As part of our recent work investigating the Twitter Userbase, we have collected data on accounts registered around the 2011 triad of natural disasters; the Queensland Floods (January), Christchurch Earthquake (22 February) and Tokyo Earthquake & Tsunami (11 March). By looking at new accounts registered between 1 January 2011 and 30 April 2011, we are beginning to investigate whether such disasters were a driver of new Twitter registrations, or whether the use of Twitter during the floods was simply among those already utilising the platform. While processing this data, we realised we had also captured the Egyptian Revolution, which showed a very similar pattern. And that pattern, as it turned out, was clear:
Let’s start by looking at the total registrations over the time period in question, a couple of different ways. The two charts below show both the average ID registered on the day (i.e. where in the Twitter ID range the users are), as well as the number of users who registered there account on each of those days.
As you can see, there are two dates where the pattern looks odd, namely 1 February and 18 February. Taking a closer look at the first of these, and registrations on a per-minute basis, we see a rather odd pattern that appears to show re-occurring outages:
We have yet to discover if this is an error with our data, or represents periods where twitter was either down or overloaded, but it does not appear to impact directly on the analysis that follows.
Registrations during Queensland Floods:
Let’s begin our detailed analysis with a closer look at the Queensland floods. There are a few different ways of attempting to locate users on Twitter, the most obvious being geolocation, although we didn’t attempt that for this analysis as it is well established that the percentage of users with geolocation on is historically around 1%. Another approach is to look for users with “Brisbane” (or similar) in their location field, while it is also possible to look at the time zone a Twitter user has set for their account, and at the UTC offset applied to their account. Let’s start with the location field and UTC offset:
It is immediately clear that “Brisbane” in the location field is not a good proxy. Even allowing for other cities and regions where the +36000 offset is applicable, the difference between 220 registrations and over 6,00 is obvious. Both charts however show a spike around the time of the floods (which peaked in Brisbane on 13 January), while the UTC offset chart also shows a peak around the time of the NZ earthquake, and the recovery that followed. Drilling deeper than the UTC offset, and not quite as deep as matching text in the location is the time zone field, so let’s take a look at that:
Here, again, we see large spikes for Brisbane, Melbourne and Sydney during the floods, and spikes particularly for Sydney and Melbourne around the NZ earthquake. So it appears clear that for the floods, not only was Twitter a source of information on the floods, but the floods were a driver of new registrations for Twitter, from users seeking to obtain that information and possibly contribute to the discussion.
Registrations during Christchurch Earthquake:
So, how about the Christchurch earthquake, which hit on 22 February? Well, we see a similar pattern. A spike in the +43200 UTC time offset, and in both the Auckland and Wellington time zones, shows the natural disaster was a clear driver of new registrations to Twitter. The second spike shown here, on 11 March, coincides with the Tokyo Earthquake & Tsunami, which posed a potential Tsunami threat to New Zealand.
Given that Twitter does not have a “Christchurch” specific location field, I repeated the text matching experiment here. But, while the spikes are obvious, the numbers are too low for this to be considered a reliable method of user identification.
Registrations during Tokyo Earthquake & Tsunami:
So, on to the Tsunami, and a familiar pattern is visible, albeit one much greater in magnitude most likely due to the sheer population in and around Tokyo:
As you can see from the above chart, the rise in new registrations as a result of the earthquake and tsunami was clearly extended here substantially into the recovery period, with increased numbers of registrations evident right through to the end of March. We already identified a spike in New Zealand registrations during this period, but the effect was also felt in other regions that may have potentially been impacted by the tsunami, as seen below:
Registrations during Egyptian Revolution.
Finally, given that we had the data, it made sense to look at Cairo and Egypt, around the time of the Egyptian revolution on 25 January. Just as with the natural disasters above, the revolution proved to drive users to Twitter, as highlighted in the graphs below:
We continue work on these to establish the longer-term effect, and to measure the significance of any residual increased publicity following the spikes, however it is evident from these graphs that not only is Twitter an important source of information during natural disasters, but the increased awareness around the platform, and its importance during the crisis events, leads to a substantial increase in new user registrations.