“To err is human … but if it’s less than 4 percent, it’s only sampling.” So went the winning slogan on the American Association for Public Opinion Research’s conference T-shirt sometime in the 1990s.
The first message of this slogan is that the result of a sample always differs from that of the entire population, so a small difference is no surprise or shame. Of course, the bigger the sample, the better it should perform.
The slogan also reveals, implicitly, that 625 is a common sample size in an American poll of voters. This is because 4 percent = 1/25, and the square of 25 = 625. The formula for the sample size for a desired error margin (e.g., 4 percent) for a proportion (e.g., a candidate’s voting rate), with 95 percent confidence, is taken up in college statistics. The formula applies as well for a small state like Hawaii as for a large one like California; 625 respondents is equally good for both states.
The US presidential contest is decided, not by one national election, but by 51 statewide elections (counting Washington, DC as a “state”). So there are 51 separate contests for pollsters to predict. The national popularity contest is incidental, with only pride at stake.
If an American presidential candidate’s lead in several statewide polls is larger than their error margins, then the state is called “safe” for him. If his lead is below the error margin, then obviously the state is not safe. Nevertheless the odds stay in his favor; there is no 50:50 tie unless the percentage points are equal.
In this column, as in previous ones, my basis is the statewide polls found in www.PollingReport.com. I simply eyeball the most recent polls, without averaging. I take for granted that polls of likely-voters make provision for realism of the demographics (race, gender, marital status, work status, socioeconomic class) of those who will actually vote.
Throughout the campaign, the statewide polls indicated that all the 22 states that went consistently Republican in 2000, 2004 and 2008—namely Alabama, Alaska, Arizona, Arkansas, Georgia, Idaho, Kansas, Kentucky, Louisiana, Mississippi, Missouri, Montana, Nebraska, North Dakota, Oklahoma, South Carolina, South Dakota, Tennessee, Texas, Utah, West Virginia, and Wyoming, with a total of 180 Electoral College votes or ECVs—would be safe for Mitt Romney.
In fact, Romney easily won all those 22 states, by leads ranging from 8 points in Georgia to 48 points in Utah (based on www.politico.com/2012-election, 11/08/2012, about 1 a.m. Eastern time). No surprises; good work, pollsters.
By October, Romney also had the statewide lead in Indiana, whose 11 ECVs thus raised his prospective safe ECVs to 191 before election day (see my “Another close US election,” 11/03/2012). Indeed, he won Indiana by 11 points, as Indiana swung red in 2012.
The statewide polls also indicated that the 19 states that went Democratic in 2000, 2004 and 2008—namely California, Connecticut, Delaware, Washington, DC, Hawaii, Illinois, Maine, Maryland, Massachusetts, Michigan, Minnesota, New Jersey, New York, Oregon, Pennsylvania, Rhode Island, Vermont, Washington, and Wisconsin, with 242 ECVs—would all be safe for Barack Obama in 2012 as well.
In fact, Obama did win all those 19 states, by leads ranging from 5 points in Pennsylvania to 83 points in DC. No surprises; good work, pollsters. Note that the major eastern seaboard states hit by Hurricane “Sandy” were traditionally blue and were expected to stay blue.
By the end of October, the statewide polls also had swing states Iowa and New Mexico going blue. Their extra 11 ECVs raised Obama to a safe 253 ECVs before election day. Indeed, Obama won both Iowa, by 5 points, and New Mexico, by almost 10 points.
Thus, 44 out of the 51 state contests had already been predicted before election day. I would give the polling community a perfect 44/44 score on its “safe” predictions.
My column just before the election put 253-191 as the cumulative score of ECVs for Obama and Romney at that point. Lacking only 17 more ECVs for reelection, Obama had many ways to win—either Florida alone or Ohio alone would be enough—yet could not be called safe to win.
Polls in six of the remaining seven states gave small, unsafe, leads to Obama. The exception was Virginia, where Obama was at +4 in a Washington Post poll, but Romney was at +2 according to Fox News (which I probably should have discounted). All in all, the odds were with Obama.
By midnight of election day, Obama had won five of the uncallable seven states—by 1.9 points in Ohio, 3.0 points in Virginia, 4.7 points in Colorado, 5.8 points in New Hampshire, and 6.6 points in Nevada. Their combined 50 ECVs thus raised Obama to 303, as of the time Romney conceded.
Only North Carolina went red (by 2.2 points); its 15 ECVs thus raised Romney to 206. Florida (29 ECVs) was still inconclusive, with Obama up by only 0.6, but had no more relevance and ultimately went blue. Final score: 332-206.
So I would score the polling community at six out of seven, or 86 percent, on its tentative predictions in the unsafe states. And I would put its overall score at 50 out of 51, or 98 percent, of the entirety of the contests.
No surprises, except maybe for Dick Morris, who told Fox News on Oct. 31 that he projected Romney to win by a score of 325-213, and this week claimed that he believed he was right at the time he said it.
* * *
Contact SWS: www.sws.org.ph or firstname.lastname@example.org.