Statistical tipsters, I need confirmation on an issue that I resolved to
my satisfaction a year or two ago, but want to be sure I figured it out.
If this is a repost, I aplogize and will make it short.
To combine 2 studies in a meta-analysis and test for significance, you
need to get the Z value for the p reported. But the p must be
one-tailed. So, if you have a two group study where t = 2.00 and p = .08
two tailed, you'd half the p to .04. In this case the corresponding Z =
1.75.
But what if the same result came from an F test? If the same numbers
were run through a simple oneway ANOVA, F = 4.0 and p = .08. I have
always halved the p value from an F test and treated it as a one-tailed
test.
The F distribution (as a statistical distribution) is one-tailed. But if
null is true it's equally likely that mean1 > mean 2, or, mean1 < mean
2. The F distribution is used to capture both research outcomes into the
right tail (or virtually all of it). If I specify _in advance_ that
mean1 > mean2, can't we just half the p and say p = .04 one-tailed? I'm
pretty sure a statistician explained this to me this way.
Some people say "one-tailed" while others say "one-sided." I suspect
that "tails" refers to the actual distribution of F, while "sided"
refers to the possible research outcomes - and the two are separate
concepts. But, if anyone thinks I'm off my rocker, please let me know.
Thanks.
--
---------------------------------------------------------------
John W. Kulig [EMAIL PROTECTED]
Department of Psychology http://oz.plymouth.edu/~kulig
Plymouth State College tel: (603) 535-2468
Plymouth NH USA 03264 fax: (603) 535-2412
---------------------------------------------------------------
"What a man often sees he does not wonder at, although he knows
not why it happens; if something occurs which he has not seen before,
he thinks it is a marvel" - Cicero.
