Based on published collaborative research co-authored by Dr Arash Ardalan.
The research question
How can several body fat measurements be analysed jointly while accounting for the structure of the study sample?
Study context
The study included 2,538 students aged 14–20, sampled through 16 public high schools in Shiraz. Three skinfold-thickness measurements were analysed alongside individual and contextual variables.
Statistical approach
- Model the three anthropometric outcomes jointly using a multivariate multilevel approach.
- Represent the child- and district-level structure described in the study.
- Examine associations with characteristics including sex, family history of obesity and socio-economic status.
What the analysis provided
The analysis described variation at different levels and reported associations while considering the measurements together, rather than treating every outcome and observation independently.
The cross-sectional data support associations within the sampled population. The local sampling structure and measurements need consideration when applying the findings elsewhere.
Practical lessons for your project
- Follow the sampling structure
- Identify the units that were sampled and measured before deciding how to represent dependence.
- Consider related outcomes
- Discuss whether outcomes should be modelled jointly and how that choice serves the research question.
- Explain each model level
- Document what variation at each level means and how it affects interpretation and uncertainty.
Original publication
Marzieh Alamolhoda; Seyyed Taghi Heydari; Seyyed Mohammad Taghi Ayatollahi; Reza Tabrizi; Maryam Akbari; Arash Ardalan
A multivariate multilevel analysis of the risk factors associated with anthropometric indices in Iranian mid-adolescents
BMC Pediatrics · 20, 191 · 2020