x

We are happy to announce that all trial registrations will now be issued DOIs (digital object identifiers). For more information, see here.
Experimental Evaluation of a Flood Forecasting Tool
Last registered on October 29, 2019

Pre-Trial

Trial Information
General Information
Title
Experimental Evaluation of a Flood Forecasting Tool
RCT ID
AEARCTR-0004947
Initial registration date
October 28, 2019
Last updated
October 29, 2019 9:59 AM EDT
Location(s)

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Affiliation
Yale University
Other Primary Investigator(s)
PI Affiliation
Yale University
Additional Trial Information
Status
On going
Start date
2019-08-01
End date
2020-07-31
Secondary IDs
Abstract
Floods caused over 50,000 deaths in the last decade—including immediate fatalities and longer-run fatalities due to disease outbreak. 97.5% of this mortality occurred in lower-income nations. Early flood warning systems can play an important role in lowering immediate fatalities, waterborne disease spread, and mental illness in lower-income, flood-prone countries like India and Bangladesh and facilitate climate change adaptation. Currently, we lack causal assessments of early warning systems because such evaluations require pairing high-frequency geolocation data to measure evacuation behavior or movement in response to disaster alerts with extensive household surveys in disrupted communities to capture short- and long-run effects on socioeconomic indicators. In collaboration with the engineering team at Google’s flood forecasting initiative, we are conducting a multi-pronged pilot evaluation of their flood alert system in the Ganges-Brahmaputra river basin with field work in Winter 2019. The pilot study has two objectives: (a) to examine the causal effect of flood alerts on evacuation behavior, and consequently, physical and mental morbidity, and (b) to evaluate if existing local government structures can be leveraged to increase the reach and impact of disaster alerts.
External Link(s)
Registration Citation
Citation
Jagnani, Maulik and Rohini Pande. 2019. "Experimental Evaluation of a Flood Forecasting Tool." AEA RCT Registry. October 29. https://doi.org/10.1257/rct.4947-1.0.
Experimental Details
Interventions
Intervention(s)
A randomly assigned group of panchayat mukhiyas (or community leaders) across six districts in the Bihar were informed about flood alerts, once an alert was triggered by Google’s flood forecasting system, via WhatsApp message and SMS. The treatment group includes panchayats (or local governing bodies) where panchayat mukhiyas were informed about the alert, via WhatsApp message and SMS, and android smartphone users with location services enabled received an alert if and when an alert is triggered. The control group includes panchayats where android smartphone users with location services enabled received an alert when an alert was triggered but panchayat mukhiyas were not informed about the alert, via WhatsApp message or SMS.

Intervention Start Date
2019-08-01
Intervention End Date
2019-10-31
Primary Outcomes
Primary Outcomes (end points)
Ex-ante and ex-post adaptive behavior, ex-post recovery, post-flood socioeconomic indicators, and diffusion and source of flood alerts.
Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
A randomly assigned group of panchayat mukhiyas (or community leaders) across six districts in the Bihar were informed about flood alerts, once an alert was triggered by Google’s flood forecasting system, via WhatsApp message and SMS. The treatment group includes panchayats (or local governing bodies) where panchayat mukhiyas were informed about the alert, via WhatsApp message and SMS, and android smartphone users with location services enabled received an alert if and when an alert is triggered. The control group includes panchayats where android smartphone users with location services enabled received an alert when an alert was triggered but panchayat mukhiyas were not informed about the alert, via WhatsApp message or SMS.

Our sample includes 181 panchayats/476 villages: 91 panchayats/232 villages in the control group, 90 panchayats/244 villages in the treatment group. According to satellite images, of these, 139 villages were flooded. Google’s flood forecasting system sent flood alerts to location-services-enabled android smartphones in 129 of these villages with a recall rate of over 90%.

In December’19-January’20, we will administer a comprehensive survey in Bihar. We will survey 1500 households in flooded treatment (65) and control (74) villages in Bihar. The survey instrument will collect information on ex-ante and ex-post adaptive behavior, ex-post recovery, post-flood socioeconomic indicators, and diffusion and source of flood alerts. In our analysis, we will compare diffusion of flood alerts as well as the networks through which said alerts were diffused between treatment and control villages. In addition, we will compare ex-ante and ex-post adaptive responses, ex-post recovery, and post-flood physical and mental morbidity between treatment and control villages.
Experimental Design Details
Not available
Randomization Method
Randomization done in office by a computer.
Randomization Unit
Unit of randomization: panchayat; panchayats are local governing bodies that typically include 2 to 5 villages.
Was the treatment clustered?
Yes
Experiment Characteristics
Sample size: planned number of clusters
181 Panchayats/476 villages
Sample size: planned number of observations
1500 households
Sample size (or number of clusters) by treatment arms
91 panchayats/232 villages in the control group, 90 panchayats/244 villages in the treatment group
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
IRB
INSTITUTIONAL REVIEW BOARDS (IRBs)
IRB Name
IRB Approval Date
IRB Approval Number