How Does Digital Economy Promote Agricultural Development? Evidence from Sub-Saharan Africa

Abstract

Understanding the impact of the digital economy on agriculture in developing countries holds significant importance in enhancing agricultural production and addressing hunger-related challenges. This study uses panel data of 35 SSA countries from 2006 to 2021 and investigates the relationship between digital economy and agriculture using dynamic regression models. The impact mechanism is also analyzed using mediating models. The results show that the digital economy has a positive impact on agriculture through the increase of agricultural productivity, human capital accumulation and the improvement of government governance. The effects of the digital economy are larger in countries with higher and lower levels of agricultural development, significant in middle-income countries, and insignificant in low-income countries. This study may provide a better understanding of the nexus between digital economy and agricultural development, and offer valuable insights for governments in developing nations to formulate relevant and effective policies.

1. Introduction

Agriculture is the mainstay of most sub-Saharan Africa (SSA) countries, accounting on average for about 20% of GDP. In some countries, such as Sierra Leone and Chad, it even exceeds 50% (World Development Indicators, 2023). Up to half of the labor force is employed in agriculture-related activities, making it the main source of livelihood for many people. However, the performance of agriculture in many SSA countries lags far behind the world average. Since the independence of African countries in the 1960s, issues such as the mono-product economies, inadequate government financial investment, low levels of agricultural technology, weak infrastructure and poverty, have not only affected food availability, but also hindered the elimination of hunger and food insecurity in SSA. Although SSA has a large amount of arable land, a demographic dividend and other abundant natural resources, its food production capacity and ability to cope with food crises are limited. In addition, land use efficiency is low in many SSA countries and the growth rate of agricultural production is not keeping pace with population growth. Nearly 80% of the agricultural land and production is rain-fed, making the productivity very low to meet the needs of a growing population. The self-sufficiency rate for major cereals in SSA is less than 80%, well below the world average (FAO, 2023). SSA is particularly affected by the world’s greatest food security challenge, with one-third of the world’s food-deficit population by 2021. Relying on the traditional agricultural resources makes it difficult for the SSA to feed itself on its own. Digital tools have become an important factor in modern agricultural production and are expected to transform agriculture [1].

The world is in a trend of a rapid expansion of the digital economy, with great advancement in the information and communication technology (ICT) sector [2]. The development of ICT has dramatically revolutionized modern society, such as how people work and communicate. The momentum of new technologies, including ICT, provides an indispensable engine for economic growth and prosperity, in both developed and developing countries [3]. According to statistics from the International Telecommunication Union [4], the usage of ICT has shown a significant upward trend worldwide. In 2022, there were approximately 5.3 billion people using the internet, accounting for nearly 66 percent of the world’s population. There are 86.9 and 108 mobile broadband and mobile phone subscriptions per 100 people, respectively, which is 3.2 and 1.2 times higher than a decade ago. SSA is catching up in the adoption of ICT as prices and access charges for ICT use are falling [5]. Although the number of households with access to the internet, computers and telecommunications in SSA is still much lower than in developed countries, it has increased at an unprecedented rate since the 21st century. However, due to inadequate infrastructure, the high cost of the ICT price basket and unreliable energy supply, the trend of digitization in SSA is still slow and has much room for development.

From mobile applications and text messaging to satellite imagery and remote sensing, ICT has provided multiple channels for delivering agricultural information. The structure, organization and efficiency of agricultural markets can be transformed and information asymmetries can be addressed by ICT, thus reducing the incidence of market failures that are common in agriculture [6,7]. The application of ICT throughout the agricultural chain helps to connect producers and consumers and improve market integration, so that agribusiness is taken out of the farm gate and integrated into global supply chains [8]. The use of ICT helps farmers to find appropriate technologies in the production process, accelerate productivity and reduce the distance between farm and market [9]. In SSA, agricultural production is dominated by smallholder farmers, whose farms are generally small, fragmented and located far from cities. Poor access to information has led to high production costs and low technology, and has even isolated some smallholder farmers from the modern agricultural market. It is high time for SSA to embark on a digital agriculture revolution, as it could realize the progress of the agricultural sector.

Many scholars have focused on the development of the digital economy and its socio-economic effects in SSA [10,11]. Samimi and Ledary [12], Hofman and Aravena [5], Mbuyisa and Leonard [13] and Myovella and Karacuka et al. [14] hold that ICT can reduce transaction costs through interconnectivity, and create favorable conditions for economic development, investment and employment. Many researchers have confirmed a positive relationship between ICT and financial development [15,16,17,18]. The relationship between the digital economy and international trade has also been of interest to scholars, most of whom have found that digital technologies have a positive impact on trade [19,20,21,22]. In addition, researchers have also investigated the impact of the digital economy on eliminating the informal economy [23,24,25] and protecting the environment [26], as well as the determinants of the digital divide [27,28] in SSA.

However, the literature is scarce on how the digital economy affects agriculture in SSA. This study fills a cavity in the literature by contributing in three aspects. First, this paper empirically examines the impact of digital technology on agriculture in SSA. As an emerging technology, digital technology may have a greater marginal impact on economic development in underdeveloped countries, especially in SSA, where the development of agriculture is crucial for economic transformation and upgrading. Second, we believe this paper is fresh to examine the mechanisms and pathways of the impact of digital technology on agriculture in SSA by using multiple mediating regression methods. Third, this paper conducts heterogeneity analysis to examine whether the impact differs in countries with different income and natural resource endowments.

The remainder of this paper is organized as follows. Section 2 presents the conceptual framework. Section 3 presents the model, methodology, and data. Section 4 reports and analyzes the empirical results. Section 5 provides further analysis. Section 6 provides discussions. Section 7 concludes this study and draws policy implications.

2. Conceptual Framework

2.1. Digitalization, Agricultural Productivity and Agricultural Development

Smallholder farming in SSA tends to result in a fragmented market and low efficiency. However, digital economy can improve the situation by integrating the market in space and time. Digital technologies such as ICT have been researched to change farmers’ production mode and decisions, and improve farmers’ participation in e-commerce, thus providing farmers with avenues to sell their products [29]. Surplus products are converted into monetary income, which has stimulated farmers’ enthusiasm for production, activated the vitality of the agricultural market, and improved the overall competitiveness of agriculture. ICT can also integrate the link between agricultural research and development (R&D), extension, and farmers. The provision of information can be facilitated with rapid response as most of the relevant stakeholders are on the ICT platform. New technologies invented by universities, colleges and research and development (R&D) organizations can be delivered to extension workers more conveniently, and farmers can provide feedback more quickly through ICT. Therefore, the overall technological level of agriculture is improved.

ICT can reduce information asymmetry and make agriculture more productive [30]. Price information is more transparent between farmers and traders, thus making market transactions more accessible. For example, several information platforms in SSA countries, such as the Sector Information System, the Agricultural Market Information Center and the Livestock Market Information System, etc., have public and non-excludable services about price information. In addition, commercial entities have also provided more exclusive or private services that form a useful complement to the public services. In this way, market performance and efficiency are improved by matching supply and demand.

Existing studies have concluded that ICT facilitates the efficient use of inputs [31]. Farmers can find information on ICT when and where possible to obtain services on traditional agricultural production suggestions, such as the supply of seeds, fertilizers, pesticides, and farm machinery services. Digitalized information such as moisture content, pest outbreaks and weather forecasts are also sent to farmers. Besides, other information about farm management, crop monitoring, and supply chain management can be found through ICT. Moreover, unexpected shocks are common in SSA, especially in conflict-affected regions. Market actors can be protected from these shocks and improve households resilience through the use of ICT tools [32].

Market integration and the reduction of information asymmetry through ICT will contribute to the improvement of agricultural productivity in SSA and thus to agricultural development. In this regard, we predict Hypothesis 1:

Hypothesis 1.

The digital economy improves agricultural production, and one of the impact mechanisms is by increasing agricultural productivity.

2.2. Digitalization, Human Capital and Agricultural Development

Human capital, other than physical capital, has become one of the most competitive resources in modern agriculture. It is not only an important factor in production, but also a key carrier of the accumulation of knowledge, experience and technological innovation. The digital economy has stimulated investment in human capital and optimized its structure. First, the digital economy has stimulated investment in education. It has reduced the cost of acquiring new knowledge and technology for labor, and improved their ability to absorb new knowledge and technology [33]. The rapid development of online education is a good example. Digitalization and networking have reduced the cost of household investment in education and improved the return on investment in education. Second, the digital economy has encouraged enterprises to invest in human capital. The application of digital technologies has reduced enterprises’ training costs, increased the number of trainees in professional training and retraining, and improved the level of human capital [34]. Third, the digital economy has facilitated human capital flows. Advances in ICT have greatly facilitated the dissemination of information, thereby optimizing the allocation of human capital in the labor market. Networking and digitization have shortened the physical distance between places of origin and destination, and reduced the psychological costs of mobility, making the flow of human capital more fluid. Fourth, ICT contributes to the diffusion of knowledge through the “learning by doing” effect. Human capital contains a certain amount of tacit knowledge. Advances in ICT have promoted the diffusion of this implicit knowledge, the improvement of producers’ efficiency and, indirectly, the development of agriculture.

The literature has confirmed the impact of ICTs on human capital accumulation for agricultural producers in SSA. For example, Ochieng and Okello [35] discovered that farmers who participate in ICT gain access to a wider range of knowledge and increase the use of improved seeds and fertilizer in agricultural production, thereby improving productivity. Arouna and Michler [36] found that smallholder farmers who use mobile applications receive personalized advice on agronomy and nutrient management. Their knowledge and skills of fertilizer application are improved, thereby increasing the rice yields and profit in Nigeria. Silvestri and Richard [37] also show that by receiving information through radio and mobile tools, farmers become more aware of new agricultural practices such as land preparation, sowing, planting, fertilizing, weeding and harvesting. The adoption of sustainable and intensified technologies is boosted. Suri and Jack [32] show that the digital economy facilitates the flow of labor as it provides alternative jobs for famers to engage in business and increase family income, which then feeds back into agriculture. In this regard, we predict Hypothesis 2:

Hypothesis 2.

The digital economy improves agricultural production through human capital accumulation.

2.3. Digitalization, Governance Quality and Agricultural Development

Countries with weak institutions, such as weak legal systems, non-transparent administrative practices and corruption, hinder the necessary investment in agricultural production [38]. This problem is greatly ameliorated by ICTs. First, ICT-enabled e-government can facilitate public access to government services. The improvement in communication capabilities and services delivery can make governance more efficient and effective [39]. Governance can be made cost-effective and convenient through the use of ICTs, ensuring an open and transparent political environment. Government services will be more easily accessible to citizens and public trust will be enhanced [40]. Second, ICT has also been considered to combat corruption [41]. The improved transparency of government transactions increases public visibility and enhances oversight. The cost of corruption is raised as corrupt officials face a higher risk of being caught [42]. The behavior of government officials and projects can be more easily monitored and controlled by a wide range of citizens through ICT. This can promote the development of the rule of law and create a fair political environment. Third, ICT also contributes to perfecting the democratization process by building good governance and simplifying bureaucratic procedures. On the one hand, ICT can enhance the proper participation of the public in governance, and suggestion of governance proposals and new ideas providing a reference for policy-making. On the other hand, the delivery of public services such as e-education, e-health, inspection, etc., will be improved [43].

The improved quality of governance through the use of ICT tends to produce policies that support agricultural development, which are conducive to the improvement of new agricultural technologies and innovation. ICT connects SSA to the world, and a good institutional environment will attract more foreign investment and make up for the lack of capital in African agriculture. It is also conducive to the development of agricultural infrastructure. Rain-fed agriculture leads to low yields in SSA, while better governance will be able to mobilize factors of production such as labor, capital and technology to build infrastructure such as irrigation, rural roads and electricity, thereby improving overall agricultural production capacity. In this context, we predict Hypothesis 3:

Hypothesis 3.

The digital economy is conducive to agricultural development through improvements in the quality of government governance.

The mechanism of impact is shown in Figure 1.

3. Methodology and Data

3.1. Equation Specification

We consider that ICT can affect agricultural development in SSA. In exceptions for the key variable, agricultural inputs, economic growth, urbanization and foreign capital are often identified as key factors according to Bunje and Abendin [44] and Myovella and Karacuka [14]. We construct the multivariate framework as follows:

$${agd}_{it}=f\left({de}_{it},{pgdp}_{it},{land}_{it},{ind}_{it},{fdi}_{it}\right)$$

(1)

where agd represents agricultural development; de represents the level of digital economy; pgdp denotes gross domestic product (GDP) per capita; land denotes arable land per person; ind denotes the industrial upgrading; fdi denotes foreign development investment; i represents 35 SSA countries; t denotes years from 2006 to 2021.

In order to mitigate the instability of data and to remove possible heteroskedasticity, each variable is transformed into a logarithmic specification. In addition, considering the possible lag of the level of agricultural development, the first-order lagged term of agd is introduced into the equation. Equation (1) is then rewritten as follows:

$${lnagd}_{it}={\beta }_0+{\beta }_1{lnagd}_{i,t-1}+{\beta }_2{lnde}_{it}+{\beta }_3{lnpgdp}_{it}+{\beta }_4{lnland}_{it}+{\beta }_5{lnind}_{it}+{\beta }_6{lnfdi}_{it}+{\epsilon }_{it}$$

(2)

where ${\beta }_1-{\beta }_6$ represent the undetermined coefficients of the variables; ${\beta }_0$ represents the intercept term; and ${\epsilon }_{it}$ is an independent and identically distributed error term.

It is possible that agricultural development and the digital economy may cause each other, leading to a two-way causal effect. The improvement of digital economy promotes the development of agriculture, while the latter attracts more capital for digitalization, which leads to the problem of endogeneity. This study selects the Generalized Method of Moments (GMM) method to obtain unbiased estimators. Two types of GMM methods are generally used in the literature, namely the first-difference GMM and system GMM. The system GMM method allows for a richer set of instruments and tends to be more efficient, but only if the cross-section unit is sufficiently large. Therefore, the difference GMM is adopted in this study. On the other hand, after the transformation of the first-difference, value may be lost and the gaps in the sample panel may be widened. However, the orthogonal-difference GMM, developed by Arellano and Bover [45], can effectively deal with this problem. This type of GMM can include more data information by simply subtracting the average of all future available variable observations. Therefore, the orthogonal-difference GMM is applied as the benchmark regression.

The study further investigates the mediating effect of agricultural productivity, human capital and government governance. According to Baron and Kenny [46], the mediating models are written as follows:

$${lnatfp}_{it}={\gamma }_0+{\gamma }_1{lnatfp}_{i,t-1}+{\gamma }_2{lnde}_{it}+{\gamma }_3{lnpgdp}_{it}+{\gamma }_4{lnland}_{it}+{\gamma }_5{lnind}_{it}+{\gamma }_6{lnfdi}_{it}+{\sigma }_{it}$$

(3)

$${lnagd}_{it}={\delta }_0+{\delta }_1{lnagd}_{i,t-1}+{\delta }_2{lnde}_{it}+{\delta }_3{lnatfp}_{it}+{\delta }_4{lnpgdp}_{it}+{\delta }_5{lnland}_{it}+{\delta }_6{lnind}_{it}+{\delta }_7{lnfdi}_{it}+{\rho }_{it}$$

(4)

where lnatfp represents the mediating variable of agricultural total factor productivity (TFP). The mediating impact of agricultural productivity can be tested by Equations (2)–(4).

$${lnhc}_{it}={\phi }_0+{\phi }_1{lnhc}_{i,t-1}+{\phi }_2{lnde}_{it}+{\phi }_3{lnpgdp}_{it}+{\phi }_4{lnland}_{it}+{\phi }_5{lnind}_{it}+{\phi }_6{lnfdi}_{it}+{ϵ}_{it}$$

(5)

$${lnagd}_{it}={\theta }_0+{\theta }_1{lnagd}_{i,t-1}+{\theta }_2{lnde}_{it}+{\theta }_3{lnhc}_{it}+{\theta }_4{lnpgdp}_{it}+{\theta }_5{lnland}_{it}+{\theta }_6{lnind}_{it}+{\theta }_7{lnfdi}_{it}+{\pi }_{it}$$

(6)

where lnhc represents human capital, one of the mediating variables and its mediating effect can be tested by Equations (2), (5) and (6).

$${lngov}_{it}={\alpha }_0+{\alpha }_1{lngov}_{i,t-1}+{\alpha }_2{lnde}_{it}+{\alpha }_3{lnpgdp}_{it}+{\alpha }_4{lnland}_{it}+{\alpha }_5{lnind}_{it}+{\alpha }_6{lnfdi}_{it}+{\omega }_{it}$$

(7)

$${lnagd}_{it}={\tau }_0+{\tau }_1{lnagd}_{i,t-1}+{\tau }_2{lnde}_{it}+{\tau }_3{lngov}_{it}+{\tau }_4{lnpgdp}_{it}+{\tau }_5{lnland}_{it}+{\tau }_6{lnind}_{it}+{\tau }_7{lnfdi}_{it}+{\vartheta }_{it}$$

(8)

where lngov represents government governance and Equations (2), (7) and (8) can be used to test its mediating effect.

If the coefficients of the lnde in Equations (3), (5) and (7), and the coefficients of agricultural productivity in Equation (4), human capital in Equation (6) and government governance in Equation (8) are significantly above 0, the mediating effects of the three mediators are significant.

3.2. Variable and Data Sources

The study applies a balanced panel data of 35 SSA countries from 2006 to 2021 considering the availability of data. The measurement and sources of the selected variables are as follows:

3.2.1. Dependent Variable

Various indicators have been used in the literature to proxy agricultural development, such as agricultural value added [47], crop yield [48], and use of improved seed varieties [49]. However, food production capacity is low and SSA is currently facing food shortage. Food production is not increasing at the same rate as population growth. Ensuring adequate food supply for the large population is a top priority for agricultural development in SSA. Therefore, this study uses the food production index as a proxy for agricultural development. The data are obtained from the World Development Indicators (WDI). The index includes commodities that are considered edible and nutritious, and all the commodities are final products so that all intermediate primary inputs of agricultural origin are deducted.

3.2.2. Key Independent Variable

The level of the digital economy is the key explanatory variable. The digital economy has a broad scope and it is not accurate to evaluate it from only one aspect. This study constructs a digital economy index based on digital infrastructure, digital social impact and digital trade (see

Table 1. Evaluation index systems of digital economy development in SSA.

First Index	Secondary Index	Units	Data Sources
Digital infrastructure	Fixed broadband subscriptions	per 100 people	ITU
	Fixed telephone subscriptions	per 100 people	ITU
	Mobile-cellular telephone subscriptions	per 100 people	ITU
	Telecommunication Infrastructure Index	-	UN
Digital social impact	Individual internet users	% of population	ITU
	Online Service Index	-	UN
	E-Participation Index	-	UN
Digital trade	ICT goods exports	% of total goods exports	WDI
	ICT goods imports	% total goods imports	WDI
	ICT service exports	% of service exports	WDI
	ICT service imports	% of service imports	WDI

Notes: ITU represents the International Telecommunication Union, and UN represents United Nations E-Government survey.

), by referring to Chen and Ye [50], Dong and Hu [51] and Shahbaz and Wang [52]. The weights of the secondary indexes are determined by using the entropy method, which is effective in overcoming the subjective arbitrariness and obtaining objective weights, according to Wang and Zhao [53] and Wu and Xu [54].

3.2.3. Control Variables

Agricultural development is caused by various factors. Control variables are selected to avoid inaccurate regression results. National economic development represents the vitality of the whole society, and a higher level of economic development can provide more human resources, financial and technical support for agricultural development [1]. Economic development is represented by GDP per capita (pgdp). The expansion of arable land has been an important driver of agricultural growth in SSA in recent decades [55], so this study also uses arable land per capita a control variable (land). Agricultural development can be significantly improved by upgrading the industrial structure and building a modern industrial system [56]. Thus, the percentage of value added of the secondary industry in GDP is selected as the control variable (ind). Foreign capital is also an important factor influencing agricultural production in SSA, but the impact is not clear yet as existing studies reach conflicting conclusions [57,58]. The percentage of net inflows of foreign direct investment (FDI) in GDP in SSA is utilized (fdi). Data for these control variables are from WDI.

3.2.4. Mediating Variables

1.

Agricultural TFP

One of the most effective ways of measuring agricultural productivity is TFP. TFP measures the amount of agricultural output produced by the combination of inputs such as land, labor, capital, etc. Growth of TFP means that total agricultural output is increasing faster than total inputs. Agricultural TFP is obtained from the United States Department of Agriculture (USDA).

2.

Human capital

The human capital index is taken from the UN E-Government Survey and reflects the general or traditional literacy of a country. It is calculated using four indicators: the percentage of people who are literate, gross enrolment ratios, expected years and average years of schooling.

3.

Government governance

The government governance of a country determines many economic issues. A government with a high governance quality is able to formulate more appropriate policies to promote agricultural development, while at the same time receiving feedback and suggestions from the public to improve these policies. There is no generally unified definition and measurement for governance quality. This study evaluates government governance from six dimensions by using the indicators from the dataset of the World Governance Indicators (WGI) (see

Table A1. Evaluation of government governance in SSA.

The Primary Indexes	The Definition of Sub-Indexes
Voice and Accountability	Reflects the perceptions of a country’s citizens on participation in choosing their government, freedom of expression, freedom of association, and free media.
Political Stability and Absence of Violence/Terrorism	Reflects the potential for governments to be destabilized or overthrown by unconstitutional violent means, such as violence and terrorism.
Government Effectiveness	Captures perceptions of the quality of public services and policy formulation and implementation, and the credibility of the government’s commitment to these policies.
Regulatory Quality	Acquires perspectives on the government’s capacity to formulate and implement sound policies and regulations that permit and facilitate private sector development.
Rule of Law	Captures perception of the extent of trust in the rules of society, in particular the quality of contract enforcement, property rights, police and courts, and the likelihood of crime and violence.
Control of Corruption	Captures perception of the extent to which public power is exercised for private gain, including small and large scale corruption, and the “capture” of the state by elites and private interests.

). The entropy method is used to obtain a unified index of government governance according to Shahbaz and Wang [52]. A higher index indicates better quality of government governance.

Table 2. Description of the variables.

Variable	Definition	Units	Data Source
agd	Food production index		World Bank, 2023
de	Digital economy index		Authors’ calculation
pgdp	GDP per capita	constant 2015 USD ($)	World Bank, 2023
land	Arable land	Per 100 people	World Bank, 2023
ind	Percentage of value added of the secondary industry in GDP	%	World Bank, 2023
fdi	Percentage of net inflows of foreign direct investment in GDP	%	World Bank, 2023
atfp	Agricultural TFP		USDA
hc	Human capital index		UN
gov	Government governance index		Authors’ calculation

illustrates the definition and data sources and

Table 3. Descriptive Statistics of the Variables.

Variable	Obs.	Mean	Std. Dev.	Min.	Max.
lnagd	560	4.565	0.185	3.831	5.200
lnde	560	−2.496	0.609	−3.725	−0.492
lnpgdp	560	7.174	0.879	5.565	9.302
lnland	560	2.906	0.531	1.779	4.607
lnind	560	3.162	0.398	2.091	4.192
lnfdi	560	3.267	0.193	1.386	4.139
lnatfp	560	4.607	0.117	4.193	5.047
lnhc	560	1.152	0.044	1.099	1.329
lngov	560	−0.864	0.424	−2.323	−0.094

represents the descriptive statistics of all the variables by using Stata17 software (StataCorp, Lakeway Drive, TX, USA).

4. Empirical Results

4.1. Benchmark Estimates

As discussed above, the orthogonal-difference GMM is applied as the benchmark method, while the results of the first-difference GMM and fixed effect (FE) model are for comparison and robustness check.

To check the validity of the dynamic estimates, the Arellano—Bond (A—B) test is applied to examine the existence of serial autocorrelation, and the Sargan test is applied to check the validity of the instruments. As shown in the bottom of

Table 4. Benchmark regression results.

	Orthogonal-Deviation GMM	First-Difference GMM	FE
l.lnagd	0.820 ***	0.652 ***
	(0.02)	(0.03)
lnde	0.033 ***	0.064 ***	0.201 ***
	(0.01)	(0.01)	(0.02)
lnpgdp	0.101 ***	0.319 ***	0.414 ***
	(0.02)	(0.04)	(0.07)
lnland	0.054 ***	0.094 ***	−0.130 **
	(0.01)	(0.01)	(0.06)
lnind	−0.030 ***	−0.023 *	−0.110 **
	(0.01)	(0.01)	(0.04)
lnfdi	0.001	0.004	0.027
	(0.01)	(0.01)	(0.04)
Constant			2.736 ***
			(0.54)
AR(1)	0.001	0.001
AR(2)	0.177	0.267
Sargan test	0.568	0.469

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

, the results of AR(1), AR(2) and the Sargan test confirm the validity and the effectiveness of the instruments.

The lag term of the dependence variable (l.lnagd) is 0.82 in the benchmark estimation in column (2) of Table 4. This indicates that food production grows faster when the country is at a lower initial level of food production. It also implies that agriculture in SSA will continue to develop relatively slowly at a natural rate of increase in the absence of intervention policies.

For the core explanatory variable, the estimated coefficient of lnde is 0.033, indicating that the improving digital economy can contribute to agricultural development. This result confirms the above arguments. The digital transition has promoted information transfer in agricultural production. The development of ICT has become an important factor in modern agriculture and is conducive to improving productivity. In the past, it was difficult for smallholder farmers to enter the market due to the existence of information asymmetry. Today, farmers in SSA can improve their practices by actively using ICT to access information and obtain essential facilities. It is also convenient for them to learn new technologies and knowledge on land management, use of new seeds and fertilizer use [59]. ICT use has also been shown to increase income and improve household welfare. Marwa and Mburu [60] show that the use of the ICT platform has increased household income by 22%. Increased income allows farmers to invest in more advanced technologies in agriculture. Moreover, the proliferation of smartphones and computers has enabled farmers to maintain mutual communication with experts and related institutions and receive more accurate suggestions on agricultural production [61]. The result is consistent with Hou and Huo [62] and Ma and Grafton [63]. As for the control variables, the coefficient of the economic growth is positive at 1% significant level. As the economy continues to grow, more capital can be invested in the agricultural sector, including R&D and new technologies. It also creates more employment opportunities in the economy, so that agricultural labor can be transferred out of agriculture, thereby improving agricultural production productivity. The increase of arable land per capita is conducive to agricultural development, as a 1% increase in arable land per capita contributes to a 0.054% increase in food production. Agriculture in SSA today is still dominated by rain-fed farming. Low agricultural productivity and poor agricultural infrastructure have forced SSA to rely on expansion of arable land, rather than increasing yields, to ensure food supply.

Industrial upgrading has a negative impact on agricultural development, as the coefficient of lnind is −0.03 at the 1% significant level. It is somewhat surprising that industrial upgrading is not beneficial for food production in SSA. The reason may be that industrialization in SSA is still in its infancy and is not based on agriculture, as the agricultural sector in SSA is too weak to provide more surplus products for industrial development. What is worse, as industry develops, a large surplus of the rural population moves from the countryside to the urban and industrial sectors without providing feedback or support to agriculture. Another reason may be that several SSA countries are rich in oil and their industries are heavily dependent on oil exports, making agriculture susceptible to Dutch disease [64]. The impact of FDI on agricultural development is not significant in SSA. This may be because FDI is more likely to flow into secondary industries, such as minerals, energy and manufacturing, or tertiary industries, such as education and health. The return on investment in agriculture is low due to poor infrastructure, high information costs and other policy barriers [64], hence FDI in agriculture is minimal.

4.2. Robustness Check

We replaced the dependent variable with agricultural value added and re-estimated the results by using orthogonal-deviation GMM and first-difference GMM. The results are presented in columns (1) and (2) in

Table 5. Results of robustness check.

Dependent Variables	lnagv		lnagd
	Orthogonal-Deviation GMM	First-Difference GMM	Orthogonal-Deviation GMM	First-Difference GMM	Orthogonal-Deviation GMM	First-Difference GMM	FGLS
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
l.lnagv	0.655 ***	0.457 ***
	(0.03)	(0.02)
l.lnagd			0.687 ***	0.686 ***	0.814 ***	0.741 ***
			(0.03)	(0.03)	(0.01)	(0.01)
lnde	0.097 ***	0.148 ***					0.195 ***
	(0.01)	(0.01)					(0.02)
lnmtele			0.045 ***	0.052 ***
			(0.01)	(0.01)
lninteruser					0.021 ***	0.027 ***
					(0.01)	(0.01)
lnpgdp	0.200 ***	0.463 ***	0.106 ***	0.116 ***	0.044	0.177 ***	0.030
	(0.03)	(0.04)	(0.02)	(0.04)	(0.04)	(0.06)	(0.02)
lnland	0.022 ***	0.036 ***	0.031 **	0.084 ***	0.098 ***	0.157 ***	0.067 ***
	(0.01)	(0.00550)	(0.01)	(0.01)	(0.04)	(0.01)	(0.02)
lnind	−0.068 ***	−0.018 **	−0.027 ***	−0.01	−0.016 *	−0.012	−0.064 **
	(0.01)	(0.01)	(0.01)	(0.01)	(0.01)	(0.02)	(0.03)
lnfdi	0.012 *	−0.001	−0.015 **	−0.03 **	0.006	0.010	0.037
	(0.01)	(0.01)	(0.01)	(0.02)	(0.01)	(0.01)	(0.02)
Constant							5.550 ***
							(0.13)
AR(1)	0.001	0.001	0.001	0.001	0.001	0.001
AR(2)	0.995	0.811	0.296	0.177	0.161	0.252
Sargan test	0.774	0.810	0.936	0.711	0.788	0.838

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. The variable l.lnagv means the lagged term of agricultural value added. The coefficients of lnde are significantly positive, indicating that the results of the benchmark regression are stable.

Then we replaced lnde with Mobile-cellular telephone subscriptions and individual internet users, respectively, to check the robustness. The results are presented in columns (3)–(6). It is obvious that the results are basically consistent with those in the benchmark regression, indicating the reliability of the benchmark results.

Furthermore, the robustness of the benchmark regression is tested by applying an alternative estimation technique. The feasible generalized least squares (FGLS) is used to solve the possible cross-sectional correlation and to obtain more effective estimates [65]. The results are shown in column (7) and it can be seen that the coefficient of lnde is still significantly positive, thus confirming the robustness.

4.3. Mediating Effect Analysis

4.3.1. The Mediating Role of Agricultural Productivity

Furthermore, the mediating effect of agricultural productivity was checked in the digital economy—agricultural development nexus. The results are presented in

Table 6. Results of the mediating effect analysis.

Dependent Variables	lnagd	lnatfp	lnagd	lnhc	lnagd	lngov	lnagd
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
l.lnagd	0.820 ***		0.804 ***		0.815 ***		0.805 ***
	(0.02)		(0.02)		(0.02)		(0.02)
l.lnatfp		0.753 ***
		(0.03)
l.lnhc				0.974 ***
				(0.01)
l.lngov						0.717 ***
						(0.02)
lnde	0.033 ***	0.037 ***	0.042 ***	0.012 ***	0.031 *	0.020 ***	0.030 ***
	(0.01)	(0.01)	(0.01)	(0.00)	(0.02)	(0.01)	(0.01)
lnatfp			0.143 ***
			(0.04)
lnhc					0.012 ***
					(0.00)
lngov							0.110 *
							(0.06)
lnpgdp	0.101 ***	0.018 *	0.096 ***	0.003 **	0.121 ***	0.063 *	0.173 ***
	(0.02)	(0.01)	(0.02)	(0.00)	(0.02)	(0.03)	(0.04)
lnland	0.054 ***	0.037 ***	0.047 ***	−0.014 ***	0.060 ***	−0.021 ***	0.056 ***
	(0.01)	(0.01)	(0.01)	(0.00)	(0.01)	(0.01)	(0.01)
lnind	−0.030 ***	0.010	−0.035 ***	0.002 **	−0.042 ***	0.017 ***	−0.032 ***
	(0.01)	(0.01)	(0.01)	(0.00)	(0.01)	(0.01)	(0.01)
lnfdi	0.001	0.012 ***	0.008	−0.001	−0.003	0.012	0.002
	(0.01)	(0.00)	(0.01)	(0.001)	(0.01)	(0.01)	(0.01)
AR(1)	0.001	0.001	0.001	0.016	0.001	0.001	0.001
AR(2)	0.177	0.160	0.193	0.267	0.178	0.364	0.272
Sargan test	0.568	0.715	0.307	0.532	0.549	0.896	0.542

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. l.lnatfp, l.lnhc and l.lngov indicate the lagged terms of lnatfp, lnhc and lngov, respectively.

. Column (1) presents the results of the benchmark regression, and column (2) presents the impact of the digital economy on agricultural TFP. The coefficient is significantly positive and verifies the existence of the mediating effect of agricultural TFP. The main reason for low agricultural productivity in developing countries is the lack of market information. It is critical for farmers to acquire information about improved seeds and the timing of planting, fertilizers, pesticides and machinery at the growing stage, and information about the timing of harvesting. Access to this information can lead to good agricultural performance. The digital economy has given rise to a range of digital products and smart services that have changed the mode of agricultural production from mechanical to digital, significantly improving the efficiency of farming [66]. Big data can extract critical information for farmers to improve their ability to make predictions, prevent crop diseases and pests, and access accurate weather data [67]. Although there is still a long way to go to achieve smart agriculture in SSA, the use of ICT is sufficient for farmers to increase productivity. In addition, owning a mobile device allows producers to overcome geographical barriers and connect to the market, either locally or across regions. It allows farmers to identify buyers and traders and decide at what price to sell products, thus reducing communication costs and facilitating transactions. Furthermore, the results in column (3) show that agricultural TFP positively affects food production. Specifically, a 1% increase in ATFP contributes to a 0.143% increase in food production. In summary, the digital economy has improved agricultural production in SSA by increasing agricultural TFP.

4.3.2. The Mediating Role of Human Capital

Human capital is selected as another mediating variable. As shown in column (4), the digital economy has a positive effect on human capital. Human capital is a combination of knowledge and skills and is the most competitive factor in the modern economy. ICT diffusion affects not only the economy, but also the education and health of the workforce. People who use ICT will manage their activities effectively, saving time and money. Furthermore, the diffusion of ICT is conducive to improving the level of education, awareness of the adoption of new agricultural technologies and understanding of new ways of doing business. According to Ouedraogo and Ngoa Tabi [68], the human capital deficit is the worst in SSA. School enrolment rates in SSA are extremely low relative to population size, and net primary enrolment rates are below 50% in one-third of SSA countries. Gender inequality remains high, with women particularly excluded from education. However, the use of ICT can improve this situation. The flourishing of information sources and ICT equipment has promoted human capital by helping farmers to acquire knowledge and grasp advanced technologies. Moreover, the social benefit of ICT is inclusive, with women also included in the access to digital information and devices, thus shrinking the gender gap in knowledge and skills. In addition, the results in column (5) show that human capital affects agricultural development positively on a 1% significant level, confirming the mediating effect of human capital in the digital economy and agricultural development nexus.

4.3.3. The Mediating Role of Government Governance

According to Column (6), the development of a digital economy promotes government governance. In particular, the digital economy generally exerts a “double-edged sword” effect on government capacity. On the one hand, as a new technology, the emergence of the digital economy generates shocks to traditional government functions by expanding the boundaries of government regulation, thus increasing governance costs to some extent. On the other hand, digital technology can lead to the digital transformation of government, effectively promoting the reform of the traditional approach to governance and enabling the public sector to achieve higher productivity and efficiency, thereby increasing the transparency and fairness of government operations and providing high-quality, low-cost, fast and accurate public services. According to the regression results, it is clear that the digital economy improves government governance. Agricultural development cannot be achieved without proper planning and implementation of the government’s industrial policy. Agriculture is the foundation and plays a crucial role in the stability and prosperity of the national economy. Governments in various countries attach great importance to agricultural development. Strong governance is conducive to the development of agriculture. As shown in Column (7), a 1% increase in the improvement of government governance has promoted agricultural development by 0.11%. Therefore, the validity of Hypothesis 3 is verified.

5. Further Analysis

There are digital divides between countries and even within a region. According to statistics from the United Nations Conference on Trade and Development (UNCTAD), the proportion of the population with access to ICT is 90% in high-income countries, but only 20% in low- and middle-income countries. SSA is an area with uneven development across regions. Although SSA has made great progress in reducing the income gap between countries, the gap in ICT adoption is still significant [69]. Unbalanced levels of the digital economy can exert different effects on agriculture. Therefore, heterogeneous analysis was conducted to further verify such impacts. First, the quantile regression model is applied to investigate the asymmetric impact of the digital economy on agriculture. As shown in

Table 7. Results of the quantile regression.

	q10	q25	q50	q75	q90
lnde	0.182 ***	0.131 ***	0.042 ***	0.054 ***	0.161 ***
	(0.03)	(0.02)	(0.01)	(0.02)	(0.03)
lnpgdp	0.078 *	0.053 ***	0.025 **	0.052 ***	0.102 ***
	(0.04)	(0.02)	(0.01)	(0.01)	(0.01)
lnland	−0.132 ***	−0.102 ***	−0.046 ***	−0.018 *	−0.021 *
	(0.05)	(0.03)	(0.02)	(0.01)	(0.01)
lnind	0.174 ***	0.082 *	−0.001	0.005	0.032
	(0.06)	(0.05)	(0.02)	(0.01)	(0.03)
lnfdi	−0.005	0.089	0.033	0.002	0.036
	(0.15)	(0.07)	(0.03)	(0.01)	(0.04)
Constant	5.203 ***	4.932 ***	5.025 ***	5.305 ***	5.645 ***
	(0.59)	(0.22)	(0.12)	(0.11)	(0.15)

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

, the effects are significant at all quantiles, indicating that agriculture is positively affected by the digital economy across the distribution. However, in terms of coefficients, the effects are larger in regions with higher and lower levels of agricultural development, showing the characteristics of “two high ends and low in the middle”. The reason for this may be that in regions with higher levels of agricultural development, the digital economy may have made a larger contribution, while in regions with lower levels, the digital economy tends to have a larger marginal effect.

Given the differences in income inequality across SSA, this study divides the sample countries into two subsamples. There are four groups of countries defined by the World Bank on the basis of income level: high-income, upper-middle-income, lower-middle-income and low-income. As no country enters the high-income group in our sample, this study divides the sample into low-income and middle-income countries. The middle-income countries include upper middle-income and lower middle-income economies. The classification of countries is presented in

Table A2. Countries of different income levels in SSA.

Income Level	Countries
Middle income	Angola, Benin, Botswana, Cape Verde, Cameroon, Comoros, Congo, Cote d’Ivoire, Eswatini, Gabon, Ghana, Kenya, Lesotho, Mauritania, Mauritius, Namibia, Nigeria, Senegal, South Africa, Tanzania, Zambia, Zimbabwe
Low income	Burkina Faso, Burundi, Ethiopia, Gambia, Guinea, Madagascar, Malawi, Mali, Mozambique, Niger, Togo, Uganda, Rwanda

and Figure A1 and the regression results are presented in

Table 8. Heterogeneous results of countries of different income levels.

	Low-Income		Middle-Income
	Orthogonal-Deviation GMM	First-Difference GMM	Orthogonal-Deviation GMM	First-Difference GMM
l.lnagd	1.237 ***	0.993 ***	0.825 ***	0.826 ***
	(0.28)	(0.27)	(0.04)	(0.05)
lnde	0.253	0.064	0.037 **	0.046 ***
	(0.19)	(0.23)	(0.01)	(0.01)
lnpgdp	0.348 **	0.390 **	0.022	0.035
	(0.16)	(0.17)	(0.02)	(0.058)
lnland	0.006	0.270	0.052 ***	0.095 ***
	(0.24)	(0.46)	(0.01)	(0.02)
lnind	0.033	0.169	0.007	-0.005
	(0.11)	(0.14)	(0.01)	(0.007)
lnfdi	−0.450 **	−0.305 *	0.014 ***	0.008
	(0.20)	(0.16)	(0.01)	(0.02)
AR(1)	0.005	0.009	0.023	0.020
AR(2)	0.157	0.111	0.303	0.261
Sargan test	0.894	0.793	0.910	0.908

Notes: Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

. The estimated coefficients of the digital economy are significant in middle-income countries, while they are insignificant in low-income countries. The reasons for this can be summarized as follows: First, countries with higher income levels may have a high penetration of digital technology and farmers tend to have better digital skills. Adoption of ICT depends on the literacy of the population, such as the ability to use advanced applications and evaluate comprehensive information. However, farmers in low-income SSA countries earn low profits and are older, making them less technologically literate than those in middle-income countries, thus limiting the diffusion of ICT in agriculture. Second, the development of the digital economy is largely determined by the level of infrastructure. For example, lack of access to electricity is a major cause of underdeveloped communications networks. Low-income countries lack the necessary capital to improve digital infrastructure, which hinders the adoption of ICT by farmers. Third, the price of using the Internet in SSA is among the highest in the world, putting it out of reach for a large proportion of the population. Regressive taxes on digital devices are a major factor undermining usage in several countries as costs escalate. The use of digital devices and services is limited by unaffordability, especially in rural areas.

6. Discussion

This paper posits that the digital economy holds substantial potential to empower agricultural production and foster agricultural development. Firstly, the digital economy has the capability to shift agricultural production from subjective experience-driven methods to objective data-driven approaches, liberating production from the uncertainties associated with unpredictable outcomes. Secondly, the digital economy contributes to elevating farmers’ incomes by enhancing overall agricultural efficiency. Within the production chain, digital technology optimizes production factor utilization, effectively lowering agricultural production costs. Such increased income levels stimulate enhancements in farmers’ human capital, further elevating agricultural production levels, fostering a self-reinforcing cycle of improvement. Thirdly, the digital economy bolsters the capacity of government governance. Under heightened government oversight, digitally connected agricultural cooperative organizations effectively incentivize diverse agricultural entities to participate actively.

Simultaneously, there exists a practical dilemma in integrating the digital economy into African agriculture. Firstly, the progress of digital infrastructure development in rural areas lags considerably behind urban centers. Given the inherent advantages of cities in terms of location, population density, and resultant high returns on investment, digital technology applications often prioritize urban development. Consequently, the construction of essential agricultural digital infrastructure in rural areas faces relative delays. Secondly, the fusion of the digital economy with agriculture remains inadequately implemented. While the digital economy has revolutionized modern agricultural production, substantial initial investment and subsequent operational costs pose barriers that limit farmers’ willingness to participate. This financial burden weakens their enthusiasm for engaging in the digital transformation of agriculture. Moreover, the majority of African agribusinesses are small-scale with short enterprise industrial chains, offering few high-value, deep-processed products, which are less compatible with the requirements of the digital economy. Thirdly, there is a dearth of support for digital talent in the agricultural sector. Effectively promoting the digital economy to bolster agricultural modernization necessitates a pool of talent proficient in digital technology and knowledge. Presently, rural internet users primarily engage in basic applications such as instant messaging and network entertainment, limiting their grasp of digitalization. Furthermore, their ability to effectively merge digital technology with agriculture remains underdeveloped.

7. Conclusions and Policy Implications

Africa’s persistent challenges in agricultural development, coupled with limited technological advancements in the agricultural sector, have resulted in recurrent issues of hunger and malnutrition. However, the emergence and expansion of the digital economy have presented novel prospects for advancing agricultural practices. Investigating the influence of the digital economy on agricultural production bears significant implications, offering substantial opportunities to bolster agricultural productivity, especially in developing countries and specifically within the context of Africa. Therefore, this study investigates the impact of digital economy on agriculture in 35 SSA countries using multiple dynamic regression models based on panel data covering 2006–2021. The mechanism was also analyzed by incorporating three mediating variables, i.e., agricultural TFP, human capital and government governance. Heterogeneity analysis was also carried out. The conclusions are as follows:

(1)

The digital economy has a positive effect on agriculture; specifically, a 1% increase in the digital economy increases agricultural development by 0.033%.

(2)

The results of the mediation analysis confirm that the digital economy improves agriculture by promoting agricultural TFP, human capital and government governance.

(3)

The heterogeneous results show that the effects of the digital economy are larger in countries with higher and lower levels of agricultural development; the effects are significant in middle-income countries, while insignificant in low-income countries.

The conclusions highlight policy implications as follows. First, it is recommended for SSA countries to improve investment and construction of digital infrastructure, and vigorously develop digital industries to enhance the support to agriculture. The application of digital technologies such as the ICT should be promoted especially in agriculture and rural areas, to improve the transformation of digitalized agriculture. SSA should develop “Internet + Agriculture” and conduct e-commerce to transform and upgrade the whole industrial chain of traditional agriculture, and achieve higher agricultural production capacity and quality.

Second, SSA countries need to focus on improving agricultural efficiency, human capital and governance. Promoting agricultural TFP can maximize the use of digital technology and make it play a key role in agricultural production. There is also a need to increase training in the use of digital technology and professional smart equipment to improve the overall digital skills of the agricultural workforce, thus promoting their understanding and application of the digital technology in agricultural operations. Governance quality should not be neglected as it is important for the government to issue targeted policies for the development of digital economy, such as guiding enterprises in the digital market, improving the construction of digital projects, etc.

Third, differentiated digital economy development strategies are recommended for SSA to narrow the regional “digital divide”, and improve the degree and efficiency of integration between the digital economy and agricultural development, especially in low-income countries. In the process of promoting agricultural development, efforts should be made to strengthen inter-regional exchanges, interactions and cooperation, so as to form an inter-regional development pattern of agriculture that is jointly shared and promoted.

There are some limitations in this study which provide guidance for future research. First, this study only conducts research at the country level from the macro perspective and lacks micro level analysis. Therefore, the use of a microscopic perspective at the farm household level will be considered in future studies. Second, this study does not consider the spatial spillovers of the impact of the digital economy on agriculture. In this regard, future research will take spatial econometric models into consideration to address this issue.