Author: ln

Can Financial Rewards Encourage Reporting of Electoral Fraud: Experimental Evidence from Moldova

Posted on October 19, 2026 | Policy Brief

Moldovan leu banknotes illustrating reporting electoral fraud in Moldova

How can young democracies protect electoral integrity against vote buying and foreign interference? In Moldova, allegations of widespread electoral fraud and Russian interference have made this question particularly pressing. Efforts to curb electoral misconduct have commonly relied on moral appeals and formal monitoring. Authorities have also increasingly encouraged citizens to report vote buying, yet securing their cooperation remains a challenge. We test two policies to encourage reporting in a survey experiment with 1,550 adults in Moldova shortly before the September 2025 parliamentary election: a brief moral message and a guaranteed financial reward. Our study shows that while the moral message has no effect on willingness to report, the financial reward reduces reporting both a community member who accepted money for a vote and a political organisation paying people to attend a rally. These findings highlight the importance of testing reporting policies before implementation. Identifying interventions that fail or backfire can help authorities avoid costly mistakes in their efforts to strengthen democratic institutions.

Which Policies Can Reduce Electoral Fraud?

Moldova’s recent elections illustrate the importance of detecting electoral corruption. The presidential election and constitutional referendum on the country’s European integration took place on 20 October 2024, followed by a presidential run-off on 3 November. International observers documented credible evidence of illicit financial inducements and foreign interference, predominantly attributed by Moldovan authorities and other interlocutors to Russia and pro-Russian political forces (OSCE/ODIHR, 2025).

Before the first round of the 2024 presidential election and the referendum on Moldova’s European integration, police reported a scheme involving approximately 130,000 voters allegedly paid to oppose the referendum and support preferred candidates. This number represented almost 10 percent of normal voter turnout (Balmforth, 2024). Ahead of the parliamentary election on 28 September 2025, there were concerns that electoral fraud could again occur on a large scale. These concerns were reinforced by police investigations into renewed vote-buying activities linked to networks implicated in the previous year’s election. Public authorities had launched initiatives ranging from civic appeals not to sell one’s vote to campaigns encouraging citizens to report electoral corruption through established reporting channels.

Encouraging citizens to report wrongdoing can help authorities detect electoral fraud at a lower cost. Neighbours, colleagues and other witnesses often know who distributed money, where votes were bought and who organised transport—information that can be difficult and expensive for authorities to obtain on their own. But providing a reporting channel is only a first step: people must also be willing to use it. The challenge is to obtain cooperation where citizens distrust authorities or view reporting as disloyal. Our earlier laboratory experiments in Moldova and France found that Moldovan participants considered reporting a tax evader less socially acceptable and reported less frequently. Nevertheless, making reporting possible increased tax compliance in both countries (Romaniuc et al., 2022). Thus, disapproval of reporting can coexist with a deterrent effect.

Against this background, we conducted a survey experiment with a representative sample of Moldovans to test two policies intended to encourage reporting: a moral appeal highlighting the social harm of electoral corruption and a financial reward for reporting. Financial rewards have proved effective in other settings. In Sweden, Campos-Mercade et al. (2021) found that a modest payment increased actual COVID-19 vaccination uptake by 4.2 percentage points. Rewards also receive support in the whistleblowing literature: Butler et al. (2020) found that financial rewards substantially increased reporting of wrongdoing in a laboratory experiment.

However, financial payments could also change how reporting is perceived: someone who would otherwise be seen as protecting the community might instead be viewed as acting for personal gain. Bénabou and Tirole (2006) explain how rewards can discourage prosocial behaviour by casting doubt on people’s motives. In the case of electoral fraud, this concern may be particularly strong when reporting involves a neighbour or another community member. Accepting money to report someone one knows may feel like profiting from that person’s punishment

Our chosen methodology – a randomized experiment – is a gold standard for identifying causal effects of interventions because they create comparable groups exposed to different policies. Survey experiments are particularly appealing as an early stage of policy evaluation: they allow researchers to vary a message or proposed incentive while reaching a broad population at relatively low cost. As Stantcheva (2023) explains, surveys can create controlled variation and reveal beliefs and reasoning that administrative records do not capture. Their limitation here is equally important: we identify effects on stated willingness in hypothetical situations, rather than on actual reports or the incidence of electoral fraud.

A Survey Experiment Before the Election

We surveyed 1,550 adults aged 18 and older from different geographical areas. Interviews were conducted by telephone and in person, in Romanian and Russian. Respondents were assigned to one of the following experimental conditions: a baseline group with no additional message (521 people), a moral-message group (520), or a financial-reward group (509).

All respondents considered two hypothetical situations. In the first, they learned that someone from their locality had accepted money in exchange for voting for a particular candidate. They knew who the person was and when and where the payment occurred, and were told that authorities could substantiate it, for example through bank records. In the second, they witnessed a political group distributing cash to people transported by minibus to attend a rally and hold placards. They could identify the location and a vehicle registration number. The second situation concerns paid political mobilisation, rather than a direct exchange of money for a ballot.

In both situations, reporting was described as anonymous. Respondents rated their likelihood of reporting on a scale from 1, meaning “not at all likely,” to 10, meaning “definitely will report.” They were then asked about reporting channels or reasons for hesitation, depending on their response.

Before the first situation, the moral-message group heard that three out of four Moldovans in a recent survey had said corruption was never justified, and that corruption, including electoral corruption, harmed the country. This combines information about a prevailing moral view with a reminder of social harm. The financial-reward group was told in each situation that reporting would bring a guaranteed reward of 400 lei (about 20 EUR).

We considered two situations because the target of a report may matter. Reporting someone from one’s locality can expose an identifiable community member to punishment. A financial reward may then appear to turn disloyalty into a source of income. Reporting the political group that organises payments could be viewed differently, because the immediate target is an organisation rather than an individual neighbour. Fiorin (2023) provides closely related evidence from a field experiment with Afghan education employees asked to report absent colleagues. When participants were reminded that reports could lead to penalties, rewards reduced reporting from 15 to 10 percent. When reports were guaranteed not to reach the government, rewards instead increased reporting from 6 to 12 percent. The findings show that the anticipated harm to a peer can change how people respond to payment.

Financial Rewards Reduced Willingness to Report

Figure 1 reports the share of respondents selecting scores 8-10, on a 0 to 10 scale, which we describe as a high stated likelihood of reporting. The percentages use the survey weights; the error bars show 95 percent confidence intervals.

Figure 1. High stated likelihood of reporting by treatment and scenario

Source: Authors calculation

For the community-member scenario, high willingness to report was 38 percent in the baseline, 36 percent after the moral message, and 30 percent with the reward. For the political-group scenario, the corresponding shares were 43, 39, and 34 percent. The reward therefore reduced high reporting intentions by approximately eight percentage points in each situation.

Respondents were somewhat more willing to report the political group than the community member. However, the reward reduced willingness in both situations by almost the same amount. We find no evidence of the predicted larger reward effect when the reporting target is an organisation. This does not establish why the reward backfired. The follow-up answers nevertheless show that concerns regarding harm to in-group members are relevant. Among the 771 respondents selecting “not at all likely” in the community-member scenario, 23 percent said they did not want to harm those involved (18 percent in the political organisation scenario); 14 percent did not trust authorities to act; and 13 percent believed nothing would change. The largest group, 31.4 percent, said it was not their business.

Policy Implications

These findings highlight the importance of evaluating what works and under which conditions before introducing a policy more widely. An intervention that succeeds in one setting may be ineffective, or even counterproductive, in another. Evidence about policies that do not work is therefore valuable: it helps authorities avoid spending resources on ineffective measures and design policies with greater impact.

First, financial rewards should be tested before they are introduced. The guaranteed 400-lei reward in our experiment reduced willingness to report, even though reporting was explicitly described as anonymous. This does not mean that all reporting rewards will fail. It does show that their effectiveness cannot be assumed from successful applications elsewhere. Different amounts, payment conditions, or ways of presenting the reward would require further evaluation.

Second, low-cost moral messages also need to demonstrate their value. Reminding respondents that corruption is wrong and harms the country did not increase reporting intentions. Condemning corruption may be easier than reporting someone from one’s community. Future communications could address this decision more directly, explaining how reporting protects the community, what happens after a report, and how witnesses are protected. Testing these approaches would help authorities identify which messages encourage cooperation.

Third, more work is needed to understand the relationship between trust in public institutions and citizens’ willingness to cooperate with the state. A large proportion of respondents expressed low trust in the Central Election Commission or the police. Higher trust was associated with greater willingness to report, but our experiment cannot establish whether increasing trust would itself encourage cooperation.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

The Middle Corridor: A Useful Option, Not a Replacement

Posted on October 12, 2026 | Policy Brief

Geoeconomic concerns have made routes that avoid Russia a priority. The Middle Corridor – the land route connecting China to Europe while bypassing Russia – is the main land example. The history of energy transit shows both why such projects attract support and why many struggle. The recent interest in the Middle Corridor is due to the recent growth in traffic; freight has grown five- to tenfold since 2021. But this is from a very small base. Russia-transit rail still carries about 90% of China–EU rail freight, and sea carries the great majority of the total trade. Because of physical and institutional limits, the corridor will remain a small share of total China–EU trade, even if its role were to continue to grow. It deserves targeted support as insurance against disruption, and to increase connectivity and development in the South Caucasus and Central Asian region. But it is important to keep in perspective the role it can play in aggregate trade between China and Europe. More important changes are more likely to come at sea.

Since Russia’s full-scale invasion of Ukraine in 2022, the so-called Middle Corridor has moved from the margins of transport policy to the center of attention. The route runs from China through Kazakhstan, across the Caspian Sea to Azerbaijan, and on through Georgia to Turkey and Europe; see Figure 1. Sometimes other alternatives are included, such as routes through Uzbekistan and Turkmenistan, and also through Armenia, but the common point is routes that avoid Russia. The EU has pledged billions of euros for connectivity with Central Asia, the World Bank has backed an upgrade of Kazakhstan’s railways, China has begun to engage with the route, and freight has grown from roughly 0.5–0.84 million tons in 2021 to about 4.5–5 million tons in 2024–25 (the exact figures vary across sources and also depend on the metrics used, but the bottom line remains: the growth has been 5-10 times the volume since 2021).

Figure 1. The Middle Corridor

Source: https://middlecorridor.com/en/route

Recent events have brought increased attention to the route and its importance. Poland’s closure of its border with Belarus in September 2025 briefly stopped most China–EU rail traffic, and attacks in the Red Sea and disruptions around Hormuz have raised doubts about the sea lanes. Also, recent events around Georgian harbors, such as the delay of the Anaklia deep-sea port can be seen in light of the Middle Corridor discussion. It is tempting to read these developments as a new trade route displacing the old ones. This brief argues for a more cautious reading.

A Geoeconomic World: Lessons from Energy Transit

The interest in the Middle Corridor is an example of a broader shift. Policy makers and academics alike are trying to navigate a new world order. The “flat” world of just-in-time global production and transport chains is giving way to a more complicated one in which resources, production capacity and geography can all be “weaponized”. Interest in geoeconomics, the use of economic means for strategic ends, has grown accordingly (Hirschman, 1945; Farrell and Newman, 2019; Aiyar et al., 2023a, 2023b; Mohr and Trebesch, 2025; Snower, 2026a, 2026b).

For those who study Eastern Europe and Russia, and energy and resource economics in the region in particular, this perspective is not entirely new. Control over chokepoints in transit routes has long been central to strategic concerns, and energy transit offers a long record of how geography becomes leverage. Three lessons stand out.

Transit Creates Rents and Exposure on Both Sides

Russia has repeatedly used gas and oil transit as a bargaining chip and to influence neighboring states. Just to give a few examples from the past decades: gas flows through Ukraine were stopped for a few days in January 2006 and for almost two weeks in January 2009, affecting more than 20 countries (Bloomberg, 2025); in 2006 Russia also shut the oil pipeline to Lithuania’s Mažeikių refinery (UNIAN, 2008); disputes with Belarus in 2004, 2007 and 2010 interrupted transit twice, gas in 2004 and oil in 2007 (OIES, 2010). Moscow said it was moving neighbors from subsidized to market prices, while many European politicians saw political pressure (RFE/RL, 2010). Leverage also ran the other way. Ukraine declined to renew the gas transit contract, which expired on 1 January 2025 after carrying 148 bcm in five years and paying Ukraine about $6.2 billion in fees (GasUA, 2025). Incidents such as these created a concern for European energy security, leading to important research attempting to quantify this (Le Coq and Paltseva, 2009).

Those Who Depend on Transit Try to Bypass It

Russia’s answer to its disputes with transit states was to propose, and later build, pipelines that avoid them, such as Nord Stream under the Baltic and TurkStream across the Black Sea (UNIAN, 2008; Bloomberg, 2025). The search for routes that avoided Russia followed the same logic, including the routes that overlap with the Middle Corridor, such as the Nabucco pipeline.

Political Ambition Does Not Replace a Commercial Anchor

The Nabucco pipeline, meant to bring Caspian gas to Europe without crossing Russia, was abandoned in June 2013 when the Shah Deniz II consortium chose the shorter and cheaper Trans-Adriatic Pipeline (Elcano, 2013). The chosen pipeline had lower capacity than Nabucco would have had, and EU member states were divided, with several signing bilateral deals with Russia on the rival South Stream project (ECFR, 2014). The Southern Gas Corridor was built, but in a smaller form than its political backers had hoped.

Each lesson has a counterpart in the Middle Corridor: transit states that gain rents and leverage, users with options and a rival with an interest in complicating them, and a gap between political support and commercial viability.

What to Make of the Interest in the Middle Corridor

Before 2022 the northern route through Russia and Belarus carried up to 86% of China–EU rail freight, while the Middle Corridor carried less than 1%. In 2025 about 90% of China–EU rail freight still crossed into the EU via Belarus and Poland, and the Middle Corridor’s capacity is only about 5–6% of the northern route’s (see, e.g. Atlantic Council, 2025). Seen in this light, the recent interest in the Middle Corridor has three drivers.

As in the energy case, the route also redistributes leverage. Kazakhstan, Azerbaijan, Georgia and Türkiye become transit states with rents to gain and pressures to manage, and Armenia may join them. The policy question is how much of the old geoeconomics the new route can actually change. The rest of this brief takes up that question: what has happened so far, how large the corridor is in proportion to the alternatives, what limits its growth, and what follows for policy.

What Has Happened

The corridor is growing steadily, and two episodes, the Polish border closure and the Georgian port saga, show both its value and its limits.

Current Volumes

Container traffic on the route totalled 53,574 TEU in January–August 2026, up 11% from 48,326 TEU a year earlier, carried on 265 container block trains. About 80% moved westward, and only 20% back to the east (Times of Central Asia, 2026; Report.az, 2026, citing Azerbaijan Railways’ ADY Express). Industry sources cited in October 2025 put monthly capacity at under 10,000 TEU, against demand of about 40,000 TEU (Times of Central Asia, 2025).

The Polish Border Closure, September 2025

Poland closed its border with Belarus on 12 September 2025, after Russian–Belarusian exercises and a drone incursion, and reopened it two weeks later. About 90% of China–EU rail freight passes through the crossing at Małaszewicze, so the closure stopped most of the flow, with more than 130 trains stranded. China’s response was mainly diplomatic: Foreign Minister Wang Yi went to Warsaw and Politburo member Li Xi to Minsk (Times of Central Asia, 2025). Some forwarders offered the southern route as a contingency, and a Chinese carrier sent a container ship through the Arctic that month, but the alternatives absorbed only single-digit percentages of the flow (GetTransport, 2026). The disruption came from a transit state acting on security grounds, a reminder that transit leverage is not Russia’s alone.

Georgia and New Branches

Georgia is the gateway to Europe and the most visible bottleneck. The Anaklia deep-sea port has been delayed by a decade of political dispute. In July 2026, Georgia announced that its Chinese-led partner had exited, with first operations now targeted for 2029 and no confirmed operator (OC Media, 2026). Two additions would widen the corridor: the US-brokered TRIPP route through Armenia, which bypasses Georgia and could open around 2028, and a southern entry point via Uzbekistan and Turkmenistan. Both add redundancy rather than a different order of magnitude.

Keeping Proportions

Even spectacular growth would leave the Middle Corridor a small share of China–EU trade, because the routes it competes with start from a vastly larger base, see Table 1.

Table 1. Relative scale of routes between China and the EU

Route Rough scale Note
Sea (all China–EU) Over 90% of volume Estimate; the default mode for bulk and most containers
Rail via Belarus and Poland (Northern route) About €25 bln in 2024, 3.7% of EU–China trade by value About 90% of China–EU trains cross at Małaszewicze
Rail via the Middle Corridor About 54,000 TEU in Jan–Aug 2026; 4.5–5 mln tons in 2025 (all cargo types) A minor share even within rail

Source: Northern route figures come from Asia Times, and Middle Corridor figures come from Times of Central Asia, 2026; the sea share is approximate.

Three points follow.

  1. Russia transit is still the main land route. Rail through Belarus has not been legally blocked and remains the cheapest, fastest and highest-capacity option. Volumes dipped in 2025, partly because of the Polish closure and partly because sea freight regained price competitiveness (Upply, 2026), not because traffic moved south.
  2. Fast growth from a small base is compatible with a small share. If the Middle Corridor reached one-third of the northern route’s current value, it would be about 1% of EU–China trade, and matching the northern route would be about 3.7%. Sea would still carry the great majority.
  3. Rail suits particular cargo. High-value, time-sensitive goods and e-commerce, not bulk, which limits how much can shift to land.

Why the Corridor Will Not Scale into a Replacement

The binding constraints are physical, financial, and institutional, and none can be removed quickly.

  • Capacity and cost. Monthly capacity is under 10,000 TEU, and the World Bank estimates that upgrades will cost $27–29 billion over 15 years (Times of Central Asia. 2025).
  • Multiple handovers. Cargo changes mode and jurisdiction several times, and each handover adds delay and risk. Border and customs congestion is common, including on the Chinese–Kazakh border.
  • Georgian ports. Until a deep-water port operates, the western sea leg cannot use the largest Black Sea vessels.
  • Directional imbalance. About 80% of container traffic runs westward, while return trips on the China–Europe rail network as a whole now match outbound trips (GetTransport, 2026). That weakens the corridor’s commercial case without subsidies.

The Bigger Picture: Sea Routes and the Arctic

The larger uncertainty for China–EU trade lies at sea, where the choice between Suez, the Cape and an Arctic lane will matter far more than anything on the Caspian. After the Red Sea attacks the Cape became the default for many carriers, and prices eased once supply chains adjusted (Upply, 2026).

The Northern Sea Route is the most talked-about new option. In summer 2026 Sea Legend launched what it calls the first scheduled seasonal container service between China and northern Europe, with eight roughly weekly sailings and a targeted transit of 20–22 days, about 15 days faster than Suez on a Shanghai–Rotterdam voyage (Cyprus Mail, 2026; Arctic Today, 2026).

The evidence so far supports reading it as a growing seasonal complement, not yet a replacement for southern lanes. Announced container capacity this season is about 20,000 TEU, small beside Asia–Europe flows of many millions of TEU (approximate), and the season runs only from roughly July to October. The route also follows Russia’s coast, requires Rosatom (Russian State Atomic Energy Corporation) permits, and in 2026 relies partly on Russian nuclear icebreakers (Port de Barcelona, 2026), so it is not a way around Russian influence. If global warming and investment continue, the Arctic could matter far more in the 2030s, which is another reason to keep the Middle Corridor in proportion.

Policy Implications

The Middle Corridor deserves targeted support as insurance against disruption, but policy should not assume that it can replace routes through Russia or the sea.

  1. Frame it as resilience, not replacement. The 2025 episode showed that the northern route has a single point of failure and that the alternatives absorb only a small fraction of the flow. The goal is an alternative good enough to cushion a shock.
  2. Fund the binding bottlenecks first, and insist on a commercial anchor. Border and customs procedures, Caspian and Georgian port capacity and rail throughput are likely to offer the best returns. Nabucco and Anaklia are warnings against large, politically driven projects without a committed supplier or operator.
  3. Plan for transit-state risk on every route. The energy record shows that leverage over transit runs in several directions, and the Polish closure shows that it is not Russia’s alone. Contingency plans for the northern route, and spreading risk across several transit states, matter more than hoping the Middle Corridor can absorb a closure.
  4. Be aware of the backhaul problem. The corridor’s economics also depend on European and Turkish cargo going east. At current levels the lower west to east demand for freight capacity increases cost for this route.
  5. Build a common evidence base. Operators should agree consistent statistics, in tons and TEU, with shared definitions.
  6. Keep the maritime picture in view. Red Sea security, Suez recovery and Arctic shipping will shape China–EU trade far more than any land corridor in the foreseeable future.

Notes on Data and Sources

The proportions in this brief are robust to the uncertainty in the data, but individual figures should be quoted with care.

  • Units and baselines. Tons include bulk, while TEU counts containers only. Sources put 2021 freight at roughly 0.5–0.84 million tons, so the rise to 4.5–5 million is five- to tenfold.
  • Container volumes. The ADY Express count (53,574 TEU in January–August 2026, about 80,000 annualised) is lower than the 130,000–300,000 TEU a year found in some commercial datasets that define the route more broadly. The conclusion holds under either.
  • Train counts. The 265 Middle Corridor block trains in January–August 2026 (about 400 annualised) compare with 20,022 China–Europe train trips in 2025 on all routes, roughly 2%, though definitions differ.

References

Internet sources

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

Mothers’ Silent Exit: Bereavement in Armenia after the 2020 Nagorno-Karabakh War

Posted on October 5, 2026 | Policy Brief

This policy brief examines how wartime bereavement affected political participation in Armenia after the 2020 Nagorno-Karabakh war. Using linked administrative voter-roll and casualty records and an original Computer-Assisted Telephone Interviewing (CATI) survey of 1,676 parents, we show that losing a conscript son reduced turnout much more for mothers (28.7 percentage points) than for fathers (12.4 percentage points) in the 2021 election. The effect reflects bereavement rather than military service alone, and a within-household comparison shows mothers withdrew 10.6 percentage points more than fathers. Survey evidence points to emotional loss as the mechanism, especially among women, rather than logistical barriers or ordinary dissatisfaction. Post-war democracies should therefore treat bereaved women’s abstention as a democratic inclusion problem and create credible, non-partisan, trauma-informed channels for voice.


Why the gender gap matters

Wars are often expected to reshape elections through blame and punishment. In Armenia, the conditions seemed ideal for retrospective accountability: defeat in the 2020 Nagorno-Karabakh war, territorial loss, and roughly 3,800 dead soldiers. Yet the incumbent government survived the 2021 snap parliamentary election, winning 54.0 percent of the vote after receiving 70.4 percent in 2018. A vote-share reading suggests that the public did not punish the government enough. Turnout evidence suggests a different interpretation: many with the strongest personal grievance left the electorate.

This brief focuses on the gender gap in that withdrawal. The central finding is not simply that bereaved parents voted less. It is that mothers withdrew much more than fathers, including within the same affected households. This matters because elections only show the choices of people who vote. Grief may instead lead some citizens to stay at home, so their dissatisfaction does not show up in election results. In Hirschman’s (1970) terms, post-war grievance can become exit rather than voice.

Research on grief suggests that men and women may cope with the loss of a loved one in different ways. Some studies find that men are more likely to respond through practical action, while women are more likely to express grief emotionally (Glick et al., 1974; Martin & Doka, 2000; Martin & Wang, 2006; McDowell et al., 1988).

Research on elections also shows that men and women can differ in whether they vote and which parties they support (Abendschoen & Steinmetz, 2014; De Vaus & McAllister, 1989; Duverger, 1955; Inglehart & Norris, 2000; Jelen et al., 1994; Morgan-Collins, 2023; Norris, 1996; Randall, 1987; Rokkan & Lipset, 1967; Skorge, 2023; Spierings & Zaslove, 2015). However, we know much less about whether war-related loss changes men’s and women’s voting behavior in different ways, and what explains these differences.

Voting Records and Survey Evidence from War-Affected Parents

Our analysis in this policy brief uses two different sources of information. First, we linked official voting records from the 2017, 2018, and 2021 elections, covering about 2.5 million registered voters, with records on military casualties and their parents. This allows us to see whether people actually voted, rather than relying on what they later report in a survey. This is important because people often overreport voting in surveys (Ansolabehere & Hersh, 2012). We then compare parents whose sons died in the war with similar parents whose sons were of conscription age during the war and survived. We define the war cohort as men born on or before 30 June 2002, who were old enough to be called up in spring 2020 and were therefore already of service age when the war began on 27 September 2020. Men born after this cutoff generally reached conscription age only after the fighting ended.

We focus on parents rather than wives or partners for two main reasons. First, most of the soldiers who died were very young, typically around ages 18–20, and many were therefore not married. Second, available records allow us to identify parents with relatively high accuracy. In particular, sons’ surnames and recorded father names allow for precise matching with fathers. Identifying wives or partners is more difficult with the information currently available and would require additional data collection.

Second, we collected an original Computer-Assisted Telephone Interviewing (CATI) survey of 1,676 parents of war-cohort sons in April-May 2026. The survey records vote recall for 2018 and 2021, voting intentions for 2026, blame attribution for the defeat, reasons for not voting, and detailed information on the son’s service type and wartime status. Because recall and stated voting can be biased, the survey is not used as a standalone causal estimate of turnout levels. Its purpose is to identify the mechanism behind the administrative result: whether lost incumbent support becomes opposition voting, non-voting, or broader political disaffection.

Figure 1. Turnout trajectories for bereaved and control fathers and mothers

Source: Authors’ calculations using linked administrative records.

The results show a clear difference between fathers and mothers. Between 2018 and 2021, losing a son reduced fathers’ turnout by 12.4 percentage points and mothers’ turnout by 28.7 percentage points, as illustrated by Figure 1.

These first estimates come from separate analyses for fathers and mothers. This means that we compare bereaved fathers with other fathers, and bereaved mothers with other mothers. We are not necessarily comparing the mother and father of the same soldier.

We therefore also look at true couples, meaning mothers and fathers who live in the same household. This allows us to compare parents who experienced the same loss and shared the same household environment. In this within-couple comparison, mothers were 10.6 percentage points less likely to vote than fathers after the loss of their son; see Figure 2. The similar patterns before the war, between 2017 and 2018, also suggest that bereaved and comparison families were not already moving in different directions before the war.

Several additional results help explain the gender gap. First, the decline in turnout seems to come from losing a son, rather than simply from having a son who was old enough to serve during the war. Parents whose sons were of conscription age during the war but survived did not vote less than parents whose sons reached conscription age only after the ceasefire. Since we do not observe actual conscription, we cannot identify exactly which sons were called up to serve. Second, most bereaved parents did not disappear from the voter register. About 99 percent of the effect is intensive: bereaved parents remain registered but do not vote. Third, the effect is not limited to parents. Sisters in bereaved families also became less likely to vote after the war, even though they were not themselves at risk of conscription; see Figure 2.

Figure 2. Estimated effects on post-war turnout

Source: Authors’ calculations using linked administrative records. The figure shows the estimated change in the probability of voting after the war for different groups. The vertical axis reports the effect in percentage points. Negative values indicate lower turnout after the war. Points show the estimated effects, and vertical lines show 95% confidence intervals.

Our findings also add to what we already know about how war affects political behavior. Earlier studies show that wartime casualties can change how people vote, how they view governments, and how they take part in politics (Gartner & Segura, 1998; Grossman et al., 2015; Karol & Miguel, 2007; Koch, 2011; Kriner & Shen, 2010). Other research finds that experiencing conflict can sometimes make people more politically active (Acemoglu et al., 2022; Bellows & Miguel, 2009; Blattman, 2009; Gilligan et al., 2014; Voors et al., 2012). Our results show an important difference: while war may encourage participation among some people, the personal loss of a child can instead lead bereaved family members to withdraw from voting. This also differs from earlier research showing that political participation can spread within families (Bhatti et al., 2017; Dahlgaard, 2018; Fieldhouse & Cutts, 2012; Golder & Golder, 2016; Nickerson, 2008). In our case, the family effect works in the opposite direction: withdrawal from voting appears to spread within the household.

Mechanism: emotional loss, not ordinary dissatisfaction

The survey evidence we collected helps explain why mothers withdraw more. The mechanism graph (Figure 3) shows coefficients for several self-reported reasons for not voting. War casualty exposure is especially strongly associated with female respondents citing emotional loss as compared to males (p=0.001). By contrast, logistical barriers, personal reasons, distrust in politics, and disappointment with candidates do not show the same clear female-specific increase.

Figure 3. Mechanism evidence from the original survey

Source: Authors’ original CATI survey, April-May 2026.

This distinction matters for policy. If abstention were mainly logistical, transportation and registration assistance would be enough. If abstention were mainly candidate disappointment, party-system reform would be the central issue. But if abstention is driven by emotional loss, then democratic inclusion after war also requires trauma-informed, gender-sensitive institutions. The point is not to pressure bereaved mothers to vote. It is to make sure the state does not misread withdrawal as consent.

Policy recommendations

These findings have implications for institutions responsible for democratic participation and support for war-affected families. Election authorities, public institutions, and civil-society organizations should ensure that political withdrawal among bereaved citizens does not become invisible. The objective is not to mobilize grief for electoral purposes, but to provide credible and non-partisan opportunities for bereaved families to remain heard in public life.

  1. Monitor gendered turnout after conflict. Election authorities, observers, and researchers should look beyond aggregate vote shares and report turnout patterns by gender and affectedness where data protection allows.
  2. Build trauma-informed civic outreach. Non-partisan information campaigns should avoid blaming abstainers and instead lower psychological and administrative barriers to participation, especially for bereaved mothers.
  3. Create non-electoral channels for voice. Public hearings, local consultation forums, and memorial-policy processes can give bereaved families ways to be heard between elections, reducing the burden on a single election day.
  4. Use turnout gaps as an early-warning indicator. Election authorities and researchers should monitor whether turnout falls disproportionately among war-affected women after conflict. Persistent gaps can identify groups whose political preferences may be becoming underrepresented and can help target further research, consultation, and non-partisan civic outreach.

Conclusion

The gender gap after war is not only about which parties men and women support. It is also about whether they vote at all. In Armenia, losing a son reduced turnout for both fathers and mothers, but the decline was much larger among mothers. The evidence suggests that this was mainly a response to bereavement itself, rather than simply having a son of military age, and that emotional loss played an important role.

This has an important policy implication. Election results do not tell the full story if people who are most affected by war stop voting. When bereaved mothers withdraw from elections, their views become less visible in the political process. Post-war policies should therefore pay attention to these differences in participation, keep support for bereaved families separate from party politics, and create trusted, non-partisan ways for affected families to make their concerns heard.

References

  • Abendschoen, S. and Steinmetz, S. (2014). The gender gap in voting revisited: Women’s party preferences in a European context. Social Politics, 21(2):315-344.
  • Acemoglu, D., De Feo, G., De Luca, G., and Russo, G. (2022). War, socialism, and the rise of fascism: An empirical exploration. Quarterly Journal of Economics, 137(2):1233-1296.
  • Ansolabehere, S. and Hersh, E. (2012). Validation: What big data reveal about survey misreporting and the real electorate. Political Analysis, 20(4):437-459.
  • Bellows, J. and Miguel, E. (2009). War and local collective action in Sierra Leone. Journal of Public Economics, 93(11-12):1144-1157.
  • Bhatti, Y., Fieldhouse, E., and Hansen, K. M. (2017). It is the family that votes: How voting is transmitted across generations. Political Behavior, 39(3):529-552.
  • Blattman, C. (2009). From violence to voting: War and political participation in Uganda. American Political Science Review, 103(2):231-247.
  • Dahlgaard, J. O. (2018). Trickle-up political socialization: The causal effect on turnout of parenting a newly enfranchised voter. American Political Science Review, 112(3):698-705.
  • De Vaus, D. and McAllister, I. (1989). The changing politics of women: Gender and political alignment in 11 nations. European Journal of Political Research, 17(3):241-262.
  • Duverger, M. (1955). The Political Role of Women. UNESCO, Paris.
  • Fieldhouse, E. and Cutts, D. (2012). The companion effect: Household and local context and the turnout of young people. Journal of Politics, 74(3):856-869.
  • Gartner, S. S. and Segura, G. M. (1998). War, casualties, and public opinion. Journal of Conflict Resolution, 42(3):278-300.
  • Gilligan, M. J., Pasquale, B. J., and Samii, C. (2014). Civil war and social cohesion: Lab-in-the-field evidence from Nepal. American Journal of Political Science, 58(3):604-619.
  • Glick, I. O., Weiss, R. S., and Parkes, C. M. (1974). The first year of bereavement. Journal of Consulting and Clinical Psychology, 41(2):181-186.
  • Golder, S. N. and Golder, M. (2016). Exit, voice, and political change. Political Science Research and Methods, 4(2):411-424.
  • Grossman, G., Manekin, D., and Miodownik, D. (2015). The political legacies of combat: Attitudes toward war and peace among Israeli ex-combatants. International Organization, 69(4):981-1009.
  • Hirschman, A. O. (1970). Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Harvard University Press, Cambridge, MA.
  • Inglehart, R. and Norris, P. (2000). The developmental theory of the gender gap: Women’s and men’s voting behavior in global perspective. International Political Science Review, 21(4):441-463.
  • Jelen, T. G., Thomas, S., and Wilcox, C. (1994). The gender gap in comparative perspective. European Journal of Political Research, 25(2):171-186.
  • Karol, D. and Miguel, E. (2007). The electoral cost of war: Iraq casualties and the 2004 U.S. presidential election. Journal of Politics, 69(3):633-648.
  • Koch, M. T. (2011). Casualties and incumbents: Do the casualties from interstate conflicts affect incumbent party vote share? British Journal of Political Science, 41(4):795-817.
  • Kriner, D. L. and Shen, F. X. (2010). The Casualty Gap: The Causes and Consequences of American Wartime Inequalities. Oxford University Press, Oxford.
  • Martin, T. L. and Doka, K. J. (2000). Men don’t cry … women do: Transcending gender stereotypes of grief. Death Studies, 24(1):1-18.
  • Martin, T. L. and Wang, J. (2006). Gender and bereavement. Omega: Journal of Death and Dying, 53(3):227-240.
  • McDowell, E. E., Saylor, C., and Taylor, M. J. (1988). Patterns of grief in men and women. Omega: Journal of Death and Dying, 18(3):213-222.
  • Morgan-Collins, M. (2023). The electoral impact of newly enfranchised women. Journal of Politics, 85(3):1041-1055.
  • Nickerson, D. W. (2008). Is voting contagious? Evidence from two field experiments. American Political Science Review, 102(1):49-57.
  • Norris, P. (1996). Mobilizing the women’s vote: The gender-generation gap in voting behaviour. Parliamentary Affairs, 49(2):333-342.
  • Randall, V. (1987). Women and Politics: An International Perspective. Macmillan, London.
  • Rokkan, S. and Lipset, S. M. (1967). Party Systems and Voter Alignments: Cross-National Perspectives. Free Press, New York.
  • Skorge, O. S. (2023). Mobilizing the underrepresented: Electoral systems and gender gaps in political participation. American Political Science Review, 117(2):538-552.
  • Spierings, N. and Zaslove, A. (2015). Gendering the vote for populist radical-right parties. Patterns of Prejudice, 49(1-2):135-162.
  • Voors, M. J., Nillesen, E. E. M., Verwimp, P., Bulte, E. H., Lensink, R., and Van Soest, D. P. (2012). Violent conflict and behavior: A field experiment in Burundi. American Economic Review, 102(2):941-964.

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

Recognizing Unpaid Care: Family-Friendly Policies for Armenia

Posted on September 29, 2026 | Policy Brief

Unpaid care work is a cornerstone of Armenia’s care economy, sustaining families, supporting human development, and underpinning the functioning of society. However, unpaid care work remains largely invisible in economic policymaking and continues to fall disproportionately on women, limiting their participation in the labor market and reinforcing gender inequalities. As Armenia faces population aging, a shrinking labor force and growing care needs, recognizing and investing in the care economy has become both a demographic and an economic imperative. 

Drawing on recent evidence from Armenia, including National Transfer Accounts and National Time Transfer Accounts Surveys, as well as international experience, this policy brief examines how family-friendly policies can help recognize, reduce, and redistribute unpaid care work. It argues that expanding childcare services, promoting shared parental responsibilities, fostering family-friendly workplaces, and integrating care into national policy planning can strengthen women’s economic participation, improve family well-being, and enhance demographic resilience. Investing in the care economy is therefore not only a matter of gender equality, but a strategic investment in Armenia’s sustainable and inclusive development.

Why Unpaid Care Matters for Armenia

Armenia is experiencing profound demographic changes, including below-replacement fertility, population aging, sustained outmigration, and a shrinking working-age population (Statistical Committee of the Republic of Armenia, 2025). These trends are increasing pressure on labor markets, social protection systems, and public finances while raising the importance of maximizing the country’s human capital. In this context, strengthening women’s economic participation is not only a matter of gender equality but also a demographic and economic imperative.

Unpaid care work lies at the center of this challenge. Every day, households rely on unpaid childcare, eldercare, care for persons with disabilities, and domestic work to sustain families and communities. Although these activities are essential for social and economic well-being, they remain largely invisible in conventional economic statistics and continue to be performed predominantly by women (Ameria Management Advisory & United Nations Population Fund, 2024). As a result, women face a disproportionate burden of caregiving responsibilities that limits their participation in paid employment, reduces lifetime earnings, constrains career progression, and contributes to persistent gender inequalities in the labor market. Recent demographic evidence suggests that Armenia’s future prosperity will depend less on increasing population size than on investing in people and creating conditions that enable individuals to reconcile work and family life. Family-friendly policies, including accessible childcare services, equitable parental leave, flexible work arrangements, and measures that promote shared caregiving, can help reduce barriers to women’s employment while supporting families in realizing their reproductive intentions (Fluchtmann, van Veen, & Adema, 2023).

Recognizing, reducing, and redistributing unpaid care work should therefore become a central component of Armenia’s demographic and socio-economic policy agenda. Investing in the care economy is not solely a social welfare measure; it is a strategic investment in human capital, labor productivity, gender equality, and long-term demographic resilience. By integrating care considerations into employment, demographic, and social protection policies, Armenia can better respond to the challenges of population aging while strengthening inclusive and sustainable economic development.

What Evidence Shows

Recent evidence from Armenia’s first National Transfer Accounts (NTA) (2024) and National Time Transfer Accounts (NTTA) (2026) shows that unpaid care work is one of the country’s largest yet least recognized economic resources. Together with national labor market statistics, these findings demonstrate that the unequal distribution of unpaid care is not only a gender equality issue but also a constraint on economic productivity, labor force participation, and demographic resilience (Statistical Committee of the Republic of Armenia, 2025).

 Figure 1. Unpaid Care Work in Armenia: Key Facts

Sources: UNFPA (2024, 2026), National Transfer Accounts (NTA), National Time Transfer Accounts (NTTA); National Statistical Committee of Armenia (2025); OECD, UN Women, ILO, ESCAP.

Unpaid care is a major contributor to Armenia’s economy

The NTTA survey has revealed that women in Armenia perform 24.1 hours of unpaid care and household work per week, compared with 6.3 hours for men, nearly four times more. Women’s unpaid care commitments peak at approximately 44-49 hours per week during the ages of 28-36, coinciding with the years of greatest career advancement and family formation. This overlap highlights how unequal caregiving responsibilities reduce women’s economic participation precisely when their contribution to the labor market is most critical for productivity and demographic resilience. Household tasks account for the largest share of this work, while women also spend 3.8 times more time on childcare than men. The disparity is particularly pronounced during the peak parenting years (ages 23-27), when women devote more than ten times as many hours to childcare.

When this work is assigned an economic value, its contribution becomes impossible to ignore. Depending on the valuation method, unpaid care represents between 16.8% and 46.9% of Armenia’s GDP, reflecting whether care time is valued at low-cost market replacement wages or individuals’ higher potential earnings, exceeding the output of several major formal sectors, including manufacturing and trade. Women generate approximately 80% of this economic value, showing that the care economy continues to rely overwhelmingly on women’s unpaid labor.

Unpaid care constrains women’s economic participation

The unequal distribution of care responsibilities has direct consequences for women’s participation in the labor market. The NTA survey reveals that, unlike men, women do not experience a positive lifecycle surplus at any age, meaning that at no point in the lifecycle does women’s labor income exceed their consumption. Because heavy unpaid care duties limit their paid employment, women’s immense economic contribution remains uncompensated in formal accounts.

This pattern is also reflected in national labor market statistics. Although women have higher rates of tertiary education enrolment than men, 51% of women remain economically inactive, compared with 29% of men (Statistical Committee of the Republic of Armenia, 2025). Women are also significantly underrepresented in leadership and decision-making positions. These disparities indicate that educational attainment alone is insufficient to ensure equal economic opportunities when unpaid care responsibilities remain unequally distributed.

Demographic change increases the urgency for reform

Population aging is expected to increase demand for caregiving while reducing the working-age population (Lee, 2025). NTA projections show that intergenerational transfers are shifting away from investment in children toward supporting older persons. By 2050, an estimated 60% of intergenerational transfers will go toward supporting older adults. This represents a complete reversal of current trends, which prioritize investments in children. Without reforms that strengthen formal care services and enable greater labor market participation, particularly among women, the growing demand for unpaid care is likely to intensify existing labor shortages and place additional pressure on families and public finances.

Collectively, these findings demonstrate that unpaid care work should be recognized as a productive economic activity rather than a private household responsibility. Addressing the unequal distribution of care through family-friendly policies is therefore essential not only for advancing gender equality but also for strengthening Armenia’s economic resilience and long-term demographic sustainability.

Current Policy Gaps

Armenia has established an important policy foundation for addressing demographic change, gender equality, employment, and family well-being. The Demographic Strategy of Armenia (2024-2040), the Employment Strategy (2025-2031), the Gender Policy Implementation Strategy (2025-2028) (Government of the Republic of Armenia, 2024), and Labor Legislation (National Assembly of the Republic of Armenia, 2004) all recognize the importance of supporting families, promoting women’s employment, and improving work-life balance. Collectively, these policy frameworks provide a strong basis for advancing family-friendly policies and responding to demographic challenges.

However, important implementation gaps continue to limit their effectiveness. While maternity protection and childcare leave are well established, fathers’ participation in caregiving remains limited because paternity leave is short (5 working days) and there is no dedicated, non-transferable parental leave. As a result, caregiving responsibilities continue to fall disproportionately on women, reinforcing traditional gender roles and contributing to career interruptions following childbirth.

Access to affordable and quality childcare services also remains uneven, particularly for children under three years of age, children with disabilities, and families living in rural communities. Limited childcare availability constrains parents’ – especially mothers’ ability to enter or remain in the labor market, despite high levels of educational attainment among women (United Nations Children’s Fund & Global Partnership for Education, 2022).

Although Armenian labor legislation provides opportunities for flexible work arrangements, their implementation remains inconsistent across sectors, and family-friendly workplace practices are not yet widely institutionalized. Flexible working hours, hybrid work models, phased return-to-work arrangements and workplace support for employees with caregiving responsibilities remain largely dependent on individual employers rather than forming part of a broader national approach.

Finally, policy reforms continue to operate within a social context where caregiving is widely perceived as a woman’s primary responsibility. Persistent gender stereotypes influence both household decision-making and workplace expectations, reducing the effectiveness of existing policy measures and limiting fathers’ engagement in childcare and other caregiving responsibilities. International experience demonstrates that legislative reforms are most effective when accompanied by measures that promote behavioral change, employer engagement, and shared caregiving norms. (de la Porte et al., 2022; UNFPA EECARO, 2021)

Addressing these gaps requires moving beyond individual policy measures toward an integrated family-friendly policy framework that recognizes unpaid care as a shared societal responsibility. Such an approach would better align demographic, labor market, social protection, education, and gender equality policies while creating enabling conditions for both women and men to participate fully in paid work and family life.

Five Priority Policy Actions

  1. Expand Access to Affordable and Inclusive Childcare Services

Expanding access to quality, affordable, and inclusive childcare services should be a national priority for strengthening women’s labor force participation and supporting demographic resilience. Although Armenia has made progress in increasing preschool enrollment, significant gaps remain for children under three years of age, children with disabilities, and families living in rural communities. These gaps often force women to reduce working hours, postpone career advancement, or leave employment altogether.

International experience demonstrates that investment in early childhood education and care is one of the most effective policy instruments for increasing female employment, reducing poverty, supporting child development, and improving long-term economic productivity (Fluchtmann et al., 2023; UNICEF & Global Partnership for Education, 2022). Armenia could adopt a phased approach that expands community-based childcare services, increases public investment in early childhood education, strengthens inclusive service provision, and introduces targeted childcare subsidies for vulnerable families.

Figure 2. Investing in the care economy for gender equality and demographic resilience in Armenia

Source: Authors’ analysis based on data from Armenia NTA 2024, NTTA 2026, and international evidence

  1. Promote Shared Caregiving through More Equitable Parental Leave

Achieving a more balanced distribution of unpaid care requires greater participation of fathers in childcare from the earliest stages of parenthood. Armenia’s current parental leave framework continues to place primary caregiving responsibilities on mothers, while existing paternity leave provisions remain too limited to influence caregiving patterns.

International experience, particularly from the Nordic countries, demonstrates that dedicated, adequately compensated, non-transferable leave for fathers substantially increases paternal involvement in childcare and contributes to more equal sharing of unpaid care throughout children’s lives (United Nations Population Fund [UNFPA EECARO], 2021). Although replicating Nordic-style universal wage replacement is limited by public budget constraints and entrenched traditional caregiving norms, Armenia can take high-impact incremental steps by introducing dedicated, non-transferable leave for fathers and strengthening targeted financial benefits.

  1. Foster Family-Friendly Workplaces

Creating family-friendly workplaces is essential for enabling parents and other caregivers to reconcile paid employment with family responsibilities. Flexible work arrangements, including hybrid work, flexible scheduling, phased return-to-work after parental leave, and breastfeeding support, have become increasingly important tools for improving work-life balance and retaining skilled workers.

Beyond legislative provisions, Armenia could encourage employers to voluntarily adopt family-friendly workplace practices through incentive mechanisms, recognition programs, or certification schemes. Such approaches would benefit employers by improving staff retention and productivity while supporting greater labor market participation among women and other caregivers.

  1. Integrate Unpaid Care into National Economic and Demographic Policy

The findings from Armenia’s National Transfer Accounts and National Time Transfer Accounts demonstrate that unpaid care constitutes a substantial contribution to the national economy. Yet this contribution remains largely invisible in policy planning and economic decision-making.

Recognizing unpaid care as a productive economic activity would support more evidence-based policymaking across employment, demographic, social protection, and fiscal policies. Armenia should institutionalize regular Time Use Surveys, continue developing National Time Transfer Accounts, strengthen sex-disaggregated data collection and integrate unpaid care indicators into national development monitoring. Better measurement would enable policymakers to assess the economic returns of investments in childcare, family-friendly policies, and gender equality.

  1. Transform Social Norms through Public Awareness and Institutional Change

Policy reforms alone cannot achieve a more equitable distribution of unpaid care. Deeply rooted gender stereotypes continue to assign caregiving responsibilities primarily to women, limiting both women’s economic opportunities and men’s participation in family life.

Changing social norms requires sustained public investment in awareness campaigns, positive fatherhood initiatives, education programs and employer engagement. Promoting caregiving as a shared responsibility between women and men should become an integral component of demographic, education, labor market and gender equality policies. Strengthening partnerships with civil society, employers, academia and the media can further support behavioral change and increase public acceptance of family-friendly reforms.

Conclusion and Strategic Way Forward

Unpaid care work underpins Armenia’s households and economy, yet its heavy reliance on women limits female labor force participation, deepens gender inequalities and constrains national productivity. Addressing this challenge requires moving beyond fragmented interventions toward a comprehensive, evidence-based strategy. The five priority policy actions highlighted in this brief- expanding affordable childcare, introducing equitable parental leave for fathers, fostering family-friendly workplaces, embedding unpaid care into economic planning and state policies, and shifting societal norms- form an interconnected framework to redistribute the care burden.

Because these family-friendly measures are mutually reinforcing, isolated implementation will yield limited gains. Embedding this package directly into Armenia’s Demographic, Employment and Gender Policy strategies, backed by cross-ministerial coordination, will maximize public investment, strengthen human capital and drive long-term demographic and economic resilience.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

Gender Gaps in Access to Startup Capital in Armenia

Posted on September 28, 2026 | Policy Brief

This brief examines whether women in Armenia have less access to early-stage startup funding and what may explain that gap. The available evidence points to a striking imbalance: in a sample of 45 funded Armenian startups, only 7 (15 percent) include a woman founder or co-founder. The structural features of the ecosystem likely reinforce this gap. First, Armenia’s startup market is small and network-dependent — roughly 144 active startups and about USD 164 million in total funding — so access to capital runs heavily through personal connections, and each gatekeeper carries more weight. Second, women bear a heavier share of care and time burdens, leaving less room for the sustained networking on which relationship-based funding depends. The brief, therefore, makes a broader policy claim. Armenia should treat gender gaps in startup capital not only as a funding problem, but also as a market design and care infrastructure problem. A stronger response would combine better gender-disaggregated funding data, more inclusive pathways into investor networks, and more practical support for shared care responsibilities.

Armenia’s startup ecosystem is young, small, and highly network-dependent — the kind of market in which who gets funded can hinge on who is known and trusted. Gender is one such line of division. This brief asks whether women in Armenia have weaker access to early-stage startup funding, and if so, what explains the gap.

The concern is grounded in broader evidence on women entrepreneurs in the country: the IFC/World Bank Women Entrepreneurship Study in Armenia finds that access to finance is the most prominent obstacle entrepreneurs report (International Finance Corporation, 2021). And the barrier is not only a matter of investor choice or founder readiness — women’s access is also shaped by collateral constraints and by unequal access to productive assets tied to inheritance and land ownership.

This brief looks first at what international evidence suggests about gender gaps in startup capital. Then it turns to the Armenian evidence: what the visible funding landscape reveals about women’s access to startup capital, how the small size and network dependence of the ecosystem shape that access, and how unequal care and time burdens reinforce it. It closes with a set of policy recommendations spanning funding-data transparency, more inclusive investor access, and shared care support.

What International Evidence Suggests

International research shows that women founders are disadvantaged in many early-stage funding markets, particularly where investor decisions rely on networks, informal signals, and subjective judgments of founder potential (Kanze et al., 2020; Koch et al., 2025; Liao et al., 2024). Women are less likely to secure external funding and, when they do, tend to raise less: across OECD and BRICS countries, startups with at least one woman founder are 5 to 10 percent less likely to be funded and receive about one third less when funded (Lassébie et al., 2019). Part of this reflects how investors evaluate founders — women are more often asked prevention-focused questions about risks, men promotion-focused questions about growth (Kanze et al., 2018).

The gap is not driven by investor bias alone; pipeline, startup orientation, and signaling explain part of it, though a substantial share remains unexplained. Fewer women may enter the funding process at all; women-led ventures are more often presented as smaller or local businesses rather than the fast-growth companies investors seek; and investors lean on visible signals — prior startup experience, early sales, a strong team, accelerator participation, trusted referrals. Yet these factors do not close the gap: even after accounting for them, women founders still face additional barriers to capital. In the United States, female-led ventures are 63 percent less likely to obtain venture capital, about 65 percent of that difference linked to initial startup orientation, with the remainder persisting after controls (Guzman and Kacperczyk, 2019). That gap narrows when stronger growth signals are available, and, conditional on funding, women- and men-led ventures achieve exits at similar rates. This last result should be read with caution, as it covers only ventures past the selection stage — but it suggests weaker access to capital should not be taken as evidence of weaker post-funding performance.

Investor composition matters too. Bellucci et al. (2024) find that female-owned businesses receive lower funding only when the angel investor is a man — a warning sign for markets where angel investing is relationship-driven and male-dominated. A recent OECD review makes a similar point: angel investment (i.e., that by individual early-stage investors who invest their own money in startups) is especially important for women entrepreneurs, yet they are disadvantaged by thinner networks and the underrepresentation of women among angel investors (OECD, 2025).

Taken together, this literature suggests that gender gaps in startup capital are shaped by both supply-side and demand-side forces: investor perceptions and investor composition matter, but so do pipeline conditions, access to networks, and the quality of the signals founders can present.

This framing is especially relevant for Armenia. Its startup ecosystem is small and tightly networked, with a limited pool of early-stage capital concentrated among relatively few investors — precisely the conditions under which the international dynamics of network reliance, informal signaling, and a male-dominated investor base are amplified rather than offset. When capital flows through a handful of relationships, being outside those networks is more costly than in deep, liquid markets. The rest of this brief examines how these forces appear in the Armenian evidence, and what they imply for who gets to build and grow.

Evidence Base and Approach

The approach in this brief is exploratory and relies on triangulation rather than causal identification. It combines international academic and institutional research on gender gaps in startup finance, Armenian studies on women’s entrepreneurship and access to finance, and ecosystem-level evidence from Armenia’s startup landscape originating from 2carrots VC Database, an Armenian startup database.

One of the key reasons for the chosen approach is data limitations. Armenia does not yet have a systematic public dataset showing founder gender across grants, accelerators, angel investment, and venture capital. The currently visible funding landscape should therefore be read as a descriptive signal rather than a complete national census of startup finance. Even so, the available evidence is strong enough to justify policy attention, especially when combined with signs of a narrow startup market, limited capital depth, and weak supportive conditions around childcare and time use.

Key Findings

Women Are Underrepresented among Funded Startups in Armenia

The clearest visible ecosystem signal is straightforward: women rarely appear among funded founders. To examine this, we used the 2carrots VC Database, a public database of Armenian startups that lists startup profiles and ecosystem information. In a sample of 45 funded startups across 8 venture capital firms and angel investor groups in Armenia, only 7 (15%) appear to include a woman founder or co-founder. Even with all the caveats that come with partial data, this is a striking imbalance. It suggests that women are not achieving the same visible investment outcomes as men.

This pattern matters because it appears in a small, shallow startup market. An article published in the Caucasus Business Journal reports around $164 million in total startup funding across the Armenian ecosystem, while Seedstars describes only 144 active startups in the country. In a larger, deeper capital market, founders may have more routes to funding and a greater chance to recover from early exclusion. In a smaller market like Armenia, each gatekeeper matters more. Under those conditions, underrepresentation among funded startups becomes more consequential because exclusion at an early stage is more likely to have a lasting effect.

Armenia’s Narrow Early-stage Market May Make Network Access Particularly Important

In Armenia’s small early-stage funding market, founders appear to have fewer formal and alternative routes to startup capital than they would in a larger, deeper market. This makes access to investor networks especially important. The broader venture capital literature helps explain why: Alexy et al. (2011) show that venture capitalists’ social networks can shape funding decisions because networks give investors access to information about investment opportunities and help reduce uncertainty.

Armenian evidence also highlights the relevance of these connections. In a Seedstars interview, EBRD’s Armenuhi Arakelyan identifies diaspora ties as a source of funding, expertise, and networking opportunities (Seedstars, 2025). The IFC’s qualitative research reports limited interaction among women entrepreneurs and insufficient connections with sectoral associations and specialized NGOs, restricting their access to information and opportunities. However, it notes that networking is more prevalent in the IT sector (International Finance Corporation, 2021).

Applied to Armenia, this suggests that when the number of visible investors and funding channels is limited, referrals, repeated contact, and personal visibility may carry greater weight. In practice, access to the right circles can therefore matter more. Such a market structure can disadvantage women founders even without explicit exclusion if they are less embedded in investor networks and therefore have fewer opportunities to build the repeated visibility and trust on which relationship-based funding depends.

Unequal Care and Time Burdens Help Explain Why Access to Those Networks Is Not Equal

Fundraising is not just about having a strong idea or writing a pitch deck. It requires repeated meetings, networking, travel, follow-up, and sustained visibility — all of which take time. And time is precisely what women in Armenia are more likely to lack. A 2021 IFC and World Bank study of more than 400 Armenian businesses found that women were more likely than men to report lack of time and work-life balance as obstacles (International Finance Corporation, 2021).

These pressures become more consequential when supportive conditions around care are weak. The World Bank’s Women, Business and the Law profile scores Armenia at 75 out of 100 on legal frameworks, but only 45 out of 100 on supportive frameworks, with childcare supportive frameworks at 0 (World Bank, 2026). The practical implication is that care remains unevenly distributed and women bear more of the time cost. When fathers are not strongly supported or incentivized to share childcare responsibilities, women have less flexibility for the intensity that early-stage entrepreneurship often demands.

The problem is likely exacerbated by the network design. The problem may be exacerbated by the timing of networking events. For example, Startup Grind Yerevan’s Pitch Battle on 15 January 2025 was scheduled for 18:30–20:30, with representatives of angel investor networks among its judges (Startup Grind, 2025). Armenia’s National Services Gateway states that most kindergartens close at 17:30, with some extending to 18:30 (National Services Gateway, n.d.). Attending such evening events may therefore require additional childcare arrangements, potentially limiting participation among founders with substantial care responsibilities.

In that sense, the care burden is not separate from the capital access problem. It is one of the conditions that helps explain why women may have weaker access to the networks, meetings, and repeated relationship-building through which early startup funding often flows. This is why childcare policy should not be treated as unrelated to startup finance. In the Armenian context, it forms part of the wider structure that shapes who can participate fully in entrepreneurial life.

Policy Recommendations

The evidence in this brief suggests that Armenia should address gender gaps in startup capital on two fronts simultaneously. The first is the funding market itself: who gets seen, introduced, and funded in a small, network-dependent ecosystem. The second is the time constraints brought about by uneven allocation of household duties and childcare that shape who can participate fully in that market. Policy should therefore focus not only on improving women founders’ access to investors, but also on reducing the structural constraints that make that access unequal in the first place.

Build a Gender-disaggregated Startup Funding Dashboard

The first priority is visibility. Armenia should begin systematically tracking founder gender across grants, accelerator cohorts, startup competitions, angel-backed deals, and venture investments that involve public or donor support. Without basic visibility on who gets funded, the debate will continue to rely on partial signals and anecdotes. This tracking should cover not only the number of women-led teams funded, but also ticket size, stage, and follow-on outcomes. A stronger evidence base would make it easier to identify where the gap is largest and whether existing support programs are helping to close it.

Make Investor Access More Inclusive and Less Dependent on Informal Evening Networking

A second priority is to widen the channels through which founders can reach investors. Public and donor-backed startup programs should create more daytime networking opportunities, more online investor sessions, more structured introductions, and less reliance on informal evening gatherings as the main route into capital. In a small ecosystem, relationship-building may remain essential. But if that is the case, then the formats through which relationships are built should not systematically advantage founders with fewer care constraints and easier access to late-evening social environments. Making investor access more inclusive is not a peripheral adjustment. It is part of improving how the market works.

Strengthen Shared Childcare Support as Part of the Economic Infrastructure

Better access to startup capital cannot be separated entirely from the wider care environment, although childcare support should be understood as an enabling condition rather than a stand-alone solution to the funding gap. The evidence does not show that better childcare arrangements automatically translate into better startup funding outcomes. However, it does show that care and time constraints shape women’s ability to participate fully in entrepreneurial activity. Armenia should therefore treat shared childcare support as part of the economic infrastructure that makes entrepreneurship more feasible, not as a substitute for finance and network reforms. This means improving childcare availability and affordability, encouraging fathers’ active participation in care, protecting men who use care-related leave or flexible work arrangements, and using public messaging to frame childcare as a shared responsibility. For women founders, the expected benefit is practical: more time and flexibility for networking, travel, investor meetings, and sustained visibility. Startup programs can reinforce this by offering hybrid participation, flexible scheduling, and childcare support where feasible.

Overall Recommendation

Taken together, the evidence reviewed in this brief supports a combined policy response rather than a single intervention. Armenia should improve gender-disaggregated startup funding data, broaden women founders’ access to investors through more inclusive and structured networking channels, and treat shared childcare support as complementary economic infrastructure. This approach reflects the literature’s broader finding that funding gaps can arise from both investor-side mechanisms and unequal access to the networks, signals, and conditions needed to compete for startup capital. In Armenia’s small and network-dependent ecosystem, addressing these constraints together is likely to be more effective than treating the funding gap as a stand-alone financing problem.

Concluding Remarks

This brief suggests that gender gaps in startup capital in Armenia are not simply a matter of who pitches better or who investors prefer. They reflect a deeper interaction between market structure and care structure: a small, network-dependent funding ecosystem on the one hand, and unequal time and childcare burdens on the other. As a result, women’s weaker access to startup funding should be understood not only as a gender gap in outcomes, but also as a question of how entrepreneurial opportunity is organized.

The implication is clear. Armenia should address this problem not only by widening women founders’ access to investors, but also by reducing the structural constraints that make that access unequal in the first place. A more open funding market, combined with stronger support for shared care, would not only improve fairness but also strengthen the system. It would also strengthen the reach and overall effectiveness of the Armenian startup ecosystem.

References

  • 2carrots VC Database. Accessed September 20, 2026.
  • Alexy, Oliver T., Joern H. Block, Philipp Sandner, and Anne L. J. Ter Wal, “Social capital of venture capitalists and start-up funding,” Small Business Economics, April 2011, 39 (4), 835–851.
  • Bellucci, Andrea, Gianluca Gucciardi, Rossella Locatelli, and Cristiana-Maria Schena, “Gender Gap in Business Angel Financing,” Entrepreneurship Research Journal, May 2024, 15 (2), 289–324.
  • Guzman, Jorge and Aleksandra (Olenka) Kacperczyk, “Gender gap in entrepreneurship,” Research Policy, September 2019, 48 (7), 1666–1680.
  • International Finance Corporation, “Women Entrepreneurship Study in Armenia: Qualitative and Quantitative Study, Synthesis Report,” Technical Report, International Finance Corporation, Yerevan, Armenia March 2021.
  • Kanze, Dana, Laura Huang, Mark A. Conley, and E. Tory Higgins, “We Ask Men to Win and Women Not to Lose: Closing the Gender Gap in Startup Funding,” Academy of Management Journal, April 2018, 61 (2), 586–614.
  • Kanze, Dana, Mark A. Conley, Tyler G. Okimoto, Damon J. Phillips, and Jennifer Merluzzi, “Evidence that investors penalize female founders for lack of industry fit,” Science Advances, November 2020, 6 (48).
  • Koch, Laura H., Elisabeth S. C. Berger, and Andreas Kuckertz, “Gender bias and discrimination towards women entrepreneurs by venture capitalists – a randomized response survey,” Venture Capital, April 2025, p. 1–27.
  • Lassébie, Julie, Sahra Sakha, Tomasz Kozluk, Carlo Menon, Stefano Breschi, and Nick Johnstone, “Levelling the Playing Field: Dissecting the Gender Gap in the Funding of Startups,” Technical Report, Organisation for Economic Co-operation and Development April 2019. Approved by the Committee on Industry, Innovation and Entrepreneurship (CIIE) on 8 April 2019.
  • Liao, Zhenyu, Jack H. Zhang, Nan Wang, William P. Bottom, Dirk Deichmann, and Pok Man Tang, “The Gendered Liability of Venture Novelty,” Academy of Management Journal, April 2024, 67 (2), 299–330.
  • Seedstars, 2025. “How Armenia built an ecosystem bigger than its borders”. Accessed June September 21, 2026.
  • OECD, Bridging the Finance Gap for Women Entrepreneurs: Insights from Academic and Policy Research OECD Studies on SMEs and Entrepreneurship, OECD Publishing, 2025.
  • Siegrist, Felicia, “Supporting Women Entrepreneurs in Developing Countries: What Works? A Review of the Evidence Base and We-Fi’s Theory of Change,” Technical Report, Women Entrepreneurs Finance Initiative, Washington, D.C. July 2022.
  • Ubfal, Diego Javier, “What Works in Supporting Women-Led Businesses?,” Policy Research Working Paper WPS10744, World Bank Group, Washington, D.C. April 2024.
  • World Bank, “Women, Business and the Law: Armenia Country Profile,” Technical Report, World Bank Group 2026.

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

From AI Adoption to Productive Use: Building Organizational AI Capability in Latvia

Posted on September 21, 2026 | Policy Brief

Artificial intelligence is spreading rapidly across organizations, but access to AI tools does not automatically translate into productivity gains or business value. Drawing on interviews with senior executives from medium and large organizations in Latvia, this brief examines how firms adopt AI and develop the organizational and workforce capabilities required to use it effectively. The analysis identifies a consistent pattern: organizations have made substantial progress in deploying AI tools, developing management awareness, assigning responsibility, and establishing governance, while less so in areas requiring deeper organizational adaptation – data readiness, measurement of business impact, task and role redesign, strategic workforce planning, and organization change. The next stage of AI readiness lies in embedding AI into organizational processes and translating its use into measurable value. For business leaders, this requires stronger foundations and measurement, systematic redesign of work, and greater integration of workforce planning into AI implementation. For policymakers, the task is to extend Latvia’s existing infrastructure for digitalization, experimentation, and AI skills towards the organizational capabilities required for productive AI use.

AI Adoption Is Accelerating, but Is It Creating Value?

AI adoption among European businesses has accelerated sharply. In 2025, 20.0 percent of EU enterprises with at least 10 employees reported using AI technologies, up from 13.5 percent in 2024. Latvia remained below the EU average, at 12.2 percent (Eurostat, 2026). The aggregate figure, however, masks substantial differences by company size: AI was used by 19.8 percent of medium-sized Latvian enterprises and by 47.1 percent of large enterprises in 2025.

These figures indicate that AI is already relevant to most organizations with sufficient scale to make structured investments in technology and workforce development. Yet adoption statistics reveal little about what happens after an organization begins using AI. A company may purchase licenses, launch pilots, or introduce AI tools without substantially changing how work is organized or being able to demonstrate that the investment improves performance.

The medium and large organizations in Latvia examined in this study reflect this shift. Most have deployed general-purpose or specialized AI tools, and some have developed extensive portfolios of AI applications. Senior executives generally demonstrate a pragmatic understanding of AI and increasingly distinguish between applications offering concrete value and those driven primarily by technological hype. The management architecture around AI is also developing: responsibility has usually been assigned, working groups or steering structures are common, and many organizations have dedicated budgets, internal guidelines, and processes for moving AI ideas from experimentation towards common use.

However, organizations struggle to integrate fragmented data, determine whether AI investments improve performance, redesign tasks and jobs around new technologies, and translate technological change into systematic workforce planning. This is the challenge of organizational embedding – integrating AI into workflows and organizational routines, adapting tasks and roles, developing appropriate workforce capabilities and establishing mechanisms for assessing whether AI creates value. AI adoption is progressing faster than organizational embedding.

Evidence Base and Analytical Framework

The underlying research (Nīgale, Kalniņa and Vārpiņa, 2026) is based on semi-structured interviews conducted in Spring 2026 with senior executives from 22 medium and large organizations in Latvia, supplemented by interviews with AI and digital-transformation practitioners. Organizations were purposively selected across major sectors of the Latvian economy. The sample deliberately focused on organizations with evidence of AI engagement and was designed for cross-sectoral qualitative trend comparison rather than for statistical representativeness. The findings should therefore be read as patterns, not prevalence estimates.

The interviews were analyzed using an adapted Technology–Organization–People (TOP) framework, based on Tursunbayeva and Chalutz-Ben Gal (2024). The framework treats AI adoption as a multilevel process shaped by the technologies being introduced, the organizational conditions in which they are deployed, and the people who use them. The adapted framework covers 17 capability areas: six related to technology (T), eight to organization (O), and three to people (P).

Interview evidence was classified using a Red-Amber-Green (RAG) assessment. Red indicates that a capability is largely absent or dependent on individual initiative; Amber – the relevant activity exists but remains partial, informal, or inconsistently embedded; and Green – the capability is integrated into organizational practice, with defined ownership or a clear implementation path.

 AI Adoption Is Ahead of Organizational Embedding

The heatmap in Figure 1 reveals differences between organizations, but the stronger pattern appears across capability areas. Capabilities associated with the visible adoption of AI tend to be comparatively well developed. Tools are being deployed, management understands the evolving AI landscape, responsibility is assigned, governance structures are emerging, and resources are committed.

Figure 1. Patterns of AI readiness across 22 medium and large organizations actively engaging with AI in Latvia

Source: Nīgale, Kalniņa and Vārpiņa, 2026. Note: The assessment summarizes qualitative evidence from executive interviews and is intended to reveal patterns of relative strength and weakness.

Areas that depend on broader organizational adaptation show a different pattern. Measurement of business impact, data foundations, HR’s strategic involvement, and behavioral change are consistently less developed. Task and role redesign has begun in many organizations but has rarely developed into systematic practice. In the RAG assessment, these areas are dominated by emerging rather than established capabilities.

This suggests an important difference between two types of organizational capability. A management team can make a relatively discrete decision to purchase licenses, authorize expenditure, appoint an AI lead or introduce a governance policy. Integrating organizational data, redesigning work, changing employee behavior, and developing reliable outcome measures require sustained coordination across functions and changes in organizational practice.

The challenge is therefore less about acquiring AI than about what organizations can do with it.

From Adoption to Embedding

Foundations and Measurement Remain Weak

Data-related constraints emerge repeatedly as a barrier to more advanced AI use. Companies report fragmented information, interoperability problems, and unclear data ownership. In many organizations, the visible layer of AI tools has advanced more quickly than the data infrastructure and processes needed to make more integrated applications work. One AI director summarized the challenge:

“AI by itself is an enabler. Value is created when prerequisites are in order: data and its quality, technologies, processes, data governance, and AI governance.” (AI Director, Retail)

The implication is that further investment in AI tools cannot substitute for the organizational foundations on which those tools depend.

Measurement represents an equally important weakness. Companies can frequently identify how many licenses have been purchased, how many employees have attended training, or how many pilots have been launched. Much less common is systematic measurement connecting AI use with productivity, quality, costs, customer outcomes or other organizational objectives. One executive described a particularly difficult measurement problem:

“Only the employee themselves knows how much time they have actually saved using AI. The manager cannot see this. This asymmetric information problem is significant.” (CEO, Shared Services Center)

This helps explain why individual productivity gains can remain largely invisible at the organizational level. Without credible evidence about what AI delivers, management has a weaker basis for deciding which applications to scale, modify, or discontinue. The distinction between measuring activity and measuring outcomes therefore becomes central to effective AI management.

AI Is Being Added to Jobs Faster than Jobs Being Redesigned Around AI

Organizations generally recognize which business functions are being affected by AI, particularly IT, marketing, customer service, legal functions, and HR. Yet systematic task-level analysis remains rare. Respondents do not describe widespread efforts to map which individual activities within jobs can be automated or augmented, which should remain predominantly human, and how responsibilities should be reorganized as AI capabilities expand.

A similar pattern appears in workforce development. General AI training is relatively widespread, particularly around common tools and prompting. Some organizations provide more advanced or role-specific programs. Still, very few systematically map AI-related capabilities across different occupations or functions, and none of the organizations studied uses a structured capability framework of this kind.

The issue does not end with identifying tasks that AI can perform faster. Organizations also need to decide what happens to the capacity that AI releases. As one CEO put it:

“Motivation is the key: what does the organization want the employee to do with the saved time? That question must have a clear answer.” (CEO, Shared Services Center)

The answer might be client development, higher-value analytical work, innovation, additional output, or further professional development. Without such decisions, faster task completion does not automatically become higher organizational productivity. AI implementation therefore requires a conversation about work, rather than simply about technology.

Workforce Transformation Often Lacks Strategic Ownership

Most organizations have relatively clear ownership of the AI program itself. Responsibility for the workforce transformation accompanying AI is less developed. HR is frequently involved in organizing training, recruitment or communication after major AI-related decisions have already been taken. It is less consistently involved in defining future capability needs, analyzing task change or redesigning roles. In some organizations, the difficulty begins even earlier:

“Currently, HR involvement in AI processes has not been relevant for us. We ourselves cannot yet clearly define AI skills.” (Director, Manufacturing)

This suggests that limited HR involvement is more than an HR problem. In some organizations, AI implementation has not yet been translated into sufficiently clear questions about tasks, capabilities, and future workforce needs for HR to address strategically.

The finding also does not imply that HR should independently own AI transformation. AI-related workforce change requires collaboration between business leadership, technology functions, process owners and HR. The central issue is that organizations frequently have clearer ownership of technology adoption than of the organizational and workforce changes required to make that technology productive.

Employee behavior is a further constraint. Some respondents describe a substantial gap between providing access to AI and seeing employees incorporate it meaningfully into everyday work. Limited time to experiment, established routines, uncertainty about usefulness, and concerns about future roles can all impede adoption. Training matters, but it is unlikely to create sustained organizational change if employees remain uncertain about how AI should alter their work or what the organization expects them to do with the resulting efficiency gains.

Latvia’s AI Support Ecosystem Is Developing

These organizational gaps need to be considered against a policy environment that has already evolved considerably. Latvia already has a range of instruments supporting business digitalization and AI adoption.

The latest Latvian Investment and Development Agency (LIAA) program for process digitalization and AI solutions supported both digital business-process transformation and the development and implementation of AI solutions. Demand was high: applications requested €25.5 million against €18.6 million available. Applications closed in November 2025, and LIAA indicates that no new call is currently planned, although approved projects continue to be implemented (LIAA, 2026).

Latvia’s European Digital Innovation Hubs provide another layer of support. The Digital Accelerator of Latvia offers digital maturity assessments, transformation roadmaps, training, consulting, and test-before-invest services. The European Commission’s Digital Maturity Assessment covers digital business strategy, digital readiness, human-centered digitalization, data management and connectivity, automation and AI, and green digitalization (Digital Accelerator of Latvia, 2026). The Latvian IT Cluster EDIH also offers AI-oriented services, including an AI Studio implementation workshop that helps firms analyze business processes, identify AI use cases, define pilots, and develop an implementation roadmap (Latvian IT Cluster EDIH, 2026).

At EU level, the EDIH network has been refocused as Experience Centers for AI, with a stronger role in supporting companies and public-sector organizations to test, deploy, and develop skills around AI. The AI Skills Academy began operations in May 2026 and is intended to support AI and generative-AI upskilling and reskilling, including for professionals in SMEs (European Commission, 2026a, 2026b).

The competence-framework landscape has also advanced. DigComp 3.0, released in late 2025, systematically integrates AI competence across the European Digital Competence Framework. It defines competence in terms of knowledge, skills, and attitudes and is designed to be adapted for employment, education, and training contexts, including sectoral and role-specific competence profiles (Cosgrove and Cachia, 2025).

Finally, Latvia’s Artificial Intelligence Center provides an institutional platform linking the public sector, private sector, and higher education. Its statutory tasks include promoting AI application in public and private organizations, strengthening AI skills, providing consultations and guidelines, coordinating projects and supporting responsible and safe implementation (Likumi.lv, 2025).

These initiatives already address technology adoption, experimentation, skills, data, and elements of organizational development. As firms move beyond initial adoption, the support ecosystem needs to evolve accordingly, from enabling AI adoption towards enabling productive organizational use.

Priorities for Business Leaders

The findings suggest three priorities for organizations seeking to move from AI experimentation towards productive use.

  1. Strengthen foundations and measurement before scaling. Companies should ensure that reliable data, interoperability, access rights, data ownership, security and governance are in place alongside AI deployment. Expected outcomes should also be defined before pilots are scaled. Monitoring licenses, usage, or participation in training shows whether AI activity is occurring. Management also needs evidence about whether AI improves productivity, quality, customer outcomes or other business objectives.
  2. Redesign tasks and define how efficiency gains will be used. Organizations should analyze tasks and workflows to determine which activities can be automated or augmented, which should remain human-led, and how work should be reorganized. Management should explicitly consider how released employee capacity will be used. Without this step, time saved on individual tasks is absorbed rather than converted into output.
  3. Strengthen HR’s role in AI implementation. HR should participate beyond training delivery and administrative support. Strategic workforce planning, capability development, task and role redesign, and change management should form part of AI implementation from an early stage. Workforce transformation should be jointly owned by business leadership, technology functions, process owners and HR.

These priorities reinforce one another. Better measurement provides evidence about where AI creates value. Task analysis identifies how work needs to change. Workforce planning and capability development then help organizations implement those changes.

Priorities for Policymakers: From Supporting Adoption to Supporting Embedding

Three corresponding priorities emerge for national and EU policymakers.

  1. Extend AI support towards organizational embedding. Future digitalization and AI support should increasingly address the organizational capabilities required to translate technology into productive use. Existing maturity assessments, roadmaps and support instruments provide a natural starting point. These instruments could place greater emphasis on data readiness, task and process redesign, workforce implications, organizational ownership and outcome measurement. This is an extension of the current support system rather than a call for a separate new program.
  2. Translate European competency frameworks into practical workforce tools. Developing another general AI competency framework is unlikely to be the best use of resources. DigComp 3.0 and related European initiatives already provide a substantial conceptual foundation. The next step is to translate these into tools employers can use for role-specific skills mapping, capability assessment, training pathways and workforce planning. Such tools could be developed with EDIHs, employer organizations, education providers and the Artificial Intelligence Centre, with particular attention to organizations lacking dedicated workforce-planning capacity.
  3. Strengthen public sector capability. Public sector institutions should also strengthen their own capacity to adopt AI effectively. This should include systematic task and process redesign, workforce capability development, and measurement of productivity gains, allowing the public sector to demonstrate practical approaches to AI-enabled organizational change. Demonstration cases should document organizational changes as well as the technology introduced, including how tasks changed, how employees were prepared, and whether productivity or service quality improved. Latvia’s Artificial Intelligence Center is well positioned to support the dissemination of such approaches given its mandate across the public and private sectors.

These recommendations do not necessarily require a new large-scale public program. They point instead towards extending and redirecting existing instruments as organizations progress from experimentation towards wider AI use.

References

Acknowledgements

We thank Inga Gleizdāne for support and advice during the research process.

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

On the Debate Around the Swedish “Reduktionsplikt”

Posted on September 7, 2026 | Policy Brief

The Reduktionsplikt (engl. reduction duty) requires Swedish fuel suppliers to reduce the greenhouse gas emissions of their petrol and diesel, in practice mostly by blending in biofuels. It has become a much-contested policy of the 2026 election, and the debate is almost entirely about the policy’s impact on the pump price.

We argue this focus is too narrow. So long as compliance runs through blending, any price effect is simply what follows from mixing two products with different prices: the mandate level scales it, but its size depends on the gap between biofuel and fossil diesel, which is largely determined by broader market conditions rather than by the mandate level alone. What deserves attention is where that gap is heading, and whether the mandate’s cost is reasonable relative to the emissions it avoids — a question Sweden must answer if it is to choose efficiently between policies to reach its 2030 emissions goals.

The Policy and the Debate

Fuel prices have become one of the most politically charged issues ahead of Sweden’s election this autumn. Several political parties are competing to offer the largest reductions at the pump, while proposals that could raise petrol or diesel prices meet immediate resistance (SVT Nyheter, 2026). The issue has become even more salient following the recent energy shock linked to disruptions in oil flows through the Strait of Hormuz. In response to rising energy prices, Sweden reduced fuel taxation from the beginning of July until the end of November to shield households and businesses (Skatteverket, 2026). This again illustrates how central the pump price has become to both economic policy and the election campaign.

At the same time, Sweden remains far from reaching its climate target for domestic transport. The country has committed to reducing transport emissions by 70 percent by 2030 relative to 2010 (Regeringen, 2017), but emissions have so far fallen by only around one fifth and electrification of the vehicle fleet has progressed more slowly than anticipated (Klimatpolitiska rådet, 2024). Fuel policy therefore sits on a fault line between keeping driving affordable and meeting a binding climate commitment.

No policy captures this tension more sharply than the Reduktionsplikt, which was introduced in July 2018 as a central part of the policy package intended to reach the climate target. It requires fuel suppliers to reduce the average lifecycle greenhouse-gas intensity of petrol and diesel, primarily by blending renewable biofuels such as HVO, FAME, and ethanol into fossil fuels. The policy can therefore reduce emissions from the existing vehicle fleet without waiting for all drivers to switch to electric vehicles.

In this brief, we focus on the diesel mandate. Historically, it has been significantly higher than for petrol, as diesel allows much more fuel blending. The diesel mandate was introduced at 19.3 percent in 2018 and incrementally increased to 30.5 percent in 2022–2023, before being reduced to 6 percent in January 2024 following a political backlash over high fuel prices. In July 2025, it was raised again to 10 percent.

A government inquiry on policy measures for phasing out fossil fuels, chaired by the head of environmental economics research at Konjunkturinstitutet, Svante Mandell, has now proposed increasing the mandate to 21 percent in 2028 and 25 percent in 2030. Together with higher fuel taxation, the package has been estimated to raise pump prices by around three kronor per litre (Mandell, 2026). Mandell argues that stronger measures are needed because the current policy mix is unlikely to deliver Sweden’s climate commitments. Representatives of the governing parties and their parliamentary allies have rejected the proposal, arguing that it would impose excessive costs on households and businesses. The governing parties instead want to emphasize electrification and lower fuel costs.

What the Pump-price Debate Misses

The political focus on pump prices is understandable, but incomplete. The debate tends to treat the price effect of the Reduktionsplikt as if it were determined primarily by the mandate level itself: a higher Reduktionsplikt necessarily means substantially higher fuel prices.

In reality, the effect depends crucially on the price difference between renewable and fossil diesel at the time. When biofuels are priced close to fossil diesel, a higher renewable share can reduce emissions with little effect on the final blended fuel price. When renewable fuels become more expensive, the same mandate can instead raise blended diesel prices sharply.

In this way, the price effect of Reduktionplikt depends on the interaction between the required renewable share and the circumstances affecting the renewable-to-fossil price gap, including biofuel supply, international biofuel demand, feedstock availability, exchange rates and international energy-market conditions. This makes a narrow focus on the current pump-price effect short-sighted. Whether a higher mandate will be costly in the future depends on how the biofuel market develops, not simply on the mandate level.

Moreover, the extent of the price increase should not be assessed in isolation. A higher price is a cost, but in return the mandate delivers something of value: lower emissions. To judge the policy fairly, the cost paid at the pump must be weighed against the climate benefit it facilitates and weighed in a way that lets the mandate be compared with other ways of cutting the same emissions.

Mandate Effect and Market Conditions

In a recent paper (Larsson and Oberhammer, 2026), we address the above points. We first estimate how much higher Swedish diesel prices were because of the mandate from 2018 through 2023. Since there is no common price for all types of biofuels, we cannot observe it directly. Instead, we use Denmark as a counterfactual, since Sweden and Denmark participate in the same north-western European fuel market and are exposed to similar oil-price and supply shocks, but Denmark maintained a much lower and more stable biofuel obligation. The resulting Swedish–Danish price premium is therefore our estimate of the mandate’s price effect: it measures the cost of the extra biofuel Sweden blends, which is simply the extra mandate share times how much more (or less) biofuel costs than fossil diesel.

As expected, before 2018, we observe that Swedish and Danish diesel prices excluding taxes moved closely together.

Figure 1. Swedish and Danish diesel prices without tax

Source: Authors’ rendering of price data from Circle K Denmark (2026) and Circle K Sweden (2026) using the daily SEK/DKK exchange rate reported by Sveriges Riksbank (2026).

From 2018 onward, the comparison lets us estimate the price premium at each mandate level. During the early period, with mandate levels of 19.3 and 20 percent and relatively low prices, we find no increase in price at all. We first detect a price premium during the 21 percent level. It starts at 0.17 SEK per litre and then rises to 0.99 SEK per litre during 26 percent policy and 2.05 SEK per litre during 30.5 percent. When the mandate was cut in 2024, the gap fell sharply and even turned negative — mostly a currency effect, with the remainder following from Denmark’s mandate now being above Sweden’s.

Figure 2. Price gap between Swedish and Danish diesel prices without tax

Source: Authors’ rendering of price data from Circle K Denmark (2026) and Circle K Sweden (2026) using the daily SEK/DKK exchange rate reported by Sveriges Riksbank (2026).

The immediate takeaway is that a high mandate per se does not generate the same price effect in every period. Because the premium is the extra blend share multiplied by the biofuel–fossil price difference, its size depends on that price difference as much as on the mandate level. The early years make this concrete: Sweden was meeting an obligation of around 20 percent yet showed no premium over Denmark, which means biofuel and fossil diesel must have been priced at similar levels then. The large premiums emerge only in 2022–2023, when biofuel became expensive across the EU during the energy crisis (Lundberg et al. 2023). Denmark, blending far less, was largely insulated from that cost — which is why the Swedish–Danish gap widens exactly as biofuel prices climb.

Climate Benefits Vs. Price Costs

With the price effect established, we can weigh what the policy achieved against what it cost. We estimate that the Reduktionsplikt reduced emissions by around 14 million tonnes of CO₂e from 2018 until the end of 2023. On average, 94 percent of that came from substituting fossil diesel with biofuel and only 6 percent from reduced consumption during the high-price period. The split reflects how the policy works: it cleans the fuel rather than pricing people out of using it. It also reflects how little diesel demand responds to price in the short run. We estimate that a 10 percent rise at the pump cuts diesel use by only about 2 percent, since the households and firms that depend on diesel for commuting, freight, farming and construction have few immediate alternatives.

By comparing the extra amount paid due to the policy with the emissions avoided, you get the cost of avoiding each tonne of CO₂e. This single figure lets the mandate be judged on the same terms as fuel taxes or electric-vehicle subsidies. Sweden’s carbon tax offers a natural benchmark. At around 1,510 kronor per tonne of CO₂e (Government Offices of Sweden, n.d.), it is the price the country has explicitly placed on fossil carbon, and in effect society’s own valuation of what avoiding a tonne of emissions is worth.

Against that yardstick, the mandate was a very cheap way to cut carbon in its early years, when it reduced emissions with little or no effect on the diesel price. It became expensive only later, at the height of the energy crisis, when compliance costs ran well above the climate benefit. However, averaged across the whole period, including those expensive years, the cost of avoiding a tonne came out at roughly 1,574 kronor. That is close to the carbon tax itself. So, despite the unfavorable market developments, the mandate has been a reasonable exchange overall for the emissions it bought.

Reasonable compared to what, though? If the transport target is to be met, one direct alternative is higher fuel and carbon taxation. Such a policy would have to work through consumption reduction — the channel that delivered only the abovementioned 6 percent of the mandate’s reductions. Konjunkturinstitutet estimates that reaching the target through fuel taxation alone, with the mandate held at its current floor, would require pump prices of around 42 kronor per litre for both petrol and diesel (Carlén and Hill, 2023) — well above anything reached even when a record mandate coincided with expensive biofuel.

However, the mandate’s advantage over taxation rests on biofuels being available at a price not too far above fossil diesel, and that condition held comfortably in some years and barely at all in others. This is the point currently missing from the debate: both the price effect and the cost per tonne of emissions avoided depend crucially on conditions in the biofuel market, not on the mandate level alone. When biofuel markets tighten, compliance costs rise quickly.

One reason for caution is scale. Even Sweden’s historically high mandate could only address part of the transport transition. If many countries raise their blending requirements at once, they will be competing for a limited pool of sustainable biofuels. Supply may well expand in response, but whether it expands quickly enough to keep pace is genuinely uncertain. That uncertainty is what deserves attention: if biofuel can be supplied in sufficient volume at a price not far above fossil diesel, the Reduktionsplikt remains an efficient way of cutting emissions, and if it cannot, the costs of the energy crisis may return.

Future reforms should therefore focus not only on the formal mandate level, but also on resilience to such shocks, by broadening how suppliers can comply to the obligation. The redesigned low-level mandate of July 2025 already allows suppliers to meet part of their obligation by supplying renewable electricity for road transport rather than liquid biofuel. Such electricity credits reduce dependence on internationally traded biofuels and tie the policy more closely to electrification (Energimyndigheten, 2025).

Conclusion

The ongoing policy debate surrounding the Reduktionsplikt is therefore too simplified. A higher mandate does not automatically produce a large fuel-price increase, and even when prices increase, we need to compare the cost of the policy to its environmental benefits.

The policy delivered substantial emissions reductions with little or no diesel-price premium in its early years but became costly in 2022–2023 as high mandate levels coincided with expensive biofuels. Across the full period, however, the cost per tonne of CO₂e avoided was close to Sweden’s carbon tax.

The relevant policy question is therefore not whether the Reduktionsplikt is simply cheap or expensive. It is first whether biofuel supply can be expected to keep pace with demand, since a market that remains tight will produce the same price shocks again. And then, whether Sweden can design a mandate that delivers near-term emissions reductions at a cost comparable to alternative climate measures, while limiting its exposure to those shocks and supporting the longer-term transition toward electrification.

Finally, price and emission effects do not settle every question about the policy. Two broader questions, not captured by the price-and-emissions comparison above, also matter for the policy’s future. First, the mandate may either support electrification by cutting emissions from the existing vehicle fleet while electric alternatives expand, or slow it by reducing the pressure to replace fossil-fuel vehicles. Second, the climate benefit depends on the biofuels used: it is larger for fuels produced from genuine waste and residues than for crop-based feedstocks associated with land-use effects (Searchinger, 2008).

The future of the Reduktionsplikt should therefore be judged not only by its effect on today’s pump price, but by whether it can deliver cost-effective emissions reductions while supporting the longer-term transition away from fossil fuels.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes, or Swedish Energy Agency (Energimyndigheten).

Health Literacy in Latvia: A Public Health Challenge Shaped by Socio-economic Inequality

Posted on June 22, 2026 | Policy Brief

Latvia faces a serious health literacy challenge: many adults struggle to access, understand, evaluate, and apply health information in everyday decisions. The weakest domain is health promotion, suggesting that many people find it especially difficult to use information that helps maintain and improve health before illness occurs. Health literacy is unequally distributed across society: people with lower income and lower perceived social status are at greater risk of inadequate health literacy. Education shows a weaker and less consistent association than expected. The findings show that health literacy is not only a separate competence, but also one of the pathways through which socio-economic inequality translates into unequal health outcomes. Improving health literacy should therefore be treated as both public health and equity priority.  Recommended actions include targeted support for socio-economically disadvantaged groups and systematic integration of health literacy approaches into health communication and education.

Health literacy matters because modern health systems require people to access, understand, appraise, and apply health-related information. In everyday life, this means being able to find trustworthy advice, interpret instructions, assess risks and benefits, and make decisions about treatment, prevention, and health-promoting behaviours.

International evidence shows that low health literacy is associated with poorer health outcomes, weaker prevention, less effective use of health services, and higher pressure on health care systems (Berkman et al., 2011; Zheng et al., 2018). In Latvia, evidence on population health literacy is scarce. To date, the closest population‑level estimate of general health literacy in Latvia comes from a 2022 pilot study, which reported that 79% of the population had below‑average health literacy (Gatulytė et al., 2022) – one of the lowest observed health literacy levels among the research done in EU (Sørensen et al., 2015).

This brief examines evidence on health literacy in Latvia and discusses why it matters for public health and health equity. Drawing on a population survey, it highlights key patterns in health literacy, shows how these are linked to socio-economic inequalities, and proposes policy directions for making health information easier to access, understand, and use.

Evidence Base

The analysis is based on survey data from the project “Health Literacy Study in Latvia: Levels, Burdens and Interventions”, funded by the Latvian Council of Science. The sample is representative of the Latvian population with respect to age, gender, nationality, and place of residence, and includes 2,007 respondents aged 18-75. Data was collected in November 2025.

The survey contained 89 questions in five sections covering demographic characteristics, socio-economic situation, self-reported health status and behaviour, health literacy, and health resource utilisation. Health literacy is measured with the HLS-EU-Q47 questionnaire, a validated instrument used in Europe to assess and compare health literacy levels across countries (Pelikan et al., 2019). The instrument yields a health literacy index on a 0-50 scale. According to the original methodology, the index values are classified into four levels: inadequate (0–25), problematic (>25–33), sufficient (>33–42), and excellent (>42–50) (Pelikan et al., 2019).

Low Health Literacy Is Widespread

Health literacy levels in Latvia are concerning. The average health literacy index score is 29.78, with most respondents clustered near the cut‑off point between problematic and sufficient health literacy. Overall, 75.64% of respondents have limited health literacy: 20.53% fall into the inadequate category and 55.11% into the problematic category. Only 19.73% demonstrate sufficient health literacy, and 4.63% excellent health literacy.

These results are close to the earlier Latvian pilot study, where limited health literacy was estimated at 79% (Gatulyte et al., 2022). Latvian results compare unfavourably with European evidence, where inadequate and problematic health literacy levels were substantially lower – the average across the surveyed European countries is approximately 12% inadequate and 35% problematic health literacy (Pelikan et al, 2014), with the highest levels of insufficient health literacy reported at about 60% in Bulgaria and Spain.

Figure 1: Overall levels of health literacy in Latvia.

Source: Author’s calculations based on the Health Literacy Study in Latvia survey, 2026.

Where the Difficulties Are the Greatest

Across the three health literacy domains (health promotion, disease prevention, and healthcare), the most pronounced difficulties appear in health promotion. In this domain, 34.03% of respondents report inadequate health literacy, and 41.85% problematic health literacy. This suggests that people find it especially difficult to engage with information about behaviours that improve or maintain health, such as nutrition, physical activity, and other health-promoting practices.

Disease prevention and healthcare show comparatively similar patterns, although disease prevention has a slightly higher share of excellent health literacy than healthcare. This may indicate that respondents are somewhat more confident in using information about preventive behaviours than in using information related to illness or medical treatment.

Figure 2: Health literacy by domain.

Source: Author’s calculations based on the Health Literacy Study in Latvia survey, 2026.

The competency-level (access, understand, appraise, apply) results are also important for policy. The Latvian population reports the greatest difficulty in applying health information, with 27.25% showing inadequate health literacy skills. Accessing health information is another weak point, with 25.81% showing inadequate and 44.89% problematic skills. Understanding information is the strongest competency, with the highest shares of sufficient and excellent health literacy among the four competencies.

Figure 3: Health literacy by competency.

Source: Author’s calculations based on the Health Literacy Study in Latvia survey, 2026.

Socio-economic Inequality Shapes Health Literacy

Examining a wide set of potential determinants of health literacy – including age, sex, place of residence, ethnicity, language proficiency, partnership status, household size, education, labour status, income, and self-assessed social status – bivariate analysis identifies income, social status, and English proficiency as positively associated with health literacy, while age is negatively associated with health literacy.

Multivariate analysis further shows that income and social status are the most consistent and statistically significant determinants. Social status emerges as the strongest predictor, and income as the second. These findings indicate that the ability to process and use health information is unequally distributed across social and economic groups.

Education does not show the statistically significant relationship that is commonly observed in international literature (Pelikan et al., 2014).

Language proficiency is consistently associated with health literacy. Latvian, Russian, and English proficiency are all statistically significant determinants, with English proficiency showing the strongest association. Language skills shape the ability to engage with health information for all age groups, but English proficiency may be linked with younger cohorts and with other skills, such as digital skills, that were not directly measured in the study.

Health Literacy Is a Pathway into Health Inequality

Apart from assessing health literacy levels, it is also important to examine whether health literacy serves as a pathway through which socio-economic status influences health, as noted by Berete et al. (2024). The mediation analysis evaluates education, income, and social status in relation to self-assessed health status.

Figure 4: Results from the mediation analysis: decomposing the total effect of socio‑economic factors on self‑reported health into direct effects and indirect effects mediated by health literacy.

Source: Author’s calculations based on the Health Literacy Study in Latvia survey, 2026. Note: The x-axis, and therefore the bar length, shows the total effect of each socio-economic factor on self-reported health. Percentages on the right indicate the share of the association mediated by health literacy. The mediated effect is statistically significant at the 0.1% and 5% levels for income and social status, respectively, but reaches significance only at the 10% level for education.

Health literacy is found to be a partial mediator for all three socio-economic status indicators used (education, income and self-attributed social status). It accounts for 8.42% of the association between education and health status, 22.41% of the association between income and health status, and 21.71% of the association between social status and health status. The mediating effects are strongest for income and social status, reinforcing the conclusion that health literacy is closely connected to socio-economic inequalities, while level of education is not as strongly related.

What the Findings Imply

In Latvia, health literacy, the ability to navigate health information and make informed health decisions, should be a public health concern. Limited health literacy is widespread and affects three out of four individuals in the population. It resonates with the current performance of Latvia in terms of different key health indicators – life expectancy lags the OECD average by 5.5 years (OECD, 2025), preventable and treatable mortalities both remain very high, and the self-rated health status is low.

The weakest competencies are in accessing and applying health information, making them the areas that need immediate action. Without adequate access to information, other stages of information processing lose their relevance. The apply-information competency area shows the highest level of inadequate and the lowest level of excellent health literacy, meaning the respondents have trouble using health information in their everyday life.

Further, health literacy is socially patterned, affecting especially lower-income and lower social status individuals, and partly mediates the link between socio-economic status and health status.

Hence, improving health literacy is relevant both for individual behaviour change and for reducing health inequalities. Policies that make health information easier to find, easier to assess, and easier to use may be especially important for lower-income and lower-status groups, who face the greatest risk of limited health literacy.

Policy Recommendations

The evidence points to a need for targeted interventions to improve health literacy in Latvia. The following policy interventions are proposed as priorities for policy development.

  1. Shift from providing information to making information usable

Because access and application are among the weakest competencies, health communication should be redesigned around practical usability. It should be evaluated if there are sufficient and accessible public health materials. Information should help people understand what to do next, not only describe risks or services. General practitioners (family doctors) may play key role in the information management. Opportunities for the use of Artificial intelligence in information flow and easy-to-understand, short, animated videos should be developed.

  1. Target socio-economically vulnerable groups

Income and social status are the strongest determinants of health literacy. Interventions should therefore prioritise people with lower income and lower social status. This could include community-based counselling, support through primary care, cooperation with municipalities and social services, and tailored outreach in settings where vulnerable groups already receive services.

  1. Strengthen health literacy through targeted education and lifelong learning

The weak role of formal education in the Latvian data indicates that formal schooling alone does not currently translate into higher health literacy. Policy efforts should therefore prioritise targeted skill development across the life course, rather than relying on general educational attainment. Strengthening health literacy within school curricula remains important, but equal emphasis is needed on adult learning and community‑based programmes that build practical competencies: locating reliable information, critically evaluating health claims, navigating digital health tools, and applying health advice in everyday contexts. Enhancing English‑language proficiency—particularly written comprehension among adults—may further expand access to high‑quality health information beyond Latvian‑language sources, supporting more equitable engagement with health guidance.

Conclusion

Health literacy in Latvia remains a significant public health concern. Limited competencies are widespread, with many individuals struggling particularly to access and apply health information – difficulties that are closely linked to poorer health outcomes and patterned by socio‑economic inequalities. Strengthening health literacy should therefore be viewed as an integral component of broader strategies to improve population health. This need is especially pronounced among socio‑economically vulnerable groups, who consistently demonstrate lower health literacy levels and, consequently, worse health outcomes.

Acknowledgements

This research is funded by the Latvian Council of Science project “Health Literacy Study in Latvia: Levels, Burdens and Interventions”, project No. Izp-2024/1-0543.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

Towards a Russian Internet?

Posted on June 15, 2026 | Policy Brief

The internet enables information and opinions to flow rapidly and at low cost, including across national borders. It allows individuals to coordinate collective action on an unprecedented scale. Many authoritarian governments, therefore, seek to control the online information environment. This policy brief examines the evolution of internet control in Russia and documents how censorship and network disruptions have intensified since the full-scale invasion of Ukraine.

Drawing on three complementary datasets—Access Now’s Shutdown Tracker Optimization Project (STOP), the Internet Outage Detection and Analysis (IODA) initiative, and the Open Observatory of Network Interference (OONI)—we document a sharp increase in internet disruptions, platform blocking, and website censorship in Russia since 2022. We argue that this expansion of censorship was enabled by a longer-term shift from a relatively decentralized system of internet regulation towards a centralised infrastructure capable of monitoring, filtering, and controlling internet traffic at scale.

The Online War

The Russian government has been fighting its war against Ukraine on multiple fronts: aside from the battlefield in Ukraine, there is also an online front at home. Since the start of the full-scale invasion in February 2022, the Kremlin has sought to control how the war is presented to the Russian public. Within weeks of the invasion, Russia blocked Facebook, Instagram, and Twitter, and restricted access to numerous foreign and independent media outlets. In the years that followed, restrictions expanded to include VPN services and specific features of messaging platforms such as WhatsApp and Telegram.

These measures are not isolated events. Rather, they represent the latest stage in a broader effort to control the Russian internet. Over the past decade, Russia has gradually built the legal and technical infrastructure needed to monitor, filter, and disrupt online communications. The war in Ukraine has revealed the extent of these capabilities and accelerated their deployment.

This brief examines how internet control in Russia has evolved in recent years. We discuss the challenges of measuring internet censorship and analyse evidence from three complementary datasets. Together, these measures show a sharp increase in internet disruptions, platform restrictions, and censorship since the full-scale invasion of Ukraine.

Why Control the Internet?

The internet has transformed the way information is produced, shared, and consumed. It allows information and opinions to spread rapidly at low cost, including across national borders, and enables individuals to coordinate collective action on an unprecedented scale. For an authoritarian government, the internet allows the spread of information that contradicts official narratives and can facilitate political opposition. The role of social media in mobilising protests during the Arab Spring highlighted the power of the internet.

In response, many authoritarian governments have sought to exert greater control over the internet. China is the most advanced example of state control over the online information environment. Through its “Great Firewall”, the Chinese government controls access to foreign information and platforms, while influencing and monitoring domestic online activity through censorship, regulation, and the cooperation of domestic technology companies.

Russia has historically taken a different approach. Until recently, Russians retained access to many Western platforms, and internet censorship was implemented in a relatively decentralized manner by internet service providers. Rather than constructing a separate internet from the outset, Russia sought to control information flows while remaining integrated with the global internet.

Authoritarian Trade-offs

Why might the Russian government have followed this light-touch approach, and why change course now? The literature in economics and political science describes two trade-offs an authoritarian government faces when deciding how much control to exert over the internet.

The first is economic. Describing the ‘dictator’s dilemma,’ Kedzie (1997) writes, “it may now be virtually impossible for any country to maintain an open economy for expansion while remaining closed to democratic ideas“. Estimates of the economic cost of internet shutdowns support this argument. One estimate suggests that government-imposed outages cost the global economy $19.7 billion in the year 2025 (Migliano 2026).

The second is informational. Egorov et al. (2009) argue that an authoritarian government that constrains free media and communication flows too aggressively cuts itself off from information required to govern effectively. Local bureaucrats have no incentive to perform in the absence of reliable independent monitoring. King et al. (2013) provide empirical evidence for this in the context of Chinese social media censorship. They find that censors allow (potentially informative) criticism of the government but specifically target posts that could give rise to collective action.

Controlling the narrative around a prolonged, costly war necessitates a greater level of intervention. The censorship strategies discussed below can be viewed through the lens of a government seeking new ways to navigate both trade-offs.

Tracking Internet Shutdowns

Identifying government-imposed internet shutdowns is challenging. Affected users will typically have no way of verifying the extent or true cause of an outage, and their ability to report it in real-time may itself be curtailed. As a result, organizations that monitor internet shutdowns use very different methodologies. In our brief, we will describe three of the most prominent publicly available datasets that attempt to track disruptions to internet services worldwide.

Access Now, a member of the #KeepItOn coalition, provides a publicly available dataset of internet shutdowns through its Shutdown Tracker Optimization Project (STOP). The distinguishing feature of STOP is that it establishes intent. It combines technical data on internet connectivity with either official government statements or information provided by informed insiders. STOP records various types of technical disruption: from full blackouts to throttling and partial service restrictions. However, a measured disruption of internet services is only recorded as a shutdown event if it can be traced back to deliberate government intervention with a high degree of confidence. The advantage of this approach is that one can be confident that each instance in the data reflects a deliberate government-induced shutdown. The limitation is that the dataset likely undercounts shutdowns in data-scarce or highly repressive environments where establishing intent is not always possible.

Figure 1 plots internet shutdowns in Russia, and for comparison, the FREE network member countries from 2016 to 2025. The chart is dominated by the sharp upward trend in Russia, starting in 2021, and accelerating after the full-scale invasion of Ukraine. There is also an increase in Ukraine after the invasion; which includes both Russian actions that disrupted connectivity and Ukrainian measures to block specific Russian platforms. Similarly, Latvia imposed nationwide blocks of two Russian platforms after the start of the invasion. The chart also shows the internet blackouts in Belarus amid widespread protests following the 2020 election.

Figure 1. Internet shutdown events in FREE Network countries (2016-2025)

Source: Access Now, KeepItOn STOP (2016-2025) and authors’ calculations.
Note: This chart shows the number of distinct intentional internet shutdown events active during each quarter in countries of the SIDA Free Network. (Sweden, Georgia, Moldova, and Poland are excluded from the graph because no shutdown events were recorded for them over this period.)  An internet shutdown is defined as an intentional disruption of internet or electronic communications, rendering them inaccessible or effectively unusable, for a specific population or within a location, often to exert control over the flow of information (Access Now, KIO). A shutdown event can result from third-party interventions rather than be intended by the country’s government.

Figure 2 uses the same dataset to illustrate which social media and online messaging platforms were most affected by the increase in government control of the internet in Russia. Relative to China, Russia used to exert only ‘light-touch’ control over the internet, as seen in the first panel of the figure. From 2016 to 2021, the social media and online messaging platforms in the chart were largely unaffected by shutdown events. The second panel shows that since early 2022, all of these platforms have experienced shutdown events, and in the case of Facebook, Twitter/X, and Instagram, there have been active blocks throughout the entire period.

Figure 2. Platform service disruptions in Russia (2016 – 2025)

Source:  Access Now, KeepItOn STOP Dataset (2016-2025) and authors’ calculations. Note: This graph details the platforms affected by internet shutdowns in Russia (See Figure 1). Each bar shows the percentage of days in the period when at least one shutdown event affected the platform in Russia.

As discussed, the STOP dataset likely undercounts internet shutdowns. We therefore evaluate whether alternative measures show the same upward trend for Russia.

Our second dataset comes from the Internet Outage Detection and Analysis (IODA) initiative at Georgia Tech, which monitors global internet connectivity using three complementary technical signals to detect when networks go offline. IODA identifies outages at the country, regional, or network level and records their duration and severity. Importantly, unlike STOP, IODA detects outages but not their cause. An outage may reflect deliberate government action or infrastructure failure.

Panel (a) of Figure 3 compares the IODA measure of outages with the STOP measure of internet shutdowns for Russia. The IODA measure (right axis) is roughly 100 times as high as the STOP measure in any given period, as IODA records all detected disruptions regardless of intentionality. That said, the IODA data corroborate the finding that internet disruptions have become ever more frequent in Russia, with significant increases in 2024 and 2025.

Our third dataset comes from the Open Observatory of Network Interference (OONI), which focuses on censorship rather than outages. OONI relies on volunteers running tests through an open-source app, generating measurements of whether specific websites, messaging platforms, and circumvention tools are accessible or blocked.

Figure 3. Trend in internet disruption and online censorship in Russia (2022-2025)

a. Count of internet disruptions

b. Rate of websites and apps censorship

Source: Access Now KeepItOn STOP (KIO), Internet Outage Detection and Analysis (IODA), Open Observatory of Network Interference (OONI), and authors’ calculations.
Note: Panel (a) shows STOP internet shutdown events in the blue line, which records intentional disruptions of internet or electronic communication (KIO, see Figure 1), and IODA internet outages in the orange line, which are abnormal simultaneous drops in 2 or more signals measuring internet connectivity, intentional or accidental. IODA outages are filtered to only include events lasting more than 2 hours to match the KIO restriction. The red line shows a twelve-month moving average of IODA outages. Panel (b) shows online censorship rates for websites and messaging apps, measured by the monthly rate of anomalies recorded by OONI. An anomaly is detected when a measurement presents signs of potential network interference (such as the blocking of a website or app). Messaging apps data derive from OONI messaging platform availability tests (WhatsApp, Telegram, Signal, Facebook Messenger), and website data from OONI Web Connectivity tests on individual websites’ availability. Days and platforms or websites with fewer than 5 measurements are excluded for reliability.

It does so by comparing results over the user’s network against a control server, with divergences flagged as potential interference. The main limitation is uneven coverage over time, a consequence of the volunteer-based approach, though the total number of daily measurements is always known.

Panel (b) of Figure 3 plots anomaly rates for websites and messaging apps as experienced by Russian users since the start of 2022. Both lines show a clear upward trend, indicating that Russian users are increasingly encountering websites and apps that are blocked in Russia but available elsewhere.

The Centralisation of Russian Internet Control

While Russia has always exerted some degree of control over the internet, it has historically relied on what Ramesh et al. (2020) call a decentralised model. Since 2012, Russia’s internet regulator, Roskomnadzor, has maintained a national blocklist of websites and required Internet Service Providers (ISPs) to restrict their users’ access to these websites. As ISPs were granted full discretion over how to comply, the blocking mechanisms and their effectiveness reportedly varied significantly across websites and providers.  The attempted blocking of Telegram in 2018 exposed the limitations of Russia’s decentralized approach to internet censorship. To enforce the ban, Roskomnadzor blocked millions of IP addresses associated with Amazon and Google cloud services, leading to widespread disruption of unrelated online services, while failing to prevent Russian users from accessing Telegram.

Since then, Russia has moved towards a more centralised model of internet governance, aimed at increasing state control over its domestic internet and reducing dependence on the global network. In 2019, the “Sovereign Internet Law” (or  “Law on Sustainable Runet”) came into force, which provided the Russian state the legal and technical tools to centrally monitor, filter and reroute internet traffic. The law requires ISPs to install TSPU (Tekhnicheskie Sredstva Protivodeystviya Ugrozam, or “technical means of countering threats”) devices on their networks, or face fines. These devices allow the government to track and manage internet traffic across private networks in a centralised manner (Human Rights Watch 2025).

TSPUs first attracted attention in 2021 when access to Twitter was throttled, but their impact has since become more widespread (Xue et al., 2021). In February 2023, OONI and the Russian digital rights organisation Roskomsvoboda reported that numerous media outlets and websites with critical coverage of the Russian war in Ukraine had been blocked in 2022. In a striking contrast to previous decentralised censorship practices, these restrictions were implemented simultaneously across internet providers. Figure 4, originally published by OONI, illustrates the simultaneous blocking of the Human Rights Watch website after it was added to Roskomnadzor’s blocklist on April 17, 2022. The figure shows that users across multiple networks lost access at the same time. The OONI report also highlights that providers are now using the same technical methods to enforce central directives, illustrating the widespread effective use of TSPUs.

Figure 4: Network interference in Russia

Source: Open Observatory of Network Interference (OONI), Roskomsvoboda, How Internet censorship changed in Russia during the 1st year of military conflict in Ukraine.
Autonomous System Numbers (ASNs) which presented the largest volume of anomalies (more than 1,200 anomalies) in the testing of www.hrw.org in Russia between 1st January 2022 to 20th February 2023. An anomaly is detected when a measurement presents signs of potential network interference (such as the blocking of a website or app).

Recent restrictions on Telegram provide further evidence of their efficacy. In contrast to the failed block in 2018, most Russian users are now unable to access the app without VPNs or other workarounds.

The capabilities of TSPUs extend far beyond individual website blocking. Recent reports suggest that instead of just targeting specific parts of the internet, they are now used to impose temporary, near-total internet blackouts. These cut users off from much of the global internet, while preserving access to a whitelist of Russian government websites and fully cooperative platforms (Human Rights Watch, 2026). In doing so, TSPU moves Russia closer to the Chinese model of internet control, increasing the state’s ability to manage internet traffic centrally.

Conclusion

In recent years, Russia’s strategy towards internet censorship has changed profoundly. The decentralised, relatively light-touch approach, with unrestricted access to many Western platforms, has been abandoned. The government has acquired the legal and technical capability to exert tight, centralised control over service providers, and every indicator we have analysed in this brief makes clear that it is using these powers ever more aggressively.

The incremental nature of these changes and their technical sophistication mean that Russia’s internet control cannot be equated with the blunt tool of country- or region-wide shutdowns as used by authoritarian governments in other parts of the world. These types of internet shutdowns are highly visible to domestic and international users and spark outrage. Russia’s strategy is more insidious. Its citizens’ access to the global internet has been shutting down, year by year and month by month. Individual users’ experience of these changes is fragmented, making a collective response difficult.  Russia is on its way to creating and controlling its very own version of the internet.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.

Mobile Learning and Indigenous Knowledge for Climate Resilience

Posted on June 8, 2026 | Policy Brief

Climate resilience in Sub-Saharan Africa depends on more than technological innovation. It also requires the preservation and transmission of locally embedded agricultural knowledge. Drawing on evidence from the ICRAFS project in Northern Ghana, this brief examines the conditions under which voice-based mobile learning can support climate adaptation among smallholder farmers. The audio modules were delivered through Farmerline’s Mergdata platform and reached more than 50,000 farmers in six local languages. They combined mobile learning with indigenous farming knowledge and Latvian seed-saving experience. Survey and fieldwork evidence show that digital extension is most effective when it is trusted, timely, intelligible across local languages and dialects, attentive to household decision-making, and linked to farmers’ economic realities. The brief argues that climate-resilient food systems require hybrid knowledge systems: scalable digital tools combined with community-based knowledge, locally adapted farming practices, and institutional support. It also shows how comparative lessons, such as Latvia’s seed-saving experience, can inform policy without being replicated directly.

Digital Extension and Climate Resilience

Climate change is increasing production risks for smallholder farmers, particularly in regions where food systems are highly vulnerable to environmental and economic shocks. In this brief, climate resilience refers to the ability of smallholder farmers and the wider food systems they depend on to adapt to climate-related shocks while sustaining production and livelihoods. In Sub-Saharan Africa, rising temperatures, irregular rainfall, soil degradation, and growing pest outbreaks are placing increasing pressure on agricultural productivity and rural livelihoods.

At the same time, access to mobile technologies has expanded rapidly. Even in remote and hard-to-reach rural areas, mobile phones have become a central infrastructure for communication and information exchange, alongside local radio stations and community hubs. However, many smallholder farmers continue to rely on basic feature phones rather than smartphones, making voice-based communication systems especially important for agricultural extension services.

International organisations and development agencies, including the Food and Agriculture Organization of the United Nations (FAO), the World Bank, and the United Nations Development Programme (UNDP), have increasingly supported digital agricultural extension and mobile-based advisory systems as part of broader efforts to strengthen climate resilience and food security. Throughout this brief, digital agricultural extension refers to the use of mobile and digital tools to deliver farming advice and information. These systems create new opportunities for delivering agricultural knowledge at scale.

However, digital access alone does not automatically improve climate resilience. Extension systems that focus mainly on technological delivery risk overlooking social realities such as literacy barriers, local ecological knowledge, gendered patterns of information-sharing and decision-making, and trust in advisory systems. The key policy question is therefore not whether digital tools can reach farmers, but whether they can reach them in ways that are trusted, understandable, timely, and useful for real-time farming decisions.

Evidence from Northern Ghana suggests that digital agricultural extension is most effective when technology functions as a mediator that strengthens and amplifies existing community knowledge systems.

The ICRAFS Project: Evidence from Northern Ghana

The “Building Digital Education of Indigenous/Heritage Crops for the Resilience of African Food Systems in the Climate Crisis” (ICRAFS) project brought together Latvian and Ghanaian partners, including BICEPS, Farmerline, CSIR-Savanna Agricultural Research Institute, and the Latvian Permaculture Association, a leading organization behind Latvia’s seed savers movement. By combining development policy expertise, digital extension infrastructure, and local agricultural research capacity, the project explored how indigenous knowledge and digital learning can support climate adaptation among smallholder farmers. As part of the project, two educational modules were developed, each consisting of ten short audio lectures. The lectures covered regenerative agriculture, indigenous crops, seed-saving practices, soil management, pesticide reduction, and climate adaptation strategies. To ensure accessibility, the content was delivered in six local languages (Dagbani, Konkomba, Buli, Sissale, Chokosi, and English) through voice-based systems compatible with basic feature phones.

The audio content reached more than 50,000 farmers through Farmerline’s Mergdata platform, a Ghanaian digital agricultural extension system that combines mobile-based advisory services with AI-enabled tools designed for low-connectivity and low-literacy rural contexts. The educational content and broader community-based approach were developed in cooperation with the Latvian seed savers movement, whose experience in preserving heirloom crops and decentralised seed-sharing networks informed both the training materials and the wider vision for strengthening local resilience systems in Ghana.

The analysis draws on a baseline survey of 367 farmers and mixed-methods fieldwork with 60 farmers in Northern Ghana. Together, these data provide insight into the conditions under which mobile-based agricultural learning can support climate adaptation, highlighting the importance of design choices around language, trust, timing, and local knowledge.

Lessons From the Latvian Seed Savers Movement

The Latvian seed savers movement provided an important practical and conceptual reference point for the ICRAFS project.

The Latvian case shows that food-system resilience also depends on cultural memory, social networks, and ecological diversity. Community-based seed-saving systems contribute to ecological resilience by preserving genetic diversity, locally adapted crop varieties, and cultivation practices developed through long-term interaction with specific environmental conditions. Heirloom crops function not only as biological resources but also as carriers of local memory, identity, and intergenerational knowledge. In Latvia, seed-saving also helps preserve agricultural knowledge, rural identity, and community ties in the context of demographic decline and countryside depopulation.

For digital agricultural systems in Sub-Saharan Africa, the Latvian case offers a useful policy lesson. It should not be treated as a model for direct replication, but as evidence that resilient food systems depend on institutions and networks that protect local varieties, preserve practical knowledge, and maintain trust between farmers, communities, and knowledge holders. Educational initiatives are more effective when they connect climate adaptation and agricultural advice to local cultural experience and community-based knowledge systems.

Main Findings

Digital Access and Adoption

Research findings suggest that mobile phone ownership was nearly universal among surveyed farmers (98%), despite the majority of respondents relying on basic feature phones rather than smartphones. Approximately 72% reported receiving educational agricultural messages via mobile phones, while around 70% stated that they actively applied the information received to production and input-purchasing decisions.

Trust in digital advisory systems was also relatively high: among farmers who had received educational messages, approximately 78% reported trusting the source. Engagement data showed particularly high interest in traditional agricultural knowledge, regenerative farming methods, and seed-saving practices, with listening rates exceeding 70%.

These findings suggest that reach matters, but it is not sufficient. Digital extension becomes meaningful when farmers trust the source, understand the content, and can apply the information in farming and input decisions.

The strongest initial interest was recorded for content on diversification into indigenous crops. However, completion and listening rates suggest that initial curiosity is not enough to sustain engagement. For digital learning to remain useful, content on crop diversification needs to be closely connected to farmers’ immediate production choices, input constraints, market opportunities, and expected livelihood benefits.

In this context, digital agricultural extension services are most useful when they improve accessibility and personalisation, for example through voice-based interaction, local-language delivery, farmer profiling, and more targeted advisory content.

Indigenous Knowledge and Hybrid Resilience Systems

Evidence from Northern Ghana demonstrates that climate resilience is deeply rooted in locally embedded knowledge systems developed across generations. Farmers rely on sophisticated ecological knowledge to manage agricultural risks under uncertain climatic conditions.

Field observations identified a range of indigenous adaptation practices, including traditional composting systems, organic soil-fertility management, smoke-based seed preservation methods, and the use of ash to protect stored crops from pests. Farmers also rely on ecological indicators to interpret environmental change. For example, the presence of the weed Striga (“bochaa”) is widely understood as a sign of declining soil fertility or soil acidity.

High engagement with content on indigenous crops further suggests strong farmer interest in locally adapted crop systems as part of climate adaptation strategies.

Importantly, farmers do not perceive digital information as a replacement for indigenous knowledge. Instead, mobile-based advisory systems are integrated into existing local decision-making processes. This creates a hybrid resilience model in which indigenous knowledge provides contextual interpretation, while digital systems support broader dissemination, coordination, and real-time information sharing. However, the effectiveness of these hybrid resilience systems remains constrained by broader structural inequalities.

Structural Barriers to Effective Digital Extension

Despite the potential of mobile-based agricultural learning, several structural barriers continue to limit effectiveness.

Gender, language, and accessibility

Digital extension is not equally accessible to all farmers. Mixed-methods evidence from Northern Ghana shows that gender roles shape both access to technology and agricultural decision-making. Although women represented the majority of respondents, men often remained household gatekeepers, controlling mobile-device ownership and broader agricultural resource allocation. This creates asymmetries between formal access to information and effective control over decisions. It also means that digital extension must consider how agricultural information is shared within households, not only whether a message reaches a phone.

Low literacy levels further reinforce the importance of voice-based communication. Approximately 62% of respondents reported having no formal education, and farmers consistently found voice-based messages in local languages easier to understand than text-based formats. However, linguistic accessibility goes beyond translation: dialects, locally used farming vocabulary, and informal agricultural expressions strongly shape whether advice is understood, trusted, and acted upon.

Timing and economic incentives

Agricultural information is highly time-sensitive. Farmers reported that advice loses value if delivered outside key planting, harvesting, pest-management, input-purchasing, or output-selling windows. This means that digital extension depends not only on content quality, but also on synchronisation with local agricultural and market calendars.

Farmers also evaluate new practices through immediate economic outcomes. Long-term climate adaptation goals alone are often insufficient under conditions of economic vulnerability. Qualitative interviews suggest that farmers are more likely to adopt new practices when they are connected to visible benefits such as improved yields, lower input costs, reduced risk, or stronger market opportunities.

Overall, the main policy challenge is not simply expanding digital access, but designing extension systems that are trusted, locally intelligible, timely, and economically relevant. Voice-based delivery can help overcome literacy and smartphone barriers, but it is most effective when integrated with local radio, farmer groups, community leaders, and extension agents.

Policy Recommendations

The following recommendations are particularly relevant for rural agricultural regions in Sub-Saharan Africa and other low-income contexts where smallholder farming, limited digital infrastructure, low literacy rates, and dependence on feature phones continue to shape access to agricultural information and climate adaptation resources. They are addressed to agricultural ministries, extension agencies, donors, digital service providers, development partners, and farmer organisations working in rural, low-connectivity contexts.

  1. Shift voice-based systems from providing information to making information usable.

Voice-based advisory systems should be accessible through basic feature phones and combined with local radio, TV, farmer groups, and community-based extension channels. Their success should be assessed not only by reach, but also by whether farmers understand, trust, and use the information in agricultural decisions.

  1. Integrate indigenous knowledge into digital extension design.

Digital agricultural content should be co-designed with farmers, local extension agents, and community knowledge holders. Indigenous knowledge, local dialects, and farmer-used agricultural vocabulary should be integrated into digital platforms to improve trust, cultural relevance, and adoption.

  1. Account for gendered decision-making in digital extension.

Digital extension should account for the fact that receiving agricultural information does not always mean having the authority, resources, or training needed to act on it. Household decision-making over land, inputs, sales, technology use, and production priorities may still be shaped by gender roles. Advisory services should therefore support direct communication with women farmers, while also encouraging household-level information-sharing so that relevant advice reaches those involved in production and household food security.

  1. Align advisory services with agricultural and market calendars.

Advisory messages should be timed around key planting, harvesting, pest-management, input-purchasing, and output-selling periods, reaching farmers before decisions are made.

  1. Link climate adaptation to immediate economic benefits.

Climate adaptation should be communicated through immediate livelihood gains, linking regenerative agriculture, indigenous crops, seed-saving, and soil-management practices to improved productivity, lower input costs, reduced risk, and stronger market opportunities.

  1. Support community seed-saving and knowledge networks.

Community seed-saving initiatives, indigenous crop preservation, and farmer-to-farmer knowledge networks should be supported. Latvian experience shows that such systems can strengthen agrobiodiversity, reduce dependency on external inputs, and preserve locally adapted knowledge. In African contexts, these models should be adapted through local farmer organisations, extension services, and community institutions rather than replicated directly.

Conclusions

The evidence from Northern Ghana shows that digital agricultural extension can support climate resilience when it is designed around the realities of smallholder farmers. Voice-based mobile learning can reach farmers with limited literacy, limited internet access, or only basic feature phones, but technology alone is insufficient to change agricultural practices. Digital extension is most effective when it is trusted, accessible in local languages and dialects, responsive to how farming decisions are made within households, delivered at the right moment in the agricultural or market cycle, and linked to visible economic benefits.

Indigenous knowledge should not be treated as a barrier to innovation. Rather, it provides the local interpretation, ecological memory, and practical experience that make digital advisory systems more relevant and credible. The Latvian seed savers movement offers a useful comparative lesson: community-based knowledge systems can preserve biodiversity, strengthen rural identity, and maintain trust across generations.

The main policy implication is to treat digital extension as part of a broader rural resilience strategy. Investments in digital advisory systems and voice-based communication should be accompanied by support for local-language content, community institutions, farmer-to-farmer knowledge exchange, and locally adapted seed-saving and farming practices.

Acknowledgements

This research was conducted within the project “Building Digital Education of Indigenous/Heritage Crops for the Resilience of African Food Systems in the Climate Crisis” (ICRAFS), supported by the Development Cooperation Programme (LATDEV) of the Ministry of Foreign Affairs of the Republic of Latvia for 2024–2025. Grant agreement No. LV-72.

The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the Ministry of Foreign Affairs of the Republic of Latvia.

References

Disclaimer: Opinions expressed in policy briefs and other publications are those of the authors; they do not necessarily reflect those of the FREE Network and its research institutes.