This is weird criteria:
The BW-CRI study was conducted during the period of February 11-20, 2016 among 825 likely Ohio voters using online panel data with quotas in place for gender, age, and region (i.e., Northeast, Northwest, Central, Southeast, and Southwest Ohio). To qualify as a likely Ohio voter, respondents had to meet three criteria: (1) they were "absolutely certain" they were registered to vote; (2) they had given "quite a lot" or "a moderate amount" of thought to the upcoming presidential election; and (3) and they had either voted in the 2012 U.S. presidential election or were not eligible to vote.
One can argue if an online panel can really compare to a traditional poll, but the biggest methodological problem I see is the use of quotas rather than weighting to ensure a demographically representative sample. I feel like that could potentially skew the sample a great deal but I don't even know if it's been tried before
I thought what was hinky was no stratification for race - which is the big gap between Sanders and Clinton and the strict screening of having to have voted in 2012. This is the DEM result that I would expect for whites, not the overall population.
