The Times They Are Rarely A-Changin': Circadian Regularities in Social Media Use

What does the baseline use of social media look like, across countries and time zones? We investigated by analyzing data from 25,000 Twitter users and found that social media interaction is dependent on circadian patterns.

Abstract

This paper uses geolocated Twitter histories from approximately 25,000 individuals in 6 different time zones and 3 different countries to construct a proper time-zone dependent hourly baseline for social media activity studies.  We establish that, across multiple regions and time periods, interaction with social media is strongly conditioned by traditional bio-rhythmic or “Circadian” patterns, and that in the United States, this pattern is itself further conditioned by the ideological bent of the user. Using a time series of these histories around the 2016 U.S. Presidential election, we show that external events of great significance can disrupt traditional social media activity patterns, and that this disruption can be significant (in some cases doubling the amplitude and shifting the phase of activity up to an hour). We find that the disruption of use patterns can last an extended period of time, and in many cases, aspects of this disruption would not be detected without a circadian baseline.

Background

There is a quickly growing literature on the role of social media and social media users in politics and political outcomes. But there is very little work on what the baseline of use on social media looks like, how it is conditioned by attributes of the user, and how quickly disruptions to the pattern return to normal. Understanding and properly accounting for how a steady, and largely circadian, pattern impacts baseline activity is important for inference, both when explaining shifts in use, and using those shifts to explain other outcomes.

Study

To understand this pattern, we present evidence from 25,000 Twitter users in three different countries (e.g. France, England, and the United States) across six different time zones from the time around the 2016 U.S. presidential election. We establish that, across multiple regions and time periods, interaction with social media is dependent on traditional bio-rhythmic or “circadian” patterns. For example, although the users in the American sample tweet at roughly 4-6 times the rate of users in the European sample, they do so in the same cyclical pattern.

Results

We find that highly significant events can disrupt traditional social media activity patterns, in some cases doubling the amplitude and shifting the phase of activity up to an hour. The disruption of social media use patterns can last an extended period of time, and that, in many cases, the “circadian” baseline is necessary for recognizing aspects of these disruptions. While accounting for underlying usage dynamics is of methodological value to social media studies, we also hope our analyses here will serve as a launching ground for substantive empirical studies in the future.