📄 other

Effectiveness of Evacuation Notices on Household Evacuation Behavior During Floods in Kerala, India

This study of the 2018 Kerala floods demonstrates that evacuation notices significantly increase household evacuation rates, with targeted in-person warnings proving particularly effective for vulnerable populations in high-risk areas.

Original authors: Sanchita Gera, Mihir Bhatt, Vishal Pathak, Anishkumar M.S., Nishant Kishore, Satchit Balsari

Published 2026-07-27

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sanchita Gera, Mihir Bhatt, Vishal Pathak, Anishkumar M.S., Nishant Kishore, Satchit Balsari

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of disaster science as a giant, high-stakes game of "What If?" Scientists in this field are like detectives trying to solve a mystery that hasn't happened yet, or one that just finished. They want to know: when the sky turns dark and the water starts rising, what actually makes a person grab their backpack and run? Is it fear? Is it the size of their house? Or is it simply who they trust to tell them to move? For decades, researchers have known that people weigh their own safety against their ability to leave, but the exact recipe for a successful evacuation has been a bit of a blur, especially in places like South Asia where the rules might be different from the West. This paper steps into that blur to ask a very specific question: does the way you tell someone to leave matter more than just telling them to leave at all?

The story begins in Kerala, India, during the massive floods of 2018. The state was hit hard, with millions displaced and homes damaged. The government and local groups scrambled to get people to safety, but they used two very different methods to do it. Some people got a "broad" shout—like a message on the radio, TV, or a megaphone from a distance. Others got a "targeted" nudge—someone actually walking up to their door, knocking, and looking them in the eye to say, "You need to go now." The researchers wanted to know which method worked better. They didn't just guess; they went out and asked 4,719 households what happened to them. It's like checking the scorecard after a big game to see which play actually moved the ball down the field.

Here is what the data revealed, and it's a bit surprising. First, getting any notice was a game-changer. If a household heard a warning, they were 4.11 times more likely to evacuate than those who heard nothing. That's a huge jump, like going from a 20% chance of winning to an 80% chance just by hearing the starting whistle. But the real magic happened with the door-knockers. When a household received a targeted, in-person notice on top of the broad broadcast, their odds of evacuating jumped by an additional 2.17 times compared to those who only heard the radio or TV.

Think of it like this: The broad notice is like a loudspeaker at a stadium announcing a fire drill. Everyone hears it, and many people start to move. But the in-person notice is like a coach running onto the field, grabbing a player's shoulder, and saying, "Hey, you specifically, the water is rising right here, come with me." That personal touch, delivered by trusted neighbors or local officials, seemed to be the final push that convinced people to leave. The study found that these notices weren't just random; they were smartly directed at the most vulnerable people—those with lower incomes, health issues, or homes in the deepest flood zones.

Interestingly, the paper rules out some things we might expect to be the main drivers. You might think that having a big family, an elderly grandparent, or a pet would make it harder to leave, or that having a house with an upper floor would make it easier to stay safe. But in this study, those factors didn't make a significant difference in who stayed and who went. The deciding factor wasn't the size of the family or the type of house; it was the message and who delivered it.

The researchers are careful to say that while these results are strong, they come from a specific time and place. They used a special statistical method called "propensity score weighting" to make sure they were comparing apples to apples, ensuring that the people who got the door-knock notices weren't just naturally more likely to evacuate for some other reason. The results suggest that in a place like Kerala, where people trust their local government and community groups, a personal visit is a superpower for saving lives. It's a reminder that even in an age where we have smartphones and instant alerts, sometimes the most effective technology is still a human being walking up to your door, looking you in the eye, and saying, "Let's go."

Technical Summary: Effectiveness of Evacuation Notices on Household Evacuation Behavior During Floods in Kerala, India

Problem Statement
Floods in India are increasing in frequency and intensity, causing significant loss of life and infrastructure damage. While extensive literature exists on evacuation behaviors in the Global North (US and Europe), there is a scarcity of empirical studies from South Asia regarding how specific warning mechanisms influence household-level decision-making. This study addresses the gap in understanding the causal impact of evacuation notices—specifically comparing broad public broadcasts against targeted in-person notifications—on evacuation behavior during the severe 2018 monsoon floods in Kerala. The research seeks to determine if and how these notices drive protective action, controlling for household vulnerability and risk perception.

Methodology
The study utilized a large-scale cross-sectional survey conducted between May and June 2019, covering two severely affected districts: Ernakulam and Idukki.

Sampling: A two-stage cluster randomized sampling design was employed. Researchers first sampled 170 wards based on a "Flood Inundation Index" (stratified by low, medium, and high inundation) and then randomly selected households within those wards using government-provided line lists. The final sample consisted of 4,719 households.
Data Collection: The survey was implemented in partnership with the Kudumbashree Mission (a state-sponsored women-led poverty eradication mission) and the All India Disaster Mitigation Institute (AIDMI). Enumerators used Android tablets with Commcare software to collect quantitative data on demographics, flood exposure, evacuation decisions, and notification receipt.
Treatment and Outcome Definitions:
- Outcome: A binary variable indicating whether a household evacuated ( $Y=1$ ) or stayed ( $Y=0$ ).
- Treatment 1 (Aim 1): Receipt of any evacuation notice (broad or in-person) vs. no notice.
- Treatment 2 (Aim 2): Receipt of an additional in-person notice vs. broad notice only (among those who received any notice).
Statistical Analysis: To address selection bias in treatment allocation (i.e., the fact that officials likely targeted specific vulnerable households for notices), the authors employed Propensity Score Weighting.
- Propensity scores were estimated using mixed-effects logistic regression models with ward-specific random intercepts.
- Due to the presence of extreme tails in the propensity score distribution, Overlap Weights (OW) were selected over standard Inverse Probability Weighting (IPW) to minimize variance and ensure unbiased treatment effect estimates within the overlap population.
- Causal effects were estimated using Generalized Linear Mixed-Effect Models (GLMM) with logit link functions, adjusting for covariates including income, housing structure, age demographics, prior risk awareness, and flood inundation levels.

Key Results

Impact of Any Notice: Households that received any form of evacuation notice were significantly more likely to evacuate than those who received none. The overlap-weighted analysis yielded a causal odds ratio of 4.11 (95% CI [2.64, 6.38]). This indicates that receiving a notice increased the odds of evacuation by a factor of 4.11, controlling for observed vulnerabilities.
Impact of In-Person Notices: Among households that received any notice, those who received an additional in-person notice (delivered via door-knocking by officials or civil society) were 2.17 times more likely to evacuate compared to those who received only broad broadcasts (Odds Ratio: 2.166; 95% CI [1.428, 3.286]).
Targeting of Notices: The analysis confirmed that notices were effectively directed toward high-risk areas and vulnerable populations (defined by low income, health status, and prior flood exposure).
Non-Significant Predictors: Contrary to some Global North literature, standard demographic predictors such as household size, presence of elderly or children, and the presence of an upper floor in the home did not significantly differentiate evacuees from non-evacuees in this context.

Key Contributions

Causal Evidence in the Global South: This is one of the first studies of its scale and depth to estimate the causal effect of disaster notifications on evacuation behavior in a South Asian context, moving beyond retrospective or hypothetical scenarios.
Methodological Rigor: The study demonstrates the utility of large-scale observational data combined with advanced causal inference techniques (Propensity Score Overlap Weighting) to derive actionable policy insights where Randomized Controlled Trials (RCTs) are unethical or unfeasible.
Validation of Hybrid Communication: The findings empirically validate the "Rebuild Kerala Initiative's" approach of combining digital/broad media with traditional, trusted human intermediaries (Kudumbashree volunteers and local officials).

Significance and Claims
The authors claim that their study provides conclusive evidence that targeted in-person notifications are highly effective and should be examined for broader application across South Asia. The paper emphasizes that in the context of Kerala—characterized by high literacy, gender equality, and a history of trust in state institutions—personalized messaging from trusted sources significantly drives evacuation behavior.

The study highlights that while digital communication is expanding (evidenced by high mobile penetration), redundancy remains the cornerstone of disaster preparedness. The authors argue that in-person outreach by trusted intermediaries should not be abandoned in favor of purely technological alternatives, particularly given the risk of power outages and infrastructure failure during disasters. The findings suggest that locally tailored disaster response efforts, which leverage existing community networks and trust, are essential for effective risk communication.

Limitations Noted by Authors
The authors modestly acknowledge several limitations:

Sampling Constraints: The sampling frame relied on available household line lists, potentially missing unregistered, migrant, or hard-to-reach populations. The urban center of Kochi was underrepresented due to data unavailability and weather conditions during the survey.
Measurement Gaps: The survey did not explicitly capture messaging apps (e.g., WhatsApp) as a distinct communication source, despite their likely role.
Outcome Granularity: The outcome was a discrete measure (evacuated/not evacuated) rather than a time-to-evacuation metric, which might obscure nuances in response timing.
Causal Assumptions: As with all causal inference studies, the results rely on the assumptions of consistency, positivity, and exchangeability (no unmeasured confounders).

Technical Summary: Effectiveness of Evacuation Notices on Household Evacuation Behavior During Floods in Kerala, India

More like this