Profesional Development Workshop

Day 1: Tuesday, 13 September


Investigating whether exposures influence the variability of outcomes: motivation, implementation and interpretation using GAMLSS

Session convener: David Bann, University College London

This interactive workshop will outline why health and social scientists may be interested in investigating how exposures (or treatments) influence the variability of outcomes or have heterogenous effects; such methods are of aetiological and policy significance. For example, an intervention may successfully reduce the mean of a health outcome, yet increase its variability—and thus widen a form of health inequality. In other cases, an intervention may improve outcomes for some people but not others. Identifying this pattern could motivate the development of theory and of new interventions. We will briefly describe different approaches to investigate this, and provide a summary of previous evidence as applied to a health outcome of global public health importance: body mass index (BMI) and its inequality. There is emerging evidence that risk factors for high BMI such as socioeconomic disadvantage are linked with both elevated BMI and greater BMI variability. We will then provide a guide to using the GAMLSS package in R (Generalised Additive Models for Location, Scale and Shape). This modelling approach has been under-utilised in the existing health and social sciences literature. We will focus on the implementation and interpretation of results; syntax and data will be provided. The workshop will then facilitate a discussion of next steps in this under-researched field. Most health outcomes are continuous in nature, yet researchers frequently examine binary outcomes, or solely examine mean differences . Similarly, trials typically focus on differences in averages, and do not formally examine differences in variability. Should investigation of differences in variability become more widespread? What about skewness or kurtosis? Can the resulting findings have aetiological or policy significance? What sources of bias should researchers be aware of when investigating this?