Sentiment, the Pandemic, and COVID-19 Policies across the US States

[chengjun]

Mining the Cause of Political Decision-Making from Social Media: A Case Study of COVID-19 Policies across the US States

Zhijing Jin et al.

https://drive.google.com/file/d/1Y2Wcn8D9sBcSi4Or0cR7gxA2jZ5jUUQE/view

Abstract

Mining the causes of political decision-making is an active research area in the field of political science. In the past, most studies have focused on long-term policies that are collected over several decades of time, and have primarily relied on surveys as the main source of predictors. However, the recent COVID-19 pandemic has given rise to a new political phenomenon, where political decision-making consists of frequent short-term decisions, all on the same controlled topic—the pandemic.

In this paper, we focus on the question of

how public opinion influences policy decisions, while controlling for confounders such as COVID-19 case increases or unemployment rates.

Using a dataset consisting of Twitter data from the 50 US states, we classify the sentiments toward governors of each state, and conduct controlled studies and comparisons. Based on the compiled samples of sentiments, policies, and confounders, we conduct causal inference to discover trends in political decision-making across different states.

Q2: What Causal Impact Does Sentiment Have on the Policies?

Formulation by Do-Calculus. Formally, we are interested in the effect of a cause X (i.e., Twitter sentiment) on the outcome Y (i.e., policy change) in the presence of the confounder Z (i.e., case numbers, unemployment, etc.), as shown in Figure 2.

https://zhijing-jin.com/fantasy/

In Table 6, we show the top five states with highest β values, and five states with β values that are the closest to zero. The higher the β value, there exists more alignment between people’s sentiment and the state policy strictness in the state.

• Social distancing policy summaries: https://coronavirus.jhu.edu/data/state-timeline
• COVID case number data: https://github.com/KFFData/COVID-19-Data
• Monthly unemployment data: https://www.bls.gov/web/laus/ststdsadata.zip
• Population data: https://www.census.gov/programs-surveys/decennial-census/data/tables.2010.html
• Urbanization data: https://www.icip.iastate.edu/tables/population/urban-pct-states.
• Presidential election return data: https://www.nytimes.com/elections/2016/results/president

For simplicity, we collect the pre-COVID data at the time point of January 2020, and do not consider the change of governorships in two states in early 2021.

There are some associations between our results and real-world patterns. For instance, among the top five states in Table 6,

• Colorado’s high β value reflects its Democratic governor’s large net favorable rating compared to the Republican politicians.14
• Massachusetts also has a high governor approval rate, and most people support the COVID policies.
• The three Republican states, South Dakota, Texas, and Florida, also have high β, but they are in a different scenario.

The loose policies in all these states are in line with general sentiment across the states to refuse restrictions.

Code and data are publicly available at https://github.com/zhijing-jin/covid-twitter-and-policy