CNN released three polls in July: one right before the Republican National Convention (July 13-16), one between the RNC and Democratic National Convention (July 22-24), and one right after the DNC (July 29-31). The headline number read: 49 Clinton – 42 Trump (Before), 45 Clinton – 48 Trump (Between), and 52 Clinton – 43 Trump (After). This wild swing was caused by two things: shifts in partisan non-responses (which I have no reason to be believe is correlated with likeness to vote) and partisans shifting their support for “neither” or “no opinion” to their candidate.
CNN reports that the After poll had 28% Democrats and 24% Republicans. We will just ignore this number as any party identification that does include “leaners” is really just getting a small proportion of the partisans. Buried in the reports (linked above) are within party questions where they include the raw numbers of leaners. Before: 407 Democrats and 425 Republicans, Between: 397 Democrats and 431 Republicans, and After: 432 Democrats and 413 Republicans. These are raw numbers, but you can see the trend, the percentage of Democrats decreases from Before to Between, and then comes back up from Between to After.
I back-out the weighted numbers from the cross-tabs where all respondents are either “Lean Democratic” or “Lean Republican”. In the After poll, Clinton has 52% support with 95% from DEM and 8% from GOP, while Trump has 43% with 3% from DEM and 89% from GOP. Or, 50.8% Democratic and 46.6% Republican is a good approximation. Similarly for Before: 53.1% Democratic and 45.2% Republican, and Between 47.7% Democratic and 50.2% Republican.
This is just approximate, but look at the relationship between two-party party ID and two-party poll share. I am not saying party ID is 100% of the results (previous research suggests maybe 2/3), but it is really important. Remember, the poll controls for: age, gender, geography, telephone usage, and education in its weighting. But, it does not control for party ID. All of the evidence suggests that party ID is stable week-to-week; this shift is a reflection of shifting non-response on party ID. Maybe Democrats just do not want to talk politics in the weekend after the RNC? Or Republicans are eager to talk politics?
Now embedded in same line of the results was that 95% Democratic support for Clinton and 89% Republican support for Trump; Before that was 88% and 87% respectively, and Between that was 87% and 89% respectively. Clinton moved from 88% to 87% to 95% support from Democrats, while Trump moved from 87% to 89% to 89% support from Republicans. The part of Trump’s post-RNC bounce that was not driven by more Republican answering the poll, was a small consolidation of support from Republicans. They were not supporting Clinton before, but stating “no opinion” or “neither” as their choice. Similarly, the part of Clinton’s post-DNC bounce that was not driven by more Democrats answering the poll, was a sizable consolidation of support from Democrats. Her “no opinion” or “neither” went from 6% to 5% to 1% over the three polls.
To be clear, I am in no way advocating to “un-skew” the polls! There is legitimate movement over-time of sentiment (just not really between candidates as much as between candidates and other responses). And, there is some movement in likeness to vote correlated with party identification. There may even be some movement in party identification. But, that movement should be really, really small on a week over week basis! So, it something we should follow closely, be cognizant about it.