I read in a recent news article that one of the main predictors of party allegiance was which party had a president in the White House during the respondents 'formative' years. However, I am unable to find neither the news article nor the study it was reporting on. If that were true, though, any current poll taken in isolation would simply reflect political systems over the last 50 years.
According to Franklin (1984), this is called the traditional model of party identification and seems to have been mostly superseded by the revisionist model:
The traditional view that party identification develops at an early age, is remarkably stable throughout life, and is relatively unaffected by other political attitudes has been recently challenged by evidence that adult partisanship is in fact quite responsive to other political attitudes such as policy preferences, retrospective evaluations, and past votes.
From the conclusions of that paper:
The results strongly support the revised view of partisanship. We have seen that party identification is not fixed in childhood, and further that it changes in response to policy preferences. We cannot say at what age young people begin to incorporate policy preferences into their partisanship. ... These findings clearly move us away from the traditional notion of party identification.
... [T]he revisionist view both accounts for the original movements away from parental identifications and provides an explanation for the later stability.
From skimming some abstracts, newer articles seem to agree with the revisionist model but try to account in more detail what makes people change allegiance and at what time in their life.
Searching Google Scholar for party identification
should throw up a good few hits to get you started, especially if you restrict the publication date to more recent years.