Re: st: interaction in anova # or ##

[David Hoaglin <dchoaglin@gmail.com>]

Francesca,

If you are lucky, the answer is "none of the above."

The ## operator ("factorial cross") expands into a main effect for
group, an overall effect (slope) for age, and an interaction for group
and age (i.e., a separate slope for age in each group).  The #
operator ("cross") expands into only the interaction for group and age
(i.e., the model does not include the main effect for group or an
overall slope for age).

As I interpret your secondary step, you would like to adjust for age
(and gender) with the same slope (and effect) in each group.  That is,
you have an analysis of covariance.  If the adjustments are not the
same in all the groups, it usually difficult to interpret comparisons
among the groups.  Hence,

anova followup group c.age c.baseline

can be compared with the result of

anova followup group##c.age  c.baseline

David Hoaglin

On Mon, Nov 25, 2013 at 11:03 AM, Pesola, Francesca
<francesca.pesola@kcl.ac.uk> wrote:
> Hi All,
>
> I am running some analysis on RCT data to see whether  depression scores at follow-up, while adjusting for baseline scores, depend on allocation group (control vs. treatment).
>
> As a secondary step I am interested to see if the effect of group depends on age and gender of participants in my sample, as suggested by the literature.  However, I am not sure which way of including the interaction term is 'more correct' based on my aims, whether to use:
>
> anova followup group##c.age  c.baseline; or
>
> anova followup group#c.age  c.baseline
>
> Grateful for suggestions,
> Francesca

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/