PollTracker (PT) aggregates and averages all reputable public polls for a wide range of presidential, congressional, gubernatorial and state-level issue contests across the United States.
Q1: How do you decide which polls to include?
A1: Our methodology is based on maximal inclusion of a wide range of public polls. PollTracker’s core approach is not to evaluate the relative quality of different polls or different polling organizations but to put together all public polls.
The editors use their judgment to reject polls about which serious questions of methodology and/or integrity have been raised, such as an unaccepted survey type (e.g. a “callback” survey), or the sample being dramatically skewed toward one party, age group or region. In close cases, we confer with our outside consultant, Professor Charles Franklin of the University of Wisconsin, one of the nation’s foremost experts in the statistical measure of public opinion data. But PollTracker errs on the side of inclusion.
At present, PollTracker includes traditional phone polls and ‘robocalls’ on the same basis. Internet polls such as Zogby and YouGov are collected and listed but are not included in our polling averages.
Q2: How do you calculate your averages?
A2: PollTracker uses the LOESS “regression analysis” library to generate its averages. The LOESS analysis is a statistical formula that integrates more current data with past polls in order to point a line in the direction the numbers are heading. PT “averages” are the endpoint of that line. The regression analysis was customized and developed in consultation with Professor Franklin. Its function is to derive trends from scatterings of polls, represented by the curved lines, taking into account both the different poll numbers and their progression over time.
PollTracker also uses a process we call “segmenting” to allow us to present data separated by large gaps of time on the same chart. PollTracker generates a “regression analysis” — the smoothed, curvy lines on the graphs — for data sets with at least 10 data points and no gaps between data points of greater than 6 weeks. Gaps in the data of greater than six weeks are represented by a dotted line. Sets of continuous data with fewer than ten data points are calculated with a weighted mean average and represented by straight lines.
Q3: Is there any editorial input in the averages? Can they be adjusted in any way?
A3: No. Other than rejecting clearly flawed or fraudulent polls (see A1), all polls are treated equally, and there is no editorial input whatsoever.
A4: Short answer: We don’t. PollTracker lists each day’s results so readers can keep up with the latest public results. But since tracking polls are “rolling” sets of data with some of the same data included over multiple days, we never enter duplicative data into the formula that produces the averages. Since the Gallup and Rasmussen tracking polls survey much more frequently than other pollsters, we “down weight” the numbers to ensure that their numbers and “house effects” do not overwhelm the results from pollsters that do not poll as often.
Q5: Why do you count partisan polls?
A5: PollTracker counts partisan polls, including internal polls commissioned by campaigns and released to the public with full data, to keep with the principle of maximal inclusion. Partisan pollsters and polls are indicated as such in their listing.
Q6: How do you determine who is a “partisan” pollster?
A6: Pollsters have to self-identify as aligned with a certain party or outwardly state their ideology in general terms (conservative, liberal/progressive).
One of the major functions of the “Pollster” filter on PT graphs is to let users work with the data themselves and remove specific pollsters. Clicking on the tab marked “Pollster” and unchecking the pollster you wish to exclude will redraw the graph without that pollster’s data. Click “Reset Filters” to bring the graph back to its original form.
Q7: How do you determine who pops up in the poll widget in the skyboxes on story pages?
A7: PollTracker editors feature the top non-presidential races in the country at the top of all TPM pages (called “skyboxes”) except for the homepage, which has a special presidential widget. The races are chosen based on the freshness of the data, the importance of the race or the newsiness of the campaign.
Q8: How do you determine a featured poll on the PollTracker landing page?
A8: PT editors use a similar process as choosing the skyboxes — they flag particularly important polls that may show an update, milestone or change in a race.