Examining the 11 September Washington Post/ABC Poll

The Washington Post/ABC presidential poll of 11 September 2012 says that Obama leads Romney by 50% to 44% among registered voters and by 49% to 48% among likely voters.  This is not really the correct result of this poll.  In their poll, the respondents broke down by party affiliation as follows:

Democrats, 33%
Republican, 23%
Independent, 37%
Other, 4%
Don't Know, 3%

So the poll result among registered voters is that for which Democrat affiliation exceeds Republican affiliation by 10%!  According to a Rasmussen Report of polling in August 2012, the breakdown by party affiliation is actually:

Democrats, 33.3%
Republican, 37.6%
Other, 29.2%

The Republicans have a 4.3% advantage in party affiliation, not the 10% disadvantage of the WP/ABC poll.  Consequently, the poll claim that Obama leads Romney by 50% to 44% among registered voters is nonsense.  Let us assume that we can at least use the poll results for the percentage of Democrats, Republicans, and Independents who will vote for either Obama or Romney and recalculate the results with a proper weighting by party affiliation.  The poll says that:

Democrats will vote 91% for Obama and 5% for Romney.
Republicans will vote 6% for Obama and 89% for Romney.
Independents will vote 46% for Obama and 48% for Romney.

Now, I have a suspicion that just as the poll over-samples both Democrats and Independents, it also over-samples middle roader Republicans, given the higher vote percentage among Democrats for Obama than Romney gets among Republicans.  But, let us use these percentages anyway and combine them with the Rasmussen party affiliation results.  In addition, let us assume that among the Rasmussen others, about 4% are going to vote for the Libertarian, Green, Socialist Workers, or whatever other micro-party candidates are running.  I am taking the Independents to be 25.1% of the voters.  The registered voter result is then:

Obama:  (0.91)(33.3%) + (0.06)(37.6%) + (0.46)(25.1%) = 44.1%

Romney:   (0.05)(33.3%) + (0.89)(37.6%) + (0.48)(25.1%) = 47.2%

Thus by declared party affiliation, Romney leads Obama by 47.2% to 44.1%.  This is a Romney advantage of 3.1%.  When the WP/ABC poll made likely voter corrections, the big change was due to their assumption that Romney's advantage among Independent voters would increase from 2% to 11%.  Applying their expected likely voter percentages to the party affiliation percentages of Rasmussen, one gets the likely voter result:

Obama:    (0.94)(33.3%) + (0.06)(37.6%) + (0.43)(25.1%) = 44.4%

Romney:   (0.05)(33.3%) + (0.91)(37.6%) + (0.54)(25.1%) = 49.4%

So the likely voter advantage for Romney over Obama is 5.0%!  Now, I am not highly confident in the adjustments the WP/ABC poll makes for likely voters, but the Romney advantage in registered voter results of 3.1% will likely grow at the polls due to the likely higher turn-out of Republicans and those Independents favoring Romney.

Of course, if Romney does win the popular vote by 5%, it is extremely unlikely that he will not also win the electoral vote.