One of the biggest surprises of the 2015 UK general election was not the Conservative Party’s victory, but rather the gulf in seats won between them and their nearest rivals. This surprise was exacerbated by the extent to which the result varied from virtually all pre-election opinion polls, which had forecast a slim victory at best, and in many cases a tie. The chart below shows the share of total votes predicted by the final pre-election poll, for the 4 parties that would go on to win the greatest number of seats:
The British Polling Council’s plans to examine an apparent bias in polls that generated this disparity between the industry’s final snapshots of opinion and the election result (shown below) are a reminder that reported or planned behaviour has some inherent weaknesses versus passively observed, market-level data.
With many reasons cited as having potentially generated such a difference, including actual behaviours differing from reported ones due to ‘shyness’ from certain sections of the electorate, the question remains whether observed behaviour might have proved a more reliable barometer for voting intentions.
Examining users’ consumption of these 4 leading parties’ digital properties delivers some insights that could make interesting reading for any organisation with an interest in gauging consumer opinion or behaviour. Fundamentally, consumption can be divided into the spheres of scale (how many people go to a property), and engagement (how often do they go / how long do they stay / how much content do they consume).
Taking one metric from each shows that in this instance, engagement would have been the far more indicative metric. If you’d examined only the scale of the audiences in the month leading up to the election (April 2015) for these four properties, the results would have suggested a landslide victory for Labour:
The fact that Labour commands a larger audience overall makes the engagement results even more surprising – despite a significantly greater number of users, the total number of pages consumed on the Conservatives’ website surpasses all other parties. Not only that, but the correlation between the total pages viewed and total votes received in the election itself are a remarkable 0.99.
In electoral terms, it might be possible to hypothesize that scale is more aligned with an expression of political preferences, but that engagement demonstrates a greater affiliation to turnout, which proved to be a decisive issue for constituencies where seats considered ‘safe’ by weight of population proved not to be in terms of supporters turning out to vote.
With so many variables in play (not least that total votes do not equal total seats in a first-past-the-post system), and the very basic scope of this analysis, it would be impossible to suggest that this was a reliable means of anticipating election results. It is, however, a reminder that reported and actual behaviour do not always align as closely as one might expect.
With consumers engaging in more activities and time on digital platforms, the reliability of what they recall having done becomes even more prone to distortion. Combining accurate and granular consumption data with panel-observed demographics and behaviours remains an effective means of capturing the total scope of consumers’ digital activity.