The tech revolution has modified our behaviour and now it is also revolutionizing sports. More and more tools are capable of collecting a great amount of data, which can help us make better decisions on the pitch. That is why it is interesting to capture all this data because it can help win matches. Although a great effort is done to improve the collection of data for real-time information, we still lack a good accurate interpretation of this information to help improve the coaches decisions.

This limitation may be related to statistical data analysis. Typically, sports scientists have focused their reports and propositions on the P-value application, where value is significant “yes or no”. However, it seems that statistical models that allow informing quantitatively as well as qualitatively about the possible changes for a variable to be real can be much more useful.

In a recent article, Martin Buchheit¹ suggests researchers should abandon the conventional null hypothesis significance testing (NHST) and complement it with magnitude-based inferences (MBI) for different reasons:

• P-values and, at the same time, the research findings, depending on the size of the sample (with a greater n, a smaller P-value), regardless of the size of the effect.² Thus, as long as it is possible to conclude, for example, that a nutritional supplement is not efficient with a sample of 12 players (P>0,05), the same comparison makes the treatment become useful with n=14 (P>0,05).³ In other words, having two subjects more or less in the sample causes a complete change in the final results. All this is especially relevant when team sports are being studied, where the quantity of players analyzed is always relatively low. Teams provide 12, 16 or 25 players depending on the sport analyzed.
• Statistical significance doesn’t report the magnitude of the effects, which is precisely what matters most.⁴ With a sufficient size of the sample, even the smallest effects (very small), trivial (trivial) or non-practical (non-practical) might become significant (P<0,05). For example, with 200 players an improvement of 0.01% in the performance, NHST would suggest the efficiency of the nutritional supplement although the improvement it generates is irrelevant. However, it is likely that coaches and players choose to know how much this supplement allows them to improve their performance.
• MBI allows researchers to be honest with the size of the sample and it better recognizes the trivial effects.
• The magnitude exam per se helps provide better research questions. Taking into account that the effect of the size induces to a response that goes beyond a yes or a no (NHST), a typical hypothesis that doesn’t have a clear rationale (x application improves performance) might be replaced by more relevant and specific statements (x application improves performance in such percentage).
• MBI is suitable for available calculation sheets which can be found for free on the Internet⁵.
• MBI allows a better graphic representation and visualization of the data.

As a conclusion, a better data interpretation and communication to coaches about the findings through reports or charts can be complemented by adding the MBI to the conventional NSHT. Keep in mind that it is about quantifying better each time the effect of an exercise or treatment on players performance, and helping coaches make better decision to enhance the chances to win.

Carlos Lago Peñas

References:

¹ Buchheit, M. (2016). 2016, 11, 551 – 554 3 4 Title: The numbers will love you back in return – I promise, International Journal of Sports Physiology and Performance, 11, 551-554.

² Hopkins, W.G. y A.M. Batterham (2016), Error Rates, Decisive Outcomes and Publication Bias with 200 Several Inferential Methods. Sports Medicine, 46(10): 1563-1573.

³ McCormack, J., B. Vandermeer, and G.M. Allan. (2013). How confidence intervals become confusion intervals. BMC Medical Research Methodology, 13:134.

⁴ Cohen, J (1994). Things I have learned (so far). American Psychologist, 45:1304-1312.

⁵ Hopkins, W.G. A spreadsheet for deriving a confidence interval, mechanistic inference and 207 clinical inference from a P-value. Sportscience, 2007. 11, 16-20. DOI: 208 http://newstats.org/xcl.xls