In epidemiology it is common to relate the value of a risk factor X, e.g. male sex, to a health outcome Y, e.g. body mass index (BMI). The risk factor’s effect is typically summarised as the mean difference in outcome between those with and without the risk factor. But the risk factor’s impact on the outcome may extend beyond the mean to affect the variability and even the skewness of the distribution. This wider impact is not conventionally explored, and methods for analysing it are not well known.
The talk will describe GAMLSS (generalized additive models of location, scale and shape), an R regression package that estimates the moments of the distribution of the outcome. The talk will use examples of BMI and mental wellbeing from this eLife paper.