Updating Climate Beliefs: The Interaction Between Traditional and Modern Forecasting

The intervention consists of an informational video on the product attributes of a satellite-based index insurance product, explaining how
it offers protection in case of drought by providing a payout to prevent livestock losses that is conditional on satellite readings of vegetation quality on the ground. The video describes how the insurance payout scheme
is based on weather information coming from satellite data based on pictures capturing vegetation on the ground. Scores of percentiles are then used to compare the
forage quality with that of historical forage quality in previous seasons. Color coding that is similar to the color coding from the Kenyan National Drought Management Authority (NDMA) is then used to indicate the
quality of the forage as established by the satellite, in relation to forage quality in previous seasons.

2024-01-01

2024-03-31

• Reliance on traditional versus modern forecasting methods
• Beliefs about upcoming weather conditions
• Insurance uptake and adaptation strategies as proxies for risk management decisions

• Community use of forecasting methods
• Perceived accuracy of modern weather information
• Perceived reliability of traditional and modern forecasting methods

The study uses a clustered randomized controlled trial across 393 communities which are either assigned to the control group, where there is no video intervention, or the treatment group with video intervention. The first step in choosing study sites was done at the district level, referred to as woredas in Ethiopia
and sub-counties in Kenya. Sites were selected based on four main criteria: (1) whether households had
actively registered for the financial product during the January/February 2023/2024 sales season, (2) a
broad geographical distribution across the target areas, (3) safe conditions for the data collection teams, and
(4) enough communities within each district to represent all experimental groups. Based on these criteria, 26
districts were selected in total: 10 sub-counties across 6 counties in Kenya, and 14 woredas across 4 regions
in Ethiopia.
From within these 26 districts, 451 non-urban communities were randomly chosen from those that met
the criteria. To ensure relevance for studying community-level decision-making under climate and livelihood
uncertainty, urban areas were excluded. These communities correspond to sublocations in Kenya (226
selected) and kebeles in Ethiopia (225 selected). However, communities in Ethiopia’s Somali region were
removed from the sample due to operational difficulties, reducing the final sample to 393 communities—167
in Ethiopia and 226 in Kenya. Before collecting the baseline survey data, a list of households and identified cooperatives (in Ethiopia) and
pastoral groups (in Kenya) within each selected community was created. They focused on groups with at
least 15 members that had not yet received any information about the financial product. From within these
groups, households were listed if they were involved in managing livestock, and if the household head had a
spouse. As all participating households were thus involved in livestock herding or management, the sample
provides a relevant setting to examine how pastoralists interpret and act on forecast information related to
drought risk. On the day of the survey, group leaders helped choose 8 households from this list. In
total, 60% of the 393 study communities were randomly assigned to the treatment arm and 40% were assigned to the control arm. The randomization was stratified by districts and units areas of
insurance.

Not available

Randomization was done in office by a computer

We use a cluster randomized controlled trial across 393 communities where 60% is randomly assigned to the treatment group and 40% to the control group.

Yes

393 communities

8 households per community, so 3144 household heads

168 communities in the control group and 225 in the treatment group