Effects of digital agriculture solutions on smallscale wage workers and employment

Abstract

The study examined how the use of digital agriculture solutions affected farm worker pay and on-farm employment. Data gathered, was from 1,199 randomly sampled Small-scale farmers in the Bono East Region of Ghana. To account for potential selection bias resulting from unobservable factors that can influence the adoption of digital agriculture solutions, an endogenous switching regression model was employed. The results show that farmer demographic characteristics (age), farm characteristics (farm size and contract with employees), and institutional support services (group membership, access to training, access to credit and access to agricultural extension services) influence the adoption of digital agricultural solutions by Small-scale farmers in the Bono East region. We provide evidence that a high household income, a large farm size, access to information and participation in FBOs encourage farmers to employ more farm workers and pay higher wages. The study reveals that digital agricultural solutions have the potential to reduce the number of farm workers employed by 52% as well as reduce the amount of wages paid farm workers by 2%. The findings suggest the need for government and development partners to devise strategies that would increase the financial endowment of households and Farmers by making credit more available, expanding the availability of farmers productive resources like land, and increasing household income through livelihood diversification strategies. This move can enhance the use of digital farming technologies and empower farmers to increase on-farm employment and wages.

Keywords:

• Digital agriculture
• farm employment
• sustainable agriculture
• adoption
• wages
• working conditions

REVIEWING Editor:

• George Mudimu, Development Sciences, Marondera University of Agricultural Sciences and Technology, Marondera, Zimbabwe

SUBJECTS:

• Economics
• Development Economics
• Social Work
• Development Policy
• Economics and Development
• Sustainable Development
• Development Studies
• Agriculture & Environmental Sciences
• Agricultural Development
• Agricultural Economics

1. Introduction

According to Shang et al. (2010), digital agriculture entails utilizing computers and other electronic gadgets to boost the economic viability and sustainability of agricultural production. The development of digital agriculture is spread by the innovation of various solutions by agricultural technology (agritech) companies that aim at solving challenges that threaten sustainable agriculture. These solutions target improved access to information/low-cost extension services, financial services, early warning systems for crops and livestock health, and market information.

Digital agricultural solutions are functional in facilitating farmworkers’ and farmers’ cooperation, especially in developing nations into the global value chain and system of food production. This view is supported by the principle of the neoliberal theory that emphasizes the globalization and modernization of agricultural activities (Byerlee et al., 2009). Digital agriculture can promote the transformation of traditional production systems to modern and innovative production systems (Andrade-Sanchez & Heun, 2010). This agrarian transformation can reduce the cost of production, provide employment avenues for farm labour and increase farm income. Digital agricultural solutions can also reduce the problem of asymmetric market information (Zhu et al., 2011), thereby making farmers less susceptible to the opportunistic behaviour of other value chain actors. According to Shang et al. (2010), small-scale farms would become more productive, efficient, and sustainable if they adopted digital agriculture. Leveraging digital agricultural would therefore enable farmers to use labour, seeds, water, fertilizer, and pesticides more precisely (Rotz et al., 2019). This also means that adopting digital agriculture allows for flexible capital accumulation at the expense of labour and this could affect the demand for farm labour especially in developing countries where labour is abundant.

Digital agricultural solutions can affect wages and labour by streamlining and improving the labour processes in agriculture. This potential manifests as a decrease in the time and effort needed to complete specific farming activities. This is further expressed in a reduction in the amount of time and effort required to perform certain farming tasks. Also, some of these solutions have the potential to address labour shortages, particularly during the peak period of farming. Such solutions enhance farmers access to inputs which could adversely affect labour use by reducing the number of workers required to perform a specific farming activity (Ivus & Boland, 2015; McMichael, 2013; Pant & Odame, 2017; Weiss et al., 2010). Examples of the solutions available to farmers in developing countries like Ghana include solutions that provide agronomic advice, market information, weather information etc. despite this intervention it is unclear how they impact the society.

In Ghana, one of the most significant groups within the rural population is agricultural labourers, who do their labour on someone else’s farm in exchange for cash, kind, or a portion of the crop and cattle. In some cases these Labourers own small farms but doubles as a farm labourer to augment farm income. Available data indicate that more than half of the population receives the majority of their basic food and income from the agricultural sector, which also employs a sizable section of the labour force and contributes to the paid wage workforce (Asuming-Brempong, 2004). Agricultural jobs in Ghana are mostly concentrated in the rural areas (71%), compared with the urban areas (Ghana Statistical Service [GSS], 2018).

The introduction of digital agriculture in Ghana has the potential of simultaneously altering the system of agricultural production and the working conditions of farmers and farmworkers. However, to date, its effect on employment is uncertain. This is because in some cases it may lead to job losses due to the automation of farming activities which were hitherto done by labour. In other situations, it may expand agriculture production capacity, improve the skills of farmers and eventually increase their employment opportunities (Luo et al., 2023). Whatever the ramifications are, there is no doubt that technological innovation such as digital agriculture would shape the labour processes such that it would have different consequences for different types of workers (Prause et al., 2021).

The Bono East region, as part of efforts to increase agricultural production, has experienced the implementation of the Ghana Rural Telephony and Digital Inclusion Programme. In addition, in December 2021 a Digital Extension Project was implemented in the Bono East by Farmerline and funded by Adventist Development and Relief agency (ADRA) and United States Agency for International Development (USAID). The project aimed to provide cost-effective tailored digital solutions to ensure sustainable agricultural production. Other digital agriculture companies and start ups have rolled out various programs aimed at meeting the farming needs of farmers with the region. Furthermore, the Ghanaian government, through its Planting for Food and Jobs initiative in the region is making advances at providing a mobile-enabled digital identity for farmers. These digital interventions are expected to promote the adoption and usage of the solutions by the farmers.

There are some difficulties when analyzing how the use of digital agriculture technologies affects labour wages and employment from an econometric perspective. One is the observed heterogeneity that exists while the second is the possible endogeneity. The endogenous switching regression (ESR) model, according to Lokshin and Sajaia (2004), Kleemann et al. (2014), Ngoma (2018), and Senou and Manda (2022), corrects the selection bias in evaluating the treatment effects. Studies on the adoption of digital agricultural solutions have mainly estimated probit models and models for count data (Azumah et al., 2020; Isgin et al., 2008) to analyze the drivers and intensity of adoption of such solutions. Meanwhile, a composite understanding of the role of digital agricultural solutions requires an estimation of the potential effects of the adoption of such technologies on employment and wages paid to farm workers. The results of using digital agricultural solutions are shown in the ESR model, both in the real and hypothetical scenarios.

Though digital agriculture is growing rapidly in Ghana and the Bono East region, predictions over its effect on farm labour employed and wages paid remain unsettled. The fundamental questions arising from this process of adoption are (1) what are the significant differences between adopters and non-adopters of digital agriculture, (2) and how do those differences affect the wages paid and employment? Because endogenous switching regression can reduce biases from both observable and unobservable variables—a factor that other research did not take into account—it is considered appropriate to address these problems. This paper therefore seeks to fill these gaps by using the endogenous switching regression model to analyse the effect of the adoption and non-adoption of digital agricultural solutions on wages paid to farmworkers and the number of farm workers employed. By embarking on this study, it is expected that policymakers, development partners and civil society organisations will develop the necessary policies and advocacy that will improve and sustain the wages farmworkers receive as well as provide employment opportunities for the teaming youth through the digitalization of the agricultural sector.

A review of earlier research on the impact of digital agriculture on farmworker employment and pay is covered in the paper’s next part. The procedures for gathering and analyzing data are described in Section 3. The paper’s results and discussion are presented in Section 4, and Section 5 offers policy implications as a conclusion.

2. Literature review

The CTA and Dalberg Advisors (2019) report that almost 400 distinct digital agricultural solutions are available throughout the continent, serving 33 million enrolled farmers, indicating the market’s dynamism and potential. Some of these solutions in Africa range from access to market, credit, inputs, and weather to other agronomic services that enhance farming (Tsan et al., 2019). Most of these services in Ghana, according to Akorsu and Britwum (2022) are dominated by private providers with some few government innovations.

Researchers have focused a lot on the use of digital agricultural technology in agriculture and its effects on productivity. A number of these studies have shown how socio-demographic factors such as farm size, age, gender, education, status of marriage, and household income affect the adoption of agricultural technology (Abubakari, 2022; Adaku et al., 2023; Donkor et al., 2021; El Intidami & Benamar, 2021; Hirpa Tufa et al., 2022; Quartey et al., 2023). For instance, Jiménez et al. (2015) argue that higher educational attainment increases the ability for entrepreneurs to employ more. Also, Munćan and Božić (2017) argue that farm size and employment are positively related. Zhu et al. (2015), also found that older farmers pay less to their farm workers. According to some research, younger farmers are more open to embracing innovation than elderly farmers (Shang et al., 2021). There are also some studies that have shown differences between those who adopt innovations and those who do not. Ahmed and Anang (2019) for example identified that adopters and non-adopters differ significant among maize varieties in terms of marital status. Engwali et al. (2019), Ahmed and Anang (2019), Donkor et al. (2021), Quartey et al. (2023), Mhango (2023), Abubakari (2022), and Dey and Singh (2023), show that adopters of various innovations significantly vary from non-adopters in terms of farm size, age, education, gender, access to credit, production or ouput.

Research has also indicated that institutional elements such as farmer-based organizations, training, credit availability, and extension services could impact the adoption of agricultural technology (Abegunde et al., 2020; Ayanwale & Adekunle, 2008; Kudi et al., 2011; Kumar et al., 2019; Ma et al., 2023; Sulo & Chumo, 2012; Wossen et al., 2017). The is also significant evidence that differences between adopters and non-adopters in terms of group membership, training and access to credit are highlighted in the studies of Lawal et al. (2004), Rahayu and Day (2015), and Akudugu et al. (2023). Also, access to information improves the ability for farmers to access credit and adopt digital agriculture (Kofarmata & Danlami, 2019). This could enable him expand his and employ more farm labourer. In addition to the variables that could influence adoption or non-adoption of digital agriculture is the level of training farmers receive on digital agriculture. Y. Liu et al. (2022) argue that training programmes educate farmers on how to use an innovation and that has a significant impact on the adoption and non-adoption and their demand for labour. According to Bontsa et al. (2023), access to information enables farmers to organize and that increases the likelihood of adopting digital agriculture.

From the discussion so far, it can be observed that socio-demographic and institutional factors that influence adoption vary among adopters and non-adopters of various innovations across the world. These studies however have not focused on the factors that influence the adoption of digital agriculture which is a growing area in developing countries but how the adoption or non adoption due to these factors is impacting on the demand for labour and wages paid farm labourers. Additionally, since labour is an important factor that contributes significantly to agricultural production in developing countries, its effect on employment and wages paid to farm workers is equally important (Deininger & Xia, 2016; Herrmann, 2017; Ostermeier, 2020). Accordingly, increasing employment and wages could significantly contribute to poverty reduction. Globally several conflicting studies have been done on technology’s effect on workers earnings and employment opportunities. There are also few studies on how these effects vary among adopters and non-adopters of agriculture technology. Some of these studies have presented the positive and negative effects of agriculture innovation on employment generation and wages paid to farm workers. According to studies by Basok (2002), Acemoglu and Restrepo (2018), and Wang et al. (2021), adopting agricultural innovation enhances the efficiency of capital and reduce the competitiveness of labour, thereby causing large numbers of workers to be unemployed. Y. Liu et al. (2022) predicts that this influence, which was mostly argued during the Industrial Revolution, will manifest itself in the United States, the United Kingdom, and Japan. This shows that any innovation such as digital technology has the potential to reduce labour intensity and create unemployment. This would mean that the adopter of agricultural innovations tends to employ less and have a higher cost of production.

According to Birner et al. (2021), digital tools increase the profitability of farmers by allowing them to reduce or save labour costs and facilitate the employment of less-skilled labour for farming activities. The ILO in 2007 reported that wage agricultural workers (hired or permanent) are poorly paid, with wages generally below the poverty line and they form part of the core rural poor (ILO, 2007). With the adoption of agricultural technologies over the years, agricultural workers who earn less face several hazards at work ranging from toxic pesticides to dangerous and unsafe farming practices.

According to the literature thus far examined, the bulk of research papers concentrate on how digital agricultural solutions affect productivity (Alston, 2010; Alston et al., 2009, 2010; Fuglie, 2012; Pardey et al., 2010). A small number of studies have also examined the direct impact of innovation on social sustainability at the farm level (Ghazalian & Fakih, 2017; Harvey et al., 2017; Materia et al., 2017) while some have concentrated primarily on factors that influence innovation within the agri-food sector (Karafillis & Papanagiotou, 2011; Läpple & Thorne, 2019). However available literature on the effect of adoption of digital agricultural solutions on employment and wages are scanty, hence the need for a study of this nature.

3. Methods

3.1. Study area

The Bono East region of Ghana was the study’s location (Figure 1). The region occupies 33,654.54 km², or 10% of Ghana’s total land area, and is located in the country’s centre (GSS, 2016). Due to the region’s fertile soil and bimodal rainy season, which begins in June and September respectively, agricultural operations can be carried out there. Because most of the land is level, it can be used to grow crops including groundnuts, maize, and yams (GSS, 2016). The area also cultivates vegetables such as tomatoes, onions, okra, and garden eggs. Agriculture is the highest employer in the region, with a significant proportion of migrants forming part of its labour force. Additionally, some young people especially from northern Ghana choose this area above others for their agricultural pursuits (Sumberg & Okali, 2013). Notwithstanding the region’s reliance on agriculture, it is marked by high rates of illiteracy, poverty, and inadequate infrastructural development (Klutse et al., 2020; GSS, 2015). Due to the agrarian nature of the study area companies like Frameline, Esoko, Trotro Tractor, Tec Shelta, and AgroCenta find the region appealing for the application of agricultural policy and the deployment of digital agriculture solutions. The region was selected for the study because of the high concentration of agritech businesses there that provide smallscale farmers with a range of digital agricultural solutions, including market-oriented solutions, digital agricultural extension and climate advisory solutions, and digital finance solutions. Furthermore, the region is home to one Cyber lab and two Ghana Rural Telephony and Digital Inclusion Programme (GRT&DIP) sites. The Ghana Investment Fund is carrying out the project to enhance ICT access in underprivileged communities in the area through electronic communication. Furthermore, the region has seen the implementation of government programmes like Planting for Food and Jobs and the Block Farm, the success of which mostly rely on farmers embracing modern farming practices, which include embracing digital agricultural solutions.

Figure 1. A map of the Bono East region of Ghana. Source: Authors construct (2023).

3.2. Sampling and data collection

A threefold sampling method was employed to choose research participants. Six (6) districts in the region were purposively selected as part of the first stage (see Figure 1). The factors that were considered in selecting these districts include the dominant presence of digital agriculture companies as well as the active participation of non-governmental organizations in promoting agricultural development. With the assistance of agricultural extension officers in the various districts, fourteen (14) communities where the majority of activities are related to agriculture output were specifically chosen. These extension officers worked as field officers in the study. The next stage was to identify the household to interview. With the assistance of these extension officers households that engage in agricultural production were identified. This formed the sampling frame. Based on this sampling frame, 1,210 farmers were randomly selected for the survey. Households in each community were randomly selected and visited for data collection based on the corresponding sample size. During the data collection process households code were assigned to enable the field officers to do any follow ups in case of data gaps. Since there wasn’t much farming going on during the survey period, a majority of the selected farmers were interviewed at their residences. A farmer must be at least eighteen years old to be eligible to participate in the data collection unit. It is also required that these farmers have resided in the neighbourhood for the 2022 farming season. A total of 1,199 farmers agreed to be interviewed. Extension officers had training to administer the interview schedule after being hired as enumerators. The preparation and data collection took two months. After data was collected, the data was cleaned to check for errors and inconsistencies and follow-ups were made to fill data gaps. Ethically all personal information that will be traced to respondents during the data cleaning stage were removed to avoid results from been connected to any individual farmer. In addition, an ethical clearance was taken from the University of Cape Coast.

Data was gathered from the chosen Small-scale farmers in the Bono East Region using a survey questionnaire. The data gathered covered the traits of farmers’ homes and farms, the digital agricultural technologies and companies they were exposed to, and the availability of institutional support services. The survey questionnaire was configured in Computer-Assisted Personal Interviews (CAPI) using Kobo Collect. The data collection exercise consisted of three teams. Each team was composed of extension officers and field agents recruited for the data collection process. The authors and a few colleagues from the University of Cape Coast, Ghana, oversaw this approach. All study participants gave their informed consent before data collection and were guaranteed anonymity.

3.3. Analytical framework

Small scale farmers generally adopt digital agricultural solutions based on some expected satisfaction. As a result, the outcome of implementing these solutions can be predicted using a random utility framework. Through the works of Aakvik et al. (2005), we allow farmers to make a binary decision about the adoption of digital agricultural solutions, with the adoption of those solutions determining the utility of those solutions or not. Let this difference be denoted as $I_i^{*}=I_{1i}^{*}-I_{0i}^{*}$, where $I_{1i}^{*}$ is the utility derived from adopting the solutions and $I_{0i}^{*}$ is the utility derived from not adopting. The farmer will adopt the solution if $I_i^{*}>0$. Although a farmer’s utility cannot be observed, it can be deduced from the decision they make, which is illustrated by $I_i$, where $I_i\left(I_i\in \{0,1\}\right)$ is a binary variable., with $I_i=1$, for the treatment, i.e., the decision to adopt, and $I_i=0$ for non-adoption: $$\begin{array}{c}{I}_i^{\text{*}}=Z_i^{\prime }\alpha +{\epsilon }_i\\ I_i=1\text{\hspace{1em}if\hspace{1em}}{I}_i^{\text{*}}>0\\ I_i=0\text{\hspace{1em} if\hspace{1em}}\hspace{0.17em}{I}_i^{*}\le 0\end{array}$$

(1)

Where, $I_i^{*}$ is a binary variable dependent on a vector of the farmer’s observable traits Z and an error term ${\epsilon }_i$, with mean = 0 and variance = ${\delta }^2$.

The probability of ‘adoption’ is given by Equation (2)(2) $$\mathrm{\text{Pr}}\left(I_i=1Z_i\right)=\mathrm{\text{Pr}}\left(I_{1i}^{\mathrm{*}}>I_{0i}^{\mathrm{*}}\right)=\mathrm{\text{Pr}}\left(I_i^{\mathrm{*}}>0\right)=F\left(Z_i^{\prime }\alpha \right).$$(2) : (2) $$\mathrm{\text{Pr}}\left(I_i=1Z_i\right)=\mathrm{\text{Pr}}\left(I_{1i}^{\mathrm{*}}>I_{0i}^{\mathrm{*}}\right)=\mathrm{\text{Pr}}\left(I_i^{\mathrm{*}}>0\right)=F\left(Z_i^{\prime }\alpha \right).$$(2)

F is the cumulative distribution function for ${\epsilon }_i$. The estimating models are based on the functional form assumption for F. According to Greene (2012), in the event where the cumulative distribution function conforms to the conventional normal distribution, the probit model is applied. On the other hand, the logit model exists if the cumulative distribution function is logistic.

The study was interested in evaluating the impact of adoption on the number of people hired on the farm (employment) and wages paid (wages). In addition to the probability of adopting digital agricultural solutions, the study was interested in estimating the effect of adoption on the number of workers hired on the farm (employment) and wages paid (wages). The relationship between adoption and employment and wages is given by Equation (3)(3) $$Y_i=f(X_i;I_i).$$(3) : (3) $$Y_i=f(X_i;I_i).$$(3) where the variable of outcome is Y (employment and wages), X stands for the variables that explain., and I represents a binary variable for the decision to adopt. If $Y_i$ is the outcome variable (employment and wages) of individual i as a function of adoption I, Y can take two forms; $Y_{1i}$ and $Y_{0i}$.

Selection bias is still a major problem in research on effect assessments. Because untreated persons may self-select into treatment in the case of non-random therapy, it is only possible to estimate the average treatment on the treated (ATT). Following Lokshin and Sajaia (2004) and Akrong et al. (2022), the ATT is given by Equation (4)(4) $${\tau }_{\mathit{\text{ATT}}}=E\left(Y_1\mathrm{I}=1\right)-E\left(Y_{0}I=1\right),$$(4) : (4) $${\tau }_{\mathit{\text{ATT}}}=E\left(Y_1\mathrm{I}=1\right)-E\left(Y_{0}I=1\right),$$(4) where $\tau$ represents treatment effect (ATT), and E[.] denotes an expected value operator. The change in the outcome due to adoption can be specified as the difference between adoption and non-adoption (Akrong et al. 2022) Thus, the expected outcomes are employed to obtain unbiased estimates of the effects of digital agricultural solutions. In the literature on impact assessments, these figures are referred to as the ATT (Akrong et al. 2022; Lokshin & Sajaia, 2004).

Since randomization is not possible in this study, a quasi-experimental method was used to correct selection bias in the estimation of treatment effects. This method was the endogenous switching regression model (ESR) (Lokshin & Sajaia, 2004), which has been used in studies (Akrong et al., 2022; Kleemann et al., 2014; Ngoma, 2018; Senou & Manda, 2022). When decision-makers make a decision based on comparative advantage, selection bias occurs (Greene, 2012). This comparative advantage could stem from unobserved factors such as the farmers’ abilities and motivation. Regression approaches are employed when selection bias is caused by observable variables, such as income (Kleemann et al., 2014). On the other hand, selection bias emerges in the problem of omitted variables when it is caused by unobservable characteristics (risk aversion, natural technical or management aptitude) that affect both the adoption and the outcomes (employment and wage).

The ESR model uses two separate equations for adopters and non-adopters. Here, the model uses the inverse mills ratio to control the selection process via a selection equation in the initial step. An instrumental variable is included in the selection equation to avoid collinearity since the selection equation’s explanatory variables are included twice into the second stage estimation: once as a linear coefficient for employment and earnings and once as a non-linear one through the inverse Mills ratio.

. In this study, we use the group membership and ownership of a working radio as instruments. This is because digital agricultural solutions are introduced to the farmers in their groups, thereby increasing the propensity of a member to adopt them. However, being a member does not directly affect wages but only through the adoption of digital agricultural solutions. Similarly, access to information through ownership of a working radio increases the propensity that a farmer would adopt more advanced digital solutions that can provide production and marketing information. However, this does not directly influence the number of people employed.

The outcome variable is linked to two distinct regimes, contingent on the adoption choice, which is determined through the use of a probit model.

From Equations (1)$$\begin{array}{c}{I}_i^{\text{*}}=Z_i^{\prime }\alpha +{\epsilon }_i\\ I_i=1\text{\hspace{1em}if\hspace{1em}}{I}_i^{\text{*}}>0\\ I_i=0\text{\hspace{1em} if\hspace{1em}}\hspace{0.17em}{I}_i^{*}\le 0\end{array}$$ and (3)(3) $$Y_i=f(X_i;I_i).$$(3) , the two regimes for the decisions to adopt and not to adopt are given by Equations (5)(5) $$Y_{0i}=X_i^{\prime }{\beta }_0+{\epsilon }_{0i}\text{\hspace{1em}if\hspace{1em}}{I}_i=0$$(5) and (6)(6) $$Y_{1i}=X_i^{\prime }{\beta }_1+{\epsilon }_{1i}\text{\hspace{1em}if\hspace{1em}}{I}_i=1$$(6) : (5) $$Y_{0i}=X_i^{\prime }{\beta }_0+{\epsilon }_{0i}\text{\hspace{1em}if\hspace{1em}}{I}_i=0$$(5) (6) $$Y_{1i}=X_i^{\prime }{\beta }_1+{\epsilon }_{1i}\text{\hspace{1em}if\hspace{1em}}{I}_i=1$$(6)

$Y_0$ and $Y_1$ are the different outcomes for the two regimes of the decisions to adopt and not to adopt. ${\epsilon }_0$ and ${\epsilon }_1$ are the residual terms. Selectivity bias which is caused by unobservable factors can contribute to a correlation between ${\epsilon }_i$, and ${\epsilon }_0$, ${\epsilon }_1$. Standard regression methods, such ordinary least squares (OLS), would produce skewed results in these situations (Senou & Manda, 2022; Akrong et al., 2022). To address this, we derive the IMRs ${\lambda }_0$ and ${\lambda }_1$ which transforms Equations (5)(5) $$Y_{0i}=X_i^{\prime }{\beta }_0+{\epsilon }_{0i}\text{\hspace{1em}if\hspace{1em}}{I}_i=0$$(5) and (6)(6) $$Y_{1i}=X_i^{\prime }{\beta }_1+{\epsilon }_{1i}\text{\hspace{1em}if\hspace{1em}}{I}_i=1$$(6) into Equations (7)(7) $$Y_{0i}=X_i^{\prime }{\beta }_0+{\sigma }_{0i}{\lambda }_{0i}+u_{0i}\text{\hspace{1em}if\hspace{1em}}{I}_i=0$$(7) and (8)(8) $$Y_{1i}=X_i^{\prime }{\beta }_1+{\sigma }_{1i}{\lambda }_{1i}+u_{1i}\text{\hspace{1em}if\hspace{1em}}{I}_i=1$$(8) : (7) $$Y_{0i}=X_i^{\prime }{\beta }_0+{\sigma }_{0i}{\lambda }_{0i}+u_{0i}\text{\hspace{1em}if\hspace{1em}}{I}_i=0$$(7) (8) $$Y_{1i}=X_i^{\prime }{\beta }_1+{\sigma }_{1i}{\lambda }_{1i}+u_{1i}\text{\hspace{1em}if\hspace{1em}}{I}_i=1$$(8)

${\sigma }_0=\mathit{\text{cov}}\left({\epsilon }_{0,}{ \epsilon }_i\right)$ and ${\sigma }_1=\mathit{\text{cov}}\left({\epsilon }_{1,}{ \epsilon }_i\right)$ and $u_{0i}$ and $u_{1i}$ are the residual terms that follow a normal distribution with zero means. This work estimates the selection and outcome equations concurrently using the full information maximum likelihood (FIML) approach (Akrong et al., 2022; Lokshin & Sajaia, 2004). An endogenous switch occurs when either the correlation coefficient of ${\epsilon }_{0 }$ and ${ \epsilon }_i$, or ${ \epsilon }_1$ and ${ \epsilon }_i$ is statistically significant.

The ATT ${\tau }_{\mathit{\text{ATT}}}^{\mathit{\text{ESR}}}$ is given by: (9) $${\tau }_{\mathit{\text{ATT}}}^{\mathit{\text{ESR}}}=E\left(Y_{1}I=1\right)-E\left(Y_{0}I=1\right)=X^{\prime }\left({\beta }_1-{\beta }_0\right)+\left({\sigma }_{1i}‐\hspace{0.17em}{\sigma }_{0i}\right){\lambda }_1.$$(9)

3.4. Description of variables

Table 1 presents the variables considered in the econometric models and their hypothesized signs. According to the study’s hypothesis, adopting digital agriculture solutions is positively correlated with age. This is because older framers are risk averse and are less likely to adopt a completely novel technique (Achukwu et al., 2023). Also, compared with younger farmers, older ones do not have the time to experiment new technologies especially if they believe that the benefits of the technology may take longer to materialize (Lai-Solarin et al., 2021). As a result, we hypothesized that, younger framers are more likely to adopt digital agricultural solutions than older farmers. Also, wages, years of schooling, training and farm size were hypothesized to have a positive relationship with the adoption of digital agricultural solutions. This is because according to Y. Liu et al. (2022) and Achukwu et al. (2023) education and training makes it easier to have access innovations, and as such contribute to the frequency of adoption of modern technology. Also, farmers who pay wages are assumed to be wealthier and are willing to invest in improved services.

Table 1. Description of variables and hypothesized signs.

Variable	Description	Measurement	Sign
Age	Age of the farmer in years	Continuous	+/−
Ln(wages)	Log of wages paid to farmworkers in Ghanaian Cedis	Continuous	+
Migratory Status	Migratory status of the farmer	Dummy: 1 = migrant; 0 = otherwise	+/−
Household head	The farmer is the head of the household or makes Farming decision.	Dummy: 1 = Yes; 0 = otherwise	+
Ln(schooling years)	Log of number of years of formal education	Continuous	+
Ln(farm size)	Log of farm size in acres	Continuous	+
Gender	Sex of farmer	Dummy: 1 = male; 0 = female	+/−
Ln(income)	Log of total household income in Ghanaian Cedis	Continuous	+
Credit access	Whether the farmer had access to credit from a financial institution/farmer group/friend in the past 1 year	Dummy: 1 = Yes; 0 = otherwise	+
Group membership	Whether the farmer is a member of a farmer-based organization	Dummy: 1 = Yes; 0 = otherwise	+
Extension visits	Whether the farmer was visited by an extension officer in the past 12 months	Dummy: 1 = Yes; 0 = otherwise	+/−
Training	Whether the farmer participated in any training related to farming	Dummy: 1 = Yes; 0 = otherwise	+/−
Marital status	Whether the farmer is married	Dummy: 1 = Yes; 0 = otherwise	+
Employment	Number of waged workers employed	Continuous	+/−
Wages paid	Amount paid to farm workers in Ghanaian Cedis	Continuous	+/−

Source: Field survey, 2023.

Sex is a proxy of decision-making and resource accessibility in agricultural households in Ghana. Male farmers were hypothesized to adopt digital technology compared with their female counterparts. This is because according to Achukwu et al. (2023) female farmers face difficulty in accessing or acquiring farming resources in developing countries. In addition, we hypothesized that household income, access to credit, group membership and marital status positively influenced the adoption of digital agriculture. This is because studies from Donkor et al. (2021), have shown that an increase in household income promotes the likelihood of farmers adopting a technology since it come with it some cost elements. Abubakari (2022) also emphasized the positive role group membership has on agriculture technology adoption. This is true because being a part of a group builds social capital, which facilitates communication and trust (Miine et al., 2023).

The rate of adoption of digital agriculture was hypothesized to positively relate to extension visit that farmer received. This is because extension visits play a pivotal role in disseminating agriculture technologies to farmers (Agumagu & Nwaogwugwu, 2008; Xu et al., 2022).

4. Results and discussion

4.1. Descriptive statistics

Four categories were created from the many kinds of digital agriculture solutions that farmers used. Market-oriented solutions (price and market linkage) account for 54% of these; finance-oriented solutions (financial services, insurance, and payment platforms) account for 4%; input-access solutions (mechanisation and input linkage) account for 30%; and extension-oriented services (climate data, agronomic counsel, farm management, insurance, and veterinary services) account for 32%.

The study drew on past studies (El Intidami & Benamar, 2021) on the adoption of agricultural innovation to select variables that represent demographic factors, socio-economic factors, farm characteristic. The study also makes enough reference to institutional support services such as access to credit, training group membership that can influence farmers’ adoption of digital agricultural solutions. Table 2 presents the breakdown of the descriptive statistics across smallholder farmers. T-tests and chi-square tests were used to elucidate statistically significant differences in the characteristics of the sampled small-scale farmers in the Bono East Region of Ghana. In this study, we found that the features of adopters and non-adopters of digital agricultural solutions vary significantly in terms of age, size of farm, education, belonging to a farmer group, gender, credit accessibility, marital status, household income and access to training. This finding is consistent with the findings of Engwali et al. (2019); Ahmed and Anang (2019), Donkor et al. (2021), Quartey et al. (2023), Abubakari (2022), Mhango (2023), and Dash et al. (2023). Also, Small scale farmers who adopted digital agricultural solutions significantly paid higher wages to farmworkers than farmers who did not adopt any digital agricultural solution.

Table 2. Descriptive statistics of study variables.

Variable		Adopters (762)	Non-adopters (437)	t-Value/$\chi 2$ value
Outcome variables
Employment	Number of waged workers employed	7.083 (0.164)	5.073 (0.146)	8.234***
Wages paid	Amount paid to farm workers in Ghanaian Cedis	998.70 (45.018)	519.831 (31.972)	7.460***
Continuous variables
Age (years)	Age of farmer in years	41.257 (0.372)	46.114 (0.669)	6.878***
Schooling years (years)	Number of years of formal education	9.667 (0.183)	8.222 (0.243)	4.761***
Farm Size (acres)	Size of farm in acres	10.129 (0.383)	12.371 (0.464)	3.643***
Household income (Ghanaian cedis)	Total monthly household income in Ghanaian Cedis	1491.28 (40.748)	932.641 (18.040)	10.061***
Number of Extension contacts	Number of visits received from extension officers in the past 12 months	4.058 (0.166)	4.135 (0.338)	0.229
Categorical variables
Marital status (Yes)	Marital status of the farmer	83%	81%	0.445
Gender (Male)	Sex of farmer	72%	63%	10.348***
Contract (Yes)	Whether the farmer is a contract farmer	91%	61%	154.979***
Radio (Yes)	Whether the farmer has a working radio	75%	3%	570.28***
Credit Access (Yes)	Whether the farmer received credit in the past 12 months	88%	3%	813.567***
Group membership (Yes)	Whether the farmer is a member of a farmer-based organization	81%	2%	688.322***
Training (Yes)	Whether a farmer has received training on the use of digital agricultural solutions	83%	2%	734.178***
Access to off-farm income	Whether a farmer has access to off-farm income	59%	75%	31.724***
Household head	Whether the farmer is the head of the household	68%	63%	2.89*

Notes: *, and *** represent statistical significance at 10%, 5%, and 1% levels, respectively. Numbers in parentheses represent standard errors. Source: Field survey, 2023.

The study showed that adopters were relatively younger than non-adopters. Compared with younger farmers, older farmers are relatively risk averse.

Most Small-scale farmers who adopted digital agriculture technology had more years of schooling than non-adopters. Since education capacitates farmers to understand the requirements of technologies and also increases yield by reducing the cost of production, it would entice farmers with longer educational backgrounds to use cutting-edge technologies. The results show that household income earned by adopters was higher (GH₵1491.28) than income earned by non-adopters (GH₵932.641). This implies that higher-income households are better able to bear the expenses related to the adoption of agricultural technology than households with lower incomes.

Across gender lines, the results from the study show that there were more males in the adopter group (72%) than the non-adopters (63%). The differences in gender in terms of adopters and non-adopters of digital agriculture from the study were also noted in the study by Adaku et al. (2023) and Hirpa Tufa et al. (2022). These gender gaps in the study could be attributed to rural women’s limited access to productive resources and relatively lower income levels which further limit their propensity of adopting improved technology. These gender gaps could also limit their ability to employ more farm workers and the wages they pay to these workers. Additionally, the results indicate statistically significant variations in household head status between adopters and non-adopters. There were relatively more household heads who were adopters (68%) compared with non-adopters (63%). In terms of farmers’ access to off-farm income, our study results showed that most non-adopters (75%) had sources of income other than the farm income earned compared to adopters of digital agriculture solutions. A household’s resource endowment may rise with access to off-farm income, making it possible for them to pay for the adoption of agricultural technologies. This finding is consistent with the study of Diiro (2013), who found that off-farm income opportunities increased the use of improved seed due to competition in labour time. The study’s findings further demonstrated that more of the adopters had a formal contract with the farmworkers compared with non-adopters. Access to a working radio helps in receiving agronomic information which can increase the likelihood that a farmer would adopt digital agricultural solutions. The results show that more adopters (91%) of digital agricultural technology than non-adopters (61%) had access to a working radio than non-adopters.

Institutional support services such as membership in farmer-based organizations, training or capacity-building opportunities and agricultural credit can influence farmers’ decisions to adopt digital agricultural technologies (Lawal et al., 2004; Kudi et al., 2011). The results show a significant variation regarding access to credit. Non-adopters (3%) of digital agricultural solutions had less access to credit compared with adopters (88%). This means that adopters had more access than non-adopters. Studies from Gupta et al. (2020) and Ullah (2020) also shows that access to credit increases the resource endowment of Small-scale farmers, thereby enabling them to afford improved agricultural technologies (Gupta et al., 2020; Ullah, 2020). In addition, regarding institutional support, most adopters (81%) belonged to farmer groups that facilitate the use and adoption of digital agriculture while very few (2%) non-adopters belonged to farmer groups. This study is also in line with the findings Ayanwale and Adekunle (2008), Sulo and Chumo (2012), Wossen et al. (2017), Abegunde et al. (2020), and Kashiwagi and Kamiyama (2023). The differences in the group membership status of adopters and non-adopters are that by enrolling into groups, farmers would be supported through the provision of technological inputs such as digital agricultural tools.

Our study shows only a few (2%) of the farmers who did not adopt digital agricultural solutions received training while most (80%) of the adopters received training regarding digital solutions and other farming practices. The finding on receiving training is not different from other findings in other innovation studies such as Y. Liu et al. (2022). Available literature indicates that training on the various digital agricultural tools/solutions in rural communities educates farmers on how to use technology and introduces them to sophisticated agricultural strategies to obtain the highest economic benefit (Y. Liu et al., 2022) as compared to non-adopters.

4.2. Endogenous switching regression (ESR) estimates of determinants of adoption of digital agricultural solutions

The study used endogenous switching regression to analyze the impact of the adoption of digital agricultural solutions on wages paid to farm workers and employment or the number of farm labour employed during the farming season. The full information likelihood estimates for the determinants of adoption of agricultural solutions are presented in columns (1) and (4) of Table 3. The findings indicate that farmers’ adoption of digital agricultural solutions was substantially influenced by age, farm size, having a written contract with staff, owning a functional radio, having access to financing, being a member of farmer-based organizations, and having access to agricultural extension services. Age and the uptake of digital agricultural solutions have a negative correlation. This suggests that younger farmers were more likely to adopt digital agricultural solutions. This may be because younger farmers are more educated, and adventurous and have more time to explore new solutions. Moreover, the adoption of digital agricultural solutions was shown to be 2.4 percentage points higher among farmers with smaller farms. The results show a positive relationship between having a contract with waged workers and the adoption of digital agricultural solutions. This implies that farmers who had written contracts with their waged workers were more likely to adopt the solutions. The adoption of digital agricultural solutions had a positive relationship with having a functional radio. Ownership of a working radio increased farmers’ probability of adopting digital agricultural solutions compared to non-adopters of digital agriculture solutions. This could be because farmers who have a working radio may have access to agricultural information that could influence their interest in such solutions.

Table 3. ESR estimates of impacts of adoption of digital solutions on working conditions.

	Wages			Number of employees
Variables	Selection equation (1)	Adopters (2)	Non-adopters (3)	Selection equation (4)(4) $${\tau }_{\mathit{\text{ATT}}}=E\left(Y_1\mathrm{I}=1\right)-E\left(Y_{0}I=1\right),$$(4)	Adopters (5)	Non-Adopters (6)
Age	−0.017***	0.00253	0.0293***	−0.016***	0.0136	−0.039***
	(−0.006)	(−0.007)	(−0.009)	(−0.006)	(−0.019)	(−0.012)
Marital status	0.016	0.468**	1.113***	0.0271	0.818**	−1.058***
	(−0.203)	(−0.217)	(−0.304)	(−0.205)	(−0.356)	(−0.292)
Gender	−0.057	−0.738***	−1.721***	−0.006	0.129	0.636***
	(−0.203)	(−0.213)	(−0.236)	(−0.21)	(−0.315)	(−0.211)
Schooling years	−0.011	−0.031**	0.004	−0.007	0.0241	−0.0717*
	(−0.014)	(−0.013)	(−0.02)	(−0.014)	(−0.033)	(−0.039)
Farm Size	−0.024***	−0.0006	0.007	−0.030***	0.0912*	0.191***
	(−0.008)	(−0.003)	(−0.008)	(−0.011)	(−0.053)	(−0.051)
Logincome	0.255	1.396***	1.391***	0.197	1.467***	0.278
	(−0.161)	(−0.122)	(−0.238)	(−0.152)	(−0.283)	(−0.203)
Contract	0.340**	−0.317	−0.425**	0.317**	−0.839*	−0.343*
	(−0.158)	(−0.248)	(−0.196)	(−0.158)	(−0.443)	(−0.183)
Radio	0.771***	0.362**	0.563	0.784***
	(−0.212)	(−0.177)	(−0.439)	(−0.209)
Credit Access	1.558***	0.069	−2.460***	1.555***	−0.807*	0.625
	(−0.172)	(−0.206)	(−0.76)	(−0.172)	(−0.427)	(−0.948)
Group membership	1.277***			1.355***	0.515*	2.507
	(−0.213)			(−0.234)	(−0.306)	(−1.956)
Training	1.203***	0.283	2.954***	1.224***	−0.583	8.688***
	(−0.22)	(−0.226)	(−0.808)	(−0.236)	(−0.36)	(−2.689)
Access to off-farm income	−0.147	−0.665***	−0.741***	−0.104	−1.610***	−1.449***
	(−0.189)	(−0.133)	(−0.22)	(−0.186)	(−0.327)	(−0.352)
Household head	0.024	0.499**	1.340***	0.041	0.263	−0.429
	(−0.267)	(−0.212)	(−0.27)	(−0.265)	(−0.397)	(−0.283)
Number of Extension contacts	−0.019**	0.001	−0.003	−0.019***	0.010	−0.003
	(−0.0073)	(−0.009)	(−0.002)	(−0.007)	(−0.020)	(−0.005)
Constant	−2.266**	−3.794***	−5.523***	−1.970*	−3.46	5.075***
	(−1.09)	(−1.007)	(−1.653)	(−1.023)	(−2.193)	(−1.637)
Model diagnostics
Lns0	0.496***			0.597***
	(−0.044)			−0.093
Lns1	0.517***			1.448***
	(−0.042)			−0.0819
$\rho$0	−0.131**			0.146
	(−0.06)			−0.126
$\rho$1	−0.125			−0.355***
	(−0.111)			−0.0717
Wald test of independent equations $\chi 2$ (2)	6.52**			24.86***
	(0.038)			(0.000)
Observations	1,199				1,199

Notes: *, **, and *** represent statistical significance at 10%, 5%, and 1% levels, respectively. Numbers in parentheses represent standard errors. The outcome variable Wages is estimated in a natural log. The endogenous switching regression model’s Movestaty STATA command does not provide post-estimation of odds ratios or marginal effects.

Institutional support services including agricultural credit, training, and participation and engagement in farmer-based groups have a favorable impact on the uptake of digital agriculture solutions. Farmers who have received credit can afford the costs of adopting digital agricultural solutions. Those who had training may have gained knowledge on how to leverage digital agricultural solutions for enhanced productivity. Those who participate in groups are more aware of these technologies given the social capital derived from being a member of such groups. Those who had extension visits were less likely to adopt digital agricultural solutions as compared to non-adopters.

4.3. ESR estimates the determinants of the number of people employed and wages paid

Columns 2, 3, 5, and 6 present results on the determinants of the number of people employed and wages paid under the ‘adopter’ and ‘non-adopter’ regimes (Table 3). The estimated coefficients for some of the explanatory variables show different signs and magnitudes, a sign that the endogenous switching regression technique, which accounts for the variability between the two adoption groups, is favoured above straightforward treatment effects models. For example, the marital status of the farmer was positively related to the number of people employed by adopters and negatively related to the number of people employed by non-adopters. This implies that married people who are adopters of digital agricultural solutions employed more people whereas married people who were non-adopters employed fewer people. The results reveal that younger people who were non-adopters were more likely to employ more people. This may be because non adopters do not have the benefits that digital agriculture provides such as effective farm management, easy access to information that will limit labour use. While the adoption of digital agriculture provides information that can be used to analysis the potential need for labour demand, non-adopters do not have that advantage, so they may turn to employ more farm workers. Similarly, our study shows that males who were non-adopters were more likely to employ more people during a farming season than adopters. Furthermore, a new technology is associated with uncertain returns and upfront cost of adoption. We are argued that whether an individual adopts digital agriculture or not depends on the human capital and their knowledge of the changing technology. Since the adoption or non-adoption of digital agriculture influences employment, the level of education farmers had on their adoption needs to be assessed. Our results show that education was negatively related to employment by non-adopters. This finding is contrary to studies from Jiménez et al. (2015) who argue that higher educational attainment increases the ability for entrepreneur to employ more because educational attainment enlightens the entrepreneur to know the type of skills required at a time for higher productivity.

The findings demonstrate a positive relationship between farm size and employment for both adopters and non-adopters. This finding is consistent with the studies of Munćan and Božić (2017) who argue that farm size and employment are positively related. Our study also shows that adopters and non-adopters who had larger farm sizes were more likely to employ more people. This is intuitive since farmers with large farms would require more labour for cultivation. Income was positively associated with the number of people employed by adopters. Adopters who had a high household income were more likely to employ more people than non-adopters. A high income capacitates farmers to afford labour costs, thereby influencing them to meet the labour needs of the farm. Farmers who had contracts with their workers were less likely to employ more workers. This could be because the since the contract is binding and has certain demands to be meet which are not favourable, it would discourage adopters to employ. In addition, the contract normally defines rights and obligations such as hours of work, sick pay, and pay intervals and this act as a disincentive for work. In addition, adopters and non-adopters who had access to off-farm income had a higher propensity to employ more workers. This could be because access to off-farm income increases the financial resource base of the farmer which enabled him to employ farm workers.

The availability of institutional assistance programs influenced the employment of farm workers. Adopters who accessed credit were less likely to employ more workers as compared to non-adopters. This finding is consistent with the Ayagari et al. (2016) finding that increased access to finance results in increased job growth in developing countries. The results further show that adopters who belong to farmer-based organizations were more likely to employ more workers. This is because FBOs provide trainings on good agronomic practices and collective mutual labour activities (Abokyi, 2013). Non-adopters who had access to training on agricultural technologies were more likely to employ more workers. This is because training provide the farmer with skill, know how and experience that enable them to employ the required person in order to increase productivity (Mamaqi et al. 2011)

Regarding wages paid to workers, marital status positively influenced the amount paid to farm workers. This implies that married people were more likely to pay higher wages. This could be because marriage can increase household resource endowment which capacitates farmers to pay higher wages. We also argue that age differentials have a greater influence on the production function of farmers. Our study results show that older farmers who were non-adopters were less likely to pay higher wages. This is because older farmers understand the labor market dynamics and could negotiate better terms with workers. This finding confirms the studies of Zhu et al. (2015) who argue that older farmers tend to pay less for wages because they reduce the labour they employ. The negative relationship between gender and wages paid implies that males had a reduced propensity to pay higher wages. This could be because men have a higher bargaining power and could negotiate for reduced wages among their peers. This finding is similar with the studies of Tran (2021) who concluded that gender differentials could influence the input needed for production which could include a reduction in the cost on labour. The findings showed that there was a positive relationship between the wages paid and heading a household. Adopters who were household heads paid higher wages to farm workers. This implies that farmers who are decision-makers in their households can make decisions on how much of households’ financial resources should be allocated to farm labour. In addition, since household heads are mostly decision makers, the likelihood of them increasing farm size is high and as such might spend more on farm wages. This argument is supported by Aubron et al. (2022) and Dufty et al. (2019) who argue that, as farm size increased many household heads cannot provide all the farm labour requirement and such pay more wages to get farm activities performed.

Since education has an effect on the adoption of digital agriculture technology, the relationship between education and wage paid was also observed from the study among adopters. More educated farmers who were adopters were paid lower wages. This is because education would lead to technological change and this brings about a substitution of labour for capital. This leads to relatively lower wages paid for labour (Enu et al., 2014). The results show a positive relationship between income and wages paid. Intuitively, wealthy adopters and non-adopters paid higher wages to farm workers. This is consistent with Jelsma et al. (2019), who argue that wealthy farmers implement better agricultural practices which include reducing labour cost. Furthermore, income diversification is an essential livelihood strategy among small-scale farmers in low-income countries (Anang & Apedo, 2023). Finding form our study show that farmers who were able to make money off the farm paid their farm workers less wages. This could be because off-farm work, according to Ahmed and Melesse (2018), Anang et al. (2020) and Tenaye (2020), is meant to complement on farm income to enhance total farm earning as a such farmers would not want to reduce their earning by spending more on wages of farm employee (Sundar et al., 2019). According to Eaton and Shepherd (2001), contract farming is progressively becoming a fundamental part of a successful agri-business in both developed and developing countries. As such our study found out that non-adopters who had contracts with their employees paid lower wages. The results reveal a positive relationship between ownership of a working radio and wages paid by adopters of digital agricultural solutions. This implies that adopters who owned a working radio paid higher wages to farm workers. This is because the owning a working radio provide farmers to listen to community education on best farming practices which in some case will require the use of skilled labour that might increase the wages paid to these skilled farm workers.

Access to institutional support services showed alternate signs for non-adopters as to regards wages paid. For example, access to credit was negatively related to wages whereas access to training was positively related to wages paid. This implies that non-adopters who had access to agricultural credit paid lower wages. This finding is consistent with the finding of Franklin et al. (2020), who concluded that credit supply reduces the wages paid within firms. The results further show that farmers who had access to training on digital agricultural solutions paid higher wages to farm workers. This is because the training provides a range of skills and knowledge that can increase productivity and income generating capabilities (Wonde, 2022). This equips the farmers to spend more on wages paid to farmers.

4.4. The labour outcomes of the adoption of digital agricultural solutions

The study hypothesized that the adoption of digital agricultural solutions had labour implications. The simple mean differences in employment rate and wages paid given in Table 2 do not provide the true employment rate and wages paid attributed to the adoption of digital agricultural solutions because of the heterogeneity between adoption and non-adoption in both observable and unobservable characteristics as discussed above. Estimates of the ATT for salaries paid and employment rate provide a more strong influence of the use of digital agriculture technologies (Table 4). The ATTs illustrate the change in wages paid and employment rates after accounting for selection bias resulting from systemic variations in observable and unobservable variables between adopters and non-adopters, in contrast to the mean differences in Table 1. Table 4 displays the impact assessment’s findings and highlights the employment rates and wages paid under actual and counterfactual scenarios which are represented by the main diagonal (a and b) and off-diagonal (c and d) elements in the decision stage. The real and counterfactual outcomes’ row-wise discrepancies reveal the true causal impacts. The average treatment effect on the treated (ATT) is the difference between how many workers adopters employed (a) and how many workers non-adopters would have employed had they adopted digital agricultural solutions (c). The average treatment effects on the untreated (ATU) is represented by the difference between how many farm workers adopters would have employed had they not adopted (b) and how many non-adopters employed without adopting digital agricultural solutions (d). These definitions are true for wages paid to employees.

Table 4. Impact of adoption on employment and wages.

		Decision stage
Outcome variable	Sub-sample	To adopt	Not to adopt	Treatment effects
Wages	Adopters	(a) 6.042 (0.033)	(c) 6.172 (0.068)	ATT	−0.02*
	Non-adopters	(d) 5.072 (0.040)	(b) 5.190 (0.067)	ATU	−0.02
Number of waged employees	Adopters	(a) 7.084 (0.060)	(c) 14.621 (0.167)	ATT	−0.52***
	Non-adopters	(d) 7.865 (0.083)	(b) 5.073 (0.118)	ATU	0.55***

Notes: * and *** represent statistical significance at 10% and 1% levels, respectively. ATT ((a-c)∗100/c) and ATU ((d-b)∗100/b) include respectively, the mean treatment effect on the treated and the mean treatment effect on the untreated. ATT and ATU are the row-by-row variations between the decisions ‘to adopt’ and ‘not to adopt’ for the corresponding sub-samples. Standard errors are represented by numbers in parenthesis.

We discovered that the number of farm workers hired and the amount of wages paid are influenced by digital agriculture solutions after adjusting for confounding variables and counterfactual outcomes. The results reveal that digital agricultural solutions have the potential to reduce the employment rate and wages paid by 52% and 2% respectively. This finding is contrary to the findings of Taylor et al. (2012) who argue that changes in production technology would lead to better labour management which would make employers employ fewer workers and increase farm wages. The finding is however consistent with the argument that employee use technology to reorganize and reshape production process which may have consequences on wages and job quality. This implies that companies would rearrange the jobs that employees perform using technology, and employees would notice a decrease in the variety or complexity of the duties they perform, which employers could use as an excuse to pay them less (Hammerling, 2022).

5. Conclusion and policy implication

Examining how the use of digital agriculture solutions affected the number of farmworkers employed and wages was the goal of this study. We also intended to identify the distinguishing features between adopters and non-adopters of digital agriculture solutions. The study provides evidence that farmer characteristics (age), farm characteristics (farm size and contract with employees), and institutional support services (group membership, access to training, access to credit and access to agricultural extension services) drive the adoption of digital agricultural solutions by Ghanaian farmers. These drivers were found to vary among adopters and non-adopters of digital agricultural solutions. Our study reveals that a high household income, a large farm size, access to information and have been members of farmer-based organizations encourage farmers to employ more farm workers and pay higher wages. We provide evidence that the adoption of digital agricultural solutions reduces wages paid to farm workers and the number of farm workers employed by farmers.

Our findings show that the ascendancy of the neoliberal theory has contributed to agricultural modernization through innovations such as digital solutions. The adoption of these solutions has economic, social, and environmental implications. This study provides insights into the non-cost factors that can be considered in efforts to accelerate the adoption of digital agricultural solutions by farmers in underdeveloped nations. More specifically, food security-oriented initiatives such as the Planting for Food and Jobs (PFJ) which are implemented by the Government of Ghana can focus on improving access to training for farmers and their membership in farmer group organisation through capacity development and farmers’ organizing strategies. This move can be instrumental in encouraging the adoption of digital agricultural solutions for increased farm productivity.

The study shows that collective action, social capital, transaction-cost-reducing factors and wealth indicators including large farm sizes and higher household incomes are important in enhancing rural employment. Thus, achieving key sustainable development goals around decent work and economic growth (SDG8) relies on efforts that increase household income and access to agricultural production information by farmers. The study also shows that increasing modernization of agriculture can be detrimental to rural employment. This further reflects the substitution effect of modern agricultural technologies. However, given that a high household income and a large farm size strongly increase the number of people employed and the wages paid to hired labour, it is recommended that policymakers prioritize efforts to make it easier to obtain financial and productive resources. The former will capacitate the farmers to afford increased labour costs and the latter would induce farmers to increase employment to improve farm productivity. Moreover, the part that credit availability plays in the adoption of digital agricultural solutions presents avenues for public policies that can increase farmers’ access to credit. These policies enable farmers to afford the costs of adoption without inducing farmers to substitute labour for improved agricultural technologies solutions.

Ethical approval

Ethical approval was obtained from the University of Cape Coast Institutional review board. This is attached to the Manuscript. The ethical clearance number or ID is UCCIRB/CHLS/2023/22.

Authors contributions

The authors confirm contribution to the paper as follows:

1. Study conception and design: Licarion K Miine

2. Data collection: Licarion K Miine.

3. Analysis and interpretation of results: Licarion K Miine

4. Draft manuscript preparation: Licarion K Miine, Angela Dziedzom Akorsu, Owusu Boampong, Shaibu Bukari

5. Revising it critically for intellectual content: Angela Dziedzom Akorsu, Owusu Boampong, Shaibu Bukari

6. Final approval of the version to be published: Angela Dziedzom Akorsu Owusu Boampong, Shaibu Bukari

7. All authors reviewed the results and approved the final version of the manuscript.

Informed consent

Written consent was obtained from all Participant in the study. The authors adhered to strict confidentiality and anonymity was granted and included in the Ethical clearance process. At interview section participant were informed about the objective of the study.

Disclosure statement

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

Data availability statement

Data will be made available on request

Additional information

Notes on contributors

Licarion Kunwedomo Miine

Licarion K Miine is a PhD Student at the University of Cape coast. He has a Masters degree in Development Policy and Planning. He has an appreciable understanding in development policy and planning issues globally. Licarion research interest has been rural development, Environment, education, agricultural development and Labour studies.

Angela Dziedzom Akorsu

Angela Dziedzom Akorsu is an Associate Professor of labour and gender relations, and the Dean of the School for Development Studies at the University of Cape Coast. She holds a PhD from the University of Manchester in the UK. She has been engaged in research, post-graduate research supervision and outreach activities involving non-standard workers’ rights for over 20 years. Professor Akorsu’s scholarly research interests are around informal economy organising, gender and rural women’s livelihoods as well as decent work along agricultural value chains.

Owusu Boampong

Owusu Boampong is a Senior Research Fellow at the Department of Integrated Development Studies, University of Cape Coast, Ghana. His research interests are in the areas of urban informality, Precarity of work and collective organising. Email: [email protected]

Shaibu Bukari

Dr Shaibu Bukari had his first and second degree (Mphil Development Studies) from the University of Cape Coast, Central Region, Ghana. He also had his PhD in Social Work and Social Development from the University of Sussex. His research area of interest is labour studies. He is also the Head of department for Labour Studies at the University of cape coast, Ghana. He is currently working on workplace democracy on digital platforms. His hobbies include playing tennis and swimming.