I'm not accusing you of misrepresenting polls, at all. I was just expounding on my initial musing before JVS sprung wood. I was curious to hear your thoughts. If these polling bodies are purely agnostic, then over time they should be indistinguishable in the aggregate, yes? So them skewing in the aggregate would suggest some level of manipulation-- even if it is in its most innocuous form where the little tricks they might be using are considered "kosher", to to speak, within the philosophy of mathematics as it applies to politics.
I'm just curious to see a more in-depth mathematical analyses of this phenomenon by someone like Silver, including if it actually exists, and the why & what that it indicates if it does.
The issue here is what you're terming manipulation is, for the most part, a different methodology. There are lots of places where polling firms vary. One is how they collect data - robocalling has less accuracy than other methods, and landlines tend to overrepresent older voters. So if a firm uses landlines, they might trend conservative.
There's also the types of questions asked. In the Democratic primary, polls that gave a yes/no tended to favor Clinton, but some polls ( quinnipac) asked voters how enthusiastic they were about different candidates, and Sanders tended to over perform in those.
There's also how the data is analyzed. Many of the answers are reweighed with two considerations - weighing answers by how likely that demographic is to vote (educated whites tend to vote at a high rate, so they would be weighted more than less educated asians, who barely vote). There's also reweighing demographics to fix overrepresention in polling. If you have 30% Hispanic respondents, but they only make up 15% of the voting population, some polling firms will reweigh that segment to make their poll more representative of the population. This might be what you mean by manipulation, but whether a firm reweighs or not, and how they do it, can have a substantial effect.
Finally, some polling methodology is simply bad. This came up in the dem California primary, which has a lot of immigrant households. An assumption of some polling firms was that in calling a household, you're equally likely to get any member, and that if you ask for someone, that's who you'll get. But in immigrant households, you'll have an assigned speaker who will always get given the phone for English calls, and if it is 'for' someone else, they might even answer for them. The speaker is usually younger, so a number of Cali polls showed Sanders performing unusually well with Hispanic households, which wasn't reflected in the actual vote.
Anyways, 538 discusses this already to various extents in different articles. If you haven't seen them already, start with The State of the Polls 2016, which links to a number of other articles on polling error, the 538 pollster ratings, which actually had several useful metrics for measuring pollster accuracy, and the article on how the ratings are calculated.
I'll end by noting that the pollster ratings include a measure for mean-reverted bias, which is how much the poll is biased in any particular direction compared to election results.
Rasmussen and Pew skew right, NBC, CNN, Gallup, etc skew left. So yes there first place I look typically is an aggregation. The reason I posted this Rasmussen poll is that it's the first time I've seen Trump up since the presumptive nominees shook out. It's early, and predictive polling isn't nearly as accurate as responsive polling.
A question about whether Hillary would be a good president just can't be answered as accurately as whether Obama has been a good president, I mean.
I by no means meant it as the first time any poll has shown Trump in a favorable light vs Hillary. But that said, so many polls have shown Hillary or Bernie waxing the floor with any GOP candidate but the kind of mamby pamby media-picked chump we've lost the last two elections with.
1.) CNN and Gallup have measured bias compared to election results of .1 and .8 respectively, and both of those have been towards conservative candidates. NBC runs their polls with the WSJ, and tilts left by all of .5. NBC and CNN also have substantially lower error than Rasmussen. You don't know what youre talking about, and that's probably due to assuming that their coverage, which you believe to be biased, is reflected in their polling. It's not.
2.) While Trump might not have been the candidate the media said would win, they selected him with billions of free airtime.
3.). Do you remember who the candidates were in 2008 or 2012? In 2008, you had mcCain, Romney, Huckabe, Ron Paul, Fred Thompson, and Giuliani. In 2012, the alternatives to Romney were Santorum, Gingrich, Paul, Perry, Bachmann, and huntsman. You can go ahead and tell me which of those was a better option.
Remember the first rule of holes.
