幅度推論與運動科學偽真性之探討

Inspection of magnitude-based inference and false-positive results in exercise science studies

國際運動科學領域開始關注幅度推論對研究結果容易造成偽真性推論，但臺灣相關研究人員似乎尚未明瞭幅度推論可能產生的問題；因此，本文旨在探討下列課題：一、幅度推論的背景；二、幅度推論產生偽真性推論的機率；三、幅度推論造成波動式偽真性推論的原因；四、判別及應用幅度推論研究結果的方法。本文回顧文獻後獲致下列結果：一、幅度推論強調實際影響效應，因而採用效果量發生機率進行推論，所以易於過度推論，迄今也未經統計驗證，而且創立者會不定時更新幅度推論的制式Excel表單，不斷地增加幅度推論的計算方式及其結果的不確定性。二、幅度推論會隨著樣本數變化，產生約2％至55％的偽真性推論機率，與頻率推論差異約0.4至11倍，其波動式機率取決於實驗設計、樣本數及幅度推論標準設定；除此之外，增加依變項量測的次數，會把偽真性研究結果發生機率提高至99％。三、幅度推論造成波動式偽真性推論的主要原因為方法設計不周。四、引用幅度推論研究結果時，建議請統計人員檢閱研究內容，並依95％最小獲益機會作為判斷標準，或者使用線上轉換器將數據轉換為頻率推論結果，以校正其研究結果及結論。

The field of international sports science has become concerned that magnitude-based inference tends to result in false inferences during studies. However, researchers in Taiwan have not yet explored the potential problems associated with magnitude-based inference. Therefore, this review article aimed to inspect the following topics: 1. the background of magnitude-based inference; 2. the probability that magnitude-based inference will result in false-positive inferences; 3. the factors that contribute to wave-type false-positive inferences; and 4. the identification and application of results derived from magnitude-based inferences. By reviewing studies that have employed magnitude-based inference, the following points were identified. 1. Magnitude-based inference employed effect size probabilities to perform statistical inferences, which emphasizes the actual impacts. Thus, magnitude-based inference tends to be optimistic and lacks statistical evidence. Moreover, the irregular updating of a standard Excel sheet used for magnitude-based inferences might increase the uncertainty of the calculation methods and corresponding results of magnitude-based inferences. 2. Differences in the sample size can affect the probability of false inferences, which ranged from 2% to 55%. This probability range was approximately 0.4 to 11 times as high as that inferred from frequentist inference. The fluctuation depended on the experimental design, sample size, and the standard setting of the amplitude inference. In addition, the probability of false research results increased to 99% when the testing times for dependent variables increased. 3. The primary reason for false-positive results caused by magnitude-based inference was poor methodology. 4. When citing the results of magnitude-based inference studies, asking a statistician to review the contents using the 95% minimum chance of benefit evaluation or adopting the online APP to convert the results into frequency inference is recommended to correct the results and conclusions.