When I posted earlier today, the RCP average was at Obama +3. It's now at Obama +3.9, thanks largely to the inclusion of one poll: a WT/Zogby poll showing Obama with a 9 point lead. It makes you wonder if polling averages, which are clearly superior to individual polls, could be further improved. This is what it currently says:
Obama: 49.1
Romney: 45.2
Spread: Obama +3.9
Using the kind of analysis I did previously, taking the standard deviation of the polls in the average, removing the poll with the largest deviation, running it again and repeating, you start with a standard deviation of 2.3, and the Zogby poll being 5 points out. Removing the poll farthest out and starting over, you eventually have four polls left, all with a deviation of 0.5 from the mean. The new average:
Obama: 49.3
Romney: 46.8
Spread: Obama +2.5
I think the method I used is probably more accurate than just an average, which can be skewed heavily by a single poll. The quick and dirty means of arriving at a similar result is to look at the median:
Obama: 49
Romney: 46
Spread: Obama +3
The thing that stands out when you compare the polls is that they all show a similar level of support for Obama, from 48-50. The support for Romney is where most of the deviation comes in, going from 41-47. The polls that show Obama with huge leads all show a lower level of support for Romney, not a higher level of support for Obama. What that means is that the polls with a big Obama lead show a much larger number of undecideds.
If in the Zogby poll, for example, Obama had a 9 point lead with no undecideds, it would be 54.5-45.5. Instead it's 50-41. Obama's "lead" is the same as the number of undecideds.