search

pitakakariki / simr

Power Analysis of Generalised Linear Mixed Models by Simulation
70 stars 19 forks source link

Do something about the mess that is `fixed`. #122

Open pitakakariki opened 6 years ago

pitakakariki commented 6 years ago

The fixed test can specify either a test of an entire factor covariate or just one of its dummy variables. If you specify one kind of test but give a fixef name for the other kind it just quietly runs through all the simulations producing unhelpful error messages, which are then quietly hidden in the error log.

Stopgap is to be a bit noisier when errors are caught: https://github.com/pitakakariki/simr/issues/110

Better would be helpful messages telling the user why simr is confused.

Even better: if no method is specified, choose a default based on the fixef name supplied.

pitakakariki commented 6 years ago

Binary factors should be a special case too, since one factor maps to one dummy, so either specification is probably fine.

pitakakariki commented 6 years ago

This should probably have some clear documentation somewhere too.

pitakakariki commented 4 years ago

"Couldn't automatically determine a default fixed effect for this model" https://github.com/pitakakariki/simr/issues/96

Make this error message clearer, and make it happen less often.

pitakakariki commented 4 years ago

Same with "scope is not a subset of term labels" https://github.com/pitakakariki/simr/issues/93

pitakakariki commented 4 years ago

Maybe don't use KR as default with large models? https://github.com/pitakakariki/simr/issues/179

pitakakariki commented 3 years ago

Make it clear in the documentation which options apply in which case: entire covariate, single level, continuous. See https://github.com/pitakakariki/simr/issues/193

nb: "z", "t", and "sa" work for single level (but "t" and "sa" might be the same thing).

Also document which tests are supported for lmer vs glmer.