Two-part models for repeatedly measured ordinal data with "don't know" category.

Area of Science:

• Statistics
• Biostatistics
• Behavioral Science

Background:

• Ordinal data with "don't know" responses are common in medical and social sciences.
• Standard models often ignore the ordinal nature or complicate interpretation, especially with repeated measures.
• Handling "don't know" responses requires specialized statistical approaches.

Purpose of the Study:

• To introduce and evaluate two-part models for longitudinal semiordinal data.
• To provide a statistically robust and interpretable method for analyzing data with "don't know" categories.
• To assess adolescent perceptions of JUUL e-cigarette nicotine strength using the proposed models.

Main Methods:

• Development of two-part models with random effect logistic and ordinal submodels.
• Incorporation of correlated random effects to account for within-individual dependence.
• Maximum likelihood estimation using adaptive Gaussian quadrature in SAS PROC NLMIXED.
• Simulation studies to assess model performance, bias, and efficiency.

Main Results:

• The proposed two-part models effectively accommodate longitudinal semiordinal data.
• Adolescents perceive 5% nicotine JUUL e-cigarettes as having low strength.
• This misperception is more prevalent in adolescents who are not current JUUL users.

Conclusions:

• Two-part models offer a flexible and interpretable approach for analyzing semiordinal data with "don't know" responses.
• The models provide valuable insights into complex health behaviors, such as perceptions of nicotine content.
• Findings highlight a significant misconception about JUUL nicotine strength among non-users, suggesting a need for targeted education.