It has been hypothesized and demonstrated that a number of languages (verbal, numeric, symbolic, and graphic) combine to affect respondents’ perception and comprehension of visual information. Verbal language refers to the words on a questionnaire; numeric refers to the numbers; symbolic refers to symbols on a questionnaire, such as arrows; and the graphic refers to the color, brightness, shape, and location of the information. Branching instructions were used as an initial test of this hypothesis because they yield objective measures of performance. A classroom experiment with college students, in which three of the languages (the verbal, symbolic, and graphic) were altered in two distinct ways (the Prevention and Detection methods) and tested against the Census 2000 method of branching, provided initial evidence in support of this hypothesis. This report documents an experiment conducted in Census 2000 in which the two branching instructions from the classroom experiment were revised, and two additional instructions were developed ( reverse printing the instruction and substituting the words "go to" for "skip to") and tested against the Census 2000 version.

The results of this experiment suggest that altering the languages of the branching instructions did not affect mail response rates nationally or by response area. However, it did have a substantial effect on navigational errors. The Detection method significantly reduced commission errors (respondents answering questions they were instructed to skip) from 19.7% to 13.5% and omission errors (respondents not answering questions they were supposed to answer) from 5.0% to 4.0%, and should be adopted in its present form, though further improvements may be possible.