心理研究引入贝叶斯统计推断的必要性、应用思路与领域

Bayesian Inference be Introduced to the Psychological Research: Its Necessity,Ideas and Field

本文从科学范式演变和概率统计推断特征分析出发,就贝叶斯统计推断引入心理学的必要性、怎么应用、在哪些领域应用进行了论述。首先在分析科学范式演进各阶段统计推断任务需要、经典的频率概率统计推断不足基础上,得出科学范式处于危机和革命阶段时需要对相关理论做可信度检验。然后,在介绍贝叶斯定理及其统计推断的基础上,进一步分析了贝叶斯推断能解决理论可信度的多种假设检验情景。最后本文还就贝叶斯统计推断在心理学理论争鸣与建构、心理技术产品开发与评估具体领域应用做了分析。

It is not adequate to only use the frequency probability inference in research for four reasons. Firstly, the tasks of statistic inference is different in stages of scientific paradigm progress, statistic inference should help to find some new knowledge based on the normal distribution, test the reliability of theories in the crisis and select more reasonable theory. Secondly, the frequency probability inference can＇t answer the reliability of theory when the science paradigm progress is in the crisis and revolution stage. Thirdly, the dependency on the sample of frequency probability inference leads to II or I error even in considering the test power. Lastly, it can＇t test whether H0 is true or false when H1 is rejected in condition p 〉0.05.So it is necessary to use another kind of statistic inference to test theories, tell more information about test and not affected by sample size. Because of the features of Bayesian theorem which would be as the fallows, Bayesian statistics inference could be one way to those questions above. Firstly, tip = P（BIA）/P（B） in the Bayesian theorem, tip as continuous value means the probability of A event happen under different conditions, So tip could be used test which theory is better than the others in the same context. Secondly, tip value is not affected by the sample size, when tip is close to 0, it means that A event in the condition B hardly happen, when βp close to 1, it means that B condition is not sensitive, when βp is far more than 1, A event in the condition B always happen. So βp could satisfy the need to test the reliability of theories, tell that H0 is true or data is not sensitive when H1 is rejected. Bayesian statistics inference should be used in inference. In the psychology, there are more conflict theories to interpret one mental phenomenon, and there are more interventions about the same mental function, then how to select one better theory or intervention is the more important question to be answered. We can set some number of prior probabilities according to theories, we get one post probability in one operating context, then a number of βp could be gotten by comparing the former with the latter. After that, tip compared with 0 and 1, it is clear that which theories are supported or not by post probability, As for the new intervention selected, we can set a higher prior probabilities compared with the old interventions, then one post probability cold gotten in new interventional context, then tip value and its meaning could be gotten as the above, one better intervention product could be selected from some ones. Although there are a lot of criticism to Bayesian statistical inference, such as subjective probability and theoretical belief adjusted by facts, Bayesian statistical inference open a new way to statistical inference and date analysis, even lead a revolution on the statistic inference paradigm in psychology by emphasizing the theories test. After all, there are more theories built in the last one hundred years more.