How Gallup missed the mark ― again

On the eve of Election Day, the Gallup Poll indicated that Mitt Romney would become president. His margin of victory: one percentage point ― 49 percent Romney to 48 percent Obama.

This projection was based on telephone interviews of 2,551 likely voters in all 50 states and the District of Columbia conducted between Nov. 1 and Nov. 4.

The size of Gallup’s sample allows for a narrow plus or minus 2 percentage point margin of error. In other words, Romney’s anticipated 49 percent vote tally translates into him garnering anywhere between 47 percent and 51 percent of the vote.

At this point, in fact President Barack Obama received 50.6 percent and Romney 47.8 percent.

Technically, Gallup’s Romney prognosis fell within the margin of error but just barely.

More on the nose was Gallup tipping its hat to the wrong winner, a heresy among pollsters who snag paying clients based on accurate forecasts, not methodological equivocations.

Gallup has been through this before. In 1948, the Gallup Poll picked New York Gov. Tom Dewey to beat another incumbent president, Harry Truman, a predicted triumph captured for posterity in the banner headline of the Chicago Tribune. That misstep forced the organization to do a comprehensive reevaluation ― and eventual revamp ― of its methods of surveying opinions.

No such major overhaul has been announced by Gallup, yet. When recently asked about its shaky 2012 projections, Gallup’s editor in chief, Frank Newport, predictably reported that the forecast was within the margin of error. He did concede that his organization underestimated the actual voting behavior of young people, African-Americans and Hispanics.

The culprit was faulty filter questions used to identify likely voters, Gallup’s principal basis of calculating electoral outcomes. Probable voters tell Gallup pollsters they know where to cast a ballot, and that they are interested in politics, will vote and have a prior voting history. Newport notes these questions “need to be and will be reviewed by Gallup. We’ve been using those questions for decades, but times change.”

For a couple of reasons, however, there may be more fundamental problems with telephone polls than simply the questions asked to winnow out unlikely voters.

First, most survey organizations, according to the Pew Research Center, achieve only one usable survey out of every 10 households they attempt to contract. This current response rate is three times lower than it was 15 years ago. Gallup fails to report its response rate. Pew claims that respondents are still representative of the U.S. population, with one important exception ― education. Less educated people are 10 percent less likely to participate in telephone polls.

Second, one-third of U.S. households depend on cellphones, not landlines. Unlike Gallup, many survey organizations do not contain cellphone numbers in their sampling databases. Regardless, Pew documents that response rates are lower for cellphone customers than for those on landlines. Consumers who usually have a limited amount of “free” cellphone time are not inclined to burn their precious minutes talking to a pollster.

Many lower-income people are probably choosing not to sign costly, lengthy contracts with communications’ providers. Not when prepaid and “throwaway” cellphones, the use of which is not tethered to binding contractual obligations, are readily available. No polling organization has faced up to the sampling problems posed by these phones.

Low response rate and the lack of inclusion of the full array of cellphones in sample call-bases means that younger voters as well as low-income and less educated people are underrepresented in telephone surveys.

Gallup, along with other polling firms, compensates for demographic undercounts through weighting. Weighting is the practice of over counting the members of the group who are in the sample to the point where their numbers reflect their actual proportion in the universe, as derived from U.S. Census figures.

The Pew Hispanic Center estimates that Latinos constitute 11 percent of the nationwide voting-eligible population. If a sample of 1,000 eligible voters contains only 80 Hispanics — 8 percent of the total ― a polling organization will weight each Latino sampled by a factor of 1.3. This recalibrated Latino group now counts as 11 percent of the full sample ― the same proportion of voting-eligible Hispanics in the general population.

The problem is obvious. In terms of income, education and age, sampled Hispanics are more affluent, better educated and older than their ethnic counterparts. Projections from such weighted samples underestimated true Hispanic support for Obama.

This is an error that Gallup needs to fix.

Jim Lamare is an editor with Hispanic Link News Service. Reach him at jwlamare@gmail.com. Distributed by Scripps Howard News Service, www.shns.com. <The Korea Times/Jim Lamare>