Go to the Globe and Mail homepage

Jump to main navigationJump to main content

Voters wait in line at the Bobby Miller Activity Center in Tuscaloosa, Ala., Tuesday, Nov. 6, 2012. (Dusty Compton/AP)
Voters wait in line at the Bobby Miller Activity Center in Tuscaloosa, Ala., Tuesday, Nov. 6, 2012. (Dusty Compton/AP)

Wrong calls on 2012 vote prompt Gallup to review methods Add to ...

One of the world’s top polling firms said it will adjust some of its methods after facing criticism for inaccurately predicting the results of the 2012 U.S. presidential election.

Gallup said a review found four factors contributed to a final 2012 pre-election poll that forecast Republican candidate Mitt Romney winning 49 per cent of the popular vote and Barack Obama 48 per cent. Mr. Obama ended up winning the popular vote by nearly four percentage points. These are the four mistakes identified by Gallup.

Misidentifying likely voters

Gallup produced a likely-voter sample of 74 per cent of the adult population, higher than any other firm, and a sample that also gave Mr. Romney a four-point edge.

Like other firms, Gallup sorts respondents into “likely voters” and “non-likely voters” based on seven self-reported turnout questions. The firm says what likely gave Mr. Romney a lead was including the question “thought given to the election” in their scoring. Gallup says the removal of this question would have changed its final estimate.

Underrepresenting regions

Gallup conducts polling across all regions of the United States. The firm said interviewing during the election campaign was staggered throughout the evening based on local time within each time zone, and did not start before 5 p.m. But because there is no minimum sample size required in each time zone, Gallup says this opened up the door to too many interviews being conducted in a region’s earlier time zone, and too few in a region’s later time zone. Simulations after the election revealed this likely underestimated Mr. Obama’s vote share.

Underestimating ethnic vote

Gallup says the way it collected race and ethnicity information, using basic “yes” and “no” questions, overestimated the white vote while underestimating the Hispanic and black vote.

Throughout the election polling period, the percentage of respondents identifying as multi-race, as well as the combination of white and American Indian, and white and Alaska Native, were higher than the census estimates for the adult population. This, in turn, made the task of adjusting these into broader categories more difficult and could have produced “unknown effects” on overall weighting.

Land lines vs. cellphones

Gallup called both land lines and cellphones, but the firm used public directories to get its numbers, rather than the through random digit sampling (RDD). In an experiment after the election, Gallup determined that respondents from a sample of RDD numbers were younger, more democratic and more likely to approve of Mr. Obama than respondents at numbers listed in directories. Thus, the land lines pollsters were calling during the election campaign were, in all likelihood, disproportionally older, more Republican and more likely to vote for Mr. Romney.

Follow on Twitter: @danbitonti

In the know

Most popular video »

Highlights

More from The Globe and Mail

Most Popular Stories