I had an interesting conversation earlier this week. Someone challenged me on my very public experiment suggesting the outcome of the vote could be influenced by anyone who wished to elect a certain party or leader. The individual claimed a campaign, for instance, could pump out tweets mentioning their party’s leader to drive up Twitter mentions. Et voila! Victory.

It’s true anyone could do that. It would definitely increase the number of mentions and possibly the visibility of the party and leader. However, it would only serve to skew the numbers on Twitter and could certainly play havoc with the legitimacy of Twitter to predict an election (if such a legitimacy actually exists in Canada’s political system).

The purpose of this experiment it to test a theory of correlation of Tweets to election outcome. This is about correlation, not causation.

This post covers the fourth and final full week of testing the Twitter-election-prediction theory using the Nova Scotia provincial election as a test lab.

Election day is Tuesday. My company, Full Duplex, will be conduction election day and election results analysis from the digital point of view on Tuesday. You can follow @FullDuplexCA and @MarkBlevis for updates.

Analysis performed using Marketwired/Sysomos Heartbeat and Marketwired/Sysomos MAP.

Election prediction by leader mentions

There was essentially no change in the distribution of mentions by leader this week over last. The only notable change was 1% of the mentions shifted from PC leader Jamie Baillie to NDP leader John Percy.


#NSvotes tweet mentions of party leaders for the week of September 29-October 5

After the campaign heads into the homestretch for Tuesday’s vote, Darrell Dexter maintains his overall lead of 40% of the mentions ahead of Stephen McNeil’s 36%.


#NSvotes tweet mentions of party leaders for the first four weeks of the campaign

I decided to conduct an extra analysis this week to see how the numbers shake out for exclusive mentions.

The analysis to date has been driven by the theory developed by University of Indiana researchers that Twitter’s ability to predict is based on mentions alone, regardless of sentiment. This includes mentions of multiple leaders in a single Tweet. That is, any mention of Darrell Dexter and Stephen McNeil in a single tweet counts in both of their columns.

By way of comparison, I thought I’d analyze exclusive mentions — only counting tweets which mention a single leader. That is, a mention of Darrell Dexter only counts if no other leaders are mentioned in the same tweet.


#NSvotes exclusive tweet mentions of party leaders for the first four weeks of the campaign

Election prediction by party mentions

The Liberals gained up 2% to 33% and the NDP gained 1% to 48%, picking up the mention losses of the PCs and Greens.


#NSvotes tweet mentions of parties for the week of September 29-October 5

The running totals for the parties changed too little to merit mention this week.


#NSvotes tweet mentions of parties for the first four weeks of the campaign

Top issues

Healthcare continued to pick up mentions to further establish itself as the dominant election issue based on election Twitter activity. Rounding out the top three issues is energy and taxes.

Note that this week’s graph features only the top six issues rather than the top 10. This decision was made due to the cluttered appearance of the graph when it displayed the top 10 issues.


#NSvotes top six election issues based on keyword analysis of tweet content

Watch for ongoing election analysis here and on the Full Duplex blog.

Analysis performed using Marketwired/Sysomos Heartbeat and Marketwired/Sysomos MAP.