One consideration when writing survey questionnaires is using a global question that asks respondents to report a measure in one question versus decomposing the question into a series of items. Currently several U.S. Census Bureau household surveys are undergoing modernization efforts to add web self-response modes to their interviewer-administered surveys. Some of these surveys use global questions that contain definitions or details about what to include or exclude in their response. While an interviewer may use their judgement to explain these instructions or definitions to the respondent as needed, these global questions with complex instruction text could cause frustration in the web self-response mode leading to more breakoffs or satisficing that could harm data quality.

This study tests a global utilities expense question against two formats of a decomposed utility expense series to examine data quality and respondent burden. The global utilities question contains examples of utilities that should be included and excluded from the amount. The decomposed versions use two questions: the first identifies the types of utilities the household pays and the second asks for the expense amounts for each of the types of utilities reported in the first item. While the global utilities expense question is completed faster by respondents, the respondents in this question version reported including fewer utility types in their answer than in the decomposed versions and reported including utility types that the instructions directed them to exclude. Conversely, item missingness was higher for the decomposed versions than the global version. This experiment suggests that survey managers and subject matter experts should carefully weigh the tradeoffs between different measures of data quality and response time when deciding whether to use global or decomposed questions for complex concepts.