講演論文
新しい統計学とのつきあい方
2019 年 37 巻 2 号 p. 167-173
詳細
2019 年 37 巻 2 号 p. 167-173
In this paper, I clarify the problem of null hypothesis significance testing along with Kruschke (2014) and point out five advantages of Bayesian statistics. First, it is not necessary to convert data for NHST. Second, additional assumptions or corrections are not required. Third, there is no need for any preliminary design of the verification plan. Fourth, the Bayesian approach allows an intuitive interpretation of results. Fifth, the sample size does not cause critical problems. Beyond these advantages, Bayesian statistics can be used alongside frequentism and likelihoodism methods. Finally, I argue it is necessary in science communication to clearly express the researcher’s premise as a prior distribution or likelihood function.
