[maxerickson]

Isn't the reasonable definition of reproducible for a poll having overlapping margins of error? A useful poll should do that.

The result is qualified, comparisons of results shouldn't ignore that.