Social Media Monitoring Can Inform Political Parties What To Poll For

by politicalthing

This blog was originally authored by me and published on the Radian6 website on July 20, 2012. I’m posting it here for posterity.

On Wednesday, I looked at the question, “Can social media be used for political polling?” In the end, my answer was no. The most convincing evidence I found against its use was in the scientific methodology pollsters use to ensure the accuracy and integrity of the results.

I briefly discussed some industry-specific reasons social media monitoring could not be used for political polling and hinted that there was still a crucial role social media monitoring could play in polling strategy.  This is the subject of today’s blog.

A Bit of Background

In the last Canadian general election, Twitter featured prominently for the first time. All of the political parties had a presence on Twitter, as did the leaders. News organizations published stories to the web that were instantly shared. It inspired many to refer to the 41st general election as ‘#elxn41’ and “Canada’s first social media election“.

The Conservative Party of Canada, the governing party, and the Liberal Party of Canada, the Official Opposition, had been the only two parties to govern Canada in its history. The New Democratic Party of Canada (NDP), formed in 1961, had never formed the official opposition in Ottawa. The Green Party of Canada had never successfully elected a member of parliament (MP) and their leader, Elizabeth May, had already lost one bid for a seat in the previous general election. Finally, the Bloc Quebecois, a separatist party that only runs candidates in the province of Quebec, ran candidates again.

Conventional wisdom at the outset said that nothing would change. The Conservatives were expected to hold on to power. Whether they would win a majority government depended – it was thought – on how well the Liberal Party did. The NDP was expected to finish third, and the jury was out on how the Green Party would do.

Conventional wisdom, as it turned out, would be mostly wrong.

In the end, the Conservatives did hold on to power and won their first majority government in 23 years. For the first time in Canadian history, the NDP became the country’s Official Opposition, the Liberals were reduced to third place in Parliament, and the Green Party’s first MP, their party’s leader, was elected.

Now that you have some historical and political context, let’s break down the social and polling data to understand the place of social media monitoring in political strategy.

Political Conversations on Social Media are Not Representative

We know how the election turned out in reality.  Nanos Research published a tracking poll three days prior to the May 2nd, 2011 election day showing that these results were likely.

This Nanos tracking poll reflects the actual make up of the House of Commons after the election. Data gleaned from social media monitoring paints a different picture.

Monitoring all mentions of the individual parties’ hashtags with the #elxn41 Twitter hashtag, which became the center of English language political discussion (Francophones in Quebec used a separate hashtag to discuss the election), shows that the NDP would have formed the government in a landslide victory.

The problem is that the trend graph reflects the total number of mentions each hashtag received; it does not take into account voter intention. The pie chart below shows the share of voice each party enjoyed on social media.

The Canadian House of Commons is comprised of 308 seats. Had social media correctly predicted the outcome of the election using the cumulative data available, the NDP would have formed a minority government with 100 seats. The Conservatives would be Canada’s official opposition with 95 seats. The Liberals would have placed third with 78 seats, the Green Party would have 29 seats and the Bloc would have finished with six seats.

Of course, it didn’t turn out this way.

Early Warning Signals

While I hold firm to my position that social media monitoring cannot be used for political polling, I also maintain that there is an important place for it within the industry. Social media monitoring allows political strategists to see what issues are hiding just below the surface and can help inform what they should be polling for.

As the election got underway, the media began focusing more on the race between the Liberal Party and the NDP. This narrative wasn’t supported by traditional polling (see the Nanos poll above) that showed the NDP had dropped in approval ratings since the first day of the campaign.

The social data tells a different tale. If we isolate the data on the trend graph to show only the Liberals and the NDP, we see their share of voice was extremely close throughout the election.

In fact, the Liberals and the NDP battled it out for the lead on a near daily basis until April 21st, when Canada’s national media reported that polls showed the NDP over-taking the Liberals.

What Does It Mean?

The social data was telling a story that was not being picked up by the national polls. This is not a fault of the polling agencies. There is little in the social data that would give any reliable indication regarding voter intention.

The social data in this trend line tells us only that both parties were being talked about roughly the same. When compared to the national polling data, it tells us that something is going on, not what exactly, but enough to commission a poll to find out if voting intention is shifting.  In retrospect, this was what was happening, as the April 21st poll revealed, but conventional wisdom said it would never happen.

These last two blog posts have been longer then normal. I appreciate you reading and please feel free to leave your thoughts in the comment section.