Farmer-led innovations and rural household welfare: Evidence from Ghana

Journal of Rural Studies 55 (2017) 263e274

Contents lists available at ScienceDirect

Journal of Rural Studies journal homepage: www.elsevier.com/locate/jrurstud

Farmer-led innovations and rural household welfare: Evidence from Ghana Justice A. Tambo a, *, Tobias Wünscher a, b a b

Center for Development Research (ZEF), University of Bonn, Walter-Flex-Str. 3, 53113 Bonn, Germany EARTH University, P. O. Box 4442-1000, San Jos
e, Costa Rica

a r t i c l e i n f o

a b s t r a c t

Article history: Received 29 March 2017 Received in revised form 18 August 2017 Accepted 28 August 2017

It is well recognized that agricultural innovations could emerge from many sources, including rural farmers. Yet the numerous micro-level studies on impacts of agricultural innovations have largely focussed on externally promoted technologies, and a rigorous assessment of impacts of farmer-led innovations is lacking. We address this issue by analyzing the effect of farmer-led innovations on rural household welfare, measured by income, consumption expenditure, and food security. Using household survey data from northern Ghana and applying endogenous switching regression and maximum simulated likelihood techniques, we find that farmer-led innovations significantly increase household income and consumption expenditure per adult equivalent. The innovations also contribute significantly to the reduction of household food insecurity by increasing food consumption expenditure, by decreasing the duration of food shortages, and by reducing the severity of hunger. Furthermore, we find that these effects are more pronounced for farm households whose innovative activities are minor modifications of existing techniques. Overall, our results show positive welfare effects of farmer-led innovations, and thus support increasing arguments on the need to promote farmer-led innovations (which have been largely undervalued) as a complement to externally promoted technologies in food security and rural poverty reduction efforts. © 2017 Elsevier Ltd. All rights reserved.

Keywords: Farmer-led innovations Household income Household expenditure Food security Impact assessment Ghana

Despite increased food production in the last decade, nearly 850 million people (12% of global population) continue to be hungry and food insecure, and many more are micronutrient deficient (Godfray et al., 2010; FAO et al., 2013). Most of these undernourished people are smallholders, who live in rural areas and on less than US$1.25 a day and derive their livelihoods from agriculture (McIntyre et al., 2009). Agricultural innovations can play essential roles in tackling the global food security challenge (Brooks and Loevinsohn, 2011) and in reducing rural poverty (de Janvry and Sadoulet, 2002). Over the past years, there has been increased development and diffusion of technological innovations by scientists, and farmers are being encouraged to adopt these innovations (Gatzweiler and Von Braun, 2016). With the rapidly changing economic environment, however, farmers have gone beyond the

adoption of the externally promoted innovations to develop their own technologies and to modify the externally introduced technologies to suit their local environments (Reij and Waters-Bayer, 2001; Sanginga et al., 2009; Tambo and Wünscher, 2015). Such innovation-generating practices among farmers, which are commonly referred to as farmer-led innovations, are claimed to play an important role in building local resilience to changing environments and in addressing food insecurity challenges (Reij and Waters-Bayer, 2001; Kummer et al., 2012; Tambo and Wünscher, 2017). Following Waters-Bayer et al. (2009), we define a farmer-led innovation to be a new or modified practice, technique or product that was developed by an individual farmer or a group of farmers without direct support from external agents or formal research.1 Thus, farmer innovators are farm households who have developed new techniques, tools or practices; have added value to common or traditional practices; or have modified external

* Corresponding author. E-mail addresses: [email protected], [email protected] (J.A. Tambo).

1 Other terms for farmer-led innovations include farmer innovations and farmerdriven innovations.

1. Introduction

http://dx.doi.org/10.1016/j.jrurstud.2017.08.018 0743-0167/© 2017 Elsevier Ltd. All rights reserved.

264

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

techniques or practices to suit their local conditions or farming systems.2 Therefore, simply adopting externally promoted technologies is not part of farmer-led innovations. There has in recent years been a surge of interest in analyzing the role of agricultural innovations in reducing poverty, hunger and malnutrition in developing countries. Many micro-level studies (e.g., Kijima et al., 2008; Minten and Barrett, 2008; Kassie et al., 2011; Asfaw et al., 2012) have shown that agricultural innovations have positive productivity, income, food security, and poverty reduction effects among adopters. These studies are, however, based on innovations developed and disseminated by National Agricultural Research System (NARS) and the Consultative Group on International Agricultural Research (CGIAR), and there is little evidence on the contribution of farmer-led innovations to economic well-being of farm households. Considering the numerous challenges hindering poor rural smallholders' adoption of externally promoted innovations (Barrett et al., 2004), it has been argued that farmer-led innovations might form the basis for rural livelihoods and food security (Reij and WatersBayer, 2001; The Worldwatch Institute, 2011). However, the few studies that have examined the potential impacts of farmer-led innovations (e.g., Reij and Waters-Bayer, 2001; Leitgeb et al., 2013; Kummer et al., 2017) were based on farmers’ subjective perceptions of the outcomes of their innovations or on selected case studies of farmer innovators, and did not account for possible selection bias. Thus, a rigorous assessment of the impact of farmer-led innovations is still lacking. Robust evidence is needed to be able to support increased arguments on the need for policy supports on farmer-led innovations as a complement to externally introduced innovations. Using survey data from farm households in northern Ghana, this study attempts to fill the void on the impacts of farmer-led innovations. Specifically, we examine whether farmer innovators are better off than non-innovators in terms of household income, consumption expenditure, and food and nutrition security. On the one hand, farmer-led innovations may improve productivity or save labour for non-farm activities and subsequently increase household income and food security. On the other hand, it is possible that the innovations developed by farmers may be unsuccessful or may not produce immediate result, hence, has negative effect on household income and food security. We employ endogenous switching regression and a maximum simulated likelihood estimator to account for potential non-random selection bias. We complement the regression results with analysis of perceived outcomes of farmer-led innovations as reported by the innovators. The rest of the paper is organised as follows. The next section presents the conceptual framework and estimation techniques. In section 3, we describe the welfare outcome indicators, followed by a presentation of the data and descriptive statistics in section 4. The empirical results are presented and discussed in section 5, while the last section summarises and concludes the paper.

2 The questions used to elicit farmer-led innovations can be summarized as follows. Have your household in the past year develop any new agricultural technique or did you modify or make any changes to farming techniques or practices in your community, on your own or jointly with other farmers without direct external assistance (e.g., from extension agents, researchers, NGOs, etc.)? If yes, please describe the practice. Note: All the practices described by the farmers were verified by confirming if they can be considered as farmer-led innovations. With the assistance of extension agents and experts who are knowledgeable about agricultural practices in the sample communities, we confirmed if a practice described by a farmer was not a common practice but rather a modified, an improved or a novel practice.

2. Conceptual framework and empirical approach In order to assess the effect of farmer-led innovations on household well-being, the farm household model that posits that households maximise utility subject to income, production, and time constraints (Singh et al., 1986) is used as a framework. The model integrates in a single framework, the production, consumption and work decision-making processes of farm households (Sadoulet and de Janvry, 1995). Following Weersink et al. (1998) and Fernandez-Cornejo et al. (2005), households are assumed to derive utility (U) from purchased consumption goods (G) and leisure (L), and the level of utility obtained from G and L is affected by exogenous factors such as human capital (H) and other household characteristics (Z). Thus:

MaxU ¼ UðG; L; H; ZÞ

(1)

Utility is maximised subject to:


 Time constraint : T ¼ F If þ M þ L; M  0

(2)

h

 i Production constraint : Q ¼ Q X If ; F If ; H; If ; R ; If  0

(3)

Income constraint : Pg G ¼ Pq Q  Wx X 0 þ WM 0 þ A

(4)

The total time endowment (T) of each household is allocated to leisure (L), working on the farm (F), or off-farm work (M). The level of farm output (Q) depends on the quantity of farm inputs (X), the innovativeness of farm household (If), F, H, and a vector of exogenous variables that shift the production function (R). X and F are functions of If since some of the farmer-led innovations are labour or input saving, hence, freeing some time and money for other uses. If in turn is determined by households’ experience of shocks, social capital, household assets, risk preference, H and Z. Equation (4) depicts the budget constraint on household income where Pg denote price of goods purchased. Thus, PgG is the income available for purchase of consumption goods, and it depends on the price (Pq) and quantity (Q) of farm output, price (Wx) and quantity (X) of farm inputs, off-farm wages (W) and the amount of time spent working off-farm (M) and exogenous household income such as government transfers, pensions and remittances (A). Substituting Equation (3) into Equation (4) yields a farm technology-constrained measure of household income:

h

0 i Pg G ¼ Pq Q X If ; F If ; H; If ; R  Wx X 0 þ WM0 þ A

(5)

The Kuhn-Tucker first order conditions can be obtained maximising Lagrangean expression (L ) over (G, L) and minimising it over (l, h):

L ¼ UðG; L; H; ZÞ n h

0 i o þl Pq Q X If ; F If ; H; If ; R  Wx X 0 þ WM0 þ A  Pg G h
 i þh T  F If  M  L (6) where l and h represent the Lagrange multipliers for the marginal utility of income and time, respectively. Solving the Kuhn-Tucker conditions, reduced-form expression of the optimal level of household income (Y*) can be obtained by (Fernandez-Cornejo et al., 2005):

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274


 Y * ¼ Y If ; Wx ; Pq ; Pg ; A; H; Z; R; T

(7)

and household demand for consumption goods (G) can be expressed as:


 G ¼ G If ; W; Pg ; Y * ; H; Z; T

(8)

Thus, the reduced forms of Y* and G are influenced by a set of explanatory factors, including If. The main aim of this paper is to estimate the effect of If on household income, household consumption of goods and other related outcome variables such as food security. Our regression model is a reduced form equation that relates farmer-led innovations and other explanatory variables to household welfare. This can be expressed as:

y ¼ 4V þ dIf þ m

(9)

where y denotes household welfare indicators such as income, consumption expenditure and food security. V is a vector of explanatory variables that influence the outcome variables with the associated parameters f. V includes household, farm and contextual characteristics such as age, gender and educational level of household head, household size, farm size, asset endowments, social network variables, risk preference and district dummies. If is a dummy capturing whether or not a household is an innovator, and the coefficient d measures the effect of farmer-led innovation on household well-being. This variable is potentially endogenous since farmer-led innovation is not randomly assigned and farmers may decide whether or not to innovate. In other words, innovative farmers may be systematically different from non-innovators, and these differences may obscure the true effect of innovation on household well-being. To address this issue, we use the endogenous switching regression (ESR) technique. In the ESR method, separate outcome equations are specified for treatment and control groups, conditional on a selection equation. Thus in our case, we estimate separate household well-being indicators for farmer innovators and non-innovators, conditional on the innovation decision. This can be specified as follows.

If ¼ gD þ ε y1 ¼ 41 V þ m1 y0 ¼ 40 V þ m0

if If ¼ 1 if If ¼ 0

(10)

where D is a vector of farm and household characteristics and V is

y1 ¼ 41 V þ sm1 ε l1 þ x1 y0 ¼ 40 V þ sm0 ε l0 þ x0

if If ¼ 1 if If ¼ 0

(11)

where l1 and l0 are the inverse mills ratios evaluated at gD; ¼ cov (m1, ε); and sm0 ε ¼ cov (m0, ε). Thus, estimates from the selection equation are used to compute l1 and l0 which are then added to the outcome equations to correct for selection bias, and this can be estimated using a two-stage method (Maddala, 1983). However, we use the full information maximum likelihood (FIML) estimation approach (Lokshin and Sajaia, 2004), which estimates the selection and outcome equations simultaneously.3 This is more efficient than the two-step procedure. If sm1 ε and sm0 ε in Equation (11) are statistically significant, we have endogenous switching. While the FIML ESR model is identified through non-linearities of l1 and l0 (Lokshin and Sajaia, 2004), a better identification requires an exclusion restriction. That is, we need at least one variable that affects farmers' innovation decisions but does not directly affect any of the households’ well-being indicators. Our identifying variable is the distance between a household and the nearest farmer field fora (FFF) meeting place. FFF is a platform for innovation, mutual learning and knowledge sharing among agricultural stakeholders, which has been implemented in the study region with the aim of building the capacities of farmers to become experts in the development of location-specific technologies and managerial practices. Participating in FFF has been found to be an important determinant of farmer innovation (Tambo and Wünscher, 2016), and farmers living in close proximity to FFF meeting place are expected to be more likely to participate due to more exposure to information about FFF, and might consequently develop innovations. Thus, proximity to FFF meeting place is a good predictor of farmer-led innovations. However, distance to FFF meeting place is not directly related to household well-being, making it a suitable identifying variable.4 Following Di Falco et al. (2011) and Asfaw et al. (2012), the admissibility of distance to FFF meeting place as a valid instrument is established by performing a falsification test: if a variable is an appropriate selection instrument, it will affect innovation decision but it will not affect the welfare outcomes of non-innovating households. The results (see the Appendix) indicate that the distance to FFF is a statistically significant determinant of farmer-led innovations (Table A.1) but not any of the welfare indicators of non-innovative households (Table A.2). Thus, distance to FFF meeting place can be regarded as a valid selection instrument. The coefficients from the ESR model can be used to derive the expected values of well-being, which are then used in estimating the average treatment effect on the treated (ATT):


 
      ATT ¼ E y1 If ¼ 1  E y0 If ¼ 1 ¼ Vð41  40 Þ þ l1 sm1 ε  sm0 ε

defined as in Equation (9). y1 and y0 represent a vector of welfare indicators for innovators and non-innovators, respectively. 41 and 40 are parameters to be estimated for the innovators and noninnovators regimes, respectively. When the error term of the selection equation (ε) is correlated with the error terms of the outcome equation of innovators (m1) and non-innovators (m0), then we have a selection bias problem. The outcome equations in Equation (10) can further be specified as (Fuglie and Bosch, 1995):

265

(12)

The ATT compares the well-being of farmer innovators with and without farmer-led innovations. In the estimation procedure outline above, the farmer innovators are lumped together and compared with non-innovators, which may obscure important information. We try to address this limitation and account for the uniqueness of the innovations by

3

The models were estimated using the movestay command in Stata. Another potential instrument would have been participation in FFF, but this variable itself is endogenous. 4

266

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

categorizing them into two groups e major and minor innovations e based on the innovators' own descriptions and expert judgement.5 Based on the experts’ consensus, we categorise the innovations into major and minor innovations depending on the extent to which an innovation differs from existing practices or the degree of newness of the innovation. Thus, innovations that are novel or relate to substantial improvement of existing practices or techniques are classified as major innovations, while those involving slight modifications to existing practices and techniques are deemed minor innovations. Consequently, we analyse the differential effects of farmer-led innovations on household welfare by disaggregating our sample into major innovators, minor innovators and non-innovators. Our treatment variable now assumes a multinomial distribution; hence, estimation techniques that can account for endogeneity issues and the multinomial nature of the farmer-led innovation variable are required. Consequently, we employ the multinomial treatment effects model proposed by Deb and Trivedi (2006). The model allows for the estimation of the effects of an endogenous multinomial treatment variable on binary, count or continuous outcomes, while accounting for selectivity bias. This model also consists of two parts (selection and outcome equation) that are jointly estimated. Following Deb and Trivedi (2006), we assume that the probability of observing a farm household i in one of the innovation category j has a mixed multinomial logit structure:

Pr




Ifi jDi ; li




exp gj D0 i þ lij ¼ P 1 þ Jk¼1 expðgk D0 i þ lik Þ

(13)

The expected outcome equation can be specified as: J J 
X X E yi jIfi ; Di ; li ¼ 40 i V þ dj Ifi j þ lj lij j¼1

(14)

j¼1

where j ¼ 0,1,2, which corresponds to non-innovation, major innovation and minor innovation, respectively. dj is a set of dummies denoting treatment effects of major and minor innovations relative to the base category (non-innovation). lij are latent factors, capturing the unobserved factors affecting household's innovation-decision process and the outcome variables. lj are coefficients associated with unobservable characteristics and can be interpreted in terms of selection effects. The rest of the variables are defined as in Equation (10). We estimate the parameters in the multinomial treatment effects model using the maximum simulated likelihood (MSL) procedure (Deb, 2009).6

3. Measuring household welfare As already mentioned, we evaluate the effect of farmer-led innovations on a number of welfare outcomes, such as household income, consumption expenditure and food security. Below, we explain these outcome measures in detail.

5 We convened a participatory stakeholder's workshop and invited 12 experts who have knowledge on farmer innovations and farming systems in the study region. The experts were selected from research institutions, Ministry of Food and Agriculture, NGO's and farmer organizations and agricultural extension officer that are operating within the sample communities. The purpose of the workshop was to evaluate all the reported farmer-led innovations. 6 This was implemented using stata routine mtreatreg. The MSL procedure was run with 200 replications.

3.1. Household income and consumption expenditure Most of the farmer-led innovations are yield-related, hence, are expected to affect productivity and consequently farm income. However, farmer innovation may result in resource reallocation, which could have indirect effect on household income. For instance, a household involved in labour-saving innovations could have surplus labour for non-farm activities and earn extra income. To capture these potential indirect effects, we analyse the effect of farmer innovation on total household income, which comprises farm and off-farm income. The household income was expressed in annual per adult equivalent (AE) basis.7 While household income can be used as a measure of household well-being, consumption expenditure is often preferred because it is less prone to seasonal fluctuations and measurement errors, hence, more reliable (Deaton, 1997). It is expected that innovative practices of households will result in increased yields or outputs, and thus more consumption of farm products or more income from sales of products for the consumption of other goods. In addition, the resource allocation effects of innovation may induce changes in consumption expenditure. Household consumption expenditure consists of different sub-components, including food consumption, housing, energy, transportation, communication, health, and educational expenses; expenditures on other consumer durables and non-durables; and transfer payments made by households. The survey questionnaire captured the value of household consumption out of purchases, home production and, all items received in kind. The non-purchased goods were valued at local market prices. A 7day recall period was used to capture food expenditure, and a 30day recall period was used for frequently purchased items or services and non-durable goods, while a 12-month recall period was used for durable items and transfer payments. All the recall periods were standardized to one year, and the different sub-components were aggregated to obtain total household consumption expenditure, which was expressed in per AE terms. 3.2. Food and nutrition security There is no unified measure of food and nutrition security, and this is partly due to its complexity and multidimensionality (Pinstrup-Andersen, 2009; Barrett, 2010). Many studies have used different measures ranging from caloric intake, dietary quality, and anthropometric estimates in order to capture the key dimensions of food security: availability, accessibility, utilization and stability. Most of these measures are, however, relatively time-consuming and costly to implement (de Haen et al., 2011). In this study we employ the standard food security measure  food consumption expenditure, as well as three other indicators which are relatively quick and easy to measure. These are food gap/deficit, Household Hunger Scale (HHS) and Household Dietary Diversity Score (HDDS). The food consumption expenditure forms part of the total household consumption expenditure discussed above. The food gap/deficit is a subjective measure of food security, and it refers to the number of months in the past 12 months that households have difficulty satisfying their food needs due to depletion of own food stocks or lack of money to purchase food. This measure is also known as the months of inadequate household food provisioning (MIHFP) (Bilinsky and Swindale, 2005). Another perception-based measure of food insecurity we employed is the HHS, which is suitable to use in highly food

7 We use the OECD adult equivalent scale which is given by 1 þ 0:7ðA  1Þ þ 0:5C , where A and C represent the number of adults and children in a household, respectively.

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

insecure areas (Ballard et al., 2011), as in our case. The HHS is a subset of the Household Food Insecurity Access Scale (HFIAS) developed by Food and Nutrition Technical Assistance (FANTA) project of the US-AID, but unlike the HFIAS, the HHS has been validated for cross-cultural use (Ballard et al., 2011). The HHS is related to food access dimension of food security, and it is based on three questions. That is, how often in the past 30 days: 1) was there no food of any kind in the house; 2) did a household member go to sleep hungry; and 3) did a household member go a whole day without eating. The response to each question was coded: 0 ¼ never; 1 ¼ rarely or sometimes; and 2 ¼ often.8 The sum of these responses yields the HHS score, which ranges from 0 (no hunger) to 6 (severe hunger). Households were interviewed in April 2012 which is around the peak period of the lean season in the study area, and hence, an appropriate period to use the HHS, which measures severe level of food insecurity. Finally, we use a dietary diversity indicator, HDDS as another measure of the access facet of food and nutrition security. We assess whether the potential improvement in food production or household income though innovation translates into better nutritional quality of diets. The HDDS, which was also developed by the FANTA project, is obtained by simply summing the total number of 12 food groups consumed by household members in the home during the past 24 h (Swindale and Bilinsky, 2006). The food groups include cereal, roots and tubers, legumes and nuts, vegetables, fruits, fish and seafood, eggs, meat and poultry, milk and milk products, oils and fats, sweets, and miscellaneous such as spices.9 As suggested by Swindale and Bilinsky (2006), we made sure that there were no special occasions, such as funeral within the sample households, which might influence their food consumption pattern during the 24-hour period.

4. Data and sample characteristics The empirical analysis is based on data for the 2011e2012 agricultural season obtained from a household survey conducted within the research programmedWest African Science Service Center for Climate Change and Adapted Land Use (WASCAL)d funded by the German Federal Ministry of Education and Research (BMBF). Data collection took place in Bongo, Kassena Nankana east and Kassena Nankana west districts in upper east region, one of the poorest administrative regions of Ghana. Overall, our sample consists of 409 farm households (101, 156 and 152 from Bongo, Kassena Nankana east and Kassena Nankana west districts, respectively) randomly selected from the three districts. Interviews were conducted with the aid of pre-tested questionnaires and were supervised by the first author. The data collection took place in two phases due to the bulky nature of the questionnaire and the potential differences in perceived food insecurity across the three districts as a result of different survey days. The first phase was conducted between December 2012 and March 2013. The questionnaire used in this phase captured data on household and plot characteristics, crop and livestock production, off-farm income earning activities, innovation-generating activities, access to infrastructural services, information and social interventions, household experiences with shocks, climate change adaptation strategies and risk preferences.10 The second phase of

8 For data collection, “rarely” and “sometimes'’ categories were separated as recommended by Ballard et al. (2011). 9 We use a disaggregated set of food groups, which were then combined into 12 food groups to generate the HDDS (Swindale and Bilinsky, 2006). 10 We measured households' subjective risk preferences using the Ordered Lottery € m, 2008). Selection Design with real payoffs (Harrison and Rutstro

267

the survey took place just after the end of the first phase and was conducted simultaneously in the three districts so that the households’ subjective responses to food insecurity are not influenced by differences in survey days. In the second phase, the same households were revisited and all but one household were reinterviewed. Thus, the sample size in the second phase is 408. The second phase was used to obtain data on the food security indicators (HHS, HDDS and food consumption) as well as household consumption expenditure. Table 1 outlines the description of the variables used in the regression and their mean values. The explanatory variables were motivated by literature on agricultural innovation, and they include household and farm characteristics as well as institutional and access-related variables. We also include district dummies to control for district fixed effects. The table shows that an average household has 7 people. Majority of the households are maleheaded, and household heads are mostly middle-aged with very low level of education. Households generally have about 5 acres of land, and many households have been affected by shocks, particularly climatic shocks. The summary statistics of the outcome variables, which are presented in the lower part of Table 1, indicate that the average household income per AE is about 532 GH¢. The average food consumption expenditure of nearly 454 GH¢ accounts for about 58 percent of average total consumption expenditure. On average, households experience about 3 months (mostly from April to June) of inadequate food provisioning. The average HHS of about 1.13 suggests that severe food insecurity or hunger is not pervasive in the study region. The table also shows that about 41 percent of the sampled households implemented farmer-led innovations during the past year. Disaggregating the innovations, we find that about 16 percent of the households are major innovators while 25 percent of them are minor innovators. Majority of the farmer-led innovations are related to agronomic practices. Example of these innovations include modification of cropping pattern, seeding rate and planting spacing; soil fertility measures such as new methods of compost preparation or methods to prevent soil nutrient loss; and control of weeds, pests and diseases using biopesticides made from local plants. Some of the innovations are related to livestock production, and they include new formulations of animal feed and applying herbal remedies in the treatment of livestock diseases. Other innovations include developing and using new farming tools, and storage of farm products using local plant extracts or grasses. Some specific examples of the most promising farmer-led innovations are presented in Tambo and Wünscher (2015). 5. Results and discussion In this section, we present the results of the effect of farmer-led innovations on economic-well-being of farm households. We first look at the outcomes of innovation practices as subjectively stated by the innovative farmers before presenting the econometric results. 5.1. Subjective outcome of farmer-led innovations To corroborate the results from the regression analysis, all the farmer innovators were asked about the outcomes observed from their innovative practices, and their subjective responses are summarised in Fig. 1. The figure shows that increased production is the major outcome of the farmers' innovations. Most of the innovative practices listed by the farmers are yield-related (e.g., crops and crop varieties, soil fertility, and pest and disease control); so, it is not surprising that increased production is the most mentioned

268

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

Table 1 Definition of variables in the regression and summary statistics Variable

Description

Mean SD

Treatment variables Farmer-led innovation Household has modified or developed a novel farming practice, tool or technique without direct support from formal research or external agents Major innovation Innovation is novel or is a substantial modification of existing practices Minor innovation Innovation is a minor modification of existing practices Explanatory variables Age Age of household head Gender Gender of household head (1 ¼ male) Household size Number of household members Education Education of household head (years) Land holding Total land owned by household in acres a Livestock holding Total livestock holding of household in Tropical Livestock Units (TLU) a Assets value Total value of non-land productive assets in 100 GH¢b Off-farm activity Household engage in off-farm income earning activities Road distance Distance from household to nearest all-weather road in km Group membership A household member belongs to a group Climate shock Household suffered from droughts or floods in the past 5 years Labour shock Death or illness of a household member one year prior to survey c Risk preference Risk attitude of household FFF distance Distance from household to the nearest FFF meeting place in km Bongo District Household is located in Bongo District KNW District Household is located in Kassena Nankana west district KNE District Household is located in Kassena Nankana east district Outcome variables Household income Total household income per adult equivalent Consumption Total household consumption expenditure per adult equivalent expenditure Food consumption Total food consumption expenditure per adult equivalent Food gap/deficit Number of months of inadequate household food provisioning HHS Household Hunger Scale Score HDDS Household Dietary Diversity Score a b c

0.41

0.49

0.16 0.25

0.37 0.43

49.42 0.86 6.64 1.67 4.56 2.92 4.54 0.76 0.54 0.64 0.91 0.60 2.53 4.35 0.25 0.37 0.38

14.88 0.35 2.59 1.10 4.15 3.41 6.92 0.43 0.84 0.48 0.29 0.49 1.17 7.07 0.43 0.48 0.49

531.69 768.68 779.08 627.29 453.83 2.85 1.13 7.14

330.66 1.68 1.27 1.96

Assets or livestock acquired during the 2011e2012 agricultural season were excluded to avoid problems of endogeneity. The exchange rate at the time of the survey was US $ 1 ¼ 1.90 GH¢ (OANDA, 2017). This ranges from 1 (extremely risk averse) to 6 (neutral to risk preferring).

Fig. 1. Subjective outcome of farmer-led innovations.

outcome. Increased income and improved food security are also important outcomes observed by the farmer innovators. These two outcomes may stem from the increase in production, and together, they point out the potential positive welfare effects of farmer-led innovations. Another positive effect of the farmers’ innovations is

labour saving, and thus reduction in production costs and freeing of labour for off-farm employment. Some farmers implement innovation-generating practices in order to make better farming decisions, and others discover innovations out of curiosity or serendipity; hence, this explains the significant number of

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

innovators asserting increased knowledge or satisfaction as outcomes of their innovations. A few of the farmers indicated that their innovations were unsuccessful, and this is expected since innovation generally involves decision making under uncertainty, which can result in positive or negative outcomes. Similar subjective outcomes were obtained by Leitgeb et al. (2013) and Kummer et al. (2017) in studies on farmer innovators in Cuba and Austria, respectively. 5.2. Econometric results Analysis of farmers’ perceptions in the previous section shows potential positive effects of farmer-led innovations. To properly analyse impacts, we use an econometric technique, the FIML ESR. The first-stage estimation results of the FIML ESR model, which show the determinants of farmer-led innovations, are presented in Table A.1 in Appendix.11 Our excluded instrument, FFF distance, is statistically significant in all the models, thus satisfying the instrument relevance condition. We now look at the results for the various welfare indicators. 5.2.1. Effect of farmer-led innovations on household income and consumption The second-stage estimates of the FIML ESR model for household income are presented in Table 2. The correlation coefficients between the error terms of the selection and outcome equations (r1 and r0) reported at the bottom part of the table provide an indication of selection bias. A statistical significance of any of them suggests that self-selection would be an issue if not accounted for. The correlation coefficients for the innovators (r1) and noninnovators (r0) equations are both negative but only r1 is statistically significant, suggesting that there is self-selection among innovators. Thus, farm households with lower than average household income for innovators are less likely to develop farmerled innovations. The significance of the likelihood ratio tests for independence of equations also indicates that there is joint dependence between the selection and income equations for innovators and non-innovators. The results show that household size, value of assets, off-farm activity and livestock holding significantly affect the income of both innovators and non-innovators. An increase in household size results in a decline in income, while access to off-farm incoming earning activities, larger livestock holding and higher value of household assets are positively associated with income. There are some differences between what determines income among innovators and non-innovators, and this justifies the use of the ESR model. For example, labour shock is significantly and negatively correlated with income of non-innovators, but the effect is insignificant among innovators. Conversely, gender of household head significantly influences the income of only innovators. The predicted household incomes from the ESR model are used to compute the treatment effects of farmer-led innovations on household income, and the results are presented in Table 3. The ATT measures the mean difference between the actual income of farmer innovators and what they would have earned if they had not innovated. The results show that farmer-led innovation has a positive and statistically significant effect on household income of the innovating households. Specifically, farmer-led innovation increases per adult equivalent household income of innovators by about 9 percent, and this effect is statistically significant. This

11

The first-stage results are not discussed in this paper since a detailed analysis and discussion are presented in another publication, which is available upon request.

269

finding confirms the farmers’ subjective reports of the positive income effects of their innovations.12 Table 2 also shows the estimation results of the consumption expenditure model. The results show that household size significantly reduces consumption expenditure of both innovators and non-innovators, but the effect is more pronounced for innovators. The value of household assets also significantly increases consumption expenditure for both groups, but the coefficients for other wealth-related variables (for example, livestock holding and off-farm activity) are not statistically significant. The positive and significant coefficient of the district dummies in both innovation regimes suggests that farm households in the KNE and KNW districts have higher consumption expenditure than those in Bongo district. This is not surprising since Bongo district is one of the poorest districts in the upper east region of Ghana. The results also show some differences between innovators and non-innovators with respect to some of the covariates. For instance, group membership and educational level of household head have significantly positive effects on the consumption expenditure of innovative households, but their effects are insignificant for noninnovators. The statistical significance of the correlation coefficient (r1) suggests that there is selection effect. Thus, unobserved factors affect both the innovation decision and household consumption expenditure, and this would have caused a bias if not controlled for. The result for the treatment effect of farmer-led innovations on consumption expenditure per AE is presented in Table 3. The ATT result shows that farmer innovators significantly increased their consumption expenditure per AE by about 6 percent as a result of their innovations. This positive consumption effect may stem from the revenue increase or cost reduction potential of farmer-led innovations. This also implies that the positive income effects of farmer-led innovation reported earlier translate into increased household consumption. 5.2.2. Effect of farmer-led innovations on food and nutrition security As already indicated, four different measures of food security are used in the estimation of the effect of farmer-led innovations on food and nutrition security. The second stage results for all the four indicators are presented in Table 2. The correlation coefficient (r1) in the food gap, food consumption expenditure and HDDS models are statistically significant while that of the HHS model is not significant, suggesting heterogeneous results depending on the food security indicator employed. The estimated coefficients of the determinants of the four food security measures further highlight the presence of heterogeneous sample and effects. For instance, the included covariates largely influence the various food security indicators differently. Similarly, the variables that explain food security of innovators do not affect that of non-innovators, and vice versa. Only the location variables are statistically significant in all the four models. Similar to the results in the consumption expenditure model, the coefficient of the district dummies suggests that households located in KNE and KNW districts are more food secure compared with households in the relatively poor Bongo district. Among the determinants of household food security are gender of household head, household size, value of household assets, and climate and labour shocks. The results indicate that female-headed households are more

12 Since most of the innovations are farming-related, we also analysed their effect on only farm income. We found that farmer-led innovations significantly enhance the farm income per AE of innovators by about 16 percent. The estimated results are not presented in Tables 2 and 3 for brevity but are available upon request.

270

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

Table 2 ESR estimates of determinants of household welfare indicators Household income

Age Gender Household size Education Land holding Livestock holding Assets value Off-farm activity Group membership Climate shock Labour shock Road distance Risk preference KNW District KNE District Constant

s1, s0 r1, r0 LR test of indep. eqns. No. of observations

Consumption expenditure

Food gap

Innovators

Non-innovators

Innovators

Non-innovators

Innovators

Non-innovators

0.003 (0.006) 0.427* (0.222) 0.118*** (0.036) 0.025 (0.022) 0.000 (0.017) 0.124*** (0.026) 0.028** (0.012) 0.666*** (0.184) 0.104 (0.194) 0.186 (0.244) 0.217 (0.152) 0.080 (0.087) 0.011 (0.045) 0.770*** (0.213) 0.824*** (0.220) 5.575*** (0.653) 0.995*** (0.158) 0.704*** (0.237) 14.20*** 409

0.003 (0.004) 0.244 (0.177) 0.135*** (0.023) 0.006 (0.018) 0.068*** (0.022) 0.068*** (0.021) 0.037*** (0.009) 0.400*** (0.139) 0.132 (0.132) 0.288 (0.247) 0.336** (0.130) 0.075 (0.076) 0.002 (0.038) 0.157 (0.169) 0.467*** (0.168) 5.067*** (0.480) 0.884*** (0.052) 0.283 (0.233)

0.001 (0.003) 0.111 (0.113) 0.154*** (0.019) 0.022** (0.011) 0.009 (0.009) 0.002 (0.013) 0.014** (0.006) 0.022 (0.092) 0.219** (0.094) 0.258** (0.125) 0.118 (0.078) 0.057 (0.045) 0.041* (0.022) 0.260** (0.107) 0.304*** (0.112) 6.638*** (0.305) 0.522*** (0.063) 0.766*** (0.134) 5.47** 408

0.003 (0.002) 0.096 (0.082) 0.096*** (0.011) 0.001 (0.009) 0.012 (0.011) 0.007 (0.010) 0.015*** (0.004) 0.002 (0.066) 0.104 (0.071) 0.120 (0.119) 0.018 (0.062) 0.011 (0.035) 0.023 (0.020) 0.397*** (0.081) 0.530*** (0.081) 6.443*** (0.250) 0.427*** (0.020) 0.043 (0.429)

0.013 (0.010) 0.244 (0.356) 0.034 (0.061) 0.005 (0.037) 0.009 (0.031) 0.049 (0.041) 0.010 (0.019) 0.148 (0.290) 0.262 (0.384) 0.243 (0.416) 0.053 (0.246) 0.137 (0.140) 0.129 (0.086) 0.086 (0.334) 0.005 (0.375) 2.678** (1.128) 1.454*** (0.084) 0.049 (0.649) 3.52* 409

0.006 (0.007) 0.947*** (0.304) 0.033 (0.040) 0.019 (0.031) 0.046 (0.038) 0.028 (0.037) 0.036** (0.016) 0.133 (0.243) 0.180 (0.228) 0.271 (0.420) 0.018 (0.228) 0.070 (0.130) 0.035 (0.066) 0.679** (0.295) 1.393*** (0.297) 5.630*** (0.811) 1.609*** (0.101) 0.433*** (0.168)

Innovators

Non-innovators

Innovators

Non-innovators

Innovators

Non-innovators

0.003 (0.007) 0.036 (0.253) 0.028 (0.042) 0.001 (0.025) 0.000 (0.019) 0.022 (0.029) 0.018 (0.013) 0.166 (0.204) 0.374 (0.232) 0.305 (0.274) 0.042 (0.173) 0.051 (0.098) 0.074 (0.053) 0.393 (0.238)* 0.508 (0.256)** 1.828

0.007 (0.006) 0.441* (0.246) 0.002 (0.033) 0.002 (0.026) 0.030 (0.034) 0.051 (0.032) 0.010 (0.013) 0.218 (0.198) 0.009 (0.213) 0.950*** (0.365) 0.347* (0.186) 0.105 (0.105) 0.080 (0.061) 0.222 (0.243) 0.785*** (0.240) 3.239***

0.001 (0.003) 0.124 (0.109) 0.117*** (0.020) 0.003 (0.012) 0.005 (0.011) 0.004 (0.013) 0.006 (0.006) 0.042 (0.091) 0.049 (0.146) 0.186 (0.139) 0.098 (0.0749) 0.012 (0.043) 0.029 (0.027) 0.446*** (0.102) 0.466*** (0.109) 6.935***

0.005** (0.002) 0.088 (0.090) 0.084*** (0.012) 0.000 (0.010) 0.011 (0.011) 0.001 (0.011) 0.004 (0.005) 0.048 (0.073) 0.159** (0.079) 0.061 (0.129) 0.060 (0.068) 0.030 (0.039) 0.032 (0.023) 0.511*** (0.088) 0.508*** (0.091) 6.620***

0.017 (0.011) 0.060 (0.414) 0.015 (0.067) 0.016 (0.039) 0.061** (0.048) 0.028 (0.048) 0.028 (0.022) 0.038 (0.334) 0.208 (0.343) 0.142 (0.438) 0.727*** (0.283) 0.127 (0.160) 0.208*** (0.080) 1.913*** (0.391) 1.950*** (0.416) 4.764***

0.002 (0.008) 0.397 (0.314) 0.022 (0.042) 0.003 (0.033) 0.081** (0.040) 0.039 (0.039) 0.024 (0.016) 1.036*** (0.251) 0.232 (0.247) 0.948** (0.445) 0.276 (0.235) 0.012 (0.134) 0.041 (0.071) 1.545*** (0.305) 1.826*** (0.307) 5.879***

HHS

Age Gender Household size Education Land holding Livestock holding Assets value Off-farm activity Group membership Climate shock Labour shock Road distance Risk preference KNW District KNE District Constant

Food expenditure

HDDS

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

271

Table 2 (continued ) HHS

s1, s0 r1, r0 LR test of indep. eqns. No. of observations

Food expenditure

HDDS

Innovators

Non-innovators

Innovators

Non-innovators

Innovators

Non-innovators

(0.715)** 1.031 (0.062)*** 0.101 (0.402) 2.39 408

(0.744) 1.276*** (0.667) 0.119 (0.432)

(0.393) 0.460*** (0.109) 0.475 (0.681) 4.77** 408

(0.264) 0.494*** (0.063) 0.643** (0.287)

(1.089) 1.819*** (0.186) 0.624*** (0.160) 9.68*** 408

(0.901) 1.666*** (0.141) 0.458* (0.278)

***, **, *Represent 1, 5, and 10 percent significance levels, respectively. Values in parentheses are standard errors.

Table 3 Treatment effects of farmer-led innovations Outcome

Household income per AE Consumption expenditure per AE Food gap/deficit (months) Household Hunger Scale Score Food expenditure per AE Household Dietary Diversity Score

Innovation decision Innovating

Not innovating

5.77 6.82 2.62 1.03 6.27 7.31

5.29 6.44 3.88 1.37 5.86 8.47

ATT

ATT in %

0.48*** 0.38*** 1.26*** 0.34*** 0.41*** 1.16***

9.07 5.90 32.47 24.82 7.00 13.70

***Represents 1 percent significance level.

likely to have extra months of food inadequacy and their household members are more likely to experience hunger, but the coefficients are only significant for non-innovators. This is probably because women in the study region have limited access to land and other resources needed to achieve food security (Apusigah, 2009). This is also in line with studies in Africa that found that female-headed households are more likely to be food insecure than male-headed households (e.g., Kassie et al., 2014). The value of household assets significantly decreases the number of months of food shortages for non-innovators. This is plausible since households in the study region have a tendency of depleting their productive assets as a coping mechanism to food insecurity (Quaye, 2008). The results for the treatment effects of farmer-led innovations on food and nutrition security are presented in Table 3. The results indicate that farmer-led innovations play a key role in food insecurity reduction among innovators. The innovations of farm households help to reduce the length of food gap periods by one month. In other words, if households that innovated were not to innovate, they would have had an extra month of food insufficiency. Analogously, farmer-led innovations significantly reduce household hunger by 0.34 index points, which amounts to about 25 percent reduction in the severe level of food insecurity for innovators. In addition, the innovations significantly caused an increase in food consumption expenditure per AE by about 7 percent for innovative households, which further confirms the positive food security effects of farmer-led innovations. The ATT estimate for the HDDS, however, suggests that farmer-led innovations do not increase household dietary diversity. Specifically, the innovations significantly decrease dietary diversity by 1.16 index points (or about 14%) for innovators. This suggests that the income benefits of farmers’ innovations do not necessarily translate into nutritious diets. Thus, the increased food consumption expenditure reported earlier is related to availability, and not diversity of food. In fact, the data on household expenditure indicates that a large share of the expenditure on food is devoted to cereal staples such as millet, maize and sorghum. Overall, farmer-led innovations improve food

security for innovative households, and this corroborates the subjective outcomes reported by the innovators as well as other qualitative evidences on the impact of farmer-led innovations (e.g., Reij and Waters-Bayer, 2001; Reij et al., 2009).

5.2.3. Treatment heterogeneity We have shown above that farmer-led innovations are significantly associated with increased household income and consumption, and reduced food insecurity. In this section, we examine if these impacts vary depending on the degree of innovativeness of the farmer-led innovations by comparing major and minor innovators with non-innovators. Table 4 displays the results for the MSL estimations of the differential effects of the farmer-led innovations. The results show that there is no significant difference between major innovators and non-innovators in terms of household income and consumption expenditure. However, minor innovations are significantly associated with higher income and consumption expenditure. In particular, compared to noninnovations, minor innovations increase household income per AE and consumption expenditure per AE by 79 percent and 41 percent, respectively.13 Similarly, we find that minor innovations are significantly associated with better household food security. The results show that minor innovations contribute to a decrease in the duration of food shortages in the household by nearly one month, a decrease in the severity of food insecurity by 0.91 index points, and an increase in food consumption per AE by 21 percent relative to noninnovations. By contrast and similar to the results of the income and consumption effects, we find that the major farmer-led innovations do not significantly enhance household food security. In fact, in comparison to non-innovators, major innovators are worse off in terms of all the food security indicators. The results also

13 Dummy coefficients in models with a log-dependent variable are interpreted as in Halvorsen and Palmquist (1980).

272

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

Table 4 MSL estimates of the differential impacts of farmer-led innovations

Major innovators Minor innovators Age Gender Household size Education Land holding Livestock holding Assets value Off-farm activity Group membership Climate shock Labour shock Road distance Risk preference KNW District KNE District Constant

l(major innovators) l(minor innovators) No. of observations

Income

Cons. Exp

Food gap

HHS

Food Expend.

HDDS

0.226 (0.288) 0.580* (0.335) 0.004 (0.006) 0.900*** (0.231) 0.096*** (0.032) 0.010 (0.022) 0.035* (0.021) 0.117*** (0.027) 0.038*** (0.012) 0.765*** (0.187) 0.241 (0.172) 0.492* (0.287) 0.282* (0.165) 0.117 (0.095) 0.036 (0.047) 0.248 (0.218) 0.623*** (0.226) 4.866*** (0.606) 0.623*** (0.214) 0.330 (0.325) 408

0.131 (0.127) 0.342*** (0.097) 0.002 (0.002) 0.161** (0.073) 0.107*** (0.011) 0.011 (0.007) 0.002 (0.008) 0.011 (0.010) 0.007 (0.004) 0.038 (0.061) 0.129** (0.051) 0.137 (0.083) 0.130** (0.053) 0.067** (0.028) 0.017 (0.015) 0.412 (0.007) 0.410 (0.008) 7.999*** (0.175) 0.211 (0.123) 0.378** (0.130) 408

0.394 (0.667) 0.884** (0.393) 0.001 (0.006) 0.735*** (0.236) 0.030 (0.033) 0.018 (0.023) 0.010 (0.024) 0.004 (0.028) 0.025** (0.012) 0.001 (0.189) 0.082 (0.173) 0.234 (0.285) 0.035 (0.168) 0.093 (0.096) 0.073 (0.049) 0.511** (0.220) 0.794*** (0.228) 4.591*** (0.613) 0.280 (0.739) 0.764* (0.426) 409

0.764*** (0.045) 0.914*** (0.040) 0.000 (0.001) 0.213*** (0.043) 0.062*** (0.005) 0.021*** (0.004) 0.024*** (0.003) 0.050*** (0.005) 0.017*** (0.002) 0.108*** (0.034) 0.267*** (0.031) 0.602*** (0.067) 0.189*** (0.031) 0.018 (0.017) 0.099*** (0.009) 0.233*** (0.044) 0.649*** (0.047) 3.430*** (0.110) 0.953*** (0.015) 0.937*** (0.012) 408

0.382*** (0.073) 0.194** (0.091) 0.002 (0.002) 0.110 (0.070) 0.091*** (0.011) 0.004 (0.007) 0.007 (0.006) 0.004 (0.008) 0.005 (0.003) 0.005 (0.054) 0.092* (0.054) 0.107 (0.089) 0.092* (0.052) 0.028 (0.028) 0.033** (0.014) 0.465*** (0.063) 0.494*** (0.066) 6.469*** (0.186) 0.364*** (0.046) 0.217** (0.094) 408

0.970*** (0.035) 1.212*** (0.028) 0.012*** (0.001) 0.340*** (0.025) 0.024*** (0.004) 0.001 (0.002) 0.045*** (0.002) 0.059*** (0.003) 0.035*** (0.001) 0.549*** (0.020) 0.049* (0.025) 0.137*** (0.024) 0.376*** (0.023) 0.087*** (0.009) 0.038*** (0.006) 1.535*** (0.036) 2.179*** (0.035) 5.686*** (0.070) 0.575*** (0.010) 1.652*** (0.008) 408

***, **, *Represent 1, 5, and 10 percent significance levels, respectively. Values in parentheses are standard errors. The first-stage results are not presented here for brevity.

indicate relative to non-innovations, both major and minor farmerled innovations do not lead to an improvement in dietary quality, and this is analogous to the findings based on aggregation of the innovations. Overall, these results suggest that the positive welfare effects from farmer-led innovations are largely driven by innovations that are slight modifications of existing practices, tools and techniques. One plausible explanation for this finding is that novel practices or substantial modifications of existing practices may not yield immediate rewards, while minor adjustments to existing and externally promoted techniques to make them fit into farming systems may generate higher benefits in the short run. 6. Conclusion We have analysed the effect of farmer-led innovations on household welfare, measured by household income, consumption expenditure and food security. With this, we contribute to the agricultural innovation literature since previous studies that look at the impact of agricultural innovations on household welfare have largely focused on externally promoted technologies. Using data from a field survey of rural farm households in northern Ghana and applying econometric techniques that control for selection bias, we

estimate the average treatment effects of farmer-led innovation on household well-being. The results show positive and statistically significant welfare effects of farmer-led innovations, confirming farmers’ perceptions as well as the numerous anecdotal reports of the significant role of farmer-led innovations in the livelihoods of rural farm households. First, we found that farmer-led innovations significantly improve household income per AE for innovators. Moreover, the innovations significantly increase household consumption expenditure per AE. Using both objective and subjective measures of food security, we also found that, on average, farmer-led innovations contribute significantly to the reduction of food insecurity. Specifically, they significantly increase household food consumption expenditure per AE, and contribute substantially to a reduction in the length of food shortages as well as a decrease in the severity of hunger among innovative households. Furthermore, we found that these effects are more pronounced for farm households whose innovative activities are minor modifications of existing techniques. Finally, we found that the positive contribution of farmer-led innovations to income do not significantly translate into nutritious diet, measured by household dietary diversity. Overall, our findings imply that farmer-led innovations have the potential of improving the livelihoods of rural households. Thus, it

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

273

study the long-term effects of farmer-led innovations. Nonetheless, our study has shown that rural poor farmers who are resourceconstrained go beyond adoption of externally promoted technologies and creatively implement location-specific innovations, which generate some positive welfare outcomes.

would be beneficial to support farmer-led innovation processes, which have often been neglected or undervalued. For instance, creating institutional arrangements that permit interactions and learning among agricultural stakeholders may play an essential role in stimulating farmers to innovate (Tambo and Wünscher, 2017). The significant contribution of farmer-led innovations to all the outcome indicators except dietary diversity suggests that further efforts are needed to ensure that the positive income effects translate into better nutrition for households in the study region. Thus, food security policies for the study region should go beyond food availability, and also focus on nutrition security. It is important to emphasize that our findings do not imply the promotion of farmer-led innovations at the neglect of modern agricultural technologies. Our results only strengthen arguments for better support for farmer-led innovations as a complement to externally promoted technologies in efforts to reduce poverty and attain food security. In this study, we either lumped the farmer-led innovations together or only disaggregated them into major and minor innovations based on their levels of originality, and this is due to limited samples. However, it will be interesting to assess how specific types of farmer-led innovations contribute to household well-being. Future research comprising large sample size will permit such analysis. Also, innovation is generally a dynamic process so further research involving panel data would be needed to

Acknowledgements Funding received from the German Federal Ministry of Education and Research (BMBF) (Grant no. 01LG1202A) through the West African Science Service Center for Climate Change and Adapted Land Use (WASCAL) research programme is gratefully acknowledged. Writing this article was also made possible by financial support of the German Federal Ministry for Economic Cooperation and Development (BMZ) under the Program of Accompanying Research for Agricultural Innovation (PARI). We also thank Joachim von Braun, Holger Seebens, Mekbib G. Haile and two anonymous reviewers, as well as seminar and conference participants at ZEF, IFPRI, 2014 AAEA Annual Meeting in Minneapolis, ICAE 2015 in Milan, and 2016 AAAE conference in Addis Ababa for comments on earlier drafts of this paper. Appendix

Table A.1 First stage results of the FIML ESR models

FFF distance Age Gender Household size Education Land holding Livestock holding Assets value Off-farm activity Road distance Group membership Climate shock Labour shock Risk preference KNW District KNE District Constant

(1)

(2)

(3)

(4)

(5)

(6)

0.052*** (0.015) 0.008 (0.005) 0.284 (0.204) 0.029 (0.029) 0.042** (0.019) 0.058** (0.025) 0.028 (0.024) 0.012 (0.010) 0.067 (0.167) 0.086 (0.082) 0.401*** (0.147) 0.403* (0.241) 0.150 (0.147) 0.101** (0.039) 0.213 (0.206) 0.405* (0.217) 0.504 (0.515)

0.056*** (0.012) 0.009* (0.005) 0.119 (0.200) 0.022 (0.029) 0.046** (0.019) 0.037* (0.022) 0.012 (0.023) 0.007 (0.010) 0.078 (0.163) 0.057 (0.084) 0.412*** (0.145) 0.208 (0.234) 0.090 (0.143) 0.095** (0.038) 0.212 (0.204) 0.503** (0.217) 0.279 (0.505)

0.059*** (0.013) 0.008 (0.005) 0.162 (0.199) 0.024 (0.028) 0.039** (0.019) 0.045** (0.023) 0.017 (0.023) 0.008 (0.010) 0.109 (0.164) 0.085 (0.082) 0.380*** (0.146) 0.305 (0.235) 0.143 (0.143) 0.096** (0.039) 0.199 (0.201) 0.414* (0.212) 0.368 (0.509)

0.057*** (0.013) 0.008 (0.005) 0.132 (0.201) 0.022 (0.029) 0.039** (0.019) 0.044** (0.021) 0.016 (0.023) 0.009 (0.010) 0.100 (0.164) 0.057 (0.081) 0.340*** (0.147) 0.305 (0.236) 0.114 (0.143) 0.098** (0.039) 0.209 (0.203) 0.057* (0.213) 0.285 (0.511)

0.051*** (0.015) 0.009* (0.005) 0.153 (0.200) 0.026 (0.029) 0.037** (0.018) 0.049** (0.023) 0.014 (-0.024) 0.006 (0.011) 0.096 (0.170) 0.069 (0.081) 0.414*** (0.145) 0.337 (0.235) 0.114 (0.142) 0.094** (0.040) 0.166 (0.212) 0.316 (0.259) 0.336 (0.502)

0.065*** (0.013) 0.007 (0.005) 0.139 (0.201) 0.025 (0.029) 0.042** (0.019) 0.046** (0.022) 0.015 (0.023) 0.009 (0.010) 0.109 (0.163) 0.097 (0.083) 0.406*** (0.147) 0.325 (0.234) 0.127 (0.142) 0.098** (0.039) 0.254 (0.204) 0.441** (0.215) 0.293 (0.511)

***, **, *Represent 1, 5, and 10 percent significance levels, respectively. Values in parentheses are standard errors. Note: Models 1 to 6 refer to first-stage estimates for household income, consumption expenditure, food gap, HHS, food consumption expenditure and HDDS, respectively.

274

J.A. Tambo, T. Wünscher / Journal of Rural Studies 55 (2017) 263e274

Table A.2 Falsification test

FFF distance Constant Wald X2/F-Stat No. of observations

(1)

(2)

(3)

(4)

(5)

(6)

0.008 (0.008) 5.336 (0.462) 11.05*** 241

0.003 (0.004) 6.809 (0.222) 12.08*** 241

0.001 (0.004) 1.649 (0.282) 63.18*** 241

0.003 (0.010) 1.553 (0.528) 30.64** 241

0.005 (0.004) 6.360 (0.229) 11.14*** 241

0.001 (0.003) 1.639 (0.192) 44.83*** 241

***, **, *Represent 1, 5, and 10 percent significance levels, respectively. Values in parentheses are standard errors. Note: Models 1 to 6 refer to household income, consumption expenditure, food gap, HHS, food consumption expenditure and HDDS, respectively. Models 1, 2 and 5: Ordinary Least Squares. Model 3 and 6: Poisson Regression. Model 4: Negative Binomial Regression. We control for other variables but only report parameters for the variables of interest.

References Apusigah, A.A., 2009. The gendered politics of farm household production and shaping of women’s livelihoods in northern Ghana. Fem. Afr. 12, 51e68. Asfaw, S., Shiferaw, B., Simtowe, F., Leslie, L., 2012. Impact of modern agricultural technologies on smallholder welfare: evidence from Tanzania and Ethiopia. Food Policy 37 (3), 283e295. Ballard, T., Coates, J., Swindale, A., Deitchler, M., 2011. Household Hunger Scale: Indicator Definition and Measurement Guide. Washington, DC: FANTA-2 Bridge, FHI 360. Barrett, C., Moser, C., McHugh, O., Barison, J., 2004. Better technology, better plots, or better Farmers? Identifying changes in productivity and risk among malagasy rice farmers. Am. J. Agric. Econ. 86 (4), 869e888. Barrett, C.B., 2010. Measuring food insecurity. Science 327 (5967), 825e828. Bilinsky, P., Swindale, A., 2005. Months of Inadequate Household Food Provisioning (MIHFP) for Measurement of Household Food Access: Indicator Guide. Food and Nutrition Technical Assistance Project, Academy for Educational Development, Washington, DC. Brooks, S., Loevinsohn, M., 2011. Shaping agricultural innovation systems responsive to food insecurity and climate change. Nat. Resour. Forum 35 (3), 185e200. de Haen, H., Klasen, A., Qaim, M., 2011. What do we really know? Metrics for food insecurity and undernutrition. Food Policy 36 (6), 760e769. de Janvry, A., Sadoulet, E., 2002. World poverty and the role of agricultural technology: direct and indirect effects. J. Dev. Stud. 38 (4), 1e26. Deaton, A., 1997. The Analysis of Household Surveys: a Microeconometric Approach to Development Policy. Johns Hopkins University Press for the World Bank, Baltimore. Deb, P., 2009. MTREATREG: Stata Module to Fits Models with Multinomial Treatments and Continuous, Count and Binary Outcomes Using Maximum Simulated Likelihood. Statistical Software Components. Deb, P., Trivedi, P.K., 2006. Specification and simulated likelihood estimation of a non-normal treatment-outcome model with selection: application to health care utilization. Econ. J. 9 (2), 307e331. Di Falco, S., Veronesi, M., Yesuf, M., 2011. Does adaptation to climate change provide food security? A micro-perspective from Ethiopia. Am. J. Agric. Econ. 93 (3), 829e846. FAO, IFAD, WFP, 2013. The State of Food Insecurity in the World 2013. The Multiple Dimensions of Food Security. FAO, Rome. Fernandez-Cornejo, J., Hendricks, C., Mishra, A., 2005. Technology adoption and offfarm household income: the case of herbicide-tolerant soybeans. J. Agric. Appl. Econ. 37 (3), 549e563. Fuglie, K.O., Bosch, D.J., 1995. Economic and environmental implications of soil nitrogen testing: a switching regression analysis. Am. J. Agric. Econ. 77 (4), 891e900. Gatzweiler, F.W., Von Braun, J. (Eds.), 2016. Technological and Institutional Innovations for Marginalized Smallholders in Agricultural Development. Springer. Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., Toulmin, C., 2010. Food security: the challenge of feeding 9 billion people. Science 327 (5967), 812e818. Halvorsen, R., Palmquist, R., 1980. The interpretation of dummy variables in semilogarithmic equations. Am. Econ. Rev. 70 (3), 474e475. € m, E.E., 2008. Risk aversion in the laboratory. Res. Exp. Econ. Harrison, G.W., Rutstro 12, 41e196. Kassie, M., Ndiritu, W.S., Stage, J., 2014. What determines gender inequality in household food security in Kenya? Application of exogenous switching treatment regression. World Dev. 56, 153e171. Kassie, M., Shiferaw, B., Geoffrey, M., 2011. Agricultural technology, crop income,

and poverty alleviation in Uganda. World Dev. 39 (10), 1784e1795. Kijima, Y., Otsuka, K., Sserunkuuma, D., 2008. Assessing the impact of NERICA on income and poverty in central and western Uganda. Agric. Econ. 38 (3), 327e337. Kummer, S., Leitgeb, F., Vogl, C.R., 2017. Farmers’ own research: organic farmers’ experiments in Austria and implications for agricultural innovation systems. Sustain. Agric. Res. 6 (1), 103e119. Kummer, S., Milestad, R., Leitgeb, F., Vogl, C.R., 2012. Building resilience through farmers’ experiments in organic agriculture: examples from eastern Austria. Sustain. Agric. Res. 1 (2), 308e321. Leitgeb, F., Kummer, F., Funes-Monzote, F.R., Vogl, C.R., 2013. Farmers’ experiments in Cuba. Renew. Agric. Food Syst. 29 (1), 48e64. Lokshin, M., Sajaia, Z., 2004. Maximum likelihood estimation of endogenous switching regression models. Stata J. 4 (3), 282e289. Maddala, G.S., 1983. Limited-dependent and Qualitative Variables in Econometrics. Cambridge University Press, Cambridge. McIntyre, B.D., Herren, H.R., Wakhungu, J., Watson, R.T., 2009. Agriculture at a Crossroads e Global Report. International assessment of agricultural knowledge, science and technology for development (IAASTD), Washington, DC. Minten, B., Barrett, C.B., 2008. Agricultural technology, productivity, and poverty in Madagascar. World Dev. 36 (5), 797e822. OANDA, OANDA.com currency converter, 2017. (Accessed 3 July 2017). Pinstrup-Andersen, P., 2009. Food security: definition and measurement. Food Secur. 1 (1), 5e7. Quaye, W., 2008. Food security situation in northern Ghana, coping strategies and related constraints. Afr. J. Agric. Res. 3 (5), 334e342. Reij, C., Tappan, G., Smale, M., 2009. Re-greening the sahel: farmer-led innovation in Burkina Faso and Niger. In: Spielman, D.J., Pandya-Lorch, R. (Eds.), Millions Fed: Proven Successes in Agricultural Development. International Food Policy Research Institute, Washington, DC, pp. 53e58. Reij, C., Waters-Bayer, A. (Eds.), 2001. Farmer Innovation in Africa: a Source of Inspiration for Agricultural Development. Earthscan, London. Sadoulet, E., de Janvry, A., 1995. Quantitative Development Policy Analysis. John Hopkins University Press, Baltimore. Sanginga, P., Waters Bayer, A., Kaaria, S., Njuki, J., Wettasinha, C. (Eds.), 2009. Innovation Africa: Enriching Farmers' Livelihoods. Earthscan, London. Singh, I., Squire, L., Strauss, J. (Eds.), 1986. Agricultural Household Models: Extensions, Applications and Policy. The John Hopkins University Press, Baltimore. Swindale, A., Bilinsky, P., 2006. Household Dietary Diversity Score (HDDS) for Measurement of Household Food Access: Indicator Guide (v.2). Food and Nutrition Technical Assistance Project, Academy for Educational Development, Washington, DC. Tambo, J.A., Wünscher, T., 2015. Identification and prioritization of farmers’ innovations in northern Ghana. Renew. Agric. Food Syst. 30 (6), 537e549. Tambo, J.A., Wünscher, T., 2016. Building farmers' capacity for innovation generation: insights from rural Ghana. Renew. Agric. Food Syst. http://dx.doi.org/ 10.1017/S1742170516000521. Tambo, J.A., Wünscher, T., 2017. Enhancing resilience to climate shocks through farmer innovation: evidence from northern Ghana. Reg. Environ. Change 17 (5), 1505e1514. The Worldwatch Institute, 2011. State of the World 2011: Innovations that Nourish the Planet. The Worldwatch Institute, Washington, DC. Waters-Bayer, A., van Veldhuizen, L., Wongtschowski, M., Wettasinha, C., 2009. Recognizing and enhancing processes of local innovation. In: Sanginga, P., Waters-Bayer, A., Kaaria, S., Njuki, J., Wettasinha, C. (Eds.), Innovation Africa: Enriching Farmers’ Livelihoods. Earthscan, London, pp. 239e254. Weersink, A., Nicholson, C., Weerhewa, J., 1998. Multiple job holdings among dairy families in New York and Ontario. Agric. Econ. 18 (2), 127e143.