We assembled a set of 22801430 tweets (8307314 excluding RTs) containing "Brexit" from1/1/2015 to present. This isn't the entire volume (occasional gaps of a few hours), but is over half and we believe it to be representative. Accounts in the legend are the 11 with the most RTs. 
Here are the top 20 accounts (in terms of number of times retweeted) both before and after the day of the Brexit vote. A quick gander at the these two lists suggests that the discussion prior to the vote was mostly driven by pro-Brexit accounts and became more balanced after.
There are 1193067 accounts with Brexit tweets in this dataset. It's obviously not practical to manually categorize them all by stance on Brexit, but it's possible to come up with a reasonable estimate by looking at a statistically significant subset. How?
1) Inspect the 200 most frequently RT'd accounts; classify them by stance on Brexit.
2) Download tweets for a sample of the overall set of accounts (used 6000, margin of error is 1.2%).
3) Classify those accounts based on the stance of the top accounts they most frequently RT'd.
2) Download tweets for a sample of the overall set of accounts (used 6000, margin of error is 1.2%).
3) Classify those accounts based on the stance of the top accounts they most frequently RT'd.
This chart shows the estimated proportion of the Brexit traffic attributed to each stance. Note that Leave almost completely controls the discussion until after the vote. The discussion flips decidedly in the pro-remain direction around August/September 2017.
Here's a network diagram constructed from a sample of 103151 tweets from the month immediately preceding Brexit. The largest node is DavidJo52951945 (now @DavidJoBrexit), which we previously identified as a Kremlin propaganda account in late August 2017.
Previous thread examining DavidJo52951945/DavidJoBrexit here:
Another thing worth looking for is bots. This chart shows the proportion of traffic from accounts with varying characteristics more common among bots than human-operated accounts. Of particular interest is the higher rate of activity from 24/7 accounts in June and July 2017.
Let's take a look at a couple of examples of 24/7 accounts tweeting about Brexit around the time of the referendum. First up: @SurviveEnd. This one has a heavy conspiracy theory theme. It stopped tweeting in November 2016, but was quite active prior to the vote.
One more: @piohifelne. In the days leading up to the Brexit vote, it tweeted about Brexit in English. Interestingly, it has now transitioned to Russian.
Conclusion: between a Kremlin agitprop account with 100K followers, various bots, and whatever we didn't notice, influence operations drove much of the pre-vote Brexit Twitter discussion. If social media affects voting decisions, these operations were relevant to the outcome.
Much appreciation to @ZellaQuixote for various assistance with this thread, including but not limited to the willingness to sanity check the methodology.
