In Colorado, two more polls show that the Republican is ahead. We look more closely at the strange ways one of them does things. Most Americans still don’t worry about getting Ebola. Today is Thursday, October 23, 2014, and this is HuffPollster.
COLORADO: NEW POLL DATA BENEFIT GARDNER – The race for the Senate seat in Colorado has been one of the closest and most watched of the whole campaign. Two new polls that came out on Wednesday show that Republican Cory Gardner is pulling ahead of Democrat Sen. Mark Udall. A telephone poll from USA Today and Suffolk University gives Gardner a 7-point lead (46 to 39), while a new Internet poll from Reuters/Ipsos gives Gardner a 2-point lead (47 to 45 percent). Overall, nine of the eleven polls done in October have given Gardner a lead of at least one point. The only things that have been different are two internal polls that came out last week and were paid for by Democratic groups that support Udall.
Based on all public polls but tuned to better-performing, non-partisan polls, the HuffPost Pollster poll tracking model puts Gardner’s lead at just over 2 percentage points (48.4 to 46.1 percent as of this writing). As of this writing, Gardner has just over a 60% chance of winning the race. This is because the margin is still close, polls aren’t always accurate, and late changes and polling mistakes are possible.
Part I: Calling Quotas: How the Suffolk Poll Was Done – The USA/Today/Suffolk University polls, like many other state-level public polls, use samples of randomly generated landline phone numbers (called “random digit dial” or “RDD” by pollsters) along with samples of mobile phone numbers taken from directories kept by data vendors. Suffolk, on the other hand, deals with the problem of low response rates in an unusual way. The Suffolk poll doesn’t correct for statistical bias in the demographics by weighting the data after it’s been collected. Instead, “calling quotas” are used as survey calls are made to correct for statistical bias. David Paleologos, who is in charge of the Suffolk Poll, told HuffPollster that for each survey in Colorado, quotas for each region, gender, and political party are filled by calling people on the last night of calling. “If quotas aren’t met,” he said, “we reserve the right to weight the data or field another day/night to meet the quotas.” The Colorado poll’s final results were not based on any weights.
Paleologos said that most of the time, these quotas make sure that the number of people from each region in the completed samples and the number of people from each party registered in each region match statistics from official voting records from previous elections. For the midterm elections in 2014, Suffolk also set quotas so that each region would have about 48% men and 52% women. This was done to match the results of previous exit polls. So, the most important demographic factors for it are region, gender, and party registration.
What about age and skin colour? Paleologos tells HuffPollster, “It does get harder with age and race.” “We do our best to make sure that age and race are filled in correctly based on the best data we have, whether it’s from the Census or from exit polls.”
Methodology of the Suffolk Poll, Part II: Screen for Likely Voters. The Suffolk Poll also uses a fairly simple screen to choose likely voters. Based on a single question, it interviews registered voters who say they are “very likely” to vote or who have “already voted,” but not those who say they are “somewhat likely,” “not very likely,” or “not at all likely” to vote. This way of doing things got attention on Twitter: