Why we should rethink ‘adoption’ in agricultural innovation: Empirical insights from Malawi

The challenges of land degradation, climate change and food insecurity have led to the introduction of conservation agriculture (CA) aimed at enhancing yield and soil quality. Despite positive biophysical results, low adoption rates have been the focus of studies identifying constraints to wider uptake. While the adoption framework is popular for measuring agricultural innovation, objective adoption measurements remain problematic and do not recognize the contextual and dynamic decision‐making process. This study uses a technographic and participatory approach to move beyond the adoption framework and understand: (a) how agricultural decision‐making takes place including the knowledge construction, (b) how agriculture is performed in a context of project intervention and (c) how practice adaptation plays out in the context of interacting knowledge. Findings confirm that farmer decision‐making is dynamic, multidimensional and contextual. The common innovation diffusion model uses a theory of change, showcasing benefits through training lead farmers as community advocates and demonstration trials. Our study shows that the assumed model of technology transfer with reference to climate‐smart agriculture interventions is not as linear and effective as assumed previously. We introduce four lenses that contribute to better understanding complex innovation dynamics: (a) social dynamics and information transfer, (b) contextual costs and benefits, (c) experience and risk aversion, and (d) practice adaptation. Investments should build on existing knowledge and farming systems including a focus on the dynamic decision process to support the 'scaling up, scaling out and scaling deep' agenda for sustainable agricultural innovations.

. However, CA adoption rates in countries such as Malawi have remained low, with a reported 5-6% of the arable land farmed using CA (Kassam et al., 2019). This has been the subject of various studies measuring adoption, identifying adoption constraints and understanding dis-adoption (Chinseu et al., 2019;Ngwira et al., 2014;Thierfelder et al., 2015;Ward et al., 2018).
Agricultural innovations are often conceptualised as a technical package of practices, distributed to new areas with the help of instruction (Glover et al., 2017), with adoption rates representing a primary way of measuring success and impact of this distribution measured (Glover et al., 2016(Glover et al., , 2019. The processes of adoption and diffusion, that is, expanding the use of the agricultural innovation, are often characterised as 'scaling'. However, recent literature has highlighted that scaling occurs across multiple levels and dimensions, which are not always considered (Sartas et al., 2020;Wigboldus et al., 2016). To acknowledge these multiple ways in which scaling can take place, specific scaling types have been defined: upscaling refers to extension of the innovation to higher levels (e.g., national), outscaling to expansion within the same level (e.g., within the community) and deep scaling to a change in the mindset and culture (Moore et al., 2015;Schut et al., 2020). This 'scaling up, scaling out and scaling deep' discourse, a linear diffusion of innovation model, remains popular among development initiatives despite various critiques (Chambers et al., 1989;Glover, 2011). It is embedded in the idea that farmers mainly make individual yes or no decisions with a linear development of replacing old methods with new ones (Glover et al., 2016).
A broad literature on the diffusion of agricultural innovation recognises the importance of context and enabling conditions on shaping technology transfer and adoption dynamics Zanello et al., 2016). Moreover, attention is required on the dynamic connection between the farmer and the system context, which coevolve and adapt in relation to each other . Drawing on science and technology studies (STS), there is also an emergent critical response to simplistic narratives around the 'rational' adoption of successful technologies, highlighting the socially constructed and contested nature of agronomic knowledge (Sumberg, 2017). A focus on metrics of adoption overlooks the important processes and decision-making through which innovation happens on farms and may miss out on considering the prerequisite conditions (Sumberg, 2005), namely if the technology is needed and suitable to potential users and local contexts. It also fails to recognise the multiple ways in which farmers do not simply adopt, but continually experiment with and adapt technologies to these contexts (Whitfield, 2015). Therefore, both technology implementation constraints, and the ways in which farmers engage with these constraints, also termed tinkering (Higgins et al., 2017), are contextual and heterogeneous.
Objective measuring of CA adoption remains problematic (Andersson & D'Souza, 2014;Giller et al., 2015) due to the definition of practices that constitute CA and the spatial (e.g., area covered), quality (e.g., how many principles of what) and temporal (e.g., how many seasons) thresholds when it 'counts' as adoption. For example, a systematic review has shown that few papers discussing technology adoption adequately define what adoption is (Loevinsohn et al., 2013). Therefore, questions have been raised in terms of the validity of adoption statements (Andersson & D'Souza, 2014;Brown et al., 2017;Giller et al., 2015).
Recent studies have also called for exploring the adaptation of CA to agro-ecological and socio-economic contexts of the targeted smallholder farmers to increase the CA uptake (Brown et al., 2018b(Brown et al., , 2018aThierfelder et al., 2015). In order to 'measure' adoption, the question of 'what is CA' is important and often found to be challenging (e.g., land size, time, all practices) ranging from technical definitions to farmers self-defining CA (Hermans et al., 2020). With adoption or non-adoption used as a measure, adoption in itself has become a metric of success for policies or development programmes.
There is a building portfolio of evidence across southern Africa that the science of new agricultural practices does not directly translate into farmers' implementation Giller et al., 2009;Ndah et al., 2018;Ngwira et al., 2014;Ward et al., 2018). The agronomically designed top-down 'fixed' package is designed with a focus on biophysical improvements and is often not fully suitable for the local adaptation it will undergo. Methodologies and research are needed that acknowledge the differences, negotiations and conflicts in processes of agricultural decisionmaking including contextualization (Thompson & Scoones, 1994).
Technography is the social science describing the technology-in-use and can support other approaches, such as participatory approaches or system theories (Glover, 2011;Jansen & Vellema, 2011). It can be used as a tool to understand the contextualized processes through which agricultural practices are decided upon, insights into how and why certain practices are implemented, and how they differ between farmers (Glover, 2011). It also enables the understanding of the temporal aspect in farmer decision-making. The approach uses a social constructivist underpinning, namely that knowledge and realities of farmers are continually shaped by contextual interactions and experiences. This is supported by the analytical framework of 'agriculture as performance', which emphasizes that farmer decision-making is a reaction in a certain moment embedded in a social and ecological context (Richards, 1989(Richards, , 1993. The technography approach promotes more open questions about how farmers make decisions when the new technologies are introduced and how this leads to agricultural practice change. In this paper, we use a method based on the technographic and participatory approaches, to rethink and move beyond the concept of 'adoption' or 'non-adoption'. Our aim is to understand farmer decision-making after the introduction of CA in two communities in Malawi and to explore the dynamics and nuance of decision-making processes. The paper seeks to understand: (a) how agricultural decision-making takes place and how the knowledge for process is constructed, (b) how agriculture is performed in a context of development project intervention, including the interaction around this intervention and (c) how CA practice adaptation plays out in the context of interacting knowledge.

| CA in Malawi
Malawi depends on rain-fed agriculture with maize being the major staple food crop, covering 80% of the cultivated land area and the major calorific intake (Ngwira et al., 2012). The traditional practice is to prepare the land manually with a hand-hoe. Planting is often done on ridges made annually with approx. 75-90 cm row spacing (Bunderson et al., 2017;Fisher et al., 2018). This traditional practice results from the focus on soil degradation of colonial policy in southern Africa since the 1930s (Andersson & D'Souza, 2014). Residues are burned, removed or buried in furrows.
Malawi, besides Zambia and Zimbabwe, has been on the forefront of CA promotion in southern Africa since the late 1990s (Andersson & D'Souza, 2014). The first CA initiative was established by the NGO Sasakawa Global 2000 in 1998 and supported by the Malawian government (Dougill et al., 2017;Thierfelder et al., 2013). The Sasakawa initiative promoted minimum tillage and mulch cover among smallholder farmers and provided resources packages, similar to national government starter packs, including NPK fertilizer, urea and improved hybrid maize seeds funded by various donors (Dougill et al., 2017). The set of management practices included planting population instructions (1 seed per station in 75 cm ridges and an in-row spacing of 25 cm) and herbicides, which farmers had to buy themselves (Ito et al., 2007;Ngwira et al., 2014). The "SG2000 package" also received extension support to improve"production management" (Ito et al., 2007:420).
This support has become a characteristic of CA promotion initiatives leading to the association and accusation that CA requires high inputs, and critique on the sustainability of such systems and its resulting adoption (Andersson & D'Souza, 2014;Dougill et al., 2017).
The Malawi CA introduction process was renewed in 2004 through a collaboration between the International Maize and Wheat Improvement Centre (CIMMYT), the Malawi Government Extension Services, and later the NGO Total LandCare (TLC) (Ngwira et al., 2014;Thierfelder et al., 2013). This effort focused on the establishment of demonstration trials in communities that enable discussions on CA technologies to prevent land degradation and yield decline (Ngwira et al.,2014). The theory of change that drove this agricultural research for development project in the communities is that demonstrating benefits through 'demonstration trial plots' and training lead farmers to become community advocates, will lead to a snowballing of rational adoption decisions, building on local interactions and innovation systems.
Currently, CA has been widely promoted by NGOs, government, international research centres and development organisations to improve maize yields and drought resilience. Initial CA advocacy has taken place without the development of a national strategy or guidelines, resulting in agreement about CA as an approved technology in 2013 and the formulation of National Guidelines for its promotion in 2016 through a National Conservation Agriculture Task Force (NCATF) (Dougill et al., 2017). This agenda is still being promoted now.

| Study sites
This study was carried out in two Malawian communities, which are  (Table 1).

| Methods
A pilot study based on four focus groups and community visits was conducted in October 2018. Subsequently a triangulation of methods was used to examine agricultural decision-making and drivers of change in agricultural practices. Firstly, focus groups were organized using participatory methods including timelines, mapping and ranking exercises. The focus groups were conducted with the trial farmer group (six farmers) and groups of non-trial farmers (8-10 farmers).
One focus group per community was conducted with trial farmers, and two for each community with groups of non-trial farmers. In total, six focus group discussion events were organized.
This was followed up with semi-structured interviews to understand individual and household decision-making. Interviews focused on diversity and depth to build understanding of farmer variable decisionmaking. Timelines of agricultural decisions focusing on changes in practice and drivers of these decisions were constructed during interviews.
This timeline approach using oral history enabled discussing changes in  (Whitfield & Marshall, 2017). In addition, it approached decision over a longer time to avoid bias of the fieldwork year's particular wet season.
The one-on-one interviews were based on the six trial farmers and a subsequent snowball methodology to select 12-14 farmers with different relations to the trial per community. In total, 38 interviews were conducted. In addition, ethnographic observation in the farming communities for a duration of 3-4 months was conducted (Jansen & Vellema, 2011).
Written consent was obtained from all participants before interviews. It was clarified that the interview had no influence on the participation in any programme. Ethical consent for this research was The case-studies presented were selected to showcase the diversity, multidimensionality and complexity in farmer decision-making and practice experimentation and adaptation. The cases were selected from both communities regardless of its agro-ecology and social makeup (patrilineal/matrilineal) to support exploring this diversity, since the theory of change for the diffusion model is applied in both communities. While the cases are diverse and contextual, they represent the (non-linear) ways in which farmer decision-making and practice implementation take place for the wider population. Therefore, case-study analysis still provides relevant representation and validity for a bigger scale (Flyvbjerg, 2006).

| RESULTS
The following case-studies are the stories of seven individuals from the CA trial hosting communities. Their relation to the on-farm trials differs from trial farmers to farmers with no direct connection to the trials (see Figure 1).
It is important to note that the definition of promotional 'packages' such as CA and Sasakawa is sometimes defined differently by the farmers, who may just refer to sub-practice (components) from the package. Sasakawa, among the farmers, in this case just refers to the spacing introduced with Sasakawa Global 2000 (75 × 25 cm ridges and one seed per station), thus not the practices of residue retention or minimum soil disturbance. In the case of CA, the practices are named separately when referred to, or as all three practices in the full CA package.

| Case 1: The 'lead' farmer
One of the farmers who maintains a demonstration trial is Albert. The main income of his household is farming groundnut, maize, pigeon pea, sweet potato and cassava. He runs a CA trial, for which he had the 'courage' to start because he was told he would receive fertilizer, seeds and herbicides.
"In the third year of the trial, was when they told us we need to do what we do in the trial also in our own field." Following this idea, outside the trial he practices 0.1 ha of CA and on the remaining 0.8 ha of maize, he plants on ridges with burying crop residues ("...for soil fertility") due to a variety of reasons including land tenure. He rents land every year although the size depends on the money available. He mentions that custom land law prescribes that they do not rent for more than 3 years because otherwise the owners are afraid the renters start to treat it like their own land. Due to this, he does not see the benefits of a practice change to invest in soil fertility and will only practice conventional agriculture on the rented land.
The unpredictable weather is problematic for his choice of agricultural practice. He knows CA is good when it is dry, which is why he promotes it since there have been more dry spells. However, he also stresses that: "CA is not good when the heavy rains come, but and then plant, whereas on CA you have to do the same in the first place -make 75cm planting rows but then also import residues." If he has enough fertilizer from the subsidy, he uses Sasakawa for 0.1 ha, which he finds manageable in terms of resources and breaking up the ridges from 90 cm to 75 cm. On the rest of the fields, he continues with making ridges and burying the residues, like most of them in the community do.
Burying residues, which he learned improves soil fertility, is not more work, whereas residues on top like in CA. He explains that: "Ridges is what farmers believe in. They make ridge and then planting the seeds, then weeding, then banking. So, it becomes hard to adopt a new system." At the same time when CA was introduced, they were told that if they feel CA is too difficult, then they can keep ridges. Others may adopt CA because they see the benefits of CA and find it worth the effort.
In his own experience, the soil gets hard on the flat land, especially when there are insufficient crop residues, whereas the ridges make the soil soft again, which makes it easier for maize to grow. This way the whole field enjoys improvement in soil fertility. If he sees the residues are not sufficient or the weeds are problematic, he decides to heap up the soil (bank) to control the weeds.

|
Since he knows the soil needs to be well covered, he imports the residues and also takes some from the neighbours who would burn them otherwise. This collection is enough for 0.2 ha in order to cover the field to the level that ridges are not needed, as observed on the trial.
For all his other fields he just plants the maize on old ridges, without renewing them and banks when weeding is needed. In the past, when he made new ridges, the rain would come and wash them away.
So, when TLC introduced the planting on old ridges, many of the farmers in the community liked it, making it now a common practice.
To help his work on the land, he hires labour but he would never do that for his 0.2 ha CA because they mess it up or ask for more money.
3.5 | Case 5: The 'age adapter' farmer Mary is excited to talk about the 3-year system she uses to cultivate because she wants to minimize the labour due to her husband's and her poor health. She thought of this in 1994 when she was late with land preparation due to her teaching job. She notes that the first year is the most work when new ridges are made including the burying of residues. In the next 2 years, she leaves the ridges without splitting them to make new ridges and places the residues between them.
Once she completes weeding, she places them on the ridges. For these 2 years of no-tillage, she also does not need to spend money on hiring labour. The old ridges are also good for her land because the strong old ridges will not wash away easily on the slope.
Since she had to pay school fees for children, she could never buy fertilizers, so she liked the idea of burying crop residues that still improve soil fertility. She started burying residues when she moved away from her parents, after learning from neighbours that residues improve soil fertility.
"Adding residues is the only way people can cultivate without fertilizer." Despite her preference, due to poor health, to avoid making ridges, she sees it as necessary to make new ridges every 3 years because otherwise her clay soil gets too hard.
When she is lucky to be part of the fertilizer subsidy programme, she can do Sasakawa on a smaller piece of land she rents, which will give her more yield than normal, particularly when there is a drought.
She tried doing this since she was invited to a field day at a trial 5 minutes from her house. For her other field, she never considers Sasakawa because it is too big.
"The big field is fertile, but Sasakawa can only be done with hybrid seeds and these seeds need fertilizer." She tried hybrids on the big field 4 years ago but without fertilizer, which resulted in very poor yields. Based on her parents farming she continued to intercrop through the fields. For the groundnut fields, she noticed on the demonstration trials that farmers are applying residues, but she believes residues are not good for groundnut so she has not changed the practices. While these practices are described as normal, she does admit that she gets mocked as being lazy for her 3-year system by others. She does not like this since "...people want to be admired to work hard" -but her health does not give her many options.
3.6 | Case 6: The 'female family caregiver' farmer In a house far from the main road and not easily accessible lives Violet. This divorced farmer has five children but takes care of nine people in total in her household. She farms, burns charcoal and works in other people's fields and on a roadside development. Furthermore, she had to rent out 1.6 ha because of her financial problems.
Due to all her livelihood supporting jobs, she wants as little work as possible on her fields. That is why she burned the residues this year and planted them on old ridges. On the fields where the children helped her, they made new ridges, because her children oppose to not making new ridges despite her own observation that maize does better when planted on old ridges. In 2008, she did Sasakawa and CA on 1 acre, but she felt intimidated by others. People were laughing that the plants were so close to each other and will not do well.
They said: "...it takes you more time to plant 1 seed per station so you will be the last to finish planting." She also heard residues will bring fall armyworm. The next year she did it only on 0.1 ha. She still kept the 0.1 ha Sasakawa because the yield was good. The others still disparaged but 0.1 ha was acceptable by them as a test.
Right now, peoples' mindset is changing, due to the trials. She She was a member of NASFAM, for which she paid a membership fee but received free groundnut seeds. She only did this for one season because NASFAM did not get back to her about it and she was not reimbursed. She just followed what they told her to do but she did not observe a change. Overall, she liked the trial system but did not expand and burned the residues again, which she continues to do now. Since nobody put effort in the trial or told her the objectives, she did not feel like continuing the practice. With the current CA trials, she mentions that "Most people think only the trial [lead] farmer was chosen to do that farming. He was chosen by TLC." The extension officer never comes to her area so she struggles to contact him and would not know how to start the new practice by herself.
In particular, planting with a marked string looks complicated and too involving. She never asked anything herself to the lead farmer, but the extension officer could tell her more in detail because he went to school and was trained.
On her own field, she has good maize so she does not feel compelled to change but she would like to know from the extension officer about how to do certain things.

| DISCUSSION
The various stories of individuals in these communities hold within them themes that contribute to a more nuanced understanding of adoption and innovation dynamics, which are often overlooked in linear innovation diffusion discourse. In the following section we highlight and discuss four lenses that can contribute to our understanding of farmer decision-making: social dynamics and information transfer, contextual cost and benefits, experience and risk aversion, and practice adaptation.

| Lens 1: Social dynamics and information transfer
Farm-level knowledge and decision-making are socially constructed have been recognised in an emergent STS literature (Glover et al., 2016;Whitfield, 2015) and critical extension studies (Leeuwis & Van den Ban, 2004 that not making ridges is still associated with 'laziness', whereas 'hard-working' is seen as the virtue for a farmer to be food secure. This is contradicting, since a perceived increase in labour, related to the planting without ridges and residue retention, is also seen as discouraging CA. On the other hand, the release from making ridges is also a motivation in favour of CA. Therefore, it seems labour remains a contested topic with beliefs, consideration of total season labour (Thierfelder et al., 2016) and its timing.
Social acceptability is associated with community group dynamics and connected flow of information. Farmers observed from the trial that support was given to start CA. This makes farmers think they need that same support to make the change work, leading to a belief that it is not worth trying on one's own. The trial farmers are part of the club and the farmers receive extension officer's attention and support. Even farmers who implemented CA on their own feel they are part of the club with access to information on modern technology.
A distinct problem is that while the theory of change of demonstration trials and farmer to farmer distribution assumes homophily (i.e., people in the community are equal) (Rogers, 2003), the group dynamics create heterophily, which makes the diffusion of innovation not as effective.
There are beliefs and social dynamics in the community that are also of importance to farmers' decision-making. For example, the general belief that residues are not good for groundnut, despite data showing more harvest under CA (Bunderson et al., 2017). Similarly, the increase of planting population under Sasakawa creates the belief of higher fertilizer need. However, less fertilizer per plant leads to similar fertilizer need per area. The consensus of what is sufficient residue is different among farmers, and based on the CA introduction and trials, residue import to create a thick layer was needed. These instructions have now changed to just leaving leftover residues but the idea of 'sufficient' seems to still differ between farmers. The concept of 'residues being a limiting factor' may therefore be based on the belief on how much is sufficient. In the narrative of residues, the belief of residue import risking disease transfer (e.g., fall armyworm) is widely accepted, although proliferation of fall armyworm through crop residues is uncommon and only applies to stalk borers. This shows that having access to information can support practice change but common beliefs may counteract this.
The closeness to a trusted source of information affects the belief in the validity of the information Holden et al., 2018). Farmers in direct contact with the extension officer trust and implement more of the information, than when it comes to indirect ways such as trial observation or other community farmers. Some state that the lead farmer dissemination approach works since they are closely connected, whereas others note that this does not work.
As previously reported in Brown et al. (2020), farmers report problems with information sources and lack of training due to lack of contact with extension officer and lead farmers. Alternatively, studies by Cofré-Bravo et al., (2019) have shown that there is a wide variety in the configuration of knowledge and support networks used by farmers, depending on livelihood, farm and innovation goals. In this light, the focus on lead farmers to instigate innovation diffusion does not fully accommodate the diversity in knowledge and support networks. The assumed model of technology transfer, which relies on expanding social connections, leading to information transfer that turns into implementation, as illustrated in Figure 1 may not be as linear and effective.

| Lens 2: Contextual costs and benefits
As recognised in diffusion theory (Rogers, 2003), sustained engagement with a new innovation depends on whether or not there is a relative advantage of the new practice over the current practice. An assessment of relative advantage includes a consideration of the compatibility of innovation with the existing context. While diffusion theory acknowledges that context plays a role, this is often limited to biophysical or technical factors or assuming linear and rational decision-making, thereby not addressing the full multi-dimensionality and dynamic decision-making process. The case of CA in Malawi helps to demonstrate that there are complex set of contextual costs and benefits that shape decision-making, and that these are themselves socially constructed.

Farmers consider the balance between costs and benefits for their
context. This is not only economic but also includes social and ecological aspects and the intangible 'cost' of changing to something new. Two economic elements that increase the 'costs' or lower the benefits are rented land and hired labour. On rented land, the benefits of practices perceived as CA are not experienced, and in hiring labour, oversight is needed or more remuneration. Another economic aspect is that practice implementation is dependent on the fertilizer subsidy received that year. In most cases, the major challenge to agricultural improvement is identified as access to the resources. This challenge is associated with the belief that CA systems can only be applied with high input packages.
Farmers do not have the 'courage' to try new practices because they do not get the resource or knowledge support, they feel they need.
Other factors also play a role in the contextual balance. Farmer experimentation and adaptation are based on health and labour concerns (e.g., ridge making labour, residue import, string planting) and agro-ecological dimension (e.g., soft soil, land slope). Some farmers know the benefits but the perceived effort costs are too high. Benefits from residue are most evident during droughts, which provide a convincing entry point. However, it was also mentioned that the year after a drought there are very little residues, thereby increasing the challenge of residue retention. Over the farming season, these factors interact and are affected by the context's institutions and structures, creating reinforcing cycles of productivity, health, resource access and labour (Jew et al., 2020). The benefits need to be sufficient and address the farmers' needs and challenges, which are dynamic and focused on short-term benefits rather than longer-term sustainability.
The balance of costs and benefits is contextual and can be dependent on the introduction of other changes in agricultural practices, such as planting on old ridges, Sasakawa planting or residue burying. The common methods of old ridges and banking are also seen as an improvement, which saves work. The observation of the trial farmer importing the residues, the agro-ecological observations and the government message that Sasakawa planting is already an improvement forms the beliefs of costs and benefits. The burying of residues for soil fertility improvement was easily adopted than the CA package because the cost was low compared to the benefit. Mentioning of 'others may find it worth it' shows that the cost and benefit balance is individualistic, addressing the challenges given by Glover (2011) that decision-making is multidimensional and dynamic.
The contextualization and livelihood dependency of the costs and benefits balance (Farnworth et al., 2016;Mutenje et al., 2019) can especially be elaborated in Violet's case. It is representative of various female farmers interviewed who are divorced, separated or widowed.
They have additional jobs, which become the focus of cash income.
There is shortage of labour for their fields and there is no money for herbicides or hired labour to replace that work, particularly weeding.
A change of practice is observed as too much work and effort (including the learning process). This shows the livelihood context of decision-making and shows that there is a risk in change, which comes with intangible costs that for some are not worth the benefits.

| Lens 3: Experience and risk aversion
In the context of complex costs and benefits, particularly for resource-constrained farmers, a risk-averse approach to new technologies and investments may predominate (Whitfield, 2015). We also see, in this case, how past experiences of technologies and interventions can contribute to an aversion to risk. This is evident in the cases of disengagement or small-scale and incremental experimentation with CA practices.
Individual experiences play a role and show that current decisionmaking is not only rational. For example, disappointment with a previous trial project, not understanding its purpose, lack of observable improvement and contact with extension officer all create less willingness to change practice again. There is a lack of feeling involved or ownership of the trial. This was also reported in Brown et al. (2020), who highlighted that lead farmers did not understand that they can expand beyond the trial. The farmer stories present that decisionmaking can result from information flow interacting with personal (sometimes accidental) experimentation.

Risk-averse behaviour to keep options open also guides farmers'
decision-making. One main challenge is the uncertainty of the weather. Risk is spread by using both the conventional practice in case of heavy rains and the perceived CA practices, of which the main focus is residue retention, in case of droughts (Ngwira et al., 2013).
The conventional method is seen as leaving options open in case the resources cannot be found because banking and weeding with a hoe can be done. Other strategies are the back-up plan of banking in case the weeds still get through the residue layer.

| Lens 4: Practice adaptation
In agricultural innovation, we rarely see a linear perfect and wholescale replacement of old practices by new ones (Glover, 2011). The adaptation or 're-invention' of practices shows that there is change in the used agricultural practices, which can be beneficial for sustainability of the implementation of new practices (Rogers, 2003). As such, there may not be a single moment of technology adoption or a clear distinction between those that do and those that do not adopt a technology, which emphasizes the dynamic process (Kiptot et al., 2007).
Rather, as in the case of CA in Malawi, we might observe a continually changing mosaic picture of resultant practices, across space and time, which reflect the socially constructed knowledge, local costs and benefits, and risk aversion and experimentation of different farmers.
Farmers use CA information and experimentation, and implement this in various manners, as has also been mentioned in CA adaptation literature (Brown et al., 2018b(Brown et al., , 2018a. There is hybridization of old and new practices. In particular, Sasakawa planting is seen as a modern agricultural improvement and a step towards the perceived CA package but without removing the ridges. The CA package introduction included the first year with Sasakawa planting with residues retention and the conventional field in the on-farm trials is also Sasakawa planting. There are associated costs with Sasakawa planting such as fertilizer and labour for breaking up the ridges for the first time. However, it is seen as using improved modern techniques, but does not meet the costs or investment that comes with perceived CA practices (e.g., residue retention). Planting on old ridges and banking is also a variation moving forward from the old practices and can be found in the CA package introduction where ridges should not be remade. Therefore, farmers, in their own way, negotiate and work with constraints, a process also called tinkering (Higgins et al., 2017), to use new information on agricultural innovation.
Other dynamic implementations are on temporal and spatial scales. New practices are done on limited land areas, most frequently in 0.1 or 0.2 ha, the usual trial size, for various reasons including social acceptance and labour limitations. Alternative strategies include moving the 0.1 ha around so that the entire land can be improved. On the temporal scales, conscious choices are made to change practices every season due to rainfall or health affecting resources.
While re-invention is often not considered good, it is not necessarily bad once the reasoning behind the choices is understood. Considering the adaptation of practices that is occurring, including an increase in the 'left-over' information from the Sasakawa introduction, crop diversification or residue retention, we notice that farmers are interacting with the introduction of new practices. This response is dynamic and resulting from the interaction of the individual farmer and system context . The use of information is not always in the exact introduced form but it does allow for the customization to local context (Rogers, 2003

| Recommendations
Establishing this dynamic process and moving away from adoption measuring framework, thereby provide empirical insights to the work of Glover (2016Glover ( , 2019, which shows that there is need to shift investment away from perfecting a technology and instead focus on the process and farming system the innovation can adapt to. This requires considering and exploring the relationship and co-evolution of the farmer decision-making and the system context, which will be increasingly important when scaling agricultural innovation Sartas et al., 2020;Wigboldus et al., 2016). Furthermore, this should be paired with a shift to focusing on the end goal, namely the extent needs are met through innovations, instead of the extent of adoption. Funding structures and incentives often reinforce the situation of organisations being tied to the promotion of specific technologies and innovations, and competing to demonstrate the relative advantage, often using adoption rates as a metric of success that reinforces their claim to success (Sumberg et al., 2012). However, shifting focus and incentives to the end goal of innovation could encourage a movement away from narrowly conceived technological solution and focus efforts on the quality of innovation processes. For example, building on adaptation that farmers already implement, such as planting on old ridges, any form of residue retention or the Sasakawa planting. This also provides the opportunity to change the approach to focus on supporting farmers' intrinsic motivation to adapt practices and experiment, thereby acknowledging the differences in farming styles and goals. Projects could therefore learn from these case studies to improve farmers' ownership, empowerment, develop 'complexity-aware' non-linear theory of change and evaluation (Douthwaite & Hoffecker, 2017) and become process facilitators (Kessler et al., 2016) in the change towards improving livelihoods and sustainable agriculture.
Innovation platforms, as also suggested in Schut et al. (2016) and Brown et al. (2020), including farmer and extension officers can support further development of existing extension, knowledge and practice systems. They can also provide better connection between introduced agricultural packages and community-based agricultural development. To capture and work with dynamic farming systems, including the non-predictable contextual emerging challenges and opportunities, continuous reflection and feedback is important to match the needs and actions (Kilelu et al., 2014). This requires evolving learning processes, through a dynamic learning agenda (Kilelu et al., 2014), in which extension services play an important role.
For the 'scaling up, scaling out and scaling deep' discourse, it will be of importance to take into account these dynamic interactions and the ways in which new innovations can be processed into implementation.

| Reflection on the approach
The qualitative approach enabled going beyond the adoption measuring framework and associated challenges with CA definitions. It uncovered the diversity in adaptation of practice and how farmers process and interact with agricultural innovation information and interventions. Its focus on depth over large area representativeness has supported the concept of agriculture as performance and the contextualised process of dynamic and multidimensional farmer decisionmaking, including the temporal aspects (Glover, 2011;Richards, 1989Richards, , 1993. The challenges of the adoption measuring framework are embedded in the agricultural systems' problem (Glover et al., 2016), in terms of how these systems are defined, and its dynamics, diversity and complexity acknowledged. This farmer-centred approach, including ethnographic informed interviews, enables a cross-disciplinary look, considering these system challenges for the diffusion of innovation and associated theory of change.

| CONCLUSIONS
In this study, a method based on the technographic and participatory approach was used to rethink the concept of 'adoption', understand how agricultural decision-making takes place and how the knowledge is constructed after the introduction of CA in two Malawian communities. The approach has shown that farmer decision-making is dynamic, multidimensional and contextual. There is a large range of interacting factors that play a role in the decision-making at a particular point in time: agro-ecology, health, labour, economics, resource endowment, family size, age, gender, experience, risk aversity, alternative practices available and social dynamics. The trade-offs of these are different for individual farming systems and livelihoods at a certain time. This is dependent on the relative advantage in the individual farmer's perception change to farming practice.
The theory of change underpinning the common agricultural innovation diffusion model is based on demonstrating benefits through 'demonstration trials' and training lead farmers to become community advocates. Our study has shown that social dimensions, including acceptability and group dynamics, play an important role in the farmer decision-making and efficiency of the diffusion model. The level of closeness and trust in the source of information influence the agricultural decisions, which balance between new information, level of trust, common beliefs and experience. The assumed model of technology transfer is, therefore, not as linear and effective as often assumed.
Moving beyond the adoption measuring framework has shown that there is a wide diversity in practice adaptation and re-invention.
While the re-invention of introduced practices is not always consid-