We've been tracking social media activity around the #migrantcaravan currently making its way north through Mexico. We analyzed over 2 million tweets from Oct 23–26. Here's what we found... 
There is a significant imbalance in user activity. 1% of accounts are responsible for 18% of the tweets, 5% are responsible for 44%, and 10% are responsible for 57% of the tweets. We've seen similar proportions around live, politicized events before, but it is still striking.
This imbalance leads to misleading trends in the overall data. Scrolling through the tweets would give you the impression that most are opposed to the refugees entering the US. But that's not necessarily the case.
The most retweeted posts about the caravan come almost exclusively from the political right — both real and fake accounts. The same goes for the articles and websites most linked to and the hashtags most used. BUT that's without correcting for bots and other high-volume users.
When we compare the most active 1% of accounts with the least active 50% of accounts (roughly the same amount of content), we see significant differences.
The highest-volume accounts (likely including many bots) tend to link to right-wing sites and a few conspiracy theory sites. The low-volume accounts, however, link to more mainstream media, social media, and a few left-leaning sites.
Likewise, the hashtags most representative of high-volume accounts tend to be right-leaning: critical of the left and of the migrants. The hashtags distinctive of the more organic, low-volume accounts are more generally descriptive, and more likely to include Spanish content.
(The unfamiliar hashtags in these graphs tend to be associated with tweets about a video game website that just happen to contain some of the same search terms we used.)
We are also seeing *very preliminary* signs of coordinated manipulation around some of the hashtags distinctive of high-volume and bot accounts.
With the election just days away, it's no surprise that the issue of immigration is being politicized and amplified, including on social media. But it's important to note, in this case, the vast difference between the majority of *tweets* and the majority of *users*.
When such a difference exists, we must be careful how we measure public sentiment, and mindful of how perceived general trends influence our own opinions. At this stage in the election news cycle, any misunderstanding or misrepresentation can have significant consequences.
