This article was cross-posted on VoxEU.
Mobile-based digital initiatives have emerged as a popular tool for governments, NGOs, and the private sector to influence individual behaviour. With thousands of initiatives spanning sectors like education, agriculture, finance, health, and governance (GSMA 2020), these programmes take advantage of the widespread adoption of mobile phones to deliver messages at scale and at a low cost. In many low-income countries (LICs), where basic phones are still more common than smartphones (Pew Research 2019), text-message-based programmes are a particularly practical and inexpensive option. However, its limitations —such as character constraints and reliance on top-down communication—raise questions about the effectiveness of these programmes. As a result, there is significant interest in evaluating their cost-effectiveness and exploring whether specific design and implementation choices could strengthen their impact.
In agriculture, mobile-based extension programmes hold significant promise for disseminating information about modern inputs and management practices to millions of smallholder farmers who would otherwise be difficult to reach through in-person visits. While we are still learning about the impacts of various of these digital initiatives, existing summaries of these evaluations have characterised the results as mixed (Aker et al. 2016, Deichmann et al. 2016). Some studies report positive and significant changes in farmer behaviour, while others find statistically insignificant effects. However, mixed results and null effects could stem from measurement issues or small sample sizes that lack sufficient power to detect impacts. Low statistical power is particularly concerning when evaluating low-cost programmes, such as those relying on text messages, because even small impacts might be sufficient for these programmes to be considered cost-effective.
Six digital programmes that aimed to encourage farmers to experiment with agricultural inputs
In Fabregas et al. (forthcoming), we evaluate six different text-message-based agricultural extension programmes that collectively reached over 128,000 farmers in Kenya and Rwanda. These programmes were implemented by three organisations: a public entity (the Kenya Agriculture and Livestock Research Organization, KALRO), a partnership between non-profits (Innovations for Poverty Action and Precision Development, IPA/PxD), and a social enterprise (One Acre Fund, 1AF). All programmes targeted smallholder maize farmers, sending text messages with information about recommended inputs, focusing particularly on agricultural lime—a soil additive used to reduce soil acidity that was relatively unknown in these areas.
While the programmes shared a common objective—encouraging farmers to experiment with locally appropriate inputs—their implementation differed based on the specific constraints and opportunities faced by each organisation. These variations included farmer recruitment strategies, the design and framing of messages, the range of recommended inputs and practices, the emphasis ontailoring information to local soil conditions, the use of behavioural nudges, the frequency of messages, and the availability of supplementary support, such as phone calls.
We partnered with these organisations to conduct experimental evaluations of each programme. To increase statistical power and formally test for impact heterogeneity across evaluations, we also conducted a meta-analysis. A key strength of our approach is the use of large sample sizes, which allow us to detect small effect sizes. Additionally, for each project, we can use objective measures of farmer behaviour, such as administrative data on input purchases, to reduce concerns that observed impacts are simply driven by experimenter demand effects. For some projects, this administrative data is complemented by self-reported survey data.
Figure 1: Effects on recommended inputs and practices
Effects on recommended inputs and practices
Note: This figure shows a meta-analysis of all six experimental evaluations of text-based agricultural extension programmes implemented by KALRO, IPA/PxD, and 1AF either in Kenya or Rwanda. The dependent variable captures whether the farmer followed the recommended inputs or practices mentioned by the programme, as measured by administrative data (input purchases or coupon redemption) when available or survey data otherwise. Effects are in odds ratios, a relative outcome measure.
Key insights from evaluating six text-message-based agricultural extension programmes
Text messages have modest impacts but can be extremely cost-effective ***.***The meta-analysis showed a combined 1.22-fold increase in the odds of following the recommended inputs and practices, corresponding to a two-percentage point increase using an absolute measure of impact. While some projects had statistically significant impacts and others did not, we cannot reject the hypothesis that projects had similar effects.
Focusing on agricultural lime, an input recommended by all programmes, and using administrative purchase data to measure uptake, we estimate a combined significant 19% increase in the odds of farmers following the recommendation. For fertiliser, a more widely known input in these areas, the combined increase in odds is 27%.
Overall, despite the modest absolute impacts, the low cost of text messaging makes these interventions highly cost-effective. A back-of-the-envelope calculation suggests a benefit-cost ratio of about 46:1 if operated at a sufficiently large scale, such that the per-farmer fixed costs become trivial.
No signs of crowding out of other inputs or message fatigue. We find little evidence to suggest that these programmes crowded out the purchase of other, non-recommended inputs. Additionally, while the effects tended to diminish in the following season, they persisted as long as farmers continued to receive messages, suggesting that these programmes may work by keeping the recommendations at the top of farmers’ minds.
Other lessons for designing phone-based programmes
We take advantage of randomisation in individual projects to learn additional lessons from the designs of these programmes.
Emphasising information granularity did not significantly increase impacts. More localised agricultural information is likely to be of higher quality. However, emphasising that recommendations targeted local soil conditions did not significantly persuade farmers to follow the advice compared to broader messages. This aligns with work in other contexts (Corral et al. 2020, Beg et al. 2024).
Behavioural framings were equally effective. In two projects, messages randomised behavioural framings such as loss/gain, social comparisons, sense of urgency, etc. However, all framings had effects similar to those of a simple message. It is possible that larger sample sizes are needed to detect framing effects. Since the cost of optimising messages is very low, this is an area that warrants further exploration.
Message repetition increased effects. Re-sending messages had a very small but statistically significant effect on increasing the likelihood of adopting recommended inputs.
Phone calls did not strengthen impacts. One project experimentally added a phone call with a field officer to explain the messages, but no new information was delivered. This add-on did not have any additional statistically significant effects.
Accounting for spillovers will likely increase the estimated benefits. We find suggestive evidence of spillovers, particularly among those without phones who were part of active farmer groups. This implies that the benefits of these types of interventions are likely underestimated when only considering direct impacts.
**Beware of administrative vs. survey data.**Significant differences were found between survey and administrative data sources. While this could indicate that farmers obtained inputs from sources not captured in the administrative data, there is also some evidence pointing to potential misreporting in the survey responses.
Learning from mobile phone-based interventions to harness the benefits of digital agriculture
The growing adoption of mobile phones provides a valuable opportunity for organisations to reach people at scale. Our evaluation of six text-message-based agricultural extension programmes shows that, while these interventions have modest impacts on farmer behaviour, they can be highly cost-effective due to their low delivery costs. Although it is challenging to draw definitive conclusions about the extent of heterogeneity from just six programmes, the interventions appear to have similar impacts despite differences in their design and implementation.
The limitations of basic text messaging suggest that these impacts are likely lower bounds for what digital agriculture could achieve. As more advanced technologies—such as smartphone apps, interactive platforms, and precise weather and soil data—become increasingly accessible worldwide, the potential forstronger, more personalised, and interactive interventions will grow. The challenge will lie in optimising their design at scale, ensuring that farmers in LICs can fully benefit from these innovations.
References
Aker, J C, I Ghosh, and J Burrell (2016), “The promise (and pitfalls) of ICT for agriculture initiatives,” Agricultural Economics, 47(S1): 35–48.
Beg, S, M Islam, and K W Rahman (2024), “Information and behavior: Evidence from fertilizer quantity recommendations in Bangladesh,” Journal of Development Economics, 166: 103195.
Corral, C, X Gine, A Mahajan, and E Seira (2020), “Autonomy and specificity in agricultural technology adoption: Evidence from Mexico,” National Bureau of Economic Research.
Deichmann, U, A Goyal, and D Mishra (2016), “Will digital technologies transform agriculture in developing countries?” Agricultural Economics, 47(S1): 21–33.
Fabregas, R, M Kremer, M Lowes, R On, and G Zane (Forthcoming), “Digital information provision and behavior change: Lessons from six experiments in East Africa,” American Economic Journal: Applied Economics.
GSMA (2020), “10 years of mobile for development,” Available at: https://www.gsma.com/mobilefordevelopment/10yearsofm4d/.
Pew Research Center (2019), “Smartphone ownership is growing rapidly around the world, but not always equally,” Available at: https://www.pewresearch.org/global/2019/02/05/smartphone-ownership-is-growing-rapidly-around-the-world-but-not-always-equally/.