The following graph shows all the polls I currently have in my spreadsheet in the POTUS race, without the normal pWP calculation:
As you can see, right now there is very little agreement between the polls as to what level of support each candidate has, as a percentage of the population. Is Obama at 47% or 48%? Is Romney at 46%?
I don’t know. And to say with certainty, by looking at the above graph, that you have a good idea as to who is leading the race and who isn’t, is pure foolishness. This is why I’m incredibly skeptical of the process of averaging out multiple polls with a rolling average, like RCP does. Doing a raw average simply ignores the way polls work.
This is why I believe pWP is a better way to aggregate polls. I’m not trying to figure out if Romney has 46% support. I only care about what the polls are telling me about the propensity of the electorate to support one candidate over the other. If polls consistently show one candidate getting more support than his opponent, that is a better indication of who is likely to win than trying to intepret the results based on percentage of support found in the particular sample.
For example, if I have three polls that show candidate A is winning a race by 42%-41%, 50%-45%, and 49%-48%, averaging these polls gives me a race of 47%-45% in favor of candidate A. It looks like Candidate B is within 2% of overtaking A, giving Candidate B some much needed hope. And based on averages this would be true, but I see the race much differently. To me, Candidate A is ahead in all three polls, which means the electorate is far more likely to elect Candidate A than the 2% lead suggests. The propensity of the electorate is firmly behind Candidate A. Instead of “generally ahead,” I would say Candidate A is “Strongly” ahead (i.e. more likely to win than our intuition would suggest).
Averaging polls is simply misleading. It allows for too much subjective interpretation, revealing the bias of the user, instead of predicting the bias of the electorate.