can non parametric tests be modified for one tailed tests?

jesshodgson

New Member
Hello all,
super newbie here.
just a quick question:
I have been doing one tailed independent two sample t tests but now on other data I no longer have normal distributions so I wanted to use Mann Whitney U test, but I am not sure if these can be one tailed.
I am using R so is it legitimate to specify the "alternative" argument as greater and if so (Im assuming so since it is an available option in R) do I then divide p/2 as I would for the t test to get a one tailed p value?
cheers
J

Karabiner

TS Contributor
I have been doing one tailed independent two sample t tests
Normally, you should not perform one-tailed tests.
You make yourself half blind.
but now on other data I no longer have normal distributions so I wanted to use Mann Whitney U test,
If your total sample size is > 50 or so, non-normal
distributions shouldn't matter much. One has to be
more concerened with unequal variances (if samples
have unequal size).
do I then divide p/2 as I would for the t test to get a one tailed p value?
Seems so. But normally you shouldn't do so.

With kind regards

K.

jesshodgson

New Member
Thanks for that.

My samples are fairly small. I'm testing the direction hence the one tail.

Cheers
J

Karabiner

TS Contributor
I'm testing the direction hence the one tail.
Many people test the direction, but you hardly ever
see one-tailed tests. Cleary, the usual two-tailed
test is absolutely ok for testing directional working
hypotheses. Just having in mind what the direction
of an effect might be does not justify one-tailed testing,
because, as mentioned before, it makes the researcher
half blind without good reason.

With kind regards

K.