paradigm.

3.3 Applying the framework to interdisciplinary research

The advantage of combining research styles is not only triangulation (see [6] and [32]) for the purpose of increasing validity, a more comprehensive understanding or to find discrepancies, but also using the different styles to understand better which methods are needed to test different candidate mechanisms relating to a single research question in the context of an interdisciplinary project. Fig. 4 facilitates a better understanding of how the various research methods used in interdisciplinary research relate to each other, and what their strengths and weaknesses are in terms of interpretation, explanation, and representativeness. This may facilitate the development of interdisciplinary methods combining the strengths of the various styles. Fig. 4 may also be applied to reflect which styles are being used in a research project, and what interdisciplinary methods integrating social and natural science research methods may need to be developed.

An interdisciplinary field that has been combining social and natural science research methods for a long time is ethnobotany, which is a combination of anthropology and botany, both having research style B in common. Integration of research methods in this field may be relatively simple because of the qualitative and interpretative aspects of both anthropology and botany. Depending on the nature of the data in an ethnobotanic research, the social science and biological science data sets may be even integrated into a single data set. Integrating natural and social sciences using the same research style is relatively easy. Integrating different styles is the big challenge, in particular integrating styles B, C and D, assuming that style A is an intrinsic element of the research. Interdisciplinary research using different research styles will yield different data sets that usually cannot be integrated. In that case each research style (and data set) can be used to answer specific sub-questions or the methods can be integrated into a single method. In both cases the research approach is strengthened.

An example of the integration of research methods belonging to different styles of the social and natural sciences is the integration of interviews (style B) on the use and origin of rice varieties with a molecular analysis (style C) of the rice varieties collected in farmer fields in West Africa by Nuijten et al. [39]. The units of analysis (rice varieties) for the molecular analysis were defined based on farmers’ descriptions of the morphology of the rice varieties. An advantage of this approach is that there is a clearer understanding of which plants a particular rice variety consists of. This is important because it allowed us to include for a particular variety only those plants that fitted farmers’ descriptions. Even though farmers say they grow a particular variety in a particular field, mixtures of rice varieties from two rice species (Oryza sativa and O. glaberrima) are common in many fields in West Africa. Together with the descriptions, farmers were asked questions about the origin and use of these varieties. This integrated approach led to the identification of a new type of rice varieties intermediate between O. sativa and O. glaberrima, and a better understanding of how natural and cultural selection pressures may have shaped this new rice type, which also provided some insights into the domestication of rice in West Africa. A disciplinary approach would probably not have led to these findings. It may also be argued that these insights were obtained because scientific and farmer knowledge were integrated.

In some types of research, the use of advanced statistics is just not feasible or very limited, irrespective of the kind of discipline (whether in the natural or social sciences or in the humanities). In that case a researcher needs to be pragmatic and to use research styles B, A and/or D. An example is a study on consistency in variety naming of rice and millet in The Gambia [40], for which samples of inflorescences of varieties were collected during a survey on farmer management of crop varieties (both research style D). Comparing the morphological features with the names of the varieties showed a clear difference between villages in consistency in variety naming for rice. In-depth interviews with farmers, agricultural researchers and NGO workers provided the information to understand the underlying mechanism explaining these differences in naming consistency.

3.4 Two steps to help identify the most appropriate research methods

Ideally, all four styles are integrated in a research project. For some interdisciplinary research projects only three, sometimes only two styles may be used integrating methods from the social and natural sciences. To know which research methods and styles to integrate, a researcher will formulate (sub) research questions, candidate mechanisms (hypotheses) and candidate outcomes and contexts, following the CMO configuration described by Pawson and Tilley [31]. This is the first step. Technography [35,36], to understand interactions between a technology and its wider agro-ecological and socio-political context, may provide additional entry points to develop a comprehensive list of candidate mechanisms, outcomes and contexts. The next phase is to build an integrated theory using insights from different disciplines to discard those candidate mechanisms and outcomes that can be falsified based on existing literature. The literature search may also yield other candidate mechanisms and outcomes. In most cases the researcher will go through several cycles of formulating candidate mechanisms and outcomes, reading literature (and possibly some exploratory research) and discarding the most implausible candidate mechanisms and outcomes. During the course of this process the researcher, together with his co-researchers, may have also integrated different languages and concepts belonging to different disciplines into a single framework. When this initial iterative process is completed, the researcher will have a set of candidate mechanisms and outcomes to be tested during the actual research.

The second step is choosing the appropriate research styles and methods for testing the identified candidate mechanisms. If only one or two research styles are identified the researcher may want to reflect whether any candidate mechanisms were overlooked. This applies in particular to development-oriented research. As such the research styles are helpful in reflecting on the candidate mechanisms and the research methods to be used. If it is impossible to use advanced statistics (style C) to test a candidate mechanism, an interdisciplinary study may consist of only research styles A, B and D, but using research methods from both biological and social sciences. If one would want to know how people living in a tropical rainforest use biodiversity in comparison with people living in a desert, it seems farfetched to use research style C. Instead, one can first conduct interviews and participant observation (style B) after which biodiversity surveys are done to measure diversity used and total diversity in the forest (both styles D). It may also be possible to include style C in the testing of one candidate mechanism (for example on testing genetic distances between varieties or languages) but not for another candidate mechanism (on peoples’ concepts and believes, or describing the essential differences between related plant species). In case a researcher may be able to use different research methods to test a candidate mechanism, he may opt to choose methods that belong to different research styles for stronger triangulation. It seems plausible to assume that the more diverse the research methods, and the more diverse the scientific disciplines (from both the social and biological sciences), the bigger the chances of an increased validity of the research results, a more comprehensive or an advanced understanding, or diverging results that may lead to falsification of earlier theoretical assumptions.

The validation of complex mechanisms often results from the outcome of the testing of the biological and sociological mechanisms that together form a complex mechanism. In agricultural research, technology often plays a central role and usually all research styles are needed to test complex mechanisms. Research style C may be applied to better understand the material aspects of technology and research style B may be used to better understand the social aspects of a technology. To test complex mechanisms it is important that all aspects of those complex mechanisms are covered using the various research styles. It should be emphasized that no research style is better than another, and that qualitative and quantitative data are equally important. Which data to collect first, the experimental or interview data, depends on the research question and the nature of the complex candidate mechanisms. To optimize the usefulness of a laboratory analysis of seeds, soil and other materials, it is better to collect these materials after interviews with farmers. But if the laboratory analyses are done as early as possible, these results may be used to optimize those interviews. For every project, a researcher needs to decide which parts of the research should be done first to be able to feed the results into the other parts of the research, or whether they can be done simultaneously and a two-way feedback may be possible.

3.5 Development-oriented research

When conducting development-oriented research, it is important to integrate farmers’ perspectives on possible candidate mechanisms. The whole set of candidate mechanisms could then be tested using participatory experiments and various types of interviews. The experiment could be based on a farmer design (for example three crop varieties sown next to each other) using single replications in several farmer fields in the same village and still be analysed using advanced statistics. Mutsaers et al. [41] show various examples of such designs that can be analysed with advanced statistics. Mother-and-baby trial is another experimental approach in which an experiment is set up together with farmers [42]. The mother trial is set up accordingly to be tested with advanced statistics (research style C). For example, the treatments consist of various combinations of fertilizer application rates, crop varieties and different levels of labour input. The baby trials are set up in farmer fields using single replications and testing a few treatments per farmer field. A sufficient number of baby trials allow using advanced statistics for the testing of the results obtained in farmer fields. At the same time the baby trials are used for qualitative comparisons by and with farmers. This approach facilitates a tight integration of research styles B and C. It also helps bridging the gap between farmer and scientist experimentation [43]. Participatory experiments (such as Participatory Varietal Selection and Participatory Plant Breeding) have demonstrated such potential [44]. Such experiments can also be used as entry points to discuss more contextual issues (which would be usually addressed through research style B). As was mentioned by Galmiche-Tejeda [6], interdisciplinary approaches may help to understand the ‘interdisciplinary’ language of the farmers better. A technology tested in the field and grounded in the local context allows an in-depth understanding of the mechanisms, context and potential outcomes [45].

4 Conclusions

Because of the complex nature of agricultural systems, certain kinds of agricultural research (particularly those with development-oriented research questions) need to integrate research methods of the social and natural sciences. The identification of four research styles that exist in both the natural and the social sciences helps to understand how to integrate these fields of knowledge within a single research project. In combination with the CMO configuration, it helps to understand which research styles may be lacking in an interdisciplinary research project. It may also reduce the misunderstanding that exists between the natural and the social sciences. At present, the natural science disciplines in agricultural research may not take enough notice of the ‘real world out there’ and overlook the need of research style B. At the same time, the fields of anthropology and development studies tend to overlook the importance of research style C, needed to understand hidden mechanisms. We should also realize that the dominant research styles within disciplines may change over time. It is important to emphasize that no single research style is intrinsically better than another. In principle, all research styles and methods from the natural and social sciences can be integrated within one research project.

What research styles to use depends on the specificities of a question. This means there is no precooked methodology for interdisciplinary research. Every time researchers start an interdisciplinary project they need to think what research styles and methods are necessary to answer their questions. At this stage it is not possible to develop a methodology that is useful for interdisciplinary agricultural research that is neither too general nor too specific at the same time. As more interdisciplinary research is conducted over time, ideas may develop about how to order the various types of interdisciplinary research (basically changing them into disciplinary research such as plant breeding, which can be considered an integration of agronomy, crop physiology, genetics, soil science, phytopathology and other sciences) and to understand whether doing that is useful or counter-productive. One of the strengths of interdisciplinary research may be that it makes researchers realize that for each research question a specific theoretical framework needs to built, linking different theoretical perspectives as deemed necessary. Every time a specific methodology needs to be designed, and every time a researcher needs to decide which methods to include, and possibly which new methods to develop. In that way, interdisciplinary research may help stimulate creativity in scientific research, build bridges and develop alternative paths in science.

Acknowledgements

I wish to thank Paul Richards for his encouragement to write this article, and Dominic Glover and two anonymous reviewers for their critical comments and suggestions.

Notes

1 Sometimes interdisciplinarity and multidisciplinarity are used interchangeably. In this paper the following definitions are used: multidisciplinarity means combining different disciplines as subprojects in one project and integrating the data at the end of the project, whereas interdisciplinarity implies integrating different disciplines within subprojects from the start. Whereas in a multidisciplinary team each member remains in his or her own field of expertise, in an interdisciplinary team members cross the borders and integrate research methods belonging to different disciplines (see Klein [8]). This paper focuses on interdisciplinary approaches.

2 This article may also be useful to scientists who are interested in linking different methods within the natural or social sciences. The emphasis on the linking of natural and social sciences is because of a supposed boundary that makes crossing very difficult.

3 There are different view points whether the social sciences include economics or not. In this paper, it is recognized that economics is part of the social sciences. As such, economics seems to be an odd one out amongst the social sciences, as it has more interaction with the natural sciences than the other social sciences.

4 Some may argue it is possible, though very difficult, to define boundaries. From a pragmatic point of view, one may want to define boundaries, although it can be considered impossible from a theoretical point of view, particularly given increasing globalization.

5 See also Jansen [30] who comments on the hierarchical aspects of system thinking.

6 Abduction is often treated as a special case of induction. It is generally considered an inference to the best explanation. It comes in two forms: it can be assessed in terms of probability (relating to style C), and by logically evaluating the available qualitative information (relating to style B).