No document available.
Abstract :
[en] In HBSC, standardized data cleaning procedures ensure comparability across countries. Two procedures are applied to substance use items: Lifetime responses inconsistent with 30-day use are recoded to -99 and missing 30-day items are recoded to NEVER, if lifetime is never. These procedures assume that missingness is at random. This study examines whether missingness in 30-day substance use items subject to re-coding occurs at random in Luxembourg.
Methods: The sample included 6002 secondary students (ages 12–18) in Luxembourg. Missingness in substance use items was coded as a binary outcome and analyzed using hierarchical logistic regression with predictors entered in two blocks: sociodemographic characteristics (Block 1) and proximal influences (Block 2). Intraclass correlation coefficients assessed clustering at class and school levels.
Results: Missingness varied across variables, with the highest observed drunkenness (7.2%), followed by cannabis (6.4%), alcohol (6.3%) and smoking (5.6%). After recoding to NEVER missingness decreased to 2.9%, 1.9%, 3% and 1.7% respectively. All three models included over 80% of cases, showed good fit, and improved with Block 2. The full models explained 1.5%, 2.4%, 2.8% and 1.8% respectively. Living with a single parent or stepfamily, and being younger, increased odds of missingness in all models. Perceiving low family support raised odds of missing for drunkenness; whereas not liking school did so only for cannabis use. Being a girl lowered odds in alcohol use and cannabis but had no effect for drunkenness or smoking. Clustering was minimal at the class level (1.3% to 2% of variance).
Conclusion: In Luxembourg, missingness in the 30-day use items is not completely at random and is mainly an individual phenomenon. These findings highlight the importance of examining missingness patterns across HBSC countries and potential implications for reported prevalences estimates.
