Page 155 - Contributed Paper Session (CPS) - Volume 7

CPS2050 Marijke Welvaert
the flagship journal of the American College of Sports Medicine) have changed
their policy and no longer accept submissions using MBI stating it is not an
acceptable method of statistical analysis (including labelling it as Bayesian).
The result is a polarised debate that is predominantly held outside of peer-
reviewed publications.
MBI gained popularity because it was providing the ultimate solution to
two problems that are claimed to be common for sport scientists: (1) sport
science studies have limited sample sizes because of practical limits in the
availability of its population of interest (e.g. elite athletes), and (2) sport science
studies are typically investigating small effects. As a net result, within a
frequentist null hypothesis significance testing (NHST) analysis framework, this
results in non-significant findings due to underpowered studies. MBI
completely rejects using p-values as an inference method, and promotes
interpreting magnitude of effects using effect sizes and confidence intervals.
The latter statisticians will agree with, but the problem arises with the
implementation of the method, in which probabilistic statements are asigned
to confidence intervals (see Sanaini, 2018 and Welsh & Knight, 2015 for a
thorough statistical review).
Going back to the original issues that MBI was aiming to address, the
question remains whether small sample sizes are indeed as common as was
suggested. Secondly, given that effect sizes and confidence intervals have
been well researched within the field of statistics, why is there the impression
that there is a need to create a “new” statistical method. This paper aims to
provide insight in recent statistical practices in sport science and a rough
estimation of the impact of MBI within the field. Based on those findings,
suggestions for further statistical education within sport sciences with the goal
of elevating statistical practice in the field are formulated.

2. Methodology
The statistical methods sections of all papers published in MSSE in 2018
were analysed. The journal published 329 papers in total, of which 291 were
research articles. A further 36 papers were excluded because they did not
report human subject data (e.g. animal studies, mechanistic modelling, meta‐
analyses, etc.). The remaining 255 articles were assessed on the following:
study design (within‐and/or between‐subject; statistical analysis method (or
family in case multiple techniques were used); statistical school (frequentist,
bayesian or other); reports effect sizes (Yes‐No); reports confidence intervals
(Yes‐No); and sample size.
Additionally, a keyword frequency analysis of the literature was performed
using the Web of Science (WoS) database to provide an estimation of
popularity of selected statistical methods within sport science. The search
included all indexed publications within the research field from 1991 to 2018.

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