Reviewers' comments 1. Comment on
Clinical vs Statistical Significance: Alan Batterham 2. Comment
on Qualitative vs Quantitative Research Designs: Keith Davids |
Comment on Clinical vs Statistical
Significance This item
is a very useful short critique of the poor scientific practice of
over-reliance on tests against the null hypothesis with arbitrary P values. I
hope that authors and reviewers take note. I wondered if some of the key
points could be highlighted with a pertinent quote or two to hammer home the
message? For example, the oft-quoted "Surely God loves the 0.06 nearly
as much as the 0.05?" (Rosnow and Rosenthal, 1989) would illustrate well
the point that one's article is much more likely to be accepted if p<0.05.
I have found that such quotes really help the lay reader and statistically
naive to grasp the point. The
points regarding a confidence interval approach to estimation are well made
and will help get across the key question in research: is the effect big
enough to be scientifically/clinically/practically relevant/important? I like
the comments regarding the potential problems with using 95% confidence
limits. This echoes your critique of the Bland/Altman 95% limits of agreement
for quantification of reliability and the superiority of the typical error.
Incidentally, Sterne and Smith (2001) have also opted for 90% confidence
limits, but they did not overtly provide a justification for their
recommendation. My reading of it is that they were proposing it more as a
deterrent to the practice of using the 95% limits as a surrogate means of
testing against the null hypothesis at the 0.05 alpha level and thus falling
into the same trap. I agree that choice of limits less than 95% may help
overcome the firmly entrenched 0.05 alpha level. Hopefully,
your article will help people to put the research question ahead of the straw
man of the null hypothesis and thus not allow the statistics to detract from
the ultimate vehicle for making inferences–the data themselves. I think the
key obstacle to this process is that, unlike a yes/no decision based on some
arbitrary alpha level, it requires genuine thought and intellectual rigor to
determine the smallest worthwhile effect for the variable in question! Yet,
that is, or should be, the crux of our scientific endeavor. Finally,
I wondered if the points in the last paragraph could be stated even more
emphatically or combatively? The misconceptions about what null hypothesis
testing does and does not tell us are close to universal and will not be
overcome without radical and persistent challenge! Rosnow RL, Rosenthal R (1989).
Statistical procedures and the justification of knowledge in psychological
science. American Psychologist 44, 1276-1284 Sterne
JAC, Smith GD (2001). Sifting the evidence–what's wrong with significance
tests. BMJ–British Medical Journal 322, 226-231 Comment on Qualitative vs
Quantitative Research Designs It is
good for scientists to have a range of methodological approaches to tackle
the large variety of experimental and practical questions in sport. Practical
work in coaching of team sports is often biased towards quantitative analysis
of the group, whereas historically psychotherapists and movement
rehabilitation therapists have preferred to treat each patient on an
individual basis. Single case studies are still relatively rare in the sports
sciences, although they are more common in the behavioral sciences
(Schöllhorn & Bauer, 1997). They are particularly useful in researching
performance of elite able-bodied and disabled athletes who are available only
in small numbers. The
emphasis on single subject designs and case studies recognizes the
significant amount of variability in human behavior. For example, in sport
science one aim of group work has been to identify key commonalities in
movement patterns that can act as a reference point for learners in skill
acquisition. During the modeling process in skill acquisition, these
reference values can take the form of an optimal kinematic pattern for
learners. Problems with the group approach can arise if the reference values
for a common optimal pattern for all learners are based on the performance of
one individual, for example a skilled athlete. The weaknesses with this
approach are based on the well-documented problems of providing average or
summarized feedback to groups of learners. Due to the unique constraints on
each individual learner, it is likely that group-based feedback will provide
a large proportion of irrelevant information for each individual. Traditional
group-based analyses tell us only part of the story, as each individual
attempts to find their own solutions to typical movement problems. A dynamical
systems perspective on movement coordination and control encourages a case
study approach by treating each individual performer as a unique system
learning to interact with the environment. Newell, Liu and Meyer-Kress (2001)
have recently shown how the traditional approach of averaging data for groups
across conditions and trial blocks may have masked the presence of different
types of learning curves in individuals (e.g., exponential, S-shaped, sudden
discontinuous), and they have questioned the ubiquity of the power law of
(motor) learning. Averaging scores over individual participants and trial
blocks ignores the fact that laws of learning should reflect both transitory
and persistent changes in behavior over time, whereas the power law approach treats
transitory effects as random-like behavior that can mask the persistent
trend. Statistical techniques of pooling group data or blocking trials have
an important role to play in quantitative research methods, in order to
examine central tendencies and dispersion, but they may limit insights into
the way that individuals solve coordination problems. For this reason, new
case-study methodologies such as coordination profiling (Button and Davids,
1999) and self-organizing maps (Bauer and Schöllhorn, 1997) are emerging in
motor behavior research to determine how each individual solves coordination
problems. Bauer HU, Schöllhorn W (1997).
Self-organizing maps for the analysis of complex movement patterns. Neural
Processing Letters 5, 193-199 Button
C., Davids K (1999). Interacting intrinsic dynamics and intentionality
requires coordination profiling of movement systems. In: Studies in
Perception and Action V (edited by MA Grealy and JA Thomson), pp.314-318.
Mahwah NJ: Lawrence Erlbaum Associates Newell
KM, Liu Y-T, Mayer-Kress G (2001). Time scales in motor learning and
development. Psychological Review 108, 57-82
|