Mark Blumenthal | June 7, 2007
Topics: 2008 , Divergent Polls , The 2008 Race
As promised, Gallup's Frank Newport has posted an update that responds to questions raised here and elsewhere about the composition of those asked the Democratic primary vote question in this week's USA Today/Gallup poll. The key paragraphs:
There has thus been no change in this methodology which would account for Obama's gain against Clinton in this latest poll. More importantly, a careful analysis of the data shows that there has also been no significant change this year across 9 polls in the percent of the overall Democratic sample who are Independents leaning Democrat compared to the percent who are "pure" Democrats...
In other words, there were just as many independents in the previous samples showing Hillary Clinton with a healthy lead over Obama as they are in the latest sample showing Clinton and Obama tied. So an argument that some change in the sample composition (as defined by these political measures) accounted for the shift in voter preference is not supported by the data.
Newport also provided me with the data he references above. As Pollster readers might be interested, I have reproduced it here (omitting results for candidates other than Clinton and Obama):
Newport's data supports his contention. Independents that lean Democratic were 35% of the subgroup that answered the Democratic primary horse-race question. That is exactly the same percentage as the Democratic leaner composition I get when I average all the nine surveys.
Of course a bigger question is whether a national survey on the Democratic primary contest should include that many independent identifiers. On this score, Gallup is not alone. Most of the other national surveys do the same (though some, unlike Gallup, include only registered voters).
This is a tough issue to resolve since the notion of a national "likely Democratic primary voter" is so abstract to begin with. We do not have a national primary, of course, and the rules for participation and turnout levels vary so wildly across states that it is next to impossible to try to use a national survey of this sort to model the entire nation.
However, we might suggest that all national pollsters begin regularly reporting both their sample composition (% of independent leaners) as well as the results among pure Democratic identifiers. The more we know, the more we know.