Individual Submission Summary
Share...

Direct link:

Machine Learning for Predicting Congressional Environmental Voting Records

Sat, October 2, 6:00 to 7:30am PDT (6:00 to 7:30am PDT), TBA

Abstract

Fighting climate change requires green policies at every level of government. In democracies such as the United States, this can only occur if elected lawmakers are committed to protecting the environment, reducing carbon emissions, and more. In the United States, one major barrier to enacting climate change legislation is political campaign contributions and lobbying. This money is often spent during the campaign season and when key bills are up for votes in the Congress. From 2000 to 2016, over 2 billion United States Dollars (USD) were spent on climate lobbying on all sides, including from the fossil fuel industry, the transportation industry, and environmental organizations. The fossil fuel industry, transportation companies, and utilities outspent environmental groups and the renewable energy industry by 10 to 1. Iacobuta et. al outlines ways in which countries around the world are fighting climate change with legislation, or "climate action." These include greenhouse gas emissions targets and energy efficiency targets. Recent research has analyzed monetary efforts to oppose climate change mitigation. However, current literature is very limited in its utilization of machine learning methods to assess the impact of such lobbying. In this work, we take a machine learning approach because of the abundance of clean, applicable data that has not yet been studied in this context. We believe it is best to utilize supervised machine learning not only because it is a novel method to approach this problem, but because it is effective to quantify the correlation between two variables, and because it is a useful tool to enter new data into and obtain the predicted voting record (assuming high accuracy).

In this paper, we propose a random forest model to predict the voting records of members of Congress on legislation related to the environment and climate change from campaign finance data as it relates to climate lobbying. Our research will contribute to the understanding of the effect of campaign finance on the passage of significant climate legislation, which in effect carries great weight on humanity's ability to fight the crisis, particularly in the United States. The Center for Responsive Politics (www.opensecrets.org) provides campaign finance and lobbying data, as reported by the Federal Election Commission (FEC), for a variety of industries, such as agribusiness, defense, transportation, labor, pro-Israel, and gun control, from election cycles 1990 to 2020 (every other year). The Center provides information about the amount of contributions in dollars, the contributors, and the recipients (senators and representatives in Congress). Only contributions from Political Action Committees (PACs) and individuals giving $200 or more are included. For our purposes, we sample data from contributors in the following industries: alternate energy production and services, coal mining, electric utilities, mining, natural gas transmission and distribution, oil and gas, air transport, airlines, auto dealers, auto manufacturers, automotive, cruise ships and lines, railroads, sea transport, trucking, transportation unions, and environmentalist groups. The League of Conservation Voters calculates a National Environmental Scorecard for each member of Congress, for each year since 1971 (scorecard.lcv.org).

We train a random forest (RF) classifier (an ensemble of decision trees) to model the voting records of members of Congress on environmental and climate-related legislation (based on the National Environmental Scorecard), given a number of factors, but most importantly campaign contributions from pro-environment and anti-environment groups (like the fossil fuel industry). To maximize the number of data points, each member of Congress in each election year is considered a separate data point (for example, data for Representative Nancy Pelosi in 2018 is separate from her data in 2016). Each data point is labelled with a number from 1 to 10, representing the average of their annual scores across the years of the term they are elected to.

The model inputs are twofold, consisting of background information of the particular member of Congress and campaign finance data from the relevant industries aforementioned. The background information consists of solely the house of Congress (House of Representatives or Senate) and the state that the lawmaker represents. This information is included because the upper and lower houses of Congress differ in the legislative process and the bills considered and because unique environmental circumstances stateside/locally may influence a lawmaker's willingness to tackle climate change or environmental problems. The main focus of our input is the campaign finance data, wherein a monetary amount of contributions, in USD, is noted for each industry (coal mining, oil and gas, air transport, environmentalist groups, etc.).

Author