We've had many inquiries and comments about the latest Gallup Poll trial heat results on the presidential race. Our editorial team will be responding to as many of the issues raised as possible here over the next day or two.
One question that comes up frequently (and apparently is based on various statements bouncing around the Net) concerns the party identification of the respondents in our sample. The supposition on the part of some is that these party identification figures from poll to poll should be constant and the same as some standard established from previous polling.
That's simply not the correct way to look at party identification. At Gallup (as is the case for many other polling firms), we ask party identification at the end of the survey using this wording: " In politics, as of today, do you consider yourself a Republican, a Democrat, or an independent? " Our experience tells us that this is not a fixed demographic measure (like age or gender or ethnicity), but rather is a variable in and of itself. While many Americans are hard-core Republicans or hard-core Democrats and never would call themselves anything different, there is a group of Americans who have no firm party allegiance and whose political identification can and does shift during an election season.
In fact, if one candidate is doing particularly well, it is usually the case that more people in the sample will identify with that candidate's party. Thus, if Kerry is having a good period of time in the campaign (as was the case after the Democratic primaries last February and March, and again in June and July of this summer), then more people will identify as Democrats at the end of the questionnaire when we ask with which party they identify "as of today." If Bush is doing better, as he is now, then more people at the end of the questionnaire will identify as Republicans.
Furthermore, there are no Census or official figures on party identification nationally. A number of states do not require party registration, and what a person calls himself or herself can vary significantly from week to week or month to month.
So it is incorrect to say that a poll's showing one candidate to be ahead is the result of the fact that there are too many members of his party in the sample. In fact, that there are more people identifying with a leading candidate's party is a result of the same forces that are pushing that candidate into the lead.
One final note. Gallup (and other reputable pollsters) do carefully analyze the compositions of each sample on known demographic measures for which there are solid Census figures: age, gender, region of country, ethnicity, and education. And we do weight each sample to each of these if necessary, using complex and accepted statistical procedures. So our samples are remarkably constant from poll to poll on known demographic and regional measures. But in a political year we don't expect that samples will be the same from poll to poll in terms of party identification, any more than we expect the samples to be the same from poll to poll in terms of the choice of candidate for whom the respondents are voting.
I would call it a "prior" distribution, whereas the estimate is the "posterior" distribution. 37+37+36 = 110. Perhaps it's 26% independent?
Anyway, the posterior is sum over the prior. So if your prior is 37% Republican, 37% Democrat, and 26% Independent,
and the crosstabulation is:
R D I
R .90 .10 .50
D .10 .90 .40
I 0 0 .10
Columns indicate with which group the voter associates, and the rows indicate for which party the voter intends to vote.
Then the posterior is:
R = .90*.37 + .10*.37 + .50*.26 = 0.50
D = .10*.37 + .90*.37 + .40*.26 = 0.474
I = .10*.26 = 0.026
Now yes turnout does vary from one election to the next. So it may not make sense to set the prior equal to the result of the previous election. However it shouldn't be the arbitrary choice of the pollster either. Maybe one response should be "I intend to stay home for this election" and then the prior should be equal to the previous election.
Hmmm...you'd have to count voters leaving the system (deaths, expired registrations, moving) as well as those entering the system (coming of voting age, new registrations, moving) somehow.
Yeah, I know, in Philadelphia and Chicago, voters never leave the system.