hankyoreh

Illustration by Jaewoogy.com

The Literary Digest’s 1936 poll that inaccurately predicted the result of the US presidential election that year is considered one of the most unsuccessful cases of election polling in history.

The magazine, which had correctly predicted the results of US presidential elections since 1916, forecast that Alfred Landon of the Republican Party would be elected president with 57% of votes against Democrat Franklin Roosevelt, who would only garner 43% of votes. However, actual election results proved the exact opposite. Roosevelt was reelected to the presidency with 60.8% of votes, and the disastrous failure of its poll ultimately drove the publication into the ground.

The Literary Digest had obtained the addresses of over 10 million individuals from motor vehicle registers and phone books, sending them virtual ballots from Sept. 6 to Oct. 31 to eventually receive 2.36 million responses. In contrast, Gallup selected a sample pool of 50,000 individuals and received responses from 1,500, going on to accurately predict that Roosevelt would win the election. This shows that sample size, beyond a certain level, is not a core variable deciding the accuracy of polls.

According to a 1988 study by Peverill Squire, The Literary Digest systematically excluded from its poll voters of low socioeconomic background who tended to be avid supporters of Roosevelt, skewing its sample; but more importantly, it failed to account for the fact that Landon supporters more actively participated in its poll than voters in favor of Roosevelt. According to Gallup’s post-election survey, of the voters who received a virtual ballot from The Literary Digest, 55% were Roosevelt supporters while 44% supported Landon. This substantially deviated from the response distribution of the magazine’s poll.

The poll results for South Korea’s upcoming presidential election seem to be all over the place. Even polls announced on the same day or just a day apart show drastically different results beyond the margin of error.

For example, while Gallup Korea’s survey conducted between Jan. 11 and 13 showed the Democratic Party’s Lee Jae-myung polling at 37% and the People Power Party’s Yoon Suk-yeol lagging at 31%, Realmeter’s survey commissioned by OhmyNews conducted from Jan. 9 to 14 showed the opposite, with Lee polling at 36.7% while Yoon led the match at 40.6%.

Likewise, the Korea Society Opinion Institute’s survey from Jan. 14 to 15 showed Yoon polling at 40.6% while Lee trailed behind at 36.2%, but Embrain Public’s survey commissioned by News1 Korea conducted between Jan. 16 and 17 showed Lee in the lead with 35.6% of support while Yoon fell behind with 34.4% of support.

Could public opinion really be so fickle? Rather, paying close attention to the participation rate of voters with different political leanings can elucidate the puzzling fluctuation in poll results. Polling companies collect their samples with sex, age and regional voter distribution in mind. In spite of this, the political orientation of such participants greatly varies from survey to survey.

Of the sample pool for the Gallup Korea survey that showed Lee polling at 37% and Yoon at 31%, 25.1% identified as conservatives and 24.0% identified as liberals, the percentage of conservative voters marginally higher than that of liberals by only 1.1 percentage points.

In contrast, of the sample pool for the Realmeter survey that showed Lee polling at 36.7% and Yoon at 40.6%, 28.4% identified as conservatives while 21.2% identified as liberals, the percentage gap between the two voter groups much higher at 7.2 percentage points.

The Korea Society Opinion Institute also surveyed a higher proportion of conservative voters, the percentage gap between the number of conservative and liberal voters at a staggering 10.8 percentage points.

Assuming that 60% of liberals support Lee while 60% of conservatives favor Yoon, one can conclude that with every percentage point advantage one voter group has over the other in sample pool representation, the candidate favored by the voter group with higher representation will have a polling advantage of around 0.6 percentage point.

Of the sample pool for the Embrain Public survey in which Lee polled at 35.6% and Yoon at 34.4%, the percentage of liberal voters was higher than that of conservative voters by 2.6 percentage points. This kind of distribution in which liberals make up more of the sample pool than conservatives is rarely observed in polls these days.

Whether solutions to such sampling biases can be devised remains to be seen, but until that day, one should take poll results with a grain of salt.

By Jeong Nam-ku, editorial writer

Please direct questions or comments to [english@hani.co.kr]