Page 157 - Contributed Paper Session (CPS) - Volume 7

CPS2050 Marijke Welvaert
4. Discussion and Conclusion
Magnitude-based inference became popular in Sport Science as a remedy
towards underpowered studies and small studies. Recent statistical reviews
(Welsh & Knight, 2015; Sanaini, 2018) invalidated the method for not being
founded in statistical principles and as a result Sport Science has been
victimised for using "shoddy statistics". The data presented in this paper
support a more nuanced picture. The Web of Science frequency analysis
provides some evidence that MBI shows up in only a small proportion of
publications, and definitely far less than more traditional frequentist methods.
It should be noted though that the database search might not necessarily
correlate with the actual usage statistics of the method as there were only 75
hits in the WoS search while the original MBI paper (Hopkins et al., 2009) has
over 2,000 citations thus far.
Sample sizes within Sport Sciences as captured by the 2018 MSSE
publications vary a lot. Interestingly though, when we combine the reported
sample size with the usage of inference beyond p-values, using either effect
sizes or confidence intervals or both, 60% of studies reported at least one of
those. However, it were especially studies with a vary small sample (<8) that
did not include any magnitude information for their inference. It could be
argued that these studies in particular would benefit to supplement their
analysis with effect sizes and/or confidence intervals.
Given the majority of studies utilising a within-subject design, it is not
surprising that RM-ANOVA is a frequently chosen analysis technique.
However, in the context of smaller samples, more complex designs and
dealing with missing data, the linear mixed model (and by extension Bayesian
hierarchical modelling) would provide a more powerful solution.
Statistical education within Sport Science should maintain a focus on the
importance of magnitude-based inference as a concept but using statistically
validated tools. By utilising effect sizes and confidence intervals, which are
standardly available in statistical software packages, sport scientists can still
draw conclusions from their data that are informative beyond statistical
significance. In addition, promoting more modern methods that handle
missing data would further enhance the ability to design powerful studies
within the constraints of the population of interest.

144 | I S I W S C 2 0 1 9