Real Clear Politics with their averaging of polls claim it makes them more accurate, I think they are deliberately ignoring the problem of their method helps reduce random error, but exacerbates error due to unrepresentative samples. I also wonder if the response rates have gotten low enough that the multiple polls are no longer using independent samples.
They are partially right. To the extent that different polling methods will have slightly different response groups, they have a point.
The error due to unrepresentative samples is not exacerbated, although it becomes a larger proportion of the total error under RCP's methods.
Both are, however, a very small factor in polling error. As for independence, they are still nowhere near losing that.