It is mainstream that to share a news on Twitter one should repeat it a few times, for the simple reason that your followers aren’t up online on twitter all day and that they are likely to miss your great news.
AtÂ WiselyticsÂ we like to look at statistics from lots of data points, and since we have our hands on around 1B tweets, we thought we could get some concrete numbers about if one should repeat tweets.
Some went into doing a fact checking analysis of their past tweets which they did repeat as featured in marketinglandÂ orÂ Guy Kawasaki’s blog.Â Tomasz Tunguz, venture capitalist at redpointÂ even did a thorough statistical analysis of his full history of tweets and all concluded that indeed it is very valuable to repeat his tweets.
Our Data perspective
By repeat we don’t mean “copy-paste” of the same tweet. It’s against Twitter policy. Repeating a tweet needs a bit of human editing, would it be for a different wording, or emphasizing a different part of your news.Â Thus we had to start by devising an algorithm which could identify for 2 tweets if they talk about the same news een if they are not identical copies. It performs very well, with 90% precision thus giving us the ability to scan through tweets and automatically identify those which are repeats.
Then we sample 1.5K accounts, looked at 3 months of their tweets and ended up analyzing 1M tweets. We identified all the tweets with repetition, and for each repetition, we normalized its performance with the performance of the initial tweet. Performance being RT + Favorites.
A first interesting statistics is that 55% of the twitter accounts are repeating their tweets. And among the 1M tweets, 3% are repeated.
Â Now here are the main findings:Â On average, the second tweet about a news get 86% as much performance as the first one. And in the following graph, you can see the performance decreasing slowly as number of repeats increaseÂ (see at the end of blog post for the blue confidence interval explanation)
The more one repeat the less the performance he gets, but even after 6 repetitions, we’re still at 67% of the first tweet.
Read from another perspective, we can estimate that on average, when you repeat a tweet, only 14% of your audience is seeing your tweet for the second time.
Comparison with Facebook
A while back, we ran a similar analysis for Facebook posts and had found that one can safely repeat important news on Facebook, but to a lesser extent that on Twitter. This is confirmed on the graph below:
The performance of a repetition on Facebook decreases much faster than on Twitter: 38% loss at the 1st repetition on Facebook vs 14% on Twitter.
Note on confidence intervals
“Performance of a tweet repetition” is showing the average performance with a confidence interval of only 10%. We do have a strong case about the decrease in performance repetition after repetition, however it depends much on the user’s twitting behavior.Â Basically if a tweet is terribly timed for its first iteration, its repetitions can perform way better than Â this first tweet (or the opposite)
Our analysis was run over enough tweets and users to cancel out for the impact of timing, however, asÂ we can see on the graph below (performance of the 5th tweet across all tweet-news) there is indeed quite a bit of variance.
The median performance for the 5th repetition is 76% but it does happen frequently that it performs better (>100%) and even sometime 2 to 3 times better than the first time the news was tweeted.