Big Data Is Decision Science: The Case of COVID-19 Vaccination

Introduction

Big data is the handling of vast amounts of data through a flexible, mostly cloud-based IT architecture. Big data is here to stay for many reasons. The first is that digitization is data extension, with new digital content creation growing at a rate between 40 and 60% a year. This rate means that Google, which was indexing a million pages for a few million searches in 1998, now indexing more than a trillion pages ten years later, for1.2 trillion searches a year. Second, big data can fuel new powerful machine learning techniques to uncover otherwise hidden relationships between data, that can support new powerful insights. As an example, Netflix shifted its content recommendation engine, based on customer rating, to a machine-learning algorithm fed by a large set of big data ten years ago. The tool offered a revolution to support personalized recommendations, in such a way, that now, four out of five Netflix movies and TV series viewed by Netflix subscribers originate from machine-based suggestions (Amatriain, 2013).

The seminal work by Brynjolfsson, et al. (2011) reveal that companies leveraging big data for more fine-tuned decision making, could increase their revenue productivity by more than 5%. A few years later and using a more global sample, Bughin (2016) had reached essentially the same conclusion- that is big data is usually responsible for an uplift of more than 5% in labor productivity for firms globally. Besides being used for business, big data is also important when it comes to helping major social issues, in particular when it comes to complex issues such as predictions of climate change (Bauer et al., 2021), vote intent (Mavragani and Tsagarakis, 2019), or traffic congestion (Teseng and colleagues 2018).

One another current case in point is the vaccination issue linked to the covid-19 pandemic. While the covid-19 pandemic continues unabated across the world, affecting hundreds of million people worldwide, the only serious way to eradicate the pandemic is mass- vaccination. However, large vaccination acceptance is not warranted. Social movements against vaccination have increased during the covid-19 pandemic, fueled by a series of conspiracy theories (Nguyen and Catalan, 2020). There are also possibly multiple, and interacting elements that affect vaccine intent, which make predictions of vaccine used complex—A recent poll performed by the high profile organization Pew Research in the US also emphasizes that the shape of vaccine intent is driven by « rather complex and interrelated factors » that might require more deeper data analytics to sort out a clear ranking of actions to support mass-vaccination¹.

We thus leverage machine learning to profile vaccination intent for covid-19, concentrating outside of known product factors (such as the effectiveness and safeness of the vaccine) to determine additional factors that could be boosted or alleviated to enhance fast vaccination against the Covid-19. Specifically, after aligning with recent theories of vaccine determinants to build an extensive database of behaviors and vaccine intent across 5 Continental European countries—Germany, France, Spain, Italy and Sweden, we use Random Forest (RF) techniques to provide a granular view of vaccine intent. RF is one of many machine learning techniques, but is found to generate rather accurate predictions regarding protective uptake (Bughin and Cincera, 2021)

Reaching for advanced analysis techniques has been complemented by a dedicated approach to data collection. Predicating behavior in a sensitive context such as vaccination intent in the face of the covid-19 pandemic required the usage of tools that would help decrease the natural tendency to distort declarations.