Digital Advisory Channels for Women Farmers in Malawi

Participants attend a village-based lab-in-the-field session in which they are randomly assigned to receive short agricultural advisory content on soils and soil fertility management through one of two digital delivery channels: (i) SMS and voice messages, or (ii) a smartphone application (Zaulimi). A third group serves as an active control and views a short WASH (water, sanitation, and hygiene) video with no agricultural advisory content. The advisory content is identical across treatment arms; only the mode of delivery differs. The intervention is designed to test whether short-run exposure to digital agricultural advisory services, delivered through alternative mobile-based channels, affects women farmers’ agricultural knowledge, perceived usability of digital services, and short-run investment-related decision-making.

The study is implemented through village-based lab-in-the-field sessions with women members of Village Savings and Loan Associations (VSLAs). Participants are randomly assigned within village sessions to one of three arms: (i) SMS and voice-based delivery of agricultural advisory content (Treatment 1), (ii) smartphone application-based delivery of the same advisory content via the Zaulimi platform (Treatment 2), or (iii) an active control group that views a short WASH (water, sanitation, and hygiene) video unrelated to agriculture.

The agricultural advisory content focuses on soil types, soil fertility management practices, and related input use decisions, and is held constant across treatment arms to isolate the effect of delivery modality. Participants in the treatment arms receive the content through the assigned digital channel during the session.

Following exposure, all participants complete an immediate post-intervention survey measuring agricultural knowledge and perceptions of digital service usability. In a sub-set of randomly selected villages, participants also complete a discrete choice experiment (DCE) designed to elicit stated preferences over key attributes of digital agricultural advisory platforms (e.g., delivery format, interactivity, language, and cost).

To measure short-run behavioural responses, all participants also undertake an incentivized revealed preference task in which they make a real choice between agricultural inputs (fertilizer) and a consumption good (sugar), allowing assessment of how information exposure affects investment-related decision-making.

Randomization occurs at the individual level within village sessions, with equal assignment across the three study arms.

2026-01-19

2026-03-13

Primary outcomes for the main RCT (measured at endline: immediate post-exposure):
1. Soil knowledge index (endline, immediate post-exposure): Standardized index of responses on soil types, soil fertility indicators, and soil management practices (measured at baseline and endline; main endpoint is endline).
2. Revealed choice – fertilizer selected (endline, immediate post-exposure): Binary indicator = 1 if participant chooses fertilizer over sugar in the incentivized choice task; 0 otherwise.
3. Comprehension index (endline, immediate post-exposure): Standardized index of understanding/recall and perceived clarity/reliability of the advisory content (treatment arms only).
4. Usability index (endline, immediate post-exposure): Standardized index of perceived ease of use, accessibility, relevance, affordability, and comparative usefulness of the delivery channel (treatment arms only).
5. Self-efficacy index (endline, immediate post-exposure): Standardized index of confidence in applying the advisory information and expecting improved outcomes (treatment arms only).
6. Intention-to-act index (endline, immediate post-exposure): Standardized index of intended use and continued engagement with the advisory information/channel (treatment arms only).
7. Intended number of changes (endline, immediate post-exposure): Count response to “How many changes are you likely to make?”
8. DCE choice outcome (DCE villages; baseline and endline): Choice of preferred DAT option in each discrete choice task, used to estimate preferences/WTP for platform attributes.

Several primary outcomes are constructed as standardized indices from multiple survey items measured at baseline and/or endline. In particular, the Soil Knowledge Index aggregates responses to questions on soil types, soil fertility indicators, and soil management practices. The Comprehension, Usability, Self-efficacy, and Intention-to-Act outcomes are similarly constructed as standardized indices combining multiple Likert-scale items capturing understanding of advisory content, perceived ease of use and relevance of the delivery channel, confidence in applying the information received, and stated intentions to adopt recommended practices.

For each index, individual items will be coded so that higher values correspond to improved knowledge or more positive evaluations, standardized (z-scored), and combined into an additive index following standard procedures used in field experiments. Further, details of on index construction is in the uploaded pre-analysis plan document. The revealed choice outcome (fertilizer vs. sugar) is analysed as a binary indicator and is not constructed from multiple items.

1. Dissemination and spillover index (endline, post-exposure): Standardized index of items capturing participants’ confidence in explaining the advisory information received and intention to share this information within their household or community.
2. Knowledge of written-only (SMS-only) content (endline, SMS/Voice arm only): Binary indicator equal to 1 if the participant correctly answers a knowledge question based on content presented only in the SMS message (and not in the audio message); 0 otherwise.
3. DCE attribute attendance (baseline and endline, DCE villages only): Binary indicators for whether each digital advisory platform attribute (e.g., cost, format, support, contextualization, market information) is reported as driving the participant’s choices in the post-DCE question.
4. Preferred DAT design (post-DCE; endline, DCE villages only): Participant responses to “build your preferred service” questions capturing preferred device, delivery format, level of contextualization, type of in-service support, language, and additional information features.
5. Knowledge Index – DATs (baseline and, where repeated, endline): Standardized index of items capturing awareness and understanding of digital agricultural technologies and current sources/types of agricultural information received via DATs.
6. DAT utilization (baseline and, where repeated, endline): Binary indicator equal to 1 if the participant reports receiving agricultural information through any digital agricultural technology modality within the questionnaire reference period; 0 otherwise.

The Dissemination and Spillover Index and the Knowledge Index – DATs are constructed as standardized additive indices from multiple survey items coded so that higher values correspond to greater knowledge or higher likelihood of information sharing. Further, details of on index construction is in the uploaded pre-analysis plan document.

The study is an individual-level randomized controlled trial conducted among women members of Village Savings and Loan Associations (VSLAs) in Malawi. Participants attending village-based lab-in-the-field sessions are randomly assigned within session to one of three study arms: (i) receipt of agricultural advisory information via SMS and voice messages, (ii) receipt of the same advisory information via a smartphone application (Zaulimi), or (iii) an active control group that views a non-agricultural video. The advisory content is held constant across treatment arms in order to isolate the effect of delivery modality.

Outcomes are measured through baseline and immediate post-exposure surveys administered during the same session. Primary outcomes include agricultural knowledge and short-run investment-related decision-making, while additional measures capture perceptions of comprehension, usability, and intended use of digital advisory services. A subset of villages also participate in a discrete choice experiment to elicit preferences over digital agricultural advisory platform attributes.

The study is implemented through village-based lab-in-the-field sessions conducted with women members of Village Savings and Loan Associations (VSLAs). Within each session, participants first complete a baseline survey capturing demographic characteristics, agricultural practices, prior exposure to digital agricultural technologies (DATs), and baseline knowledge of soil fertility management practices.

Participants are then randomly assigned at the individual level within session to one of three arms: (i) SMS and voice-based delivery of agricultural advisory content on soil types and soil fertility management practices (Treatment 1), (ii) delivery of the same advisory content via a smartphone application (Zaulimi) (Treatment 2), or (iii) an active control group that views a short WASH (water, sanitation, and hygiene) video unrelated to agriculture. The informational content is held constant across treatment arms in order to isolate the effect of delivery modality.

Immediately following treatment exposure, all participants complete an endline survey measuring agricultural knowledge, perceived comprehension of the information received, perceived usability of the delivery channel, confidence in applying the information (self-efficacy), and intended behavioural responses.

Participants then undertake an incentivized revealed preference task in which they make a real choice between an agricultural input (fertilizer) and a consumption good (sugar), allowing assessment of short-run investment-related decision-making following exposure to advisory information.

In a subset of villages, participants also complete a discrete choice experiment (DCE) module at baseline and endline to elicit stated preferences over key attributes of digital agricultural advisory platforms, including delivery format, language, interactivity, contextualization, market information, and cost. The DCE is followed by additional questions capturing attribute attendance and preferred DAT design features.

Random assignment is implemented using computer-assisted survey software, with approximately equal allocation across study arms within each session.

Within each village lab-in-the-field session, eligible participants were randomly assigned at the individual level to one of three experimental arms using a public lottery conducted in the field (1:1:1 allocation; 5 participants per arm).

Primary unit of randomization: individual participant (woman farmer) within each village lab-in-the-field session (1:1:1 assignment; 5 participants per arm per session).

No

Not a cluster-randomized design; treatment assigned at individual level within each village session.

1,050 women farmers (individual participants)

350 women (Control – WASH video), 350 women (Treatment 1 – SMS/Voice advisory), 350 women (Treatment 2 – App-based advisory).

Minimum detectable effect size (MDES): 0.15 standard deviations (SD) for the main outcomes, with N = 1,050 individuals (350 per arm), 80% power and α = 0.05, for comparisons of each treatment arm versus control (and T1 vs T2). Equivalent to a 15% of a standard deviation change in the outcome.

Digital Agricultural Technologies for Women Farmers in Malawi: How Do Delivery Channels, Platform Attributes, and Revealed Priorities Shape Learning and Decision-Making?