Nonresponse

Classification

Calculating nonresponse is not as simple as it initially seems. One survey may have mainly nonrespondents who could not be reached (i.e. contacted) by the [Page 528]researchers. Another reaches most of the sampled persons, but a large proportion refuse the survey. A third has an easily reached, cooperative set of people in a sampling frame, but many do not possess a characteristic that is part of the definition of the desired survey population. These reasons for nonresponse, usually termed noncontact, refusal, and ineligibility, respectively, are differentiated within most classification procedures for nonresponse, such as the Standard Definitions established by the American Association for Public Opinion Research (AAPOR). The components of nonresponse may differ across the various contact types or modes of survey. For example, the noncontact count may be higher on a telephone survey in which calls to some numbers are never answered than on a postal survey in which a letter mailed is presumed to be a contact, except for a few nondeliverables returned by the post office. On the street, as distinguished from the logical world of survey methodology research articles and textbooks, the distinctions among noncontact, refusal, and ineligibility are not always watertight. Is a foreign-born person who opts out of a survey due to language problems, or an elderly citizen with impaired hearing or eyesight, an ineligible or a refusal? Only the person himself or herself really knows. What of those who, in a face-to-face household survey, recognize the approaching person as a survey taker and decline to answer the doorbell? Refusal or noncontact?

Predictability

Nonresponse, like many social behaviors, is only weakly predictable at the level of the individual yet becomes more ordered at the aggregate. Any experienced survey methodologist, told the type of population to be sampled, the mode of contact (i.e. telephone, in-person, mail, Web, or multi-mode), the length and content of the questions, the resources available for callbacks and follow-ups, and whether or not payments or other incentives are to be used, can give an informed estimate of the final response and nonresponse rates. Information on the sponsor, and whether the fieldwork is to be conducted by a government agency, a survey unit at a university, or a commercial firm, will add further precision to the estimates. At the same time, and in most cases, the attempt at prediction of whether one person versus another will respond or not generates only slight probabilities. Always depending on the particulars of the survey, people of greater or lesser education, male or female, racial majority or minority, young or old, may be disproportionately among the nonrespondents. Such effects often appear in surveys, but few generalities are possible. Perhaps the safest is the tendency for middle, rather than upper or lower, socioeconomic status people to respond: the middle-class bias sometimes attributed to surveying. Even this, however, is not a certainty. For example, on a survey whose results will help generate funds for social programs, the lower socioeconomic strata may have good reason to respond. If the survey is conducted by an agency of government, there may be some deference to the authority of the sponsor among the lower strata, also contributing to probability of response. Offsetting those advantages might be a subcultural hesitancy and anxiety about participating in surveys. The probability of an individual responding to a survey remains inherently hard to predict, because so many factors enter the mix. (An exception being panel surveys in which response behavior in a previous wave can be highly predictive of subsequent wave participation.)

...