I have found that if I use pollster and just eliminate blindly internet, web, and robos that what I end up with looks much more sane. I think that disposes of all of those you cited e.g. It shrinks the sample a great deal, but the real outliers seem to depend on robos and web/internet.
It seems web/interent is insane, one has to see the poll first, which I assume is a function of going to a page where it is at, which seems like a massively biased event.
I suspect that if TheTrustMatrix and POPEII both ran web polls they would see different results, just as larryflint.com would undoubtedly massively over-represent tea party members of congress. Or maybe I don't get how web polling is done.
| 