[R] simple main effect.

[DrorD]

I tried to implement Ista's procedure and would like to provide it as
a working example, with the intention to get feedback from the R
community:

The data contains three variables:
One dependent var: t.total
and two independent vars: group (between: D2C2, C2D2) and present.type
(within: C2, D2).

# First I do the "overall" ANOVA:
m.full=aov(t.total ~ group * present.type + Error(subj/present.type),
data=dat.net)
summary(m.full)

Error: subj
          Df Sum Sq Mean Sq F value Pr(>F)
group      1   1430    1430  0.4224  0.528
Residuals 12  40634    3386
Error: subj:present.type
                   Df  Sum Sq Mean Sq F value    Pr(>F)
present.type        1   603.1   603.1  0.7988 0.3890145
group:present.type  1 22775.8 22775.8 30.1661 0.0001379 ***
Residuals          12  9060.1   755.0
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Error: Within
           Df Sum Sq Mean Sq F value Pr(>F)
Residuals 840 148493     177
-------------------------------

Now, since the interaction is significant, I want to compute two
simple main effects: to find out if there is a significant difference
between C2 and D2 (present.type var) (i) in group D2C2 and then also
(ii) in group C2D2 (won't be shown to avoid redundancy). To achieve
that:

(1) I run the model separately for each level of group:

dat.g1 = subset(dat.net, group=="D2C2")
m.g1 = aov(t.total ~ present.type + Error(subj/present.type),
data=dat.g1)
summary(m.g1)

Error: subj
          Df Sum Sq Mean Sq F value Pr(>F)
Residuals  6  22788    3798
Error: subj:present.type
             Df  Sum Sq Mean Sq F value   Pr(>F)
present.type  1 15395.8 15395.8  18.694 0.004963 **
Residuals     6  4941.4   823.6
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Error: Within
           Df Sum Sq Mean Sq F value Pr(>F)
Residuals 420  80658     192
-------------------------------

(2) I use the error term from the overall model (dat.net) to calculate
the MS-Error term:

MS-Effect(from model m.g1) for present.type = 15395.8 with df = 1
MS-Error(from model m.full) for present.type = 755.0 with df = 12
(from Error: subj:present.type)

so we have F(1,12) = 15395.8 / 755.0
which means F = 20.4 and to calculate p-sig:
1 - pf(20.4,1,12)
->> p=0.0007070375

Well, is this the way to do it?
Is it equivalent to or different from using the HH package?

Thanks in advance and best to all,
dror

----------------------

On Feb 27, 4:18 pm, Or Duek <ord... at gmail.com> wrote:
> I am very new to R and thus find those examples a bit confusing although I
> believe the solution to my problems lies there.
> Lets take for example an experiment in which I had two between subject
> variables - Strain and treatment, and one within - exposure. all the
> variables had 2 levels each.
>
> I found an interaction between exposure and Strain and I want to compare
> Strain A and B under every exposure (first and second).
> The general model was with that function:
> aov(duration~(Strain*exposure*treatment)+Error(subject/exposure),data)
>
> in summary(aovmodel) there was a significant interaction between exposure
> and strain.
> how (using those HH packages) can I compare Strains under the conditions of
> exposure?
>
> BTW - I don't have to use aov (although its seems to be the simplest one).
>
> Thank you very much.
>
> On Mon, Dec 21, 2009 at 12:16 AM, Richard M. Heiberger <r... at temple.edu>wrote:
>
>
>
>
>
> > For simple effects in the presence of interaction there are several
> > options included in the HH package.  If you don't already have the HH
> > package, you can get it with
> >  install.packages("HH")
>
> > Graphically, you can plot them with the function
> >  interaction2wt(..., simple=TRUE)
> > See the examples in
> >  ?HH::interaction2wt
>
> > For tests on the simple effect of A conditional on a level of B, you
> > can use the model formula B/A and look at the partition of the sums of
> > squares using the split= argument
> >  summary(mymodel.aov, split=<put your details here>)
>
> > For multiple comparisons from designs with Error() terms, you need to
> > specify the same sums of squares with an equivalent formula that doesn't
> > use the Error() function.  See the maiz example in
> >  ?HH::MMC
> > Read the example all the way to the end of the help file.
>
> > Rich
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-h... at r-project.org mailing listhttps://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guidehttp://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.