Why likelihood to vote matters when one looks at American pollsters

April 17, 2008

Why you should be sceptical of polls that focus on registered voters.

Normally, I am sceptical that the differences between individual firms makes compiling polling averages problematic, especially since the British pollsters produced very similar results in 2005. However, given that one the few points of difference between the otherwise nearly identical methodologies of the American pollsters, is whether all registered voters, or only likely voters, should be sampled at this point. Including the preferences of voters who are registered to vote is cheaper and quicker than only including likely voters and supporters argued that it is unrealistic to expect people to know whether they are likely to vote several months before the election. However, the counter-argument is that there is little point in asking the opinions of people who won’t ultimately put a cross on a ballot (or pull a lever in a voting machine). The fact that higher income groups tend to vote more often than lower income groups makes this question of who to include even more pertinent as does the fact that it is customary for even the holdouts to switch to likely voters closer to the date of the election.

The simplest way to find out how this dynamic impacts on polling trends is to compare the Obama-McCain margins produced by surveys of registered voters with those produced by survey of only likely voters. Using Samplemiser, which although possibly flawed is the best way to create daily figures from irregularly spaced polls, I have created a time series for McCain-Obama margins for registered voters and one for likely voters between February 4th and April 6th. I have excluded polls that use all adults as their sampling frame and the poll carried out by the Republican National Committee. Eyeballing this chart suggests that restricting the sample to likely voters helps the Senator from Arizona whereas including registered voters helps Senator Obama.

However, the strength of the correlation and the sensitivity of the margins must also be established. Linear regression suggests that the equation is: y = 0.8565x – 0.043944 (where y is the likely voter margins and x is the registered voter margins) with the R2 a reasonably high 0.4344. This suggests that poll which measure likely voters will generally reduce Obama’s margin over McCain by nearly 4.5%, and are slightly less sensitive, when compared to polls of registered voters. This will not solve the debate over whether registered voters or likely voters are the more appropriate measure, but it does give an indication of the impact of the differing methodologies on the reported leads. Indeed, Mary Matalin and James Carville’s memoir of the 1992 campaign points out that the change from registered to likely voters in the closing days of the election caused Clinton’s lead in polls to go from double digits to 6% – raising the question of whether this could happen again.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: