There a large body of research about the sources of error in surveys and how to get the most accurate data by minimizing sources of error. One of the important reasons for working with CSR is the knowledge Center staff members bring to the design and execution of a survey. Moreover, the Center has taken a leading role in doing many studies of survey error and how to reduce it.
INTERVIEWERS can affect the quality of data collected, and CSR did one of the first and largest studies of the role of interviewer training and supervision in reducing interviewer-related error. CSR researchers also teamed with a group at the University of Michigan to develop behavior coding, which is an innovative way to use pretest interviews to identify questions that increase interviewer-related error, as well as identifying questions that pose problems for respondents to answer.
COGNITIVE TESTING of survey questions before using them in a survey is one of the best ways to identify questions that are not consistently understood and for which respondents have problems providing valid answers. Cognitive testing has been a routine part of surveys at CSR for nearly 30 years.
THE MODE OF DATA COLLECTION, whether the survey is carried out by in-person or telephone interviewers or using the mail or the Internet to collect data without using interviewers, is one of the most important aspects of survey design. Each approach has strengths and drawbacks. CSR has been continuously studying how best to use different modes and how the mode of data collection affects the answers provided and the characteristics of those who do and do not respond to surveys.
NONRESPONSE is perhaps the biggest problem facing survey research in recent years. Response rates have dropped dramatically. Sometimes low response rates have big effects on survey estimates, and sometimes they do not. CSR has been continuously doing studies of how estimates are affected by nonresponse, how that varies with the content of a survey and the mode of data collection, and how best to design surveys to maximize response rates and minimize the effects of nonresponse on the accuracy of survey estimates.