Innovation is essential for agricultural and economic development, especially in today’s rapidly changing global environment. While farmers have been recognised as one of the key sources of innovation, many studies on agricultural innovations continue to consider farmers as adopters of externally-driven technologies only. This thesis, in contrast, analyses the innovation-generating behaviour among rural farmers. Specifically, the study looks at the determinants, impacts and identification of farmer innovation. The study is based on primary data obtained from a survey of 409 smallholder farm households in the Upper East region of northern Ghana. Additional data were collected through an innovation contest and a stakeholder workshop conducted in the region. Employing recursive bivariate probit and endogenous treatment-regression models which control for selection bias, it was found that participation in Farmer Field Fora − a participatory extension approach with elements of the innovation systems perspective − is a key determinant of innovation behaviour in farm households. Other important determinants are education, climate shocks and risk preferences. These results are robust to alternative specifications and estimation techniques. The study also found no spillover effect of FFF on farmers’ innovation capacity and discussed its implications. Using endogenous switching regression and propensity score matching techniques, the effect of farmer innovation on household welfare was analysed. The results show that farmer innovation significantly improves both household income and consumption expenditure for innovators. It also contributes significantly to the reduction of food insecurity among innovative households by increasing household food consumption expenditure, reducing the length of food shortages, and decreasing the severity of hunger. However, the findings show that the positive income effects of farmer innovation do not significantly translate into nutritious diet, measured by household dietary diversity. The results also indicate that though households innovate mainly to increase production, their innovations indirectly contribute to building their resilience to climate shocks. Overall, the results show positive and significant welfare effects of farmer innovation. Through an innovation contest that rewards farmers’ creativity and a household survey, 48 outstanding innovations developed by smallholder farmers were identified in the study region. The innovations are largely extensive modification of existing practices or combination of different known practices in unique ways to save costs or address crop and livestock production constraints. While some of the identified innovations can be recommended or disseminated to other farmers, most of them may require further validation or research. The multi-criteria decision making analysis − based on expert judgement ¬− is proposed as a simple and useful method that can be applied in prioritising high-potential innovations. Using this method, it was found that among the most promising innovations involve the control of weeds, pest and diseases using plant residues and extracts, and the treatment of livestock diseases using ethnoveterinary medicines. In conclusion, this study provides empirical evidence that smallholder farmers develop diverse and spectacular innovations to address the myriad challenges they face. These innovations also contribute significantly to household well-being, hence, need to be recognised and promoted. An institutional arrangement that permits interactions and learning among stakeholders may be a potential option for strengthening farmers’ innovation capacity.

This survey aims to collect data on farmers in the Niger basin of Benin. The data collected are relative to: (i) demographic information; (ii) crop production; (iii) livestock production; (iv) off-farm activities, and wages; (v) Access to extension, markets, credit, food consumption and social capital; (vi) climate change perception and shocks; (vii) adaptation strategies; (viii) household assets, and basic services. Three-stage sampling was used. First, municipalities were randomly selected within each agro-ecological zone (AEZ) based on their number of agricultural households. Second, villages were randomly selected within selected municipalities. Finally, random farm households were selected within selected villages. Therefore, the municipalities were randomly selected within each AEZ (AEZ I: one municipality, AEZ II: two municipalities, AEZ III: three municipalities, and AEZ IV: one municipality). The choice of the number of municipalities per AEZ is linked to the number of municipalities covered by AEZs I and IV (they covered two municipalities, and it has been decided to select one of the two). The number of municipalities for the AEZs II and III was determined proportionally to their size, referring to the size of AEZ I as a reference. Only four out of the five AEZs covered by the basin are considered, namely AEZ I (totally), AEZ II (totally), AEZ III (partially), and AEZ IV (partially). AEZ V was disregarded because only one of its municipalities is located within the Niger basin and it is a small part of the municipality that is included in the basin. Moreover, Pèrèrè was disregarded, because this municipality is partially covered by the basin (just a small part). Similarly, two municipalities that are partially included in the basin within AEZ III were avoided (Kouandé was maintained because its major part is within the basin). The municipalities were selected within each AEZ by the means of probability-proportional-to-size (PPS). Finally the municipalities that were chosen are: Malanville in AEZ I, Banikoara and Kandi in AEZ II, Bembèrèkè, Kouandé, and Nikki in AEZ III, and Natitingou in AEZ IV. The sample size was 545 agricultural households allocated across selected municipalities by the means of N-proportional allocation . Moreover, some adjustments have been made due to logistical constraints. Based on the allocation of the sample size across municipalities, it has been decided to allocate twenty households per villages, meaning four villages should be surveyed in each municipality, except Natitingou (three villages) and Kandi (due to the fact that one village of Kandi was already randomly selected for the pilot survey). At the end of the process, 28 villages had been surveyed. Due to logistical constraints, twenty agricultural households were not surveyed in every village.

The dataset consists of socio-economic data obtained from the survey of 409 farm households in the Bongo, Kassena Nankana East and Kassena Nankana West Districts in Upper East Ghana. It also includes constructed data from the main dataset for the analysis of the drivers and impacts of farmer innovation. A description of all the variables in the data is also included.