Hi Tom,
I agree that it would be highly beneficial if such a modification could be done and I am rather sad that it did not happen so far because I am still an absolute MatLab-beginner and incapable of doing it myself.
Would highly appreciate it.
Beate
"Armin" wrote in message <ih57h3$jlu$1@fred.mathworks.com>...
> Hi Tom,
>
> Getting back to this question, I was wondering if you could think of any modification in the multcompare code so that it can measure p-values for all the pairwise comparisons that it is performing. if not, do you know any other code which it can do that?
>
> Thanks
> Armin
>
>
>
> "Tom Lane" <tlane@mathworks.com> wrote in message <gf9jlj$5fh$1@fred.mathworks.com>...
> > > When using Anova1 to test for differences between group, Anova 1 returns a
> > > p-value that shows whether there is a difference or not. To make a further
> > > pairwise comparison between groups you can use the multcompare-function.
> > > Is this some sort of post hoc test? It uses ‘tukey-kramer’ as
> > > default ..
> > > How is it possible to get the p-values of each pairwise comparison?
> >
> > Markus, the p-values of individual comparisons aren't computed in
> > multicompare and so there's no way to get them from that function. The
> > function basically computes intervals around each comparison and looks for
> > ones that overlap zero, instead of computing p-values and looking for ones
> > less than 0.05.
> >
> > I haven't thought much about this before. The idea behind multcompare is to
> > strive for an overall significance level of, say, 0.05, and test individual
> > differences preserving this overall level if the null hypothesis is true. I
> > suppose the p-values for the individual tests would be defined as the
> > overall p-value such that that individual difference is just significant.
> > Is that right?
> >
> > This might be something we could consider for the future, but we don't have
> > it now.
> >
> > -- Tom
> >![]()
I agree that it would be highly beneficial if such a modification could be done and I am rather sad that it did not happen so far because I am still an absolute MatLab-beginner and incapable of doing it myself.
Would highly appreciate it.
Beate
"Armin" wrote in message <ih57h3$jlu$1@fred.mathworks.com>...
> Hi Tom,
>
> Getting back to this question, I was wondering if you could think of any modification in the multcompare code so that it can measure p-values for all the pairwise comparisons that it is performing. if not, do you know any other code which it can do that?
>
> Thanks
> Armin
>
>
>
> "Tom Lane" <tlane@mathworks.com> wrote in message <gf9jlj$5fh$1@fred.mathworks.com>...
> > > When using Anova1 to test for differences between group, Anova 1 returns a
> > > p-value that shows whether there is a difference or not. To make a further
> > > pairwise comparison between groups you can use the multcompare-function.
> > > Is this some sort of post hoc test? It uses ‘tukey-kramer’ as
> > > default ..
> > > How is it possible to get the p-values of each pairwise comparison?
> >
> > Markus, the p-values of individual comparisons aren't computed in
> > multicompare and so there's no way to get them from that function. The
> > function basically computes intervals around each comparison and looks for
> > ones that overlap zero, instead of computing p-values and looking for ones
> > less than 0.05.
> >
> > I haven't thought much about this before. The idea behind multcompare is to
> > strive for an overall significance level of, say, 0.05, and test individual
> > differences preserving this overall level if the null hypothesis is true. I
> > suppose the p-values for the individual tests would be defined as the
> > overall p-value such that that individual difference is just significant.
> > Is that right?
> >
> > This might be something we could consider for the future, but we don't have
> > it now.
> >
> > -- Tom
> >