In September I put forth a theory that people identify their party affiliation differently to exit pollsters than they do in pre election polls, which is just one more reason that weighting for party id is not a good idea.
Here's my basic point: there are a lot of registered Democrats in North Carolina who vote Democratic for local offices, a fair number of state offices, but then vote Republican for President and sometimes Governor or Senate. When we poll them on a bunch of statewide offices and they're mostly choosing Democrats, they identify as Democrats on our polls. But when they're filling out an exit poll just about the top of the ticket stuff where they may have voted Republican, they're more likely to identify as independents or Republicans then, regardless of their registration status.
The North Carolina exit poll this year I think backs up my point. On our final poll, which was dead on, we found 49% of respondents identifying as Democrats, 38% as Republicans, and 13% as independents. The exit pollsters found 42% identifying as Democrats, 31% as Republicans, and almost twice as many, 27%, describing themselves as independents.
Here are some key differences between how groups within our poll stacked up and how groups within the exit poll did:
-We found Obama getting 81% of the Democratic vote, while the exit poll found him getting 90%.
-We found McCain getting 90% of the Republican vote, while the exit poll gave him 95%.
-We found Obama up 50-46 with independents. The exit poll found McCain winning them 60-39.
So to review, in the exit poll we have fewer Democrats and fewer Republicans both showing a higher degree of party unity than our pre election polls indicated and independents looking much more Republican.
What happened? I think a lot of the registered Democrats supporting McCain identified as independents on the exit poll. And probably some of the Republicans supporting Obama did too. But there are a lot more registered D's voting Republican for President than vice versa, hence the strong advantage for McCain among indies there even as our polls told a different story.
The bottom line? The folks identifying as independents on our polls were a completely different universe than the folks identifying as independents on the exit poll. And that's why using exit poll numbers as a baseline for weighting standard pre election polls isn't a good idea because the way people respond to each of those types of surveys is not necessarily the same.