> For this particular question, you could reduce data for each person to
> number of falls out of 18 trials and try simple 3-group analysis of
> variance.  I am assuming that you have enough in each group to make
> this sensible and would pay some attention to whether distribution for
> each group is roughly bell-shaped with approximately same spread.  To
> include condition in analysis, see others posts.
> 
> Terry J. Reedy

Setting aside time travel and changing ther  design. I'd try what
Terry suggested first. Repeated measures ANOVA should peform pretty
well as long as the mean number of falls is around 0.5 (0.2-0.8 is
often cited). Before the popularity of logistic regression the arcsin
transformation was also widely used in these situations. There are
repeated measures "versions" of logistic regression*, but they might
be overkill if the ANOVA model is reasonable (standard
diagnostics/checks of assumptions apply).

* multilevel/heirarchical logistic regression is available in MlWin,
SAS, HLM etc. and will do the job (but there are probably other
options).

Thom
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to