Breaking into the Echo Chamber: what Twitter says about the UCU Ballot

While the role of Cambridge Analytica in the Brexit vote is in dispute, including whether their analysis of Facebook data actually provided any meaningful information, for the UCU ballot it’s twitter that’s been the social media platform of choice. Hashtags have shifted from #NoCapitulation to #NoDetriment, with #ReviseAnd Resubmit emerging and now #RejectUUKdeal used to publicly display the ballot paper.

Once balloting opened, twitter users have been declaring themselves as crossing the no box, and less commonly the yes. The errors in using this data to estimate the outcome are obvious: for example there are very few politically engaged STEM academics making declarations on twitter, despite their making up a much larger proportion of the membership. However, even without taking into account these biases, it is possible to track how the no vote has been shifting over the course of the ballot.

There is a public starting position, for at least a portion of the no vote, a UCULeft intiated #NoChange petition signed by 850 members. Of course this cannot be expected to include everyone who will vote no in the ballot: some will on the grounds of #NoDetriment, whether they signed the petition or not, but there are clearly many other reasons that members would want to reject the proposal. However, if the ballot is for reject, there will have to be many more votes than the 850 signing the petition — the turnout was over 20,000 in the intial vote to take strike action. Twitter gives a way of estimating how many more: if, and it’s a big if, reject voters are as likely to declare on twitter whether or not they signed the petition, we can estimate their voting numbers. It’s simply 850 multiplied by the ratio of all twitter-declared voters to twitter-declared #NoChange voters on the petition.

These declarations can be readily found, not by name but through hashtags. This gives a random sample of declarations that can be matched to the initial petition. I’ve been keeping a running total since the ballot started plotted below. I’ll not give a number before the ballot closes tomorrow, and I don’t believe it’s so reliable, but even without knowing the size of the no vote we can see how it has evolved. Also, from the breakdown of twitter declarations by branch it’s possible to see which branches have been most successful at building their vote beyond the #NoChange number on the petition.

How the no vote estimate has changed over the course of the ballot

I’ve marked the error bars on how the proportion of the final estimate has changed, and as the counts have increased through the ballot the errors have shrunk. What hasn’t changed has been the no vote, at least by this way of estimating it. Within the statistical accuracy possible, all the tweets, the three emails from the GS, the exchanges on the UCU activist list, the blog articles written by the score and read by the thousands, all of them have made no discernible difference to the number voting no during the course of this ballot.

It’s just possible countering effects have been playing off against each other. For example, as the ballot has proceeded those that signed the #NoChange petition might have increasingly decided to hashtag their vote, and this would have masked an increasing proportion not on the petition that hashtagged as soon as they voted: there are indeed tweets following each of the GS emails stating it to be the last straw to a previously undecided voter, though how undecided is rather doubtful if their name matches to the #NoChange petition.

But Occam’s razor suggests the most likely explanation is simply that after reading all we have, engaging on line and at branch meetings and most certainly feeling better informed, we have not substantially changed our position with regard to our vote. Of course this is just a measure from twitter, so it’s possible that those members off twitter, the majority, have shifted their positions more dramatically, but if so clearly twitter is neither the reason or gauge of those shifts.

Breaking it down to the branch level, we can see how far each branch has been successful in its efforts to build the votes for reject beyond the core members that signed the #NoChange petition. For each branch the number of declarations is low, but still there are clear trends visible, for example in how the branch recommendation attracted votersbeyond the core. I know that one branch committee has been agonising over whether to state a branch position, while others immediately communicated in the strongest terms that they called for a reject vote from their members.

The twitter declarations suggest the very worst way for a branch to encourage a no vote beyond their petition core was to state a branch position to reject. Kent gained no reject declarations on twitter beyond those on the petition, though they admittedly had by far the largest number of petition signatories of any branch. Kent therefore had a very strong core no vote, but one that didn’t appear to expand during the ballot. In contrast, the largest growth of reject declarations off the petition came from Cardiff, Oxford, York and QMUL, with only QMUL calling for a no vote. Overall there is a clear trend, that declaring for reject does not build the no vote; if a branch wanted to encourage an accept vote, the best approach would have been to call for a reject! Clearly those that work in universities do not like being told how to vote by their branch.

It also appears that those outside the core are also not encouraged to vote reject by branches that strongly criticise the General Secretary during the ballot. Ten branches signed a letter sent last Friday, very strongly questioning the conduct and implementation of this ballot. They were later joined by at least one additional branch. Those branches have seen their reject vote outside of the #NoChange petition core statistically significantly weaker than in general. Or course correlation doesn’t imply causality, but this result does suggest more care has to be taken before deciding how branches might make such criticisms in the middle of a ballot, if it is not to adversely affect their ability to build the vote.

We will know the actual ballot tomorrow, and I’ll post the data I used here a little afterwards, with names removed. I should emphasis I haven’t used any data that have not been posted publicly, either on the #NoChange petition or as voting hashtags.

However, whatever the results tomorrow, there are some important conclusions to be made. First of all, we shouldn’t flatter ourselves that we have influenced the actual vote with a few well chosen tweets and incisive blog entries. Providing information to us voters is one thing, but it looks like once we view that information as propaganda, it has no or even a negative effect on our voting. Secondly, branches might want to consider refraining from giving their members a recommendation to vote, or at least not agonise over doing it. Lastly, strong internal criticism within the union during the ballot has not gone down well in terms of building the reject vote within those critical branches, even if the criticism might be welcomed by their core supporters.

If this analysis is correct, it suggests that a less aggressive approach to the ballot might have been both more persuasive as well as sometimes less unpleasant. I would suggest it might increase turnout as well. Whatever the ballot result, we need to all work together after this, not continue the divisions we’ve seen during this ballot, and of course we need to concentrate on the major job in hand, to protect our pensions.