(10-25-2016 10:37 AM)TechRocks Wrote: (10-25-2016 12:06 AM)Tom in Lazybrook Wrote: The USC Poll is a hot mess. It starts from a very unorthodox weighting mechanism. That, combined with its' outlier status, has resulted in questions. It also uses the same sample of poll responders. If they got a bad original sample, and I think they did, that bad sample will continue throughout the process.
And who is in that video? Is he some sort of mathematician or statistician?
----
That being said, if the poll was being weighed incorrectly towards Trump due to the weighting mechanism (where HRC supporters didn't assign probabilities that they'll vote in equal measure to DJT supporters), and now it turns out that the HRC voters actually are voting...that some of that pro-Trump bias is evaporating. As it should as election day approaches. Everyone will be voting. And will be voting for either DJT or HRC. The massive change in the poll shows that the HRC voters voted, while they'd be assigning a 70% probability to voting, and they weren't moving up their 'likelihood' percentage up rate ably during the Fall. It just went up - all in one go - on one day when early voting started/approached.
You seem to be an expert in everything.
I'm not an expert on polling, but I've been around enough of it to be conversant in it. I consider myself very competent in statistics and statistical analysis.
There are polls that concern me. This isn't one of them.
---
I'm not a big fan of tracking polls. They tend to confirm any bias that was present in the original sample. And the USC poll doesn't appear to control for it all. As a result, you end up with fat tail events. The ABC tracking poll showing HRC up by 12 is probably on the other tail.
There's a reason why groups that are in the business of getting polling right (Real Clear Politics, and the other firms) go to a 'poll of polls' in order to increase the likelihood of getting a good result as the inherent errors in any individual poll will be mitigated by the other polls.
Polling firms are divided into three categories
1) Reputational polling groups - These groups earn money by getting it right. They get to sell their services to more people, at higher prices in the next cycle based upon their historical performance. They have every motivation to get it right. This would also include Real Clear Politics, which gets the vast majority of its web traffic from its polling features. Also groups like CNN, which markets itself as a straight news site. By the way, I think CNN probably has the best polling - and it shows HRC up by 5-6.
2) Bespoke polling groups - These groups create polls for specific purposes, such as to provide an underdog with a poll that they can use to gin up donations, create clickbait/contrarian poll stories, etc. Gravis and Civitas would be two on the right. PPP, SurveyUSA,IBD and Rasmussen are a mix between 1 and 2. When a poll comes out from any of them, I usually wait to find confirmation from another poll.
3) Academic polling groups - Usually these can be classified in group 1. But sometimes, these can be more of the inquisitive - 'hey lets do some research and try to find something new' type polls. The USC poll, with its unorthodox methodology appears to be one of these types of polls.