Summary of the Video
What percent of the American work force was unemployed last month? Dr. Janet Norwood, Commissioner of Labor Statistics, is seen presenting the latest data to Congress and the media. These data come from a large sample survey conducted monthly by the Census Bureau. How are our views on social issues changing over time? An important source of data is the yearly General Social Survey (GSS) conducted by the National Opinion Research Center (NORC) at The University of Chicago. The GSS interviews a sample of about 1500 individuals each year.
[Dr. Norwood, head of the Bureau of Labor Statistics since 1979, retired on January 1, 1992. The New York Times said (December 31, 1991) that she left with “a near-legendary reputation for non-partisanship and plaudits that include one senator's designation of her as a ‘national treasure.' ”]
One foundation of careful sample surveys such as these is good sample design. The design describes how the sample is selected. Without good design, even a large sample can give erroneous results. In 1932, the Literary Digest took a large but poorly designed sample, while young George Gallup took a smaller but well-designed sample before the presidential election. Gallup predicted the Roosevelt landslide, while the Literary Digest predicted a Landon victory.
Random sampling is fundamental to good sampling design, but for practical reasons national surveys don't use simple random samples. They take their samples in stages: first counties, then towns within counties, then households within towns, for example. Some of the stages in such a multistage sample often use a stratified sample. Counties are first classified as urban, rural, or suburban, then a separate random sample is selected from each stratum. This ensures that cities are in our sample even though most counties are rural.
Good statistical design is only part of a careful sample survey. Questionnaire design and skillful interviewing are also essential. We see some bad examples: leading questions, interviewing people where others can overhear responses, ignorant responses, and the effect of the interviewer on responses. Then we see NORC testing questions in “think-aloud” sessions and training its interviewers through role-playing.