Implications of choice of statistical approach when analysing multiple health behaviours
AuthorsKareena McAloney, Hilary Graham, Julia Hall, Catherine Law, Lucinda Platt and Heather Wardle
Introduction: Public health research and policy are moving from a focus on analysing single health behaviours to multiple health behaviours. There are two approaches for analysing multiple behaviours – co-occurrence, which investigates concurrent but independent participation in multiple behaviours, and clustering, where account is taken of dependence between the behaviours. This paper investigates how the choice of analysis strategy influences the characterisation of multiple risk behaviours.
Method: The youth cohort of Understanding Society: the UK Household Longitudinal Study (UKHLS) at wave 1 (2009/2010) provided information on physical activity, fruit and vegetable consumption, smoking and alcohol consumption. Data from 4899 young people (10 – 15 years) were used to explore and compare four statistical approaches to the analysis of multiple health behaviours – a summed risk index; identifying specific behaviour patterns; the observed/expected ratio method; and latent class analysis.
Results: Only 1% of young people participated in all four behaviours, while 8% participated in none of the behaviours. The risk index revealed 44% of young people with two behaviours, and 11% with three; however this method gives no information on which behaviours combined. The observed-expected ratio method revealed 13 significantly clustered behaviour patterns. By contrast latent class analysis uncovered 4 underlying classes of youth health behaviours. Different associations with gender, ethnicity and household income were identified for the different methods.
Discussion: Analyses of co-occurrence produce different information and results to those produced in analyses of clustering. These differences are not confined to the relationships among the four behaviours, but extend to the social patterning of multiple risk behaviours. The paper emphasizes the need for consistency in the use of terminology, and an appreciation of the relative strengths of the different analytic approaches, to better support the developing evidence base for multiple health behaviour research and interventions in public health.