It has only been a few months since Twitter became a public company, and its reported growth continues to be under intense scrutiny. In its first post-IPO financial statement on February 5, the micro-blogging firm attempted to keep things looking positive, much as it has done in when the company went public in November 2013.
Back then, in its filing with the U.S. Security and Exchange Commission, Twitter revealed that its average monthly active users in the previous quarter was 232 million. Impressive, no?
But there is much information hidden in this huge figure. Chiefly, these 232 million users don’t reflect the company’s churn (how many people are leaving the service each month). Nor does this total reveal Twitter’s growth, as in how many new people are joining on a monthly basis. These numbers are essential if we want to make an educated projection of Twitter’s prospects.
This short paper will take a look at the findings from a study of Twitter users’s behaviour, revealing the company’s true growth and churn over the two-year period prior to their initial public offer on Nov. 7, 2013. While we find that user growth is slow but steady, we also discover that only half of those who started accounts two years ago are still active.
We used Conditional Independence Coupling to create a sample of 36,435,499 tweets from 51,266 authors between November 1, 2011 and October 31, 2013. A 30-day windowed correlation was performed across the sample. This was done by taking a cohort of users who actively posted in a 30-day period and then measuring that group’s activity in successive 30-day periods. The results were averaged over several different cohorts.
Posting was used as a proxy for activity, though there are other ways to engage with Twitter. Users can log in to Twitter to read posts from their friends, or to access Twitter programmatically through Twitter’s Application Programming Interface (API). Twitter counts all of these actions as activity when they are calculating their Monthly Active Users (MAU). To judge the accuracy of posting as a proxy for activity, we compared the 30-day MAUs that we measured in the past 15 months to those reported by Twitter. Numbers were normalized using the value the company reported in the previous quarter (June 30, 2012). Twitter’s reported values (illustrated by bars in Figure 1) and our measured values (illustrated as points in Figure 1) agree well within the experimental error of the measurement. We were confident that the measured sample accurately represented the evolution of Twitter’s growth during that observed period.
Figure 1: Twitter’s Monthly Active Users
Easy Come, Easy Go
In this way, we were able to calculate Twitter’s churn and growth as a function of time. Churn is measured as the ratio of people from the initial cohort that are active within the subsequent month. Growth is defined as the ratio of new users active in a subsequent month that were not part of the initial group. Figure 2 shows the results of averaging the growth and churn for several different cohorts.
Figure 2: Twitter’s Growth and Churn
Both the growth and churn are approximately linear and can be fit using a linear model. From this model, the average month-over-month churn is 1.2±0.1% (R2 = 0.88) and average month-over-month growth is 9.8±0.2% (R2 = 0.98). There is a noticeable increase above the average for both churn and growth going from the first to the second month. The first month churn is 14.6±1.2% and first month growth is 20.3±0.4%, indicating that each month there are many people joining and leaving Twitter. This single churn rate can be estimated by subtracting the month-over-month baseline from the first month churn and is found to be 11.95±0.6%.
Counting Idle Accounts
Our study into Twitter users’ behaviour over a two-year period sheds some light on the firm’s short and long-term growth. The short-term result focuses on users’ initial first month behaviour. Each month, about 12 per cent of active Twitter users are new accounts who will stop using the micro-blogging service after that month. With current monthly active users at 232 million, that translates to 28 million accounts. These could be users that are trying out the service and finding it not to their liking, spam accounts suspended by Twitter, or accounts that are created as one-offs. Using Twitter’s published growth rate, this sharp first month drop-off suggests that there should be 659 million (±10 million) Twitter accounts that are no longer active as of October 2013. This agrees quite well with the estimate done by Twitter statistics tracking site Twopcharts showing 651 million inactive Twitter accounts for that same month.
As for the long-term, the measurements predict that the half-life of the average Twitter user is 29 months. This means that almost two-and-a half years after an account is made, there is a 50 per cent chance that that user is no longer active on Twitter.
There are two things Twitter doesn’t want you to know about its active users: not only are there fewer than they say there are, but roughly half of those who are active today won’t be around by mid-2016. Our two-year analysis of active user’s behaviour indicates that Twitter’s growth over the next few months and the next couple of years may be slow, but is steadily climbing. Each month there are many millions of people joining and leaving Twitter, but active user growth is superseding its churn. This sounds like good news, which begs the question why Twitter feels the need to hide their idle accounts.
- K. White, J. Li, N. Japkowicz, “Sampling Online Social Networks Using Coupling From the Past,” 2012 IEEE 12th International Conference on Data Mining Workshop on Data Mining in Networks, pp. 266-272↵