Determinants of unit nonresponse in multi-mode data collection: A multilevel analysis

Area of Science:

• Survey methodology
• Public health research
• Statistical analysis

Background:

• Multi-mode data collection, using sequential data collection stages (e.g., interview followed by questionnaire), is common in surveys.
• Partial nonresponse, where participants do not complete all survey stages, can significantly bias results but is understudied.
• This research focuses on understanding why individuals do not complete a self-administered questionnaire after a face-to-face interview.

Purpose of the Study:

• To investigate the factors contributing to partial nonresponse in a multi-mode survey.
• To identify participant, interviewer, and contextual characteristics associated with nonresponse to a self-administered questionnaire following an initial interview.

Main Methods:

• Utilized data from the Belgian Health Interview Survey 2013.
• Analyzed partial nonresponse (n=1,464) among initial interview completers (n=8,133) using multilevel logistic regression.
• Included random interviewer effects to account for interviewer-level variability.

Main Results:

• Interviewers accounted for a substantial portion (46.6%) of nonresponse variability, even after controlling for area.
• Partial nonresponse was higher among younger individuals, non-Belgians, those with lower education/income, and residents of Brussels and Wallonia.
• Nonresponse was also linked to perceived poor health, interviews conducted later in the survey year, and the total number of interviews an interviewer conducted.

Conclusions:

• Interviewers play a critical role in partial nonresponse in sequential multi-mode surveys.
• Participant and interview circumstances explain only part of the interviewer effect, highlighting the need to explore other interviewer characteristics.
• Emphasizes the importance of interviewer training and motivation to mitigate nonresponse in complex survey designs.