Tweet
As biologists we should always strive to obtain a sense of variation
within a subject population. Variation is real -- the mean is a,
sometimes inappropriate, statistical artifact.
That's the ideal, now the reality. It depends upon the purpose of the
gait data -- is it a clinical evaluation of a individual or a collection
of many individuals aimed at creating a population statement. It also
depends on how the gait data is collected, and what is collected. Some
data formats lend themselves to easy calculation of the mean, while in
other formats the method is not so obvious. Ease of calculation,
however, does not mean that the statistic is appropriate.
Generally, what you should strive for is a measure of central tendency
and the variation about that measure. In the clinical example,
collecting a few trials for each subject is typical. Sometimes
eyeballing the results to develop a sense of what is the typical outcome
for the patient is as valid as any calculation based statistic you might
want to employ.
When multiple trials are collected for each subject in a study that
includes several subjects I think it is best to represent each subject
by its median value. The median does not "assume" a normal distribution
for each data point and many times gait data will show some widely
divergent values. Those outliers would effect the calculation of the
mean, but not the median. By using the median for each subject you don't
have to employ preconceived notions about what is an outlier, which is
something that can be difficult to justify in a rigorous sense. An added
benefit -- if the sample population is normally distributed the median
and the mean will basically be the same value.
Once each subject value is determined, the next step is to create the
population statements, and perhaps comparisons across populations. If
comparisons are the goal then you are stuck with the mean, since very
few statistical methods are available that make use of other expressions
of central tendency. If, however, you are just aiming to make a
descriptive statement of the gait parameters within a single population,
then I would recommend use of the median again.
Thomas M. Greiner, Ph.D.
Assistant Professor of Anatomy
Dept. of Health Professions
University of Wisconsin - La Crosse
-----Original Message-----
From: * Biomechanics and Movement Science listserver
[mailto:BIOMCH-L@NIC.SURFNET.NL] On Behalf Of Anne-Marie Willems
Sent: Wednesday, May 03, 2006 5:41 AM
To: BIOMCH-L@NIC.SURFNET.NL
Subject: [BIOMCH-L] good practice: average gait trials?
Dear all,
Could you maybe share your thoughts on the pro's and con's of
averaging gait trials?
I am collecting spatio-temporal parameters of several gait tasks in
Parkinson's disease patients (straight line walking, turning).
The test-protocol was designed to comprise different (2 or 3 trials)
from each gait task in a randomized order.
The randomization was applied to exclude learning effects or changes
due to fatigue.
For statistical analysis, an average was calculated of all trials
within one gait task.
In my opinion averaging gait trials minimizes the variation that
exists between trials and increases reliability.
That is, as long as there are no systematical difference, e.g. learning
effect.
On the other hand, the gait parameters can vary between trial 1 and
trial 2, and such an averaging procedure may discard important
information.
Are there any agreements on 'good practice' on this issue?
I would really appreciate all replies.
Anne-Marie Willems
within a subject population. Variation is real -- the mean is a,
sometimes inappropriate, statistical artifact.
That's the ideal, now the reality. It depends upon the purpose of the
gait data -- is it a clinical evaluation of a individual or a collection
of many individuals aimed at creating a population statement. It also
depends on how the gait data is collected, and what is collected. Some
data formats lend themselves to easy calculation of the mean, while in
other formats the method is not so obvious. Ease of calculation,
however, does not mean that the statistic is appropriate.
Generally, what you should strive for is a measure of central tendency
and the variation about that measure. In the clinical example,
collecting a few trials for each subject is typical. Sometimes
eyeballing the results to develop a sense of what is the typical outcome
for the patient is as valid as any calculation based statistic you might
want to employ.
When multiple trials are collected for each subject in a study that
includes several subjects I think it is best to represent each subject
by its median value. The median does not "assume" a normal distribution
for each data point and many times gait data will show some widely
divergent values. Those outliers would effect the calculation of the
mean, but not the median. By using the median for each subject you don't
have to employ preconceived notions about what is an outlier, which is
something that can be difficult to justify in a rigorous sense. An added
benefit -- if the sample population is normally distributed the median
and the mean will basically be the same value.
Once each subject value is determined, the next step is to create the
population statements, and perhaps comparisons across populations. If
comparisons are the goal then you are stuck with the mean, since very
few statistical methods are available that make use of other expressions
of central tendency. If, however, you are just aiming to make a
descriptive statement of the gait parameters within a single population,
then I would recommend use of the median again.
Thomas M. Greiner, Ph.D.
Assistant Professor of Anatomy
Dept. of Health Professions
University of Wisconsin - La Crosse
-----Original Message-----
From: * Biomechanics and Movement Science listserver
[mailto:BIOMCH-L@NIC.SURFNET.NL] On Behalf Of Anne-Marie Willems
Sent: Wednesday, May 03, 2006 5:41 AM
To: BIOMCH-L@NIC.SURFNET.NL
Subject: [BIOMCH-L] good practice: average gait trials?
Dear all,
Could you maybe share your thoughts on the pro's and con's of
averaging gait trials?
I am collecting spatio-temporal parameters of several gait tasks in
Parkinson's disease patients (straight line walking, turning).
The test-protocol was designed to comprise different (2 or 3 trials)
from each gait task in a randomized order.
The randomization was applied to exclude learning effects or changes
due to fatigue.
For statistical analysis, an average was calculated of all trials
within one gait task.
In my opinion averaging gait trials minimizes the variation that
exists between trials and increases reliability.
That is, as long as there are no systematical difference, e.g. learning
effect.
On the other hand, the gait parameters can vary between trial 1 and
trial 2, and such an averaging procedure may discard important
information.
Are there any agreements on 'good practice' on this issue?
I would really appreciate all replies.
Anne-Marie Willems