Technological priorities in rice production among smallholder farmers in Ghana

Highlights

•	Rice farmers were more likely to adopt fertilizer only after adopting the improved rice variety but not vice versa.
•	Adopters of the improved rice varieties were more likely to use inputs separately or simultaneously adopt all technologies.
•	Education and complementary input availability consistently predicted the adoption of five technology bundles.
•	Extension contacts and farm size consistently and significantly predicted the adoption of all technology bundles.
•	Complementary input availability and participation in field demonstrations were the strongest predictors of adoption.

Abstract

Technological innovations in agriculture have transformed the farming systems of smallholder farmers, leading to the realization of economic incentives of higher outputs, profits, and sustainability. However, attempts to model their adoption behaviors have failed to consider the possible interdependencies and simultaneities in the adoption decision process. This may cause invalid inferences and incorrect conclusions to be made regarding smallholder farmers’ adoption decisions. This study, therefore, models smallholder farmers’ adoption decisions while taking into account the possibility of interdependencies and simultaneities of adopting rice productivity-enhancing agricultural technologies in Ghana. Cross-sectional survey data on 1016 smallholder rice farmers sampled using a three-stage sampling method from four municipalities in two regions in Ghana were used for the study. We fit a multinomial probit model to analyze the decisions to adopt or not to adopt varied combinations of improved rice variety, fertilizer, and herbicide. Our study finds that when faced with improved rice variety-fertilizer-herbicide technological options, smallholder rice farmers’ adoption decisions fell into any one of four possibilities: (a) reject all technological options, (b) adopt a particular technology only, (c) adopt two of the technologies, and (d) adopt all technological choices. We show that such adoption decisions were consistently and most significantly predicted by farm size, extension contacts, education, participation in field demonstrations, complementary input availability, and extent of commercialization. Modeling smallholder farmers’ adoption decisions to capture interdependencies and simultaneities provides useful information for integrating diverse perspectives into the design and promotion of technological innovations based on their nature of adoption, the specific technological innovations, and influencing factors. We recommend that to achieve the maximum benefits, future technology promotion activities should concurrently reduce farmers’ subjective uncertainties and risks, promote credit access and subsidies, as well as ensure effective input delivery systems while targeting specific technological innovations.

Keywords:

• Multinomial probit model
• Technology adoption
• Productivity-enhancing technologies
• Simultaneity
• Interdependencies

1 Introduction

Globally, increasing population growth, urbanization, changing consumer demands and climate change have resulted in dramatic changes in farming practices. Consequently, policy makers are confronted with the challenge of tackling agricultural productivity and food security challenges as well as alleviating poverty among smallholder farmers. Advancement in agricultural technological innovations has generated renewed interest, leading to policy directives for addressing these global challenges to increase productivity and sustainability in the sector. Recent years continue to witness greater investment in productivity-enhancing technologies for smallholder farmers in developing countries. Therefore, policy makers and academicians have intensified their interest in understanding the fundamental and complex decisions that smallholder farmers have to make in adopting such technologies.

In Ghana, agricultural modernization remains the central focus of agricultural policy and the strategy for private sector development. This policy directive emphasizes the role of the government and the private sector in transforming agriculture from a low-productivity subsistence-based sector to one characterized by high-productivity (Ministry of Food and Agriculture [MoFA], 2016) and market-orientation. However, the present political and social climate facing smallholder farmers restrain them from adopting high productivity and sustainable technologies in most aspects of farming practices. In response, governments, donors, and agricultural researchers continue to develop and promote new production techniques, which promote productive and sustainable agricultural systems with less environmental damage. Such policy directions are more pronounced in the production of cereals (especially, maize and rice) because they are the major staples for millions of people around the world and in Ghana, in particular (Food and Agriculture Organization, 2006; MoFA, 2012).

Smallholder farmers’ adoption of a combination of agricultural technologies presents a paradigm shift from low-productive farming methods towards high-productivity techniques. Accordingly, smallholder farmers may benefit from less chemical leaching to groundwater, less soil erosion, and lower water requirements (Dorfman, 1996) as well as higher yields with lower input costs [from more efficient use of chemicals and fertilizers] and household food security and income (Dixon et al., 2001; Wanyama et al., 2010). Notwithstanding their possible benefits, existing studies have shown low rates of adoption of high-productivity and sustainable agricultural practices in developing countries (Jansen et al., 2006; Kassie et al., 2009; Wollni et al., 2010), leading to soil degradation and nutrient depletion as major constraints to productivity and sustainability gains in the sector.

Productivity-enhancing technologies are increasingly being promoted as a bundle of technologies to be adopted as one potential bundle or subset of bundles that smallholder farmers could choose from some larger set of technologies. For instance, the Government of Ghana has recently promoted an improved rice variety – NERICA – as a complementary input with established optimum fertilizer requirement levels, weed management regime, and planting density for farmers to adopt. However, smallholder farmers may face several distinct technological options regarding the adoption of agricultural technologies that are introduced in the package. Adoption decisions may be to adopt either the complete package or the individual subsets of the package resulting in simultaneity in the adoption decisions. Hence, such adoption processes may follow specific (and predictable) sequential patterns (Dorfman, 1996; Feder et al., 1985). Often, smallholder farmers adopt multiple agricultural technologies, rather than just a single technology, to manage biological and production risks. Considering the fact that such adoption choices inherently involve multivariate decisions, univariate models exclude useful and connected economic information related to the interdependent and simultaneous adoption decisions (Dorfman, 1996; Feder et al., 1985).

Several farm, farmer, and socio-economic characteristics have featured extensively in both the theoretical and empirical literature in explaining farmers’ adoption decisions. Farmer’s age, farm size, education, on-farm/off-farm income, extension contact, and access to credit have been found to be significantly related to the adoption of agricultural technologies (Dhital and Joshi, 2016; Ghimire et al., 2015; Mittal and Mehar, 2016; Simtowe et al., 2016). However, most existing studies have limited the measurement of smallholder farmers’ adoption choice to a single technology, such as an improved variety or fertilizer (Ghimire et al., 2015; Simtowe et al., 2016) while only a few studies have focused on two or more technologies (Abay et al., 2016; Mittal and Mehar, 2016; Yirga et al., 2015). Also, prior studies have failed to consider the possible interdependencies and simultaneities in smallholder farmers’ adoption decision process.

Ignoring the possibility of interdependencies and simultaneities in the adoption of productivity-enhancing technologies may cause invalid inferences and incorrect conclusions to be made regarding smallholder farmers’ adoption decisions (Dorfman, 1996; Feder et al., 1985). As a result, policy-makers may have limited or incorrect understanding of the prospects for sustainable production technologies and integrating diverse perspectives into the development, dissemination, and promotion of future agricultural technologies. This article, therefore, models smallholder farmers’ adoption decisions while taking into account the possibility of interdependencies and simultaneities of adopting rice productivity-enhancing agricultural technologies in Ghana. In particular, the article examines the (a) nature of simultaneities in the adoption of rice productivity-enhancing technologies, and (b) factors that influence smallholder farmers’ adoption decisions regarding multiple productivity-enhancing technologies.

The remainder of this article proceeds as follows. The conceptual framework is presented in the next section. This is followed by the methodology. The empirical results are then presented and discussed. Policy implications and conclusions are presented in the final section.

2 Conceptual frameworks

This study modeled smallholder farmers’ choice of productivity-enhancing technologies based on expected utility theory (Coble et al., 2000; Ke and Wang, 2002; Sherrick et al., 2004). Expected utility theory assumes that smallholder farmers assess their own specific risk environments (i.e., production risk) and choose between technological bundles by comparing their expected utility values. That is, smallholder farmers adopt a new technology or bundle of productivity-enhancing technologies only when the expected utility of using such a technology is significantly greater than the traditional method subject to some constraints (Feder et al., 1985; Rogers, 2003).

Consider a set of bundles of rice productivity-enhancing technologies. Each of the technology bundles is a possible adoption decision by the smallholder farmer. He/she would choose the bundle that maximizes his/her expected utility. That is, the smallholder farmer makes the decision of whether or not to adopt any, some, or all of the productivity-enhancing technologies, j, available to him/her ($\text{j}=0,1,\dots ,s$). The smallholder farmer assesses each of these j productivity-enhancing technologies in view of its effect on the returns distribution to a set of assets, $\mathrm{A}$, used in production. Denote this expected utility conditional on adoption of a particular productivity-enhancing technology bundle by U(T_i, z) where $T_i$; denotes the i^th productivity-enhancing technology bundle and z is a vector of individual and environmental characteristics including education, extension contact, credit use, etc. Implicit in the function U(T_i, z) are the profit, cost, and risk impacts of the adoption decisions. The farmer’s problem is then a simple one:$\begin{array}{c}\mathrm{C}\mathrm{h}\mathrm{o}\mathrm{o}\mathrm{s}\mathrm{e}{T}_i:{\text{T}}_i=\arg \max [U(T,z)],\\ i=0,1,\mathrm{...},\mathrm{s}.\end{array}$

Conventionally, productivity-enhancing technologies bundle zero represents the adoption of none of the productivity-enhancing technologies being studied; the s^th bundle represents the adoption of all possible technologies. According to Feder et al. (1985), agricultural technologies that are introduced in packages involve simultaneous decision-making about either the whole package or subsets of the package. Modeling such adoption decisions provides a better understanding of the interdependencies among adoption decisions and, thus, helps determine appropriate specifications for simultaneous adoption models. In spite of the considerable array of models that are used to predict the adoption decision-making of farmers, the multinomial probit regression model – an extension of the standard probit model – remains the most appropriate method for modeling such a multivariate decision (Dorfman, 1996; Feder et al., 1985).

The standard probit model for a single-choice adoption decision has a response variable, Y, which is binary with only two possible outcomes, denoted as 1 and 0. The model can be represented by two equations. First, a latent (unobservable) variable is described by a linear function of a set of regressors and a normally distributed stochastic error,(2) $U=z\beta +\epsilon$(2) where U represents the latent variable, z is a vector of the regressors that influence the level of the latent variable through the coefficient vectorβ, and ε is the stochastic term. The second equation describes the observable choice of the decision maker, (3) y = 1, if U > 0 ; y = 0, otherwise, where U is the level of expected utility.

The multinomial probit model is a generalization of the probit model and assumes that the stochastic error terms follow a multivariate normal distribution that is correlated across choices. The model extends the latent variable equation in (1) to a system of equations with latent dependent variables. Consider a set of Junordered alternatives that are modeled by a regression of both case-specific (i.e., the information on one decision maker) and alternative-specific covariates. Underlying the model is the set of J latent variables (utilities) represented by (Powers and Xie, 2000):(4) $U_{ij}=\beta x_{ij}+{\alpha }_j{z}_i+{\epsilon }_{ij}$(4) Where i denotes cases (i^th smallholder farmer) and j denotes alternatives of productivity-enhancing technology bundles. Further, x_ij is a 1 × p vector of alternative-specific variables, β is ap × 1 vector of parameters, z_i is a 1 × q vector of case-specific variables, α_j is a q × 1 vector of parameters for the j^th alternative, and ε_i = (ε_i1, ..., ε_ij) is distributed multivariate normal with mean zero and covariance matrix Ω. The stochastic terms are assumed to be jointly distributed multivariate normal random variables. The unknown parameters in Eq. (4(4) $U_{ij}=\beta x_{ij}+{\alpha }_j{z}_i+{\epsilon }_{ij}$(4) ) are estimated using a maximum likelihood procedure. The decision-maker selects the alternative for which the latent variable is highest. The observed dichotomous choice variables are:(5) $\begin{array}{c}{y}_i=1,\text{if}{\text{U}}_i=\max [{\text{U}}_j,\text{j}=0,1,\dots ,s];\\ \hfill y_i=0,otherwise\end{array}\mathrm{.}$(5)

A typical representation of the probability that individual i selects alternative jis given by:(6) $P_{ij}=\mathrm{Pr}ob[U_{ij}>U_{ik}forallj\ne k]$(6)

The multinomial probit (MNP) model permits the analysis of decisions across more than two categories, allowing the determination of choice probabilities for different categories (Greene, 2003; Wooldridge, 2002). Also, the MNP model does not impose the independence of irrelevant alternative assumption underlying multinomial logistic and conditional logistic models because it allows for a general covariance structure for stochastic error terms (Dow and Endersby, 2004). This implies that the MNP model permits the probability of choosing one alternative over another to depend on the remaining alternatives. The multinomial probit has been used extensively in literature for similar studies (Abay et al., 2016; Dorfman, 1996; Velandia et al., 2009).

3 Methods

3.1 The study areas

The study was confined to four municipalities from two regions in Ghana. The municipalities were the Ejisu-Juaben and Ejura-Sekyedumasi Municipalities of the Ashanti Region, and the Atebubu-Amantin Municipality and Pru District in the Brong-Ahafo Region. These municipalities are part of the 27 districts in the respective regions. The Ejisu-Juaben and Ejura-Sekyedumasi Municipalities exhibit derived deciduous forest vegetation and savannah vegetation. These municipalities produce major crops such as maize, yam, beans, rice, plantain, cassava and groundnuts (Ghana Statistical Service, 2013). In the Brong-Ahafo Region, the two municipalities are located within the transitional zone between the wet semi-equatorial and tropical savannah climate regions. Crops such as maize, cassava, yam and rice are grown in commercial quantities in these two municipalities (Ghana Statistical Service, 2014). The four municipalities have an average household size of five persons per household. Over 60% of the active labor force in the municipalities are engaged in agriculture and related activities. Farming activities are all-year round due to the two rainfall seasons. Farming is predominantly small-scale with average cultivated land ranging between 1.6 and 2.4 ha for all crops (Ghana Statistical Service, 2013, 2014).

3.2 Data and sampling procedure

The data used in this study originated from a 2013 cross-sectional in-person survey of smallholder rice farmers in the four municipalities in the Ashanti and Brong-Ahafo Regions. The sample for the study included selected smallholder rice farmers in the four municipalities. A multistage (i.e., a three-stage) sampling procedure using simple random sampling techniques was employed in selecting three operational areas and six communities from each municipality as well as 1016 smallholder rice farmers from the four municipalities. This method was utilized to ensure a high degree of representativeness by providing the elements with equal chances of being selected (Babbie, 2007). The lottery sampling method was employed and involved the selection of the sample at random from the sampling frame through the use of random numbers (Saunders et al., 2003). A structured questionnaire was used to collect data on variables such as farm, farmer, and socio-economic characteristics. Data were collected on the adoption of biochemical (improved rice varieties, fertilizer and herbicide), land tenure, education, rice farming experience, extension contacts, participation in field demonstrations organized by extension agents, access to credit, membership of an association, household size, availability of complementary inputs and extent of commercialization affecting technology adoption. Trained agricultural extension agents facilitated the task of data collection.

3.3 Nature of simultaneity in adoption

Following Nerlove and Press’ (1973) contingency tables and odd ratios approach, we evaluate the propensity of smallholder rice farmers to adopt the full package of technologies (consisting of the improved rice variety, fertilizer, and herbicide) or to adopt certain components selectively. Each technology was treated dichotomously. The odds ratio is a bivariate measure of association. Higher odds ratios are associated with more likelihood of adopting the technologies involved. Extending the Nerlove and Press (1973) approach to a 3*3 contingency table, we simultaneously treated two technologies – fertilizer and herbicide – to examine the impact of the adoption of the improved rice variety on the adoption of these technologies. Pearson Chi-Square statistics were computed to establish whether a statistically significant relationship exists among any two or the three technologies.

3.4 The MNP model

In the MNP model, the adoption of the productivity-enhancing technologies was treated as a discrete choice decision with the choice variable being qualitative in nature. Since the decision-making involves alternatives that are interrelated (and in this case, complementary), all possible combinations of the productivity-enhancing technology bundles should be included in the decision set for the rice farmer. In this study, the adoption decision relates to an improved rice variety, fertilizer, and herbicide. These productivity-enhancing technologies were chosen because they are the most frequently promoted and adopted complementary technologies by smallholder rice farmers. Also, they are important technologies for Ghana’s agriculture as they form key aspects of the government’s policy directives for boosting productivity in the rice sector. The technology adoption variable is modeled as a discrete decision variable consisting of various possible mutually exclusive combinations (2³) of the productivity-enhancing technologies. Accordingly, eight technology bundles constituted the sample space facing the smallholder rice farmer (see below); rather than three distinct dichotomous decisions (improved variety: yes/no; fertilizer use: yes/no; herbicide use: yes/no). This should lead to a better understanding of the characteristics associated with the adoption decision and improve the ability to forecast adoption of a particular technology bundle (Dorfman, 1996). The eight combinations of rice productivity-enhancing technologies serve as the dependent variable in the MNP model.

The MNP model for this study is, thus, specified as:(7) $\begin{array}{l}{Y}_{ij}={\beta }_0+{\beta }_{1}te{n}_i+{\beta }_{2}ed{u}_i+{\beta }_3{\exp }_i+{\beta }_{4}ex{t}_i+{\beta }_{5}dem{o}_i\\ +{\beta }_{6}ass{o}_i+{\beta }_{7}cre{d}_i+{\beta }_{8}fsiz{e}_i+{\beta }_{9}hsiz{e}_i+{\beta }_{10}inava{l}_i\\ +{\beta }_{11}com{m}_i+{\beta }_{12}ES{M}_i+{\beta }_{13}AA{M}_i+{\beta }_{14}EJ{M}_i+{\epsilon }_i\end{array}$(7) Where Y_ij(j = 1, …, m) represent the technology bundle or choice (in this study, m = 8). The technology bundles are 1 if none of the technology bundle is adopted, 2 if the farmer uses improved rice variety only, 3 if the farmer uses fertilizer only, 4 if the farmer uses herbicide only, 5if the farmer uses improved rice variety and fertilizer, 6 if the farmer uses improved rice variety and herbicide, 7 if the farmer uses fertilizer and herbicide, and 8 if the farmer uses all the technology bundles (i.e., improved rice variety, fertilizer, and herbicide) faced by the i^th farmer (i = 1, …, m); ten_i denotes tenure (1 if the farmer owns farmland and 0 otherwise [+]); edu_i is the years of education of the farmer [+]); exp_i is the years of rice farming experience of the farmer [+]; ext_i represents number of extension contacts [+]; demo_i indicates farmer participation in field demonstrations organized by extension agents or services (1 if the farmer participated and 0 otherwise [+]); asso_idenotes membership of farmer-based association (1 if the farmer is a member and 0 otherwise [+]); cred_i is access to credit (1 if the farmer had access to credit and 0 otherwise [+]); fsize_i denotes farm size in hectares [+]; hsize_iindicates household size [+]; inaval_i represents the availability of complementary inputs (if the farmer considers complementary inputs to be available and 0 otherwise [+]);comm_i indicates the extent of commercialization (i.e., the fraction of output the farmer sold [+]); ESM_i is location dummy for Ejura-Sekyedumasi Municipality (1 if the farmer lives in this Municipality and 0 otherwise [±]); AAM_i is location dummy for Atebubu-Amantin Municipality (1 if the farmer lives in this Municipality and 0 otherwise [±]); EJM_i is location dummy for Ejisu-Juaben Municipality (1 if the farmer lives in this Municipality and 0 otherwise [±]); and ε_i denotes the unobserved error term. The fourth location dummy – Pru District – was excluded in the model so the coefficient estimates of the other three locations are relative to Pru District.

The choice of the explanatory variables was based on findings from economic theory and prior empirical studies on agricultural technological adoption decisions (Feder et al., 1985; Mittal and Mehar, 2016; Rogers, 2003; Simtowe et al., 2016). To confirm the basis for selecting these explanatory variables, preliminary diagnostics involving binary probit models were performed on the three technological options (improved rice varieties, fertilizer, and herbicide) with varied model specifications. It was observed that the location dummies measuring the municipal effects were not statistically significant in explaining the adoption decisions regarding the three technologies. Moreover, the MNP model did not converge with the locational dummies included. Consequently, the location dummy variables were excluded from the MNP model.

4 Results and discussion

4.1 Descriptive results

Table 1 presents the means and standard deviations of the variables used in the MNP model. The results showed that of those participating in the study, only 31% owned their farmland. This was due to the large proportion of migrant farmers in the study areas. Most of the rice farmers had obtained about six years of formal education and had been rice farmers for about 10 years. Also, most of the rice farmers had contact with the extension agents at least once during the farming season. Similarly, 48% of the rice farmers participated in field demonstrations organized by extension services or agents. Providing extension contacts and field demonstrations for smallholder rice farmers would enable them to acquire needed information regarding improved technologies and expose them to the practice of using those technologies. Ultimately, this may help shape their subjective probabilities and risk aversion towards technological innovations. It is also evident that only 16% of the rice farmers had access to credit. This finding showed that access to credit was still a challenge for smallholder rice farmers and suggests the need to remove all impediments to access to credit and provide specialized credit facilities to rice farmers to undertake greater investments in technological innovations. The low access to credit among the rice farmers is consistent with the findings of Dittoh (2006) and Tsinigo et al. (2016) that access to credit was a major constraint facing smallholder farmers. Further, it can be seen from the survey that only 26% of the rice farmers reported that complementary inputs were available. The reported low availability of complementary inputs clearly suggests that smallholder rice farmers face supply constraints and explains the need for effective input integration efforts with the promotion of technology adoption.

Table 1 Means and standard deviations of the variables used in the MNP model.

Indicators	Mean	Bundle 1	Bundle 2	Bundle 3	Bundle 4	Bundle 5	Bundle 6	Bundle 7	Bundle 8
Tenure	0.31	0.31	0.31	0.31	0.31	0.31	0.30	0.31	0.31
	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)	(0.46)
Education	5.60	5.60	5.60	5.60	5.60	5.60	6.08	5.60	5.60
	(4.78)	(4.78)	(4.78)	(4.78)	(4.78)	(4.78)	(4.57)	(4.78)	(4.78)
Experience	10.29	10.29	10.29	10.29	10.29	10.29	9.51	10.29	10.29
	(9.36)	(9.36)	(9.36)	(9.36)	(9.36)	(9.36)	(8.55)	(9.36)	(9.36)
Extension	1.27	1.27	1.27	1.27	1.27	1.27	1.22	1.27	1.27
	(1.91)	(1.91)	(0.09)	(1.91)	(1.91)	(1.91)	(1.32)	(1.91)	(1.91)
Demonstration	0.48	0.48	0.48	0.48	0.48	0.48	0.61	0.48	0.48
	(0.50)	(0.50)	(0.23)	(0.50)	(0.50)	(0.50)	(0.49)	(0.50)	(0.50)
Association	0.52	0.52	0.52	0.52	0.52	0.52	0.52	0.52	0.52
	(0.50)	(0.50)	(0.15)	(0.50)	(0.50)	(0.50)	(0.50)	(0.50)	(0.50)
Credit	0.16	0.16	0.16	0.16	0.16	0.16	0.15	0.16	0.16
	(0.37)	(0.37)	(0.22)	(0.37)	(0.37)	(0.37)	(0.36)	(0.37)	(0.37)
Farm size	1.26	1.26	1.26	1.26	1.26	1.26	1.55	1.26	1.26
	(0.83)	(0.83)	(0.09)	(0.83)	(0.83)	(0.83)	(0.95)	(0.83)	(0.83)
Household size	6.66	6.66	6.66	6.66	6.66	6.66	6.73	6.66	6.66
	(3.38)	(3.38)	(0.02)	(3.38)	(3.38)	(3.38)	(3.77)	(3.38)	(3.38)
Input availability	0.26	0.26	0.26	0.26	0.26	0.26	0.06	0.26	0.26
	(0.44)	(0.44)	(0.19)	(0.44)	(0.44)	(0.44)	(0.24)	(0.44)	(0.44)
Commercialization	74.14	74.14	74.14	74.14	74.14	74.14	72.78	74.14	74.14
	(14.26)	(14.26)	(0.01)	(14.26)	(14.26)	(14.26)	(14.15)	(14.26)	(14.26)
Observations	1016	100	232	69	134	81	236	79	86

Note: Figures in parentheses are standard deviations.

The results further indicated that the average family size among the sampled rice farmers was seven people, presenting sufficient scope to rely on family labor in the adoption of productivity-enhancing technologies. Some improved technologies, such as improved rice varieties, require more labor inputs, and hence, the availability of family labor will encourage the adoption of such technologies. Moreover, average farm size was one hectare; indicating that the rice farmers were largely smallholders. The findings are consistent with MoFA’s (2011) reported average farm holdings of farmers in Ghana. The small farm holdings among the rice farmers could be due to the demand for land rental due to the large number of migrant farmers in the study areas and the increasing demand for resources for the cultivation of larger farms. On average, most of the rice farmers sold 74% of their harvested rice output, indicating that the rice farmers were predominantly commercial-oriented in their farming activities.

4.2 Adoption of technological choices among the rice farmers

Table 2 presents the probability distribution of joint and marginal adoption decisions of the technology bundles in rice production. The results showed that rice farmers simultaneously adopting improved rice variety and herbicide constituted 23% of the total, while those adopting only improved rice accounted for about 23%. The joint probability of using the improved rice variety, fertilizer, and herbicide was 8%. The MNP model predicted 31% for the adoption of the improved rice variety and herbicide. This finding is surprising given the many opportunities offered to smallholder rice farmers to use recommended production technologies in rice farming. It may also be worrying considering that government’s efforts to increase contact with extension services and agricultural information systems are central to rice productivity goals.

Table 2 Joint and marginal probabilities of adoption of technology bundles.

Technology bundle	Joint distributions			Marginal distributions
	Observed	Predicted	Difference	Rice	Fertilizer	Herbicide
(1) None	9.84	3.56	−6.28	9.84	0.00	0.00
(2) Improved rice varieties only	22.83	22.48	−0.35	22.83	0.00	0.00
(3) Fertilizer only	6.79	2.87	−3.92	0.00	6.79	0.00
(4) Herbicide only	13.19	16.58	3.39	0.00	0.00	13.19
(5) Improved rice and fertilizer	7.97	9.82	1.85	7.97	7.97	0.00
(6) Improved rice and herbicide	23.23	31.31	8.08	23.23	0.00	23.23
(7) Fertilizer and herbicide	7.78	9.44	1.66	0.00	7.78	7.78
(8) Improved rice, fertilizer & herbicide	8.37	3.94	−4.43	8.37	8.37	8.37
Total	100.00	100.00	0.00	72.24	30.91	52.57

4.3 Simultaneity of adoption of technological choices among the rice farmers

Table 3 presents the simultaneity in the adoption of the technologies in rice production. The results showed that there was a close positive association between fertilizer and the use of the improved rice variety, fertilizer and herbicide as well as between fertilizer and herbicide use. For instance, comparing the decision to adopt the improved rice variety with fertilizer indicated that 61% of the rice farmers used neither the improved rice variety nor fertilizer, while 26% used both technologies. For the adoption of fertilizer, the odds ratio for the use of the improved rice variety was 1.26, for the adoption of herbicide it was 1.01. Also for the adoption of herbicide, the odds ratio for the adoption of the improved rice variety was 1.08. The adoption of fertilizer and the improved rice variety had the highest odds ratio of 1.25. This means that the odds of adopting the improved rice variety were 25% higher for those who adopted fertilizer compared to those who did not adopt fertilizer.

Table 3 Simultaneity of adoption among rice farmers – a 2*2 contingency table.

Technology type	Choice	Improved rice variety		Herbicide
		Not used	Used	Not used	Used
Fertilizer	Not used	234(61.26)	468(73.82)	332 (68.88)	370 (69.29)
	Used	148(38.74)	166(26.18)	150 (31.12)	164(30.71)
	Odds ratio	1.26		1.01
	χ^2(1)	17.61^***		0.02
Herbicide	Not used	169(44.24)	313(49.37)	–	–
	Used	213(55.76)	321(50.63)	–	–
	Odds ratio	1.08		–
	χ^2(1)	2.51		–

Note: Figures in parentheses are percentages.

*** p < 0.01.

The Pearson Chi-Square statistics, χ²(1) = 17.61, p < 0.01, suggest that there was a statistically significant association between the adoption of improved rice varieties and fertilizer. That is rice farmers who adopted improved rice varieties were more likely to adopt fertilizer. It is obvious that smallholder rice farmers are more likely to adopt fertilizer only after adopting the improved rice variety but not necessarily vice versa. It is also evident with the odds ratio that rice farmers were more likely to adopt the improved rice and fertilizer compared to improved rice variety and herbicide and than fertilizer and herbicide. This indicated that fertilizer and herbicide were only adopted by rice farmers if they had adopted the improved rice varieties. These findings provide some insight into a dependency relationship among the improved rice variety, fertilizer, and herbicide.

Fertilizer and herbicides are key inputs in the production of improved rice varieties. These inputs are often considered as complementary to the adoption of improved rice varieties. But do the adoption of fertilizer and the adoption of herbicide depend on the adoption of improved rice varieties? To answer this question, fertilizer and herbicide were treated simultaneously using a 3*3 contingency table (Table 4). The results indicated that the adoption of fertilizer and herbicide depended on the use of improved rice varieties. For rice farmers who did not adopt improved rice varieties, 41% of those who did not use fertilizer adopted herbicide. However, for rice farmers who adopted improved rice varieties, 26% used both fertilizer and herbicide. This means that rice farmers who adopted improved rice varieties were more likely to use either fertilizer or herbicide separately, or simultaneously adopt both fertilizer and herbicide. This pattern of adoption highlights the complementary nature of these inputs in rice production. The Pearson Chi-Square statistics, χ²(1) = 17.61, p < 0.01, suggest that there was a statistically significant association in the adoption of the improved rice varieties, fertilizer, and herbicide; signifying the simultaneous adoption of the three technologies.

Table 4 Simultaneity of adoption among rice farmers – a 3*3 contingency table.

Fertilizer	Herbicide	Improved rice variety		χ^2
		Not used	Used
Note used	Not used	100 (59.17)	232 (74.12)	11.44^***
	Used	69 (40.83)	81 (25.88)
Used	Not used	134 (62.91)	236 (73.52)	6.77^***
	Used	79 (37.08)	85 (26.48)
Overall χ^2(1)		17.61^***

Note: Figures in parentheses are percentages; χ² = Chi-square test.

*** p < 0.01.

The results revealed that technology promotion activities including extension services must first target the adoption of improved rice varieties since the adoptions of either fertilizer or herbicide or both are likely to follow. Also, smallholder rice farmers are more likely to adopt certain technologies individually than simultaneously. The following pairs of technological options are more likely to be adopted simultaneously: improved rice varieties and fertilizer, improved rice varieties and herbicide, fertilizer and herbicide, and improved rice varieties, fertilizer and herbicide. The simultaneous adoption of these technologies indicates that rice farmers respond positively to economic incentives to increase yield and profit. This is driven by the use of improved rice variety with a combination of fertilizer and/or herbicide to increase yield, suggestive of the need to design integrated development programs in promoting technologies.

4.4 Determinants of adoption of the technology bundles

Given the simultaneous nature of the decisions to adopt the technological options, we fitted an MNP model for the decision to adopt some combination of the technological options in rice production. Using the variance inflation factor to test multicollinearity, the results showed that none of the explanatory variables exceeded the threshold value of 10; implying no problem of multicollinearity in the data. Table 5 presents the parameter estimates from the MNP model on the impact of the 11 explanatory variables on the rice farmers’ adoption decisions. The improved rice variety [IRV]-herbicide (bundle 6) was the base outcome, by default, given its highest frequency of occurrence. This means that the ensuing results and discussions are based on the impact of the explanatory variables on the specific technology bundles choice relative to the IRV-herbicide bundle. The Wald test for the hypothesis that all coefficients in each technology bundle equation are jointly equal to zero is rejected (Wald χ²(77) = 624.27, p < 0.01), signifying that the variables included in the model explained significant portions of the variations in rice farmers’ technological choices. At least four explanatory variables were statistically significant in predicting the adoption of the technology bundles. The signs of the significant explanatory variables were consistent with a priori expectations and economic logic.

Table 5 Parameter estimates from the MNP model.

Indicators	Bundle 1	Bundle 2	Bundle 3	Bundle 4	Bundle 5	Bundle 7	Bundle 8
Tenure	0.13	0.01	−0.08	−0.25	0.21	−0.08	0.47^**
	(0.19)	(0.17)	(0.23)	(0.18)	(0.19)	(0.20)	(0.23)
Education	−0.07^***	0.04^**	−0.14^***	−0.07^***	0.04^**	−0.09^***	−0.01
	(0.02)	(0.02)	(0.03)	(0.02)	(0.02)	(0.02)	(0.02)
Experience	0.02^*	0.01	0.01	0.01	0.01	0.02	0.00
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Extension	0.23^**	−0.15^*	0.21^*	0.22^***	0.24^***	0.29^***	0.43^***
	(0.09)	(0.09)	(0.12)	(0.08)	(0.07)	(0.08)	(0.07)
Demonstration	−1.89^***	0.22	−2.34^***	−1.92^***	−0.79^***	−1.52^***	−0.23
	(0.29)	(0.23)	(0.39)	(0.26)	(0.25)	(0.27)	(0.35)
Association	0.17	−0.03	−0.68^***	0.16	0.21	−0.34^*	0.42^*
	(0.17)	(0.15)	(0.21)	(0.16)	(0.18)	(0.18)	(0.23)
Credit	−0.88^***	−0.02	−0.15	−0.37	0.13	−0.20	1.12^***
	(0.34)	(0.22)	(0.33)	(0.26)	(0.25)	(0.27)	(0.26)
Farm size	−0.29^***	−0.16^*	−1.24^***	−0.32^***	−0.54^***	−0.72^***	−0.89^***
	(0.11)	(0.09)	(0.24)	(0.10)	(0.13)	(0.14)	(0.17)
Household size	−0.04	0.01	−0.12^***	0.02	−0.05	−0.01	0.05
	(0.03)	(0.02)	(0.04)	(0.02)	(0.03)	(0.03)	(0.04)
Input availability	0.39	2.21^***	1.44^***	0.20	0.64^***	0.89^***	2.30^***
	(0.26)	(0.19)	(0.26)	(0.26)	(0.25)	(0.25)	(0.27)
Commercialization	−0.00	−0.00	0.02^*	−0.01	0.014^*	0.01	0.02^*
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Constant	0.93^*	−0.64	1.33^*	1.31^***	−1.29^*	0.63	−3.63^***
	(0.54)	(0.47)	(0.75)	(0.49)	(0.68)	(0.58)	(0.88)
Observations	1016	1016	1016	1016	1016	1016	1016
Log likelihood	−1487.18
Wald χ^2 (77)	624.27
Prob > chi2	0.00

Note: The base outcome is bundle 6; Figures in parentheses are standard errors.

*** p < 0.01.

** p < 0.05.

* p < 0.10.

The parameter estimates associated with education, demonstrations, credit, and farm size significantly predicted the probability of choosing the “no” technology bundle. For instance, higher education, frequent participation in field demonstrations, greater access to credit, and larger farm size predicted that a rice farmer was less likely to choose the “no” technology bundle. The strongest impacts were related to participation in field demonstrations, with a magnitude of 1.89; followed by credit with a coefficient of 0.88–both significant at the 1% level. Farm size showed a negative effect on the adoption of the “no” technology bundle, meaning that farm size was an important factor in the adoption of improved technologies in rice production. Smallholder rice farmers would adopt improved technologies if they had the opportunity to allocate portions of the farmland to the improved technologies without compromising their prospects to utilize unimproved technologies such as traditional rice varieties. However, higher rice farming experience and more extension contacts increased the chances of selecting the “no” technology bundle, even though the magnitude of the effects were small. This result may reflect that more frequent contacts with extension services may not necessarily exert the required behavioral influence on rice farmers, or past extension services as well as farmers’ past experiences with improved technologies might have failed in solving their problem, thereby, eroding their confidence.

The parameter estimates for the improved rice varieties only bundle generated some interesting insights. Education, extension contacts, farm size, and complementary input availability were the significant factors in the adoption of the improved rice variety bundle. More extension contacts and larger farms were less likely to influence the adoption of the improved rice varieties among the rice farmers. Conversely, farmers who were educated and had available complementary inputs were more likely to adopt the improved rice variety. For instance, if the availability of complementary inputs were to increase by one point, the chances of rice farmers adopting the improved rice variety only, compared to the IRV-herbicide bundle, would increase by 2%, while holding other factors constant. The results underscore the importance of information and the availability of complementary inputs in influencing the adoption of improved rice varieties. This implies that rice farmers who were educated as well as had adequate information and available complementary inputs were in the best position to decipher information and comprehend the economic incentives – higher yields, increased profits, and greater sustainability – of improved rice varieties, leading to adoption. The findings are consistent with previous studies (Dhital and Joshi, 2016; Ghimire et al., 2015; Mittal, and Mehar, 2016) that found education, farm size, extension services and seed access to influence the adoption of improved technologies.

Regarding the fertilizer-only bundle, the statistically significant predictors were education, extension, demonstrations, association, farm size, household size, complementary input availability, and commercialization. Unlike extension, complementary inputs availability, and commercialization, the other explanatory variables tended to decrease the likelihood of rice farmers to adopt the fertilizer-only bundle. For example, higher participations in field demonstrations and association meetings reduced the likelihood of adopting the fertilizer-only bundle. In fact, if farmers’ participation in field demonstrations were to increase by a point, the propensity of farmers to adopt the fertilizer-only bundle, compared to the IRV-herbicide bundle, would fall by 2%. However, the availability of complementary inputs would increase the likelihood of adopting the fertilizer-only bundle by 1% times. It is also obvious that smallholder rice farmers who were commercial-oriented will adopt fertilizer to increase yield and sell a higher proportion of output for financial rewards. Overall, demonstrations and complementary inputs availability were the strongest statistically significant predictors of the fertilizer-only bundle. In earlier studies, Ghimire et al. (2015), and Mittal and Mehar (2016) established that higher education, farm size, extension services, and seed access influenced the adoption of improved technologies. These results are largely consistent with those found in this study.

The parameter estimates associated with education, demonstrations, and farm size for the herbicide-only bundle were significantly different from zero at a 1% level of significance. This means that higher education, more participation in field demonstrations and having larger farms were associated with less likelihood of adoption of the herbicide bundle. For instance, frequent participation in field demonstrations (β = −1.92, p < 0.01) reduced the likelihood of adopting the herbicide bundle by about 2%. Higher extension contacts increased the likelihood of adopting the herbicide bundle by 0.2%. The study highlights the role of information (education and extension services) in influencing the adoption behavior of rice farmers.

Higher education, frequent extension contacts, complementary input availability, and a higher degree of commercialization appeared to increase smallholder rice farmers’ chances of adopting the IRV-fertilizer bundle. The importance of the higher degree of commercialization and the availability of complementary inputs was also evident in the adoption of the IRV-fertilizer bundle. Likewise, with the incentive to sell a greater proportion of harvested rice outputs to achieve higher profits, farmers were more likely to adopt the IRV-fertilizer bundle. However, higher participation in field demonstrations and having larger farms tended to reduce their propensity to adopt the IRV-fertilizer bundle. Rice farmers who were more educated, informed, and exposed to improved production practices were more likely to have adequate knowledge about the economic benefits of adopting the IRV-fertilizer bundle.

The last but one technology bundle is the fertilizer-herbicide bundle. Smallholder rice farmers’ decisions to adopt the fertilizer-herbicide bundle, relative to the IRV-herbicide bundle, were significantly influenced by their education, extension contacts, demonstrations, farm size, and complementary inputs availability. Higher education and availability of complementary inputs tended to lead to the adoption of the fertilizer-herbicide bundle. Fertilizer and herbicide help to replenish soil nutrient and reduce the negative effects of weeds on yields. More exposure to extension services and the availability of these inputs would, therefore, helped rice farmers to appreciate these potential benefits and concurrently adopt them. However, it appeared that higher education, more participation in field demonstrations, frequent participation in association meetings, and larger farm size tended to reduce the probability of adopting the fertilizer-herbicide bundle.

Tenure, extension contacts, credit, commercialization, farm size, complementary inputs availability, and association tended to be directly related to the adoption of the IRV-fertilizer-herbicide bundle. Access to credit and availability of complementary inputs increased the probability of adopting the IRV-fertilizer-herbicide bundle by almost 112% and 230%, respectively. This indicates that access to capital for cash-constrained farmers is key to not only purchasing complementary sets of inputs but also for simultaneously adopting such inputs given their combined effects on production. Also, increasing farmers’ knowledge of complementary inputs through frequent extension contacts and participation in association meetings, given their commercial orientation, tended to enhance their chances of adopting the IRV-fertilizer-herbicide bundle. However, larger farm size tended to reduce the likelihood of adopting the IRV-fertilizer-herbicide bundle. Our findings suggest that rice farmers who are economically inclined to achieving higher yields would invest in productivity-enhancing technologies and sell higher proportions of outputs to meet their profit objective.

The marginal effects for the MNP model are shown in Table 6. The marginal values were evaluated at the mean values of the explanatory variables. It is evident that most of the significant variables both have significant marginal effects and show modest magnitudes. For instance, the probability of selecting the “no” technology bundle (i.e., the farmer chooses no such improved technologies as improved rice variety, fertilizer, and herbicide) is on average about 19% points lower for the farmers with more participation in field demonstrations, all other variables held constant. Similarly, a marginal change in the number of extension contacts from the average of two times was associated with a 1% increase in selecting the “no” technology option. Moreover, the likelihood of adopting the IRV-fertilizer-herbicide bundle was on average about 5% points higher for access to credit than no access to credit.

Table 6 Marginal effects from the MNP model.

Indicators	Bundle 1	Bundle 2	Bundle 3	Bundle 4	Bundle 5	Bundle 7	Bundle 8
Tenure	0.02	0.00	−0.00	−0.06	0.04	−0.01	0.02^*
	(0.02)	(0.04)	(0.01)	(0.03)	(0.02)	(0.02)	(0.01)
Education	−0.01^***	0.02^***	−0.01^***	−0.01^***	0.01^***	−0.01^***	0.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Experience	0.00	0.00	0.00	−0.00	0.00	0.00	−0.00
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Extension	0.02^*	−0.09^***	0.01	0.03^**	0.03^***	0.03^***	0.01^***
	(0.01)	(0.02)	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)
Demonstration	−0.19^***	0.34^***	−0.08^***	−0.25^***	−0.00	−0.12^***	0.02^*
	(0.04)	(0.05)	(0.02)	(0.04)	(0.03)	(0.03)	(0.01)
Association	0.027	−0.02	−0.03^***	0.03	0.03	−0.06^**	0.02^*
	(0.02)	(0.03)	(0.01)	(0.03)	(0.02)	(0.02)	(0.01)
Credit	−0.13^***	0.04	−0.00	−0.05	0.05	−0.01	0.05^***
	(0.04)	(0.05)	(0.01)	(0.04)	(0.03)	(0.03)	(0.02)
Farm size	0.01	0.05^**	−0.05^***	−0.00	−0.04^**	−0.06^***	−0.02^***
	(0.02)	(0.02)	(0.01)	(0.02)	(0.02)	(0.02)	(0.01)
Household size	−0.01	0.01	−0.01^***	0.01	−0.01	−0.00	0.00^*
	(0.00)	(0.01)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Input availability	−0.08^**	0.47^***	0.03^**	−0.14^***	−0.04	0.01	0.06^***
	(0.03)	(0.04)	(0.01)	(0.04)	(0.03)	(0.03)	(0.02)
Commercialization	−0.00	−0.00	0.00^*	−0.00^**	0.00^**	0.00	0.00^*
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Observations	1016	1016	1016	1016	1016	1016	1016

Note: The base outcome is technology bundle 6; figures in parentheses are standard errors.

*** p < 0.01.

* p < 0.10.

4.5 Heterogeneity of the impacts of the predictors across technology bundles

The parameter estimates from the MNP model showed several heterogeneous impacts of the explanatory variables on the technological options. As shown in Table 7, with the exception of tenure, experience, and household size, all the explanatory variables had significant heterogeneous impacts across the technology bundles. The strongest indicators of the probability of adoption of all technology bundles were complementary input availability and field demonstrations. Complementary input availability consistently predicted the probability of adopting five of the technology bundles while education predicted six of the technology bundles.

Table 7 Predictors showing significant heterogeneous impacts on technology bundles.

Indicators	Bundle 1	Bundle 2	Bundle 3	Bundle 4	Bundle 5	Bundle 7	Bundle 8
Tenure							(+)
Education	(−)	(+)	(−)	(−)	(+)	(−)
Experience	(+)
Extension	(+)	(−)	(+)	(+)	(+)	(+)	(+)
Demonstration	(−)		(−)	(−)	(−)	(−)
Association			(−)			(−)	(+)
Credit	(−)						(+)
Farm size	(−)	(−)	(−)	(−)	(−)	(−)	(−)
Household size			(−)
Input availability		(+)	(+)		(+)	(+)	(+)
Commercialization			(+)		(+)		(+)

Note: Summarized from. ;+ and − indicate statistically significant positive and negative effects, respectively.

Frequent contacts with extension services increased the chances of adopting fertilizer-only, herbicide-only, the IRV-fertilizer bundle, the fertilizer-herbicide bundle, and the IRV-fertilizer-herbicide bundle, but reduced the probability of adopting improved rice varieties. The largest effect of extension contacts was on the adoption of the IRV-fertilizer-herbicide bundle (β = 0.43, p < 0.01). Similarly, higher participation in field demonstrations not only reduced the likelihood of independently adopting fertilizer and herbicide but also decreased farmers’ chances of adopting the IRV-fertilizer and fertilizer-herbicide bundles. Field demonstrations had the strongest impact on the adoption of fertilizer and herbicide. The interdependencies of extension services (through extension contacts and organizing field demonstrations) across the technology bundles underscore the role of extension services in providing the requisite knowledge and experience required for changing the perceptions and attitudes of farmers towards technology adoption.

Education was found to be a key indicator of technology adoption, haven consistently predicted the probability of adopting six of the technology bundles. For example, better-educated rice farmers were less likely to independently adopt either the fertilizer or the herbicide bundles, and simultaneously adopt fertilizer and herbicide bundle. However, better-educated rice farmers were more likely to either independently adopt the improved rice varieties only or adopt the IRV-fertilizer bundle. Higher educational attainment enables rice farmers to understand and appreciate the benefits of adopting improved technologies. Education had the strongest impact on the fertilizer bundle (β = −0.14, p < 0.01); implying that educated farmers were more likely to adopt fertilizer compared to the other technology bundles. The consistent effect of education in predicting six of the technology bundles reveals that rice farmers with higher education appear to have a higher allocative ability to adjust faster to favorable economic environments such as availability of complementary inputs and readily available markets by simultaneously adopting improved technologies.

Participation in association meetings was related to the adoption of the combination of the technological options. A greater participation in association meetings by the rice farmers lowered the probability of adopting the fertilizer-only and the fertilizer-herbicide bundles, while the likelihood of adopting the IRV-fertilizer-herbicide bundle increases. Participation in association meeting had the strongest impact on the fertilizer bundle (β = 0.68, p < 0.01). Rice farmers’ involvement in networks through association meetings created the platform for information sharing regarding new or improved methods of production leading to adoption.

Lack of credit and unavailability of complementary inputs may act as a disincentive to the adoption of improved technologies. However, the availability of complementary inputs increased the likelihood of rice farmers to adopt the improved rice varieties and the fertilizer separately and the adoption of the IRV-fertilizer, the fertilizer-herbicide, and the IRV-fertilizer-herbicide bundles. Rice farmers who had access to credit were less likely to choose the “no” technology bundle but more likely to adopt the IRV-fertilizer-herbicide bundles. That is, cash-constrained rice farmers without access to credit were more likely to choose the “no” technology bundles while those who had access to credit were very likely to adopt the IRV-fertilizer-herbicide bundles, as they could finance the purchase of the complementary inputs. The heterogeneous impacts of the explanatory variables were also evident in the commercialization variable.

As can be seen, rice farmers who sold a higher proportion of their output were more likely to not only adopt the IRV-fertilizer bundle or the IRV-fertilizer-herbicide bundle but also to adopt the fertilizer-only bundle. This means that the commercial orientation of rice farmers had significant effects on choosing technologies that maximized their economic incentives of yield and profit. Credit and complementary input availability strongly predicted the adoption of the IRV-fertilizer-herbicide bundle (β = 1.12, p < 0.01) and the improved rice variety only (β = 2.21, p < 0.01). Moreover, rice farmers with larger farms were less likely to independently adopt the improved rice varieties, fertilizer, and herbicide, as well as adopt the IRV-fertilizer and the IRV-fertilizer-herbicide bundles or the “no” technology bundle. This means that rice farmers with larger farms were less likely to allocate some portions of their farms to improved technologies, such as improved rice while maintaining the remaining land for their traditional technologies as a safety net.

It is obvious that rice farmers were more likely to adopt part or all possible combinations of technological options in rice production when they were more educated, participated more in association meetings and field demonstrations, had more contacts with extension services, had access to credit, were commercially focused and had available complementary inputs. Past studies including Mittal and Mehar (2016), Dhital and Joshi (2016), and Ghimire et al. (2015) showed that more educated farmers, larger farms, more experienced farmers, access to extension services and inputs increased the probability of adopting improved technologies. These results generally conform to those found in this study, except for farm size. As shown in this study, farm size decreased the probability of adopting all technology bundles. Feder et al. (1985) explained that farm size had varying impacts on technology adoption. Notwithstanding, the negative effects of farm size on the adoption decisions is evident in the characteristics of the acreage of land allocated to rice production by the smallholder farmers. The average acreage cultivated was 1.26 ha. In fact, 87% of the rice farmers had allocated less than 2 ha of land to rice production, 7% had between 2 and 3 ha of land for rice production while 6% allocated more than 3 ha to rice production. With such a large magnitude of small rice farms, rice farmers were probably less likely to adopt improved varieties. Moreover, several studies had consistently shown that explanatory variables such as farm size, education, extension contact, and access to credit predicted the adoption of single technologies (Dhital and Joshi, 2016; Ghimire et al., 2015; Mittal and Mehar, 2016; Simtowe et al., 2016). However, evidence of the heterogeneous impacts of the explanatory variables on the technological choices revealed the underlying logic of avoiding the generalizing of such explanatory variables on the adoption of unspecified technologies. This is because the particular impact depended on the specific technology and whether the particular technology was adopted individually or jointly.

5 Conclusions and policy implications

In this study, we investigated the (a) nature of simultaneities in the adoption of rice productivity-enhancing technologies, and (b) factors that influence smallholder rice farmers’ adoption decisions regarding multiple productivity-enhancing technologies. Smallholder rice farmers are more likely to adopt technologies introduced as a package individually, partially, or collectively. When faced with improved rice variety-fertilizer-herbicide technological options, smallholder rice farmers’ adoption decisions exhibit interdependencies and simultaneities that need to be correctly modeled to understand their adoption behavior. Their decision-making falls into any one of four possibilities: (a) reject all technological options; (b) adopt a particular technology only (e.g., improved rice varieties or fertilizer, or herbicide); (c) adopt two of the technologies (i.e., improved rice varieties and fertilizer, improved rice varieties and herbicide, fertilizer and herbicide); and (d) adopt all technological choices (i.e., improved rice varieties, fertilizer, and herbicide). The adoptions of these technologies are consistently and significantly predicted by farm size, extension, education, demonstration, complementary input availability, and commercialization.

Evidence provided through the MNP analysis has yielded four policy-relevant implications for facilitating the adoption of technological innovations in rice production. First, technological promotion activities will be better facilitated through an integrated extension system that promote access to extension services, create knowledge, and build the expertise of rice farmers in using improved technologies. Second, technology promotion activities must integrate credit provision with an effective input supply system to facilitate adoption. In particular, rice sector policies must seek to guarantee credit availability and subsidization for rice farmers. Also, productive inputs must be readily available and affordable. This will ensure that rice farmers who have credit could purchase complementary inputs at the right time and quantity for the specific production activity. Third, technology promotion efforts must consider the orientation of rice farmers towards maximizing their economic incentives. Rice farmers who are commercial-oriented must choose bundles of technologies that maximize their economic incentives of yield and profit. Fourth, evidence of the interdependencies in the adoption of the technology bundles suggests that such interdependencies can be considered when developing technology promotion and dissemination strategies to increase productivity in rice production. It should be possible to design technology promotion activities that concurrently reduce farmers’ subjective uncertainties and risks, promote credit subsidies, as well as ensure effective input delivery systems while targeting specific technological innovations.

Conflict of interest

We declare that no competing interests exist.

Authors’ contributions

This work was carried out in collaboration between both authors. Author ET designed the study, wrote the protocol, performed the statistical analysis, wrote the first draft of the manuscript and managed literature searches. Author JRB managed the analyses of the study, literature searches and critically reviewed the final manuscript for important intellectual content. Both authors read and approved the final article.

Acknowledgements

We thank the Ghana Strategy Support Program of the International Food Policy Research Institute, Ghana for providing part funding for the research leading to this paper. The authors owe a debt of gratitude to the rice farmers in the four Municipalities who made the empirical findings of this study possible.

Appendix A

Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://doi:10.1016/j.njas.2017.07.004.