Here is some analysis I did of Registered Voter Polls...
I did use the 2/3 of the undecided go to the challenger thoery, so if you want to argue on that, please do so, but crazy Dick Morris says 85-90% based on past elections go to challengers when there is an incumbent. Example Ford was +1 in the last Gallup poll in 1976, but the result was Carter +2
As to the likely voter issue, and Rasmussen... Because you must subscribe to get demographics, it is hard to say if his sample is skewed one way or another in his state polls. Most of his state polls are done in one day, so time of day, is important. His better polls are the ones done over several days, but even there there is sometimes wide fluctuation which are not the most comfort building in terms of accuracy.
There was a truly the King of Bogus polls put out by WMUR/UNH which gets to the attempt to discourage issue. Unfortunately, for the pollster the Obama lead of 9 with a huge oversampling of Dems, which you can compare to the state party figures. If that is `adjusted` to the proper ratio, you get a Romney +2/+3 number... Which coincides with the Dartmouth Coll poll, who did use proper party ID figures when conducting their poll..
Also in the chart, above you can see several cases of oversampling of Dems, most notably PPP in Virginia, and Survey USA in NC...