Modelling count, bounded and skewed continuous outcomes in physical activity research: beyond linear regression models