Slate and the Industry Standard join forces to examine the effect of the Internet on Campaign 2000.
The weekly poll on the Web site of the Democratic National Committee asked visitors: “As the nation approaches a new millennium, what are the most important priorities facing our next president? Saving Social Security, strengthening Medicare and paying down the debt or implementing George W. Bush’s $1.7 trillion risky tax scheme that overwhelmingly benefits the wealthy?”
Thanks to an organized Republican effort, more than two-thirds of the respondents favored Bush’s tax cuts, prompting an embarrassed DNC to remove the poll from its site. News coverage of the incident explained that the poll was non-binding and non-scientific. But you could go further than that. Online polls aren’t even polls.
A poll purports to tell you something about the population at large, or at least the population from which the sample was drawn (for example, likely Democratic voters in New Hampshire). Surprising though it may seem, the results of a scientific poll of a few hundred randomly sampled people can be extrapolated to the larger population (to a 95 percent degree of confidence and within a margin of error). (For a primer on “margin of error” and “degree of confidence,” see this Slate “Explainer.”) But the results of an online “poll” in which thousands or even millions of users participate cannot be extrapolated to anything, because those results tell you only about the opinions of those who participated. Online polls are actually elections, of a kind. And elections, while a fine way to pick a president, are a decidedly poor way to measure public sentiment.
Why aren’t online polls an accurate measure of public opinion?
1. Respondents are not randomly selected. Online polls are a direct descendent of newspaper and magazine straw polls, which were popular in the 19th and early 20th centuries. The print-media straw polls (very different from today’s political straw polls but equally inaccurate) featured clip-out coupons that readers sent in to cast ballots for their preferred candidate. Other organizers of straw polls mailed ballots to people on a list of names. The most infamous of these took place in 1936 when Literary Digest sent 10 million presidential ballots to people, based on telephone directories and automobile registration lists. More than 2 million of the ballots were returned, and based on the results, the magazine predicted Republican Alf Landon would carry 57 percent of the popular vote and defeat Franklin Delano Roosevelt in a landslide.
Literary Digest was wrong, of course, and straw polls never recovered, at least as a predictive tool. Reader and viewer surveys continue to prosper, however, in magazine contests, on TV shows like CNN’s TalkBack Live, and on Web sites.
2. Socioeconomic bias. Some of the common criticisms of online polling could be lobbed at the Literary Digest survey. In 1936, only a relatively small and wealthy portion of the electorate owned a telephone or an automobile. Likewise, many have criticized online polling because Internet users tend to be wealthier, more educated, and more male than the population at large. For this reason, many people assume Internet poll results to be biased in favor of the viewpoints of relatively wealthy, highly educated males.
But even saying that gives such polls too much credit. A scientific poll of the political opinions of Internet users would be subject to that socioeconomic bias (even random-digit telephone polls are only valid for the population of Americans owning telephones). An online poll–even one that eliminates the problem of multiple voting–doesn’t tell you anything about Internet users as a whole, just about those users who participated in the poll.
3. Questions and answers are always given in the same order. Pollsters speak of both the “primacy effect” and the “recency effect,” meaning that the first and last choices are more likely to be chosen, particularly when there is a long list of possible answers. In addition, the order in which questions are given can affect the respondents’ answers. For example, a question about “the longest economic expansion in history” might affect respondents’ answers to a subsequent question about the president’s job approval. Scientific polls account for these effects by rotating the order of both the questions and the answers.
Of course, even scientific polls are subject to error, and not just to the standard “margin of error” that is due to assumed errors in sample selection. As in the DNC poll, questions can be biased. Errors can also be made by interviewers and by data processors. Despite these possibilities, scientific polling has a long, reliable history, whereas “straw polling” has a long history of total unreliability.
As long as they are meant as entertainment, and as long as users understand what their results communicate, there’s no reason to lose much sleep over online polls. What is worrisome is the failure of pollsters themselves to learn from the history of their profession. Even if they bill themselves as “voting sites” rather than “polling sites,” Web sites such as Dick Morris’ Vote.com tacitly imply that the results of their online polls are reliable and valid. Otherwise, why would Morris bother to send Vote.com’s results to members of Congress?
Another online pollster, Harris Interactive, is using its Harris Poll Online to learn about the public’s views on the 2000 election. In order to overcome socioeconomic bias, Harris is using what is known as “quota sampling,” which ensures that the poll’s respondents are an accurate reflection of the population’s demographics. Quota sampling assumes that the answers of a particular demographic group such as white, 18-to-25-year-old Internet users can be projected to describe the opinions of white 18-to-25-year-olds at large. This technique was in widespread use until 1948, when the major national polls based on this technique all predicted that Republican Thomas E. Dewey would defeat incumbent Democrat Harry S. Truman.