writeup.Rmd

---
title: "Final Project: Write-Up"
subtitle: "Data and Programming for Public Policy II - R"
author: "Nolan Villasmil"
date: "2024-12-08"
output: pdf_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Background

Despite public concern over climate change, global greenhouse gas emissions
(GHG) have shown sign of no abatement. These reached a new record in 2023. While
the focus has mostly been on carbon dioxide, all GHG have shown a marked
increase - including methane. Methane is particularly concerning as it is 80
times more powerful than carbon dioxide over a 20-year period.

The energy sector is the main source of GHG emissions produced by humans. In
particular, 25% of anthropogenic methane emissions come from venting, leakage
and flaring in the oil and gas industry. These industry processes consist of 
the release and burning of associated gas during oil processing. While flaring
is to some extent inevitable for safety reasons, gas continues to be wasted
across the world due to limited economic incentives, infrastructure or poor
regulation. In addition to economic losses negative environmental impact, gas 
flaring is especially associated with worsening health outcomes as a result of
poor air quality.

Oil output in the United States has seen considerable growth over the last two
decades. According to the U.S. Energy Information Administration (EIA), 
production rose from 8.7 million barrels per day (mbpd) in 2003 to 
21.9 mbpd in 2023. This has contributed to the positioning of the United States
as the top oil producer in 2023, concentrating nearly 21.5% of global output.
At the same time, the United States has also become one of the top flaring 
nations in the world, together with Russia and Iraq.

# Research question

In light of the recent transformation of the oil and gas industry in the United 
States, there is merit in exploring how its associated levels of gas flaring
have interacted with air quality. This research presents an overview of the
evolution of oil output, gas flaring, and air quality in the United States from
2012 to 2023. Using more granular data, it intends to give an account of the
differential spatial relationship by focusing on smaller administrative 
geographical units such as states and counties.

# Coding approach

This coding project largely relied on three (3) major sources: the Annual Gas
Flared Volume, by the Earth Observation Group (EOG)/Colorado School Mines; the 
Air Quality Index (AQI), by the U.S. Environmental Protection Agency (EPA); and
the Air Quality Life Index (AQLI), by the Energy Policy Institute at the 
University of Chicago (EPIC). 

The Annual Gas Flared Volume data by the EOG comprised information about every
gas flare site detected globally through satellite monitoring, including country
of origin, year, estimated volumes of flared gas (expressed in billion cubic 
meters), longitude and latitude. This initial data was saved in multiple Excel
files which divided the data on upstream and downstream. Given the focus on oil
production, upstream data from the United States were extracted and merged into
a single dataframe. In addition to cleaning, this combined dataframe was merged
with United States geographic information in order to identify the spatial
and administrative distribution of flaring sites across the country.

The Air Quality Index (AQI) by the EPA comprised a time series of index 
measurements for each reporting county in the United States. This included the 
number of days with different levels of air quality, as well as the total 
number of days in the year for which AQI observations were recorded. This .csv
file was cleaned and matched with United States and geographical information. 
After doing this, it was clear that the county-level coverage was low, as only
large statistically metropolitan areas reported to the EPA. While helpful to 
understand county-level dynamics, these data omitted a large number of gas 
flaring counties (especially in Texas, the state with the highest number of
flaring sites, with barely 17% of coverage). State-level data was not
available.

The Air Quality Life Index (AQLI) by EPIC comprised a time series containing
measurements of population-weighted atmospheric particulate pollution 
(particulate mater (PM) 2.5 micrograms per cubic meter or less), which is the 
most serious measure of air pollution, alongside estimations of additional days
of live saved if WHO and U.S. air quality standards were met. This information 
was available for all U.S. states. The dataset was cleaned and merged with 
United States geographic information.

In addition, national crude oil field production data from the U.S. Energy 
Information Administration (EIA) and aggregated gas flaring volumes data from 
the Global Gas Flaring Reduction Partnership were used with descriptive 
purposes to understand the evolution of these indicators in the United States.

Lastly, a press release summarizing a scientific article by Tran et al. (2024) 
was retrieved using web scraping for sentiment analysis. It was not possible to
obtain news from local outlets due to paywall barriers. Similarly, it was not
possible to conduct sentiment analysis for the scientific article due to 
restrictions on the publisher's webpage. Content was tokenized by word and 
sentence, and analysis was produced using the `sentiment` function as well as 
the AFINN, NRC and Bing lexicons.

# Results

Using a dummy variable for the existence of a gas flaring site as the only 
explanatory variable in a linear OLS regression, a positive and statistically 
significant (at the 5% significance level) relationship was found. Having gas 
flaring is associated with an increase of 0.4 micrograms per cubic meter of PM 
2.5. With an R-squared of 0.021, this model has low explanatory power. 

With a similar approach but controlling for year- and state-fixed effects, the
existence of flaring in a state is associated with an increase of 0.15 
micrograms per cubic meter of PM 2.5. This is not statistically significant at
the 10% confidence level.

Two additional regressions using these approaches were run for the number of 
flaring sites in each state. These were negative and not statistically 
significant. While counterintuitive, this is an indication of the underlying
data issues mentioned previously and the need for more granular data for a
better understanding. Accessing county-level data and adding relevant controls
as sources of air pollution is a next research step for further inspection.

Text analysis results indicate mostly a positive sentiment of the press 
release. This is consistent for word-based analysis using AFINN and NRC, as well
as sentence-based analysis. However, using the Bing lexicon indicates that 
sentiment is mostly negative, which would be consistent at face value with the
conclusions of the report. This might be a suggestion of the change in tone of 
the press release, emphasizing the opportunities for collaboration. Examining
a wider sample of documents, particularly news reports from top flaring states
such as Texas, would be another step to be further explored in research about
public perceptions of gas flaring.