Date:         Wed, 19 Jan 2005 17:07:41 -0500
Reply-To:     "Hill, Steven" <hills@SFWMD.GOV>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Hill, Steven" <hills@SFWMD.GOV>
Subject:      Paired T-test on data populations that are really independent,
              but can be paired.
Content-Type: text/plain; charset="US-ASCII"

SAS-Ler's:



Would it be incorrect to use paired t-test to compare independent
populations?  Picture the following:  You have a water body from which
you take measurements on 30 consecutive days, with two different
machines (located next to each other).  I know that you can use the two
sample t-test, to statistically compare the data from these machines, to
see if there is a significant difference between the measurements of the
machines.  But would it be wrong to use a paired t-test to statistically
compare the measurement between the machines?  This would be treating
the measurement for each day as a pair?  Is there anything wrong with
this?  This came up in a meeting and at first thought I didn't see this
as a problem, but the other way around (using two sample t-test on truly
paired (dependent) data) would be inappropriate.  Using the paired test,
in the situation I describe above, would yield a more powerful test,
correct?



Steven