National Public Radio has released a new poll that shows Republican Mitt Romney with a slim one-point lead over President Barack Obama--an eight-point swing since before the First Presidential Debate. NPR says the presidential race is "too close to call." And Democrat pollster Stan Greenberg, one-half of the bipartisan duo behind the poll, points to a 4-point Obama lead in "battleground" states as proof that Obama would win the election if it were today. But NPR's poll of "battleground" states actually suggests an advantage for Romney.
That is because the twelve battleground states tested by NPR include several in which Obama has long held a significant lead, even if several of them (such as Michigan and Pennsylvania) are now "in play" for Romney. There is a common statistical phenomenon, called Simpson's paradox, in which what is true in small groups is false when those groups are combined. Romney could theoretically be leading slightly in most of the battleground states, taken individually, but still remain behind in battleground states overall if Obama leads significantly in one or two.
(The poll does not provide results for individual states--and those results would be useless anyway, since NPR included an average of fewer than forty voters in each battleground state.)
There is at least one "battleground" state, New Mexico, in which Obama leads by double digits in the latest RealClearPolitics average (which rates the state as "Likely Obama"). That could be skewing the margin significantly towards the president, especially as all eleven of the other NPR states are in RealClearPolitics's "Toss Up" column.
In addition, both Michigan and Pennsylvania--which just recently became serious targets for Romney--show Obama with average leads over 4 percent. In every other battleground state, the margin is far tighter than the 4 percent indicated by NPR's poll.
So Romney could be doing far better than indicated--especially since one of the most important new targets for both campaigns is Minnesota, which shows Romney closing rapidly on Obama but is not included in NPR's poll. Nor is Oregon, where Obama has been advertising because of the possibility that a third party candidate, Libertarian Gary Johnson, could cut into his lead.
In addition, it does not matter how many battleground states Romney or Obama wins. It matters which ones--i.e. it matters how many electoral votes they win. NPR's twelve battleground states have 151 electoral votes between them--and four states (Florida, Pennsylvania, Ohio and Michigan) account for well over half of them. Romney could win most of these, and still lose the election, if Obama wins the key states. However, recent polls suggest that Romney is winning in Florida and Virginia, and is tied or close in Ohio.
The only real takeaway, therefore, from the NPR poll is how much the race has changed. The 8-point swing overall--and the 2-point narrowing in "battleground" states, even with likely Obama states such as New Mexico in the mix--suggest Romney has achieved a remarkable turnaround. In addition, the partisan sample of the poll favors Democrats by four points, which may be out of line with the 2-point edge that Gallup suggests Republicans will enjoy at the polls.
This is likely to be the last credible media poll for several days, given the difficulty of polling in the midst of Superstorm Sandy--and it is good news for the challenger.