The Why & How of Likely Voters – Part I

Before exploring the "how" of selecting likely voters, I want to explore the "why."  Why do we care about selecting "likely voters" anyway?

Hopefully, it is obvious that a pre-election survey should aim to sample those who will actually vote on November 2, not the much larger population of adults.  If not, consider the numbers:  Four years ago, according to the Census, the United States had a voting age population of 203 million, of whom an estimated 130 million were registered to vote.  Of these, 105 million cast a ballot for one of the candidates for president.  Thus, the vote for president was 52% of the voting age population and 80% of registered voters. 

Voters are demographically distinct.  As Census estimates show, voters tend to be significantly older, better educated and white than non-voters.  Consider age, a subject of considerable importance this year. In 2000, only 36% of citizens between the ages 18 and 24 voted compared to 50% of those between 25 and 34 and 68% of those over 35. 

While deciding to sample only likely voters is easy, the task of doing so is quite hard.  First, at any given time, many likely voters (and non-voters) do not know who they are.  As political scientist Joel Bloom put it:

When pre-election polls are in the field voters are a population that technically does not yet exist. While some individuals are nearly certain to vote and others are nearly certain not to (with a great many in between), the population of those who will actually vote is, prior to the deadline for voting, is not yet a population, but in the process of becoming one."  [Emphasis and typo in original]

Second, even if respondents know for certain how they plan to vote, they often exaggerate their true intentions. Voting is something society expects, and for that reason many respondents are reluctant to tell a stranger they will not vote.  Thus, while only 80% of registered voters (as estimated by the Census) cast a ballot in 2000, 97% of registered voters told the most recent CBS/New York Times survey they are likely to vote this time.

This combination of uncertainty and over-reporting vastly complicates the pollster’s task of identifying the likely electorate.  Here pollsters rely more on art than science, and no two pollsters use exactly the same method.  The most common practice, particularly in news media polls, is to rank respondents on their likelihood to vote then choose the subgroup of the most likely voters whose size corresponds to expected turnout.  If the pollster thinks the turnout will be, say 55% of eligible adults, they will try to create a subgroup of likely voters that is somewhere near 55% of their adult sample.   

Sounds simple, right? 

Not a chance.  More in the next post…