In the run-up to the launch of our new platform, we thought it would be interesting to create an Election-specific version. We’ve been collecting and analysing data from a number of sources and the results have been fascinating. In this post I’m going to address “Is there data on Twitter that correlates with current polling predictions?”
For every candidate standing in the UK general election 2015 that has a Twitter account with more than 50 tweets we gathered:
We used Predicted Share of Vote consistently for our x-axis and produced 3 charts comparing it to each of the Twitter datasets. We have assumed that, especially with the ‘non-celebrity’ candidates, that the majority of people who follow a candidate are from that constituency and would consider voting for them.
We will use a statistical measure (the Correlation Coefficient: r) to quantify the dependence between each set of variables. This is a number in the range -1 to +1 inclusively, where -1 is total negative correlation, +1 is total positive correlation and 0 is no correlation. See here for a deeper explanation.
As we can see from the chart, there is seemingly no correlation between election prediction and the proportion of time that a candidate spends engaging their constituents in conversation. With a Correlation Coefficient of -0.11 I might even suggest that candidates that do reply to constituents tend to make matters worse.
Now that’s far more interesting. There does seem to be a visual correlation (from bottom left to top right) between election performance and what proportion of a candidates tweets get retweeted by others. With a Correlation Coefficient of 0.50, this is showing a moderate positive relationship.
A very similar visual correlation with favourites as with retweets. It also seems that election performance and what proportion of a candidates tweets get favourited by others are related. With a Correlation Coefficient of 0.48, this is also showing a moderate positive relationship.
If we were to look at individual parties for evidence of the same correlations between votes and retweets or favourites, do we see the same patterns?
I think we can both agree that the Conservative party continues the correlation. r is 0.57 for retweets and 0.49 for favourites.
And again, the Labour party continues the correlation. r is 0.58 for retweets and 0.59 for favourites.
However, these 2 make up the majority of candidates. So, what if we look at a smaller party?
Interesting, this is the only party with negative r for replies, retweets and favourites. r is -0.40 for replies, -0.10 for retweets and -0.18 for favourites.
It seems clear that a candidate’s openness to communicating with potential voters has no impact on their projected performance in the election. But why? Possibly we, the voters, aren’t using Twitter yet to communicate with our candidates. Perhaps candidates are far more interested in towing the party line and just broadcasting this week’s message plan? Either way, it does seem to show that ignoring voters is not hamstringing candidates!
But what of “cause and effect” regarding the proportion of a candidate’s tweets that get either retweeted or favourited by others? Well, given our initial assumptions, we could assume that this is a measure of how well a candidate taps in to what their electorate are concerned about. If nobody cares about what they’re saying, nothing gets shared. If, however, a candidate is pushing the right buttons, they will see increased shares and favourites. It’s then not unreasonable to suggest that their facade on Twitter is a mirror of how they talk to their electorate through other channels.
All of our constituency and candidate analysis is available, for free, online at election.awedience.com. Please have a look and share it with all your friends!
Does this mean that Awedience’s Twitter measures can actually be used as an indicator of polling results? We’re going to update our analysis daily and see how it changes over time. Maybe we can even predict the results :).