The impacts of climate change forecasts on expectations, labor market and migration plans, and policy demand in rural Bangladesh

Last registered on August 25, 2022


Trial Information

General Information

The impacts of climate change forecasts on expectations, labor market and migration plans, and policy demand in rural Bangladesh
Initial registration date
August 23, 2022

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
August 25, 2022, 2:08 PM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.



Primary Investigator

University of Warwick

Other Primary Investigator(s)

PI Affiliation
London School of Economics
PI Affiliation
London School of Economics
PI Affiliation
London School of Economics

Additional Trial Information

On going
Start date
End date
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
The impacts of climate change are forecasted to vary widely across space and across occupations. Hence, occupation and location change are believed to be promising strategies to foster climate-change adaptation in vulnerable communities. However, it remains unclear whether individuals are informed about the looming changes to their climate, so that they can make the best occupation and location decisions. In this project, we collect unique data on the climate change beliefs of a large sample of individuals in rural Bangladesh. We then experimentally provide forecasts about local climate change and investigate how the provision of these forecasts changes expectations, occupation and migration plans, and policy demand.
External Link(s)

Registration Citation

Bryan, Gharad et al. 2022. "The impacts of climate change forecasts on expectations, labor market and migration plans, and policy demand in rural Bangladesh." AEA RCT Registry. August 25.
Experimental Details


In this project, we provide climate-change information to individuals in rural Bangladesh. We leverage recent, high-quality grid-level forecasts from the World Climate Research Program’s Coupled Model Intercomparison Project (CMIP6) to design five different information packages. The packages give quantitative information on:
1. the likely and worse-case scenario changes to the local climate;
2. the local damage function (the local relationship between warming and yields, and warming and mortality);
3. The combination of (1) and (2): the likely and worst-case scenario changes to the local climate and the local damage function;
4. The combination of (1) and (2) + information about expected climate change in Dhaka.
5. The combination of (1) and (2) without information on the worst-case scenario.

All forecasts provided refer to the year 2050.

Individuals are randomly assigned to one of these five information packages, or to a control group that does not receive any information.
Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
1. Expectations
2. Intentions related to oneself
3. Intentions related to one's children
4. Policy demand

We will study reduced form impacts on these outcomes.

We will also leverage the exogenous variation in expectations generated by the experiment to identify the causal link between expectations (an index of expected impacts) and intentions.

For the comparison between treatment group 3 (information on the likely changes to the climate and the damage function) and control, we will also be able to study the persistence of treatment effects over a period of about one week.

We will explore treatment effect heterogeneity with respect to the following dimensions:

1. Severity of forecasted local climate change
2. Pre-treatment expectations
3. Age
4. Gender
5. Income
Primary Outcomes (explanation)

1. Expectations

Changes to the local climate
i. Expected changes in the number of hot days by 2050 (quantitative question).
ii. Expected frequency of large floods by 2050 (quantitative question).

The damage function
iii. Changes in crop yields due to one degree of warming (quantitative question).
iv. Changes in mortality due to one degree of warming (quantitative question).

Expected impacts
v. Expected change in earnings without adaptation (quantitative question).
vi. Expected change in earnings with adaptation (quantitative question).
vii. Expected cost of adaptation (quantitative question).
viii. Expected change in amenity value of locality (quantitative question).

Changes with respect to Dhaka
ix. Expected change in the number of hot days by 2050 relative to Dhaka (quantitative questions).
x. Expected change in mortality due to one degree of warming relative to Dhaka (quantitative questions).
xi. Expected change in earnings without adaptation relative to a city worker (quantitative questions).

2. Intentions relative to oneself
i. Likelihood of changing occupation
ii. Likelihood of migration
iii. Likelihood of investing in adaptation

3. Intentions related to one's children
i. Likelihood child will work in different occupation than the parent
ii. Likelihood child will migrate
iii. Likelihood investment in male child education will increase
iv. Likelihood investment in female child education will increase

4. Policy demand
i. a dummy variable for preferring an adaptation-through-migration intervention to a local adaptation intervention.
ii. a dummy variable for preferring a human capital intervention targeted to children to a local adaptation intervention
iii. willingness to pay for a climate-change information intervention.

We will study the impacts of each intervention on each one of these variables. For interventions (1)-(3) the focus of the analysis will the treatment-control comparison. For interventions (4) and (5) the focus will be the comparison against treatment (3).

To address concerns related to multiple-hypothesis testing, we will calculate indices for each category and sub-category above. We will then report q-values using the p-values on the statistical tests obtained for each summary index. Note that, for the expectation variables, we will construct a different index for each intervention, reflecting the information that was given in that particular intervention. For example, we will exclude expectations about Dhaka from the index calculated for each intervention that does not provide information about Dhaka. While those spill-over effects are of interest and will be reported in the paper, the primary aim here is to test whether each intervention affected the expectations it targeted.

In addition to the indices we will separately estimate the effect of the interventions on expectation 1.a (i.e. expected changes in the number of hot days by 2050). From our experience with the pilot, it has become clear that this question is best understood by the respondents

We will explore treatment effect heterogeneity with respect to the following dimensions:

1. Severity of expected local climate change: an index of village-level dummies for having above median value of all climate variables reported in the information interventions, weighted by sensitivity weights (discussed below).

2. Pre-treatment expectations:
A county-level index obtained using only control observations capturing the average misperception of the likely physical changes and damage function, weighted by sensitivity weights (discussed below).
3. Age: Dummy for whether the age of the respondent is above the median age of the sample
4. Gender: Dummy for being female
5: Income: a dummy for having income above the median

We will focus the analysis of treatment effect heterogeneity on the four main indices listed above. We will also report some exploratory regressions on individual variables if appropriate.

“Sensitivity weights” will let us re-weight different dimensions of climate change according to how threatening individuals perceive them to be. We will calculate these village-level weights by regressing a variable capturing whether individuals think climate change will result in negative damages for the local economy and the lives of local people on climate-change and damage-function expectations. The coefficients from this regression will be used as weights. The larger the coefficient, the larger the weight a given component will be given in the index. For these regressions, we will only use control observations.

Structural Model
We plan to use the exogeneous variation of the experiment to estimate a structural model of migration and occupation decisions. This will enable us to quantify optimal migration and occupation choices and how those are constrained by inadequate information.

Secondary Outcomes

Secondary Outcomes (end points)

1. Confidence in climate beliefs
2. Climate anxiety
3. Experimenter demand

We will also look at the following, secondary dimensions of heterogeneity:
1. Pre-treatment expectations (alternative measure)
2. Trust in science
3. Whether the respondent is employed in agriculture
Secondary Outcomes (explanation)
Confidence in climate beliefs: an index of two questions where we ask respondent how confident they are about their climate change beliefs.
Climate anxiety: a variable capturing whether the respondent felt anxious during the interview.
Experimenter demand:
a. the extent to which respondent felt a pressure to reply as the experimenter wanted them to reply
b. the extent to which respondent felt confident about the objective of the experimenter
Pre-treatment expectation (alternative measure): an individual-level dummy for thinking that climate change is likely to have a severe negative impact on the economy and lives of people in their community.
Trust in Science: a dummy for whether the respondent feels confident in scientist’ ability to predict the climate.

Experimental Design

Experimental Design
We randomize individuals into one of six conditions. These conditions include:
- The 5 interventions discussed above;
- A control condition, where respondents do not receive any information.

Individuals are first asked some demographic questions and some general climate change belief questions. Then, treated individuals are given one of the information interventions. Finally, individuals are asked questions about expectations, intentions and policy demand. We also randomize whether the questions about expectations follow or come before the questions on intentions and policy demand.

We also randomly assign a small share of the control group to an experimenter demand condition, where respondents are explicitly given information on the hypotheses of the experimenter. We will drop these individuals from the main analysis, and will only use them to measure experimenter demand effects.
Experimental Design Details
Randomization Method
Randomization Unit
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
Sample size: planned number of observations
Sample size (or number of clusters) by treatment arms
1,000 individuals will be assigned to treatment arms (i), (iv), (v) and to the control group.
500 individuals will be assigned to treatment arms (ii) and (iii).
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
In a standard treatment-control comparison, we have 98 percent power to detect, at the five percent significance level, an impact of 0.2 standard deviations on a standardized index with mean zero and standard deviation one.

Institutional Review Boards (IRBs)

IRB Name
LSE Research Ethics Committee
IRB Approval Date
IRB Approval Number


Post Trial Information

Study Withdrawal

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information


Is the intervention completed?
Data Collection Complete
Data Publication

Data Publication

Is public data available?

Program Files

Program Files
Reports, Papers & Other Materials

Relevant Paper(s)

Reports & Other Materials