The Reuters-Ipsos interactive interface for the organization’s national presidential tracking poll is a composite of the results obtained from the 50 states and DC. The national sample of likely general election voters totals 22,846; the sum of the states over this same period of time comes to 23,310. It’s unclear where those additional 464 participants come from. My guess is that it’s the consequence of some coding errors or a few respondents being double-counted, but at 2% of the total, it’s a trivial difference.
The three most heavily sampled states are, from the top, New Jersey, Virginia, and Minnesota. These states are the 11th, 12th, and 21st most populous in the country, respectively. The latter two are blueish-purple states and the first is, while relatively amenable to Trump compared to previous Republican candidates, deeply blue.
It’s an inconvenient week to have the schedule loaded up because this really feels like something worth getting a handle on. R-I only applies weights to the following four variables: Gender, Age, Education, and Ethnicity. A 40 year-old white male college graduate from Mississippi and a 40 year-old white male college graduate from Minnesota aren’t necessarily interchangeable.
I remember hearing exasperation during the primaries from pundits who couldn’t figure out why Trump’s performance among “white Evangelicals” in South Carolina was so much better than it was in Wisconsin (ie relative to previous Republican presidential candidates, Trump resonates especially well with ‘Scots-Irish’ from Appalachia). If these professional prognosticators don’t have a clue, how unreasonable is it to wonder if the pollsters do?
How much is the fact that Trump isn’t a generic Republican candidate disrupting the utility of standard polling models? It’s an important question. As commenter pyrrhus alluded to, polling outfits have to draw the line somewhere, but the unusual nature of this election could be causing these polling models to be missing the mark in a systematic way.
Then again, this could just be wishful thinking on this Trump supporter’s part. This weekend I’ll compare the R-I samples by state with votes cast in 2012 by state to present a fuller quantitative picture. Any help making sense of this in the comments is appreciated.