On Separating Church and Lab
批评Rosenthal将伦理判断与科研质量挂钩的观点，认为这既错误又危险，并给出四点理由。适合关注研究伦理、实验偏差和科学方法论的学者阅读。
Robert Rosenthal has made highly original and important contributions in a number of different areas of psychology, including conditions for and effects of experimenter bias, the effects of expectancies on children's performance, the principles of nonverbal communication, and the evaluation of the importance and statistical reliability of empirical effects. He is certainly one of the most distinguished graduates of my own department (though before my time!). And he is an old friend, going back to our co-teaching of the introductory proseminar in social psychology at Harvard University in 1967, and reinforced by a year spent together as fellows at the Center for Advanced Study in the Behavioral Sciences. However, there is room for disagreement among friends, and this commentary presents a case in point. In his recent article on science and ethics, Rosenthal (1994) made a number of important points, picking up on themes he has developed through a long and illustrious career. He expressed proper concern about theoryconfirming actions such as selective dropping of data or subject or variable selection. He noted that extreme ''snooping or massaging of data can yield distorted p values. And he reiterated one of his favorite and most important themesthat small effects can be important if they are reliable, and that standard significance tests may underestimate that reliability. He emphasized the value of meta-analyses as methods for determining the reliability of effects. Indeed, one of his most consistent and most important contributions has been to force our attention away from evaluating results in single, isolated studies, and toward evaluating the pattern of findings over a long series of researches on the same hypothesis. On this latter point, one could argue that even he perhaps did not go far enough in this article. He argued that it is important to report all data, whether or not they support the original hypothesis. But built into the original conceptualization of questions are biases in favor of positive results, nonobvious findings, and novel findings, and these biases may yield quite misleading generalizations about human behavior. How far would Kahneman and Tversky have gotten if they had devoted their careers to showing that individuals normally make rational choices in simplified laboratory decision-making situations? Having said all this, I would like to take quite strenuous objection to Rosenthal' s views in some other particulars. He concluded his article saying, A central theme has been that the quality of our is not independent of the scientific quality of our research (p. 133). I think that in taking this position, Rosenthal, so clearh aded and farsighted in so many ways, made a serious e ror. I think it is not only wrong but also dangerous to apply ethical judgments to so many different areas of our work, for four reasons.