,
23 tweets,
11 min read
Read on Twitter
1/ [Data #Thread] *Klaxon for all those interested in potential Twitter manipulation*. Here I compare a large sample of pro and anti-Trump Twitter accounts. Sample A includes around 137,000 pro-Trump Twitter accounts, and sample B, 65,000 anti-Trump accounts #disinformation
2/ Both Samples involved downloading Twitter accounts based on data from their Twitter bios. Pro-Trump accounts were selected if the contained terms such as #MAGA, #KAG #WWG1WGA #AmericaFirst (and similar). Anti-Trump accounts were chosen to see if they contained #Resist
3/ #DumpTrump #ImpeachTrump etc. While obviously not exhaustive, the sample is large. The first obvious potential finding is that there are clearly more overtly pro-Trump Twitter accounts in terms of biographical detail. Interesting when you consider that he lost the popular vote
4/ Next up I analysed when the Twitter accounts were created. In previous analysis, I highlighted that large numbers of accounts being created at the same time could be indicative of a) troll or bot activity b) a huge event driving ppl to sign up to Twitter...
5/ Some will remember that a lot of the pro-Trump accounts were created in January 2017. While this was the month of Trump's inauguration, I found it strange that a similar spike in such accounts weren't created when he wont the election (December). As you can see from ....
6/ the graph below, 4205 pro-Trump accounts were created in January 2017. This is much higher than the average and the standard deviation. Indeed, it is obvious to the naked eye. (FYI the bars shows the accounts created per month since Twitter started, the colours represent days 
7/ of the month). The general number of accounts being created remains quite high. Another spike in January 2018 is also somewhat anomalous. However what surprised me was when I examined the #Resist accounts...
8/ When I analysed the account creation pattern of anti-Trump accounts I noticed the same spike in January 2017! The same as with the pro-Trump accounts. It is also hugely anomalous. So why are we seeing huge spikes in anti-Trump accounts being created at the same time? 😬
9/ Well there are a number of potential explanations, for which we can make assumption. 1) Let's assume all the Twitter activity is organic, i.e. real people signing up to Twitter to cheer on or condemn Trump. That is possible, after all January 2017 was both the month
10/ of Trump's inauguration and the month of the global Women's march. I am still not sold on this explanation, as I still think both pro and anti Trump accounts would have ballooned upon him winning the election in December. However, let's keep assuming the assumption
11/ that they are organic is true. Which came first, the MAGAs or the RESISTS? After all we might assume that because the inauguration and women's march were a day apart, then the accounts would have been created at similar times. However this does not appear to be the case...
12/ If we 'zoom in' on the dates. We can see that the pro-Trump accounts (red graph) peaked on 21st January, while anti-Trump accounts (blue graph) peaked on 26th January. Again, if organic, does that suggest the anti-Trump accounts were due to the post-Inaugration + WMs
13/ backlash against Trump. Or was it people creating Twitter accounts to combat the increasing number of MAGA accounts? That would explain the lag. However, I still find it somewhat unusual that we see these different peaks - an interesting finding in itself...Buttt
14/ If we assume again, that many of the accounts are inorganic, i.e. trolls or maybe bots designed to promote polarization with a pro-Trump bias (as has been demonstrated before by the @DFRLab ) then that could also make sense. It would explain why there are a) more pro-Trump
15/ accounts, and b) the priority of getting the pro-Trump message out there first. It would also require some organization, presumably explaining why there was such a surge in January. (Some people have tried to attribute the January spike to a new year, people starting
16/ work again or becoming more engaged in current affairs but I still don't buy it). Of course it could be a mixture of both, large numbers of legitimate people signing up with trolls jumping on the bandwagon and inflating numbers. There are a few more interesting points...
17/ First for some reason I've been writing Trump won the election in December, when I meant November 2016 (Thanks @susiebillings ) for the reminder!
@susiebillings 18/ 44% of the pro-Trump accounts created since around 2007 were created after December 2016, while only 35% of the anti-Trump accounts created since 2007 were done so since December 2016. Of course that number is per sample, but the pro-Trump crowd seem to be more voluminous
@susiebillings 19/ Assuming my data gathering techniques have been relatively inclusive, and I have got a good spread of pro and anti-Trump accounts based on key phrase sampling, we can say that the pro-Trump crowd are winning at Twitter. Whether this reflects the 'offline world' is unclear
@susiebillings 20/ But January 2017 does require further scrutiny. Are we seeing mass troll and/or bot creation? It is highly plausible. @NetwarSystem has been gathering more stats on this. 200 accounts per hour were created during that key week in Jan 2017
@susiebillings @NetwarSystem 21/ On the other hand we know pro-Trump Twitter activists are highly vocal and tactical, as this amazing work on Trump trains by @3r1nG shows.
@susiebillings @NetwarSystem @3r1nG 22/ However, with extant evidence of Twitter manipulation by malicious actors we can make an assumption that the use of influence campaigns is highly probable. That is not to understate the existence of real people engaging in activism.
@susiebillings @NetwarSystem @3r1nG 23/ End #thread. Thanks for listening, feel free to share and ask questions! #disinformation
