|
ErIog posted:Population stability isn't really the key issue. The key issue is that nobody ever got around to figuring out who shows up on election day and why. Pollsters treat it like a black box. They modeled turnout based on 2012 and 2008. They were wrong, and being wrong with the turnout just by that little bit made them fail to see Trump narrowly swinging the EC. There aren't really any better assumptions that could be made. Dealing with that kind of issue can only be the job of the aggregators.
|
# ? Dec 7, 2016 16:53 |
|
|
# ? May 3, 2024 16:53 |
|
ErIog posted:Population stability isn't really the key issue. The key issue is that nobody ever got around to figuring out who shows up on election day and why. Pollsters treat it like a black box. They modeled turnout based on 2012 and 2008. They were wrong, and being wrong with the turnout just by that little bit made them fail to see Trump narrowly swinging the EC. I remember so many people here bashing polls that had internals arguing for an electorate more like 2004, with older whites turning out in greater force. Of course, the white vote did not actually grow as a share of the electorate, but it stayed mostly identical and Trump getting ~61% of the white vote rather than Romney's ~59% actually explains a good chunk of the shift in the electorate. This election also had some of the weirdest turnout patterns. "Record high turnout in Florida, nearly 15% more votes cast than four years ago!" And "Record low turnout in WI, fewest votes cast since 2000!" Should not be takeaways from the same election. Turns out that this election had 58.8% turnout compared to 2012's 58.6%, but it's not really stagnant, it's that turnout was way up in some places and way down in others. Edit: up to 59.1% turnout according to electproject. Even more true. Patter Song has issued a correction as of 17:25 on Dec 7, 2016 |
# ? Dec 7, 2016 17:23 |
|
Fangz posted:There aren't really any better assumptions that could be made. Dealing with that kind of issue can only be the job of the aggregators. Pollsters should be more honest about their error bars. For instance, specific whole numbers (the way most poll results get reported) have more precision than actually exists in the data the pollsters are drawing their numbers from. They report MoE, but even the MoE is being skewed by how they choose to model the electorate. This had the affect of creating essentially a systemic error across nearly all pollsters. If pollsters were being honest the polls for most of the race would have just been ¯\_(ツ)_/¯. So the better way to go would be to stop reporting single whole numbers as gospel. If everybody's worried about predicting turnout then how about instead of modeling turnout a single way and lying about the precision of their conclusions they run their numbers through multiple turnout models and, again, report the results as a range. I understand that predicting things is hard. My critique is more with modern pollsters' self-aggrandizing penchant for fake precision. We don't need "better," assumptions. We just need more honesty about the limits of the models these pollsters are using.
|
# ? Dec 8, 2016 01:56 |