The CRUX of accuracy in presidential election polling now is entirely a matter of how to correctly determine if a likely voter will actually vote, AND if there are registered voters who will vote who are not captured by the "likely" formula. So, Fred, if I understand you correctly the key to accurate polling is to measure intensity and to accurately size the "likely" voting group. I do not know that any poll tells me much about intensity-is this so?
I have a gut feeling without any empirical evidence to support my view that 2012 will look very much more like 2010 than previous presidential election cycles and that is because of an inchoate understanding that intensity is on the side of Republicans and normally uninvolved citizens not regarded to be likely voters will in fact turnout and support the Republicans.
Do I understand you to say that my suspicion, if proved correct in the event, will play a far more decisive role than normally accepted?
I also note that the chart showing the relative Democrat/Republican breakdown shows no consistent correlation in the size of the victory for other party. Can the intensity factor explain this?
The ideal obviously is to poll actual voters, but since we can't do that the key is to make sure your statistical model for actual voters -- which is called "likely" voters -- is a good one.
Intensity is not the only measure but it is probably the single largest factor that campaigns can actually alter in a base-dominated election, which it looks like this will be. For example: age is a huge indicator of likeliness, but unfortunately none of us can change that. So is party affiliation; Republicans turn out in consistently larger percentages. One of the reasons that the 2006 and 2008 cycles were so deadly to us was that turnout among the base was down, in consequence of which the Dems' raw numerical superiority was more damaging to us than usual.
The polls of "likely voters" do tell you about intensity, but you don't realize it, because it is very carefully hidden in the internals, and everyone attempts to mask how they arrive at intensity. It's a closely guarded trade secret. It's actually quite interesting, because pollsters reputations are based on how closely their polls track to results and they have to show some of their cards in order to be useful to the people they sell information to. We know Gallup has been consistently good once they switch to the "likely" model, and we know that in the last three presidential cycles Rasmussen has been even better.
We also know some of the questions they ask in order to determine likelihood: did you vote in 2010? [People who vote in an off-year election are motivated voters.] How often do you vote? [Habitués are always likelies.] Did you contribute money? [Contributors are guaranteed voters.] Will you vote by absentee ballot? [Absentee participation by citizens within country is a very strong likely indicator.]
But they also ask highly proprietary questions about specific issues important to voters, and they have statistics on how strongly intensity tracks to those issues. Pollsters also gauge intensity on the basis of exit polls from previous elections, and then use those to forecast if those issues voters will be motivated to show up in current cycle. They do trend-line analysis and determine intensity on the basis of whether an issue is falling or rising on the priorities rankings. All of this goes in to the "special sauce" that forms the algorithms for determining likely voters.
I also believe that you're right about this election. I think one of the reasons that 0bama is so worried about money is that the numbers from contributions are a broad-based indicator of intensity, and that does not bode well for him. He will certainly get far more free "money" from the collaborationist press and labor unions than Romney will. That is always true. But that is not an indicator of intensity from the rank and file voter.
There is no correlation in the relative Democrat/Republican breakdown for a number of reasons. For one thing, party affiliation is a lagging indicator, and for that reason pollsters stopped asking for party affiliation about 10 years ago, and now rely almost exclusively on party self-identification (my Dad is still a registered Democrat, but the last Dem he voted for in a presidential election was Harry Truman.) The problem with party identification is that it's also a lagging indicator and there are a lot of people in the squishy middle who identified as Republican in the last election who voted for 0bama. To me that says they a) don't know what their party is about and/or b) had no clue what the modern Democrat Party is. In any event, party identification is more meaningful than party affiliation (or party registration in places -- like Pennsylvania where I live -- where it's required) as an indicator of long-term trends. Fewer and fewer people identify with either party. This makes the degree of Dem oversampling particularly misleading. But when the number of people who identify as Republicans rises to be nearly equal to the number of people who identify as Dems, the country is clearly swinging right. Then the likely models have to be adjusted for issues of more importance to right wing voters ...
The numbers indicating +-/error% in polls employ nothing more than a well known statistical inference about the standard deviation of a sample mean based on sample size -- the error is proportional to the inverse square root of the sample. That, in turn, already assumes that you are taking a purely random sample from the actual population you're trying to measure. There are two problems with that: 1) The sample is never random. There are systematic reasons and one of the most troublesome is that the refusal rate is much higher among Republicans than among independents and democrats, which makes a correction from true randomness necessary (And it is not generally agreed what that number is, or why.) 2) The VERY PROBLEM IN POLLING is making sure you are polling someone in the population. And, as this article correctly explains neither adults, nor registered voters is a good representation of the actual voting population.