The ‘24 election won’t be won by a tweet, but here’s some data anyway

Here’s a thing:

The above charts are from a JPMorgan Data Assets and Alpha Group analysis of how the launch of general election manifestos played out on social media. The idea is apparently “to better appreciate how [tweets] could move opinion polls”.

Its main conclusion is that four weeks into the campaigns, the only online subsection of the electorate getting exercised about stuff is the Faragists.

JPMorgan tells clients:

In general, it appears televised debates generated the highest level of engagement, with manifestos publication hardly comparable.

However, attention seems to be fading as the peaks reached in early June have not been matched again.

The only party able to challenge this trend was Reform UK, whose “Contract” announcement has generated significant levels of social media activity.

Tweet volumes support the possibility that Reform UK will overtake the Conservatives by share of the vote, the bank concludes:

Based on our previous analysis, an increase in the weekly ratio of tweets by a factor of 1 leads to a poll increases of 0.58 points.

While there is no sufficient data to test this model in a multi-party scenario, it can be noted that Farage has been recently outperforming Sunak by a factor of 2x/3x and Starmer by a factor of 1.5x/2x.

Layering these considerations with the momentum Farage is enjoying in polls, Reform UK could be overtaking the Conservatives party in the near future, while Labour’s lead seems harder to challenge.

Might it be useful to see tweet ratios overlaid on poll data? Whichever way, you’re going to:

An obvious question to ask relates to whether tweets are still representative of anything.

Traffic tracking agencies like Sensor Tower and SimilarWeb say X usage plummeted following Elon Musk’s Twitter takeover in October 2022 (data the rebranded company has sort-of denied) and the few financials to be made public have looked grim. Bloomberg reported this week that X generated $1.48bn in revenue in the first six months of 2023, down nearly 40 per cent year-on-year, and that it lost $456mn in the first quarter last year.

Tweet quality presents just as big a problem, with AI bots flooding the channel with hard-to-censor crap. “Are fake accounts swaying voters towards Reform UK?” asked the BBC last week, to a cacophony of protest from people who may or may not be fake, which is how these things have been playing out since at least 2016.

What’s not proven is whether the shoutiest accounts on social media are most commonly persuasive, repulsive, or irrelevant to the majority.

This NBER working paper offers a good summary of recent literature as it moves towards a conclusion that, in the 2016 and 2020 US presidential elections, Twitter users were less, rather than more, likely to hold polarised views. That might be because its base skews young though, relative both to the wider electorate and the Facebook average.

Mapping political demographics across the increasingly dispersed social media landscape would take a Venn diagram that’s beyond our abilities to create, and is probably beyond the remit of JPMorgan’s Data Assets and Alpha Group, whose work on the theme so far seems most useful for finding alpha in the vote-share novelty betting markets.

Still, it keeps them busy. That’s something.