Using Machine Learning to Measure Effective Stringency Index

Government Response Stringency Index is basically a composite measure based on nine response indicators including school closures, workplace closures, and travel bans, rescaled to a value from 0 to 100 (100 = strictest). Basically, it quantifies how strict a country’s policies are regarding coronavirus pandemic.Visit Our World In Data

I was randomly exploring some data when I noticed Bangladesh’s Stringency Index has almost never been less than 80(with a few exceptions) since March 2020. Which came to me as a surprise because of all the people out there enjoying a normal life. Thus, I wanted to measure the effective stringency level of the country using machine learning.

For this, I have basically taken the top 30 countries with the highest statistical capacity score as the benchmark and used around 30 variables such as – Cases per million, death per case, hospital facility, new tests per day, total tests per thousand, positive rate, etc. as the explanatory(input) variable. And came to the results.

According to the model – Bangladesh had the highest stringency score around April to June and after that, it has slowly gone down as we have seen. And currently, Bangladesh has a policy level stringency score of 80.9 while the effective stringency score is around 55.

This shows us how ineffective Bangladesh Govt.’s policies are and how the administration has failed to actualize the taken policies.

Summary:

1. I have trained a machine learning model to compute the effective stringency index.

2. The model shows that Bangladesh’s effective stringency score is at least 30% lower than the computed stringency score. Meaning the policies taken by the government are ineffective.

However, this might not be completely accurate as the model takes the top 30 countries with the highest statistical capacity score as the benchmark and the collected data was incomplete which needed imputation – thus resulted in a noisy prediction in some cases(Graphs added)

The whole work was done in python. Currently, the codes are in a really messy state, but if anyone wants to take a look at them please let me know!

Please note that this was done as a personal project and does not carry any inferential value.