I'd call these assumptions about the error terms,
which you can examine by looking at the residuals.
1. Dependence may be a matter of dependence within
groups for some reason, or with respect to time or
space. You'd need to say more about your set-up to
be told the best way to check for this.
2. Others might point you to tests but I'd follow
the advice of the late Rupert G. Miller quoted on
[R] diagnostic plots p.266 and do a -qnorm- on the
residuals.
3. Similar comment. Eyeball a parallel dot or box
plot.
On (2) you could do something like this:
anova mpg for rep78
predict residuals, res
qnorm residuals
On (3) I'd recommend -rdplot- from SSC:
ssc inst rdplot
egen group = group(for rep78), la
rdplot , by(group) xla(, ang(v))
Note that many of the plots listed at -regdiag-
may be used after ANOVA. In addition, various
routines on SSC may be of interest, such as
anovaplot
indexplot
ovfplot
rhetplot
Note, however, that with 80 observations
presumably the number in each cell is not
large -- so the auto data may be a reasonable
analogue for you, and it may be too much to
expect much structure to be discernible.
Nick
n.j.cox@durham.ac.uk
fabiopericolini@fastwebnet.it
> I have to apply the two-way ANOVA to a set of 80 observations.
> My question is:what are the commands to test the three
> Hypothesis under
> the Anova model?
> 1. indipendent residuals
> 2. normal distribuition of the residuals
> 3. egual std deviation of the groups
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/