"wah wah wah, but if I ignore the figures you actually linked and instead pick the ones most charitable to my argument, the numbers are different!"
That's literally what you're doing right now, I started off posting sources because this is such a tired and debunked discussion that I can shut down immediately and then you had to search for hours to find something that would agree with you. Now let's take a look at your supposed proof, ah yes, the PoQ study, that's been debunked just as much.
The authors analysed the self-reported votes of 1,837 respondents, finding that of the 15% (~275) who reported voting for Clinton in the primary, 25% (~69) claims to then have voted for McCain in the general election.
Sounds damning? Except... it's all bullcrap. See for yourself by adding up the votes for Obama and McCain: 0.76 * 30 + 0.11 * 21 + 0.33* 49 vs 0.19 * 30 + 0.86 * 21 + 0.37 * 49 => 41.28% vs 41.89%. Of course, in our timeline, instead of losing by 0.61%, Obama became president in a 7.1% (52.9 to 45.7) landslide. Further red flags: Studies typically find only 2% of primary voters vote against their own candidate. Yet, in this table only 87% of Obama's primary voters reported voting for him in the general, and for McCain it's even lower, 84%.
So why is this apparently the worst poll since The Literary Digest called the election for President Alfred Landon in 1936? Simple: because it is theunweightedresults of apanelsurvey. Normally, opinion polls try to produce representative results by getting a certain number of responses from different demographics and modelling the population. If they don't get enough responses, they keep trying until they do. In contrast, with a panel survey, a fixed cohort of panel members are selected at the start and just keeps getting re-interviewed throughout the rest of the year. Inevitably, response rates drop off a cliff. Hence, it is conventional wisdom that panel surveys are good for showing trends of the self-reporting cohort, but useless as an prediction of the absolute numbers. This gets even worse when you try to get a subgroup of a subgroup, as the author were doing in creating this table. All 69 Hillary-McCain voter it found could just be from West Virginia, for all we know.
It makes zero sense to believe that the 25% number is accurate, when we know for fact that nearly every other number on that table is off by double digits.
Well that was pathetically easy, in your defense the math is significantly harder that basic addition that you're already failed at so can't really blame you.