Does adoption of cocoa hand pollination (CHP) improve welfare of farmers? Evidence from smallholder cocoa farmers in Ghana

ABSTRACT

This paper evaluates the impact of the adoption of cocoa hand pollination (CHP) on the welfare of smallholder cocoa producers. Using data collected from 1200 cocoa farmers in Huni Valley and Kejebril districts of Ghana, we employed endogenous switching regression, propensity score matching and inverse probability weighted adjustment techniques to assess the impact. The two-step Cragg was used to examine the determinants of the decision and intensity of adoption. The results showed that education, marriage and farm size positively influenced both the decision and intensity of adoption of the CHP technique. The result further showed that the cost of labour for implementing the technology, age, education, family head, economic active members, age of the cocoa tree, off-farm work, credit access and farm size significantly determined the impact of adoption on smallholder cocoa producers’ welfare. Additionally, there is a positive impact of CHP adoption on productivity, income and food security. Careful consideration should be given to these factors including collaborations between government and stakeholders in the cocoa industry through the regular sensitization and trainings for farmers on improved technologies as the CHP to increase productivity, household income and reduce food insecurity of smallholder cocoa farmers.

KEYWORDS:

• Adoption
• propensity score matching
• hand pollination
• endogenous switching regression
• welfare

View correction statement:

Correction

Introduction

Africa continues to lead global cocoa production by accounting for 70% of the cocoa beans produced worldwide with Cote D’Ivoire, Ghana and Nigeria as the highest contributors on the continent (World Cocoa Foundation, 2020). Cocoa production and its related activities remain the main source of livelihood for more than 20 million people on the continent (International Finance Cooperation, 2022). In West Africa, the number of family farms under cocoa production is estimated to be over two million (International Finance Cooperation, 2022; World Cocoa Foundation, 2020). Ghana and Cote D’Ivoire alone account for more than 60% of the global cocoa beans supply (FAOSTAT, 2019). Cocoa production is, therefore, considered the mainstay of some West African economies (Bangmarigu & Qineti, 2018; Brako et al., 2021). For instance, in Ghana, cocoa contributes about 30% (US$2 billion annually) to export earnings and 10.6% to Gross Domestic Product (GDP) (Bangmarigu & Qineti, 2018; GSS, 2021). A recent study by OPEC Fund (2022) reveals that the cocoa sub-sector of agriculture employs over 800,000 farmers and their families in Ghana.

Despite the importance of the cocoa sector and its contribution to livelihoods, the industry faces sustainable development challenges: falling productivity, low farmer incomes and poor infrastructure among others (International Finance Cooperation, 2022). Some studies have confirmed that the cocoa-producing countries in Africa often fail to achieve their potential yield (Aneani et al., 2011; Aneani & Ofori-Frimpong, 2013; International Finance Cooperation, 2022; World Cocoa Foundation, 2015). FAOSTAT (2020) estimated the average cocoa yield in Ghana between 2015 and 2019 to be 525 kg/ha, representing only 30% of the estimated yield potential of 1889 kg/ha. A related study by Daymond et al. (2017) reported the average cocoa yield in Cote D’Ivoire to be 552 kg/ha, far below the potential yield of about 1900 kg/ha. Overall, cocoa yield, particularly in West Africa, has remained relatively stable around 400 kg/ha whilst Indonesia records an average yield of 1229 kg/ha (Danso-Abbeam & Baiyegunhi, 2018; Daymond et al., 2017; Van Vliet & Giller, 2017; World Bank, 2019). These yield gaps have culminated in low income and deterioration of the livelihoods of the cocoa farmers (Amponsah-Doku et al., 2022). A recent study by Van Vliet et al. (2021) revealed that about 2 million cocoa farmers in Ghana and Côte d’Ivoire are living in poverty with 59% of them unable to afford quality food, health care and education. Low productivity and its attendant repercussions on smallholders have been attributed to overaged cocoa farms (World Cocoa Foundation, 2020), high incidence of pests and disease (Amponsah-Doku et al., 2022; Djokoto et al., 2016), climate uncertainties, low adoption of improved technologies (Danso-Abbeam & Baiyegunhi, 2018), lack of credits and low soil fertility (Wessel & Quist-Wessel, 2015). As a result of climate change and variability, cocoa yield has been projected to further reduce by 2.3% (International Cocoa Organization, 2018).

To address the issue of low cocoa yield, particularly in West Africa, many government policies have been designed and implemented in the main cocoa-growing areas such as Cote D’Ivoire, Ghana and Nigeria (Toledo-Hernández et al., 2020). The main interventions include cocoa rehabilitation, cocoa agroforestry, introduction of drought-tolerant varieties, cocoa mass spraying and fertilizer subsidy programmes (Bangmarigu & Qineti, 2018; COCOBOD, 2020). These sustainability initiatives are necessary but not sufficient to close the cocoa yield gap especially when it has been well-established that less than 10% of cocoa flowers get pollinated under natural conditions (Toledo-Hernández et al., 2020). In recent years, hand pollination has been promoted as a reliable option to improve cocoa yield among smallholder farmers. Cocoa hand pollination (CHP) involves the removal of pollen from the cocoa flower on the same tree or nearby trees and manually attaching it on the stigma leading to cross-pollination (Chan, 2022). A field trial of CHP in Indonesia resulted in about 161% increase in yield and 69% improvement in the income of the participated farmers (Toledo-Hernández et al., 2020). A similar experiment in Cote D’Ivoire revealed that hand pollination enhanced pollen disposition rate and increased the number of seeds per fruit resulting in an improved yield (Forbes et al., 2019; Forbes & Northfield, 2017). In Ghana, hand pollination was officially introduced to smallholder farmers by the Ghana Cocoa Board in 2017 after successful trials (COCOBOD, 2020). This technology was then enrolled in all cocoa-producing regions in Ghana for farmers to subscribe through extension officers to carry out hand pollination on their cocoa farms. During the season, farmers whose farms are ready for pollination contact the extension officers to make the necessary arrangements for the commencement of pollination. According to Wongnaa et al. (2021), the smallholder farmers have a positive perception of CHP. However, the adoption rate among the sampled smallholder cocoa farmers remains low (49%). Wanger et al. (2021) further reported a positive relationship between CHP and biodiversity. Other related studies revealed that the said technology increased the number of fruits per cocoa tree as well as the quality of the fruits (Gupta et al., 2017; Sánchez-Estrada & Cuevas, 2020; Vera-Chang et al., 2016). Whilst the existing studies are credited for their significant contributions to the literature, there remains a dearth of empirical evidence on the impact of hand pollination on the welfare of particularly smallholder cocoa farmers. It is, therefore, hypothesized that the adoption of CHP increases the yield and welfare of smallholder cocoa farmers. Improvement in yield is expected to increase income for cocoa farmers and positively impact their ability to acquire the needed foodstuffs to ensure food/nutrition security (Etuah et al., 2020).

This study, therefore, makes the following specific contributions to the literature. First, it identifies and examines the factors that influence the decision of smallholder cocoa farmers to adopt the hand pollination technology. This could be useful when designing programmes or activities to sensitize cocoa farmers on the said technology to increase the rate of adoption. Second, this study empirically quantifies the impact of CHP technology on cocoa productivity, household income and food security of smallholder cocoa farmers. Further investments in any intervention or programme depend on its impact on the targeted population (Etuah et al., 2020). The findings of this study could, therefore, serve as a guide to policy makers and agricultural development agencies when making investment decisions in the cocoa sectors to improve yield and enhance the standard of living of cocoa farmers.

Rogers (1962) first defined adoption as the use and continual use of innovation that has the influence to drive change. Feder et al. (1985) also asserted that adoption is the degree of use of a new technology in which in the long run a particular farmer is aware of the full information about the new technology and its potential. Regarding this definition, adopters are grouped into innovators, early adopters, early majority, late majority and laggards (Rogers, 1962). Adoption of technology has been explained using different theories such as the theory of planned behaviour; theory of reason action; unified theory of acceptance and use of technology; diffusion innovation theory; and technology-organization-environment framework. Based on the random utility theory, it is assumed that a cocoa farmer is a rational producer and hence, makes rational decisions. These individual farmer decisions are made towards the maximization of utility in the context of cocoa technology. For an ith cocoa farmer to adopt CHP, the benefit that is expected $\text{β}\ast ji$ that a farmer derives from the adoption of the jth technology which is a latent variable determined by observed households, institutional and farm-specific characteristics (Diiro et al., 2015).

Methodology

Study area

The study was conducted in the Huni Valley and Kejebril districts located in the Western North of Ghana. The Huni-Valley District has a total population of 159,304 (Ghana Statistical Service, 2010) and a land size of 1376 km². The district is situated east of Tarkwa but lies within the southwestern equatorial zone. The district is known for economic activities such as farming, mining and trading among others. In the context of Kejebril district, located at Yabu at East and Beaho at North-East and lies within the latitude and longitude 4°54′34.1″ north and 1°50′4.1″ west, respectively. The two districts are among the leading producers of food and tree crops in Ghana. The districts recorded a bimodal rainfall with an average annual rainfall of 1878.3 mm (GSS, 2010). About 52 percent of the districts’ population are into agriculture (GSS, 2010). Crops grown in the districts consist of cocoa (Theobroma cacao L.), maize (Zea mays L), plantain (Musa spp), cassava (Manihot esculenta C.), oil palm (Elaeis Guineensis J.), rice (Oryza sativa L.), yam (Dioscorea spp) and others.

Multi-stage sampling was employed to sample cocoa farmers for the study. First, we purposively selected the Western North region due to its significant contribution (332,647 metric tonnes representing 43.37% in 2020) to cocoa production in Ghana (AgriGold Magazine, 2022). Second, within the region, the two key cocoa-producing districts: Kejebril and Huni Valley were purposively selected because the two districts are among the top producing districts in the region. Third, 10 communities under each of the two cocoa districts were selected using the simple random sampling approach. At the final stage, there was a simple random sampling of 120 cocoa farmers from each of the selected communities. At the end of the sampling exercise, 1200 were selected for the survey. Data were collected between August and October, 2021. It covers the socioeconomic and institutional characteristics of cocoa farmers, information on the adoption of CHP, food security status, cocoa production and household income as well as the constraints faced in adopting the CHP. Face-to-face interviews were done with the questionnaires translated into the local dialect of the farmers in order achieve the objectives of the study. Data gathered from cocoa farmers was analysed using Stata MP version 16.

Determinants of decision and intensity of adoption of CHP

In this study, the sampled cocoa farmers are those who adopted the CHP and those who did not adopt it. Consider $A_i$ signifies the binary decision to adopt the CHP, thus farmers who adopted the CHP denote as $A_i=1$ whiles non-adopters denote as $A_i=0$. Afterward, the intensity of adoption is captured based on the number of hectares under CHP divided by the total farm size for cocoa production. This is expressed below as; (1) $I{A}_k={\displaystyle \frac{LCH{P}_k}{T{L}_k}}$(1) where $IA$ =  Intensity of adoption, $LCH{P}_k$ =  land under CHP from the kth selected farmer, and ${\sum }_{k=1}^{n}T{L}_k$ =  total land under cocoa production from the kth selected farmer.

Since two decisions are expected to be made separately and different factors do not depict similar effects on the two decisions, the two-step Cragg is appropriate for such decisions (Cragg, 1971). This is because the two-stage Cragg best explains why some farmers adopted the CHP and why others did not, and if adopted, to what extent adopted farmers allocated land for the CHP.

Drawing on past studies, the probit model which is the first stage of the Cragg model, thus for the adoption of CHP (Asante et al., 2017; Belotti et al., 2012) was employed. Based on the utility maximum theory (Rahm & Huffman, 1984), adoption decision is mutually exclusive and may be influenced by different factors such as socioeconomic characteristics, farm, and institutional level factors and other factors that could not be captured. The latent utility of the ith sample farmer for jth adoption is represented as: (2) $Y_{ij}=X_i{F}_j+l_{ji},i=1,2,3,4\dots .,n\mathrm{a}\mathrm{n}\mathrm{d}j=0,1$(2) where $X_i$ =  variables associated with ith sample farmer’s utility. The likelihood that ith sample farmer adopt the CHP specified below as; (3) $P_r=(A_i=1/{\text{β}}_i{\alpha }_i)=1-\lambda (-{\text{β}}_i\alpha {}_i)$(3) and for observing 0, the outcomes expressed as; (4) $P_r=(A_i=0/{\text{β}}_i{\alpha }_i)=\lambda (-{\text{β}}_i\alpha {}_i)$(4) where $\lambda$ = cumulative distribution function.

In the second stage of the Cragg model, the use of the truncated model to analyse the intensity of adoption of CHP after sampled farmer adoption. The truncated model deemed the appropriateness for the intensity of adoption of the CHP assumed to be autonomous with normal random errors distribution specified as: (5) $A_i|a_i^{\ast }>\text{β}{Z}_{2i}+{ϵ}_i>Q$(5) where $A_i$ = intensity indices; $a_i^{\ast }$ = unobservable latent variable; $Q$ = unobservable threshold value (in this study, $Q$ means adoption status, i.e. adopted or not); and $Z_{2i}$ = vector of values of independent variables. Using the maximum likelihood estimator (MLE), some of the independent variables can be found to influence both decisions. To validate the use of the two-step Cragg model for this study, we conducted a likelihood ratio test, thus separate estimations of the probit, truncated, and Tobit models (Asante et al., 2017; Mal et al., 2012). To compute the likelihood ratio test (λ), we used the log-likelihood values of the models (Asante et al., 2017; Katchova & Miranda, 2004) which is specified as: (6) $\lambda =-2(L{L}_{probit}+L{L}_{Truncated}-L{L}_{Tobit})$(6) where $L{L}_{probit}$ = log-likelihood of the probit; $L{L}_{Truncated}$ =  log-likelihood of the truncated; and $L{L}_{Tobit}$ = loglikelihood of the Tobit. Therefore, the estimated λ should be greater than the χ² critical value in order to approve the Cragg two-step model usage (Asante et al., 2017; Katchova & Miranda, 2004; Mal et al., 2012).

Estimation of impact of CHP on welfare

The study further analyses the impact of the CHP on household welfare. In this study, welfare indicators were productivity, household income and food insecurity. We defined productivity as a ratio of output concerning the land area under cocoa production. The household income was measured by accumulating farm and off-farm incomes. The farm income consists of sales from cocoa, other trees or food crops as well as livestock. The food insecurity was computed using the Household Food Insecurity Access Scale (HFIAS). The HFIAS ranges from 0 to 27 where close to zero is more food secure and close to 27 is more food insecure. To account properly for the impact of the CHP on welfare, the Endogenous Switching Regression (ESR) model was employed. Many studies used this model because of the ability to demonstrate between unobserved attributes, adopted the Full Information Maximum Likelihood (FIML), allows overlap, and considered as efficient to account for both the adoption and outcome equations (Abdulai, 2016; Ngwako et al., 2021). In the ESR model, the first stage as the selection equation is the same as the probit model discussed in the above section. Hence, we focused on the second stage of the model. The second stage demonstrates the impact of adoption on the outcome variables (productivity, household income and food insecurity). This is specified below as: (1) $\mathrm{Re}gime1:=Y_1={\text{β}}_1{X}_1+{ϵ}_1\mathrm{i}\mathrm{f}A=1$(1) (2) $\mathrm{Re}gime2:=Y_0={\text{β}}_0{X}_0+{ϵ}_0\mathrm{i}\mathrm{f}A=0$(2) where $Y_1$ and $Y_0$ =  welfare indicators for adopters and non-adopters, respectively; $X$ =  vector of independent variables (socioeconomic, farm level and institutional variables); ${\text{β}}_0$, ${\text{β}}_1$ =  vector of parameters to be estimated and ${ϵ}_0,{ϵ}_1$ are assumed to have a trivariate normal distribution with mean zero and non-singular covariance matrix (Mulwa et al., 2017), specified as: (3) $\mathrm{cov}({ϵ}_1,{ϵ}_0,CH{P}_i)=\left[\begin{array}{ccc}{\sigma }_{CHP}^2& {\sigma }_{{ϵ}_{1}CHP}^{}& {\sigma }_{{ϵ}_{0}CHP}^{}\\ {\sigma }_{CHP{ϵ}_1}^{}& {\sigma }_{{ϵ}_1}^2& .\\ {\sigma }_{CHP{ϵ}_0}^{}& .& {\sigma }_{{ϵ}_0}^2\end{array}\right]$(3) where ${\sigma }_{CHP}^2$ denotes variance of $CH{P}_i$, ${\sigma }_{CHP{ϵ}_1}^{}$ denotes covariance of $CH{P}_i$ and ${ϵ}_1$, ${\sigma }_{CHP{ϵ}_0}^{}$ denotes covariance of $CH{P}_i$ and ${ϵ}_0$, ${\sigma }_{{ϵ}_1}^2$denotes variance of ${ϵ}_1$, and ${\sigma }_{{ϵ}_0}^2$denotes variance of ${ϵ}_0$. Based on other studies, both the outcomes of adopters (Y₁) and non-adopters (Y₀) are not observed concurrently. Hence, the covariance of ${ϵ}_0$ and ${ϵ}_1$ is not defined with no information regarding the covariance. The full information maximum likelihood estimation (FIMLE) method becomes appropriate for more robust estimation. For identification, information sources on the technology such as extension, FBO and colleague farmer were used as the instrumental variables. This is because they could influence adoption decisions but not the welfare indicators. We validated the instrument variables prior to running the ESR model, thus, valid test in the selection model (Asante et al., 2023; Di Falco et al., 2011). The instruments have a joint influence on decision to adopt (Model 1, χ² = 21.35; p = 0.000), however, not of the outcomes of non-adopters. Also, the multicollinearity test conducted shows no correlation and hence, we proceeded with the estimations. See Appendix A1 and A2 for the test of validity of instrument selection and multicollinearity test, respectively. Following Cameron and Trivedi (2005) and Weyori et al. (2019), the log-likelihood function of the FIML estimator is given as: (4) $\begin{array}{rl}& \ln L=\sum _i(CH{P}_i{\theta }_i\left[\ln \left\{\ln \text{β}\left({\delta }_{1i}\right)\right\}+\ln \left\{\phi \left({ϵ}_{1i}/{\lambda }_1\right)/{\lambda }_1\right\}\right]\\ & +(1-CH{P}_i){\theta }_i\left[\ln \left\{1-\mathrm{\Omega }\left({\delta }_{0i}\right)\right\}+\ln \left\{\phi \left({ϵ}_{0i}/{\lambda }_0\right)/{\lambda }_0\right\}\right])\end{array}$(4) where $\mathrm{\Omega }$(.) denotes cumulative normal distribution function whereas $\phi$ (.) denotes a normal density distribution function, ${\lambda }_i$ denotes an optional weight for observation i. After the estimation of the parameters in models, the impact of CHP on welfare defined as the average treatment effect on the treated (ATT) is computed as: $\begin{array}{rl}& AT{T}_{ESR}=E\left(Y_{i1}\right|FB{O}_i=1,X_i)-E(Y_{i1}|FB{O}_i=0,X_i)\\ & ={\sigma }_1{\eta }_1\mathrm{\Phi }\left(X_i\mathrm{\Psi }\right)/\mathrm{\Omega }\left(X_i\psi \right)-{\sigma }_1{\eta }_1\mathrm{\Phi }\left(X_i\mathrm{\Psi }\right)/\{1-\mathrm{\Omega }(X_i\mathrm{\Psi }\left)\right\}\end{array}$

Robustness check

For a robustness check of the ESR results, both inverse probability weighted adjustment (IPWRA) and propensity score matching (PSM) were used. Propensity score matching involves the pairing of treatment and counterfactual units with similar values on the propensity score and possibly other covariates while removing all the unmatched units (Rosenbaum & Rubin, 1983; Shaikh et al., 2009; Smith & Todd, 2005). PSM method incorporates two stages:x the probit model and computation of average treatment effect (ATT). The probit model for estimating the propensity scores is expressed as: (9) $p\left(X_i\right)=Pr[A_i=1|X_i]=E[A_i\left|X_i\right];p\left(X_i\right)=\mathrm{\Omega }\left\{t\right(X_i\left)\right\}$(9) where $\mathrm{\Omega }${.} denotes normal cumulative distribution and X denotes the vector of pre-treatment characteristics. Based on the Conditional Independence Assumption (CIA), the PSM performs the best selection on observables giving a potential outcome of an independent technology choice of the covariates. ATT is, therefore, computed once the propensity scores are calculated which is expressed as: (10) $ATT=E\{{\sigma }_1-{\sigma }_0|A=1\}$(10) (11) $ATT=E\left[E\right\{{\sigma }_1-{\sigma }_0|A=1,p(X\left)\right\}]$(11) (12) $ATT=E\left[E\right\{{\sigma }_1|A=1,p(X\left)\right\}-E\left\{{\sigma }_0\right|A=0,p\left(X\right)\left\}\right|A=1]$(12)

Algorithms of the PSM used in this study are the nearest neighbour matching (NNM), kernel-based matching (KBM) and radius matching (RM) methods due to their ability to give exact empirical results on impact studies (Donkor et al., 2019; Ma & Abdulai, 2016; 2019).

Inverse probability weighting (IPW) reduces selection bias in the case of missing data (Höfler, 2005). It incorporates the inverse probability weighting and regression adjustment (RA) methods to estimate averages of treatment predicted outcome. For double robust feature, the IPWRA ensures consistency in estimating the parameters of the treatment model by calculating the inverse-probability weights. The variance of the IPW estimator is consistently estimated provided the weighting is taken into account (Robins et al., 1994). Two stages are involved in the IPWRA estimation. The first stage, thus adoption in treatment support a weight of $w_i=1/e_i$ is given, while adoption in control support is given as $w_i=1/(1-e_i)$. Following Cattaneo (2010) and Tan (2010), the sample estimation vector is specified as: (1) $P\left(y\right|do(Y=y))=\sum _{x}P\left(g\right|y,x\left)P\right(x)=\sum _x{\displaystyle \frac{P\left(g\right|y,x)\ast P(x)}{P\left(y\right|x)}}=\sum _x{\displaystyle \frac{P(g,y,x)}{P\left(y\right|x)}}$(1) where; t = treatment variable, x = independent variables and y = outcome variable. In the case of the continuous outcome, fit linear regression model on treatment by computing the adjusted mean treatment group difference weighted by the inverse probability of obtaining treatment (Lunceford & Davidian, 2004). Table 1 presents the variables used in the models.

Table 1. Explanatory variables used in the Models.

Variables	Measurement	Expected sign
Sex	1 = Male and 0= Female	+/-
Age	Years	+/-
Marital status	1 = Married and 0 = Unmarried	+/-
Years of schooling	Years	+/-
Household head	1 = Yes and 0 = No	+/-
Household size	Number	+/-
Economic active members	Number	+/-
Off farm activity	1 = Yes and 0 = No	+/-
Farm size	Hectare	+/-
Age of cocoa trees	Years	+/-
Experience in cocoa farming	Years	+
FBO membership	1 = Yes and 0 = No	+
Credit access	1 = Yes and 0 = No	+
Extension contacts	1 = Yes and 0 = No	+/-
IP membership	1 = Yes and 0 = No	+
Distance to market	Kilometer	-
Land ownership	1 = Yes and 0 = No	+
CHP labour cost	Ghana Cedis	+/-

Results and discussion

Descriptive statistics of cocoa farmers

Table 2 summarizes the socioeconomic characteristics of cocoa farmers. About 78% of the farmers were male, with a higher share among the adopters (84%). The mean age of the cocoa farmers was 48 years with minimal variation across adopter groups, implying that cocoa farmers are generally younger than non-adopters, hence, more likely to adopt CHP. Overall, a typical farmer has spent 5.9 years in school with a slightly higher among adopters. Although, the differences in education were wide, indicates that adopters were more educated than their non-adopter counterparts. Generally, educated farmers increase adoption and hence, are more likely to adopt new technologies (Asante et al., 2014). Most of the farmers were family heads, specifically, higher adopters constitute 87.6% while non-adopters constitute 74.9%. The mean family size of cocoa farmers was 7.8 which is higher than the national household size of 5 persons (GSS, 2014). Out of the total family size, the mean family size for adopters was 8.4 while non-adopters were 7.5. Also, the overall mean economic active member was 2.9 out of which adopters was 3.1 and non-adopters was 2.7. The finding shows that large families denote labour availability which could possibly assist farmers in cocoa production activities such as weeding, pollination, spraying, pruning, harvesting and others (Zheng et al., 2021). It was evident that less than half of cocoa farmers (40.5%) engaged in off-farm activities. This was slightly higher among adopters constituting 41.2% compared to non-adopters with 39.6%. Overall, cocoa farmers earned about GHS1,962.52 as off-farm income. The difference among the groups was significant suggesting that adopters earned significantly GHS2,090.68 per month while non-adopters earned GHS1,811.05 per month. As expected, farmers who engaged in off-farm activities generated higher income (Anang, 2017). Additionally, the mean monthly income earned by cocoa farmers was GHS896.36 out of which the average income of adopters was GHS1,056.68 and non-adopters were GHS980.38. The results further demonstrate that less than half of the cocoa farmers (47.8%) owned land. The difference was not wide; thus, adopters and non-adopters represent 48.3% and 47.2%, respectively.

Table 2. Socioeconomic characteristics of cocoa farmers in Kejebril and Huni Valley, Western North Region, Ghana.

Variable	Adopters (N = 650)		Non-Adopters (N = 550)		Overall (N = 1200)		t-stat/X^2
	Mean	SD	Mean	SD	Mean	SD
Gender (Male)	0.840	0.367	0.727	0.446	0.788	0.408	−3.39***
Age (Years)	47.7	12.38	48.6	12.02	48.1	12.21	0.88
Years of schooling	6.4	4.9	5.2	4.8	5.9	4.96	−2.97***
Household head (Yes)	0.876	0.329	0.749	0.434	0.818	0.385	−4.09***
Household size	8.4	3.6	7.5	2.8	7.8	3.3	−1.67*
Economic active member	3.1	2.4	2.7	1.5	2.9	2.1	−2.40**
Monthly income (GHS)	1056.68	906.8	980.38	590.53	896.36	798.53	−3.53***
Off farm activity (Yes)	0.412	0.493	0.396	0.490	0.405	0.491	−0.39
Off farm income (GHS)	2090.68	1609.61	1811.05	1354.34	1962.52	1629.02	−0.82
Farm size (hectare)	4.24	3.8	3.11	2.71	3.76	3.39	−1.13***
Experience in cocoa farming	15.7	7.8	14.8	7.6	15.3	7.7	−1.51
Age of cocoa trees (Years)	14.2	5.7	13.2	5.2	13.8	5.5	−2.36**
Credit access (Yes)	0.243	0.429	0.225	0.427	0.235	0.428	−0.50
Amount received (GHS)	1160.83	836.66	912.09	182.25	1046.82	670.10	−0.64
IP membership (Yes)	0.270	0.733	0.258	0.438	0.265	0.615	−0.24
Land ownership (Yes)	0.483	0.500	0.472	0.500	0.478	0.499	−0.25
Extension access (Yes)	0.893	0.078	0.792	0.085	0.833	0.081	−0.16
FBO membership (Yes)	0.841	0.628	0.887	0.809	0.916	0.716	−0.92
Colleague farmers	0.75	0.199	0.707	0.290	0.735	0.290	−1.96**
Distance to market (km)	5.2	4.4	5.9	5.0	5.5	4.9	1.09

‘.’ used for p < 0.1. Source: Field data, 2021.

Furthermore, the mean experience of cocoa farmers is 15.3 years and varied minimally across adopter categories. More years of experience in cocoa farming suggests that farmers are knowledgeable and hence, can appreciate the benefits of the CHP technology (Obisesan, 2014). The average age of cocoa trees was 13.8 years, however, adopters had more aged cocoa trees (14.2 years) than non-adopters (13.2 years). This shows that, cocoa trees in Ghana are relatively mature compared to 15 years full potential bearing of cocoa trees (Kodom et al., 2022; Ministry of Food and Agriculture, 2021), hence, adoption of CHP to improve productivity is crucial. Adopters operated an average of 4.24 hectares of land which was significantly higher than that for non-adopters (3.11). Moreover, over 91.6% of the cocoa farmers were FBO participants and this varied significantly across adopter categories. Participation in FBO enhances farmer’s access to information on CHP. Almost all the cocoa farmers (83.3%) had extension contacts during cocoa production. Subsequently, about 23.5% of the cocoa farmers had credit with more of the adopters (24.3%) having access to credit than non-adopters (22.5%). Moreover, the mean credit amount received was GHS1,046.82 of which adopters received GHS1,160.83 whereas non-adopters received GHS912.09. About 26.5% of the farmers were members of IPs of which adopters and non-adopters constituted 27% and 25.8%, respectively. Further, farmers covered an average of about 5.5 km to both inputs and output markets.

Determinants of decision and intensity of adoption of CHP

The two-step Cragg estimates of the determinants of the adoption of CHP are presented in Table 3. This highly significant likelihood ratio test results indicate that incidence and intensity of adoption decisions did not occur simultaneously. Unsurprisingly, some variables may influence both decisions. The discrete decision is a probit regression model while the continuous is a truncated regression model. The age of a farmer, being a household head, years of schooling, farm size, age of cocoa farm and experience in cocoa farming influenced the incidence and intensity of adoption CHP.

Table 3. Determinants of adoption of CHP technique.¹

Variable	Probit estimates		Truncated estimates
	Coefficient	Std. Err.	Coefficient	Std. Err
Sex	0.057	0.040	0.069	0.103
Age (years)	0.036***	0.012	−0.073**	0.032
Marital status (married)	0.082*	0.043	0.263**	0.109
Years of schooling	0.094***	0.027	0.018**	0.007
Household head	0.129***	0.040	0.278***	0.104
Household size	−0.012**	0.005	−0.015	0.014
Economic active members	0.034***	0.008	0.023	0.021
Off-farm activity	−0.008	0.028	0.095	0.073
Farm size (hectare)	0.068***	0.018	0.028***	0.004
Age of cocoa (years)	0.078***	0.032	−0.015*	0.008
Experience in cocoa farming	0.013	0.024	0.017***	0.006
CHP labour cost (GHS)	−0.078***	0.013	−0.047	0.139
Credit access	0.044	0.032	0.264***	0.082
Distance to market (km)	−0.045***	0.013	−0.032	0.0291
Land ownership	−0.014	0.029	−0.047	0.139
IP membership	−0.012	0.023	0.011	0.057
Extension contacts	0.012**	0.005	0.014	0.015
Colleague farmer	0.039**	0.018	0.021	0.048
FBO	0.074***	0.016	0.039	0.043
Constant	0.883***	0.116	0.155***	0.024
Observations	1200		1200
Pseudo r-squared	0.366		0.771
Prob > chi2	0.000		0.000
LR chi2(19)	218.13		128.30
Log likelihood	−718.53		−1901.97

The probit estimates are a measure of probability of adoption and truncated estimates are a measure I the intensity of adoption of CHP. ‘.’ used for p < 0.1. Source: Field data, 2021.

The positive effect of age on the discrete decision to adopt suggests that older farmers are more likely to adopt CHP. Older farmers tend to be more experienced in terms of the benefits of adopting improved cocoa technologies hence, increasing the likelihood of adopting CHP. In addition, older farmers are more likely to have superior access to productive resources concerning younger farmers (Adebayo et al., 2021; Massresha et al., 2021). Conversely, due to the labour-intensive nature of the technology coupled with the generally low physical strength of older farmers, they tend to allocate smaller proportions to CHP. Younger farmers are more eager with adequate physical strength and hence, more likely to allocate larger areas to the CHP technique (Kodom et al., 2022). Also, younger farmers may have the capacity to intensify adoption as they are risk-takers and early adopters of new technology (Massresha et al., 2021).

Years of schooling increase both the incidence and intensity of adoption of CHP. The dissemination of any agricultural information with educated farmers increases the adoption of new technology (Danso-Abbeam & Baiyegunhi, 2018). Educated farmers are able to read and appreciate the benefits of CHP hence more likely to adopt the technology. The finding agrees with Adebayo et al. (2021) and Wossen et al. (2017) who asserted a positive relationship between education and adoption in Nigeria. Mostly, educated farmers have a high likelihood to be more appreciative of new ideas and innovations, hence culminating in a high adoption of improved agricultural technologies (Olawale et al., 2021; Oyawole et al., 2020).

Being a household head positively influences the probability and intensity of adopting CHP. Among rural farm households in Ghana, household heads are very dominant in most production decisions of the household including adoption of CHP. Also, the number of economically active members positively influenced CHP adoption, suggesting that the availability of such household members offers adequate labour for implementing the CHP adoption decisions. Experience in cocoa production has a positive influence on the intensity of adopting CHP, implying that an increase in years of experience in cocoa farming results in an increase in the intensity of adopting CHP technique. This is because experienced farmers have the ability and are endowed with the expertise to assess thoroughly, the benefits of new technologies due to their previous experience with previous or existing technologies (Avane et al., 2021; Obisesan, 2014). In addition, farm-level factors such as farm size increased both the incidence and intensity of adoption of CHP. This means that farmers with large farms are more likely to adopt CHP and allocate more land to the technology. Land is a superior precursor of smallholder cocoa farmers’ wealth, hence farmers with large farms are likely to adopt and allocate large areas to CHP in order to generate high income (Adebayo et al., 2021; Hui et al., 2019). Again, older cocoa trees have a positive significant influence on probability of adoption of CHP but negatively influence the intensity of adoption. Older cocoa trees (i.e. 14 years) are prone to low yields, hence enhancing the adoption of CHP (MoFA, 2021).

Furthermore, contact with extension agents has a positive significant influence on the probability of adopting CHP. Extension agents mostly assist the dissemination of improved technologies, good agricultural practices, market information, farm inputs, technical advice to others, hence frequent contact during cocoa production season could increase the adoption of CHP. In addition, FBO and colleague farmer as information sources influenced decision to adopt the technology positively and significantly. For instance, solicit information from FBO increased adoption by 7.4% while colleague farmer source increased the likelihood of adopting the technology by 3.9%. Most non-governmental agencies who support farmers may prefer extension agents and identify FBOs to disseminate their support packages to farmers (Kolade & Harpham, 2014; Kumar et al., 2020; Oyawole et al., 2020), hence the probability of a farmer being informed about the technology could increase adoption. Consequently, the credit access increased the intensity of adopting CHP. This result shows the importance of credit access suggesting that farmers are more likely to intensify adoption. The finding corroborates with Nwaru and Onuoha (2010) and Ammani (2012) that credit access improves farmers’ capacity to adopt improved technologies. Moreover, the negative influence of distance to market indicates a decrease in the probability of adopting CHP. For instance, with an additional distance of 1 km, farmers are less likely to adopt the CHP by 4.5%. Longer distances covered by farmers to markets discourage adoption, hence, a negative relationship between adoption and market distance (Avane et al., 2021). The results further show that the probability of adopting CHP is reduced by 7.8% with labour. The labour costs associated with CHP techniques reduces the likelihood to adopt the technology. As labour cost is a major component of the total cost of adopting which has the tendency of reducing adoption of improved technologies such as the CHP.

Determinants of impact of CHP on welfare

Table 4 presents the Full Information Maximum Likelihood Estimates (FIMLE) in the ESR for the impacts of adoption of CHP on welfare indicators (productivity, income and food insecurity). As discussed earlier, information sources such as colleague farmer, FBO membership and extension passed the validity test. Thus, these variables affected adoption, however, but did not affect productivity, income and food insecurity directly. Furthermore, the likelihood ratio test for joint independence (Table 4) indicates that the equations are dependent. The results of the covariance show that non-adopters may not have the same effect as the adopters (Lokshin & Zurab, 2004).

Table 4. Determinant of Impact of CHP on Welfare among farmers.²

Variables	Productivity				Household income				Food insecurity
	Adopters		Non-adopters		Adopters		Non-adopters		Adopters		Non-adopters
	Coeff	SE	Coeff	SE	Coeff	SE	Coeff	SE	Coeff	SE	Coeff	SE
Age	0.056***	0.018	−0.079***	0.024	0.052**	0.023	−0.088***	0.025	0.065**	0.029	−0.003	0.027
Gender	0.012	0.057	0.051	0.066	0.035	0.090	0.140*	0.072	0.497	0.711	0.025	0.535
Marital status (married)	0.190***	0.047	0.030	0.068	0.085	0.061	−0.015	0.067	−0.575	0.757	−0.228**	0.091
Years in schooling	0.088**	0.038	0.165***	0.056	0.072	0.050	0.204***	0.061	0.421	0.414	−0.160***	0.042
Household head	0.203***	0.061	0.305***	0.061	0.183**	0.080	0.261	0.679	−0.026	0.121	0.123	0.904
Household size	0.005	0.007	−0.007	0.010	0.019***	0.007	−0.010	0.010	0.142	0.099	0.102	0.092
Economic active	0.056***	0.018	−0.038	0.097	0.073***	0.018	−0.017	0.010	0.127	0.221	0.332	0.264
Off farm activity	0.110***	0.039	−0.083*	0.050	0.259***	0.045	0.216***	0.051	−0.373	0.433	0.318	0.419
Farm size (hectare)	0.039***	0.003	−0.050***	0.004	0.025***	0.003	−0.044***	0.004	0.025	0.035	−0.052	0.042
Years in cocoa farming	0.010***	0.003	−0.083*	0.044	0.020	0.047	−0.013***	0.004	0.017	0.037	−0.045	0.029
Age of cocoa tree	0.028	0.070	−0.074*	0.040	0.011	0.057	−0.034***	0.008	0.047	0.041	−0.040	0.054
Credit access	0.152***	0.049	−0.010	0.057	0.129**	0.056	0.034	0.065	0.262	0.565	−0.039	0.537
IP membership	−0.003	0.022	−0.045	0.055	0.011	0.022	−0.017	0.060	−0.251	0.175	−0.274	0.398
Land ownership	0.003	0.004	−0.002	0.006	0.031	0.048	−0.090	0.060	−0.118*	0.068	−0.012	0.061
CHP labour (GHS)	−0.229***	0.378	−0.039	0.056	−0.037**	0.016	−0.042	0.008	0.036	0.086	0.038	0.090
Distance to market (km)	−0.182	0.227	−0.210	0.201	−0.259	0.330	−0.065**	0.029	0.052	0.060	0.055	0.071
Constant	0.727***	0.244			−0.721***	0.244			−0.816***	0.252
athrho	−0.015	0.077			−0.039	0.088			0.058	0.572
lnsigma	−0.685***	0.020			−0.578***	0.028			0.152***	0.043
rho	−0.015	0.077			0.039	0.088			−0.058	0.572
sigma	0.503***	0.010			0.560	0.015			0.457***	0.197
lambda	0.070***	0.038			0.222***	0.049			0.267***	0.026
Observations	1200
Wald test of independent		84.33***			69.15***				13.67***

SE is the Robust standard errors. ‘.’ used for p < 0.1. Source: Field data, 2021.

Age of cocoa farmer shows a positive significant impact on all the welfare indicators – productivity, income and food insecurity for adopters. This means that older farmers are more likely to adopt new technologies that require less physical strength, hence, increasing productivity translates into higher farm income and reduced food insecurity. The finding connotes with Bellemare (2012) who found increased production through the adoption of technologies for older farmers. However, the results also indicated that age has a negative impact on productivity, income and food insecurity among non-adopters. This is because non-adopters believed that adoption of the CHP is labour intensive, hence older farmers are strength-constrained to adopt the CHP resulting in low productivity, income and poor food insecurity. The Gender variable had significant impacts on income for non-adopters. Education increased productivity, income and reduce food insecurity for adopters. This implies that through education, farmers accumulate knowledge and a better understanding of the technology, hence may increase productivity and income and translate to lower food insecurity (Umeh et al., 2022; Vogel et al., 2020). Furthermore, educated farmers are believed to typically have the ability to read and write as well as appreciate elementary advice about the CHP technique (Kotu, Alene, Manyong, Hoeschle-Zeledon, & Asamoah, 2017; Kassie et al., 2011). The positive significant effect of household heads suggests an increase in productivity and income among adopters and non-adopters, while household size also shows a positive significant impact on income. A typical cocoa farmer who is a household head with a number of economically active members, may generally influence household decisions to adopt hand pollination and hence, increase productivity and farm income. Farm size had negative and significant impact like the decrease of productivity and incomes for non-adopters, implying that farmers with smaller farm sizes are more productive (Chen et al., 2011). Conversely, farm size had positive impacts on high productivity and income for adopters.

Additionally, the results show that older cocoa farms result in low productivity and income for non-adopters. Mostly, the aged cocoa trees produced low yields (Aneani et al., 2011), hence farmers end up spending more on the cocoa farm.

This finding is consistent with Aikpokpodion et al. (2005) who argued that no use of improved technologies caused the aging of cocoa trees in Nigeria. The experience of cocoa farmer shows a positive and significant impact on productivity for adopters. Experienced farmers are expected to have good knowledge and expertise in farming that enable them to appreciate and understand the benefits of the technology better. Off-farm activities positively and significantly impacted productivity and income for adopters and non-adopters. Farmers engaged in off-farm activities are expected to earn incomes helping to meet the basic needs of the household as well as investing part of the income acquire farm inputs to enhance productivity. The finding corroborates the findings of Anang (2017) in Ghana, and Rakotoarisoa and Kaitibie (2019) in Kenya. In terms of productivity, the result shows a negative and significant impact of off-farm activity for non-adopters implying that farmers involvements in such activities does not increase cocoa yields. A possible explanation could be the fact that farmers may tend to spend less time on their farms which could affect crop performance. Institutional factors such as credit show a positive significant impact on productivity and income for adopters. The results imply that farmers with credit accessibility may increase productivity because they inject the credits received into cocoa production by purchasing necessary farm inputs such fertilizers, pesticides, etc. hence, translating to higher farm income. Similarly, other studies found that credit access may improve farmers’ capacity to adopt agricultural technologies in the long run and increase farm investment (Ammani, 2012; Nwaru & Onuoha, 2010). For adopters, the result shows that high cost of labour to assist in pollination decreases household income but increases productivity. This is not surprising because hand pollination is a physical process that is usually done manually, hence, more labour which affects farm income.

Estimation of average treatment effect on treated and average treatment effect of adoption of CHP on welfare

The ESR model further estimates the average treatment effects on treated (ATT) and average treatment effects (ATE). This shows the empirical impact of the adoption of CHP on welfare. The welfare indicators are productivity, household income and food insecurity presented in Table 5. The estimates of the ATT and ATE demonstrate that the adoption of the CHP technique led significantly to an increase in productivity and income, and reduced food insecurity. Adoption resulted in 4.32% (1054.72 kg per hectare) increase in productivity. Similarly, the adoption increased income by GHS5,368.41 (4.82%). However, adoption of the CHP reduced food insecurity by 4.4%. These findings suggest that the adoption of CHP is crucial to improving cocoa producers’ welfare. Our findings corroborate with other studies which found that the adoption of new technologies has a positive impact on farm productivity and household income in Africa (Abdulai & Huffman, 2014; Minten & Barrett, 2008).

Table 5. Impact of CHP on welfare among farmers.

Variable	ATE	Standard Error	ATT	Standard Error
Productivity (kg/hectare)	1163.52***	267.52	1054.72***	243.84
Household Income (GHS/hectare)	5556.03***	1057.19	5368.41***	1113.4
Food insecurity	−4.71***	1.06	−4.54***	1.01

Note: 64 kg is equivalent to 1 bag of cocoa. US$1 is equivalent to GHS6.32 on December, 2021. ‘.’ used for p < 0.1. Kg is kilogram. GHS is Ghana cedis. ATT is Average treatment effect on treated which means that the effect of treatment (hand pollination) on the treated subpopulation. ATE is Average treatment effect which implies that effect of treatment on a farmer randomly selected in the population. Source: Field data, 2021.

Robustness check with inverse probability weighted adjustment and propensity score matching methods

As discussed earlier, the IPWRA and PSM methods were used as a double robust check of the ESR model. The results are presented in Tables 6 and 7, respectively. The results show that adoption of the CHP improves productivity, household income and reduces food insecurity for adopters in Ghana. Accordingly, results show that yield and household income increased significantly by 4.72% and 2.74%, respectively, while food insecurity reduced by 3.3% (Table 6). The finding is consistent with Hailu et al. (2014) and Ahmed et al. (2017) who found a positive impact of the adoption of agricultural technologies on welfare.

Table 6. IPWRA results.

Welfare	ATT	Standard error
Productivity (kg/hectare)	565.76***	119.68
Household Income (GHS/hectare)	1229.69***	447.7
Food insecurity	−1.13***	0.34

Note: 64 kg is equivalent to 1 bag of cocoa. US$1 is equivalent to GHS6.32 on December, 2021. ‘.’ used for p < 0.1. Kg is kilogram. GHS is Ghana cedis. ATT is Average treatment effect on treated. Source: Field data, 2021.

Table 7. Average treatment effect from the propensity score matching.

Welfare	ATT	Standard Error
Yield
NNM	734.72***	221.44
RM	1098.88***	106.24
KBM	625.92***	154.88
Household Income
NNM	3827.50***	2150.35
RM	8015.62***	755.14
KBM	3145.23***	1352.07
Food insecurity
NNM	1.754***	(0.69)
RM	1.440 ***	(0.26)
KBM	1.663***	(0.46)

Note: 64 kg is equivalent to 1 bag of cocoa. US$1 is equivalent to GHS6.32 December, 2021. ‘.’ used for p < 0.1. NNM is nearest neighbour matching. RM is radius matching and KBM is kernel based matching. Source: Field data, 2021.

From the PSM estimates in Table 7, adoption of the CHP increased welfare indicators. For instance, productivity was significantly increased by 6.25 to 10.9%. Similarly, income increased by GHS3145.23 to GHS8015.62 while food insecurity by 1.44 to 1.75%. This concludes that adoption of CHP results in increasing cocoa farmers’ welfare. The findings agree with that of other studies (Etwire et al., 2022; Nakano et al., 2018; Wu et al., 2010). Keovilignavong and Suhardiman (2020) opined that farmland tenure security has a positive impact on food security in the rural uplands.

Figure 1 depicts the distribution of propensity scores among adopters and non-adopters of CHP. The figure reveals cocoa farmers who are on support and those off support. There was an even distribution of the propensity scores as the upper and bottom boundaries show the overlaps among adopters and adopters, respectively.

Figure 1. Histogram of propensity scores between treated and untreated groups.

Conclusion and policy recommendations

This study assessed the determinants and impact of adoption of CHP on the welfare of smallholder cocoa producers in Ghana. First, we examined the drivers of decision and intensity of adoption using the two-step Cragg model and evaluated the impact of adoption of CHP on the welfare indicators – productivity, income and food insecurity. Results showed that to increase the adoption and adoption intensity of CHP among smallholder cocoa farmers, efforts should target farmers with more years of schooling, family heads, with large farm sizes and have access to extension and advisory services. Furthermore, such efforts could be complemented by collaborative linkage with cocoa extension, cocoa research and non-governmental organizations to provide practical trainings on CHP to enhance adoption and ultimately improve productivity incomes and reduce food insecurity.

The positive impact of credit on CHP adoption suggests the need to enhance farmers’ access to credit. This could be more effective through strong linkage with organizations such as the rural development fund (RDF) which could provide credit to farmers through formal institutions to enhance their welfare. The empirical results of the impact of CHP on welfare found shows that cost of labour for the technology, age, education, family head, economic active members, age of the cocoa tree, off-farm activity, credit access and farm size determined farmers’ welfare. This affirms the potential direct role of the adoption of CHP in improving smallholder cocoa producers’ household welfare. Also, the impact results positively improved productivity and income by 4.32%, 4.82%, respectively and reduced food insecurity significant by 4.4%. The government and other stakeholders should, therefore, devise strategies such as trainings, demonstrations and subsidized labour to enhance the adoption of CHP and eventual increase in productivity, incomes and reduced food insecurity among smallholder cocoa producers.

Ethical statement

To ensure the confidentiality of the respondents’ information, the study was conducted in accordance with ethical research practices. Prior to taking part in the study, every participant gave their informed consent. It was made sure that respondents were fully informed of the nature and goal of the study and that their participation was voluntary. All information gathered for this study was kept private and anonymous. We appreciate the input from each and every respondent, and we are dedicated to responsibly and ethically using the research’s findings to advance knowledge in our field.

Acknowledgements

The authors are grateful to the smallholder farmers in the Huni Valley and Kejebril districts in the Western North of Ghana for their precious time and patience in contributing to the data collection for making this study successful. We express our deepest appreciation to all unknown reviewers for constructive suggestions to improve this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data is available upon request from the corresponding author.

Correction Statement

This article was originally published with errors, which have now been corrected in the online version. Please see Correction (http://dx.doi.org/10.1080/14735903.2023.2278947).

Additional information

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Notes on contributors

Bright O. Asante

Bright O. Asante is a Senior Lecturer at the Department of Agricultural Economics, Agribusiness and Extension of the Kwame Nkrumah University of Science and Technology (KNUST), Kumasi. He holds a PhD. Agricultural Economics degree from the University of New England, Australia; an MPhil. Agricultural Economics from the University of Ghana, Legon; and a BSc. Agriculture from KNUST.

Seth Etuah

Seth Etuah is a Senior Lecturer at the Department of Agricultural Economics, Agribusiness and Extension in the Faculty of Agriculture, Kwame Nkrumah University of Science and Technology (KNUST). He holds a PhD in Agricultural Economics from KNUST. Dr. Etuah has specialized in Agricultural Economics, Applied Econometrics, and Agribusiness Management.

Adams Faizal

Adams Faizal is a lecturer with the Department of Agricultural Economics, Agribusiness and Extension, Kwame Nkrumah University of Science & Technology, Kumasi, Ghana. He holds both MPhil and PhD in Agricultural Economics from the same University. However, part of Dr Adams' PhD education was carried out in Canada at Nova Scotia Agriculture College under the mentorship of Prof. Emmanuel Yiridoe.

Amos Mensah

Amos Mensah is a leading educator at KNUST's College of Agriculture and Natural Resources in Ghana. He completed his Ph.D. in Agricultural Economics at the Georg-August University Göttingen in Germany, and also earned a Master of Science degree in Tropical and International Agriculture from the same institution.

Stephen Prah

Stephen Prah is an MPhil student in Agricultural Economics at KNUST, Ghana, with a BSc (Hons) in Agribusiness Management. Certified in Rainforest Alliance, Fairtrade, and more, he specializes in production economics, climate tech, gender analysis, and agribusiness. He's a National Tutor for Greening Africa Together and an Agronomist at Holland Greentech Ghana.

James O. Mensah

James O. Mensah is an Associate Professor of Agribusiness Management at the Department of Agricultural Economics, Agribusiness and Extension of KNUST, Kumasi. He holds a PhD in Agribusiness Management from KNUST-Kumasi and an MSc. in Management and Economics of Innovation from Chalmers University of Technology, Gothenburg-Sweden.

Robert Aidoo

Robert Aidoo is an Associate Professor at the Department of Agricultural Economics, Agribusiness and Extension at the Kwame Nkrumah University of Science and Technology (KNUST), Ghana. He holds Ph.D and MSc. Degrees in Agricultural Economics from KNUST, and a Certificate in Agribusiness and Postharvest Management from the Galilee International Management Institute (GIMI), Israel.

Notes

1 The likelihood ratio test statistic was found to be greater than χ² critical value, therefore, the results from the Cragg two-step mode was used for the discussions.

2 In the ESR model, the instrument variables used are FBO membership, extension and colleague farmer. We assumed that these variables affect adoption decisions but does not affect productivity, income and food security not directly.

Appendix

A1: Test of validity of instruments selection

Information sources about cocoa hand pollination	Model 1	Model 2	Model 3	Model 4
	Adoption of CHP	Cocoa yield by non-adopters	Household income by non-adopters	Food insecurity by non-adopters
Extension	0.003**	0.058	36.947	0.001
	(0.001)	(0.037)	(23.668)	(0.009)
FBO	0.474***	2.960	1369.59	0.675
	(0.113)	(2.403)	(1539.778)	(0.546)
Colleague farmer	0.445***	−0.443	−2149.993	1.176
	(0.153)	3.151	(2019.551)	.73
Constant	−0.811***	23.109***	18582.738***	.339
	(0.174)	(3.537)	(2266.888)	(.811)
Wald test on information sources	χ2 = 21.35; p = 0.000 pseudo R^2= 0.0.20	F-stat = 1.277; Prob > F = 0.0.281	F-stat = 1.242; Prob > F = 0.294.	F-stat = 1.497; Prob > F = 0.214
Sample size	1200	550	550	550

Note: Model 1 denotes Probit model and Model 2, 3, 4 and 5 denote ordinary least squares estimates. Standard error in parenthesis. ***, ** and * are 1%, 5% and 10% significance level, respectively.

A2: Variance inflation factor

Variables	VIF	1/VIF
Age	1.368	0.731
Gender	1.525	0.656
Marital status (married)	1.149	0.870
Years in schooling	1.159	0.863
Household head	1.402	0.713
Household size	2.079	0.481
Economic active	1.851	0.540
Off farm activity	1.090	0.917
Farm size (hectare)	1.188	0.841
Years in cocoa farming	2.001	0.500
Age of cocoa tree	1.807	0.553
Credit access	1.072	0.933
IP membership	1.070	0.934
Land ownership	1.143	0.875
CHP labour (GHS)	1.026	0.975
Distance to market (km)	1.220	0.820
Extension	1.261	0.793
FBO	1.045	0.957
Colleague farmer	1.119	0.894
Mean VIF	1.346