R help - Jan 2008 - question regarding hypothesis testing in contingency tables

Hi R-users!

I have the following example:
a<-data.frame(cat=c(5,10,15), dog=c(5,10, 15), mouse=c(10,10,20))
b<-data.frame(cat=c(15,10,5), dog=c(15, 10, 5), mouse=c(20,10,10))
rownames(b)<-c("scared", "happy", "sad")
rownames(a)<-c("scared", "happy", "sad")

Let's say that a and b are 2 contingency tables and 5,10, 15, 20 are
proportions of the whole sample (eg there are 5% animals that are both dogs and
scared in sample a). a and b are two different samples. Now, what i want is to
test the  hypothesis that these two sample structures are  similar. I know that
the chi square test, only handles one 2-dimensional contingency table, or a
3-dimensional one (mantelhaen. test), but I haven't seen a test that handles
2 separate 2-dimensional tables.  Is there such a test?
Another thing that i want to see , is which cells differ one from another? Eg is
the 5 percent of scared dogs from sample a, different from  15% of scared dogs
form sample b? I would like something like the "adjusted standardized
reziduals" test from SPSS?
If i would compute by hand a z-statistic that tested the hypothesis that the 2
proportions are equal, would that be correct, since I have a multinomial
proportion and not a binomial one ?
Could I construct a confidence interval for my multinomial proportions in R
simultaneously ? If i would do it for each proportion by using the  binomial
distribution would that affect the power of the test?

Thank you and have a great day!


       
---------------------------------

	[[alternative HTML version deleted]]

On Thu, 2008-01-10 at 09:47 -0800, eugen pircalabelu
wrote:
> Hi R-users!
> 
> I have the following example:
> a<-data.frame(cat=c(5,10,15), dog=c(5,10, 15), mouse=c(10,10,20))
> b<-data.frame(cat=c(15,10,5), dog=c(15, 10, 5), mouse=c(20,10,10))
> rownames(b)<-c("scared", "happy", "sad")
> rownames(a)<-c("scared", "happy", "sad")
> 
(...)
> Another thing that i want to see , is which cells differ one from another?
Eg is the 5 percent of scared dogs from sample a,
> different from  15% of scared dogs form sample b? I would like something
like the "adjusted standardized reziduals" test from SPSS?
(...)
> Thank you and have a great day!
Hi Eugen!

The adjusted standardized residuals is available in package gmodels

require(gmodels)
?CrossTable


-- 
Bernardo Rangel Tura, M.D,MPH,Ph.D
National Institute of Cardiology
Brazil