Sample counts proved accurate in the 2015 general election, with actual results coming well within the error margin around what was predicted, especially for larger wards.
This year marked the first time the sample counts - an early indication of the possible result for each electoral division - were made public for each of the 16 group representation constituencies (GRCs) and 13 single-member constituencies (SMCs).
One reason for this move is to prevent unnecessary speculation and reliance on unofficial sources of information before all the votes are tallied and the final results are announced.
This year, the difference between sample results and actual results ranged between zero to about three percentage points. Most constituencies and wards reported results within one percentage point of the sample count result.
The first sample counts came in just before 10pm, two hours after the close of polls at 8pm. By 11.15pm, the last sample counts had been given to the media by the Elections Department (ELD) - giving an early hint of the national swing towards the People's Action Party (PAP). Actual results started streaming in, meanwhile, closer to midnight.
Sample counts have a 95 per cent chance that they will be within four percentage points of the actual count. To conduct the sample count, a hundred ballot papers from each polling station were randomly chosen by a counting assistant in the presence of the candidates and their counting agents.
The number of polling stations vary across SMCs and GRCs. Given the number of polling stations for an SMC or GRC, typical sample sizes are around 1,000 for SMCs and a few thousand for GRCs. The votes were added up and weighted accordingly to account for the difference in the number of votes cast at each polling station.
The single ward with the biggest difference between actual and sample counts was MacPherson. There, the PAP's final vote share - 65.6 per cent - ended up 2.6 percentage points higher than the 63 per cent it was originally reported to have. Sample counts were more accurate for GRCs, possibly due to the bigger sample sizes collected. The GRC with the biggest variance was Jurong GRC. The PAP's final vote share there - 79.3 per cent - was 1.3 percentage points higher than the sample count's 78 per cent.
This article was first published on September 12, 2015.
Get The Business Times for more stories.