R help - Oct 2013 - dunnett test questions

Hi,
    I've got a data set with a control group and a number of experimental
groups, that have unequal sample sizes, and am measuring the number of
people in each that respond yes or no.  I'd like to use a dunnett test in
R, where the syntax is supposed to be like:

library(multcomp)
test.dunnett=glht(anova_results,linfct=mcp(method="Dunnett"))
confint(test.dunnett)
plot(test.dunnett)

but:
1)  How do I run a dunnett test without doing the ANOVA (which wouldn't
have its requirements satisfied, as the measurements are not independent
due to the control group)?
2)  Do I have to tell the test what my sample sizes are, or will it
calculate the sums itself?

Here's my matrix:


       Control        A           B           C
Yes       23          19          27          53
No        623         523         823         469


thanks,

imran

	[[alternative HTML version deleted]]

On Sun, 27 Oct 2013, Imran Akbar wrote:
> Hi,
>    I've got a data set with a control group and a number of
experimental
> groups, that have unequal sample sizes, and am measuring the number of
> people in each that respond yes or no.  I'd like to use a dunnett test
in
> R, where the syntax is supposed to be like:
>
> library(multcomp)
> test.dunnett=glht(anova_results,linfct=mcp(method="Dunnett"))
> confint(test.dunnett)
> plot(test.dunnett)
>
> but:
> 1)  How do I run a dunnett test without doing the ANOVA (which wouldn't
> have its requirements satisfied, as the measurements are not independent
> due to the control group)?
But the control group is a separate independent group from the three 
treatments A-C, isn't it? Then independence should not be a problem.

For the binary response you need something different than an ANOVA, e.g., 
an analysis of deviance in a logistic regression.
> 2)  Do I have to tell the test what my sample sizes are, or will it
> calculate the sums itself?
If you supply a suitable fitted model, then glht() can infer the group 
sizes from that:

## data and table
dat <- data.frame(
   freq = c(23, 19, 27, 53, 623, 523, 823, 469),
   resp = factor(rep(c("Yes", "No"), each = 4)),
   method = factor(rep(1:4, 2), labels = c("Control", "A",
"B", "C"))
)
tab <- xtabs(freq ~ method + resp, data = dat)

## visualization and Pearson chi-squared test
spineplot(tab[, 2:1])
mosaicplot(tab, shade = TRUE, off = c(5, 0.5))
chisq.test(tab)

## Logistic regression and analysis of deviance chi-squared test
m <- glm(resp ~ method, weights = freq, data = dat, family = binomial)
anova(m, test = "Chisq")

## Odds ratios
exp(coef(m)[-1])

## Dunnett test
library("multcomp")
m_glht <- glht(m, linfct = mcp(method = "Dunnett"))
summary(m_glht)
confint(m_glht)
plot(m_glht)
> Here's my matrix:
>
>
>       Control        A           B           C
> Yes       23          19          27          53
> No        623         523         823         469
>
>
> thanks,
>
> imran
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>