Why Bother with Likely Voters

The Ipsos approach to likely voters involves asking multiple questions, assigning each person a score based on their responses, and reporting on likely voters based on expected turnout… So what? Why do we go to all this trouble to build a sophisticated  (and expensive!) likely voter model?

We do it because likely voter model construction has a huge impact on electoral survey results. In past elections, we found that traditional likely voter approaches were not flexible enough and tended to perform poorly under certain turnout conditions. For instance, in your typical November presidential election, between 55%and 70% of eligible voters show up to vote. That means in a typical presidential election survey, 6 or 7 out of every 10 people we interview are going to be voters. When that many people are voting, it is not hard for us to pick out the actual voters to feature in our survey reports.

Compare that to the typical midterm election – i.e. when the president is not on the ballot – where around 35%to 45% of eligible voters choose to cast a vote. What this means for a pollster is that finding ‘real’ voters (who will actually vote!) is much harder for a midterm election poll than a presidential one. Instead of 7 out of 10, only about 4 out of every 10 people we survey are actually going to vote. Polling on primary elections, when only 10% to 20% of eligible voters show up, is even harder. ’Standard’ ways of finding likely voters will functionally treat all these elections as the same even though they feature radically different populations.

That is why we built our likely voter model, to help find and filter down to the 2 or 4 or 7 out of every 10 survey participants who are actually going to show up on Election Day. But again, so what? Why is it important to pick the actual voters out of our survey population; if we randomly pick just anyone, wouldn’t we get the same results? Actually, no – not at all.

A good likely voter model matters because the 20% of people who show up for a primary election, or the 40% who vote in the midterms, are not just like everybody else – not politically. Below is a graph from our Sept & Oct 2014 Reuters/Ipsos Poll. It shows the results of our “generic congressional ballot” question. (The generic congressional question asks if someone would vote Republican or Democrat for their House Representative in the upcoming election without specifying the candidate names.) This graph shows the percentage who say they would vote Democrat or Republican at a series of different turnout levels. On the left we see a 30% turnout where Republicans handily defeat Democrats; on the right we see a 65% turnout where Democrats have a small advantage. Remember, all this data is from the same survey; we get these different results by selecting different likely voter populations based on the expected turnout.

Likely Voter 3 020415A low turnout shows a clear Republican victory; a high turnout shows a likely Democratic win. Why is this happening? Because these voting populations are not the same. If we take the 35% cutoff representing the actual 2014 turnout and compare it to the 55% cutoff representing a turnout level similar to the 2012 presidential election, we can see some of the difference in the voting population.

The graphs below present the age, race and household income demographic makeup of the 35% and 55% likely voter groups.

Likely Voter 4 020415At 35% turnout, the voting population is much older, whiter, and somewhat more affluent. Compare that to the 55% turnout voting population which is younger, more diverse and not quite as affluent. These demographic groups – particularly race and ethnicity – are strongly connected to partisan identification and voting patterns. Finding the right mix of likely voters is therefore critical to election polling: defining this group accurately makes all the difference in predicting the correct winner.

A good likely voter model is all about helping researchers find actual voters. The Ipsos approach to likely voters is a tool that helps us find the actual voters in a wide variety of election conditions, from the low-turnout primaries to the big presidential elections. This flexibility is a key component of our successes.