Skip to content
Prev 12002 / 20628 Next

Reformat: Assistance with specification of crossover design model

Adam Smith <raptorbio at ...> writes:
with a pre-study baseline measure [y0].
Thanks for a very clear question.  I'm going to take a stab at
these, but haven't thought *very* deeply about them; hopefully
this will inspire corrections/alternative answers.
I think you're right that the individual-level random effect
controls for pre-treatment and sequence/carryover effects.  However,
I'm not sure it will capture period or treatment-by-period effects
(i.e. residuals within the same period might be correlated, and
residuals from individuals who received the treatment in the same
period might be correlated).

  One way to see whether you might have missed something would
be to draw (e.g.) boxplots of residuals by whatever groupings
you might be concerned about.
I think so.
I think so.
I think you do want trt*period (possibly as a random effect?)
  In a setting where I had access to some regularization (e.g.
blme) I would be tempted to treat period as random -- but I wouldn't
do it in this vanilla model, because there aren't enough periods to
estimate it reliably.

  While you don't _need_ to incorporate pre-treatment covariate,
carryover effects, etc., it might be interesting -- I think these
would be partitioned out of the among-individual variation ...
methods base