thepolicylab/hx-advertisementvalences

Code for "Can Moral Framing Drive Insurance Enrollment in the US?"

Health Exchange Advertisement Valences

This repository reproduces the analyses described in our paper "Can Moral Framing Drive Insurance Enrollment in the US?" In this analysis, we looked at whether “green halo” and “noble edge” effects demonstrated in other markets would carry over to the health insurance marketplace.

Using an online advertising platform (Google), we purchased 5.6 million advertising impressions in English and Spanish, targeting higher-income Americans nationwide during the 2021 open-enrollment period. Consumers saw advertisements from a control group (representing status quo ads highlighting economic self-interest, collected from the field) versus three experimental groups (helping others, helping community, or personal responsibility). We measured whether consumers clicked to “shop now” on the healthcare.gov website (1.01% in English and 1.38% in Spanish at baseline).

This repository represents the code for our analysis of this experiment.

Requirements

Software

To run the code in this repository, you will need Python 3.8+ and Stata 16 pre-installed.

In Python, we use poetry to manage our dependencies. To install this dependency manager, you can run (assumes Mac or Linux; see tool's documentation for other operating systems):

curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python3 -

Once this is installed, you can install all dependencies with

poetry install

Census API

In order to run some of this code, you will also need a Census API Key. Once you have the key, copy the .env.sample file to .env and fill in your API key in the specified location.

Running the analysis

Once the dependencies are setup, then you should be able to run the command:

poetry run snakemake -c 1 all

This will produce output in the build folder. Note that this does not reproduce the regressions that are performed in Stata. For those, you will need to open Stata, point it at the file called data/phase2/exploded_data_by_demographics.csv.gz, and run the do file src/stata/040_ads_analysis.do.

HUD Data

Crosswalks between ZIP codes and Census tracts are made available by the US Department of Housing and Urban Development here. We have utilized the third quarter 2021 files, which crosswalk to the 2010 Census boundaries. For convenience, these files are reproduced in the data/hud folder of this repository.

Din, Alexander and Wilson, Ron, 2020. "Crosswalking ZIP Codes to Census Geographies: Geoprocessing the U.S. Department of Housing & Urban Development’s ZIP Code Crosswalk Files," Cityscape: A Journal of Policy Development and Research, Volume 22, Number 1, https://www.huduser.gov/portal/periodicals/cityscpe/vol22num1/ch12.pdf

Elections data

Election data is available from MEDSL at their GitHub repo and the dataverse. For convenience, these data have been reproduced in the data/medsl folder of this repository.

MIT Election Data and Science Lab, 2018, "County Presidential Election Returns 2000-2020", https://doi.org/10.7910/DVN/VOQCHQ, Harvard Dataverse, V9, UNF:6:qSwUYo7FKxI6vd/3Xev2Ng==

Code layout

All of our code lives in the src directory. It contains three folders described below:

hxpr: Python utilities

These are a few python utilities that are useful throughout the project. No analysis occurs in these folders.

notebooks: Data wrangling in Python

This folder contains all of the python notebooks used for data wrangling. Each notebook handles a different type of data we interacted with.

stata: Analyses in Stata

Our principal statistical analyses were performed in Stata. These may be found in this folder.