-  fit Rosa's example with predicted instead of fitted
- see if fits using cloglog.2asym can be made more stable.
