This article was originally published on September 11, 2015 for GE2015 and has been updated on July 10, 2020, to provide more information for GE2020
For only the second time in Singapore's history, sample votes will be revealed by the Elections Department to the public even before the final results are tabulated.
This means, from 10pm or earlier, Singaporeans will start receiving early indications of how the results are faring based on sample size.
How is the sample vote taken?
Very simply, at each polling station, a counting assistant will randomly pick a bundle of 100 ballot papers and count the number of votes for each candidate(s).
Since each polling station will have a different number of total ballot votes, different weightage will be assigned to each station to account for the difference in votes cast.
For example, if there is a total of 100,000 votes cast in a GRC and a certain polling station in the constituency has 3,000 votes, the weightage given to that station will be 3 per cent.
The various polling stations across the constituency will each have their weighted scores tabulated and added together.
How accurate will the sample count be?
The sample counts are "fairly indicative of the final percentages" with a confidence level of about 95 per cent, plus or minus 4 per cent points.
What this means is that there is a 95 per cent chance that the actual election results will be within a 4 per cent margin of the sample count.
It implies that any indicative results announced that falls between the 46 per cent to 54 per cent range are far from a done deal since there is a 4 per cent margin in error.
For example, and statistically speaking, candidates who win 60 per cent of their sample votes have a 95 per cent assurance that the final result will win between 56 per cent to 64 per cent.
Candidates that win 52 per cent of the sample votes have a 95 per cent probability of accumulating between 48 per cent and 56 per cent of the votes cast, so it would be fair to say that the contest is far from over.
Likewise, a candidate winning 46 per cent of the sample vote may still have a chance, statistically speaking, to win the actual vote.
According to a Mothership article, in GE2015, sample vote counts in all constituencies except MacPherson came with 2 percentage points of the official results, below the 4 per cent margin of error.
This is an indicator that the sample vote count is going to be a good proxy for the actual results. For MacPherson, it was 2.58 per cent.
It's worth noting that for constituencies with a very close fight, the sample vote count will not be a good proxy for the actual result.
It's worth noting that in GE2015, only two constituencies had a winning margin of less than 4 per cent (i.e. result may not be accurately predicted by the sample votes).
These were Aljunied GRC (won by WP at 50.96 per cent) and Punggol East (won by PAP at 51.77 per cent). Aljunied GRC did a recount since the winning percentage was lower than 2 per cent.
Automatic recount if winning margin is less than 2 per cent
For GE2020, there will be an automatic recount if the winning margin is less than 2 per cent. This is a change as compared to GE2015, where the candidate(s) have the option for a recount if the margin is less than 2 per cent.
For example, in GE2015, because the winning margin of the vote for Aljunied GRC was less than 2 per cent, the PAP candidates requested a recount of votes (this indicated that they lost the first count). As such, a recount was ordered by the Election Department.
Subsequently, the WP still won the vote with the final count being 50.96 per cent (WP) and 49.04 per cent (PAP).
In GE2020, a recount will be automatic as long as the winning margin is less than 2 per cent. Losing candidate(s) do not need to request for it.
For those who will need to get up earlier the next day, the sample count will allow you to get an early indication of the results.
For the rest of us, we will only know when the results are finally announced. However, if a recount is needed, be prepared for a late-night if you want to wait for the actual results.
This article was first published in Dollars and Sense.