Doing a review of the accuracy of each survey data collection mode in 2020, as we do basically every year.
The findings are really worth a show, which will be the topic of #InsideTheNumbers tomorrow.
Will drip some preview findings on this thread... but polling must evolve.
The findings are really worth a show, which will be the topic of #InsideTheNumbers tomorrow.
Will drip some preview findings on this thread... but polling must evolve.
Of course, data sourcing itself also had an impact on accuracy — meaning voter files vs. self-reported, etc — but the long and short of it:
Live caller, which is media preferred and typically a last resort for us in reaching groups like millennial males, is just unneeded now...
Live caller, which is media preferred and typically a last resort for us in reaching groups like millennial males, is just unneeded now...
We reached them and other difficult groups with other modes/methods, for less money, and with far more accuracy.
Strictly sourcing live caller with voter files and applying a tight LV model definitely resulted in missing more GOP-friendly “new” voters. Only spotted new D votes.
Strictly sourcing live caller with voter files and applying a tight LV model definitely resulted in missing more GOP-friendly “new” voters. Only spotted new D votes.
We’ll do this by demographic groups tomorrow, but the gist of it really is that “different groups respond to different modes at different rates.”
And those disparities got worse, which unsurprisingly resulted in even bigger misses outside sampling errors, in ONE direction.
And those disparities got worse, which unsurprisingly resulted in even bigger misses outside sampling errors, in ONE direction.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
