Party affiliation is something that varies over time just like other political opinions, it is not an immutable demographic characteristic. That is especially true when asked in the form used in this survey -- "do you consider yourself", or forms like, "which party do you identify with". I'd fully expect to see responses vary depending on whether pre- or post-survey, as the survey questions will bring to mind likes/dislikes about the parties that the respondent may not have thinking of when asked pre-survey. If anything, I'm surprised it changed so little for the two groups.

Even with asked in terms of formal registration, responses vary as other political attitudes change, though with slower rates of change than things like presidential approval.

When support for a Republican president and Republican Congress go down, we would expect self-reports of Party ID to swing towards the Democrats and away from the Republicans. This is a long-standing, well-known, well-documented phenomenon. It's one of the reason that professials look towards changes in self-reported party ID as an indicator of attitude changes, not necessarily an indication of a screwy sample (which is what I assume you're implying, Tom).

Good source to read for a summary of relevant research is Erickson, MacKuen, and Stimson, _The Macro Polity_ (Cambridge University Press, 2002)

The 9% Indepdnets in that Ipsos poll does look odd, but it's so small because they push indpendents to pick which party they are closer to (question 2d, page 1). Most polling houses just accept an initial repsonse of Independent and leave it at that, but it's not an indication of a bad sample, just different practice from most polls you see.