No abstract

No abstract

No abstract

IN SAMPLE SURVEYS conducted by probability methods, from 20 to 30 percent of the designated sample typically are either never contacted or, once contacted, never interviewed. This raises a series of questions about the nature and magnitude of nonresponse bias.The National Opinion Research Center (NORC) was commissioned by the U. S. Department of Agriculture (USDA) to study attitudes and behavior of farm operators who supply information in USDA crop and livestock surveys. 1 Hence the attitudinal, biographical, and behavioral variables measured in the survey were those we thought most likely to be related to nonresponse. These data thus provide an excellent opportunity to estimate the effects of nonresponse. If we can determine the nonresponse bias for these variables, other constructs measured in this sort of sample should not exhibit worse bias.There seem to be two main factors contributing to nonresponse, each with different causes and hence different biasing effects. One is availability to be interviewed, including the availability of other 1 See the general report on the survey in Jones et al. (1979). This paper is an elaboration of a part of Chapter 6 of that report, "Analysis of Errors." That chapter also has an analysis of question wording effects and sampling variability. The questionnaire is reproduced in the report, which is available from the NORC Library.Abstract By separating respondents in a survey of farmers into those who refused to be interviewed and were converted, those who were hard to reach, and those who gave no problems, we can estimate nonresponse bias from refusers not converted and respondents not reached. We show that for attitude questions the refusal bias is serious, ranging from 2 to 4 percent on many attitudes, and very often larger than the standard error in surveys of ordinary size.Arthur L. Stinchcombe is Professor of Sociology at

No abstract