Re: repeated measure analysis

On 10 May 2006 09:25:10 -0700, arrayprofile@xxxxxxxxx wrote:

[snip, details of repeated measures ANOVA ]

The problem is that if I do pairwise (among treatments) regular t tests
at each time point (i.e. using only the data in comparison), there are
indeed many significant comparisons; however, If I do pairwise t tests
using MSE from the ANOVA (i.e. 153230) for each time point, there are
no significant comparison at all which I think is inconsistent with the
marginally significant p value for the treatment main effect in the
ANOVA.

why?

Actually, this failure of the separate t's, with the better error
term, is fairly natural and predictable. The main test, with
p= 0.04, is a 1 d.f. test of the total across all 15 periods.

It is, in effect, a test of the composite score. When the
separate scores are measuring the same tendency, this
composite will be a (much?) more powerful test than
any of the separate tests.

Bruce suggested, earlier - when he thought you had an
interaction of Group with time - that the 1 d.f. test on
the linear *slopes* would be the powerful test of the
expected difference between groups. If the groups
*were* expected to diverge gradually from an initial
starting level of equality, then his suggestion *is* what
you ought to be testing.

What you ought to test depends on the hypothesis.
For an overall test of a single hypothesis, you usually
ought to aim at finding a single test with 1 d.f., which matches
the form of the expected differences.



--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Anova
    ... What analyses to pursue for treatments that are mathematically related ... given that there are significant differences from ANOVA? ... I was advised that it would be inaccurate to use SNK/Turkey test as the ...
    (sci.stat.consult)
  • Re: Anova
    ... With ANOVA there was significant differences in the treatments. ... applied SNK test and Tukey to determine where are the differences (40 ...
    (sci.stat.consult)
  • Re: Nested ANOVA
    ... > stats, so I'm having to play catch-up. ... Perhaps it will help to think about the difference between one-way ANOVA ... *crossed* with treatments. ... What is the consequence of leaving the nesting terms out of ...
    (sci.stat.edu)
  • Re: Using paired t-test for matched case-control subjects
    ... You can use pairwise t tests with pooled standard deviations, ... The way to go is to perform first an ANOVA with a linear model, ... > In standard experiemental design, ...
    (sci.stat.math)
  • ANOVA analysis help!
    ... So the design looks ... A,B,C,D are the different treatments and 1,2,3, 4 are the different ... Could I use 3-way ANOVA? ...
    (sci.stat.math)
Loading