If you’ve been paying special attention to the news recently, it’s just possible you may have heard or seen one or two items relating to a peculiar recurring phenomenon called a “national election.”
Like the Capistrano swallows, elections return at regular (more-or-less) intervals, and attract a great deal of attention when they do. Some of that attention takes the form of public opinion polls.
Polls were used as early as1824 by the Harrisburg Pennsylvanian and the Raleigh (North Carolina) Star, which conducted “straw polls,” so-called because a number of citizens were selected haphazardly to see which way the political wind was blowing, just like you might throw straw into the air to see which way the real wind is blowing.
However, polls didn’t really come into their own until the 20th century. One of the first serious attempts was carried out by the magazine Literary Digest, which in the 1920s and 1930s sent out postcards to up to 18 million U.S. voters, asking their preferences among the presidential candidates. They received two million replies, and correctly predicted the presidential winners up through 1932. They blew it in 1936, however, predicting that Franklin D. Roosevelt would lose to Alf Landon. Roosevelt won by a landslide. Literary Digest went out of business.
Literary Digest’s methods biased its sample. The list of people to whom they sent postcards excluded many of lower socioeconomic status, and people who choose to reply to a postcard poll are only those who want to, who are generally more extreme in their opinions than is average.
Enter George Gallup, director of research at New York’s Young and Rubicam advertising agency, who drew on new business marketing methods to invent a more scientific way of polling, founding the American Institute of Public Opinion in 1935. Gallup’s interviewers chose individual respondents in accordance with quotas designed to make the sample roughly mirror the national population in geographic location, rural or urban residence, sex, age, race and socioeconomic status.
This “quota method” correctly predicted Franklin D. Roosevelt’s election in 1936, but it still included a source of bias: interviewers had discretion in choosing the respondents within the quota categories. This could exclude, for example, individuals who are difficult to contact.
Modern polls use random selection to choose respondents, insuring that no type of individual is systematically omitted. This also puts polling firmly into the field of mathematics called statistics, which means, while you can’t be certain the poll is accurate, you can accurately predict the likelihood of error.
No matter how large the population under study, the size of the sample determines the range of error. Whether you poll a city or the whole world, a sample of 1,500 people will be accurate within three percentage points, 95 percent of the time. Larger samples yield only slightly smaller errors.
The proof is in the polling pudding: national opinion polls have only failed to correctly predict elections twice, in the U.S. in 1948, when Harry S. Truman defeated Thomas E. Dewey, and in1970 in Britain, when the Labour party unexpectedly lost. In both cases, polling ceased too long before the election to catch last-minute trends. As well, quota sampling was used, and voter turnout was misjudged. (Modern polls are designed to also determine the likelihood of the respondent actually voting.)
The success of pollsters in minimizing these and other sources of error is best illustrated by the fact that since the 1948 debacle, leading polling organizations have had only a tiny average error in predicting the results of all congressional and presidential elections in the U.S. (By the way, research studies in that time have also shown no evidence that poll results influence voters’ political choices — in other words, there is no”bandwagon effect.”)
On October 25 we’ll put the polls to the test again.