[R] Two sample t.test, order of comparions

Helmut Schütz helmut.schuetz at bebac.at
Wed Apr 18 16:17:40 CEST 2007


Dear Charilaos!

Charilaos Skiadas wrote:
>>
>> Do you know a more elegant way than the clumsy one I have tried?
>>
>>> as.numeric(exp(result$estimate[2]-result$estimate[1]))
>>> as.numeric(exp(-result$conf.int[2]))
>>> as.numeric(exp(-result$conf.int[1]))
>
> First off, those three could probably be simplified slightly as:
> as.numeric(exp(-diff(result$estimate)))
> as.numeric(exp(-result$conf.int))
>
> The simplest solution I think is to specify that resp$treatment should 
> have the levels ordered in the way you like them using this first:
>
> resp$treatment <- ordered(resp$treatment, 
> levels=rev(levels(resp$treatment)))
>
> Then the t.test will show things in the order you want them.
I applied relevel() as suggested by Douglas and Dimitri:

 > relevel(resp$treatment, ref = "Reference")
 > result <- t.test(logAUC ~ treatment, data = resp, subset = (period == 
1), var.equal = FALSE, conf.level = 0.90)
 > result

yielding

        Welch Two Sample t-test

data:  logAUC by treatment
t = 1.1123, df = 21.431, p-value = 0.2783
alternative hypothesis: true difference in means is not equal to 0
90 percent confidence interval:
 -0.0973465  0.4542311
sample estimates:
mean in group Reference      mean in group Test
               3.562273                3.383831

So right now the confidence interval in the log-domain is of the correct 
order.

Your first suggestion is working (sign changed due to reversed level)

 > as.numeric(exp(diff(result$estimate)))
 > [1] 0.8365723

But still I have to apply

 > as.numeric(exp(-result$conf.int[2]))
[1] 0.634936
 > as.numeric(exp(-result$conf.int[1]))
[1] 1.102242

because
 > as.numeric(exp(-result$conf.int))
[1] 1.102242 0.634936

in order to get the correct CI in the untransformed domain I had to sort 
the list:
 > sort(as.numeric(exp(-result$conf.int)))
[1] 0.634936 1.102242

Best regards,
Helmut

-- 
Ing. Helmut Schütz
BEBAC - Consultancy Services for
Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna, Austria
tel/fax +43 1 2311746
e-mail  helmut.schuetz at bebac.at
web     http://bebac.at
forum   http://forum.bebac.at



More information about the R-help mailing list