#2(a)

1  Check I*H*V 

2   AIC: I + H + V + H:V
    BIC: I + H + V
    which is preferred by LRT depends on data set

1   diagnostics or other model checking

2   Comments on results (boxplots help!)
    * I matters most
    * H = I-5-6 strongest influence
    * V: weak evidence that contrary is more classical
    (but a recitation of significant effects is enough)

#2(b) is pretty straightforeward; add (1|Subject) and otherwise no
 changes

#2(c)

I think the model with all three random effects at once is very hard to fit well.  Your by-hand approach worked well. For fitLMER, I would try forward selection of the (0+factor|Subject) effects one at a time rather than all three at once.

best model appears to be I + H + V + H:V + (I + H|Subject)
  (possibly without H:V if the persom used BIC on the prev problem)

There should be a different eta for each level of instr and harmony, and these etas should be correlated.  But this is OK for now.



[The instructions for the future:
  * compare fixed effects of model with mult rand effects to fixed
    effects for model with only (1|Sub).
  * find a better way to teach or ask about the equations in (iii)
  * keep X16.minus.17 out of the models.
  * the hint in 3(a) about converting to factor variables is not
    helpful as-is.
  * make clear in 3(a) that model should be continued from end of 2.



]

[look at

* lmerTest package as alternative to fitLMER.fnc from
LMERConvenienceFunctions...

* olssr as a selection and model checking tool for lm()

]


3(a)

2  Selection of fixed effects using lm() only

4  Selelction of fixed effects using lm() followed by re-checking with
   rand effects in lmer()

4  Selection of fixed effects using lmer [by hand, using fitLMER.fnc, etc.]

2  Execution


[fwd selection by hand in lmer]
Classical ~ Instr + Harm + Voice + Harm:Voice +
   ConsInstr + Instr.minus.Notes + ClsListen + Composing + GuitarPlay +
   (Inst + Harm | Subject)

[stepAIC followed by fitLMER]
Classical ~ Instr + Harm + Voice + Harm:Voice +
   Selfdeclare + X16.minus.17 + ClsListen + X1990s2000s.minus.1960s1970s
   (Inst + Harm | Subject)

[backward selection by fitLMER]
Classical ~ Instr + Harm + Voice [ + Harm:Voice ] +
   log.OMSI + X16.minus17 + ClsListen + ConsNotes +
   (1|Sub) ; didn't check (Inst + Harm | Subject)