Page 156 - Contributed Paper Session (CPS) - Volume 7

MSSE analysis CPS2050 Marijke Welvaert
Number of hits for each search term per year were recorded. The search terms
were: RM‐ANOVA OR repeated measures ANOVA, linear mixed model, t test,
magnitude‐based inference, and bayesian.

3. Result
MSSE analysis
The journal publishes articles in 5 categories: Applied Sciences, Basic
Sciences, Clinical Sciences, Epidemiology and Special Communications. Three
quarters of the publications reported a within‐subject design and the majority
of publications (92.9%) utilised a frequentist inference approach. 3.5%
reported MBI hybrid results (i.e. MBI alongside frequentist p values). 2.4% of
the articles used a machine learning approach. A few single studies used
standalone MBI, graphical analysis or confidence interval analysis.
The reported sample size was grouped in 6 categories: 10, 11‐20, 21‐100,
101‐500, 501‐2000, >2000 to differentiate between small sample studies and
larger samples. Figure 1 shows the number of studies in each group per
category. Overall, 36% of publications reports a sample size between 20 and
100 and 30% have a sample size larger than 100 participants. Of the remaining
studies, 10% reports a sample size smaller than 10 and 24% have a sample size
between 11 and 20 subjects. Figure 1 illustrates that those smaller samples are
more common within the Applied Sciences.
The majority of studies reports either effect sizes (16.1%), confidence
intervals (20.4%) or both (16.5%). However, 42.7% of publications reports
neither. Figure 2 illustrates the use of effect sizes and confidence intervals in
those studies with up to 20 subjects in the sample.
34.5% of the studies utilised repeated measures ANOVA, 16.5% reported
a linear mixed model analysis, 14.5% reported a General Linear Model analysis
and 9.0% reported using t tests for their statistical analysis. Other less
frequently reported methods were Generalized Linear Model (4.3%), Cox
Proportional Hazard model (4.3%), Magnitude‐based inference (3.1%),
Structural Equation Modelling (2.7%), non‐parametric tests (1.6%), Random
Forests (1.2%), MANOVA (1.2%) and correlations (1.2%).
Web of Science analysis
Figure 3 illustrated the trends of frequency of the selected keywords across
publication years within the field of Sport Science. "t‐tests" is the most
frequently used with repeated measures ANOVA resulting in the second most
hits. Bayesian, Linear mixed model and Magnitude‐based inference
demonstrate similar reporting rates.

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