The Associated Press-GfK Poll released Wednesday marks the start of a new era for AP polling. From now on, regular polling from the AP and GfK, a global survey research company based in Germany, will be conducted online, interviewing from a unique group of willing survey takers that is representative of all Americans.
Polling that draws on what are known as online panels is widely used for many purposes, but most of it does not meet AP's standards for publication. That's because typical online research leaves out the 15 percent of Americans who have no access to the Internet, and most online panels are made up of people who have clicked through online ads or been referred by friends to sign up to complete surveys in exchange for rewards.
No one yet knows whether the people who have signed up for online panels this way really represent the views of the broader public. But research has shown that those who lack access to the Internet are quite different from those who are connected, and there is no way for them to participate in the vast majority of online panels.
The science behind polling says that in order for survey research to work, respondents must be randomly selected rather than self-selected and everyone in the population must have a chance of being asked to participate.
GfK addresses this by recruiting its panel from a random selection of all Americans with either a telephone or a home. That's pretty much everyone. GfK calls it KnowledgePanel.
Recruiting this way means that KnowledgePanel does reach people who do not have Internet access. GfK provides a connection and a laptop to those who agree to participate and do not already have access. Once they're selected, poll participants earn rewards as they complete surveys.
The AP has used KnowledgePanel to conduct polling in the past, most notably through the 2008 AP-Yahoo! panel study tracking voters' views throughout the presidential campaign that year.
Before making the switch for our regular polling, the AP and GfK conducted parallel polls of the American public in April. One used a telephone sample including both landline telephones and cellphones; the other used KnowledgePanel. There were some differences between the two samples due to sampling error, but most significant gaps were the kind pollsters have come to expect due to what are known as "mode effects."
For example, people sometimes answer questions differently when they're talking to a person on the phone than they do when they're filling out a questionnaire on their own. Research has shown that people are more willing to express opinions that may be unpopular or private when they're filling out a survey by themselves than if they are talking to an interviewer.
Most prominent public polling is conducted using telephone samples, which cover a random selection of people who have either a cellphone or a landline telephone. Polling conducted this way has been shown to be representative of the overall population, yet falling response rates and rising costs have battered the industry.
Telephone polling has faced a rough road in the past 15 years or so as Americans have begun cutting their cords and screening their calls. And online research has not proven itself a perfect fix yet.
Using KnowledgePanel allows the AP and GfK to conduct research that relies on scientific principles while taking advantage of the new ways Americans communicate.
Follow Jennifer Agiesta on Twitter at: http://www.twitter.com/JennAgiesta
AP-GfK Poll: http://www.ap-gfkpoll.com
AP polling FAQs: http://www.ap.org/company/faqs/polling
EDITOR'S NOTE _ Digits is Associated Press Director of Polling Jennifer Agiesta's take on the numbers that reflect our world and the survey research techniques used to find them.