Jophiel wrote:
gbaji wrote:
You can see this when you look at RCP average polling. You'll see about half that show the candidates within a couple point spread (some with Romney up, some with Obama up, but all close), and another half that have Obama up by 5-8%.
Do you even look at the
RCP site or do you just mindlessly parrot whatever someone else tells you? RCP has a single poll since Oct 10th in Ohio with Romney up (a Rasmussen poll from a few days ago). Every other poll out of the nine in the main calculation right now has Obama leading. Obama's largest lead in an Ohio poll from the main calculation is +5.
I was speaking about polling in general. Not just Ohio.
someproteinguy wrote:
Whatever happened to dialing up 1,000 people at random and asking them who they were going to vote for?
They do start out that way. Then they adjust the results by weighting the party affiliation of the respondents. Let me explain how this works.
Let's say that in the exit polls from the 2008 election 40% reported themselves as democrats, 30% reported as republicans, and 30% were "other/independent/whatever". You then conduct a poll of 1000 random people and ask them who they're going to vote for in this election. As part of the poll, you ask people's party identity and 35% say they are democrats, 35% say they are republicans, and 30 say they are "other/whatever" (yes, I'm contriving these numbers for simplicity of the math).
Let's also assume that both groups who identify as republican and democrat vote their candidate and the others break for Romney 20-10. The result of your poll would be 55% Romney, 45% Obama (yeah, I'm ignoring "don't know" answers as well to also make this simple). But since the sample you polled doesn't match the ratio of the last election, we must adjust for that. So we weight each Obama vote and Romney vote according to the ratio of the sample in the poll to the sample in the last election (40/35 versus 35/40 respectively). So the results look like this:
Obama. 45*(40/35) = 51.390
Romney 55*(35/40) = 48.125
And magically, a poll that actually resulted in Romney +10% transforms into Obama +3.2%
Normally, weighting based on the last election is a good method to determine relative party turnout because usually party turnout and affiliation doesn't change that much from one election to the next. But 2008 really was an outlier in terms of Dem turnout and identification (and no one is remotely attempting to claim otherwise). Using those exit polls to weight polling in this election is going to result in significantly wrong results. But polls use the same methodology every time so as to ensure consistency. You can't blame them for this. But you really do have to understand that this means that a significant portion of polls out there are not remotely accurate.
Edited, Nov 2nd 2012 2:20pm by gbaji 
