Robust clustering of standard errors in autoregressive multi-level model (mixed)

I have a panel data of the format

I am using xtmixed where level 2 is app_id. My primary regressor of interest : changes at rank_category level. I am also using autoregressive model residuals(ar 1, by(rank_category) t(age.months)) .

I am confused if I should be clustering the standard errors at level of app_id or that of rank_category.

Another conceptual doubt that I have is, if I cluster errors at app_id level, it would take care of the serial autocorrelation between different observations for each app_id. However, that was the reasoning behind using the autoregressive model. So, is it necessary to use both approaches together, or is it alright to cluster the errors at rank_category level? To come out clean, clustering at rank_category level gives me favorable results.


Any suggestions would be greatly appreciated! Thanks!

-- Harshal