Why Are So Many Polls Getting Colorado Wrong?

All signs point to a close race for U.S. Senate between incumbent Democrat Mark Udall and Republican Congressman Cory Gardner. Only the national broadcast preemption during Sunday night's Broncos game could keep political ads off the air for a few hours in the country's most saturated media market. But, for the past few weeks, a steady stream of publicly released polls have shown Gardner in the lead, by as much as seven percent. There are many reasons to be skeptical.

Last weekend, Project New America released the results from two separate surveys that showed Udall ahead by three percent. HuffPollster suggested that the methodology of our surveys pointed to a possible "Hidden Udall Edge." This trend has a precedent. In 2010 and 2012, public poll averages in Colorado underestimated Democratic performance, especially compared to Democratic internal polls that were released.

Why is there such a difference between the public polls and our surveys -- and what does it mean?

Most pollsters are modeling their surveys to a "likely voter electorate" based on what previous mid-term elections looked like. Part of the reason many polls were wrong in '12, including Mitt Romney's campaign data, is that they predicted a much older electorate but turnout was higher than predicted. With an expensive, competitive election and Colorado's new voting procedure, the stage is set for a similar surprise this year.

Starting on October 14th, county clerks mailed out over 2.8 million ballots to registered voters. Voters can either mail them back or visit a county vote service center to vote or drop off their ballot in person. By comparison, in 2010, only 1.8 million people voted in Colorado. No one can be sure exactly who will vote this year -- but there is evidence that public polls may be significantly off in their predictions.

Let's focus on the "Dropoff Voters," the roughly 800,000 people who cast a ballot in the presidential election in 2012 but did not vote in 2010. They would have been excluded entirely from at least one recent poll because they may not be considered "likely voters." Compared to people who vote frequently, this audience is much younger, more mobile and votes Democratic more than Republican. Their vote counts exactly the same as a "likely voters" -- and both received ballots in the mail last week.

There are a few other important reasons to be skeptical of recent Colorado surveys.

According to the Catalist voter file, 51.4 percent of registered voters are women in Colorado. But, men were a majority in the recent Quinnipiac (50 percent) and High Point (52 percent) surveys. This flaw is problematic in a state where the gender gap has been hugely important and women are the focus of constant campaign communication.

Latino voters make up 14.2 percent of Colorado voters according to the Pew Hispanic Center. UNLV Professor David Damore presents a strong case for why most polls are missing Latinos - who overwhelmingly favor Udall. Some surveys interview too few people to even report their Hispanic numbers and only one offered respondents a Spanish-language option.

Coloradans are more likely to be "cell phone only" users than the national average. Some polls contact cell phone users, but most are not getting enough respondents. There are important political differences between people with landlines and cell-only voters.

When you consider the drop-off voters, women, Latinos and cell-only houses, there are plenty of ways for public pollsters to miss these voters who are predominantly supporting Udall.

No one poll is a perfect snapshot of reality -- which is why it makes sense to look at multiple surveys, averages and aggregations. In Colorado, all of the recent public polls have one or more major methodological shortcomings. One firm even used two different approaches in two polls on the same day, with similarly strange results. The bottom line is there is real reason to be skeptical of the polls (and their averages) that missed CO in '10 and '12.