Within three hours after polls closed on Sept 11, Singaporeans had a fairly reliable indicator of how votes went, thanks to sample counts released for the first time by the Elections Department (ELD).
These correctly indicated all the eventual winners, even in the close contest of Aljunied GRC, where there was a recount. The difference from the actual count was below two percentage points for all but one of the 29 constituencies.
Expect such counts to be the norm in future elections.
Mathematics aside, their success is clear from how many voters remarked online that they could have gone to bed without waiting up for the official tallies.
National University of Singapore economist Davin Chor, who has studied voting patterns in other countries, tells Insight: "The broad agreement between the sample counts and the eventual official results was reassuring that the methodology works."
Associate Professor Chor adds: "For future elections, the public will surely look to these sample counts as a reliable estimate of the likely outcome."
The results also achieved the ELD's goal of helping to prevent unnecessary speculation and reliance on unofficial sources of information while counting is under way.
Sample count results - drawn from 100 votes at each of the 832 polling stations, and weighted to account for the difference in the total number of votes cast at each - were released from about 9.40pm, with the last set coming in at 11.15pm. The first actual count result came in at around 11.30pm, and the last after 3am.
Sample count results were given in percentages, rounded off to the nearest whole number.
The GRC with the largest variance was Jurong, where the PAP's final vote share was 79.3 per cent, against the sample count's 78 per cent. Only in MacPherson SMC was the sample count result off by over 2 percentage points - at 2.58 percentage points. It showed PAP's Tin Pei Ling getting 63 per cent of the votes, but she actually got 65.58 per cent. Still, this was well within the stipulated margin of error of four percentage points.
As for why the sample counts seemed more on the mark for GRCs, Dr Walter Theseira of UniSIM College says this could be due to the larger sample volume taken in a GRC - where there are more polling stations - than in an SMC.
"The effect of tripling or quadrupling the sample volume is much greater, in terms of increasing the statistical accuracy of the sample count than any loss in accuracy from the increase in population size," he says.