Digitalisation,structural change and the demand for social programs in Sub-Saharan Africa (SSA)

Abstract

Big economic shifts often spark bold public demands. As digitalisation and artificial intelligence reshape economies, what do people expect from governments in response? While research on technological disruption has expanded in advanced economies, far less is known about how these transformations are unfolding in Africa or how Africans perceive them. This article addresses that gap by examining structural change and citizens’ responses to digital disruption in Sub-Saharan Africa (SSA). Using dynamic panel data from 36 African countries from 1995 to 2019 and Afrobarometer Round 8 surveys conducted between 2019 and 2022, covering 45,684 respondents across 32 countries, I show that higher digital diffusion is associated with declining agricultural employment shares, rising service-sector employment shares, and productivity gains that depend on complementary capabilities. I further show that respondents prioritise job creation and work-linked training over general education, business loans, or broader social spending. Rather than the monetary compensation that often features prominently in OECD debates, the dominant public demand in SSA is for productive inclusion through employment. This article discusses the implications of these findings for the kind of employment-rich, productivity-enhancing transformation needed to realise ‘digital dividends' and contributes to broader debates on the political economy of technological change, compensation, and inclusion in the digital age.

Introduction

In the busy streets of Accra’s ‘Circle’ market in Ghana, where the rhythm and tempo of buying and selling is everywhere, Kwame, a young entrepreneur, sets up his mobile phone repair shop. A few years ago, his prospects were limited to the fields of his family’s farm, far from the city, where they cultivated maize for sale and household consumption. Today, his smartphone connects him to a world of online resources, spare parts suppliers, and a customer base that spans beyond his immediate environment. He not only repairs damaged phones but also acts as an ‘agent’ in the chain of mobile money transactions, a service that has revolutionised banking in the region. When I met Kwame during my field research in Ghana in 2021, his story was not an isolated case but part of a larger narrative unfolding across the country and Sub-Saharan Africa (SSA) in general; a narrative of a youthful continent rapidly pivoting towards a digital future, with all its attendant promises and challenges.

The introduction of ICT tools such as computers, mobile phones, the internet, and artificial intelligence is reshaping products, processes, and market channels, altering not only how people behave and communicate but also overall livelihoods worldwide. ¹ This transformation extends beyond individuals and firms to governments, which must now adapt their policies to address socioeconomic dislocations in an increasingly interconnected and digitalised global economy. While digitalisation promises significant economic gains, such as increased efficiency and productivity, it also poses risks, including technological displacement^2,3 and widening inequalities. ⁴ These shifts necessitate reimagined social protection frameworks to mitigate potential disruptions and ensure inclusive outcomes.^5,6

In SSA, digitalisation unfolds against a backdrop of several systemic challenges such as inadequate infrastructure, including limited electricity and internet coverage, low literacy rates, and pervasive un(der)employment. ⁷ The predominance of informal labour markets and weak social protection systems further compounds these challenges, leaving many workers unable to access the full benefits of digital transformation.^8,9 While digital technologies hold promise for fostering economic innovation and improving efficiency, these structural constraints hinder equitable outcomes^4,10 and limit the capacity of states to respond effectively to the disruptions brought about by digitalisation. The absence of robust safety nets exacerbates vulnerabilities, particularly for those in informal and precarious employment.

Yet, the global discourse often neglects the impact of digitalisation on labour markets in the Global South, particularly in Africa, where the dynamics of policy responses differ significantly from those in advanced economies. ¹¹ In OECD countries, public attitudes toward technological change frequently emphasise redistributive policies such as unemployment benefits and income security.^5,12 By contrast, in SSA, weak social protection systems leave many without safety nets, and public preferences toward social policies and compensation in the face of increasing digitalisation remain underexplored. The role of digitalisation in driving structural economic change in SSA has received limited attention and public attitudes toward social policy responses are rarely integrated into the discussion. This study contributes to addressing these gaps by empirically examining the relationship between digitalisation and structural change in SSA while also investigating public priorities for government investment in social programs to respond to these transformations.

This contribution offers a panoramic view across SSA and sets the stage for an in-depth examination of how digitalisation relates to structural change and how social policy might adapt to a rapidly changing future of work. First, I assemble a panel for 36 SSA countries (1995–2019) drawing on the ILO, ITU and the World Bank, and estimate dynamic models using two-step system-GMM with orthogonal deviations. Digitalisation is proxied by mobile network coverage and internet use. I study sectoral employment shares and labour productivity in agriculture, industry and services. Second, I analyse public attitudes towards social policy priorities using Afrobarometer round 8 (32 countries), estimating mixed-effects (multilevel) logistic models with country random intercepts.

The panel results indicate that, as digital diffusion rises, agricultural employment shares tend to decline while service-sector employment shares increase, alongside productivity gains that are conditional on complementary capabilities rather than automatic. Several relationships with internet use are non-linear and turn positive only beyond moderate usage thresholds, suggesting that connectivity enhances efficiency only when supported by complementary inputs such as skills, managerial quality, reliable electricity, and data infrastructure. The public-opinion evidence complements this macro-level pattern. It shows that across 32 countries, respondents consistently prioritise job creation and work-linked training over general education, business loans, or broad social services. In the SSA context, this jobs-first orientation is understandable because high informality and limited contributory insurance make access to earnings more salient than income replacement; fiscal constraints and uneven administrative capacity weaken the credibility of finely targeted redistribution; and weak school-to-work transitions reduce the perceived payoff to general education relative to targeted, employment-relevant upskilling. These findings point to a policy agenda centred on productive inclusion by expanding meaningful connectivity, investing in complementary capabilities across the digital ‘ecosystem’, and deploying employment-oriented instruments such as public works, apprenticeships, and short-cycle vocational training. In short, digitally connected citizens like Kwame are not asking primarily for compensation, but for pathways into better work. The challenge is to turn connectivity into capabilities and capabilities into sustainable decent work.

The rest of the article is organised as follows. The next section reviews the present and prospective future of work in SSA, outlines how digitalisation is reshaping structural change, and synthesises evidence on public attitudes in the digital era. The subsequent section sets out the theoretical framework and hypotheses. This is followed by a section describing the data, measures, and empirical strategies for the macro and micro analyses. The penultimate substantive section presents the results, integrating the macro evidence on structural change with the micro evidence on policy demand. The final section discusses the implications for the future of work and social policy in SSA and situates the article’s contributions within the broader technological change literature.

Digitalisation and structural change in SSA

Digitalisation is driving change by revamping production and trade through improved efficiency, innovation, and diversified services. ¹³ In Africa, descriptive evidence suggests that digitalisation-induced structural changes are well underway. Digitalisation on the one hand, refers to the adoption and integration of digital technologies, which underpin the emergence of new and transformed economic and social activities. ¹⁴ Structural change, on the other hand, can be understood as the transformation of the structure and composition of an economy, encompassing changes in the share of employment, output and income across varying sectors, industries, and regions.^15,16 As Naudé W ¹⁷ avers, this transformation offers a spectrum of industrialisation prospects for Africa, particularly through the advent of automation, additive manufacturing, and the industrial internet.

Interestingly, since the early 2000s, the adoption of digital technologies has soared in SSA. For example, cellphones coverage rose from under 20 per 100 inhabitants in 2005 to approximately 90 per 100 by 2019, as shown in Figure 1. Internet penetration has also witnessed a substantial increase, with less than 5% of SSA having access in 2005, surging to about 40% by 2019 (see also Figure 7, 8, 9 and 10). These trends have even been accelerated by the COVID-19 pandemic, which necessitated a mass transition to online platforms due to mobility restrictions (see 18). Although Africa’s online penetration may lag behind more developed regions, such as the Americas, the Arab States, Asia-Pacific, and Europe, the impact of technology on the continent’s economic landscape is undeniable. New technologies have revolutionised access to communication, particularly significant in regions previously hindered by inadequate infrastructure. Not only have they facilitated financial inclusion for the previously unbanked majority through mobile banking but also enabled small and medium enterprises (SMEs) to engage in e-commerce. ¹ These developments have been instrumental in catalysing profound sectoral transformations.

Figure 1.

Cellphone coverage and internet usage in SSA over time. Note: Source: Cellphone coverage is measured as mobile-cellular subscriptions as a percentage of the population, whereas internet usage refers to the proportion of individuals who used the internet from any location in the last 3 months via a fixed or mobile network. Source: Author’s illustration using data from ITU (2020).

The trajectory of digitalisation within SSA interestingly parallels a significant shift in the economic structure, mainly in terms of employment and productivity (see Figure 12 and 13). Employment composition data from the ILO reveals a sharp transition since the late 1990s, where agriculture employed close to 60% of SSA’s labour force as shown in Figure 2. Fast forward to 2019, and we observe a reversal; over 60% of the population is now absorbed by the services sector; industry and manufacturing have experienced marginal change, while agriculture’s share has diminished to under 40%. Azenui NB et al. ¹⁵ explains this structural evolution, noting that despite disparities in employment and productivity growth across LDCs, a common trend in the majority of SSA countries is the decline in agricultural employment juxtaposed with a rise in service sector employment shares. This is evidenced in Table 5 (in the Appendix), which shows that for instance, in a country like Burkina Faso, there was a dramatic shift from agriculture to services and industry, with agriculture plunging from 89.8% in 1991 to 27.1% in 2018, and services almost doubling. Corresponding changes can be observed in productivity measured by value-added across sectors as seen from Table 5 (see also Figure 11).

Figure 2.

Sectoral employment share and labour productivity in SSA over time. Note: The left graph illustrates sectoral employment shares in SSA since 1995, showing the proportion of the labour force employed in agriculture, industry, and services. It shows a significant decline in agricultural employment and a corresponding increase in the services sector’s share. The right graph depicts labour productivity trends across sectors over the same period. Source: Author’s illustration using data from ILO (2020).

Overall, while SSA countries have experienced a considerable shift from agricultural employment toward service-oriented jobs, with a negligible shift into labour-intensive industries and manufacturing, this transition has not followed the classical trajectory of moving from agriculture directly into industry. ¹⁶ Instead, as Figure 2 (left panel) illustrates, there has been a substantial flow of labour into the services sector, which could be characterised as ‘premature deindustrialisation’. ¹⁸ The services sector, despite absorbing a growing labour force, has not comparatively demonstrated a corresponding leap in productivity (Figure 2 – right panel. See also Figure 6). Thus, the services sector productivity growth is anything but impressive; an issue often linked to the sector’s informal nature and low productivity levels. ¹⁵

These trends in SSA are significant to observe because, historically, as Matthess M and Kunkel S ¹⁶ summarise, the path of employment evolution during economic transitions has traditionally followed a distinctive pattern in most of today’s advanced economies. Agriculture initially dominated the employment sector until the advent of industrialisation, which saw a rise and eventual peak in manufacturing jobs, forming a hump-shaped employment trajectory over time. Conversely, the service sector began with a smaller employment share, which then progressively expanded as the economies matured, particularly evident in the historical data from North America and Europe.

In contemporary developing countries including SSA, the structural shift has varied significantly. Some Asian countries have echoed the development pathway of their now-developed counterparts, with a pronounced shift from agricultural to manufacturing employment, and ultimately to services, with the agricultural employment share dropping from 48% in 1960 to 21% in 2010. For SSA, the data presented above shows there seems to be a deviation from this pattern, that is, there appears to be a jump from agriculture to services, skipping industrialisation. With SSA’s low (10% in 1991, 12% in 2021) industrial sector share of employment over time, it is debatable whether the region is experiencing ‘premature deindustrialisation’, or showing a sign of ‘deferred industrialization’, labour reallocation to high productivity sector, or rather an expansion of low productivity services sector. Several studies (see below) point to the latter. Whatever the answer is, it is clear that this evolution raises pertinent questions about new challenges and opportunities that will emerge and the role of social programs in such a context.

Public attitudes towards policies in response to technological change

Historically, major economic transitions have always triggered public demand for specific policies in response. It becomes necessary to understand the public’s perception of how social policies should evolve to meet the challenges posed by structural shifts in the future of work. While there is a scarcity of research directly addressing public policy responses to technological change within Africa, the future of work, particularly in the context of the welfare state, has sparked an ongoing debate in European countries, with an emphasis on understanding public attitudes towards digitalisation and automation and the corresponding need for social policies to recalibrate in the digital era to address new challenges posed by technological transition.

In the European context, Thewissen S and Rueda D ¹² contend that individuals in routine task-intensive occupations, particularly those with higher incomes, show a heightened preference for public insurance against income loss due to technological change. Busemeyer MR and Sahm AHJ ¹⁹ take a broader look at welfare state attitudes, suggesting that technological risk increases support for redistribution but shows a negative or non-significant association with support for social investment policies. Dermont C and Weisstanner D ²⁰ extend the debate by examining support for universal basic income (UBI) as a potential policy response to automation, finding that, in contrast to the demand for redistribution, UBI preferences are not clearly associated with technological job displacement risk. Kaihovaara A and Im ZJ ²¹ introduce a new dimension to the discussion by linking occupational vulnerability with attitudes toward immigration, where the routineness of tasks increases negative sentiments toward immigration.

Similarly, Gallego A and Kurer T ²² hold that recent technological changes have created a new divide where routine workers are more likely to support populist parties, while economic winners from digitalisation tend to support the status quo. Gallego A, Kuo A, Manzano D, et al. ⁶ further examine the policy preferences, coming to a conclusion that workers at higher automation risk tend to prefer policies that slow down technological change rather than those offering compensation. Meanwhile, Busemeyer MR, Gandenberger M, Knotz C, et al. ⁵ highlight that technology-related labour market risks might lead to new cross-class coalitions as they increase support for compensatory and protective policy solutions to technological change, even among technology users, highly educated, and higher-income individuals.

While these insights are novel, in the unique socio-economic context of Africa, traditional active labour market policies and trust in the government’s capacity for effective redistribution are often met with scepticism (see Krah R and Mertens G) ²³ . Unlike in European contexts, where public trust in government-led social policies is relatively higher, the African landscape presents a different scenario.

The unique context of SSA and why policy demands May differ

The present of work in SSA

Unlike in advanced economies, the current picture of work in SSA is a blend of predominant informal employment, a dynamic vulnerable labour market with high youth unemployment, and yet weak social protection. Much of the workforce is engaged in self-employment, often under informal conditions. ²⁴ These informal workers typically receive low wages, often below the poverty line, lack social security or pension plans. ²⁵ According to 9, informal employment is most prevalent in Africa compared to other regions, with substantial variation among countries within the continent. Specifically, approximately 84.3% of the workforce in Africa is engaged in the informal economy. However, this percentage varies widely across different African countries, ranging from a low of 15.1% in Seychelles to over 90% in two-fifths of the countries as shown in Figure 3.

Figure 3.

Share of informal employment in total employment around the world. Note: This graph shows that approximately 84.3% of the workforce in Africa is engaged in the informal economy. Source: OECD (2023).

Informal employment in Africa spans a wide spectrum, from informally paid workers without social security to those in unregistered small enterprises, self-employed individuals, and family workers. ²⁶ This sector absorbs a substantial portion of the labour force, preventing higher unemployment rates, but its dominance creates challenges for policymakers trying to manage its economic impact. Labour force participation in Africa is complex due to ambiguities in defining unemployment, the growing number of students and discouraged workers, and the prevalence of domestic caregiving roles, which disproportionately involve women. ²⁵ High informality, coupled with underemployment and vulnerable jobs, often obscures the region’s relatively low official unemployment rates. ⁷

Young people, who make up 70% of Sub-Saharan Africa’s population, are disproportionately affected. They are more likely to work in informal or low-quality jobs compared to older workers, with formal employment opportunities being scarce and primarily occupied by older individuals. ⁷ For example, while youth employment rates (15–24 years) in Africa are comparable to those in other low- and middle-income regions, only 25.5% hold salaried jobs, compared to 68.8% elsewhere. ²⁷ Nearly half (49%) of African youth work in agriculture, much higher than the 25.4% in other regions, and unpaid work is more prevalent at 15.6%, over double the global average. These dynamics reveal a labour market where young people face not only limited formal opportunities but also heightened vulnerability to informal and unpaid employment.

Social protection in SSA remains unfortunately underdeveloped, leaving much of the population vulnerable in an already precarious labour market. Wage and salaried employment has historically been low, rising only marginally from just over 20% in the late 1990s to 28% by 2020, despite an increase in per capita GDP from 11,000 to 15,000 USD during the same period. Vulnerable employment, comprising self-employed and contributing family workers, remains dominant, averaging over 70% of total employment. This poses significant challenges as these workers often lack formal arrangements and are more susceptible to economic shocks. Social protection coverage remains minimal, with less than 10% of the population enrolled in social insurance programs and social safety net coverage increasing only slightly from 22.8% to 24.4% between 1995 and 2019 (see Table 1). Social security contributions as a percentage of revenue are negligible, averaging just 1.7% over the same period. Policymaking in this area has also seen little progress, as indicated by stagnant CPIA scores on social inclusion and protection. These gaps highlight a disconnect between the region’s high levels of informality and underemployment and the urgent need for stronger social safety nets to support its workforce in the face of ongoing economic transformation.

Table 1.

Social protection coverage in SSA (1995–2010 and 1996–2019).

	1995–2010			1996–2019
Variables	Mean	Min	Max	Mean	Min	Max
Coverage of social insurance programs (% of population)	5.1	0.5	59.5	6.4	0.5	59.5
Coverage of social protection and labour programs (% of population)	25.1	0.5	81.2	28.7	2.3	81.2
Coverage of social safety net programs (% of population)	22.8	0.5	80.4	24.4	0.6	80.4
CPIA policies for social inclusion/equity cluster average (1 = low to 6 = high)	3.2	1.5	4.3	3.2	1.5	4.3
CPIA social protection rating (1 = low to 6 = high)	2.9	1.0	4.5	2.9	1.0	4.5
Social contributions (% of revenue)	1.6	0.0	18.5	1.7	0.0	18.5

Note: CPIA denotes Country Policy and Institutional Assessment. Average values shown. Coverage is for 36 countries in SSA. Source: Author’s calculations using data from the WDI (2020).

The future of work in SSA

The historical and current overview of the nature of work in SSA presented above begs a compelling question of what the future of work holds for the African populace. Interestingly, the digitalisation boom since the turn of the millennium has introduced innovations such as artificial intelligence, big data, blockchain, and the Internet of Things, reshaping production, commerce and policy landscapes, and making Africans generally optimistic. ²⁸ These technologies offer the potential to enhance productivity, foster innovative employment models through digital platforms, and expand opportunities for marginalised groups, driving economic growth and skills development. However, the trajectory remains uncertain, with three scenarios outlined by Abdychev A, Alonso C, Alper E et al ²⁹ : an optimistic ‘Africa Arisen’ with digital advances empowering a youthful workforce and integrating SSA globally; a conservative ‘Africa for Africa’ prioritising local solutions; and a pessimistic ‘Africa Adrift’, where automation disrupts jobs and climate challenges persist despite localised entrepreneurship. While the exact path may be debatable, the diffusion of ICTs undoubtedly carries profound implications for the region’s future. ²⁷

To cite some examples, on the bright side, digital technologies are enabling the emergence of new job markets in SSA. For instance, the rise of the gig economy, fuelled by digital platforms, allows individuals to engage in flexible, on-demand work. ³⁰ Between 2018 and 2019, Africa witnessed a substantial 37% growth in digital platforms, with South Africa leading in numbers (over 140), Kenya showing the fastest growth rate at 71%, and other countries like Ghana, Nigeria, Uganda, Rwanda, Tanzania, and Zambia also experiencing notable increases.^31,32 Given the region’s unemployment challenges, this can be particularly transformative in areas with high unemployment rates, as it provides alternative income-generating opportunities. Johnson C, Bester H, Janse van Vuuren P, et al. ³¹ revealed that in 2018, the gig economy supported the livelihoods of 4.8 million individuals across seven African countries. Nascent, embedded AI-credit-scoring and fraud detection in fintech, route optimisation in logistics, and computer-vision tools in agtech, is diffusing from a low base in hubs such as South Africa, Kenya, and Nigeria.^33–35

Digital tools lower barriers to entry for starting businesses (see Ref. 36). E-commerce platforms, digital payment systems, and social media marketing have made it easier for entrepreneurs in SSA to establish and grow their businesses, creating jobs and stimulating local economies. The integration of digital tools in traditional industries can also lead to increased productivity.^34,37 For example, in agriculture, which is a significant sector in many SSA economies, digital technologies like precision farming and mobile apps for market pricing and weather updates can enhance efficiency and output, leading to higher incomes for farmers and related workers. ³⁸ Apart from these, digital technologies also facilitate greater access to educational resources and skills training, which is crucial for building a workforce capable of thriving in a technology-driven economy.

On the other side, the intersection of automation risk and digital infrastructure raises fears about technological displacement. AI uptake for example, can heighten displacement risks where routine-task intensity is high and protections thin, reinforcing a potential ‘services trap’ and precarious platform work dynamics.^8,37,39 Advances in digital technologies increasingly enable the automation of both physical and virtual tasks, with studies estimating a high automation potential in SSA. For instance, 42.4% of jobs in Ghana are considered automatable ³⁹ (see Table 2). This threat looms larger in SSA than in advanced economies due to the region’s high share of routine tasks readily automatable, unlike in OECD countries where many such roles have already transitioned through earlier technological shifts. ³⁷ However, these estimates must be contextualised, given SSA’s significant digital divides. The region faces barriers like limited access to affordable internet, devices, and stable infrastructure, with 50% of the region lacking electricity. Also, platform work, which offers new opportunities, is often inaccessible to those without internet-capable devices and is marked by income instability and the absence of social protections. ⁸ Empirical evidence from South Africa’s platform economy shows workers enter this sector out of necessity, drawn by flexibility but operating in a precarious ‘survivalist’ mode (see 8). All these signal the need to investigate social consensus towards government intervention to address vulnerabilities exacerbated by digitalisation.

Table 2.

Estimates of jobs/activities at risk of automation by study, context, scope, and approach.

Study (year)	Geography/Context	Scope	Estimate	Approach
Frey & Osborne (2017) 2	United States	Micro	47% of US employment at high risk of computerisation	Occupation-based automation probabilities using O*NET task profiles.
Muro, Maxim & Whiton (2019) 67	United States	Micro	25% of workers (about 36m) at high exposure	Task-exposure index by occupation/metro; ‘high exposure’ threshold.
Manyika et al. (2017) – McKinsey 68	Global	Micro	About 50% of current work activities are technically automatable; fewer than 5% of occupations are fully automatable; about 60% of occupations have at least 30% automatable activities	Activity-level assessment of technical automation potential.
Arntz, Gregory & Zierahn (2016) 3	OECD	Micro	6–12% of jobs at high risk (average about 9%)	Individual task data (PIAAC); task-based probabilities at worker level.
World Bank (2016) 37	Developing countries	Macro–micro hybrid	About 2/3 of jobs are susceptible to automation from a technological standpoint	Occupational risk indices applied to national employment structures.
Egana-delSol (2020) 39	Developing countries incl. SSA; Ghana	Micro	8–42% of jobs at high risk (median 23%); Ghana 42.4%	Task-level ‘automatable share’ within occupations.

Notes: Author’s illustration. ‘Micro’ denotes occupation/task/individual-level exposure; ‘Macro–micro hybrid’ applies micro probabilities to country structures. Estimates reflect exposure to automation, not realised job losses; methods are not strictly comparable across studies.

Theoretical framework and hypotheses

This article conceptualises digitalisation as the diffusion of ICT ¹ infrastructure and use that lowers search, coordination, and contracting costs within and across firms, thereby reconfiguring production and the allocation of labour. My digitalisation proxies, that is, cellphone coverage and internet usage, are standard in the African literature and plausibly capture information, matching, and market-integration channels that theory links to reallocation and coordination-cost reductions.^10,40 Two interlocking channels organise my expectations. First, at the task margin, digital tools substitute for routine, codifiable activities while complementing non-routine problem-solving, communication, and managerial tasks, shifting relative labour demand and wage premia within sectors.^41–43

In SSA labour markets, characterised by thin social insurance and high informality, these adjustments frequently materialise as extensive-margin changes in employment and hours rather than neat wage re-pricing. ⁴⁴ Second, at the sectoral margin, digitalisation reduces fixed and variable coordination costs, deepens the use of producer services (e.g. logistics, business services), and facilitates participation in fragmented value chains. Cumulatively, this reallocation pressure tends to raise the employment and value-added shares of services.^45,46 That said, outcomes are heterogeneous: service-led development may arise where tradable, higher-productivity services expand, while services-sector dualism can dominate where absorptions occur into low-productivity local services, in line with concerns about premature deindustrialisation.^18,47 Because diffusion exhibits complementarities and network effects, relationships between internet usage and sectoral outcomes may be non-linear (thresholds or diminishing returns), which motivates the quadratic specification.^48,49

Hypothesis 1

Higher digital diffusion is associated with a larger services employment share and a smaller agricultural share. Effects on industry are theoretically ambiguous given servicification and the risk of premature deindustrialisation.

Beyond reallocation, digitalisation can raise within-sector productivity by easing information frictions, enabling better matching and coordination, and complementing intangible organisational capital. However, these gains are typically capability-conditional. Without complementary investments in skills, management quality, and reliable infrastructure, measured productivity effects may be modest or slow to materialise.^44,50 In SSA, this study therefore anticipates at most moderate associations between diffusion proxies and sectoral labour productivity, with relatively clearer scope in services and formal segments than in agriculture.

Hypothesis 2

Digital diffusion is positively associated with sectoral labour productivity, especially in services, conditional on complementary capabilities; effects may be weak or negligible where such complementarities are scarce.

Altogether, the heterogeneity of services, the servicification of manufacturing, and the risk-shifting role of informality suggest a capability-conditional productivity response to diffusion: short-run elasticities of sectoral labour productivity to coverage and usage are expected to be modest on average, that is, clearer in producer services with complementary capabilities and weaker in agriculture and local non-tradables, consistent with reallocation outpacing measured efficiency gains.^{18,44,46,47,50} The macro productivity models therefore treat productivity as highly persistent (hence a dynamic specification) and interpret coefficients on diffusion proxies as lower-bound, complement-dependent associations rather than unconditional technology effects.

Services are not a monolith. Productivity, tradability, and learning opportunities differ sharply between tradable, skill-intensive services (e.g. ICT, finance, transport, business services) and local non-tradables (e.g. retail, personal services), with the former more likely to embed economies of scale, exportability, and knowledge spillovers. ⁴⁶ Digitalisation also servicifies manufacturing: firms bundle design, software, data analytics, logistics, and after-sales support into goods, raising the measured services share even when factory employment is flat. On Africa specifically, the expansion of producer services and logistics has created employment and upgrading opportunities even where manufacturing payrolls are modest. ⁵¹ This dual logic helps reconcile why service-led transformation and concerns about premature deindustrialisation can co-exist: value chains lengthen in services while industrial employment peaks earlier than in historical forerunners.^18,47 Relatedly, sectoral technology differences shape feasible reallocation paths. ⁵² From a late-development perspective, SSA can compress stages by adopting frontier organisational forms and infrastructures, for example, digital payments, e-logistics, cloud-enabled business services, rather than replicating legacy vintages; this is a realistic variant of leapfrogging. ⁵³ The policy corollary is that transformation hinges as much on enabling producer services and complementary capabilities (e.g. skills, management, reliable electricity, data infrastructure) as on adding plant and equipment. Where these complements are thin, employment absorption may concentrate in low-productivity local services despite rising digital connectivity, moderating average productivity gains; where complements accumulate, diffusion can tilt the economy toward higher-productivity, tradable services without requiring large industrial payrolls.^18,46

The structure of African labour markets shapes how digitalisation maps into political preferences. High informality provides an adjustment buffer, as workers can switch activities, add secondary jobs, or vary hours, but it also shifts risk onto households. Platform intermediation (e.g. ride-hailing, delivery services, online freelancing) lowers entry and matching frictions while tightening algorithmic control and externalising insurance, capital, and downtime risks. The effects on net earnings and protection are heterogeneous and context-dependent. ⁵⁴ In comparative political economy, a large (OECD and European dominant) literature finds that exposure to technological risk increases demand for compensation and active labour market policies, including training and upskilling.^5,12,22

This study extends the technological-shock–compensation logic to SSA with two contextual amendments. First, where fiscal capacity is constrained and contributory social insurance is shallow, broad income compensation is less credible or poorly targeted; people may therefore weight employment creation, public works, SME support, and short-cycle work-linked training more heavily than general monetary redistribution. ⁵⁵ Second, because informality blurs employer–employee boundaries and platform work further individualises risk, the salient margin of mitigation is often access to work (and earnings stabilisation) rather than income replacement per se. Consequently, it is reasonable to anticipate that digital exposure heightens the salience of near-term, employment-intensive interventions, including public investment with local job content, basic business services, apprenticeships, targeted upskilling, relative to broad, non-job-specific spending.^7,27,44,54

Hypothesis 3

Individual digital exposure is positively associated with support for job-related policies, for example, employment creation and work-linked training, relative to non-job-specific policy priorities, insofar as digital exposure raises perceived employment risk and adjustment needs.

Because digital tools tend to complement abstract, problem-solving, and communication-intensive tasks, one would expect larger gains, or smaller losses, for workers with greater human capital and better access to complementary inputs such as reliable infrastructure and thick labour markets. In many SSA settings these advantages are more concentrated among more educated and urban workers.^41,42 At the same time, several features of SSA labour markets make large cross-sectional gradients less pronounced in static samples. High informality and multi-activity employment mean adjustments to technological change often occur along extensive margins, for example, hours, secondary jobs, sectoral switching, rather than wage re-pricing, weakening observable differences by schooling or location. ⁴⁴ Formal credentials are an imperfect proxy for task-relevant skills where school quality is uneven, and productivity responses are likely to be capability-conditional where management quality, infrastructure, and skills are uneven. ⁵⁰ Urban advantages can also diffuse through general-equilibrium channels, such as lower search and coordination costs, platform intermediation, and cheaper producer services, benefiting a wide range of activities, including low-skill services, which compresses observable heterogeneity in the short run.^46,54 Finally, liquidity constraints and household risk sharing can push preferences toward broadly employment-intensive policies across groups, narrowing gaps one might otherwise anticipate on purely skill-complementarity grounds. These considerations suggest a directional expectation of stronger support among more educated and urban respondents, but also a realistic prior that such moderations will be modest in the short term and may emerge more clearly over time or in settings with more intensive capability investments. ⁴⁴

Hypothesis 4

The positive association between digital exposure and job-related policy support may be more pronounced among more educated and urban respondents.

Methodology

This study combines macro-level and micro-level analyses to examine the effects of digitalisation on structural change and social policy preferences in SSA. At the macro-level, a panel for 36 SSA countries spanning 1995–2019 is constructed using digitalisation indicators from the International Telecommunication Union (ITU), structural change indicators from the International Labour Organization (ILO), and controls from the Penn World Table, International IDEA, and the World Bank’s World Development Indicators (WDI). A two-step System Generalised Method of Moments (System–GMM) estimator with orthogonal deviations is employed to address endogeneity and dynamic panel issues. At the micro-level, public opinion data from Afrobarometer round 8 (between 2019 and 2022), covering 32 SSA countries, are analysed using mixed-effects (multilevel) logistic models to assess how digitalisation is associated with support for various policy options.

Data and measures

Macro

The macro analysis draws on multiple sources of data as outlined above. Digitalisation is proxied by cellphone coverage (population covered by a mobile-cellular network) and internet usage (share of individuals using the internet), both from the ITU. These proxies are standard in African applications and plausibly capture information, matching, and market-integration channels that theory links to reallocation and coordination-cost reductions.^10,40 They also track the ITU’s multidimensional approach while offering longer historical coverage than composite indices. ⁵⁶

Structural change is measured on two margins: sectoral employment shares (agriculture, industry, services) and sectoral value-added per worker (labour productivity) from the ILO. Additional controls include the log capital stock (Penn World Table), mean years of schooling (WDI), institutional quality (e.g. predictability of law enforcement, democracy; International IDEA), trade openness, FDI inflows (% GDP), inflation, log GDP per capita, and trade openness, which captures the degree of integration (all from the WDI). Descriptive statistics are reported in Table 3. The period 1995–2019 is chosen to avoid confounding COVID-19 dynamics.

Micro

The micro analysis uses Afrobarometer round 8 data covering 32 SSA countries (45,684 respondents, see Table 4). This time period has been chosen because the earliest available data (on the outcomes of interest) covers only this period. Individual digital exposure is operationalised using cellphone ownership and internet use frequency (Figure 14); a composite digitalisation index rescaled to 0–100 summarises these indicators (mean 54, SD 15). Outcome variables are binary indicators of support for youth-focused policy priorities in job creation, education, job training, business loans, and social services (see Figure 15, 16, 17, 18 and 19). Individual covariates include age, gender, urban/rural residence, education, occupation, lived poverty, and political orientation.

Empirical specifications

Macro

Dynamic sectoral outcomes are estimated with two-step System–GMM,^57,58 using orthogonal deviations to preserve observations when there are gaps and to remove unit effects. ⁵⁹ Year dummies are included. Finite-sample corrected standard errors follow. ⁶⁰ The baseline specification is:

${\text{StrucChange}}_{it}={\beta }_1{\text{StrucChange}}_{i,t-1}+{\beta }_2{\text{Digitalisation}}_{it}+{\beta }_3^{\top }{X}_{it}+{\alpha }_i+{\mu }_{it},$

(1)

where StrucChange _it denotes (i) sectoral employment shares or (ii) sectoral labour productivity for country i and year t; Digitalisation _it is cellphone coverage or internet usage; and X _it contains controls. To probe non-linearity consistent with threshold dynamics and network effects, squared terms for digitalisation are included in Equation 1; digital variables are mean-centred before squaring to mitigate collinearity.^48,49

Endogeneity and instruments

The lagged outcome variable is treated as endogenous. Digitalisation proxies are treated as predetermined (potentially correlated with past shocks but not current innovations), while time-varying macro controls are classified as predetermined or exogenous according to standard priors (e.g. inflation more plausibly exogenous; institutional measures more plausibly predetermined). Internal instruments are constructed from suitable lags of the endogenous and predetermined variables in levels and orthogonal deviations. To contain instrument proliferation and weak-instrument risks, instruments are collapsed, lag depth is restricted (e.g. L2–L3 in differences; shallow lags in levels), and instrument counts are reported with Hansen and difference-in-Hansen tests.^59,61 Serial correlation diagnostics report AR (1) and AR (2) in differences. ⁶²

Micro

Binary support for each policy is modelled via mixed-effects (multilevel) logistic regressions with a random intercept for country, based on the rationale that individuals are nested in countries.^63,64 Let Y _ij ∈ {0, 1} indicate support for a given policy by individual i in country j. The baseline model is specified in Equation 2:

$\text{logit}\mathrm{Pr}(Y_{ij}=1)=\alpha +\beta D_{ij}+{\gamma }^{\top }{Z}_{ij}+{\mu }_j,$

(2)

where D _ij is the 0–100 digitalisation index, Z _ij is the vector of individual controls (age, gender, urban/rural, education, occupation, lived poverty, political orientation), and ${\mu }_j\sim \mathcal{N}\left(0,{\sigma }_{\mu }^2\right)$ is a country-level random intercept capturing unobserved country heterogeneity.

To examine moderation without soaking effects, interactions are introduced one at a time:

$\text{logit}\mathrm{Pr}(Y_{ij}=1)=\alpha +\beta D_{ij}+{\theta }_k(D_{ij}\times W_{k,ij})+{\gamma }^{\top }{Z}_{ij}+{\mu }_j,$

(3)

where W_k,ij is a single moderator (education, gender, urban/rural, age, or lived poverty). The digital index in Equation 3 is mean-centred so that the main effects retain a clear interpretation. For continuous moderators (e.g. age), both variables are mean-centred. Average marginal effects are then reported for interpretability. Random slopes are not adopted in the main models to preserve parsimony and comparability across specifications.

Limitations of the econometric estimations

While this study provides new insights into the influence of digitalisation on structural transformation and social policy preferences in SSA, it is not without limitations. These limitations span both econometric and data-related challenges, which must be considered when interpreting or applying the findings.

First, the system-GMM estimator employed in the macro-level analysis addresses key challenges of endogeneity and dynamic panel bias. For example, the AR (2) tests do not indicate second-order serial correlation, which is a necessary condition for the validity of lagged internal instruments (Tables 8 and 9). In addition, the Hansen J test does not reject the null of joint instrument validity. However, the diagnostics in Tables 8 and 9 also report perfect Hansen p-values (1.000) alongside strong Sargan rejections. This configuration warrants caution in interpretation. The Sargan test is not robust to heteroskedasticity and can over-reject under common forms of misspecification, whereas the Hansen test is robust but may have weak power in finite samples and when the instrument set closely fits the endogenous variables.^59,61,62 Consequently, very high Hansen p-values should not be read as unambiguous proof of instrument validity. I therefore treat the specification tests as supportive but not definitive and interpret the system-GMM estimates as indicating robust associations rather than fully causal effects.^59,65

Second, the micro-level analysis is subject to several measurement and contextual constraints. Country random intercepts capture between-country heterogeneity, but subnational variation in economic conditions and institutions remains unobserved. The composite digitalisation index captures access and use, rather than the intensity or quality of digital engagement.

Moreover, the timing and structure of the available datasets do not permit a direct individual-macro linkage. I therefore treat this as complementary macro- and micro-level analyses, where the macro models estimate diffusion–structure associations across countries and years, and the micro models assess whether digital exposure is systematically related to social policy priorities in the same regional context.

Beyond these design constraints, the available macro data also impose limits on inference. Sectoral categories are relatively aggregated, potentially masking within-sector reallocation across sub-sectors and occupations. Missingness in country-year coverage restricts the sample to countries with available information, which may introduce selection concerns and limits external validity. Finally, the digitalisation indicators used in the macro analysis (cellphone coverage and internet usage) are necessarily proxies. Although they align with ITU evidence that internet use closely follows broader digital development, ⁵⁶ they do not capture other important dimensions of technological change, including the diffusion of AI applications, the intermediation of platforms and other features of the gig economy that may affect work and policy preferences through different mechanisms.

Robustness of the results

The robustness of the econometric results derived from the two-step system-GMM estimator can be assessed by comparing them with those from Fixed Effects (FE) and Random Effects (RE) models. The GMM estimator’s key advantage in the analysis lies in its ability to account for endogeneity and the dynamic nature of the relationship being examined. As evidenced in Tables 8 and 9, digitalisation, as proxied by cellphone coverage and internet usage, has differential effects on sectoral employment shares and productivity in SSA. For example, an increase in cellphone coverage correlates with a decline in agricultural employment and an uptick in (industrial and) services employment, signifying a potential reallocation of labour corresponding with digital expansion. This effect is persistent and statistically significant across the models.

Turning to the robustness checks provided by the FE and RE models, as shown by Tables 10 and 11, I find that the direction of the effect of digitalisation on sectoral shifts remains largely consistent, thereby reinforcing the system-GMM findings. However, the magnitude of the coefficients is different. This discrepancy is not unexpected as FE and RE models, while useful for controlling unobserved heterogeneity, can be less adept at addressing the endogeneity concerns that system-GMM models are specifically designed to handle. Nonetheless, the consistency of the direction of these coefficients across different econometric models suggests that digitalisation is indeed contributing to structural change in SSA, although the magnitude and significance of the coefficients vary across models and sectors. The GMM model, with its focus on dynamic relationships and controlling for endogeneity, provides a more refined analysis compared to the FE and RE models. However, the FE and RE results still offer valuable insights, especially when considering the significance of the Hausman tests that suggest the presence of fixed effects.

On public attitudes to social policy, the robustness of the results from the mixed-effects logistic regression models is also examined by assessing the consistency of findings across the various policy areas and by hierarchical testing of the effects of digital exposure on public support for social policies. The positive and significant relationship between digitalisation and support for job creation and job training policies in Table 12 is a robust finding, reinforced by the consistent direction of effect across different models and specifications. This robustness is validated further by the hierarchical examination of the individual policy areas (Tables 13 to 17), where digitalisation consistently shows a positive effect on job creation. The magnitude of the digitalisation coefficient remains stable across different specifications, perhaps signifying that job creation is a pressing need in SSA. Conversely, the robust negative relationship between digitalisation and support for education and business loans suggests a divergence in policy priorities among the digitally literate, potentially reflective of a preference for immediate, tangible outcomes such as job opportunities over longer-term investments in general education.

Results

This section presents the empirical evidence as hypothesised in Section 3. First, I examine whether higher digital diffusion is associated with sectoral reallocation, that is, smaller agricultural employment shares and larger services shares, with ambiguous implications for industry (H1). Second, I assess whether digital diffusion correlates with higher sectoral labour productivity in a capability-conditional manner, with clearer scope in services (H2) (see Figure 4(a) (and Table 8) as well as Figure 4(b) (and Table 9). Third, I test whether individual digital exposure is positively associated with support for job-related public programmes, especially job creation and work-linked training, relative to non-job-specific priorities (H3). Finally, I evaluate anticipated heterogeneity in these policy preferences (H4), probing whether the digitalisation-policy linkage is stronger among respondents with more complementary capabilities.

Figure 4.

Effects of digitalisation on sectoral employment shares and productivity. Note: The graphs illustrate the effects of digitalisation on sectoral employment shares and productivity in SSA. Increased cellphone coverage is associated with a decrease in agricultural employment and increases in industrial and service employment. For internet usage, the effects on employment shares are not statistically significant.

Digital diffusion and sectoral reallocation (macro evidence)

Figure 4(a) and Table 8 report the system-GMM estimates for sectoral employment shares. Consistent with H1, a one percentage-point (pp) increase in cellphone coverage is associated with a decline in the agricultural employment share and increases in industry and services. The point estimates indicate a reduction of about 0.43 pp in agriculture alongside increases of roughly 0.37 pp in industry and 0.32 pp in services on average (see also Table 6). Substantively, these patterns fit a reallocation story in which digital diffusion eases information frictions, reduces coordination costs and deepens producer services, enabling labour to move out of agriculture toward higher-productivity activities.^10,40,46

For internet usage, linear effects on employment shares are generally imprecise, but the significant squared term in services (Table 8, Model 4) points to a non-linearity: services absorption strengthens once usage passes a moderate scale. This is consistent with a network-effects logic, that is, ‘meaningful connectivity’ must reach critical mass before coordination gains show up in employment aggregates. The macro evidence therefore supports H1’s reallocation mechanism that digital diffusion is associated with declining agricultural employment and expanding services, while the industrial margin remains context-dependent. This reflects servicification on the one hand and the risk of a premature peak in industrial employment on the other.^18,47 The signs and qualitative magnitudes are robust to FE/RE checks (Tables 10 and 11).

Digital diffusion and sectoral productivity (macro evidence)

Turning to labour productivity (value-added per worker), Figure 4(b) and Table 9 align with the capability-conditional logic in H2. Cellphone coverage is associated with a small linear gain in agriculture (about 0.02% per pp increase in coverage), no robust effect in services, and a non-linear pattern in industry. By contrast, internet usage enters the industry and services specifications with a negative linear term and a positive squared term, implying usage thresholds beyond which marginal effects turn positive. Using the quadratic turning point U* = −β₁/(2β₂) ² , the estimates imply an industry threshold around $\approx 14.5\%$ of the population using the internet and a services threshold around $\approx 9\%$ . Below those levels, adoption appears too shallow to offset adjustment and organisational costs; above them, the slope becomes positive as coordination gains, producer-services deepening and scale effects dominate. In agriculture, internet usage is linear and positive across the observed range: a 1 pp rise in usage correlates with a $\approx 0.10\%$ increase in productivity on average (see also Table 7).

These sector-specific thresholds are precisely the kind of capability-conditionality anticipated in Section 3 where measured productivity effects are expected to materialise once complementary inputs, such as managerial quality, skills, reliable infrastructure, reach a minimal scale. ⁵⁰ The pattern also echoes African evidence that ICT–productivity links emerge once enabling policies and skills cross modest critical masses, ⁶⁶ and broader comparative work showing conditional growth effects of digital diffusion. ¹ Notably, the services threshold is lower than the industry threshold, consistent with earlier arguments about heterogeneity within services and the servicification channel. In short, H2 is supported in a qualified form: diffusion–productivity links are present but become clearly positive only beyond modest (services) to moderate (industry) usage intensities, with agriculture showing linear gains where information frictions are most pervasive.

Digital exposure and demand for job-centred policies (micro evidence)

Micro-level estimates from mixed-effects logistic models (Table 12, Figure 5) are consistent with H3 (see also Awuni ET and Kemmerling A. ⁵⁵ A one-point increase in the 0–100 digitalisation index is associated with a 0.6 pp increase in the probability of prioritising Job Creation and a 0.3 pp increase for Job Training, holding covariates constant (see also Table 15). By contrast, the likelihood of prioritising General Education and Business Loans declines with higher digitalisation (Tables 12, 14 and 16), while investments in Social Services shows no systematic relationship (Tables 12 and 17). Substantively, these results echo the compensation logic adapted to high-informality settings, namely, among more digitally engaged respondents, the salience of near-term, employment-intensive interventions (e.g. public investment with local job content, apprenticeships, short-cycle upskilling) rises relative to broad, non-job-specific spending. This micro pattern aligns with the macro dynamics of H1-H2. As digital diffusion nudges economies toward services and capability-conditional productivity, people appear to favour policy instruments that stabilise earnings and facilitate work and matching in the near term. The association is strongest for job creation, arguably the most immediate lever in environments where liquidity and insurance constraints are tight.

Figure 5.

Mixed-effects logistic regression of public priorities for investments into social programs ³ . Note: The graph presents the marginal effects of digitalisation on public preferences for investments in various social programs, based on mixed-effects logistic regression. The results indicate that increased digitalisation is positively associated with prioritising job creation and job training programs.

Heterogeneity in the policy preferences (micro evidence)

Table 18 probes moderation via one-by-one interactions, avoiding the soaking problems of stacked interactions. Two of my ex ante expectations under H4 are not realised: education does not reliably steepen the digitalisation slope for either job creation or job training, and urban residence likewise fails to amplify the association. Interaction terms for primary, secondary, and post-secondary schooling are small and imprecise, and the rural×digitalisation terms are close to zero. By contrast, two small exceptions are worth noting: the digitalisation–job-creation slope attenuates modestly with age, and the digitalisation–training linkage weakens among respondents reporting low or moderate lived poverty. These results suggest that the positive link between digital exposure and job-centred policy demand is broadly shared across groups rather than concentrated among the better educated or the urban.

There are several reasons why the education and urban moderation results may be weak in cross-section. First, formal schooling is an imperfect proxy for task-relevant, non-routine skills where school quality is uneven; as a result, ‘education’ may not map tightly onto the complementarities that digital tools reward. Second, widespread mobile diffusion compresses urban–rural gaps in basic access, while platform intermediation and cheaper producer services transmit urban advantages more broadly, muting spatial gradients in stated policy demand. Third, high informality and multi-activity employment shift adjustment to extensive margins (hours, secondary jobs, sectoral switching) rather than wage re-pricing, which weakens observable cross-sectional heterogeneity by schooling or residence. Fourth, measurement constraints bite on both sides: the digitalisation index captures access and frequency of use, not the quality or task relevance of use whereas the policy outcomes reflect broad priorities that command wide baseline support (especially job creation), leaving limited room for heterogeneous slopes. On balance, while the specific education and urban interaction hypotheses are not supported, the pattern is consistent with a high-informality setting in which the perceived need for near-term, employment-intensive interventions cuts across strata.

Diagnostics and robustness (macro and micro)

For the macro models, Tables 8 and 9 report AR (1)/AR (2) tests, Hansen p-values, and (where applicable) difference-in-Hansen tests alongside instrument counts. Instruments are collapsed and lag depth is restricted to contain proliferation; two-step standard errors use the Windmeijer correction. Signs are broadly consistent in benchmark FE/RE checks (Tables 10 and 11), though magnitudes differ as expected when dynamic endogeneity is not fully addressed. For the micro models, results are robust to alternative codings, to excluding ‘don’t know/refused’ responses, and to the one-by-one interaction battery (Appendix Table 18). Across both designs, the evidence is coherent with the expectation that digital diffusion correlates with structural reallocation and capability-conditional productivity at the macro-level (H1–H2) and with a tilt toward job-centred policy demand at the micro-level (H3), with only modest cross-sectional heterogeneity (H4).

Discussion, implications and conclusion

This study set out to provide a panoramic yet empirically grounded account of how digitalisation is reshaping work and social policy preferences in SSA. Building on prior research, two complementary pieces of evidence were tested, that is, macro panel showing how digital diffusion relates to sectoral reallocation and productivity (H1–H2), and micro evidence on how individual digital exposure maps into policy demand (H3–H4). The overarching picture is one of reallocation toward services and capability-conditional productivity gains, coupled with a clear public tilt towards job-centred interventions rather than broad redistribution.

At the macro-level, the dynamic panel estimates support H1: increases in cellphone coverage are associated with declining agricultural employment shares and rising industry and services shares (Figure 4(a); Table 8). This pattern aligns with the realignment mechanism emphasised in the theory section, involving reduced search and coordination costs, greater use of producer services, and participation in fragmented value chains.^10,40,46 Internet usage shows weaker linear effects on employment shares, but the significant quadratic in services indicates that absorption strengthens once usage reaches a moderate scale, consistent with network externalities in ‘meaningful connectivity’. In short, labour moves out of agriculture toward activities where digital tools are more productive, while the industrial margin remains context-dependent, suggesting servicification alongside the possibility that industrial employment peaks at comparatively low levels.^18,47

Turning to productivity, the results align with the capability-conditional logic in H2. Cellphone coverage is associated with small, linear gains in agricultural labour productivity and no robust average effect in services, while industry displays a non-linear relationship (Figure 4(b); Table 9). For internet usage, the signs on the linear and squared terms imply usage thresholds beyond which the marginal effect turns positive. Using the quadratic turning point U* = −β₁/(2β₂), the estimates suggest a threshold around $\approx 14.5\%$ of the population using the internet for industry and $\approx 9\%$ for services; above these levels, productivity rises with further diffusion, while below them adoption appears too shallow to offset adjustment and organisational costs. In agriculture, internet usage is linearly and positively associated with productivity across the observed range, consistent with sizeable information-friction reductions in input and output markets. These patterns are consistent with comparative work showing that ICT–productivity links in Africa are strongest where enabling capabilities, for example, digital skills, managerial quality, infrastructure, etc., cross minimal critical masses,^1,66 and they sit alongside evidence that fast internet spurred firm entry, employment and exporting on the continent. ¹⁰ They also nuance concerns about ‘premature deindustrialisation’. ¹⁸ Thus, in many SSA economies the more salient risk may be persistent industrial stagnation, with services-led transformation proceeding through producer-service deepening rather than large factory payrolls.

The micro evidence directly supports H3. Across 32 countries in Afrobarometer round 8, individual digital exposure is positively associated with support for job creation and work-linked training, with larger effects for job creation, whereas support for general education and business loans is weaker or negative on average (Table 12; Figure 5). This configuration is consistent with SSA’s institutional and labour market context. First, high informality and shallow contributory insurance blunt the credibility and reach of income-replacement schemes; for most workers the salient margin is access to earnings rather than protection of a formal wage.^24,26,44 Second, tight fiscal space and uneven administrative capacity make broad redistribution and fine-grained targeting hard to deliver at scale, lowering expected returns to programmes that require extensive means-testing or heavy state intermediation.^33,34 Relatedly, trust and transparency constraints, alongside risks of clientelism and elite capture, reduce enthusiasm for credit lines and business-loan schemes that often depend on discretionary allocation and monitoring. ²³ Third, the perceived payoff to general education is reduced by slow formal job creation, skills mismatch, and weak school-to-work transitions. Many graduates struggle to secure stable work, which elevates preferences for immediate, work-linked training or placement over broad schooling outlays.^27,34 In this environment, digitally engaged respondents, who are more exposed to online job search, platform work, and volatile gigs, prioritise employment-intensive measures with visible near-term returns (e.g. public works with local content, apprenticeships, short-cycle upskilling) over non-job-specific spending.

This pattern contrasts with OECD findings, where compensation for technology shocks often emphasises monetary redistribution (e.g. unemployment insurance, earnings-related benefits), extended income support or social investment paired with strong activation, and, in some debates, basic income.^5,12,20 Those instruments are more credible and popular where fiscal capacity, tax administration, and formal employment coverage are high, contribution histories are well recorded, and programme governance enjoys greater public trust. By comparison, in SSA’s informality-heavy settings the same instruments are harder to target and finance and may be perceived as less accessible. The divergence therefore does not imply that education or finance are unimportant; rather, it indicates that, given prevailing constraints, citizens favour productive inclusion, that is, jobs now and work-linked skills, over general redistribution or untargeted credit.

With respect to H4, one-by-one moderations between digital exposure and education, gender or rural residence are small and imprecise. Two modest exceptions are a slight dampening of the digitalisation–job-creation slope with age and a weaker digitalisation–training link among respondents with low or moderate lived poverty (Table 18). The overall picture is therefore one of shared job-centred policy demand across groups rather than sharply segmented preferences. This is consistent with the theoretical expectation that in static cross-sections, informality, uneven school quality and the diffusion of low-skill services compress observable gradients, even if skill complementarities would predict steeper slopes in more capability-rich environments.

What do the findings imply, overall, for the future of work and social policy in SSA? Three policy recommendations can be derived directly from the evidence. Firstly, if the objective is productivity and growth, push usage of digital technologies over identified thresholds, not just coverage. In practice, that means bundling connectivity with the complements that turn access into meaningful use: for example, reliable electricity, affordable internet data, entry-level devices, and stepwise capability ladders including basic and intermediate digital skills, frontline management, alongside firm-facing services such as accounting, logistics, payments, cloud back-office, that enable ‘producer services’ adoption.^46,50 Governments can sequence this with (i) targeted tariff reductions on smartphones and entry-level laptops; (ii) outcome-based subsidies for last-mile broadband in secondary cities and growth corridors.

Secondly, governments should ensure that service-led transformation creates many jobs by being explicit about the type of jobs involved and where they are located. They could prioritise tradable services that leverage skills and have strong spillover effects, such as logistics, business services, ICT support, digital finance and tourism operations, while raising productivity in local non-tradable services, such as retail and personal services, via simple digital tools and formalisation paths. Specifically, they should use public procurement to create demand for local business services, attach local job content requirements (such as apprentices, junior technicians and customer service representatives) to infrastructure and e-government contracts, co-finance firm-led apprenticeships and short-cycle, vacancy-linked training rather than generic courses, and support ‘middle-skill’ roles such as technicians, installers, call centre and back-office agents, warehouse and fleet coordinators and maintenance staff, which can be scaled up quickly and provide employment opportunities for young people. In terms of location, they should prioritise secondary cities and transport corridors where usage thresholds are cheapest to reach and agglomeration effects are available, but where congestion is lower. In rural areas, they should pair connectivity with digital market access (e.g. produce e-marketplaces, arrange input delivery and provide extension services) to raise farm and off-farm earnings.^10,46

Thirdly, policies should aim to match citizens’ preferences for job creation and work-related training with protective rules for non-standard work. Portable, contributory micro-benefits should be extended to platform and informal workers (e.g. injury insurance and basic pension tiers). Registration and presumptive taxation should be simplified, and baseline standards for platforms should be enforced (e.g. transparent pricing, dispute resolution and data portability), balancing flexibility with security.^24,26,33

In terms of policy implementation, it must reflect the governance realities of SSA. To mitigate the risks of low trust, corruption, and weak administrative capacity, subsidies and benefits should be routed through digital public infrastructure, as well as e-procurement, in order to reduce leakage. SME vouchers and training slots should be allocated via transparent, rules-based portals, and independent verification should be used in the form of third-party audits of job content and tracer studies of trainees.

The design of this contribution is informative but not without limitations. I address endogeneity in the macro panel and hierarchy in the micro data, yet causal claims remain conditional on the identifying assumptions, instrument strength, and measurement choices. These constraints point to clear research future possibilities. Firstly, future research could enhance the measurement of digitalisation beyond coverage and usage by incorporating indicators of digital intensity, platform participation and AI adoption at firm and worker levels. Secondly, future research could examine aggregation by disaggregating services into tradable producer services and local non-tradables, and by tracing movement between the two. Thirdly, identification could be improved using longitudinal and design-based strategies (e.g. staggered broadband or submarine cable rollouts, 3G/4G/5G upgrades), as well as by linking geocoded micro surveys to small-area infrastructure and market access changes. Fourthly, micro–macro evidence could be integrated more directly by building linked datasets (e.g. Afrobarometer alongside local connectivity, prices and vacancies) and by following cohorts to study adjustment dynamics and lagged complementarities. Finally, attitudinal outcomes could be paired with administrative and programme data (e.g. training uptake, placement and social protection enrolment) to connect preferences with realised policy uptake and labour market trajectories. These avenues would allow future work to speak more directly to causal mechanisms in SSA’s evolving digital economy.

In sum, the evidence points to a SSA pathway in which digital diffusion redirects labour out of agriculture and, once minimal capabilities are in place, raises productivity, earlier and more clearly in services than in industry. Public preferences mirror this structural turn. Respondents who are more digitally engaged prioritise jobs and work-linked training over broad redistribution. For governments, the strategic agenda is therefore two-handed: push usage past threshold levels by investing in complementary capabilities and producer services, and respond to citizens’ revealed priorities with labour-absorbing programmes that connect people to work while building skills for hybrid, digitally enabled roles. Done together, these moves can help convert Africa’s demographic dividend into a digital dividend that is both inclusive and resilient.^34,37

Supplemental Material

Supplemental Material - Digitalisation, structural change and the demand for social programs in Sub-Saharan Africa (SSA)

Supplemental Material for Digitalisation, structural change and the demand for social programs in Sub-Saharan Africa (SSA) by Evans Awuni in Journal of Economic and Social Measurement.