Can we predict the results of an election using science?

Is it ever possible to accurately predict an election result using maths and science?

Elections and their outcomes can have major consequences on society for better or for worse. A new government can bring major changes and reforms to a country in a relatively short space of time. It is therefore unsurprising that people wish to predict the winner before an election has happened. For the normal voter, predictions provide clarity on the current political climate and may inform voting while for politicos, predictions are their life blood providing hours of entertainment. Statistical analysis is key in predicting the results of elections, however it is a difficult task, especially in an Irish context.

The most straight forward and widely used predicting tool is polling. Polling works on the basis that by recruiting a sample from the population and we can make inferences to apply to the wider population. Arguably, in the absence of an exciting general election campaign this year, the media turned to polls to fill column inches with polling results and made stories on the basis of these polls.

Polling is not a perfect method by any means. By its very nature it relies on sampling a very large population. Polling companies have become adept at making their results as representative as possible, adjusting for locality, socioeconomic status and age, but it remains difficult to create a perfect representative sample. Even if we can be certain that populations are representative, with sampling that there is an inherent margin of error in the results – headlines during the election campaign often showed parties rise or fall by 1 or 2%, but in reality these changes are not statistically significant.

Polling has great difficulty in predicting who will actually show up to vote. While all those polled are asked about who they are voting for, it can be difficult to predict whether this will translate into actual votes on the day. Turnout in Irish general elections historically approximates 70%. Polling companies attempt to overcome this flaw by asking people how likely they are to vote on a scale of 1 to 10, but how this answer to this question is used is unclear.

Previous to the 2015 United Kingdom general election, Red C, one of Ireland’s best known polling companies used a cut-off of 4/10 to decide whether an answer should be counted but they changed this to 6/10 following concerns in polling in Britain. This may be an incorrect assumption for Irish elections.

Polling assumes that voters are being truthful – this is not always the case! The phenomenon of “shy voters” was apparent following results in last year’s British elections. Polling in the lead up to the election showed that the Conservatives were behind Labour only for the opposite to occur on election day.

While some of this was attributable to poor predictions in polling of who would turn up to vote, it was also noted that some voters were shy about telling pollsters they were voting Conservative. In the last general election and local elections, similar trends were noted with Fianna Fáil, however it is unclear to what extent this can be attributed to the “shy voter” phenomenon.

Polling suffers from a number of unique flaws in an Irish context. Proportional representation (single transferable vote) or PR-STV is a complex system and the percentages quoted in polls rarely translate directly to seat numbers. Transfers from poll-toppers and weak candidates are often crucial in deciding the fate of many of the final seats across the country.

These are not easily predicted from national polls even where second preferences are examined, as local factors contribute hugely to transfers in addition to political ideologies. It is not unheard of for left-wing candidates to transfer to the opposite spectrum if the two candidates share similar “home turf”.

While polling may provide theatrics for politicos and the media alike, it is not perfect. Some have attempted to improve on polling. “Poll of polls” examining polls averaging polls over an extended period of time have become more widely used and arguably provide more worthwhile information allowing for once off blips to be ignored and trends examined, although combining different polls with different methodologies as is sometimes done is not ideal either.

Statistical models are becoming more commonly used and may provide more accurate methods of predicting results. Internationally, FiveThirtyEight in the United States of America and Election Calculus in the United Kingdom have been developed to predict results in “first-past-the-post” systems. Statistical models applied to the more complicated Irish electoral system have now become available.

Trinity graduate David Higgins has developed, a website that employs a statistical model to predict results at a constituency level. He uses a combination of previous general election results, local election results and current political polling to predict outcomes using statistical algorithms.

The statistical analysis used is quite advanced and relies heavily on quantitative methods. Rather than provide definitive answers, the results are presented as probabilities. This is useful as many would agree that at a constituency level nothing is absolutely certain until after votes have been cast – anything could happen, but admittedly some outcomes are more probable than others.

Of course, the most unscientific of methods – asking people on the street to predict the result of the election – seems like a bad way to predict an election intuitively. However, Red C believe that this may be of use in ascertaining the levels of “shy voters”. They have christened the idea the “Wisdom of Crowds”.

Crowds were very wise during the marriage equality referendum – while most polling predicted roughly a 70-30 split, a result which was outside the margin of error compared to the result on the day. The “crowd” on the other hand predicted the result exactly. It remains to be seen whether this will translate from a referendum to a general election in Ireland.

Despite the various attempts to predict the results of an election, only one poll can be counted on to illuminate the winners of an election – the actual ballot!