Accounting for misclassification error in retrospective smoking data

Area of Science:

• Epidemiology
• Biostatistics
• Sociology

Background:

• Longitudinal surveys increasingly use retrospective questions on smoking initiation and cessation.
• This data offers insights into life course smoking behaviors but is under-utilized.
• Retrospective reports may introduce misclassification errors into smoking behavior models.

Purpose of the Study:

• To investigate the extent and consequences of misclassification errors in smoking participation models using retrospective data.
• To explore potential solutions for mitigating these errors.
• To compare different modeling approaches for accounting for misclassification.

Main Methods:

• Utilized the National Longitudinal Survey of Youth 1979 (NLSY79) data, which includes both contemporaneous and retrospective smoking status.
• Developed and compared four sets of probit models for smoking participation.
• Included baseline models with contemporaneous data, models with retrospective data, and parametric models addressing misclassification errors.

Main Results:

• Preliminary findings indicate that accounting for misclassification error is crucial for accurate smoking behavior analysis.
• The adjusted maximum likelihood estimation method for misclassification did not consistently yield expected results.
• Significant differences were observed between models using contemporaneous versus retrospective smoking data.

Conclusions:

• Retrospective smoking data requires careful handling due to potential misclassification errors.
• Standard statistical approaches may not fully resolve issues arising from retrospective reporting.
• Further research is needed to refine methods for analyzing longitudinal smoking data accurately.