Model fitting - error distribution

[kathsherratt]

From @sbfnk

I noticed that you found "The log link models the score multiplicatively but leaves some skew in the residuals. A sensitivity analysis modelling the log-transformed score directly substantially improves the residual distribution while preserving the direction of all effects and leaving the model-structure conclusions unchanged." - the skewed residuals are a concern but I don't think modelling log(score) is the right way to address this (as it loses propriety). Instead why don't we change the error family to something that allows skew e.g. gamma? Then I think there would be a case for swapping this to the main analysis as we don't want massively skewed residuals.

Agreed, to be implemented.

[sbfnk]

You could also try Tweedie(log) if zeroes cause headaches.