Access, use, and welfare effect of crowdsourced agricultural information: Evidence from Kenya

The study design is a cluster-randomized controlled trial. We will randomize 200 villages into one treatment arm and one control. We will work with 24 “contributors” to sign-up the treatment arm for crowdsourced KAZNET information about the performance of livestock markets, rangeland conditions, livestock mortality and the associated causes, and milk production. The target sample includes pastoralists groups within the selected villages. We will use a minimum distance rule for the selection of villages; for example, any two villages in the sample must be a minimum of 5 km apart. While we cannot rule out that people who live in different villages may still talk to each other, imposing a minimum distance reduces the likelihood that people will belong to more than one group in our sample or observe the behavior or practices of those in another village. All pastoralist groups in a village belong to either the treatment arm (T1) or control (T0). In each pastoralist group, five individuals will be recruited into the study. In the treatment arm, the KAZNET crowdsourcing approach will be explained. Pastoralists will be informed about the process of data collection and the information that is available for dissemination. Individuals with smartphones will be asked if they would be willing to sign-up to receive information on their smartphones. Those expressing interest will be registered on the KAZNET app for information access. Pastoralist groups in the control villages will not receive any information about KAZNET and will not be provided the opportunity to sign-up for KAZNET information.

2023-11-01

2024-09-30

Our main outcome variables measure (1) household dietary diversity score, (2) months of food self-sufficiency, and (3) reduced coping strategy index.

Household dietary diversity score (HDDS) will be measured using 24 hour recall data on the consumption of different food groups; months of food self sufficiency is a summation of all months in the last 12 months when the household had enough food to eat; reduced coping strategy index is a score used to compare the hardship faced by households due to a shortage of food. The index measures the frequency and severity of the food consumption behaviors the households had to engage in due to food shortage in the 7 days prior to the survey.

KAZNET was mentioned as a source of information on livestock markets and forage conditions, perceived usefulness of KAZNET information, trust in KAZNET data, number of people with whom the respondent discusses about markets and forage conditions, changing selling markets, selling livestock in markets with better prices, and purchasing and storing fodder, price knowledge, we look at the ability or willingness to report a price by livestock type and market, inputs expenditure, livestock selling price, livestock mortality, and income.

First, exposure to KAZNET crowdsourced livestock information can increase trust and influence the use of such information. We check whether KAZNET was mentioned as a source of information on livestock markets and forage conditions. We further test the effect of our intervention on the perceived usefulness of KAZNET information. Respondents will be asked to indicate on a scale of 0 (not useful at all) to 10 (very useful) the extent to which KAZNET information was useful to the livestock activities of their households. We also assess trust in KAZNET data. Respondents will be asked to indicate on a scale of 0 (no trust at all) to 10 (complete trust) the extent to which they trust that KAZNET information is accurate for the livestock activities of their households.

Second, we hypothesize that KAZNET information can influence behavior. Specifically, we test the effect of our intervention on the number of people with whom the respondent discusses about markets and forage conditions, the likelihood of changing selling markets, the likelihood of selling livestock in markets with better prices, and the likelihood of purchasing and storing fodder.

Third, KAZNET information can increase the knowledge of pastoralists. We focus on price knowledge and test the effect of exposure to KAZNET on the absolute value of the difference between respondent’s reported per animal price from the price sent the week the survey was carried out, in the respective unit areas of insurance. In addition, we look at the ability or willingness to report a price by livestock type and market. Specifically, we ask respondents, “If you were to sell your [livestock type] today, what would be the selling price?”. We hypothesize that pastoralists exposed to KAZNET would negotiate better for their livestock compared to their counterparts in the control group.

Fourth, access to KAZNET information can crowd-in investment in productivity-enhancing inputs and encourage commercialization of livestock production. Further, through its influence on herd management, exposure to KAZNET is expected to reduce livestock mortality. Therefore, we test the effect of our intervention on inputs expenditure, livestock selling price, livestock mortality, and income.

The study design is a cluster-randomized controlled trial. We will randomize 200 villages into one treatment arm and one control. We will work with 24 “contributors” to sign-up the treatment arm for crowdsourced KAZNET information about the performance of livestock markets, rangeland conditions, livestock mortality and the associated causes, and milk production. The target sample includes pastoralists groups within the selected villages. We will use a minimum distance rule for the selection of villages; for example, any two villages in the sample must be a minimum of 5 km apart. While we cannot rule out that people who live in different villages may still talk to each other, imposing a minimum distance reduces the likelihood that people will belong to more than one group in our sample or observe the behaviour or practices of those in another village. All pastoralist groups in a village belong to either the treatment arm (T1) or control (T0). In each pastoralist group, five individuals will be recruited into the study. In the treatment arm, the KAZNET crowdsourcing approach will be explained. Pastoralists will be informed about the process of data collection and the information that is available for dissemination. Individuals with smartphones will be asked if they would be willing to sign-up to receive information on their smartphones. Those expressing interest will be registered on the KAZNET app for information access. Pastoralist groups in the control villages will not receive any information about KAZNET and will not be provided the opportunity to sign-up for KAZNET information.

Not available

Randomization will be done in the office by a computer using a replicable stata code.

The unit of randomization are villages.

Yes

200 villages

2,000 pastoralists.

100 villages control and another 100 villages treatment (i.e., receive KAZNET crowdsourced livestock information)

We assume a minimum detectable effect of 15% and an intra-cluster correlation coefficient equal to 0.05. To achieve at least 80% statistical power and adjusting for 10% attrition, we estimate that a sample size of 2,000 pastoralists would be required.