How many followers does the average Twitter user have?  How many accounts does the average Twitter account follow?  How many times has the average account tweeted?  What about the median?  These questions seem simple, but it is not easy to find answers to them.  Twitter only discloses how many monthly active users exist, and other articles appear to use data from 2013 or 2014.  Only one report provides information on the distribution of accounts followed, and it again was written in 2013.  I could also find no report that calculates median statistics; since Twitter, like all social networks, exhibits a right-skew in the number of connections, median figures provide a more accurate understanding of the normal user than average ones.   Moreover, most statistics are for Twitter overall, but there is certainly variation by country: Americans should have more followers than Albanians simply because there are more of them.  Specialized analytics companies such as socialbakers and statista segment users, but their data require payment and provide finer segmentation than what I am interested in.  Lacking a ready answer to these questions, I answered them myself.

To answer these questions, I consulted over 18 months worth of Twitter data.  Since September 2013, I have connected to Twitter’s 1% stream and requested only tweets with GPS coordinates and, every hour, save the ingested tweets to a file.  For this post, I randomly sampled 50 hours’ data and calculated follower, following, and frequency statistics globally and by country.  50 hours equals about 10,371,979 million tweets.  I deduplicate users so that active tweeters, who are more likely to have many followers, do not skew the results; the resulting statistics therefore apply to the average account, not the average tweet. Deduplication reduces these tweets to 2,646,337 accounts.

The results are below and available for viewing and downloading on Github.  The table is sorted by number of accounts per country, with the first row being the total number of accounts. The first row of the Population column is the global population, while the other columns of the first row are averages or medians across all geographic units.  The results do not include the United States because my stream excludes the US in the bounding boxes it sends to Twitter.

So, what did I learn? The average Twitter user has 1,050 followers, though the median has 275. The average user follows 532 accounts, the median 264. The average account is responsible for 10,311 tweets, the median 4,034. These numbers are higher than what I linked to earlier, for two possible reasons. First, I have looked at Twitter 1-2 years after the studies in the links; Twitter has grown in size, and users may gain followers as a function of time (independent of Twitter’s population). Second, I look at tweets with GPS coordinates only, and those tweets may differ from those without GPS coordinates. I return to this point shortly.

Interesting patterns emerge when looking at the country rankings. Countries with the most accounts also have many people.  Brazil has the most accounts, followed by Indonesia and Turkey.  This result is actually not surprising, as each country is populous and is known for having active Twitter communities.  Surprisingly, the overall correlation between a country’s population and the number of GPS-tagged accounts is low, .235.  Few low income countries crack the top 50, none the top 25.  I was most surprised to see how many geographic units Twitter encompasses: 83 countries or territories have fewer than 100 accounts in this sample, and 9 (Western Sahara, Micronesia, Heard Island and McDonald Islands, Jersey, Kiribati, Montserrat, Pitcairn, Sao Tome and Principe, and Wallis and Futuna) have 1.  Antarctica has quadruple the number of accounts.  Fun stuff!

In a future post, I will investigate how these results change when I normalize by population. I will also see how the rankings change when using metrics besides number of accounts, i.e. some areas may have users with more per capita followers than others, suggesting they have denser social networks. Time permitting, I will see if the statistics presented are stable or change as the sample size increases.  Especially for small countries, 50 hours of tweets may not provide enough data to accurately measure the true statistics.  I will perform the same analysis for 1 hour, 10 hours, 20 hours, […], 100 hours to see if the statistics, globally and by country, stabilize.

Finally, keep in mind that the sample consists of users who elect to allow Twitter to assign GPS coordinates to their tweets. Research has shown that users with GPS coordinates skew towards urban areas and higher incomes , so they might skew on social network characteristics as well.  (If anyone knows work comparing GPS-tagged tweets to non-GPS-tagged tweets, I’d love to see that.  The finding on GPS-tagged tweets at that link may be a finding about tweets, as it accords with work from Pew on Twitter users.)  Future work should therefore calculate these statistics on the global stream to see if they differ. I expect that users who select into using GPS coordinates are heavy Twitter users, so they should have more tweets and followers than non-GPS accounts.