August 28, 2014

On Facebook when the earth shakes…

By: Ariel Evnine, Andreas Gros, Aude Hofleitner

On Sunday August 24th, 3:20 a.m Pacific time, an earthquake of magnitude 6.0 occurred in the Bay Area, 3.7 miles (6.0 km) northwest of American Canyon near the West Napa Fault. It was the largest earthquake in the Bay Area since the 1989 Loma Prieta earthquake.

During a crisis, people turn to Facebook to stay connected to their friends and family. They use it to receive social support and keep the people they care about informed on how they are doing.

We used aggregate, anonymized Facebook activity data from cities located within 300 kilometers (around 200 miles) from the epicenter and analyzed how their Facebook activity following the earthquake differed from usual activity. A recent post from Jawbone (https://jawbone.com/blog/napa-earthquake-effect-on-sleep/) of the earthquake on people’s sleep. Here we are interested in social interactions and communication.

How much does the earthquake affect Facebook activity?

On most days, Facebook activity is quiet at 3:30 a.m., PST. A magnitude 6.0 earthquake wakes up an entire region and it is interesting to see what people do once they are in such a situation.

We compare the activity on Facebook on August 24th, 2014 to the activity a week prior at the same time (August 17th, 2014).

Effect of the distance to the epicenter

The map below shows the relative difference in activity on Facebook between the 24th of August, 3:21 a.m. and 3:26 a.m., and the same time period one week earlier.

For visualization, we cluster together nearby cities which showed similar changes in activity. The color represents the percent variation in activity (red: largest activity, yellow: lowest activity). The size represents the area covered by the cluster. The blue cross indicates the location of the epicenter.


Relative variation of activity on Facebook in the five minutes following the earthquake.

Similarly, we compare the variation in the number of posts in a city to the city’s distance from the epicenter. The variation in the number of posts is computed as the ratio between the number of posts in a city within one hour following the earthquake to the number of posts on August 17th at the same time.

We ran a linear regression between the distance to the earthquake and the posting variation (in log scale).

Spikes of activity

We see very significant spikes in Facebook activity for people located in a 300 km radius of the earthquake. In the beginning of the night, the activity is very similar on both August 17th and 24th. The difference spikes at 3:21a.m., just following the shake. We notice people staying more active than usual throughout the night. The difference decreases in the early morning (more than two hours after the earthquake) but never to the usual level of activity. In the morning, the number of posts increases above normal number of posts and the additional activity remains for the entire day.

What do people talk about?

We looked at all public posts from people within 300 km from the epicenter during the hour following the earthquake on August 24th. The following word cloud shows the frequency of words used. “Earthquake” comes as the most commonly used word, also very common are “American Canyon”, which is the location where the earthquake occurred. Happening in the middle of the night, the earthquake had a strong effect on people’s sleep (“wake”, “sleep”). People also express their fear and general feelings and enquire about friends and family.