Closing the Gender Gap in Digital Agricultural Extension: Evidence from Ethiopia's 8028 Farmer Hotline

This study will include three experiments:
Experiment 1: Improving Access through Phone Sharing (men phone owners; randomized encouragement with or without incentives)

Intervention: Male phone owners are asked to nominate a spouse/another female farmer in their household (without a phone) and are encouraged to let her use their phone to call 8028. Randomized treatment status determines whether the request is accompanied by encouragement and/or a financial incentive.

Experiment 3: Gender-Tailored Messaging (women phone owners)

Intervention: The Hotline’s push-message feature sends two pre-recorded calls. Each call starts with a short narrative (story) followed by a sample of hotline content. We independently randomize whether the narrative is female-tailored (story about a female farmer successfully using the hotline) versus male-tailored (story about a male farmer). Two content topics are tested: maize (widely cultivated) and head cabbage (commonly used by women per LSMS Ethiopia 2018–2019). Each participant in a treatment arm receives two push calls; the order of the two crop messages is randomized. The pick-up rate for the second call (after exposure to the first call) will serve as an outcome. Randomization to the different experimental arms will be carried out after the screening survey and before the distribution of messages.

2026-03-27

2026-07-31

Experiment 1:
Primary outcomes: (1) callback rates during the two-week incentive period and during the next 4 months (this includes both call attempts and successful calls to the Hotline, where at least one informational message was accessed by the caller); (2) time spent on the hotline service during the two-week incentive period and during the next 4 months.

Experiment 2:
Primary outcomes: (1) likelihood of picking up the push message for the second push call, (2) likelihood of listening the push message, (3) length of the push call listened to, (4) no. of callers who listen to the whole message being delivered, (5) callback rates after the first push message, the second push message and during the next 4 months (both call attempts and successful calls) , (6) time spent on the hotline service over the study period (4 months).

Experiment 1:

Secondary outcome: (1) percentage of men who agree to share their phones with nominated female farmers (ex-ante). (2) no. of callbacks during the two-week incentive period and during the next 4 months (this includes both call attempts and successful calls to the Hotline, where at least one informational message was accessed by the caller)
Verification calls will be carried out for a random subset of callers from the experiment sample (around 20%) to verify if it was the nominated female farmer who made the call.

Experiment 2:

Secondary outcomes: (1) call back rates to the hotline by the female farmer over the study period (4 months) (2) we will also study the impact of the second experiment on farmers’ knowledge of agricultural outcomes and (self-reported) adoption of improved practices through a follow-up phone survey, subject to funding availability.

Experiment 1:
Male phone owners are asked to nominate a spouse or another female farmer in their household who does not own a phone, and are encouraged to let her use their phone to call 8028. Whether this request is accompanied by a financial incentive is determined by randomized assignment.
Arms: (1) Control: No financial incentive; no encouragement to inform the nominated female farmer; men still asked to nominate a female farmer in the household. (2) Information-only: Encouragement to inform the nominated female farmer to use the phone (no incentive). (3) Incentive 1–4: Encouragement plus a small financial incentive, with the incentive level randomly varied; payment conditional on the nominated female farmer calling within two weeks. Four incentive amounts form four distinct treatment arms.

Experiment 2:
The eligible female farmer is randomly assigned to one of three conditions: a female-tailored narrative (a story about a female farmer successfully using the hotline), a male-tailored narrative (a story about a male farmer), or a pure control group. Each participant in a treatment arm receives two push calls; the order of the two crop messages is randomized. The pick-up rate for the second call (after exposure to the first call) will serve as an outcome. Randomization to the different experimental arms will be carried out after the screening survey and before the distribution of messages.
Arms: (1) Control: No push message. (2) Female-tailored story (3) Male-tailored story.

Not available

The randomization will be conducted in office using Stata.

We will conduct an individual level randomization of farmers.

No

This will be an individual level randomization.

Experiment 1: 2000 male callers identified through a screening survey Experiment 2: 2000 female callers identified through a screening survey Study 3: 2000 female callers identified through the screening survey

Experiment 1: Control = 400; Information only = 400; Each of the 4 incentive buckets = 300 each
Experiment 2: Control = 666; Female tailored messages = 667; Male tailored messages = 667

Experiment 1: We conduct ex ante power calculations for the three pairwise treatment comparisons (Incentives vs. Information, Incentives vs. Control, and Information vs. Control) under a two-sided test with 5% significance and 80% power. The sample consists of 1,200 observations in the pooled Incentives group and 400 observations each in the Information and Control groups. For the primary binary outcome - whether a respondent calls back (control mean = 0.148) - we estimate minimum detectable effects (MDEs) of 5.7 percentage points (38.8% of the control mean) for comparisons involving the Incentives group (Incentives vs. Information and Incentives vs. Control), and 7.0 percentage points (47.5% of the control mean) for the Information vs. Control comparison. For call duration (measured in seconds, unconditional on callback; control mean = 22.22, SD = 76.53), the corresponding MDEs are 12.4 seconds (0.16 SD) and 15.2 seconds (0.20 SD), respectively. Experiment 2: We conduct ex ante power calculations for three pairwise comparisons: Female-tailored message vs. Control, Male-tailored message vs. Control, and Female- vs. Male-tailored messages. Assuming a total sample of 2,000 observations equally allocated across the three arms (approximately 667 observations per arm), a two-sided test with 5% significance and 80% power, the study is powered to detect a minimum effect of 7.7 percentage points for the binary outcome of picking up a push call (control mean = 0.48), corresponding to approximately 16.0% of the control mean. For the outcome measuring the length of the push call message heard (not conditional on picking up; control mean = 63.46 seconds, SD = 86.35 seconds), the minimum detectable effect is 13.3 seconds, equivalent to 0.15 standard deviations. These detectable effects apply symmetrically across all three pairwise comparisons under equal allocation. Given the large dispersion of the duration outcome relative to its mean, the distribution is likely right-skewed, with a mass at short durations and a long upper tail; we therefore interpret power for this outcome as conservative and will complement analyses with transformations or robustness checks (e.g., excluding very short call durations).