View Full Version : summary responses on [posthoc test for EMG]

unknown user
12-08-2002, 02:28 PM
dear biomechanists,
I would like to thanks for the responses on my query [[BIOMCH-L] post hoc
test for EMG evaluation] .
herewith I post the responses, hopefully it would be usefull for others.

Best regards,
SWIE Yuniarto W
Master Course Student
Univ. of Electro-Communication (UEC), Tokyo JAPAN
Human Engr Research Group - Sakamoto Lab.

1. From: "David Collins" , To: SWIE Yuniarto Wijaya , Sent:
Thursday, December 05, 2002 7:17 AM

I have been struggling with methods of multiple pairwise comparisons for
years. I look forward to your summary to learn of other current solutions. I
assume you are interested in comparing summary values from each EMG time
series (e.g. onset time or a max value); methods of directly comparing time
series involve a very different discussion. The statistics software, SAS,
seems to offer the best analysis (high statistical power, control of type I
error/false positives) of mixed between subjects and repeated measures
designs, including pairwise comparisons, but not necessarily with default
options. Kowalchuk & Keselman (2001) recently addressed this issue:
Kowalchuk, R.K., & Keselman, H.J. (2001) Mixed-model pairwise multiple
comparisons of repeated measures means. Psychological Methods, 6: 282-296.
From a large simulation study, they recommend using a mixed linear
model (they only discuss the SAS analysis PROC MIXED; SPSS seems to have
more limited options), with Satterthwaite degrees of freedom and an
unstructured heterogeneous variance structure, with a stepwise procedure for
multiple pairwise comparisons. For the gait projects with which I have been
working, I have departed slightly from these recommendations, noted below.
Specifying a group variable (between subjects) causes estimation of a
separate covariance matrix for each group, so there will be separate
variance estimates for each group and each measure, and covariance
components between measures for each group. Autoregressive (first-order, or
AR(1) in SAS) and unstructured (UN) covariance structures are often
appropriate for repeated measures. In some cases, I have had trouble getting
SAS to fit the mixed model with UN and a group variable, so I have been
using AR(1) with a group variable.
The online SAS documentation
(http://v8doc.sas.com/sashtml/stat/chap30/sect35.htm#glmcm) seems to
recommend a Tukey-Kramer single stage adjustment for multiple pairwise
comparisons, or Duncan for comparisons to a control. The SAS statement
lsmeans compares the least-squares estimated means (rather than the raw
means) based upon the specified model (e.g. AR(1), heterogeneous). The
Tukey-Kramer method is used to adjust the T-test statistics for multiple
For a data set dat, sn=subject number, gr=a group variable, vi=visit or
other repeated measure, and var1=outcome measure of interest, the SAS syntax
proc mixed data=dat;
class sn gr vi;
model var1 = gr vi gr*vi / HTYPE=3 DDFM=SATTERTH;
repeated vi / type=AR(1) subject=sn group=gr;
lsmeans gr*vi / DIFF adjust=TUKEY;

Best wishes, Dave.
David R. Collins, Ph.D.

2. From: "Steve McCaw" , To: SWIE Yuniarto Wijaya ,
Sent: Thursday, December 05, 2002 12:31 AM
Subject: Re: [BIOMCH-L] post hoc test for EMG evaluation

> Yuniarto:
> Your choice of follow up test(s) depends on the initial means comparison
> test(s) you are conducting. Have you utilized a MANOVA, repeated measures
> ANOVA, multi-factor or oneway ANOVA? Your follow up tests to a significant
> F ratio depend on the type of initial test used and the nature of the
> research hypothesis. There is nothing special about conducting a post hoc
> test on EMG data compared to any other type of dependent variable. A
> typical statistics textbook will provide you with guidance in the
> of the appropriate post hoc test.


To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l