R help - Feb 2009 - Adjusting confidence intervals for paired t-tests of multiple endpoints

Dear R-users,

 

In a randomized placebo-controlled within-subject design, subjects recieved
a psycho-active drug and placebo. Subjects filled out a questionnaire
containing 15 scales on four different time points after drug
administration. In order to detect drug effects on each time point, I
compared scale values between placebo and drug for all time conditions and
scales, which sums up to 4*15=60 comparisons.

 

I have summarized the results in a data.frame with columns for t test
results including confidence intervals and mean-differences:

 

df1<-data.frame(trt=gl(2,35),matrix(rnorm(4200),70,60))

 

df2<-as.data.frame(matrix(NA,60,6))

names(df2)<-c('t','df','p','lower','upper','mean.diff')

for (i in 1:60) {df2[i,1:6]<-as.numeric(

unlist(t.test(df1[,i+1]~df1$trt,paired=T))[1:6])}

 

Now, I want to adjust the confidence intervals for multiple comparisons.

 

For a Bonferroni-adjustment, I did the following:

 

df2$std.error.of.diff<-df2$mean.diff/df2$t

ci<-qt(p=1-(0.05/nrow(df2)),df=df2$df)*df2$std.error.of.diff

ci.bonf<-data.frame(lower=df2$mean.diff-ci,upper=df2$mean.diff+ci)

 

I hope this is the correct method. However, I think, the
Bonferroni-adjustment would be much too conservative. I need a less
conservative approach, perhaps, something like Holm's method, which I can
easily apply to the p-value with p.adjust(df2$p,method='holm'). Is there
package, which can do this for the confidence-interval or could someone
provide a simple script to calculate this?

 

Thanks a lot!

 

Erich


	[[alternative HTML version deleted]]

Erich Studerus <erich.studerus <at> bli.uzh.ch> writes:
> 
> 
> Dear R-users,
> 
> In a randomized placebo-controlled within-subject design, subjects recieved
> a psycho-active drug and placebo. Subjects filled out a questionnaire
> containing 15 scales on four different time points after drug
> administration. In order to detect drug effects on each time point, I
> compared scale values between placebo and drug for all time conditions and
> scales, which sums up to 4*15=60 comparisons.
> 
....
> Now, I want to adjust the confidence intervals for multiple comparisons.
For simple cases, see p.adjust or TukeyHSD (more limited).

For complex cases, see simint and friends in package multcomp.

Dieter

Thanks, but as far as I know the multcomp package and the tukeyHSD function 
can only handle one dependent variable, whereas I have 60 dependent 
variables.

Regards,

Erich


On Sat, 28 Feb 2009 17:41:11 +0000 (UTC)
  Dieter Menne <dieter.menne at menne-biomed.de>
wrote:
> Erich Studerus <erich.studerus <at> bli.uzh.ch> writes:
> 
>> 
>> 
>> Dear R-users,
>> 
>> In a randomized placebo-controlled within-subject design, subjects 
>>recieved
>> a psycho-active drug and placebo. Subjects filled out a questionnaire
>> containing 15 scales on four different time points after drug
>> administration. In order to detect drug effects on each time point, I
>> compared scale values between placebo and drug for all time conditions
and
>> scales, which sums up to 4*15=60 comparisons.
>> 
> ....
>> Now, I want to adjust the confidence intervals for multiple
comparisons.
> 
>For simple cases, see p.adjust or TukeyHSD (more limited).
> 
>For complex cases, see simint and friends in package multcomp.
> 
> Dieter
> 
> ______________________________________________
> 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.