Hi Jeff… It’s a fair point. As the saying goes with data (or really anything), garbage in, garbage out.

In this case, though, I tried to factor in three things:

1. Quality rating. I ran a bunch of analyses on past records and quickly figured out… RCP does a way better job than I could attempt to do. So, leveraging the B and above I think is a pretty good standard.
2. Depth. The data isn’t just Quinnipiac, or NBC/WSJ, or CBS… it’s a blended average, with some weights applied based on the rating of the poll. One poll can be off. Statistically, multiple polls with different methodologies all being off is possible, but not likely
3. Margin of error. This is key. It’s been pointed out on the thread that folks called Brexit and the UK election wrong. This is basically true, but the polls all had a margin of error that basically said (to me at least, statistically), it’s probable, but it’s a toss up. In this case, I only gave a lock to a party when every quality poll… every one… showed the candidate in the lead + the margin of error + 2 points for good measure. I think the statistical term for that is “CYA” :-). By that metric, both races in the UK would have been too close to call.

That said, it’s my opinion and analysis. I think it’s right. It doesn’t mean it is. All I can do is list the methodology and the raw data, and folks can judge for themselves (which is how I do wish election decisions were made in general).