In terms of projecting the final result, our model was quite accurate from the beginning. We projected that Moon always had a higher chance of victory, that grew greatly once the campaign went on. Although there was a period where the probabilities were close, our final predictions were a resounding success. This, however, was quite evident in the final days – everyone would have projected that the candidate with such a high lead would emerge victorious.
Overall, our polling averages were extremely accurate for predicting the final polling results of Moon-Jae-in and Ahn Cheol-soo. Furthermore, our predictions for Sim Sang-jung and Yoo Seong-min were within the margin of error. The average polling error of 2.4% was within reason – it was similar to the polling errors in many situations.
|Results||Sim Sang-jung (Justice)||Moon Jae-in (Democratic)||Ahn Cheol-soo (People’s)||Yoo Seong-min (Bareun)||Hong Jun-pyo (Liberty)|
However, unfortunately our projections for Hong Jun-pyo were quite far off the mark – all the polls at the time had underestimated his strength. Our projection also failed to estimate that he would come in second place, instead believing that Ahn Cheol-soo would do so instead. This is largely due to a failure to estimate the momentum of Hong Jun-pyo – was not assumed that the 9.5% of undecided or minor-party voters would back Hong Jun-pyo.
There were several possible reasons for this error. Firstly, the 5-day gap in polling between the final published polls and the actual results meant that Hong’s momentum in the last few days (see the chart) was underestimated and not taken into account. Secondly, the large amount of minor-party or undecided voters meant the race would be more volatile inherently, and combined with Hong’s momentum, this may have persuaded significant amounts of voters to move to Hong in the last few days.
In the grand scheme of things, this differential mattered very little – the placing of the third and the second place don’t erase the fact that Moon won an overwhelming victory. But the verdict shows how inaccurate polls can be, and how one should be wary when assigning a result as “definite”.