Research has shown that questionnaires with skip instructions produce higher item nonresponse than those without (Featherston and Moy, 1990: Messmer and Seymour, 1982). In addition, the failure to process skip instructions correctly leads to inconsistent responses between questions (e.g., Gower and Dibbs, 1989; Zuckerberg and Hess, 1996).

Two kinds of errors can be made as a result of skip instructions. An error of commission occurs when a respondent is instructed to skip over the following question (or questions), but instead answers it. An error of omission occurs when a respondent is supposed to answer the next question, but instead skips it. Both kinds of errors happen with substantial frequencies. For example, in an analysis of Census questionnaires, Raglin (1997) found that nearly 55 percent of the respondents who responded that they were working answered a series of questions about their non-work status (an error of commission). Conversely, 24 percent of the respondents who said that they were not working did not respond to the non-work questions (an error of omission).

A recently developed theory of self-administered questionnaire design posed visual techniques for getting respondents to follow a prescribed navigational path through a questionnaire (Jenkins and Dillman, 1995, 1997). Although this theory led to suggestions for gaining compliance with skip instructions, the suggestions were not empirically tested. Our purpose in this paper is to report results from an experiment in which two quite different skip instruction designs derived from the aforementioned theory were tested against the method of providing skip instructions used in the 2000 Census. The experiment was designed to test whether both errors of commission and omission could be reduced.