How this survey was conducted

Introduction

This study was fielded over the internet from April 13th – 19th with 665 American adults selected as a representative sample of the American public. They are part of a nationwide panel of Knowledge Networks, a company started by two Stanford professors.The Program for Public Consultation, affiliated with the University of Maryland’s School of Public Policy, conducted the survey.

The panelists were recruited through a scientific process of selection using two methods: a random selection of possible US telephone numbers (also called random digit dial sampling, or RDD); and a random selection of residential addresses using the United States Postal Service’s Delivery Sequence File (a complete list of all U.S. residential households — including households that have only cell phones, and are often missed in random-digit-dial sampling).

Persons were then invited by telephone or by mail to participate in something called the KnowledgePanel. Those who agree to participate but do not have Internet access were provided a laptop computer and Internet service. A representative sample is then chosen from all the panelists for a specific survey. Once that sample completes a survey the demographic breakdown of the sample is compared to the US census. Any variations from the census are adjusted by weighting.

Conducting surveys with this type of representative sample on line has proven superior to standard telephone surveys, because it is drawn from a pool that is more complete demographically. Also respondents can take as much time as they like to read and respond to questions, thus increasing the thoughtfulness of their answers. This method has increasingly become the preferred method for academic and government-sponsored surveys.

A recent study by the American Association of Public Opinion Research concluded that while ‘opt-in’ Internet panels have limited reliability for general population studies, probability-based Internet panels produce high quality data. This particular survey has a margin of error of plus or minus 4.8%.

More technical information is here.