December 29, 2010

How accurate are the statistics derived from Pew Research polls?

Q. How accurate are the statistics derived from Pew Research polls when applied to the population of the United States?

The accuracy of polls can be judged in different ways. One is the degree to which the sample of the public interviewed for the poll is representative of the whole population. For example, does the poll include the proper proportions of older and younger adults, of people of different races, or of men and women? Another standard for accuracy is whether the poll’s questions correctly measure the attitudes or relevant behaviors of the people who are interviewed. For both of these ways of judging accuracy, Pew Research’s polls do very well. We know that key characteristics of our samples conform closely to known population parameters from large government surveys such as the U.S. Census. Similarly, our final polls in major national elections have a very good track record of measuring intended behavior because they accurately predict the election results (we came within one percentage point of predicting the margin of victory for Barack Obama in 2008, and Republican candidates for the U.S. House in 2010).

To improve the accuracy of our polls, we statistically adjust our samples so that they match the population in terms of the most important demographic characteristics such as education and region (see our detailed methodology statement for more about how this is done). This practice, called weighting, is commonly used in survey research.

Scott Keeter, Director of Survey Research, Pew Research Center