1. update some of the statistical language for what the MSP's actually
   know

2. The fitLMER.fnc from library(LMERConvenienceFunctions) is a bit
   fragile, when you ask it to do everything.  So have students do
   some version of

   * select fixed effects
   * select rand effects
   * select fixed effects

   by hand or nearly by hand...

3. try to get students to produce something gradeable without so much
   scaffolding that the students are just monkey-see-monkey-do'ing...