[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Re: Comparisons in repeated measures anova

From   David Airey <>
Subject   Re: st: Re: Comparisons in repeated measures anova
Date   Wed, 25 Jun 2008 10:09:22 -0500


Your welcome. What I suggested is just a little easier for me in your situation and will give the same answers as contrasts, if you can figure out how to use "test" following a mixed model ANOVA. It would be nice to have more worked examples in the Stata FAQ for multiple tests following various split plot and repeated measures ANOVA models using the Stata "test" command. Even better would be to have classical ANOVA models translated to xtmixed, but continuing past the point of the omnibus tests (as is available at the ATS UCLA site) to post-hoc comparisons.


On Jun 25, 2008, at 8:54 AM, Jayne Lee wrote:

Thank you for your assistance. It did not occur to me use a corrected oneway anova after the repeated measures as I have always been taught that "contrasts" were the appropriate method (if used correctly!). The analysis now makes sense.

----- Original Message ----
From: David Airey <david.airey@Vanderbilt.Edu>
Sent: Tuesday, 24 June, 2008 9:02:57 PM
Subject: re: st: Re: Comparisons in repeated measures anova

Seems like you have one between sample factor (2 levels) and one
within sample factor (11 levels) and because you have an interaction,
you now want to test the simple effects, or the levels of system at
each time, or the levels of time within each system. For testing
different levels of the within sample factor (time) across the between
sample factor (system), you can just use a oneway ANOVA (or even ttest
since you have just two factor levels for system). You do have a
choice of errors for this test, one for just the time you are
interested in, or a pooled term, but I think most packages like SPSS
and SAS that automate this kind of ANOVA post estimation analysis (ie
give to you without asking) will use the MS at time k.
So, if you did
anova y a/s|a b a*b, repeated(b)
then if b = 1, 2, or 3, and a = 1 or 2
anova y a if b == 1
is fine, and applying a FWER as needed.
Also see the Stata FAQ of ANOVA.

I am stuck with evaluation of a comparison. I have carried out a
repeated measures anova of two systems (labeled 3 & 8) over 11 time
periods, and I want to see if the two systems differ at each time

. anova deltae sys / sample|sys time sys*time, repeated(time)

Repeated variable: time
                                         Huynh-Feldt epsilon
=  0.2594
                                         Greenhouse-Geisser epsilon
=  0.1914
                                         Box's conservative epsilon
=  0.1000

                                           ------------ Prob > F
                 Source |     df      F    Regular    H-F      G-
G      Box
                      time |     10    20.30   0.0000   0.0000
0.0000   0.0011
               sys*time |     10      5.27    0.0000   0.0076
0.0160   0.0446
               Residual |    100
The design matrix has 48 columns and I tried to compare the systems
at time 1 as follows:
. test _coef[time[1]*sys[3]] = _coef[time[1]*sys[8]]

( 1)  sys[3]*time[1] - sys[8]*time[1] = 0

      F(  1,   100) =   10.59
           Prob > F =    0.0016

I interpret this as the 2 systems being significantly different. The
problem I have is that the means of the two systems are 0.618 and
0.642. When the data is plotted there appears to be no difference at
this time and a t-test (I appreciate that it is not valid in this
design) has a probability 0.668 which seems to agree with the plot.

I would be grateful if somebody could tell what I am doing wrong in
this comparison and how to carry it out.

*   For searches and help try:

Sent from Yahoo! Mail.
A Smarter Email

*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index