Do Lockdowns Work?

Inspiration

Our project was inspired by a desire to understand underlying causal relationships in lockdown policies. Obviously, with regards to COVID-19, there are a plethora of perspectives and opinions, and it remains a very controversial topic of discussion. If we can take steps towards understanding causally why certain phenomena occur, we can make great strides to gaining a fuller understanding of the disease and being able to act more confidently in response to it. Responses to COVID-19 are particularly problematic. Lockdowns prevent small businesses from succeeding and force many into unemployment which hurts the economy. Still, advocates of lockdowns believe they will help slow down infections and allow businesses to reopen for good (i.e. long-term benefits).

What we learned

We learned that achieving causality is something that is not easy, but it can be extremely satisfying and compelling. To make sure that what we were studying was not the result of simple correlations, we had to consider the ramifications of our conclusions and match that with our common sense and understanding of the disease and its current events. This required a bit of domain knowledge and extra research which made us confident that we were on the right track.

How we built our project

This project is purely a data science project, that is, our project consisted purely of analyzing data and coming up with visualizations and technical results. To this end, our project tools consist mostly of Python, with the classic libraries used for data science e.g. numpy and pandas. We calculated Spearman correlation between cases and income, and we utilize a CausalImpact library to forecast the time-series of new cases before lockdown and then show a difference that could be attributed to lockdown. Additionally, we employed a few key models from Python that helped make our analysis robust. For example, we used a transfer entropy package to show causation between two-time series. This transfer entropy metric was important since it did not make spurious assumptions about the data being linear.

What we accomplished

We aimed to study a lockdown’s effect on decreasing potential cases and on the economy, as well as exploring how it affects various socioeconomic groups at a county level. We are happy that we were able to come up with some meaningful, robust conclusions all in the span of one weekend. This is our first time working with causal models, so being able to come up with a compelling result with them is quite satisfying. We think that in future data science projects we can continue to leverage these powerful models because the ability for other humans and experts to interpret and be convinced by our work in data science is of great importance to us.