In response to the helpful comments, I have edited the original question (where I had assumed that a for-loop and an apply-loop give different results).
I am using R to run a large number of 2-group t-tests, using input from a delimited table. Following recommendations from here and elsewhere, I tried either 'for-loops' and 'apply' to accomplish that. For 'normal' t.test, both work nicely and give the same results. However, for a paired t-test, the for-look appears to works while the apply-loop does not. Later, i found out that both loops suffer from the same problem (see below) but the for-loops deals more gracefully with the situation (only one cycle of the loop returns an invalid result) while the apply-loop fails altogether.
My input file looks like this: (the first line is a header line, the data lines have a name, 4 datapoints for group 1 and 4 datapoints for group 2):
header g1.1 g1.2 g1.3 g1.4 g2.1 g2.2 g2.3 g2.4
name1 0 0.5 -0.2 -0.2 -0.1 0.4 -0.3 -0.3
name2 23.2 24.4 24.5 27.2 15.5 16.5 17.7 20.0
name3 .....
and so on (overall ~50000 lines). The first data line (starting with name19 turned out to be the culprit.
This is the for-loop version that works better (failes on the problematic line but correctly deals with all other lines):
table <- read.table('ttest_in.txt',head=1,sep='\t')
for(i in 1:nrow(table)) {
g1<-as.numeric((table)[i,2:5])
g2<-as.numeric((table)[i,6:9])
pv <- t.test(g1,g2,paired=TRUE)$p.value
}
This is the 'apply' version that causes problems
table <- read.table('ttest_in.txt',head=1,sep='\t')
pv.list <- apply(table[,2:9],1,function(x){t.test(x[1:4],x[5:8],paired=TRUE)$p.value})
One of the ~50000 data lines is problematic in that the differences of all pairwise comparions are identical, which in a paired t-test results in an undefined p-value (essentially zero). The apply loop crashes with the error 'data are essentially constant'. To me (as an R newbie) it does not seem to be a good idea to crash the entire script just because the t.test doesn't like one piece of data. In the for-loop, this data line also results in an error message but the loop continues and all the other t-tests give correct results.
Did I do something fundamentally wrong? This behaviour a essentially prohibits the usage of apply-loops for this kind of batch analysis. Or is there a standard way to circumvent this problem. Why doesn't the t-test just return something invalid for that particular p-value instead of bailing out?
In situations like this, I catch all the warnings and errors and investigate them afterwards, as shown here: How do I save warnings and errors as output from a function?
You may also find some good ideas here: How to tell lapply to ignore an error and process the next thing in the list?