Thomas Lumley over at StatsChat has used Peter Green’s polling average code to estimate the actual margin of error for political polls after adjusting for design effects. I had no idea how this could be attempted across non-probability samples (EDIT: To be fair, I had no idea how this could be attempted across multiple polls – at all).
If the perfect mathematical maximum-margin-of-error is about 3.1%, the added real-world variability turns that into about 4.2%, which isn’t that bad. This doesn’t take bias into account — if something strange is happening with undecided voters, the impact could be a lot bigger than sampling error.
That last point is a fairly important one. There are many potential sources of error in a poll other than the sampling error.