By Martha Bozic
In the run-up to the 2016 American presidential election, predictions made from official statistics were taken, used and thrown away, with little apparent meaning.
In the aftermath, one of those lesser cited was a prediction that campaign spending on television broadcasts alone would amount to more than 4.4 billion dollars (£3.3 billion). Previously, the 2012 Obama campaign spent 65% of their entire budget on media.
Unlike many other projections, the final figures suggest that this one wasn’t too far from the truth, and such record level expenses underline a widely held belief, that running a pricey campaign can have a real impact on the results.
In many recent elections, high levels of spending have been called out as concerning, with criticism focussing on the involvement of sometimes dubious donors, and the potential for their money to influence the floating voter.
With this in mind, swaths of research have looked at the costs of campaigning during elections, and even the most basic manipulation of open data has shown correlations between votes won and a greater amount of money spent by the winning candidate.
A new study, entitled ‘The Price of a Vote’, has attempted to simulate the impact of increased spending on the outcome of an election. By analysing data sets from ‘proportional elections in Brazilian states’, researchers have created a mathematical model which behaves in a way strikingly similar to real voters.
To demonstrate the validity of their model, the team also used it to predict turnout in their simulated election and found their results to again be compatible with real-world data.
Although to begin with a linear relationship was found between spending and votes won, above a maximally efficient budget the model suggested that more money must be spent for each new voter gained. This demonstrates a phenomenon known in economics as the ‘diseconomy of scale’ – in other words, the more you spend, the more expensive something becomes.
In the study, votes were viewed as products, and assumed to be ‘buyable’, either directly by door-to-door campaigning, or indirectly, by advertising. (This may be an unsavoury way to view a democracy, but it certainly appears to be a realistic one.)
Despite the non-linearity at the top, the researchers cited ‘a consistent non-trivial dependence of votes on money spent’, confidently corroborating previous observations, and the instinctive view of many.
Additionally, the team believe that the diseconomy of scale can be explained by the competition between candidates: at some point potential voters reached by the campaign will be less easily won over, if at all.
They also suggest that the simplicity of the assumptions made in their model is likely to have played a part in its non-linearity. While the study considered the reach of both traditional and social media, the researchers admit that they have not looked at how other social or political issues may affect the results of an election.
Instead, they leave this as a task for follow up research projects, which they hope will provide more detailed insight into the price of a vote – a structure far bigger than the data set they originally started out with.
It is a scary prospect, but eventually, we may have a model which can accurately predict the results of an election, not just how much it will cost.
Photograph: Tim Evanson via Wikimedia Commons