How ReTweets Spread: The Epidemiology of Viral Messaging on Twitter

Now that my ReTweet map­ping sys­tem is func­tion­ing, I’m able to start com­pil­ing more gran­u­lar data on the actual dynam­ics of the spread of ReTweet streams. 

First, I’ll start with some sim­ple aver­ages. For the first 3 num­bers –depth, users and Tweets, I’m look­ing at entire ReTweet streams, that is the whole tree, start­ing with the orig­i­nal Tweet and all of the sub­se­quent ReTweets. What we see here is that the large major­ity of ReTweet streams only con­tain 2 lev­els of depth, that is ReTweets of the ini­tial Tweet do not them­selves pro­duce fur­ther ReTweets. They also tend to only include two par­tic­i­pat­ing users (the orig­i­nal Tweeter and the ReTweeter) and two indi­vid­ual Tweets. From this, we begin to under­stand that most RT streams are merely one user ReTweet­ing another, and never go any fur­ther.

Aver­ages

The next data point I’m look­ing at is the ReTweets per Fol­lower (RTpF) ratio for the users involved in streams I’ve indexed (just under 20,000 users). The graph below shows the dis­tri­b­u­tion of RTpF in the top 9000 most fol­lowed users in my data­base, I’ve graphed the actual dis­tri­b­u­tion line in blue, with a 30-point mov­ing aver­age over it in black. 

Here we see that while most users had an RTpF of under 1% in my dataset, some users showed much larger ratios, pos­si­bly indi­cat­ing that there are a class of users who are more “ReTweet­able” than oth­ers. In the future, as I have more data indexed, I plan to release a list of those users with the high­est RTpF ratios.

To explore the fol­low­ers to ReTweet issue a bit more, I then ana­lyzed cor­re­la­tions between fol­low­ers and stream depth and total Tweets. I used two fol­lower num­bers, the first is the num­ber of fol­low­ers the “root” level of the stream had, that is how many peo­ple were poten­tially exposed to the “seed” Tweet. The sec­ond fol­lower num­ber is the com­bined total of fol­low­ers of every user who par­tic­i­pated in each stream. 

While we must remain cau­tious not to assume a causal rela­tion­ship between these num­bers, it does become clear that there is no sig­nif­i­cant cor­re­la­tion between either fol­lower num­ber and the depth of a stream. There is on the other hand a sig­nif­i­cant, though weak, pos­i­tive cor­re­la­tion between the num­ber of users exposed to a Tweet and the num­ber of times it was ReTweeted. What this (and the dis­tri­b­u­tion graph above) tells me is that while users who have more fol­low­ers get ReTweeted more often, the num­ber of fol­low­ers plays a less-than-expected role in pre­dict­ing how widely some­thing is ReTweeted. I expect to find that the actual con­tent of Tweets explains more of its “ReTweetability”.

Cor­re­la­tions

The last data point I looked at in this stage of research are aver­age repro­duc­tion rates, that is how many ReTweets in turn trig­gered fur­ther ReTweet­ing. This is com­pa­ra­ble to the bio­log­i­cal Repro­duc­tion rate (R0) con­cept in that it rep­re­sents the aver­age num­ber of addi­tional infec­tions a sin­gle case of infec­tion results in. 

Of those streams with 2 or more lev­els, only 7.57% even­tu­ally gain an addi­tional level, yet, of those streams with 3 or more lev­els, nearly 11.5% grow another level. This trend con­tin­ues out to the 5th level (I did index some streams with more than 5 lev­els, but not enough to gen­er­ate any sig­nif­i­cant data). The more lev­els a ReTweet stream has, the more it is likely to accu­mu­late.

What this may indi­cate is that social proof (or imi­ta­tion more specif­i­cally) plays a role in a user’s deci­sion to ReTweet. The more users a Twit­terer sees ReTweet some­thing, the more likely they are them­selves to ReTweet it. Another fac­tor in the deci­sion to ReTweet that this data point (as well as the pre­vi­ously noted higher occur­rence rate of the “please” call-to-action) may be high­light­ing is that when the act ReTweet­ing is called to a user’s atten­tion, they may be more likely to ReTweet.

Repro­duc­tion Rates by Depth

Look for more posts on this sub­ject, as I’ll be devel­op­ing more func­tion­al­ity into my ReTweet tools and I’ll also start inves­ti­gat­ing content-based cor­re­la­tions, that is what are the fac­tors of the con­tent of a Tweet that make it more or less likely to be ReTweeted.