Calgarypuck Forums - The Unofficial Calgary Flames Fan Community - View Single Post - American Politics Thread: The Hunt for Blue November

GGG · 09-06-2020, 10:05 PM

You have to really dig into the aggregators methods before using them. For example the Princeton one form 2016 with Hillary at 98% assumed that polling errors at the state level were independent events wheras as the 538 model assumed that they were correlated. So 538 considered that if there was a 2% error in Wisconsin in one direction it would likely also be a 2% error in Michigan in the same direction.

The other difference usually is when they assume the election is. Some aggregators use of the election was held today whereas others dampen with the assumption that races tighten as elections approach.

But given the accuracy of the polling and models 84% and 70% and this point are almost the same number.

09-06-2020, 10:05 PM	#3233
GGG Franchise Player Join Date: Aug 2008 Location: California Exp:	You have to really dig into the aggregators methods before using them. For example the Princeton one form 2016 with Hillary at 98% assumed that polling errors at the state level were independent events wheras as the 538 model assumed that they were correlated. So 538 considered that if there was a 2% error in Wisconsin in one direction it would likely also be a 2% error in Michigan in the same direction. The other difference usually is when they assume the election is. Some aggregators use of the election was held today whereas others dampen with the assumption that races tighten as elections approach. But given the accuracy of the polling and models 84% and 70% and this point are almost the same number. Last edited by GGG; 09-06-2020 at 10:15 PM.