Three-part random effect models for longitudinal skewed survey data with "not applicable" responses.

Area of Science:

• Statistics
• Biostatistics
• Survey Methodology

Background:

• Survey data often include questions for only a subset of participants.
• Floor effects, where responses cluster at the lowest scale value, are common.
• Analyzing such data requires methods that account for selection and response patterns.

Purpose of the Study:

• To propose a novel three-part statistical model for analyzing survey data with missingness and floor effects.
• To improve the unbiased and efficient estimation of trends over time.
• To address challenges in analyzing complex longitudinal surveys like the PATH study.

Main Methods:

• A three-part model comprising two logistic sub-models and a truncated normal model.
• Incorporation of random effects to handle correlations in repeated observations.
• Maximum likelihood estimation using SAS PROC NLMIXED.

Main Results:

• The proposed three-part model demonstrated significantly lower bias compared to simpler models.
• The model achieved better coverage probabilities for regression coefficients in simulations.
• Application to the PATH young adult data illustrated its practical utility.

Conclusions:

• The three-part model provides a robust approach for analyzing complex survey data with missingness and floor effects.
• It offers superior performance in terms of bias and efficiency over traditional methods.
• This methodology enhances the accuracy of trend analysis in longitudinal health research.