Identifying patterns of item missing survey data using latent groups: an observational study.

Area of Science:

• Health survey methodology
• Data analysis and statistics
• Epidemiology

Background:

• Missing data in surveys is a common challenge.
• Understanding patterns of item missingness is crucial for data integrity.
• Previous research has not fully explored respondent grouping based on missing data patterns.

Purpose of the Study:

• To identify and characterize distinct groups of survey respondents based on their item missing patterns.
• To investigate the relationship between participant characteristics and their latent classes of missing data.
• To determine if item missingness in one survey wave predicts missingness in subsequent waves.

Main Methods:

• Latent class analysis with a mixture of multinomial distributions was employed.
• Bayesian information criterion was used for model selection.
• Logistic regression analyzed associations between participant characteristics and latent classes, and predicted wave missingness.

Main Results:

• Four percent of participants exhibited significant item missingness, often in the middle of the survey.
• A large majority completed most questions, with a notable exception for a poorly presented sleep question.
• Younger, more educated participants were more likely to complete nearly all questions.
• Higher completion rates in one wave predicted lower missingness in the subsequent wave.

Conclusions:

• Identifying patterns of item missingness provides valuable insights into data generation processes.
• This understanding can inform the design of future surveys to minimize missing data.
• Targeted interventions based on identified patterns can enhance data quality in longitudinal studies.