The P-Value Debate and Implications for Management Research : The Significance of A Priori Probability
Résumé
This paper argues that the P-value debate is grounded in the confusion between statistical significance testing and statistical inferences. Using results of statistical significance testing to make statistical inferences can lead to the validation of many specious theories. While errors in statistical significance testing refer to the probability of false positives, errors in statistical inferences lie in the inverse probability of false positives. The relationship between two probabilities is highly affected by a priori probability. A higher a priori probability decreases the probability of errors when using results of statistical significance testing for statistical inferences. By contrast, a lower a priori probability increases the probability of statistical inference errors even if the probability of false positives is low. In order to improve the reliability of statistical inferences, this paper draws on management research and proposes to better establish a priori probability through a dual dialogue with practical and theoretical worlds. By doing so, this paper underlines the pervasiveness of underdetermination and sheds light on the interplay of the a priori and the empirical in management research.
P值之争的本质是人们把统计检验和统计推断混淆，直接用统计检验结果来进行统计推断。这导致许多似是而非的理论被接受。统计检验犯错的是假阳性概率。统计推断犯错的是假阳性之反概率。两种概率之间的数值关系受先验概率影响很大。先验概率越高，利用统计检验的结果进行推断所犯错误概率就越小。如果先验概率很小，即使假阳性概率很小，利用该结果进行推断所犯错误概率也很大。本文以管理学研究为例，提出解决该问题的办法是，通过与现实和与理论的双重对话来掌握先验概率，以提高基于统计检验的结果进行统计推断的可靠性。本文的科学哲学意义是，揭示了科学研究是先验与经验相统一的本质，以及科学研究的不充分决定性的特征。