Political Networks and COVID-19 Mitigation

Question: What factors are most correlated with COVID-19 mitigation policies in the 50 United States?

A reasonable expectation might be that the states that had the worst incidences rates had similar policies to help mitigate the spread of the epidemic. Over the course of the pandemic, it became apparent that this was not the case. Politics seemingly had more impact than actual incidence values.

GOAL: Understand when political similarity impacted similar policies.

We first find a method for best classifying the politics of each state and then create networks of the 50 states with connections based on various factors. We see which of these relationship structures best matches one created from similar mitigations.

How do we measure mitigation policy Strength?

We use the measures to the right to describe each state's mitigation policies. The data is available from the Google COVID-19 Open Data Repository.

For each of the measures, we first normalize (so everything is between 0 and 1) and then create a vector these. We measure the length of this vector as the strength of the policy. So, states with strong policies will have larger values and states with 0 policies will have 0 length. We can visualize what this looks like for all 50 states on the right.

Measuring Each State's Politics:

We use 4 measures of political affiliation to categorize a state's politic measure.

2021 Governor Affiliation
2021 State Legislature Majority
2021 State Senate Majority
2020 Electoral College

We consider a 5th measure which is an aggregate of all 4 and is displayed on the right.

What's the best measure of politics?

To determine which political measure is best. We calculate the correlation between each of our political measures and the mitigation policies. High correlation would indicate that the political affiliation corresponds with mitigation policy strength.

We also look to see when democrat states and republican states have statistically significantly different policy intensities. This is present in the figure to the left.

Generating Networks

We now represent various factors that could impact mitigation strategies as networks. We consider

Geographic proximity
Incidence Similarity
Political Similarity

Each of the graphs is shown to the right as well as a Mitigation Policy Similarity network as well. We are going to look at how these graphs are related. If the connections between the graphs are correlated, then we can say the graph structures influence each other.

Graph Correlation

To calculate graph correlation, the most simple method is to consider how many edges the two networks have in common. However, to know if this result is unusual (ie. not random), we need to compare this with randomly generated networks.

To better this analysis, we will consider Quadratic Assignment Procedure for our correlation. Like above, we look at how many edges the graphs have in common, but instead of comparing our graph to random graphs, we compare it to all graphs of the same structure (just mixing the state's names around). A significant result now implies that the labels themselves (ie. where the states are in the network) matters.

Political Similarity Network

Geographic Network

Incidence Similarity Network, May 1, 2020

Mitigation Similarity Network, May 1, 2020

Do the Relationships Matter?

We can visualize the results of the QAP analysis on the left. There is a strong positive correlation between the mitigation similarity network and political similarity networks from May 2020 through June 2021. This implies that states with similar political backgrounds had similar mitigation policies.

This result might seem fairly obvious, given the correlations above and our personal experiences of the pandemic. HOWEVER, not that the Incidence similary networks have little to no correlation with the Mitigation Similarity networks. This implies that states with similar incidence levels did NOT have similar strategies. So, if two states were seeing an increase or decrease in case load, they did not respond similarly. That is, the epidemic did not drive policies creation.

Google Sites

Report abuse