Introduction

This project

This is the documentation for the replication package of the paper “Information frictions, overconfidence, and learning: Experimental evidence from a floodplain” (PDF, pre-analysis plan), by Sofia Badini.

Abstract. I use an online experiment to study whether offering information to floodplain residents is sufficient to change their perceived risk exposure and demand for insurance. The participants are offered information on the flood risk profile at their address and on the rules over compensation of flood damages. I find that respondents tend to misperceive their risk category according to publicly available flood maps, but express high levels of confidence in their guesses. When not prompted to engage with the information they are offered, one third of them read nothing. Respondents who are asked to read information on their risk profile tend to stop reading any further and report a lower willingness-to-pay for insurance. However, this effect does not seem to be driven by respondents learning more from the information they are provided with, at least based on how they update their beliefs. Instead, I find suggestive evidence of backlash to information among residents of high risk areas and individuals who initially underestimated their risk category.

Structure of the project

This project uses the econ-project-templates by Hans-Martin von Gaudecker, which in turn use pytask, a workflow management system for reproducible data analysis. Therefore, this project is fully and automatically reproducible conditional on having conda, a modern LaTeX distribution (e.g. MikTex), and the text editor VS Code installed. You will also need to install R. Please see the documentation of econ-project-templates for an overview of the dependencies that you need to install (and for detailed explanations on why the template looks the way it does).

The minimum information that you need to know is the following:

  • The important scripts in this project are stored in the folder src, specifically in the sub-folders experiment_floodplain/data_management, experiment_floodplain/analysis, and experiment_floodplain/final.

  • Upon replication of the project, a new folder named bld will appear in the root folder (where src is). bld contains all the output of this project.

  • bld also contains this website, under the sub-folder documentation. The documentation is also produced in PDF version: it is the file documentation.pdf (under bld/documentation/latex, a copy is in the root folder).

  • The single most important file produced by this replication package is the .pdf paper_results_replication.pdf, which reproduces all the figures and tables in the paper (under bld/latex, a copy will appear in the root folder).

Issues with data sharing

This project uses survey data collected with the platform Qualtrics. The survey respondents were contacted via postcards mailed to their addresses, as the Dutch government maintains a database of publicly available geocoded addresses for the entire country. Each postcard included a unique code, which served both as a password to log-in to the online survey, and as an identifier (survey respondents were displayed address-specific information). I additionally determine whether each address was flooded in July 2021 using a dataset of postcode-level data provided by the government (which is not available to the public).

These two datasets will be shared once this paper is published, and will be delivered as a dataset in .csv format (named original_survey_data.csv and rvo_pc6.csv). The original data collected via Qualtrics will be anonymized and will include only the subset of the original Qualtrics data that I used for the analysis.

Note

The analysis will be automatically performed on the real data if the .csv files are placed in the folder src/experiment_floodplain/data. By default, this replication package uses fake data, based on the original dataset and created with the Python library Synthetic Data Vault (SDV), in place of the real survey data. Moreover, by default the flood status is attributed at random. See also the section Data Management.

Replication

To reproduce, first navigate to the root of the project (where src is). Then, open your terminal emulator and run, line by line:

conda create --name condalock python # create empty environment
conda activate condalock # activate environment
pip install conda-lock # install conda-lock

The lines above install conda-lock in a clean environment. conda-lock is a Python package to generate reproducible conda environments – i.e., directories containing the packages needed to run code. To install the environment for this project, after the lines above you additionally need to run:

conda-lock install --name experiment_floodplain environment.conda-lock.yml # create project environment with conda-lock
conda activate experiment_floodplain # activate project environment

Finally, to execute the project run:

pip install -e . # install the project
pytask # execute the project

If your machine runs Windows11

… you may want to use the file environment_win11.conda-lock.yml instead of environment.conda-lock.yml. The former file uses an older version of pytask, as the newest releases appear to… behave capriciously with Windows11.