As the world is taking cautious steps to reopen the economy without increasing the pandemic risk, researchers have discovered an unlikely ally: social media.

We’ve learned over the course of the pandemic that when it comes to both health and the economy, we need all the information we can get. For all the progress we’ve made in recent years, society is still not as robust and knowledgeable as we’d like to think, at least in a situation of crisis.

The challenge of assessing the impact of natural hazards on people and businesses is of the utmost importance in order to understand the ability of such systems to recover, as well as to direct policies that offer support.

Traditionally, recovery estimates are done through surveys and interviews, which is time-consuming and costly. But a team of researchers from the University of Bristol believes there’s another way.

In a new study published in Nature Communications, they describe how social media can be an accurate indication of downtime and recovery time for local businesses following a natural hazard — particularly small, at-risk businesses.

The practice is called nowcasting — the ‘prediction’ of the present (or in a broader sense, the very near future and the very recent past). Using social media posts, the researchers were able to nowcast the downtime of small businesses for three natural hazard events that took place in Mexico, Nepal, and Puerto Rico.

It has never been more important to assess the impact of natural hazards on businesses, says Dr. Filippo Simini, Senior Lecturer and lead author of the study — in particular, for small businesses who are more vulnerable and at-risk from hazards.

“The challenge of nowcasting the effect of natural hazards such as earthquakes, floods, hurricanes, and pandemics on assets, people and society, has never been more timely than ever for assessing the ability of countries to recover from extreme events,” he says.

“Often, small to medium-sized businesses slip through the net of the traditional monitoring process of recovery. We noticed in areas struck by natural hazard events that not all areas and populations react in the same way.”

Studies on business recovery generally use large-scale economic indicators (such as the total number of money spent on sales, or the number of active businesses). But economic indicators are not as readily available or as relevant for events on a more moderate scale, or events that affect different areas in different ways. This is where alternative statistics can come in handy, as a proxy for this type of information. For instance, changes in pedestrian foot traffic or parking transactions could offer some indication of how businesses are doing.

Social media posts can also serve as indicators of economic activity. The method assumes that businesses tend to publish more when they are open and less when they are closed. When researchers analyzed public Facebook data collected before, during, and after three natural disasters (the 2015 Gorkha earthquake in Nepal, the 2017 Chiapas earthquake in Mexico, and the 2017 hurricane Maria in Puerto Rico), they were able to correlate the number of Facebook posts from small urban businesses to their post-disaster recovery. Facebook text analysis also supported this view.

Dr. Flavia De Luca, Senior Lecturer in Bristol’s Department of Civil Engineering and lead author, added:

“We had the idea of supporting post-emergency deployment of resources after a natural hazard event using public Facebook posts of businesses to measure how a specific region is recovering after the event. It was amazing to find out that the approach was providing information on the recovery in ‘real-time’.”

De Luca continued, “We would like to test the method to measure the economic impact of the COVID-19 pandemic.”

Simini and De Luca complemented the method by using satellite imagery. They counted the number of vehicles in imagery collected by Google Earth, correlating the presence of vehicles in business zones to the activity of businesses. Other data (such as anonymized mobile phone data or data from other social networks like Twitter or Flickr) can also serve as a complement here, the study notes.

If the method can be confirmed on larger datasets, it could serve as a way to see which businesses need the most help, directing policy support where it is most needed.

The fact that the method works in real-time with public data, and can be adapted to all languages and cultures makes it particularly appealing

The analysis algorithm is also freely accessible and open source. You can reproduce the results presented in this paper (and perhaps try it on different datasets) by checking this Git: https://github.com/roberteyre/Business-Recovery.