• Presentation

Nexus Between Survey Breakoff and Unit Nonresponse

Citation

Peytcheva, E., & Peytchev, A. A. (2010, May). Nexus Between Survey Breakoff and Unit Nonresponse. Presented at AAPOR 2010, .

Abstract

Breakoffs in interviewer-administered surveys are different from those in self-administered modes. They are more likely to occur at logical breaks in the survey, arguably because they can be interpreted to some degree as separate survey requests (Groves and Kahn, 1979). Unlike in self-administered surveys where breakoffs occur more uniformly over the course of the survey (e.g., Peytchev, 2009), in interviewer-administered surveys breakoffs occur when perceived separate survey requests are introduced (e.g., between screening and main interview, and at section breaks such as the introduction of new topics and modules). Therefore, it is possible that breakoffs in interviewer-administered surveys share common causes with unit nonresponse. If so, methods to reduce as well as adjust for unit nonresponse may also be effective in addressing bias in face-to-face survey estimates due to breakoff; if not, new methods and separate models will be needed to address breakoff and nonresponse, as has been suggested for self-administered surveys.

We address this question using paradata from the National Health Interview Survey available on complete respondents, breakoffs, and nonrespondents. Paradata come from the Contact History Instrument, which includes summary descriptions of the contact attempt as well as survey organization actions such as leaving a note, “staking out” the household, and checking with neighbors. The paradata to be employed also include timing data and keystroke data that can also be used to identify common causes, when related to the likelihood of breakoff and likelihood of nonresponse.

We conduct two sets of analyses to address the degree of similarity between breakoff and unit nonresponse; we examine a causal link through common covariates and compare the direction and magnitude of bias in estimates due to likelihood of breakoff and of unit nonresponse. Implications for survey practitioners are discussed.