This project concerns the development of statistical methodology suitable for modelling accelerometer measured physical activity.

Aim: Our objective is to develop a general framework of statistical methodology for exploring the relationship between physical activity and a health outcome which allows to take the pattern of physical activity into account. We will use functional data analysis approaches (Ramsey and Silverman, 2005) for this task. In particular we will consider models where either the response (outcome) variable or the predictor variables are functional summaries. That is, we will need to find suitable summaries of the physical activity profile for a specific modelling task. For instance, if the aim is to model fat mass with physical activity as a predictor a function such as the histogram might be appropriate. In the case where the outcome is a risk factor for cardio-vascular disease, a function respecting the temporal pattern of physical activity may be more appropriate. In the cases where physical activity is treated as a predictor for a health outcome such as obesity, the methodology should allow to test whether the effect of physical activity is constant over the activity range, i.e. whether a single summary statistic such as MVPA is adequate rather than using a functional summary of the physical activity profile. We will use the ALSPAC data for developing the methodology. We have already carried out some work with the physical activity data from ALPAC where the outcome variable was fat mass, this work is summarised in Augustin, Mattocks et al.