Hello, for my meta-analysis I try to test if two varainces are equal without using the raw scores. I have is the SD's, N's and the Means. I want to test the variances from dependent and independend samples. I assume I can use the var.test procedure for the independent samples, but what about the dependent samples ? Has anyone an idea how to realise this with R ? Thanks in advance Markus
I don't see a way to use var.test without data vectors. However,
you could trick it as illustrated by the following:
SD1 <- SD2 <- N1 <- N2 <- 5
var.test(SD1*rnorm(N1), SD2*rnorm(N2))
For more than two variances, you could use bartlett.test
similarly. However, Bartlett's test, and presumably also var.test, is
highly sensitive to non-normality. I don't have a citation, but I
remember hearing George Box say that Bartlett's test is almost a better
test of non-normality than of inhomogeneity of variance. If you needed
a citation for that, I would look first at various papers and book
sections discussing robustness and Bartlett's test, especially in the
index of Box on Quality and Discovery (Wiley, 2000) or his earlier
collected works volumes.
This may answer to "independent samples" question. However, we
would need to know more about the nature of the dependence to answer the
dependent samples question, and a sensible answer to the latter may
require untenable assumptions.
hope this helps. spencer graves
Markus Koesters wrote:
>Hello,
>
>for my meta-analysis I try to test if two varainces are equal without
>using the raw scores. I have is the SD's, N's and the Means.
>I want to test the variances from dependent and independend
>samples.
>I assume I can use the var.test procedure for the independent
>samples, but what about the dependent samples ? Has anyone an
>idea how to realise this with R ?
>Thanks in advance
>
>Markus
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>
Thank you very much for the quick answer, I'll try that. The dependence for the other samples are a result of the designs of the studies I want to metaanalyse - most of them are single group studies with pre-post measures. regards, Markus> I don't see a way to use var.test without data vectors. However, > you could trick it as illustrated by the following: > > SD1 <- SD2 <- N1 <- N2 <- 5 > var.test(SD1*rnorm(N1), SD2*rnorm(N2)) > > For more than two variances, you could use bartlett.test > similarly. However, Bartlett's test, and presumably also var.test, is > highly sensitive to non-normality. I don't have a citation, but I > remember hearing George Box say that Bartlett's test is almost a better > test of non-normality than of inhomogeneity of variance. If you needed > a citation for that, I would look first at various papers and book > sections discussing robustness and Bartlett's test, especially in the > index of Box on Quality and Discovery (Wiley, 2000) or his earlier > collected works volumes. > > This may answer to "independent samples" question. However, we > would need to know more about the nature of the dependence to answer the > dependent samples question, and a sensible answer to the latter may > require untenable assumptions. > > hope this helps. spencer graves > > Markus Koesters wrote: > > >Hello, > > > >for my meta-analysis I try to test if two varainces are equal without > >using the raw scores. I have is the SD's, N's and the Means. > >I want to test the variances from dependent and independend > >samples. > >I assume I can use the var.test procedure for the independent > >samples, but what about the dependent samples ? Has anyone an > >idea how to realise this with R ? > >Thanks in advance > > > >Markus > > > >______________________________________________ > >R-help at stat.math.ethz.ch mailing list > >https://www.stat.math.ethz.ch/mailman/listinfo/r-help > > > > >
"Markus Koesters" <Markus.Koesters at uni-jena.de> wrote:>for my meta-analysis I try to test if two varainces are equal without >using the raw scores. I have is the SD's, N's and the Means. >I want to test the variances from dependent and independend >samples. >I assume I can use the var.test procedure for the independent >samples, but what about the dependent samples ? Has anyone an >idea how to realise this with R ? >Thanks in advance > > >For independent samples there is a simple technique described by Larsen, Am. Stat, May 1992 for turning summary statistics back into a data set with an equivalent ANOVA summary. This is implemented and described in my STAT2DAT macro, http://www.math.yorku.ca/SCS/sasmac/stat2dat.html It should be easy to translate into R. -- Michael Friendly Email: friendly at yorku.ca Professor, Psychology Dept. York University Voice: 416 736-5115 x66249 Fax: 416 736-5814 4700 Keele Street http://www.math.yorku.ca/SCS/friendly.html Toronto, ONT M3J 1P3 CANADA
> > > For independent samples there is a simple technique described by Larsen, > Am. Stat, May 1992 > for turning summary statistics back into a data set with an equivalent > ANOVA summary. > > This is implemented and described in my STAT2DAT macro, > http://www.math.yorku.ca/SCS/sasmac/stat2dat.html > > It should be easy to translate into R.This is an interesting way - but most software packages test homogeneity of variances with the Levene Test. This can't work in this way, becas eit uses the median for calculation of the test statistics. But the real problem are the dependent measures... Markus> > -- > Michael Friendly Email: friendly at yorku.ca > Professor, Psychology Dept. > York University Voice: 416 736-5115 x66249 Fax: 416 736-5814 > 4700 Keele Street http://www.math.yorku.ca/SCS/friendly.html > Toronto, ONT M3J 1P3 CANADA > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://www.stat.math.ethz.ch/mailman/listinfo/r-help