A few weeks ago, the Reuters/Ipsos poll, along with other pollsters, began reporting on likely voters in addition to registered voters. In past years, we would make this change and go about our day with little comment, but as polling is increasingly scrutinized, and given Ipsos’ total commitment to transparency, we think it’s important to explain our thinking. The main reason we, and other pollsters, use likely voters now is because we are trying to get a better picture of what the electorate will look like on election day.
Let’s start with the basics. Every poll is designed to represent a population. Most polls refer to all American adults (18 and over), who are represented as “all adults” or “all Americans.” However, sometimes polls instead represent specific subpopulations, such as just women or just parents or just cat owners.
Of critical importance to consider in thinking about election polling is the fact that in the United States not everyone votes in elections. The number of Americans voting in an election, usually referred to as the turnout, fluctuates election to election. In some years more people are engaged and more people turned out. In other years, and in lower-salience election such as midterms, fewer people have engaged and fewer people turn out to vote.
And, unlike pollsters and people who read articles written by pollsters, most Americans do not spend a lot of time thinking about the election all year round. For most Americans, the election does not become a major topic of interest until the few months leading up to it – usually represented in presidential years as the time between the conventions and Election Day.
Also important to note is that there are no reliable and current source of information about who is actually going to vote in November – it is a future event and therefore unknown. The Census Bureau is a vital and high quality source of information about Americans, but it looks back on the last election rather than looking forward to the next one.
Therefore, one of the most difficult tasks pollsters face is trying to represent the portion of the American electorate who will show up to vote in November even though we lack any sort of external benchmark to reliably control for turnout. The solution most pollsters have come to is to use a likely voter filter to identify the Americans we believe are most likely to actually vote; ie. to best represent what the American electorate will look like in November.
Since there is reliable data from the Census on who has voted in the past, we combine that information with survey data about how enthusiastic people are about voting in the current election. We at Ipsos do this by calculating the odds of a particular demographic group to vote. For example, there is good Census data on how many young white men in Missouri voted in the last several elections. So, for our surveys today, we give every young white man in Missouri a score based on that historic data. Then, we ask everyone in our surveys to tell us on a 1-to-10 scale how likely they are to vote in November. We use this self-identified ‘enthusiasm to vote’ score to modify the likelihood of a person showing up on Election Day. So based on this technique, a young white man in Missouri who says he is a 10 out of 10 in his likelihood to vote will be weighted heavier than a young white man in Missouri who says he’s a 0 out of 10 to vote on Election Day.
We then use these ‘odds of voting’ scores to identify the people who are most likely to vote. The Reuters/Ipsos poll is currently showing the 60% of Americans who are most likely to vote as our likely voter population. As we get closer to Election Day, we will fine-tune our likely voter population to try to get closer to what we expect the turnout level to be.
However, since we are using the stated enthusiasm of respondents in our poll as one of the ways to identify likely voters, we need to wait to start using this method for the American public to start paying attention to the election. As the conventions tend to be the shorthand marker for the start of the election season, we wait until after the conventions to start using our likely voter filter.
How someone plans to vote or what party they identify as is completely separate from the likely voter calculation. Calculation of likely voters is solely based on statistics of how groups have voted in the past and current poll data of how enthusiastic individuals are about voting in the future. At the end of the day, our goal with the Reuters/Ipsos poll is to be as accurate as possible and reflect what the actual votes in November will be. That is our sole focus and the yardstick for which we measure our performance.