【作者】 马奕帆； 贺冠瑞； 罗照盛；

【Author】 Yifan Ma;Guanrui He;Zhaosheng Luo;School of Psychology,Jiangxi Normal University;

【机构】 江西师范大学心理学院； 江西师范大学心理学系；

【摘要】 在实证研究中,研究者需要进行包含了若干分析决策(analytical decisions)的数据分析过程。然而,这些决策的做出往往具有武断性,这是因为1)研究者会根据自己的观点和理解进行决策,不同的研究者之间可能存在着差异;2)针对同一批数据,研究者可能会做出多种分析决策,但是研究者一般会选择其中一部分分析决策下的假设检验结果进行报告,而不是报告所有研究者认为有意义的决策下的结果。这些情况叫做研究者自由度(researcher degrees of freedom),也使得研究结果面临诸如事后假设、P值操纵(p-hacking)等方面的质疑。对于这一问题,有研究者提出使用决策曲线分析(specification-curve analysis, SCA)技术来尝试降低研究过程中的武断性。SCA是一种统计检验方法,它可以藉由检验所有合理的统计决策组合的结果,来检测主效应的稳健性,即检验研究结果显著是因为主效应本身确实存在,还是由于特殊的统计决策组合所造成的。同时,使用SCA进行统计检验可以使研究的结果更可靠,这尤其体现在对大样本、结构复杂的研究数据分析上。SCA包括三个主要步骤,1)列出某个研究所有合理且不冗余的分析决策;2)分析所有合理的分析决策下假设检验的显著性,并绘制决策分析曲线展示结果;3)对第二步的数据和结果进行统计检验,以判断主效应的稳健性。本研究对SCA实际应用的一般步骤进行了展示,为心理学研究统计分析过程提供了一个新的视角。更多还原

【Abstract】 In empirical researches, researchers need to complete a data analysis process that includes some analytical decisions. However, these decisions are always arbitrary because 1) Researchers make decisions according to their thoughts and understandings. Different researchers may make diffident decisions. 2) Researchers may make several different decisions in one study. But they would choose only a part of their hypothesis significance test results to report, not the results derived from the whole analytical decisions. This phenomenon called researcher degrees of freedom. It also makes the researches face the doubt of post-hoc selection, p-hacking and so on. To solve this problem, researchers develop a new method called specification-curve analysis(SCA) to try to decrease arbitrary in studies. SCA is a kind of statistical test method. It can test all the combinations of reasonable analytical decisions to verify the robustness of the main effects. It means this method can test whether the main effect exists or not. At the same time, the use of SCA for statistical tests can make the results of the research more reliable, especially in the analysis of large sample and complex research data. Specification-Curve Analysis is carried out in three main steps. 1) Define the set of reasonable and non-redundant analytical decisions to estimate. 2) Estimate all analytical decisions and report the hypothesis significance test results in a descriptive specification curve. 3) Conduct joint statistical tests to results from the step2 to verify the robustness of main effects. We demonstrated the main steps of SCA. Specification-Curve Analysis provides a new view of psychological statistical analysis.更多还原