The Why & How of Likely Voters – Part III

Legacy blog posts Likely Voters

Since the second installment of this series was a bit complex and abstract, even for me, I thought we might take a break and look at the difference between results among likely voters and registered voters. How much difference does it make?

At the end of the 2000 presidential race, three polling organizations released results among both likely voters and registered voters (that I could find). All three used a very similar method of selecting likely voter models.

The results show the classic pattern. The Republican, Bush, did a few points better among likely voters than among registered voters. Again, since registered Republicans typically turn out at levels a few points higher than registered Democrats, this pattern was not unusual. Given the closeness of the race, this meant that Bush led by a few points among likely voters, but trailed or ran even among registered voters. The outcome of the election was, of course, somewhere in between: 47.9% for Bush, 48.4% for Gore and 2.7% for Nader (with the remaining 1% going to Buchanan and other candidates).

Posting these results begs the “BIG, BASIC question” that emailer Bill S asked this morning: “HOW did the various polling outfits do in 2000?” I am not sure we can find a definitive answer, but to let you be the judge, let’s look at the rest of the results among “likely voters” as defined by the pollsters:

Before getting to Bill’s question, notice a few things. First, the average of all the likely voter results (47% Bush, 45% Gore) was not much different than the average likely voter result (46% Bush, 44% Gore) for the three organizations that released results among both likely and registered voters.

Second, notice the similarity across all the polls. Recall the lesson on sampling error from my very first post, and notice that every poll but one had Bush and Gore within the expected 3% of the average. For all the variation in polls during the 2000 race (and there was a lot of it), they converged at the end. (The exception was Rasmussen poll. According to a story quoted on this page, Scott Rasmussen came away from the experience convinced he had too many Republicans in his sample, and vowed to weight by party identification in the future so he would not be similarly off again).

Third, many point to these results as evidence that the likely voter models were wrong, since they typically had Bush ahead by a few points, and of course, the result was essentially a 48% to 48% tie. I’m not persuaded for two reasons: First, Most of the polls were conducted over the final weekend, although a few were done as much as a week before the election. Over the final weekend, four of the five daily tracking surveys charted by the Polling Report (Gallup, Zogby, Battleground and TIPP) showed Gore gaining ground. Second, most of the surveys gave Ralph Nader more support (4% on average) than he received on Election Day (2.7%). Presumably, some Nader supporters switched to Gore at the last moment when the final polls showed that Gore closing on Bush.

The likely voter models may have been off a bit in Bush’s favor, but the results among registered voters (in the first table above) were off a bit in Gore’s favor. Reality may have been somewhere in between.

So which survey was best? I am not comfortable picking a single “winner.” Let’s remember sampling error. Once a survey gets to within a few points of reality, the difference between getting it exactly right and missing by a point or two is pretty much random chance. The National Council on Public Polls calculated the error for ten of these surveys and found the Bush/Gore margin was within 1.5 percentage points for nine of ten surveys – all well within sampling error.

Finally, let’s go back to the difference between likely and registered voters. Are the likely voter models showing the usual pattern of a slight advantage for the Republican candidate? The following table shows results for eight organizations that have released results for registered and likely voters over the last two weeks.

Oddly, the results do not following the usual pattern. On four surveys (Gallup, Newsweek, CBS and Marist) Bush runs at least a point or two better among likely voters. However, on four different surveys (ABC, Time, AP-IPSOS, and NBC) Kerry does at least a point better among likely voters. On the Pew survey, there was no difference.

Why? I’m not sure I know the answer, and I know at least some of the pollsters who produced these results are puzzled too. Perhaps we can find some answers in the different ways these organizations define likely voters. But don’t count on it.

Mark Blumenthal

Mark Blumenthal is political pollster with deep and varied experience across survey research, campaigns, and media. The original "Mystery Pollster" and co-creator of Pollster.com, he explains complex concepts to a multitude of audiences and how data informs politics and decision-making. A researcher and consultant who crafts effective questions and identifies innovative solutions to deliver results. An award winning political journalist who brings insights and crafts compelling narratives from chaotic data.