This article originally appeared at Politico.com
Is it possible that Colorado’s Cory Gardner is shaping up to be this election cycle’s Sharron Angle? You might recall what happened to Angle in the Nevada Senate race in 2010. Almost every pre-election poll had Angle, the tea-party-supported Republican challenger to Senate Democratic majority leader Harry Reid, leading the race handily. On the eve of the election, no less a polling personage than Nate Silver of FiveThirtyEight pegged Angle’s chances of winning at 83 percent. But not only did Reid win, he beat Angle by a comfortable 5.7 points .
What went wrong with the polling four years ago should be a cautionary tale for what’s about to take place on Tuesday, with future control of the Senate expected to tilt to the Republicans. In Colorado, according to the pre-election polls, Republican Gardner is leading incumbent Democrat Mark Udall in the Centennial State’s Senate race. But dig a little deeper and there seems like there’s something off about the numbers—potentially the same problem that plagued Nevada’s race four years ago.
A few pollsters did manage to predict a Reid victory in 2010—among them our firm, Latino Decisions, which based its projection in part on our pre-election polls of Latinos. The reason we got it right is that we survey Latinos properly, supported by bilingual callers who administer the surveys in the language respondents use when answering the phone. We also check respondent samples against known, Census-based demographics of the Latino electorate (nationally or in specific states) to safeguard against the results being comprised by higher subsamples of English-speaking, college-educated or American-born Latinos who, on average, are more likely to vote Republican.
The familiar “garbage in, garbage out” maxim applies: Bad sampling methods tend to yield bad estimates. Cultural competency is the critical currency of survey design, sampling and administration, and until pollsters take the surveying of the bilingual, bicultural Latino community seriously, surprises like the 2010 Reid/Angle race will continue.
Indeed, in his post-election analysis of how he and other prognosticators had gotten Nevada wrong, Nate Silver confirmed that biased Latino samples likely produced the misleading forecasts.
We compare the 2010 Nevada result with this year’s Colorado Senate race not to boast, but to forewarn: Until and unless pollsters begin to properly sample and survey minority voters—and in particular the Latinos and Asian-Americans who comprise most of the so-called “rising electorate”—there will be other election day surprises like the Reid-Angle outcome.
Flawed polls of Latinos this cycle have also yielded unrealistic sub-samples and curiously low levels of Latino support for Democrat Charlie Crist, in the race for governor of Florida, and for Michelle Nunn, a Democrat who is in another tight race for U.S. senator in Georgia, although in those cases the mistakes might not be large enough to alter the outcome.
Sure, there’s biennial hand-wringing about whether landline-based polls are keeping up in an era where so many people—especially young people—are switching to cell phones. But today’s polling challenge goes much deeper than mere technology becausepolling hasn’t keep up with an electorate that increasingly looks—and thinks—differently.
Too often call centers are staffed with English-only interviewers. Samples of minority respondents may not reflect the religious identity, educational attainment or income levels of the broader community. Discrepancies like these remain common in polls of minority voters.
In ten statewide (mostly media-sponsored) polls taken in Colorado in the past five weeks, for example, nine showed Gardner leading Udall by single-digits among Latinos. Given that our recent statewide poll of Colorado Latinos shows Udall leading Gardner by a whopping 41 points, 55 percent to 14 percent, we’d be shocked if Gardner carries the Latino vote.
In a close race like the Udall-Gardner contest, polling discrepancies of this magnitude—especially in a state where Latinos were 12 percent of the electorate in the 2010 midterm and comprise an even larger percentage today—can alone account for a seemingly last-minute reversal from an apparent Gardner “lead” to an actual Udall victory, when in fact Udall may have been ahead all along.
As fivethirtyeight.com’s Harry Enten notes, more recent statewide polls show Udall leading Gardner among Latinos by margins of 22, 28 and even 73 points. These numbers seem somewhat more plausible to us. But it’s difficult not to conclude that inconsistent polling contributed to Udall’s Latino margin vacillating, in the two most extreme cases, from -8 points (Quinnipiac, Sep. 10-15) to +73 points (Suffolk, Oct. 18-21) within just five weeks.
Maybe Gardner will be the latest victim of systematic polling bias; maybe not. But does it really make sense that his support from Latinos in the statewide Colorado polls has been dropping—and in some cases plummeting—yet he has held or even widened his lead over Udall?
Perhaps Gardner has broadened his support among the more than 80 percent of non-Latino Coloradans. Or perhaps there’s something fishy in those polls.
On Tuesday, we’ll find out.
Dr. Matt Barreto of the University of Washington is the co-founder of Latino Decisions polling firm, and co-author of Latino America: How America’s Most Dynamic Population is Poised to Transform the Politics of the Nation (Public Affairs, 2014). Dr. Thomas F. Schaller of the University of Maryland at Baltimore County, is Political Director for Latino Decisions, and author of Whistling Past Dixie.
