modeling temporal variation


I have 6 years of patient-level data and I'm using a hierachical (2-level) regression model. Could anyone suggest ways to model the temporal change in hospital-level variation in patient outcomes, in a multilevel model ?

So far, my approach has been to run the model on separate years of data, and simply inspect the variance in random effects for each year. This was a good initial approach, but now I am looking for ways to explicitly model the change in variance using the whole dataset. I've considered using a GAM with an indicator variable for year, but I would like to find a more sophisticated solution, perhaps one where I can provide some functional form for the variance over time, for example a linear relationship with unknown boundary points, where the boundary points (variance at t=0 and t=6) are estimated by the model by the model.

Any suggestions of comments would be most welcome.

Edit: I would also like to know if anyone can provide a reference to any published research that investigates temporal change in random effects.

Last edited: