[Iowa_Flames_Fan]

I mean, the other reality is the poll aggregators have a better history in the US than they do here simply because they have enough recent data to make the older (and frankly useless) data fall out of the average. 338’s model now includes 3 recent polls, which if you take a simple average of just those three would show an NDP lead of about 0.5% (so basically tied).

The problem is 338’s model ALSO includes some extremely old data in the set, like (as just one example) an Angus Reid poll from last September showing the UCP leading by 5.

To me that’s a huge problem with the model, but the issue is probably that if you drop all the old data out of the set, you’re left with not very much to go on.